UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Government intervention in financial markets Xiao, Kairong 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2017_september_xiao_kairong.pdf [ 4.92MB ]
JSON: 24-1.0348741.json
JSON-LD: 24-1.0348741-ld.json
RDF/XML (Pretty): 24-1.0348741-rdf.xml
RDF/JSON: 24-1.0348741-rdf.json
Turtle: 24-1.0348741-turtle.txt
N-Triples: 24-1.0348741-rdf-ntriples.txt
Original Record: 24-1.0348741-source.json
Full Text

Full Text

GOVERNMENT INTERVENTION INFINANCIAL MARKETSbyKairong XiaoA THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Business Administration)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)July 2017© Kairong Xiao 2017AbstractGovernments play an important role in financial markets around the world. This thesis studies theoreti-cal mechanisms and empirical consequences of government actions in financial markets in order to betterunderstand the organization of the financial sector and the inner working of governments. The first essay“Shadow Banks, Deposit Competition, and Monetary Policy” studies the transmission mechanism of mon-etary policy through the shadow banking system, a group of non-bank financial intermediaries conductingbanking business in the economy. This essay shows empirically and theoretically that the shadow bankingsystem partially offsets the impact of monetary policy on the traditional commercial banking system andmay lead to unintended consequences in terms of the stability of the financial system. The second essay“Regulation and Market Liquidity” (co-authored with Professor Francesco Trebbi), explores whether thepost-crisis financial regulations, including the Dodd-Frank Act and Basel III, have caused liquidity deterio-ration in the U.S. fixed income market. Against the popular claim that post-crisis regulations hurt liquidity,this essay finds no evidence of liquidity deterioration during periods of regulatory intervention. Instead,liquidity seems to have improved in this period. The third essay, “Factions in Nondemocracies: Theoryand Evidence from the Chinese Communist Party” (co-authored with Professor Patrick Francois and Profes-sor Francesco Trebbi), investigates theoretically and empirically the factional arrangements and dynamicswithin the Chinese Communist Party (CCP), the governing political party of the People’s Republic of China.This essay documents a set of new empirical findings showing how factional politics affects the promotionof individual politicians within the CCP hierarchy. This essay proposes a theoretical model to rationalizethese findings and conduct a set of counterfactual analyses of possible institutional changes within the CCP.iiLay SummaryThis thesis studies the interaction between governments and financial markets. The first essay “Shadowbanks, Deposit competition, and Monetary Policy” shows that the shadow banking system, a group of non-bank financial intermediaries conducting banking business in the economy, can partially offset the impactof monetary policy. The second essay “Regulation and Market Liquidity” (co-authored with ProfessorFrancesco Trebbi), shows that the post-crisis financial regulations have not caused liquidity deteriorationin the U.S. fixed income market. The third essay, “Factions in Nondemocracies: Theory and Evidencefrom the Chinese Communist Party” (co-authored with Professor Patrick Francois and Professor FrancescoTrebbi), investigates how informal groups of politicians known as factions affect the promotion of individualpoliticians within the Chinese Communist Party (CCP). In summary, this thesis helps us to better understandthe important role of governments in financial markets.iiiPrefaceThe research project in Chapter 2 was identified and performed solely by the author. The essay in Chapter3 is based on unpublished research with Francesco Trebbi (University of British Columbia). The essay inChapter 4 is based on unpublished research with Francesco Trebbi (University of British Columbia) andPatrick Francois (University of British Columbia). In the coauthored work of Chapters 3 and 4, I worked onthe development of the research question, data collection, empirical analysis, structural estimation, and partof the writing of the manuscript. While it is hard to quantify exactly, my personal shares of contribution toChapters 3 and 4 amount to about 1/2 and 1/3.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Shadow Banks, Deposit Competition, and Monetary Policy . . . . . . . . . . . . . . . . . . . 22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22.2 Deposit Creation by Shadow Banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 A Structural Model of Bank Competition . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4 Structural Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.5 Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.7 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Regulation and Market Liquidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503.3 Econometric Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.4 Results for Market Liquidity of U.S. Corporate Bonds . . . . . . . . . . . . . . . . . . . . 583.5 Results for Market Liquidity of U.S. Treasuries . . . . . . . . . . . . . . . . . . . . . . . . 643.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.7 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674 Factions in Nondemocracies: Theory and Evidence from the Chinese Communist Party . . . 834.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.2 Institutional Background: the CCP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 874.4 CCP Factional Politics: Reduced Form Results . . . . . . . . . . . . . . . . . . . . . . . . 884.5 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.6 Maximum Simulated Likelihood Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 100vTable of Contents4.7 CCP Factional Politics: Structural Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014.8 Counterfactuals and Model Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1044.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1074.10 Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1084.11 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1165 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142viList of Tables2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.2 Effect of Monetary Policy on Aggregate Deposit Growth Rates . . . . . . . . . . . . . . . . 372.3 Demographic Determinants of Shadow Bank Deposit Holding . . . . . . . . . . . . . . . . 382.4 Demand Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392.5 Supply Parameter Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402.6 Own-rate Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412.7 Cross-rate Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422.8 Monetary Policy and MMF Lending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432.9 Monetary Policy and Asset Growth of Shadow Banks . . . . . . . . . . . . . . . . . . . . . 443.1 Summary Statistics of the U.S. Corporate Bond Liquidity (Aggregate-level) . . . . . . . . . 783.2 Correlation Table of the U.S. Corporate Bond Liquidity (Aggregate Level) . . . . . . . . . . 793.3 Difference-in-Difference Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803.4 Summary Statistics of the U.S. Treasury Liquidity . . . . . . . . . . . . . . . . . . . . . . . 813.5 Correlation Table of the U.S. Treasury Liquidity . . . . . . . . . . . . . . . . . . . . . . . . 824.1 Summary Statistics of Elites in China . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1284.2 Summary Statistics of Central Committee Members . . . . . . . . . . . . . . . . . . . . . . 1294.3 Geographical Distribution of Factions and Groups . . . . . . . . . . . . . . . . . . . . . . . 1304.4 Factional Mix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1314.5 Factional Mix (Shanghai vs. CYCL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324.6 Leadership Premia in Promotion and Retirement . . . . . . . . . . . . . . . . . . . . . . . . 1334.7 Leadership Premia in Power Score and Seat Shares . . . . . . . . . . . . . . . . . . . . . . 1344.8 Anticorruption and Factional Affiliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1354.9 Parameter Estimates of the Faction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 1364.10 Parameter Estimates of Alternative Models . . . . . . . . . . . . . . . . . . . . . . . . . . 1374.11 Share of promotion of Each Faction by Level of the Central Committee . . . . . . . . . . . 1384.12 Tests of Xi’s Factional Affiliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1394.13 Out-of-sample Forecast of 19th Central Committee . . . . . . . . . . . . . . . . . . . . . . 140viiList of Figures2.1 Deposit Growth Rates and the Fed Funds Rates . . . . . . . . . . . . . . . . . . . . . . . . 242.2 The U.S. Banking System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3 Deposit Rates and the Fed Funds Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262.4 Model Predicted Deposit Growth Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.5 Distribution of Estimated Convenience . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.6 Distribution of Estimated Demand Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . 292.7 Difference in Deposit Rates (MMF-CB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.8 Difference in Markup and Marginal Cost (CB-MMF) . . . . . . . . . . . . . . . . . . . . . 312.9 Choice Probability of Depositors by Type . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.10 Counterfactual Aggregate Money Growth Rates . . . . . . . . . . . . . . . . . . . . . . . . 332.11 Demand Elasticity and the 2008 Runs on MMFs . . . . . . . . . . . . . . . . . . . . . . . 342.12 Change in Depositor Surplus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.1 Timeline of Crisis and Post-Crisis Regulatory Activities . . . . . . . . . . . . . . . . . . . . 673.2 Simulated Illiquidity Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.3 Time Series of Liquidity of U.S. Corporate Bonds (Aggregate-level) . . . . . . . . . . . . . 693.4 Breaks in the Means of Liquidity (Disaggregate-level) . . . . . . . . . . . . . . . . . . . . . 703.5 Breaks in the Means of Liquidity by Underwriter (Disaggregate-level) . . . . . . . . . . . . 713.6 Breaks in the Means of Liquidity by Underwriter (Disaggregate-level) . . . . . . . . . . . . 723.7 Liquidity Index of the U.S. Corporate Bond Market . . . . . . . . . . . . . . . . . . . . . . 733.8 Liquidity of Volcker Rule and Non-Volcker Rule Bonds (Matched Sample) . . . . . . . . . 743.9 Primary Dealer Corporate Bond Holding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753.10 Fraction of Agency Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763.11 Time Series of Liquidity of the U.S. Treasury Bonds . . . . . . . . . . . . . . . . . . . . . 774.1 Geographic Distribution of Factions or Groups (1956-2014) . . . . . . . . . . . . . . . . . 1164.2 Leadership Premium in Promotion Rates of Each Faction or Group . . . . . . . . . . . . . 1174.3 Leadership Premium in Power Score of Each Faction or Group . . . . . . . . . . . . . . . . 1184.4 Power Score of Each Faction or Group in the Central Committee . . . . . . . . . . . . . . . 1194.5 Power Score of Each Constituency in the Central Committee . . . . . . . . . . . . . . . . . 1204.6 Seat Shares at Each Level of the Central Committee . . . . . . . . . . . . . . . . . . . . . . 1214.7 Aggregate Share of Promotions over Time . . . . . . . . . . . . . . . . . . . . . . . . . . . 1224.8 Model Fit (In Sample) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1234.9 Meritocracy (In Sample) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1244.10 Model Fit (Out of Sample) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1254.11 Counterfactual Aggregate Share of Promotions over Time (Leadership Premium × 2) . . . . 1264.12 Counterfactual Aggregate Share of Promotions over Time (Li Keqiang Presidency) . . . . . 127viiiAcknowledgementsI would like to express my special gratitude and thanks my thesis advisors, Adlai Fisher and Lorenzo Gar-lappi, for their invaluable guidance throughout my graduate studies. Your advice, both on my research andmy career, has been most useful to help me grow as a researcher. I am also grateful to Carolin Pflueger andFrancesco Trebbi for joining my committee and providing many constructive suggestions. In addition to mycommittee, I would also like to thank the rest of the UBC Finance faculty for their insightful feedback induring my PhD study. Many thanks go to my parents and my wife, Stacy. None of this would have been pos-sible without your unconditional love and sacrifice over these years. Finally, I would like to acknowledgethe financial support provided by UBC, St John’s College, and Canadian Securities Institute that fundedparts of the research in this thesis.ixChapter 1IntroductionThis thesis is a collection of three essays at the intersection of finance, industrial organization, and politi-cal economy. Although the topics are diverse, they share the common objective of studying the interplaybetween governments and financial markets. The first essay studies the transmission mechanism of mon-etary policy through the shadow banking system, a group of non-bank financial intermediaries conductingbanking business in the economy. Using the U.S. money aggregate data from 1987 to 2012, I find that theshadow banking system partially offsets the impact of monetary policy on the traditional commercial bank-ing system. I construct a structural model of bank competition and show that this new channel is a resultof deposit competition between commercial and shadow banks in a market with heterogeneous depositors.The second essay empirically examines the effects of the post-crisis financial regulations, encompassing theDodd-Frank Act and Basel III, on market liquidity of the U.S. fixed income market. Against the popularclaim that post-crisis regulations hurt liquidity, this study finds no evidence of liquidity deterioration duringperiods of regulatory intervention. The third essay investigates theoretically and empirically the factionalarrangements and dynamics within the Chinese Communist Party (CCP), the governing political party ofthe People’s Republic of China. This study presents a set of new empirical regularities within the CCP anda theoretical framework suited to model factional politics within single-party regimes. Because each essayinvestigates a different topic, chapters were designed to be self-contained. I thus leave a more exhaustivediscussion of the research question and contribution to the introduction specific to each chapter.1Chapter 2Shadow Banks, Deposit Competition, andMonetary Policy2.1 IntroductionThe U.S. banking system has experienced significant structural changes over the past thirty years. A groupof non-bank financial intermediaries, collectively known as the shadow banking system, has grown outsideof the traditional commercial banking sector. Important components of the shadow banking system includemoney market funds (MMFs), securitization vehicles, broker-dealers, and mortgage companies. Shadowbanks compete with commercial banks in many traditional banking businesses. For example, MMFs com-pete in the deposit market by creating liquid claims which, in many ways, are similar to commercial bankdeposits, yet provide a higher yield. In recent years, more than 30% of deposits have been created by shadowbanks.The rapid growth of shadow banks has raised two main concerns for policy makers.1 The first concern re-gards the effectiveness of monetary policy in the presence of a sizable shadow banking sector. Traditionally,commercial banks play an important role in transmitting monetary policy to the real economy. However, alarge proportion of deposits are now created outside of the commercial banking sector. How does the depositcompetition from shadow banks affect the transmission of monetary policy? The second concern regardsthe effect of shadow banks on financial stability. Deposit creation involves the risk of bank runs. The riskof bank runs is more severe for shadow banks, because their deposits are not insured by the government.By creating uninsured deposits, shadow banks may have negative impacts on financial stability. Motivatedby these concerns, this paper examines how shadow bank deposit competition affects the transmission ofmonetary policy, and offers an analysis of the implications of such competition for financial stability.Unlike commercial banks which combine deposit creation and loan origination under one roof, theshadow banking system separates the intermediation process into different entities. MMFs provide depos-itory services for households and businesses, and then pass the proceeds to other shadow banks such asmortgage companies which specialize in loan origination. This paper focuses on MMFs, as these are themain entities which create deposits in the shadow banking system.I first document a new transmission channel of monetary policy in the shadow banking system. Standardtheories of monetary transmission predict that high interest rates are associated with low deposit creation(Bernanke and Blinder, 1988; Drechsler, Schnabl, and Savov, 2016).2 This prediction has been verified em-pirically by previous literature in the commercial banking sector (Kashyap and Stein, 1995, 2000; Drechsler,1For instance Federal Reserve Board Chair, Janet Yellen, in response to a question by IMF Managing Director Christine Lagardeon Shadow Banking in July 2014, said that “we won’t be able to detect them (shadow banking), and if we can, we won’t haveadequate regulatory tools. That is a huge challenge to which I don’t have a great answer”.2The shared idea of these two theories is that high interest rate policy increases the opportunity cost of holding liquid deposits,which reduces the amount of bank deposits in economy. The difference between these two theories is how high interest ratepolicy increases the opportunity cost of holding liquid deposits. Bernanke and Blinder (1988) suggest the reserve requirement ofcommercial banks as an important channel, while Drechsler, Schnabl, and Savov (2016) show that market power of commercialbanks can also play a role.22.1. IntroductionSchnabl, and Savov, 2016). However, using aggregate U.S. money supply data from 1987 to 2012, I findthe opposite of commercial banks happens for shadow banks.3 When the Federal Reserve wants to reducedeposits by raising interest rates, shadow bank deposits expand dramatically, and as a result, dampen theimpact of monetary policy. The contrast between shadow and commercial banks can be easily seen in atime-series plot of the deposit growth rates as shown in Figure 2.1. This finding contradicts conventionalwisdom that high interest rates are contractionary for deposit creation. It suggests that the monetary trans-mission channel in the shadow banking sector is different from the traditional channels in the commercialbanking sector. Moreover, my results show that monetary policy not only affects the total amount of bankdeposits, but also the relative shares between the shadow and commercial banking sectors. Because shadowbank deposits are outside of government safety nets such as the deposit insurance and the discount window,shifts in the relative shares of deposits have important implications for financial stability. To the best of myknowledge, the present study is one of the first to document this counter-intuitive results of shadow bankdeposit creation.In order to understand the underlying mechanism, I develop a structural model of bank competitionfollowing the industrial organization (IO) literature on oligopoly markets.4 I show that the expansion ofshadow bank deposits during periods of monetary tightening is a result of deposit competition betweencommercial and shadow banks. In my model, banks are differentiated by their respective degrees of trans-action convenience and yields. Commercial banks provide superior transaction services such as branchnetworks and payment systems, while shadow banks compete on yields due to the lack of bank charters tooffer transaction services similar to those offered by commercial banks. Banks compete for a continuum ofdepositors with different preferences over transaction convenience and yields. Commercial banks attract agroup of transaction-oriented depositors who value transaction services but are insensitive to yields. Typicalexamples of transaction-oriented depositors include small and unsophisticated depositors who choose banksmainly based on geographical proximity rather than the potential yields. In contrast, shadow banks attracta group of yield-oriented depositors such as wealthy individuals and corporate treasurers. These yield-oriented depositors are not primarily concerned with transaction convenience, but instead are very sensitiveto yields.Depending on their depositor clientele, deposit rates of different banks exhibit different sensitivitiesto changes of market interest rates. When the Federal Reserve increases interest rates, commercial banksare reluctant to increase their deposit rates. This is because their main clientele, the transaction-orienteddepositors, view cash as the main alternative for transactions. Because cash bears no interests, commercialbanks are able to keep paying low interest rates without losing many of their transaction-oriented depositors.As a consequence, commercial banks can earn higher spreads between lending rates and deposit rates inperiods of high interest rates. In contrast, shadow banks have to raise their deposit rates together withthe market interest rates, because otherwise their yield-oriented clientele will switch to other high-yieldingliquid assets such as short-term bonds. Because shadow bank deposit rates are more sensitive to marketinterest rates than commercial bank deposit rates, high interest rate policy usually widens the difference indeposit rates between shadow and commercial banks, inducing some marginal depositors of commercialbanks to switch over to shadow banks. This explains why shadow banks expand their deposit creation whenthe Federal Reserve tightens monetary policy.In order to assess the quantitative importance of the shadow bank channel of monetary policy, I estimatemy model using institution-level data on U.S. commercial banks and MMFs. The estimation shows that3In this paper, I use “MZM” (money zero maturity) as the measure of money supply in the economy. This measure is amodification of M2 after the usefulness of previous measures became comprised in the 1990s. This measure includes currency,traveler’s checks of non-bank issuers, demand deposits, other checkable deposits, saving deposits, retail and institutional MMFshares. Choosing a specific definition of money aggregate, however, is not important, because my question is about each componentof the money aggregates, rather than the sum.4See Berry (1994), Berry, Levinsohn, and Pakes (1995) and Nevo (2001).32.1. Introductioncommercial bank deposits provide significantly higher convenience than shadow bank deposits. The esti-mation also shows that depositors exhibit significant differences in their preference over convenience andyields. Different types of depositors self-select into different types of banks. The heterogeneity in depositorclientele quantitatively explains the different responses to monetary policy of the two banking sectors.I use my model to study the implications of shadow banking for monetary policy. I simulate a coun-terfactual economy without shadow banks using estimated parameters. Comparing the real data with thecounterfactual economy, I find that the presence of shadow banks reduces the sensitivity of the aggregatemoney supply to the Fed Funds rates by 40 percent. I further use my model to study the implications ofshadow banking for financial stability. I show that shadow banks may be more prone to bank runs becausea small shock to asset values may trigger large redemption from their yield-oriented clientele. Using theruns on MMFs in 2008 as a case study, I find that MMFs with higher demand elasticity before the financialcrisis subsequently suffered more severe runs. Finally, my results suggest a cautious stance towards a recentpolicy proposal which suggests using monetary tightening as a tool for promoting financial stability (Stein,2012; Borio and Zhu, 2012; Ajello et al. 2016). I show that this policy proposal may unintentionally drivedeposits from the insured commercial banking sector into the uninsured shadow banking sector, and in doingso, heighten the risk of bank runs.This paper contributes to three strands of literature. The first strand studies the monetary transmissionmechanism in the banking system. Traditionally, this literature has focused on commercial banks (Bernankeand Blinder, 1988; Kashyap and Stein, 1995, 2000; Drechsler, Schnabl, and Savov, 2016). This paper bringsshadow banks into the forefront of the theoretical and empirical analysis of monetary policy. I documenta new transmission channel of monetary policy in the shadow banking system, which is contrary to theconventional wisdom in the commercial banking sector. This new channel partially offsets the traditionalchannels in commercial banks and dampens the impact of monetary policy.I further provide quantitative estimates of monetary transmission mechanisms for both commercial andshadow banks. Previous literature has made reduced form arguments about how monetary policy is transmit-ted in the commercial banking system. Bernanke and Blinder (1988) and Kashyap and Stein (1995, 2000)argue that monetary policy works through the opportunity cost of bank reserves, while more recent worksuch as Scharfstein and Sunderam (2016) and Drechsler, Schnabl, and Savov (2016) acknowledges that themarket power of the banking sector may also play a role in the transmission of monetary policy. Sincethe aforementioned literature relies on reduced-form models, their approach cannot quantify the impact ofeach channel. The present study complements the previous literature by providing a structural IO modelto quantify these different channels. My estimates suggest that the market power channel has been playinga dominant role in commercial banks since the 1990’s. This structural framework is also used by Egan,Hortaçsu, and Matvos (2015) to study bank run risks in the commercial banking sector. In contrast to theiranalyses, my paper introduces shadow banks and heterogeneity in depositors’ preferences with the aim ofdemonstrating the heterogeneous transmission of monetary policy in the modern banking system.The second strand of literature to which this paper contributes concerns the interaction between mone-tary policy and macro-prudential policies. Prior to the 2008-09 financial crisis, the consensus among policymakers was that monetary authority should focus on price stability and employment (Smets, 2013). How-ever, this consensus has been challenged by an alternative view that took shape after the financial crisis,arguing that monetary policy should also be used to promote financial stability (Stein, 2012; Borio and Zhu,2012; Ajello et al. 2016). Proponents of this view contend that by tightening monetary policy, the centralbank can curb the creation of money-like liabilities within the banking system and reduce the appetite ofinvestors for risk. My findings contribute to this debate by showing that monetary tightening may lead tothe unintended consequence of driving deposits to the shadow banking system. Since shadow banks arenot protected by deposit insurance, such a policy may actually increase systemic risk. My paper supportsthe view that “monetary policy is too blunt a tool to address possible financial imbalances” as argued byBernanke (2011) and Yellen (2014).42.2. Deposit Creation by Shadow BanksThe third strand of literature studies shadow banking, and particularly, the sources of fragility in theshadow banking sector. Previous research finds that the lack of deposit insurance (Gorton and Metrick,2012), high leverage (Adrian and Shin, 2010; Moreira and Savov, 2016), and information opacity (Dang,Gorton, and Holmström, 2016) create fragility in the shadow banking system. My paper contributes tothis line of research by showing that the yield-sensitive clientele of shadow banks can also be a source offragility. My paper also adds to a third group of papers on MMFs (Kacperczyk and Schnabl, 2013; McCabeet al., 2013; Schmidt, Timmermann and Wermers, 2016; Parlatore, 2016). I show that monetary policyhas a strong impact on the deposit flows of MMFs by changing the competition environment of the depositmarket.The remainder of this paper is organized as follows. Section 2.2 presents several new stylized facts ondeposit creation of the shadow banking system. Section 2.3 presents a structural model of bank competitionto rationalize the empirical findings. Section 2.4 presents the estimation procedure and results. Section 2.5discusses policy implications and Section 2.6 concludes.2.2 Deposit Creation by Shadow BanksIn this section, I provide a brief description of the institutional background of the shadow banking system. Ithen present several stylized facts to motivate my study.Institutional BackgroundThe shadow banking system is a collection of financial intermediaries which conduct maturity, credit, andliquidity transformation outside the traditional commercial banking system.5 Examples of shadow banksinclude securitization vehicles, asset-backed commercial paper (ABCP) conduits, MMFs, investment banks,and mortgage companies. Like commercial banks, shadow banks transform long-term illiquid assets intoshort-term money-like claims. Since households and business have a preference for liquidity, issuing money-like claims allows shadow banks to lower their financing costs.Figure 2.2 provides a simplified representation of the U.S. banking system.6 The upper branch repre-sents the commercial banking sector, while the lower represents the shadow banking sector. Unlike com-mercial banks, which combine deposit creation and loan origination under one roof, the shadow bankingsystem separates the intermediation process into different entities. MMFs constitute the first stage of theshadow banking intermediation process. MMFs take deposits from households and businesses and thenpass the proceeds to other shadow banks such as securitization vehicles, mortgage conduits, broker dealers,and mortgage companies, which specialize in loan origination. In this process MMFs create money-likeliabilities, MMF shares, which resembles commercial bank deposits.MMF shares are widely (though not necessarily accurately) regarded as being safe as bank deposits, yetproviding a higher yield. Similar to commercial bank deposits, MMFs provide intraday liquidity, and someof them even allow depositors to write checks on their deposits. Unlike other shadow banking liabilities,which are generally held within the shadow banking system, MMF shares are directly held by householdsand businesses. Due to their similarity with commercial bank deposits, MMF shares are included in theofficial money supply statistics, while other types of shadow banking liabilities are generally not. Theamount of MMF shares also provides a good proxy of the quantity of funds flowing into the shadow bankingsector.5Former Federal Reserve Chair Ben Bernanke provided a definition of the shadow banking system in April 2012: "Shadowbanking, as usually defined, comprises a diverse set of institutions and markets that, collectively, carry out traditional bankingfunctions–but do so outside, or in ways only loosely linked to, the traditional system of regulated depository institutions.”6A more detailed description of shadow banking intermediation process can be found in Pozsar et al. (2010).52.2. Deposit Creation by Shadow BanksOn the asset side, MMFs hold various money market instruments. The asset holdings of MMFs can begrouped into three major categories. According to the iMoneyNet data, the majority 50 percent are investedin short-term debts of other shadow banks such as repurchase agreement (repos), asset backed commercialpapers (ABCPs), commercial papers (CPs) and floating rate notes (FRNs).7 20 percent are invested inTreasury and agency securities. Lastly, 18 percent of the shadow bank deposits go back to the commercialbanking sector in the form of large denomination commercial bank obligations.Over the past thirty years, the shadow banking sector has become increasingly important in the econ-omy. Based on the aggregate money supply statistics from the Federal Reserve, the share of shadow bankdeposits has increased from around 15 percent in the 1980s to around 40 percent in 2007, while the share ofcommercial bank deposits is on a downward trend.Data SourceThe first main database used in this paper is iMoneyNet. This data provides monthly share class level datafor the U.S. MMFs dating back to 1985. After cross check with the aggregate money supply statistics fromthe Federal Reserve Board, I find that this database covers essentially all the MMFs which are included inthe official statistics after 1987. The data contains rich information on fund characteristics such as depositamounts, charged expense ratio, yields, check-writing privilege, and fund sponsors. The data also providesinformation on fund operating cost such as incurred management fee, share service fee, 12b1 fee and otherfees. Portfolio holding information becomes available since 1998, which includes average portfolio maturity,and portfolio weight by asset class. As data on shadow banks are generally very scarce, this data set providesa rare opportunity to look into the inner working of the shadow banking system.The second main data is the Consolidated Report of Condition and Income, generally referred as theCall report. This data provides quarterly bank-level data for every U.S. insured commercial bank, includingdetailed accounting information such as deposit amounts, interest income, salary expense, and fixed assetexpenses. I complement the Call report with FDIC Summary of Deposits, which provides branch-levelinformation on deposit amounts in annual frequency since 1994. Following the literature, deposit rates areimputed from bank financial statements by dividing deposit interest expense over total amount of deposits(Dick, 2008; Hannan and Prager, 2004). In the following analysis, I focus on “liquid deposits” which aredefined as the sum of checking and savings deposits8.In addition to the above two main data sources, I also use the Survey of Consumer Finance (SCF) 2013to obtain depositor-level deposit holdings and demographic information. Lastly, aggregate time series of theamount of cash held by households and the Fed Funds rates are retrieved from Federal Reserve EconomicData (FRED). Aggregate time series of the amount of Treasury bills held by households is retrieved fromthe Financial Accounts of the United States.Shadow Bank ChannelIn what follows, I document a new transmission channel of monetary policy using aggregate money supplydata from 1987 to 2012 from the Federal Reserve. I break down the aggregate money supply into cash,commercial bank deposits, and shadow bank deposits. Commercial bank deposits include demand andsaving deposits. Shadow banking deposits include retail and institutional MMF shares. Figure 2.1 plotsthe Fed Funds rates and the annual deposit growth rates of each banking sector over time. Conventional7Some of large industrial corporations also issue commercial papers to obtain short term financing. These commercial papersare mainly used to finance their captive finance companies, which are also considered as shadow banks. For example, one of thelargest issuers of commercial paper, General Motors Acceptance Corporation (GMAC), is a captive finance company that providesfinancing for the customers of its parent company, General Motors.8Previous literature has shown that the pricing and quantities of “liquid deposits” are quite different from “illiquid deposits”such as small time saving deposits (Driscoll and Judson, 2002; Drechsler et al., 2009).62.3. A Structural Model of Bank Competitionmonetary transmission channels predict that high Fed Funds rates have tightening effects on the moneysupply (Bernanke and Blinder, 1988; Kashyap and Stein 1995, 2000; and Drechsler, Schnabl, and Savov2016). Indeed as shown in the top panel of Figure 2.1, high Fed Funds rates are associated with low growthrates of commercial bank deposits. However, the opposite happens in the shadow banking system. Asshown in the bottom panel of Figure 2.1, high Fed Funds rates are associated with high growth rates ofshadow bank deposits. This finding implies that monetary policy seems to have a different transmissionchannel in the shadow banking system. High interest rate policy, which is intended to reduce money supplyin the economy, surprisingly increases deposit creation by shadow banks.To assess the magnitude of the new transmission channel, I regress deposit growth rates of each bankingsector on the Fed Funds rates, controlling for a list of macroeconomic variables such as GDP growth rates,inflation, VIX, and TED spread:Deposit Growth Ratest = α+βFed Funds Ratest + γXt + εt (2.1)Table 2.2 presents the results. Consistent with the graphical observation, monetary policy has oppositeeffects on these two sectors: a 1 percent increase in the Fed Funds rates is associated with a 1.4 percentdecrease in the growth rates of commercial bank deposits, but a 3.9 percent increase in the growth rates ofshadow bank deposits. The estimates are both statistically and economically significant.Column 4 and 8 show the results for the total money supply. The coefficients of the Fed Funds rates areinsignificant different from zero. This result shows that deposit creation by shadow banks partially offsetsthe reduction of commercial bank deposits and attenuates the impact of monetary tightening on aggregatemoney supply. As shadow banks create more deposits, they obtain more loanable funds for lending. InSection 2.5.2, I further show that shadow bank lending also increases as the Fed tightens monetary policy.To summarize, the above results show that shadow banks may dampen the impact of monetary policy bycreating more money-like liabilities when the Fed wants to reduce money supply. Furthermore, this channelimplies that monetary policy not only affects the total amount of money supply, but also the relative shares ofmoney supply between the shadow and commercial banking sector. Since shadow banks do not have accessto government safety nets such as deposit insurance and discount window, such shifts in the composition ofmoney supply have important implications for financial stability.2.3 A Structural Model of Bank Competition2.3.1 IntuitionThe previous section documents that monetary policy has very different impacts on the amount of depositscreated by commercial and shadow banking sector. In this section, I develop a structural IO model to ratio-nalize the above empirical findings. There are two key ingredients of the model. First, banks are differenti-ated in the dimensions of convenience and yields. Commercial bank deposits offer a lot transaction servicessuch as branch networks, ATMs, and payment system. In contrast, shadow banks compete on yields becausethey do not have bank charters to provide those transaction services.9 In addition to heterogeneous banks,the second key ingredient is that depositors exhibit heterogeneous preference over convenience and yields.There are a group of “transaction-oriented” depositors who care a lot about transaction convenience, but arenot sensitive to yields. For example, “mom and pop” depositors choose banks mainly based on geographicalproximity rather than deposit rates paid by banks. There are also a group of “yield-oriented depositors”who are very sensitive to yields but are relative insensitive to convenience. For example, large corporations9In addition, the deposit insurance on commercial bank deposits also increases their convenience relative to shadow banks. Thedeposit insurance of the commercial bank deposits is less relevant for for very large depositors because the FDIC only insurescommercial bank deposits up to a certain amount.72.3. A Structural Model of Bank Competitionand wealthy individuals usually have large amounts of deposits. A small difference in yields can make abig difference in the dollar value of income. Moreover, these depositors are often more sophisticated than“mom and pop” depositors. Therefore, they are better-equipped to find the highest yielding options in themarket and monitor the risks associated with them.These two groups of depositors are likely to self-select into different types of banks. Commercial banksare likely to attract more transaction-oriented depositors because of the superior transaction services offeredby them, while shadow banks attract more yield-oriented depositors because of the high deposit rates. Con-sistent with this idea, using the Survey of Consumer Finances (SCF) 2013, I find that depositors who are richor more sophisticated (proxied by college education) are more likely to choose shadow banks. The result isreported in Table 2.3.Facing different clientele, deposit rates of different banks have different sensitivities to monetary pol-icy. When the Federal Reserve increases interest rates, commercial banks are reluctant to increase theirdeposit rates. This is because their main depositor clientele, transaction-oriented depositors, view cash asthe main alternative transaction medium. Since cash bears no interests, as long as commercial banks paysome interests, the transaction-oriented depositors will stay in commercial banks. This allows commercialbanks to keep deposit rates relatively low and earn higher spreads between the rising lending rates and thestagnant deposit rates. In contrast, shadow banks have to raise their deposit rates together with the marketinterest rates. Otherwise, their yield-oriented clientele will switch to other higher yielding liquid assets suchas short-term bonds. As a result, when the Fed raises interest rates, the difference in deposit rates betweenshadow and commercial banks widens, which induces some of the marginal depositors of commercial banksto switch to shadow banks. This could potentially explain why shadow banks expand their deposit creationwhen the Federal Reserve raises interest rates.This explanation seems to be consistent with the data. Figure 2.3 plots the average deposit rates ofcommercial banks and MMFs over time. I find that when the Fed raises interest rates, shadow banks passthrough more rate hikes to depositors than commercial banks do. The changes in relative deposit ratesare economically significant. For example, in the 2004 tightening cycle, the difference in deposit ratesincreased from less than 0.5 percent to nearly 3 percent. Since transaction convenience and safeness of bankdeposits are relatively stable over monetary cycles, such big changes in relative yields may significantlyaffect depositors’ choice between these two banking sectors.In what follows, I introduce the setting of the structural empirical model. The structural empiricalmodel allows me to uncover the nature of competition in the deposit market using data of deposit rates andquantities. I will be able to estimate the extent of product differentiation between the two types of banks, andsee how they relate to alternatives such as cash and bonds. I will also be able to estimate demand functionsfor each commercial bank and MMF, which shed light on how banks set deposit rates in response to changesof monetary policy. More importantly, the structural model can quantify the magnitude of proposed channel.This is crucial given that there are alternative explanations which give the same qualitative results. Lastly, thestructural approach allows counterfactual simulations which are useful for examining policy implications.2.3.2 Model SettingHaving shown the basic intuition that different clientele can lead to different responses to monetary policy, Inow proceed with offering a full structural model to quantify the magnitude of this channel. The model usesthe discrete choice framework of oligopoly competition developed by Berry, Levinsohn, and Pakes (1995).This framework, by defining consumer preference over product characteristics as oppose to actual products,can endogenously generate a demand system for a large set of differentiated products with only a smallnumber of preference parameters. In the context of this paper, the demand for commercial and shadowbank deposits will be derived endogenously as function of their product characteristics, instead of beingexogenously specified. The discrete choice framework has been successfully applied to many industries82.3. A Structural Model of Bank Competitionsuch as the automobile, cereal, and airline industries. It has been a workhorse model in the quantitative IOliterature over the past 20 years. Early applications of this framework to the commercial banking industryincludes Adams, Brevoors, and Kiser (2007) and Ho and Ishii (2010). My paper contributes to this line ofliterature by introducing competition from shadow banks, a sector which has become increasingly importantin the modern banking system. In addition, I use this framework to study the impact of monetary policy onthe banking system, an aspect which has not yet been explored in this literature.I first introduce the basic setup of the framework. It is useful to show how the structural model connectsunobservable quantities such as utility and marginal costs to observable quantities such as market shares anddeposit rates in the data. I then add two features that are important to the deposit market. Lastly, I estimatethe model with the data.2.3.3 DepositorsThere are I depositors. Each of them is endowed with one dollar. Depositors make a discrete choice amongoptions including Treasury bills, cash, commercial bank or shadow bank deposits. Each individual bank is adistinctive option as each provides a differentiated product. Each option is indexed by j, and the choice set is{0,1, ...,J}. Each depositor can choose the option which gives him/her the highest utility. The assumptionsthat each depositor has only one dollar and can choose only one option are not as restrictive as they mayappear. We can imagine that depositors make multiple discrete choices for each dollar that they have, and theprobability of choosing each of the options can be interpreted as portfolio weights. The utility for depositori to choose product j is given bymaxj∈{0,1,...,J}ui, j = αr j +β ′x j +ξ j + εi, j (2.2)x j is a vector of product characteristics of bank j. Examples of product characteristics of a commercialbank include branch density, number of employees, and the age of the bank. Examples of product char-acteristics of a MMF include a rating dummy indicating whether the MMF is rated by three major ratingagencies, a bank fund dummy indicating whether the MMF is affiliated with a bank holding company, anda check-writing dummy indicating whether the MMF allows depositors to write checks. r j is the depositrate. ξ j is an unobservable demand shock. εi, j is a mean-zero idiosyncratic shock to utility, which followsthe extreme value distribution with a probability density function f (ε) = exp{−exp(−ε)}. This distribu-tion assumption is standard in structural IO literature. It basically allows closed-form solution of the choiceprobabilities. Finally, α and β are sensitivities to deposit rates and product characteristics. β ′x j capturesthe convenience of holding deposits from bank j. I normalize the convenience of Treasury bills to 0. Noticethat the linear form of utility does not mean that depositors have to be risk neutral. The aversion to riskcan be incorporated as disutility to risk, similar to the mean-variance utility formation. For example, Egan,Hortaçsu, and Matvos (2015) use CDS spreads of banks as a proxy of risk. Since CDS spreads are notavailable for MMFs, in this study the riskiness of deposits depends on whether the deposits are insured bythe deposit insurance, which further depends on whether the deposits are issued by a commercial bank or ashadow bank.I define δ j as the mean utility of product j across all depositors.δ j = E [ui, j] = αr j +β ′x j +ξ j (2.3)Under the assumption that the idiosyncratic utility shock follows the extreme value distribution, theexpected probability that product j is the best choice is given by the following formula:s j = E[1{ui, j≥ui,l∀l}]=exp(δ j)∑Jl=1 exp(δl)(2.4)92.3. A Structural Model of Bank CompetitionNotice that the expected probability that product j is the best choice for a depositor is also the marketshare of the product. Intuitively, the higher mean utility that a product generates, the greater market sharethat it has. In the baseline model, the market share is a simple logit function of the mean utility. Therefore,this model is often referred as “the logit model of demand” in the literature. Later, I will introduce featuresthat are important to fit the deposit market.2.3.4 BanksA bank is represented by a vector of product characteristics. The differences in product characteristicsbetween commercial and shadow banks are mainly driven by regulatory constraint. For example, shadowbanks are not allowed to operate branches or to provide checking accounts because these activities requirebank charters. I assume that product characteristics are exogenous, which leads to exogenous demandfunction s j (r j). Facing the demand, the decision of bank j is to choose a deposit rate r j to maximize profitsmaxr j( f − r j− c j)s j (r j) (2.5)where f is the Fed Funds rates, r j is the deposit rate of bank j, c j is the cost of providing depository services.s j (r j) is the market share of bank j. I assume that marginal lending rates of all the banks equal to the FedFunds rates. This assumption is quite close to reality, since the inter-bank market is usually quite efficientto equalize the marginal lending rates of different banks.10Banks’ optimal pricing decision is given by the following FOC:FOC: f − r j =(∂ log(s j)∂ r j)−1+ c j (2.6)On the left hand side, the spread between the Fed Funds rates and deposit rates, commonly refereed asdeposit spread, represents the price that banks charge for their depository services. On the right hand side,the first term(∂ log(s j)∂ r j)−1is the markup that a bank can charge on its depository service over the cost ofproviding it. It is inversely related to the demand elasticity. If the demand is inelastic, then the bank cancharge a higher markup. In contrast, if the demand is elastic, then the markup is likely to be low.I specify the marginal cost as a linear function of cost shiftersc j = γ ′w j +ω j (2.7)where w j is a vector of observable supply shifters. Examples of supply shifters of a commercial bankinclude salary paid to employees and fixed asset expenses. Examples of supply shifters of a MMF includemanagement costs and other operating costs. γ is the sensitivity of marginal cost to these cost shifters. ω j isan idiosyncratic supply shock.2.3.5 EquilibriumThe pure-strategy Bertrand-Nash equilibrium is a set of deposit rates, r∗, chosen by banks, and a set ofproduct, j∗, chosen by depositors such that each bank maximizes its profits, each depositor maximizes theirutility, and the deposit market clears.The main difference between commercial and shadow banks arises from their demand functions. I donot make any a priori assumptions on how these demand functions should be different. Instead, the demand10For instance, if bank A has higher lending rates than bank B, bank A will borrow from bank B and keep lending until the twolending rates converge.102.3. A Structural Model of Bank Competitionfunctions and their differences will be determined endogenously in the equilibrium by the exogenous productcharacteristics and preference parameters which will be estimated from the data. In addition, the cost toproduce shadow and commercial bank deposits may also be different, but as the estimation will show later,the demand side difference is the key reason why banks set deposit rates differently. In summary, to fullycharacterize the equilibrium, I need to know a set of primitive parameters, α , β , γ , which governs howdepositors value different products and how much it costs to produce them.Ideally, if I observe mean utility, δ j, and marginal costs, c j, I can pin down these parameters by estimat-ing the following two equations.δ j = αr j +β ′x j +ξ j (2.8)c j = γ ′w j +ω j (2.9)The first equation is the “mean utility equation” which describes how deposit rates and product char-acteristics are valued by depositors, and the second is the “marginal cost equation” which describes howobservable cost shifters affect the marginal cost of providing depository services. The challenge here, how-ever, is that neither mean utility, δ j, nor marginal costs, c j, are observable. Here is how the structural modelcan help. From the optimal decisions of depositors, I can link unobservable utility to observable marketshare. Using equation 2.4, I can solve unobservable mean utility as a closed-form function of observablemarket sharesδ j︸︷︷︸Unobservable= log(s j)− log(s0)︸ ︷︷ ︸Observable= αr j +β ′x j +ξ j (2.10)From the optimal decisions of the bank (equation 2.6), I can solve unobservable marginal costs as thedifference between deposit spreads and markups. Markups can be further derived from the market shareequation 2.4 as a function of observable market shares and yield sensitivity, α , which can be estimated fromthe mean utility equation.c j︸︷︷︸Unobservable= f − r j−(∂ log(s j)∂ r j)−1︸ ︷︷ ︸Unobservable= f − r j− (α (1− s j))−1︸ ︷︷ ︸Observable= γ ′w j +ω j (2.11)2.3.6 Depositor Heterogeneity and Adjustment CostWhat I have shown above is a basic setup of the discrete choice framework. Now I will add two features thatare important to the deposit market. The first one is depositor heterogeneity. In the basic setup, depositorshave homogeneous tastes over yields and convenience. In reality, depositors may exhibit strong heterogene-ity, as evident in Table 2.3. As argued in Section 2.3.1, different clientele can lead to different exposure tomonetary policy for the banks. As a result, it is important to incorporate this empirical feature in the model.Second, there is considerable stickiness in the adjustment process of deposits. As shown in Figure 2.1and 2.3, when shadow banks offer higher deposit rates than commercial banks, deposits do not switch intoshadow banks immediately. Instead, deposits flow into shadow banks gradually. To capture this feature, Iintroduce a switching cost for depositors when they change their choices.Formally, the depositors’ problem is modeled as the following maximization problemmaxj∈{0,1,...,J}ui, j = (α+σvi)r j +β ′x j +ξ j−ρ1{k 6= j}+ εi, j (2.12)σvi captures the heterogeneous response to deposit rates, where vi follows a standard normal distributionand σ is the magnitude of dispersion. −ρ1{k 6= j} captures the adjust cost, where k is choice of last period andj is the choice of current period. This term means that if the current choice j is different from the previouschoice k, the depositor incurs a cost of ρ .112.4. Structural EstimationNotice that such formulation of adjustment cost assumes that depositors are myopic: they simply maketrade-off between the contemporaneous utility and adjustment costs instead of trying to forecast future pathof interest rates. Ideally, it would be more realistic to have forward-looking depositors. However, suchdynamic model with heterogeneous agents would be very difficult to estimate. Therefore, this paper takes amore reduced form approach to model adjustment cost.With the adjustment cost, the past choice k becomes a state variable for the current decision. DefinePk→ ji,t as the transition probability from product k to j for depositor i at time t. The transition probability isgiven by the following formula:Pk→ ji,t = E [ui, j ≥ ui,l∀l|k] =exp(δ j + r jσvi−ρ1{k 6= j})∑Jl=1 exp(δl + rlσvi−ρ1{k 6=l}) (2.13)Define si, j,t as the expected choice probability for depositor i to choose product j. The current expectedchoice probability is the sum of the products between transition probability, Pk→ ji,t , and the probability dis-tribution of last period, si,k,t−1.si, j,t =∑ksi,k,t−1Pk→ ji,t (2.14)The aggregate market share of product j is the obtained by summing over different depositor typess j,t =∑ipiisi, j,t (2.15)where pii is the frequency of type i depositors.Note that after introducing adjustment costs, the banks’ problem become dynamic.maxr j,tE∞∑t=0exp(−rt)( ft − r j,t − c j,t)s j,t (2.16)where r is the discount rate for the profits.With adjustment costs for depositors, the optimal deposit spread is determined by long-run elasticityinstead of short-run elasticity.ft − r j,t =(E[∞∑τ=0exp(−rτ) ∂ s j,t+τ∂ r j,t1s j,t])−1+ c j,t (2.17)11One complication introduced by these two new features is that I can no longer solve mean utility, δ j,as a closed form function of market shares. Instead, I need to solve it numerically from the market shareequation 2.15 using the fixed-point algorithm introduced by Berry, Levinsohn, and Pakes (1995).2.4 Structural EstimationIn this section, I take the model to the data. The goal here is to pin down the primitive structural parametersand quantify effects of depositor heterogeneity on banks’ response to monetary policy. This will set thestage for the counterfactual analysis that ensues.11The above deposit rates are solved in a stationary equilibrium where the state variables remain constant over time so thatft+τ − r j,t+τ − c j,t+τ = ft+τ − r j− c j .122.4. Structural Estimation2.4.1 IdentificationThe set of primitive parameters are α,β ,σ ,ρ,γ . Some of the parameters enter linearly (α,β ,γ) in the twostructural equations below, and the rest enter non-linearly (σ ,ρ) (the time subscript is suppressed here forsimplicity).δ j (σ ,ρ) = αr j +β ′x j +ξ j (2.18)c j = γ ′w j +ω j (2.19)Two alternative procedures can be implemented to estimate the above structural equations. I can estimatethem simultaneously using a joint-equation GMM, or I can estimate them sequentially. I choose the sequen-tial approach, since in a joint estimation the misspecification of the marginal cost equation may contaminatethe estimation of preference parameters. More concretely, I first estimate the mean utility equation (2.18).Given the estimated demand side parameters, I calculate the marginal costs implied by equation (2.17). ThenI estimate the marginal cost equation (2.19).Since σ and ρ enter the mean utility equation non-linearly, the estimation of the mean utility equationrequires more discussion. I use the Nested Fixed Point (NFP) algorithm as detailed in Nevo (2000). Thealgorithm first searches over the non-linear parameter space of σ and ρ . Second, for a combination ofσ and ρ , it solves δ j (σ ,ρ) through fixed-point algorithm using equation (2.15). Third, I find a set of linearparameters α,β which minimize the moment condition which I will describe later. The above three stepsare repeated until the optimal set of parameters α,β ,σ ,ρ is found.A key challenge in identifying the demand parameters is that deposit rates are correlated with unobserv-able demand shocks ξ j. As a result, yield sensitivity α will be biased in an OLS regression of mean utility,δ j, on deposit rates, r j. The solution is to find shocks that affect deposit rates, but are exogenous to unob-servable demand shocks, ξ j. I follow the literature to use a set of cost shocks, z j, as instrument variables.Examples of instrument variables include salary, rent, and other operating costs. The intuition is that theseshocks shift the supply curve so that the demand curve can be traced out.The moment condition of the mean utility equation is given by the orthogonality condition between theunobservable demand shocks, ξ j, and the product characteristics, x j, and cost shifters, z j:E [ξ j (x j,z j)] = 0 (2.20)The moment condition of the cost equation is given by the orthogonality condition between the idiosyn-cratic supply shock, ω j, and observable cost shifters, w j:E [ω jw j] = 0 (2.21)2.4.2 Data for Structural EstimationThe data used for the structural estimation are a panel of commercial banks and MMFs from 1994 to 2012.Following the literature, a market is defined as a MSA-year combination. Since commercial banks attractdeposits mainly through local branches, the choice set of depositors of a MSA includes commercial bankswhich have local branches in the MSA. In contrast, MMFs generally compete in a national market throughdistribution channels of brokerage firms and over the internet. Therefore, local depositors can access to allthe MMFs in the market. In addition, depositors can also choose cash or Treasury bonds.I calculate market shares of a commercial bank by summing up deposits of local branches of the bankin the MSA. For MMFs, cash, and Treasury bonds, no MSA-level information on quantities is available. Iimpute MSA-level deposit amounts assuming that deposit amounts are proportional to local personal income132.4. Structural Estimationlevel.12 The total market size is the sum of cash, commercial bank deposits, MMF shares and Treasury bondsin a MSA.13 Following the literature, I combine tiny banks and MMFs (market share less than 0.2 percent)with Treasury bonds as the outside option.Product characteristics are chosen based on the belief that they are important and recognizable to de-positors’ choice. Product characteristics of commercial banks include deposit rates, branch density in thelocal market, average number of employees in a branch, bank age, and single-market dummy (whether abank operates in a single market or multiple markets).14 Product characteristics of MMFs include depositrates, rating dummy (whether a fund is rated by three major rating agencies), bank fund dummy (whethera fund is affiliated with a commercial bank), check-writing dummy (whether a fund allows depositors towrite a check), and fund age. I include product fixed effects to absorb unobservable time-invariant productcharacteristics. Notice that bank fixed effects also absorb observable time-invariant product characteristics.To retrieve the taste coefficients on these product characteristics, I follow the minimum-distance procedureproposed by Chamberlain (1982) to estimate coefficients of time-invariant product characteristics. Lastly, Iinclude time fixed effects to absorb aggregate demand shocks, and MSA fixed effects to absorb cross-marketdifferences in demand.The marginal cost equation includes product characteristics and cost shifters. The set of product char-acteristics is the same as the demand function. The cost shifters of MMFs include management costs andother operating costs. The cost shifters of commercial banks include salary expenses and expenses of fixedassets.15 Lastly, I include bank fixed effects to absorb time-invariant bank-specific cost shocks, time fixed ef-fects to absorb aggregate shocks to marginal costs, and MSA fixed effects to absorb cross-market differencesin the cost of providing depository services.As discussed in 2.4.1, I need a set of instruments to identify the yield sensitivity, α . Following previousliterature, I use the cost shifters and their second-order polynomials as instruments for the mean utilityfunction. I use Chamberlain’s (1987) optimal instruments in the second stage of estimation to increase theestimator’s efficiency and stability (Reynaert and Verboven, 2014). The optimal instruments are definedas the conditional expectation of the derivatives of the residuals with respect to the parameter vector. Thedetails of constructing the optimal instruments can be found in Reynaert and Verboven (2014).Table 2.1 provides summary statistics of the sample used for the structural estimation. A commercialbank typically has larger market shares than a MMF: the average market share is 3.5 percent for a commer-cial bank and is 0.44 percent for a MMF. A commercial bank also tends to offer lower deposit rates thanshadow banks: the average deposit rates are 1.79 percent for commercial banks and 3.04 percent for MMFs.A commercial bank on average has 22.5 branches per million population in a MSA, and each branch has18.45 employees. 52 percent of MMFs are rated by a least one of three rating agencies, 45 percent areaffiliated with bank holding companies, 34 percent allow depositors to write checks.12More specifically, I first compute the percentage of a MSA personal income in the national total personal income. Then Icalculate the historical average of this percentage over the sample period to calculate the weight of the MSA. Lastly, I impute theMSA-level deposits according to the weight. Alternatively, I estimate the model using national-level data. This alternative approachgenerates similar results.13Treasury bills are more appropriate for the model setting. However, the information of the aggregate Treasury bills outstandingis not always available in the sample period.14For commercial banks which operate in multiple markets, I only have bank-level rather than branch-level information ondeposit rates, so there is no cross-market variation for these multi-market banks. Nevertheless, this may not be a major issue sinceprevious empirical studies have shown multi-market banks usually use uniform pricing across local markets within a state (Radecki,1998).15This set of cost shifters of commercial banks are also used in previous literature such as Dick (2008) and Ho and Ishii (2011).142.4. Structural Estimation2.4.3 Parameter EstimatesThis section presents the results of the structural estimation. First, it discusses the demand parameters.Column 1 of Table 2.4 reports the estimates of demand parameters of the full model. The estimated yieldsensitivity are positive and significant. Most importantly, there is statistically significant dispersion in de-positors’ sensitivity to yields. Later, I will explore the economic implications of such dispersion. Depositorsalso face a significant adjustment cost: the pecuniary cost of changing choices for an average depositoramounts to 8.81 percent. Depositors prefer banks with higher branch density and more employees perbranch. Depositors also prefer single-market banks and younger banks. For MMFs, depositors prefer oldfunds and funds sponsored by bank holding companies. Interestingly, funds with check-writing privilegeand credit rating do not gain favor among depositors on average.Column 2 of Table 2.4 shows a simple logit model in which depositors are homogeneous and face noadjustment costs. One can see that depositor heterogeneity and adjustment costs are important to get thecorrect sign for the yield sensitivity: without these two features, the model implies that depositors counter-intuitively derive negative utility from deposit rates.Regarding the supply side parameters, column 1 of Table 2.5 presents the estimated cost coefficients ofthe full model. The cost coefficients for MMFs are precisely estimated: a 1 percent increase in managementcosts is associated with 1.069 percent increase in marginal costs, and a 1 percent increase in other operatingcost is associated with 1.151 percent increase in marginal costs. These magnitude is very close to thetheoretical value, which is 1. For commercial banks, greater branch density and more employees per branchare associated with lower marginal costs, implying that there is an increasing return to scale. Higher salaryexpense is associated with higher marginal costs, which is intuitive. The estimated coefficient of expenseof fixed assets, however, has counter-intuitive sign. This could be due to the fact that expense of fixedassets is quite sticky over time. It is difficult to identify its impacts when bank fixed effects are includedin the regression. The reserve cost is the product between the Fed Funds rates and the amount of reservesnormalized by total deposits. This variable captures the effect of monetary policy on the cost of providingdeposits for commercial banks. Theoretically, a 1 percent increase in the reserve cost should lead to a 1percent increase in the marginal cost. The estimated coefficient of the reserve cost is 1.765, which is closeto the theoretical value.16Although the positive coefficient of reserve cost shows that monetary tighteningindeed increases the marginal costs of commercial banks relative to shadow banks, the magnitude of thischannel is likely to be small. The summary statistics in Table 2.1 show that commercial banks on averagehold 1.1 cents of reserve per dollar of deposits. This tiny amount of reserve will have limited quantitativeimpacts on marginal costs of commercial banks. Lastly, the interaction between the Fed Funds rates andcash dummy has a coefficient of 1, implying that the cost of holding cash increases one for one with the FedFunds rates.Column 2 of Table 2.5 presents the estimated cost coefficients of the logit model. In this regression,the cost function has exactly the same specification as column 1. The only difference is that the depen-dent variable, marginal cost, is calculated from a logit model of demand. Comparing with the full model,the logit model implies counter-intuitive signs and magnitudes of these cost shifters. For example, highermanagement costs are associated with lower marginal costs. These results imply that the logit model doesnot generate as good estimates of the markups as the full model. This again confirms the importance toincorporate depositor heterogeneity and adjustment costs in the model.2.4.4 Model FitFigure 2.4 compares model predicted market shares with the data. The full model successfully generatesthe counter-cyclical deposit growth rates for commercial banks, and pro-cyclical deposit growth rates for16The reason why it is greater than 1 could be that it picks up other transmission channels not specified by the supply equation.152.4. Structural Estimationshadow banks. The magnitude matches the data closely.What drives the different responses to monetary policy for commercial and shadow banks? In the fol-lowing subsection, I will use the structural model to understand the underlying mechanism. The proposedmechanism consists of the following four steps:1. Banks offer differential products. Shadow banks provide lower convenience than commercial banks.2. Different products attract different depositor clientele. Shadow banks attract more yield-orienteddepositors than commercial banks.3. Facing different depositor clientele, banks have different interest rate pass-through. Shadow bankdeposit rates are more sensitive to market interest rates than commercial bank deposit rates.4. Different types of depositors respond to changes in relative prices differently. Yield-oriented deposi-tors move in and out shadow banks, while transaction-oriented depositors stay with commercial banks.In the following analysis, I examine the above four steps one by one.2.4.5 Differential ProductsThe underlying assumption of the proposed channel is that commercial bank deposits offer greater con-venience than MMFs. In this subsection, I verify this assumption. I construct a composite measure ofconvenience by x′jβˆ . x j is the vector of bank characteristics which is related to transaction services. Ex-ample of these characteristics includes the density of branches, number of staffs in a branch, check-writingprivileges and so on. I also include bank fixed effects to absorb residual time-invariant difference in transac-tion services. βˆ is the vector of estimated sensitivity to these characteristics. This composite measure showshow much the average depositor values each product if deposit rates are zero.In Figure 2.5, I plot the distribution of the estimated convenience. Each observation is a MSA medianfor each sector. Consistently with the assumption that commercial bank deposits are safer and offer moretransaction services, commercial banks have higher estimated convenience than MMFs. In addition, cashhas the highest convenience, which is also intuitive. The figure also shows the relative market position ofcommercial and shadow bank deposits in relation to cash. Commercial bank deposits are a closer substituteto cash than shadow bank deposits. This could potentially allow commercial banks to attract a group oftransaction-oriented depositors, a conjecture to be examined in the next section.2.4.6 Depositor ClienteleWith commercial and shadow banks offering differentiated products, I expect different types of depositorsto self-select into different types of banks. The estimates show that this is indeed the case. Figure 2.6plots average demand elasticities of commercial banks and shadow banks. The detailed summary statisticsof demand elasticities can be found in Table 2.6. The median own-rate elasticity of commercial banks is1.0668, which has the same magnitude as previous literature. The median own-rate elasticity of MMFs is1.8112, which is almost twice as large as that of commercial banks. This is consistent with the idea that theclientele of MMFs is more yield-sensitive than commercial banks.Next, I examine the cross-rate demand elasticity. The cross-rate elasticity measures the percent changeof market shares due to changes in deposit rates of a competitor. Table 2.7 presents the median and standarddeviation of cross-rate elasticity. The entry of the i-th row and j-th column shows the percent change of themarket share of a product in category i (cash, CB, MMF) with one percent change of the deposit rates of arival product in category j (cash, CB, MMF). Comparing across columns, price changes of a product withhigh convenience seem to have greater effects on other products. Comparing across rows, MMFs in generalhave the highest cross-rate elasticities, which is also likely a result of their yield-sensitive clientele.162.4. Structural Estimation2.4.7 Banks’ Responses to Monetary PolicyI have shown above that shadow and commercial banks face different clientele. Now the question is whetherthe difference in clientele can lead to different responses to monetary policy. Figure 2.7 plots the averagedifference in deposit rates between the two banking sectors in the data and predicted by the model. Themodel generates similar pro-cyclical patterns as in the data: as the Fed Funds rates go up, deposit rates ofshadow banks become higher than that of commercial banks.Two possible mechanisms that may explain the above pattern for deposit rates. The first possibility isthat monetary policy has differential impacts on the demand side of the banks. As the Fed Funds ratesincrease, commercial banks are able to charge higher markups on the transaction-oriented depositors, whilethe markups charged by shadow banks remain stable because their depositors are more yield-sensitive. Thesecond possibility is that monetary policy has differential impacts on the costs of providing depositoryservices. To elaborate, commercial banks are required to hold reserves while MMFs do not. High FedFunds rates drive up the cost of holding reserves. Commercial banks do not increase their deposit ratesbecause they have higher costs to cover. To separate the two channels, I decompose deposit spreads intotwo components: marginal costs and markups. The top panel of Figure 2.8 shows the average differencein markups and marginal costs between commercial and shadow banking sectors over time. It is clear thatthe markup term fully drives variations in deposit spreads. In contrast, the difference in marginal costs isalmost flat over monetary cycles. This figure shows that the differential pricing pattern is mainly driven bythe demand side, rather than the supply side. It is hardly surprising, given the summary statistics in Table2.1 showing that commercial banks on average only hold 1.1 cents of reserves for every dollar of deposits.A 1 percent increase in the Fed Funds rates only leads to a 0.011 percent increase in marginal costs.What is the key demand parameter that generates the different cyclical patterns of markups? In thebottom panel of Figure 2.8, I re-estimate the markups and marginal costs assuming that the dispersion inyield sensitivity, σ , is zero. In this case, the markup term becomes flat as well. This result shows thatdepositor heterogeneity is the key feature to explain the different pricing patterns in the data. The intuitionis that when σ goes to zero, depositors become essentially homogeneous. This means that commercial andshadow banks will have the same clientele and monetary policy will have similar impacts on their demandelasticity.The decomposition of deposit spreads into markups and marginal costs also sheds lights on the monetarytransmission mechanisms in the commercial banking sector. Traditionally, the banking system is modeledas a perfectly competitive industry. There is no role for market power because markups are always zero.It is until recently that several papers such as Scharfstein and Sunderam (2016) and Drechsler, Schnabl,and Savov (2016) start to point out that market power of the banking sector may play a role in transmittingmonetary policy. Since the aforementioned papers rely on reduced-form models, they cannot quantify theimportance of the market power channel relative to the traditional bank reserve channel. This paper comple-ments the above studies by providing a structural model to quantify the impact of market power and reservecost. As it is clear from the top panel of Figure 2.8, the markups are main sources of variations in the de-posit spreads of commercial banks. This evidence suggests that the market power channel has been playinga dominant role in the transmission of monetary policy since the 1990’s. In contrast, the reserve channelhas become less relevant due to technological and regulatory changes (Teles and Zhu, 2005). In additionto quantifying the magnitude of market power channel, my model also provides a deeper understanding onthe determinants of market power. It shows that market positioning affects the pricing power of banks. Thishelps us understanding why monetary policy may have heterogeneous impacts on different types of banks.172.4. Structural Estimation2.4.8 Choice of DepositorsLastly, I examine the choices of different types of depositors over monetary cycles. I classify depositorswith above-median yield sensitivity as yield-oriented depositors, and depositors with below-median yieldsensitivity as transaction-oriented depositors. Figure 2.9 plots their probability to choose a commercial bankor a MMF over time.The first observation is that yield-sensitive depositors are on average more likely to choose MMFs, whiletransaction-oriented depositors are more likely to choose commercial banks. The second observation is thatthe choice probability of yield-oriented depositors varies significantly over monetary cycles: they are morelikely to choose commercial banks when the Fed Funds rates are low, and switch to MMFs when the FedFunds rates go up. In contrast, transaction-oriented depositors are more likely to choose commercial banksall the time. This is consistent with the intuition that yield-oriented depositors are constantly looking forhigher yields, while transaction-oriented depositors stay in commercial banks because of the transactionconvenience and safety of commercial bank deposits.2.4.9 Alternative ExplanationsThus far, I have shown that the model produces coherent evidence that depositor heterogeneity can quantita-tively explain the different responses to monetary policy by commercial and shadow banks. One may arguethat there are many other institutional differences across banking sectors could also possibly explain theirdifferent responses to monetary policy. One intuitive candidate is the reserve requirement. When commer-cial banks take deposits, they are required to keep a fraction of the deposits as reserves instead of lendingthem out. Historically, bank reserves do not bear interests.17 Therefore, holding reserves imposes a cost forcommercial banks, and the cost of holding reserves is increasing in the Fed Funds rates. In contrast, shadowbanks are not subject to reserve requirements. As a result, monetary policy may have differential impactsacross banking sectors through the cost of providing depository services. The bank reserve channel featuresunderlying mechanism of several papers such as Kashyap and Stein (1995), Stein (2012), Sunderam (2015)and Nagel (2016).The reserve based explanation does not seem to quantitatively explain the magnitude of pricing dif-ference documented in this paper. To do a back-of-envelope calculation, I assume that 10 percent reserverequirement applies to all commercial bank deposits18. In the 2004 tightening cycle, the Fed Funds rates in-crease by 4.25 percent, which leads to an increase in the marginal cost by 0.425 percent through the reservechannel. This number is still far from explaining the 2.5 percent increase in the data. The structural modelprovides more concrete evidence. In the top panel of Figure 2.8, we hardly see any differential impacts ofmonetary policy on the marginal costs of commercial and shadow banks, despite of their different reserverequirement. This result is not surprising given extensive research has suggested that reserve requirementhas become less relevant for the current banking system due to technological innovations and regulatoryreforms.19The second potential explanation is based on asset-side differences between commercial banks andMMFs. The asset duration of MMFs are much shorter than commercial banks for both economic and17After October 2008, the Fed started to pay interest on reserves.18In practice, saving deposits face much less reserve requirement (1 percent), which further reduce the magnitude of this channel.19One example of technological innovations is the sweep technology, which allows banks to easily transfer funds from transactionaccounts to saving accounts to avoid the reserve requirement (Teles and Zhu, 2005). As a result, the amount of bank reserve inthe economy has become very small before the recent unconventional monetary policy: as of December 31, 2007, the aggregatereserve balance is only 48 billion, which accounts for less than 0.4 percent of 6,720 billion commercial bank deposits. It is hard toimagine such a small opportunity cost could quantitatively explain the substantial deposit spreads observed in the data. After thestart of unconventional monetary policy in 2008, the reserve balance grew dramatically. However, in this period, the Fed started topay interest on reserves, which essentially eliminated this reserve channel.182.5. Policy Implicationsregulatory reasons. Economically, the shadow banking system breaks down the intermediation process inseveral steps. MMFs only conduct a small amount of maturity transformation: the average maturity ofMMF assets is around 40 days based on the iMoneyNet data, while commercial banks usually have muchlonger asset maturity. In terms of regulation, Rule 2a-7 of the Investment Company Act of 1940 restrictsthe highest maturity of any debt held by MMFs to be under 13 months, and the portfolio must maintaina weighted average maturity (WAM) of 60 days or less. Therefore, a change in interest rates may lead todifferent impacts on the value of the assets due to different asset duration. However, this channel is onlyrelevant for the periods shortly after interest rate changes. It cannot explain the persistent differences indeposit rates between MMFs and commercial banks long after the change of the Fed Funds rates.The third potential explanation relies on the risk of holding shadow bank deposits. The risk of holdingshadow bank deposits is likely to be lower when the economy is doing well. This is usually the time whenthe Fed adopts high interest rate policy to cool down the economy. Therefore, the inflows to the shadowbanking sector may be driven by lower risks of shadow banks, rather than by monetary policy. Although thisexplanation can potentially explain the inflows to the shadow banking sector, it cannot explain why shadowbanks usually pay higher deposit rates in periods of high interest rates. If anything, this explanation wouldpredict that shadow banks should pay lower deposit rates because their risks are lower during these periods.2.5 Policy ImplicationsUsing the structural model, I conduct a set of counterfactual exercises to study several questions relevantto shadow banking. How does shadow banking change the effectiveness of monetary policy? What are theimplications of shadow banking for financial stability? How does shadow banking affect depositor surplus?2.5.1 Shadow Banks and Effectiveness of Monetary PolicyThere is a long-lasting concern that financial innovation may undermine monetary control of the centralbank. Such concern has intensified in recent years as the shadow banking sector has grown outside thetraditional commercial banking sector. Has the rise of the shadow banking system affected the effectivenessof monetary policy? To answer this question, I calculate the aggregate money supply in the counterfactualeconomy without MMFs. Figure 2.10 shows the growth rates of aggregate money supply in the counterfac-tual economy. In a simple regression of the aggregate money growth rates on the Fed Funds rates controllingfor a time trend, the coefficient of the Fed Funds rates decreases by around 40 percent in the actual economycomparing to the counterfactual one. This means that the presence of shadow banks reduces the responsive-ness of aggregate money supply to monetary policy.The counterfactual analysis offers insights on the monetary transmission mechanisms in an economywith both commercial banks and shadow banks. In an economy without shadow banks, commercial banksare more reluctant to pass through the rate increase to depositors. Depositors flow out of the banking system,leading to a reduction in money supply and credit supply.Shadow banks provide buffers for depositors. When depositors are unsatisfied with the low rates paidby commercial banks in periods of monetary tightening, they do not have to switch from deposits to bonds.They can switch within the banking system from commercial banks to MMFs. Having more loanable funds,MMFs pass the proceeds to other shadow banks which specialize in lending. An increase in shadow banklending can substitute the decline in commercial bank lending, which dampens the impact of monetarytightening.192.5. Policy Implications2.5.2 Monetary Policy and Credit Supply of Shadow BanksSo far, my empirical analysis has been focusing on the money supply. This section examines the creditsupply of shadow banks. As discussed in Section 2.2, while shadow bank deposit creation is conducted byMMFs, loan origination is conducted by different shadow banking entities such as funding corporations,finance companies, ABCP issuers, captive financial institution and broker-dealers.20 These loan-originationshadow banks do not issue deposits directly. Instead, they obtain funding from MMFs through issuing moneymarket instruments. There are four major categories of money market instruments issued by these loan-origination shadow banks: commercial papers (CPs), asset-backed commercial papers (ABCPs), repurchaseagreements (repos), and floating rates notes (FRNs). I regress annual changes of MMF lending throughthese four money market instruments on the Fed Funds rates, controlling for macroeconomic variables, fundcharacteristics and fund fixed effects:∆MMF Lendingi,t = α+βFed Funds Ratest + γXi,t + εi,t (2.22)Column 1 to 4 of Table 2.8 show that MMFs significantly increase their lending to the loan-originationshadow banks as the Fed Fund rates increase. The economic magnitude is significant, too: a 1 percentincrease in the Fed Fund rates is associated with a 0.17-0.45 percent increase in lending from MMFs toother shadow banks.In addition to the four types of money market instruments discussed above, MMFs also hold largedenomination commercial bank obligations (CBs), which are issued by commercial banks to obtain short-term funding. Column 6 of Table 2.8 shows that MMFs also increase the holding of large denominationbank obligations when the Fed raise interest rates. This result reveals an interesting interaction between theshadow and commercial banking system. As the Fed tightens monetary policy, commercial banks borrowmore from MMFs to compensate their loss of the core deposits. Such arrangement is profitable for both typesof banks: it effectively conducts price discrimination on transaction-oriented depositors. However, it has adownside: through this lending relationship, bank runs on the MMF industry may spread to commercialbanks. This result comprises another unintended consequence of monetary tightening on financial stability.With an increase in funding supply from MMFs, the loan-origination shadow banks should be able toexpand their credit supply. I examine five types of shadow banks which rely on MMFs to obtain financing:funding corporations, finance companies, ABCP issuers, captive financial institution and broker-dealers. Iregress aggregate asset growth rates of these five types of shadow banks on the Fed Funds rates and variousmacroeconomic controls:Shadow Bank Asset Growtht = α+βFed Funds Ratest + γXt + εt (2.23)Table 2.9 presents the results. When the Fed Funds rates are high, the assets of these shadow banksalso grow faster. The composition shift in the aggregate credit supply may also increase the systemic risk,20Finance companies are financial entities that sell commercial paper and use the proceeds to extend credit to borrowers whichusually tend to be riskier than that of commercial banks (Carey, Post, and Sharpe, 1998). In the mortgage market, these shadowlenders such as Quicken Loans, PHH and loanDepot.com accounted for 53 per cent of government-backed mortgages originatedin April, 2015. Funding corporations are subsidiaries of foreign banks and non-bank financial firms that raise funds from thecommercial paper market and pass the proceeds to foreign parent companies abroad or to foreign banking offices in the U.S..ABCP issuers are structured investment vehicles which purchase and hold financial assets from a variety of asset sellers andfinance their portfolio by selling asset-backed commercial paper to MMFs or other “safe asset” investors like retirement funds.Captive finance company is a subsidiary whose purpose is to provide financing to customers buying the parent company’s productthrough issuing commercial papers. Examples include the captive finance of the Big Three car manufacturers: General MotorsAcceptance Corporation (GMAC), Chrysler Financial and Ford Motor Credit Company. Broker-dealers include both non-bankfirms and subsidiaries of commercial banks that engages in the business of trading securities for its own account or on behalf of itscustomers. Broker-dealers heavily rely on repo to obtain funds from MMFs and then lend to their customers through reverse repo.A prominent example of broker-dealers is Lehman Brothers which went bankrupt during the 2008-09 financial crisis.202.5. Policy Implicationsbecause shadow banks usually lend to the riskier segment of borrower (Carey, Post, and Sharpe, 1998). Thepositive relation between shadow bank asset growth rates and the Fed Funds rates is also documented bya contemporaneous paper by Nelson, Pinter, and Theodoridis (2015). The main difference between theirwork and the present study is that they attribute the expansion of shadow bank assets to negative shocks ofhigh interest rate policy on equity values of commercial banks, while my paper argues that the expansion ofshadow bank assets is driven by the increase in shadow bank deposit creation. Empirically, their assumptionthat high interest rate policy reduces equity values of commercial banks seems to be inconsistent with thedata, as stock prices of commercial banks usually increase when the Fed raises interest rates. However,the increase in stock prices of commercial banks during periods of high interest rates is consistent with mymodel, which shows that high interest rate policy widens the spread between lending rates and deposit rates,which boosts the profitability of banks.2.5.3 Depositor Clientele and Risks of Bank RunThe shadow banking system played a central role in the 2008-09 financial crisis. Why is the shadow bankingsystem so fragile? Previous literature has pointed out factors such as the lack of deposit insurance (Gortonand Metrick, 2012), high leverage (Adrian and Shin, 2010; Moreira and Savov, 2016), and informationopacity (Dang, Gorton, and Holmström, 2016). In this section, I explore whether the yield-sensitive clientelecould be a source of fragility for the shadow banking system.The runs on MMFs in September 2008 provide a unique laboratory to study this question. On September16th, 2008, the Reserve Primary Fund, one of the oldest MMF, broke the buck when its net asset valuefell below $1. This event triggered widespread runs on the whole prime MMF industry. By October 7th, 2008, deposits of prime MMFs fell by $498 billion (24 percent).21 Interestingly, there was significantheterogeneity in the severity of runs across funds: some funds suffered enormous amount of withdrawal,while others were less affected.I examine whether the severity of the runs is related to the clientele of the funds. I use demand elasticityof the largest 30 prime funds to predict the severity of the runs in the following month. The demand elasticityis estimated using my structural model as of September 9th, 2008, one week before the runs. Figure 2.11shows the result. Funds with higher demand elasticity experienced greater redemption subsequently. Whydid funds with higher demand elasticity suffer more severe runs? A fund with higher demand elasticity mayface greater redemption for a given loss. Because of the lack of deposit insurance, the initial redemption mayforce the fund to liquidate its assets in fire sale, imposing negative externality on the remaining depositorsand causing further redemption.22 This means that the yield-sensitive clientele served by shadow bankscould be a source of fragility in the shadow banking sector.The above results also contribute to a recent debate on the merit of using monetary policy as a macro-prudential policy tool. After the 2008-2009 financial crisis, many argue that monetary tightening should beused to promote financial stability, because high interest rates can slow down deposit creation in the bankingsystem and reduce risk appetite of investors (Stein, 2012; Borio and Zhu, 2012; Williams, 2014; Smets,2016). My findings suggest a cautious stance towards this policy proposal. I show that while monetarytightening can reduce deposit creation in the commercial banking sector, it may unintentionally drive moreyield-sensitive depositors to the shadow banking sector as shown in Figure 2.9. Since shadow bank depositsare not insured by the government, a higher concentration of yield-sensitive depositors may amply the risk21MMFs are usually categorized into three types based on investment strategy: prime, Treasury and Tax-exempt. Prime MMFscan invest in private debts, while Treasury and Tax-exempt only invest in government securities. During the 2008 crisis, primefunds were the ones that exposed to losses of other shadow banks and suffered the runs. Therefore, my analysis focuses on primeMMFs.22From this perspective, a recent rule change adopted by the SEC which requires all prime MMFs to float their share prices mayhelp to reduce the risk of bank runs, because it reduces the negative externality of early redemption on remaining depositors.212.6. Conclusionof bank runs. Therefore, this policy may increase the fragility of the system, rather than reducing it.23In summary, my paper supports the view that “monetary policy is too blunt a tool to address possiblefinancial imbalances” as expressed by Bernanke (2011) and Yellen (2014). The modern banking system isextremely complex. It is very difficult to fine-tune monetary policy to address financial stability concerns,because different institutions may have very different responses to monetary policy. Instead, we should usemacro-prudential regulations, which can be targeted to specific institutions.2.5.4 Implication of Shadow Banking for Depositor SurplusCommercial banks have considerable market power in local depository markets. The entry of shadow banksmay increase rate competition in the deposit market and potentially bring significant gain in depositor sur-plus. To access the impact of shadow banking on depositor surplus, I use the estimated structural model tosimulate a counterfactual economy with no MMFs. I solve deposit rates and market shares of commercialbanks in this counterfactual economy and calculate depositor surplus according to the new set of choicesand prices. In absence of MMFs, commercial banks pay slightly lower deposit rates (9 basis points) but gainmuch greater market shares (3.86 percent). I follow Nevo (2001) to estimate the gain for depositor surplusfrom the entry of MMFs. I first compute the expected utility for each type of depositor i from its optimalchoice.E[maxj∈{0,1,...J}ui,k, j | k]= ln(J∑j=0exp(δ j + r jσvi−ρ1{k 6= j}))(2.24)Then, I divide expected utility by the yield sensitivity to calculate the equivalent utility in the unit ofdeposit rates. Lastly, I sum over past choices and depositor types to calculate the aggregate surplus.Depositor Surplust =∑ipii∑ksi,k,t−11αiE[maxj∈{0,1,...J}ui,k, j | k](2.25)I compare the surplus in the counterfactual economy with the actual economy. The entry of shadowbanks on average generates 0.36 cents of a dollar per year in the sample period. This amounts to 50 billionsincrease in depositor surplus with an aggregate money supply of 14 trillions at the end of 2015. The changein depositor surplus has the same magnitude as national branching deregulation in the 1990s estimated byDick (2008), which is estimated to be 0.50 cents of a dollar. I further examine the time-series variation ofthe change in depositor surplus, which is plotted in Figure 2.12. The change in depositor surplus is largerwhen the Fed Funds rates are high, which is consistent with the previous result that commercial banks enjoygreater market power during these periods.2.6 ConclusionThis paper documents a new monetary transmission mechanism: the shadow bank channel. I find thatshadow bank money supply expands when the Fed raises interest rates. This is at odds with the conventionalwisdom in the commercial banking sector that monetary tightening reduces deposit creation. I show thatthe different clientele served by shadow banks can explain their different responses to monetary policy byusing a structural model of bank competition. Fitting my model to institution-level commercial bank andmoney market fund data shows that this channel reduces the impact of monetary policy on money supply by40 percent. The macro-prudential implications of shadow banking are also explored.23On the contrary, using ultra-low interest rates to squeeze shadow banks may not be a good policy either. As interest ratesgo down, MMFs lose deposits to commercial banks. Since MMFs need to maintain a minimum scale to cover fixed operatingcosts, they may take excessive risk to increase their yields. This conjecture is supported by empirical evidence in Di Maggio andKacperczyk (2016).222.6. ConclusionThis paper highlights the complexity of the current banking system. The rise of the shadow bankingsector may have fundamentally changed the structure of the U.S. banking system and the transmissionmechanisms of monetary policy. This paper provides a quantitative framework to analyze these issues. Infuture research I hope to extend my analysis to the asset side of the shadow banking system. This willallows me to understand the potential interaction between monetary policy and risk-taking behaviors ofshadow banks.232.7. Tables and Figures2.7 Tables and FiguresCommercial Banks-1001020CB Deposit Growth Rates0246810Fed Funds Rates1987q1 1993q3 2000q1 2006q3 2013q1Fed Funds Rates CB Deposit Growth RatesShadow Banks-2002040SB Deposit Growth Rates0246810Fed Funds Rates1987q1 1993q3 2000q1 2006q3 2013q1Fed Funds Rates SB Deposit Growth RatesFigure 2.1: Deposit Growth Rates and the Fed Funds RatesThis figure shows the annual growth rates of the U.S. commercial and shadow bank deposits from 1987to 2012. The data are quarterly. Commercial bank deposits are the sum of checking and saving deposits.Shadow bank deposits include all the U.S. retail and institutional MMF shares. The data are obtained fromFRED.242.7. Tables and FiguresFigure 2.2: The U.S. Banking System252.7. Tables and Figures02468101987q1 1993q3 2000q1 2006q3 2013q1Fed Funds Rates Commercial Bank Deposit RatesShadow Bank Deposit RatesFigure 2.3: Deposit Rates and the Fed Funds RatesThis figure shows the average deposit rates of the U.S. commercial banks and MMFs from 1987 to 2012.The data are quarterly. Commercial bank deposit rates are obtained from the Call report. MMF yields areobtained from iMoneyNet.262.7. Tables and FiguresCommercial BanksMMFsFigure 2.4: Model Predicted Deposit Growth RatesThis figure shows the average deposit growth rates of commercial and MMFs predicted by the structuralmodel and in the data. The model is estimated using institution-level data on U.S. commercial banks andMMFs from 1994 to 2012.272.7. Tables and FiguresFigure 2.5: Distribution of Estimated ConvenienceThis figure shows the histogram of the estimated convenience for cash, commercial banks and MMFs. Theconvenience is defined as the inner product between the vector of characteristics, x, and correspondingsensitivities, β . Each observation is a MSA-sector median. The model is estimated using institution-leveldata on U.S. commercial banks and MMFs from 1994 to 2012.282.7. Tables and FiguresFigure 2.6: Distribution of Estimated Demand ElasticityThis figure shows the histogram of the estimated demand elasticity for cash, commercial banks and MMFs.Each observation is a MSA-sector median. The model is estimated using institution-level data on U.S.commercial banks and MMFs from 1994 to 2012.292.7. Tables and FiguresFigure 2.7: Difference in Deposit Rates (MMF-CB)This figure shows the difference in average deposit rates between MMFs and commercial banks predicted bythe structural model and in the data. The model is estimated using institution-level data on U.S. commercialbanks and MMFs from 1994 to 2012.302.7. Tables and FiguresHeterogeneous Depositors (σ > 0)Homogeneous Depositors (σ = 0)Figure 2.8: Difference in Markup and Marginal Cost (CB-MMF)This figure shows the difference in average markups and marginal costs between commercial and MMFspredicted by the structural model and in the data. The model is estimated using institution-level data on U.S.commercial banks and MMFs from 1994 to 2012.312.7. Tables and FiguresYield-oriented DepositorsTransaction-oriented DepositorsFigure 2.9: Choice Probability of Depositors by TypeThis figure shows the estimated probability for yield-oriented and transaction-oriented depositors to choosecommercial banks or MMFs over time. The model is estimated using institution-level data on U.S. commer-cial banks and MMFs from 1994 to 2012.322.7. Tables and FiguresFigure 2.10: Counterfactual Aggregate Money Growth RatesThis figure shows the observed and counterfactual aggregate money growth rates. The counterfactual sim-ulation is conducted by assuming that there are no MMFs in the economy. The model is estimated usinginstitution-level data on U.S. commercial banks and MMFs from 1994 to 2012.332.7. Tables and FiguresFigure 2.11: Demand Elasticity and the 2008 Runs on MMFsThis figure shows the scatter plot of demand elasticity against the percentage of redemption during the 2008runs on MMFs. The demand elasticity is estimated using data before the default of Lehman Brothers onSeptember 15, 2008. The percentage of redemption is calculated over four weeks following the start of therun (September 9th, 2008 - October 7th, 2008). The sample includes 30 largest prime MMFs in the U.S. asof September 9th, 2008.342.7. Tables and FiguresFigure 2.12: Change in Depositor SurplusThis figure shows change in depositor surplus due to the competition from MMFs. The change in surplus iscalculated by comparing the actual economy with a counterfactual economy where there are no MMFs. Themodel is estimated using institution-level data on U.S. commercial banks and MMFs from 1994 to 2012.35    variable mean sd p10 p25 p50 p75 p90         Cash Amount 1485.360 4284.194 150.692 224.929 402.519 1004.472 3007.820 Market share 11.345 1.474 9.653 10.432 11.270 12.228 13.180         Commercial Banks Amount 420.480 1561.996 49.098 70.789 115.569 244.662 735.627 Market share 3.512 1.707 1.752 2.370 3.200 4.294 5.533 Deposit rates 1.787 1.096 0.287 0.787 1.867 2.744 3.223 HHI 0.214 0.104 0.124 0.151 0.190 0.244 0.330 No. of banks 12.579 5.616 6.000 9.000 12.000 15.000 21.000 Branch density 22.589 11.324 10.273 14.546 20.699 28.602 37.074 No. of employee per branch 18.447 4.812 13.536 15.175 17.449 20.452 25.063 Single-market 0.181 0.165 0.000 0.000 0.154 0.286 0.417 Age 90.868 20.659 65.636 75.800 89.900 104.313 118.727 Expense of fixed assets 0.111 0.022 0.089 0.096 0.107 0.121 0.138 Salary 0.515 0.070 0.434 0.466 0.509 0.554 0.604 Reserve 1.100 1.500 0.149 0.260 0.477 1.051 3.365         MMFs Amount 65.919 224.330 4.930 8.721 15.693 40.365 127.887 Market share 0.445 0.110 0.288 0.385 0.451 0.517 0.586 Deposit rates 3.042 2.107 0.047 0.798 3.804 5.046 5.503 HHI 0.072 0.026 0.046 0.053 0.066 0.083 0.109 No. of banks 24.272 7.820 15.000 18.000 23.000 29.000 35.000 Rating dummy 0.516 0.068 0.438 0.467 0.520 0.565 0.600 Bank fund dummy 0.453 0.076 0.357 0.400 0.455 0.519 0.552 Check-writing dummy 0.343 0.063 0.261 0.296 0.333 0.391 0.438 Age 27.628 3.166 23.629 24.500 27.348 30.333 31.750 Management cost 0.182 0.052 0.123 0.166 0.191 0.217 0.228 Other cost 0.141 0.045 0.114 0.121 0.137 0.158 0.208 Table 2.1: Summary Statistics36    This table presents time series regressions of aggregate deposit growth rates on the Fed Funds rates. The data frequency is quarterly. The sample period is from 1990 to 2012. Standard errors in brackets are computed with Newey-West standard errors with 4 lags.  ***, **, * represent 1%, 5%, and 10% significance, respectively.         Dependent variable: Deposit growth rates  (1) (2) (3) (4) (5) (6) (7) (8)  CB MMF Cash Total CB MMF Cash Total          Fed Funds rates -1.896*** 3.890*** -0.0204 -0.0594 -1.407*** 3.929*** 0.340 0.177  [0.307] [0.766] [0.181] [0.326] [0.263] [0.875] [0.256] [0.369]          GDP growth    0.253 -1.289* -0.156 -0.171      [0.265] [0.691] [0.285] [0.333]          Inflation rates    0.609 -1.124 -0.337 0.0252      [0.412] [1.520] [0.419] [0.548]          VIX     0.412*** 0.161 0.106 0.282**      [0.101] [0.263] [0.0866] [0.128]          TED     -8.175*** 14.47** -2.914* 0.231      [1.735] [6.503] [1.607] [2.725]          N 92 92 92 92 92 92 92 92 adj. R-sq 0.517 0.405 -0.011 -0.010 0.662 0.636 0.072 0.194 Table 2.2: Effect of Monetary Policy on Aggregate Deposit Growth Rates37     This table presents cross-sectional regressions of shadow bank deposit holding on demographical variables. The sample includes 27,764 households in the Survey of Consumer Finance (2013). Shadow bank deposits are defined as deposits which are not insured by the government. Shadow dummy equals 1 if a household has shadow bank deposits, 0 otherwise. Shadow share is the share of shadow bank deposits in the total deposits of a household. Shadow amount (log) is the log dollar amount of shadow bank deposits held by a household. The independent variables are the demographics of the head of the household. Robust standard errors are in brackets.  ***, **, * represent 1%, 5%, and 10% significance, respectively.     Dependent variable: Shadow bank deposit holding  (1) (2) (3)   Shadow dummy Shadow share Shadow amount     Income 0.036*** 0.020*** 0.818***  [0.002] [0.001] [0.030]     College 0.040*** 0.014*** 1.392***  [0.004] [0.003] [0.096]     Unemployed 0.005 0.002 -0.210  [0.008] [0.005] [0.219]     Age -0.011*** -0.006*** 0.578***  [0.002] [0.001] [0.045]     Age2 0.001 0.004*** 0.120***  [0.002] [0.001] [0.039]     Home owner -0.010** -0.007** 0.394***  [0.004] [0.003] [0.097]     Car owner -0.010** -0.007*** 0.233**  [0.004] [0.003] [0.107]     Female 0.023*** 0.018*** -0.458***  [0.005] [0.004] [0.156]     Married 0.018*** 0.009*** -0.048  [0.005] [0.003] [0.131]     N 27764 27764 2860 adj. R-sq 0.047 0.025 0.593 Table 2.3: Demographic Determinants of Shadow Bank Deposit Holding38  This table presents the estimates of demand parameters of the structural model. The first column is the full model, and the second is the logit model in which there is no depositor heterogeneity and adjustment cost. The sample is a panel of U.S. commercial banks and MMFs from 1994 to 2012. Robust standard errors are in brackets.  ***, **, * represent 1%, 5%, and 10% significance, respectively.   Dependent variable: Mean utility  (1) (2)  Full Logit Yield sensitivity 0.764*** -0.323***  [0.019] [0.0039]    Branch density 0.026*** 0.045***  [0.00053] [0.00018]    No. of employees 0.007*** 0.012***  [0.0013] [0.00041]    Single-market dummy 0.055*** 0.011*  [0.011] [0.0056]    Age (CB) -0.076*** -0.081***  [0.011] [0.005]    Age (MMF) 0.586*** 0.352***  [0.037] [0.019]    Rating -0.020*** 0.009**  [0.007] [0.0037]    Bank fund 0.000 -0.022***  [0.0071] [0.0037]    Check writing -0.014* -0.015***  [0.0076] [0.0042]    Yield sensitivity dispersion 1.053***   [0.025]     Adjustment cost 6.681***   [0.035]     Bank F.E. Y Y City F.E. Y Y Time F.E. Y Y N 269687 269687 adj. R-sq 0.496 0.759 Table 2.4: Demand Parameter Estimation39    Dependent variable: Marginal costs  (1) (2)   Full Logit    Management costs 1.069*** -0.085***  [0.037] [0.026]    Other cost 1.151*** 0.078***  [0.027] [0.019]    Branch density -0.011*** 0.003***  [0.000] [0.000]    No. of employees -0.010*** 0.008***  [0.001] [0.000]    Expense of fixed assets -1.217*** 6.819***  [0.154] [0.106]    Salary 0.475*** 0.584***  [0.040] [0.027]    Reserve cost 1.765*** 11.449***  [0.135] [0.093]    FFR*Cash dummy 1.000*** 1.000***  [0.006] [0.004]  269687 269687 adj. R-sq 0.504 0.610  This table presents the estimates of supply parameters of the structural model. The first column is the full model, and the second is the logit model in which there is no depositor heterogeneity and adjustment cost. The sample is a panel of U.S. commercial banks and MMFs from 1994 to 2012. Robust standard errors are in brackets.  ***, **, * represent 1%, 5%, and 10% significance, respectively.   Table 2.5: Supply Parameter Estimation40        Cash CB MMF     Cash 0.1486    [0.571]       CB  1.0668    [1.0208]      MMF   1.8112       [0.4647]  This table presents the median and standard deviation (in brackets) of own-rates elasticity of cash, commercial banks, and MMFs estimated from the full model. Each entry gives the percent change of the market share of product i with one percent change of its own deposit rates.   Table 2.6: Own-rate Elasticity41     Cash CB MMF     Cash  0.0018 -0.0002   [0.0204] [0.0007]     CB 0.0073 -0.0087 -0.0021  [0.1264] [0.0247] [0.0032]     MMF -0.0062 -0.0119 -0.0048   [0.0456] [0.0156] [0.0052]   This table presents the median and standard deviation (in brackets) of cross-rates elasticity of cash, commercial banks, and MMFs estimated from the full model. The entry of the i-th row and j-th column shows the percent change of the market share of a product in category i (cash, CB, MMF) with one percent change of the deposit rates of a different product in category j (cash, CB, MMF).   Table 2.7: Cross-rate Elasticity42   Dependent variable: Change in lending/total lending  (1) (2) (3) (4) (5) (6)  CPs ABCPs Repos FRNS Treasury CB        Fed Funds rates 0.670*** 0.0848*** 0.457*** 0.270*** 0.171*** 0.217***  [0.0176] [0.00468] [0.0184] [0.0128] [0.0243] [0.00887]        GDP growth -0.102*** 0.00534 -0.144*** 0.111*** -0.534*** -0.0968***  [0.0246] [0.00648] [0.0256] [0.0178] [0.0361] [0.0122]        Inflation rates 0.139*** -0.000660 0.355*** -0.0913*** -0.0337 0.195***  [0.0411] [0.0113] [0.0448] [0.0307] [0.0606] [0.0209]        VIX 0.0315*** -0.00506*** -0.0117* 0.0186*** 0.126*** 0.00990***  [0.00670] [0.00180] [0.00706] [0.00487] [0.00949] [0.00337]        TED -0.268** -0.145*** -0.0882 -0.0463 1.568*** -0.0861  [0.126] [0.0332] [0.132] [0.0915] [0.180] [0.0619]        Fund Characteristics Y Y Y Y Y Y Fund F.E. Y Y Y Y Y Y N 41006 41006 41006 41006 41006 41006 adj. R-sq 0.102 0.093 0.098 0.069 0.094 0.076              This table presents regressions of MMF Lending on Fed Funds rates. The dependent variable is the annual change in a specific type of lending normalized by the lagged total lending (lagged one year). Fund characteristics include fund size (log), fund age, management costs, and other costs. The sample includes 1,148 MMFs in the period of 1998 to 2012. The data frequency is quarterly. Standard errors in brackets are clustered by time.  ***, **, * represent 1%, 5%, and 10% significance, respectively.  Table 2.8: Monetary Policy and MMF Lending43          Dependent variable: Asset growth rates       (1) (2) (3) (4) (5) (6)   Funding Corporations Finance Companies ABCP Issuers Captive Financial Institutions Broker-dealers Shadow Bank Total        Fed Funds rates 2.765*** 1.439** 4.528*** 0.972* 0.747 1.773***  [0.617] [0.649] [1.184] [0.500] [0.839] [0.555]        GDP growth 3.068*** 1.813*** 0.849 0.841 1.791 1.645***  [0.693] [0.574] [0.851] [0.638] [1.089] [0.369]        Inflation rates -2.853*** 0.647 -0.138 -4.271*** 1.667 -1.001*  [0.951] [0.720] [1.078] [0.784] [1.608] [0.583]        VIX 0.206 0.418*** -0.203 -0.0540 -0.528* -0.137  [0.147] [0.151] [0.336] [0.109] [0.267] [0.144]        TED 16.98*** -5.206* -4.021 11.61*** -5.235 2.261  [2.861] [2.625] [6.480] [2.036] [4.939] [2.984]        N 92 92 92 92 92 92 adj. R-sq 0.641 0.386 0.495 0.484 0.449 0.561  This table presents time series regressions of the aggregate asset growth rates of shadow banks on the Fed Funds rates. The dependent variable is the annual growth rates of the shadow bank assets. The data frequency is quarterly. The sample period is from 1990 to 2012. Standard errors in brackets are computed with Newey-West standard error with 4 lags.  ***, **, * represent 1%, 5%, and 10% significance, respectively. Table 2.9: Monetary Policy and Asset Growth of Shadow Banks44Chapter 3Regulation and Market Liquidity3.1 IntroductionThe aftermath of the 2008-09 financial crisis has witnessed one of the most active periods of regulatory in-tervention in U.S. financial history since the New Deal (Barr, 2012). A centerpiece of this sweeping reactionto the near collapse of the financial system, the Dodd-Frank Wall Street Reform and Consumer ProtectionAct (Dodd-Frank), was signed into law in July 2010. With Dodd-Frank, hundreds of regulatory rulemakingrequirements have been subsequently met, affecting virtually every dimension of modern financial activity,from derivatives trading to housing finance to capital requirements for depository institutions. In the back-drop of this intervention, a lack of rigorous assessment of the complex costs and benefits of the new ruleshas been highlighted (Cochrane, 2014). While Law scholars have been active in the regulatory debate at thequalitative level, quantitative work in Economics and Finance has been occasional and surprisingly sparse.Pertinently to this debate, this paper investigates the crucial claim that U.S. post-crisis financial regula-tory over-reach might have adversely affected the provision of market liquidity of a vast class of financialassets, structurally decreasing liquidity levels and increasing liquidity risk in fixed-income markets acrossthe board.Such claim is linked, but not uniquely, to a specific set of provisions embedded within recent legislation,the so-called Volcker Rule, statutorily delineated in Section 619 Title VI of the 2010 Dodd-Frank Act andfinalized by multiple regulatory agencies in January 2014. According to this provision, any banking entityis prohibited from engaging in proprietary trading or from acquiring or retaining an ownership interestin, sponsoring or having certain relationships with a hedge fund or private equity fund, subject to certainexemptions. Although this is in no way the only dimension of Dodd-Frank along which serious welfarelosses or liquidity shortages could have been potentially triggered, it emerged as one of the most hotlydebated, with roughly 17,000 public comments filed during the process of federal regulatory rulemaking(Bertrand, Bombardini and Trebbi, 2015). Specifically, some commentators24 have highlighted how byplacing undue artificial limits on securities inventory and retained risk and directly affecting inter-dealer24For instance regulators write in the final version of the Volcker Rule (p.5578 Federal Register / Vol. 79, No. 21 / Friday,January 31, 2014 / Rules and Regulations) "As discussed above, several commenters stated that the proposed rule would impacta banking entity’s ability to engage in market making related activity. Many of these commenters represented that, as a result,the proposed exemption would likely result in reduced liquidity[...]" and the Federal Register explicitly mentions on the matterof reduced liquidity comments received from "AllianceBernstein; Rep. Bachus et al. (Dec. 2011); EMTA; NASP; Wellington;Japanese Bankers Ass’n.; Sen. Hagan; Prof. Duffie; Investure; Standish Mellon; IR&M; MetLife; Lord Abbett; CommissionerBarnier; Quebec; IIF; Sumitomo Trust; Liberty Global; NYSE Euronext; CIEBA; EFAMA; SIFMA et al. (Prop. Trading) (Feb.2012); Credit Suisse (Seidel); JPMC; Morgan Stanley; Barclays; Goldman (Prop. Trading); BoA; Citigroup (Feb. 2012); STANY;ICE; BlackRock; SIFMA (Asset Mgmt.) (Feb. 2012); BDA (Feb. 2012); Putnam; Fixed Income Forum/Credit Roundtable; WesternAsset Mgmt.; ACLI (Feb. 2012); IAA; CME Group; Wells Fargo (Prop. Trading); Abbott Labs et al. (Feb.14, 2012); Abbott Labs etal. (Feb. 21, 2012); T. Rowe Price; Australian Bankers Ass’n. (Feb. 2012); FEI; AFMA; Sen. Carper et al.; PUC Texas; ERCOT;IHS; Columbia Mgmt.; SSgA (Feb. 2012); PNC et al.; Eaton Vance; Fidelity; ICI (Feb. 2012); British Bankers’ Ass’n.; Comm.on Capital Markets Regulation; Union Asset; Sen. Casey; Oliver Wyman (Dec. 2011); Oliver Wyman (Feb. 2012) (providingestimated impacts on asset valuation, borrowing costs, and transaction costs in the corporate bond market based on hypotheticalliquidity reduction scenarios); Thakor Study. The Agencies respond to comments regarding the potential market impact of the rulein Part IV.A.3.b.3., infra."Available at http://www.gpo.gov/fdsys/pkg/FR-2014-01-31/pdf/2013-31511.pdf453.1. Introductiontrading, the Volcker Rule could have severely limited market liquidity25. When recently the Congressionaldebate shifted on the merits of regulatory relief, one of the provisions considered for rolling back withinDodd-Frank included the prohibition of proprietary trading on the part of insured banking entities and theiraffiliates below certain thresholds26.A balanced view of the potential adverse welfare consequences of such provision is summarized inDuffie (2012): “The Agencies’ proposed implementation of the Volcker Rule would reduce the quality andcapacity of market making services that banks provide to U.S. investors. Investors and issues of securitieswould find it more costly to borrow, raise capital, invest, hedge risks, and obtain liquidity for their exist-ing positions. Eventually, non-bank providers of market-marking services would fill some or all of the lostmarket making capacity, but with an unpredictable and potentially adverse impact on the safety and sound-ness of the financial system. These near-term and long-run impacts should be considered carefully in theAgencies’ cost-benefit analysis of their final proposed rule. Regulatory capital and liquidity requirementsfor market making are a more cost effective method of treating the associated systemic risks.” Duffie (2012)further remarks on the needs for an appropriate assessment of the cost and benefits of the rule, an assessmentthat the empirical analysis we perform systematically complements. Thakor (2012) raises similar issues.Another focal point of post-crisis regulatory reform has been the Basel III framework, which was pro-duced in 2010 by the Basel Committee on Banking Supervision at the Bank for International Settlements.The Basel III final rule adopted by the U.S. federal banking regulators also implements some provisionsfrom the Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010 (Dodd-Frank Act; P.L. 111-203), which also addressed capital reserve requirements for banks (Getter, 2014). Basel III demands highercapital and liquidity buffers for banks, and imposes leverage restrictions on systemically important financialinstitutions. Despite the fact that higher levels of bank capital may reduce the probability of another finan-cial crisis, critics claim that these regulations might have unduly constrained banks’ ability to deploy capitalto market-making, and forced banks to charge clients more to use their balance sheet when they facilitatetrades or provide financing27.This paper formally assesses the effect of the U.S. post-crisis regulatory intervention, encompassing theDodd-Frank Act and Basel III, on market liquidity of a large portion of the U.S. fixed-income market.Our biggest empirical challenge is the unknown timing of regulatory impact. As vividly shown in Figure3.1, post-crisis financial regulation is characterized by protracted rulemaking processes and complicatedanticipatory responses and lagging reactions of market participants. For example, the Volcker Rule tookalmost four years to finalize, with the deadline being postponed several times. During the four years ofrulemaking, different banks wound down their proprietary trading desks at different times28. Conventionalmicro-econometric methods which compare liquidity before and after a treatment date are difficult to applyin this setting because it is unclear when regulation should have effects. The result of these methods couldbe sensitive to the assumption of the date around which the comparison is conducted29.25For example, on May 20, 2015 The Wall Street Journal in an article titled "Why Liquidity-Starved Markets Fear the Worst"reports "[..] a large part of the explanation lies in changes to regulation aimed at addressing weaknesses exposed by the financialcrisis. Banks must now hold vastly more capital, particularly against their trading books. The ring-fencing of proprietary tradingin the U.S. and retail banking in the U.K. has also squeezed liquidity. " Similar reasoning is implied by Alan Greenspan on theFinancial Times on August 17, 2015, who writes "Lawmakers and regulators, given elevated capital buffers, need to be far lessconcerned about the quality of the banks’ loan and securities portfolios since any losses would be absorbed by shareholders, nottaxpayers. This would enable the Dodd-Frank Act on financial regulation of 2010 to be shelved, ending its potential to distort themarkets — a potential seen in the recent decline in market liquidity and flexibility."26See S.1484 - Financial Regulatory Improvement Act of 2015, Title I: Regulatory Relief and Protection of Consumer AccessTo Credit. The bill is sponsored by Senate - Banking, Housing, and Urban Affairs Chairman Richard Shelby (R-AL).27See "Global Macro Research: A Look at Liquidity", Goldman Sachs, August, 2015.28The section "A Brief History of the Volcker Rule" in online appendix provides a detailed account of the rulemaking process ofthe Volcker Rule.29For example, if liquidity deterioration occurred before the regulation is implemented, a test comparing the liquidity around thedate of implementation may find no liquidity reduction.463.1. IntroductionTo address this challenge, we employ recent econometric approaches based on large factor models(Stock and Watson, 2011; Chen, Dolado and Gonzalo, 2014) to identify structural breaks in both levels anddynamic latent factors for a large set of liquidity proxies in fixed-income markets. Our empirical approachis attractive on several dimensions. First, our tests do not require a priori knowledge of the exact timing ofthe breaks. Second, we can capture not only sudden breaks in levels, but also breaks in slow-moving trends.Finally, the tests display excellent power properties and appear robust to confounding factors in a battery ofMonte Carlo simulations.We explore the market for U.S. corporate bonds, a heterogeneous asset class directly affected by theVolcker Rule and Basel III capital regulation. Exploiting the segmented nature of corporate bond market, weconstruct a large panel of liquidity measures by bond issue size, credit rating, and lead underwriter’s identity.Given that original underwriters typically tend to make markets on the specific securities underwritten, thisallows us to potentially identify bank-specific liquidity breaks and more nuanced disaggregated dynamics.We also study U.S. Treasuries, an asset class which is exempted from the Volcker Rule, but is still affected bythe stringent capital regulation of Basel III. Several commentators have ascribed recent episodes of tradingdisruption (e.g. the flash crash of October 15, 2014) to liquidity depletion.Against the popular claim that post-crisis regulation systematically hurt liquidity, we find no evidenceof liquidity deterioration during periods of regulatory interventions. While our methodologies do not allowto exactly quantify the causal effect of specific regulatory provisions on market liquidity, our portfolio oftests rebuts the hypothesis of a permanent reduction in liquidity. The empirical evidence robustly showseither no breaks or statistically significant breaks toward higher liquidity in the aftermath of Dodd-Frankand Basel III. Under the shared assumption that Dodd-Frank and Basel III represented in fact massivelyconsequential policy interventions, large negative effects on liquidity have to be rejected. This is a substantialcontribution in sharpening the debate on post-crisis financial regulatory intervention in the U.S. and Europe.We also present concordant evidence from microeconometric approaches based on difference-in-differencesof matched bonds samples that support these findings. Our work both qualifies frequent informal discussionon the lack of evidence of large deterioration in market liquidity provision, a view shared by a growinggroup of market participants and policy makers30, and is relevant to the rigorous assessment of the welfareconsequences of the Dodd-Frank Act and Basel III in terms of hindering the market making capacity oflarge financial institutions, one of the main welfare costs observers have ascribed to the recent regulatorysurge.This paper employs four different estimation strategies. First, we employ standard multiple breakpointtesting (Bai and Perron, 1998, 2003) on the level of liquidity as a first-pass examination on the potentialdates around which liquidity depletion may manifest. We find no evidence of liquidity depletion during theperiod of regulatory intervention (July 2010-December 2014), a period encompassing regulatory events suchas the passage of the Dodd-Frank Act and Basel III, the proposal and finalization of the Volcker Rule, andrelated shutdowns of proprietary trading desk by different banks. On the contrary, statistically significant30For example, the semi-annual Monetary Policy Report of the Federal Reserve in July 2015 writes: "Despite these increasedmarket discussions, a variety of metrics of liquidity in the nominal Treasury market do not indicate notable deteriorations", and"similar to the Treasury market, a range of conventional liquidity metrics in corporate bond markets also generally do not pointto a significant deterioration of market liquidity in recent years". See also Dudley (2015) and the New York Fed’s Liberty StreetEconomics blog series, in particular "Has U.S. Corporate Bond Market Liquidity Deteriorated?" by Adrian et al., Liberty StreetEconomics, October 05, 2015.This view is also echoed by some market participants. A Wall Street Journal commentary titled "Overlooking the Other Sources ofLiquidity" writes that "fortunately for investors, recent reforms and regulatory pressures have dramatically increased the number ofparticipants who can make prices and provide liquidity across many fixed-income markets. Markets that have opened to competitionnow enjoy better pricing, efficiency and resiliency". The global head of credit at Morgan Stanley, Steve Zamsky, said that "in ourday-to-day, moment-to-moment business today, marketplace works just fine". The chief investment officer of Oppenheimer Funds,Krishna Memani, the president of Bianco Research, Jim Bianco, and the president of Better Markets, Dennis Kelleher, also voicedscepticism on the "overheated" worries on bond market liquidity.473.1. Introductionbreaks toward higher liquidity are often detected during this period.Our second and third methodologies apply recent econometric approaches based on large factor models(Stock and Watson, 2011) to capture breaks in latent factor structures in the large panel of disaggregatedliquidity measures. Specifically, our second methodology focuses on single breakpoint testing for largedynamic factor models (Chen, Dolado and Gonzalo, 2014), while our third methodology extends to more arealistic multiple breakpoint case, transposing the intuition of Chen, Dolado and Gonzalo (2014) to Bai andPerron (2003) type tests. These methodologies allow flexible forms of structural breaks (including breaks intrends, in serial correlation, or in factor loadings), and help us to answer the deeper question whether marketliquidity would be higher or lower in absence of regulatory intervention. In simulations we show that ourmethodologies can successfully identify the onset of a gradual liquidity deterioration, even when masked byconfounding factors, and accurately estimate the counterfactual path of liquidity using observed data.We apply these methodologies to a large panel of disaggregate liquidity measures for corporate bonds.Our tests robustly capture breaks in latent liquidity dynamics at the start and at the end of the 2008-09 crisis(and indeed these tests can be employed to precisely time the beginning and end of the liquidity crisis). Thisreassures us on the tests having sufficient power within this specific empirical application. However, wefind no systematic statistical evidence of structural breaks leading to lower liquidity during the period ofregulatory intervention (July 2010-December 2014).As opposed to time-series approaches delineated above, our fourth estimation strategy relies on a stan-dard microeconometric approach in estimating liquidity deterioration around salient regulatory events, namelydifference-in-differences matching (Heckman, Ichimura, Todd, 1997; Heckman, Ichimura, Smith, and Todd,1998; Smith and Todd, 2005). In this part of the analysis we focus on the finalization of the Volcker Rulealone. We construct a dataset of bonds matched by issue size and credit rating, split between treatment andcontrol based on whether the original underwriter is covered or not by Volcker Rule provisions. Matchingallows for balancing between covered and non-covered bonds, assuaging concerns of attenuation due toheterogeneity across the two groups of securities.Consistently across all four estimation strategies, this paper finds no systematic evidence of deteriorationin liquidity levels or structural breaks in dynamic latent factors of the U.S. fixed-income market duringperiods of heightened regulatory interventions. This is in stark contrast to the popular claim that post-crisiswould cause severe depletion in market liquidity. Instead, consistent with the view shared by an increasinggroup of policy makers and market participants, we find breaks toward higher liquidity during these periods,possibly due to entry of non-banking participants and increase in competition between market makers. Wealso document some changes in the market structure, notably the diminishing dealer inventory and the shiftfrom principal-based trading towards agency-based trading. These evolutions in market structure startedbefore the regulatory intervention, and do not appear to be associated with deterioration in commonly usedliquidity measures. To the best of our knowledge, this is one of the very first studies to statistically assessliquidity depletion related to regulatory activity post-2008.Our work is related to several strands of literature in both economics and finance. The first strand ofliterature studies the determinants and measurement of market liquidity. A recent comprehensive surveyon this literature can be found in Vayanos and Wang (2012). Theoretical works such as Grossman andStiglitz (1980), Kyle (1985), Roll (1984), Grossman and Miller (1988), Amihud and Mendelson (1986), Gromb and Vayanos (2002), Duffie, Garleanu, and Pedersen (2005), and Brunnermeier and Pedersen(2009) relate illiquidity to underlying market imperfections such as participation costs, transaction costs,asymmetric information, imperfect competition, funding constraints, and search frictions. Many empiricalworks have since studied various measures of market liquidity across different asset classes, such as priceimpact (Amihud measure), price reversal (Roll measure), and bid-ask spreads. It has been shown that theseliquidity measures are related to market frictions as suggested by theory, and can explain asset returns inboth cross section and time series (see Amihud, Mendelson and Pedersen (2006) for a recent survey). Recentstudies of fixed-income market liquidity can be found in Edwards, Harris, and Piwowar (2007), Bao, Pan483.1. Introductionand Wang (2011), Feldhütter (2011), Dick-Nielsen, Feldhütter, and Lando (2012), Krishnamurthy (2002),and Hu, Pan and Wang (2012).A second strand of connected literature studies statistical tests of structural changes31. These methodolo-gies have been widely used in the macroeconomic literature to study structural changes in inflation-outputrelations, labor productivity, and monetary policy regimes32. Our paper contributes to this literature by em-ploying a test of multiple breaks with unknown dates in dynamic factor models, transposing the intuition ofChen, Dolado, and Gonzalo (2014) to Bai and Perron (1998) type tests. We show that this type of tests isparticularly useful when the timing of regulatory impact is unclear.A third and important strand of literature pertains to the cost-benefit analysis of financial regulation.By every stretch of imagination, this literature remains considerably underdeveloped relative to potentialwelfare benefits of rigorous and data-driven regulatory intervention. Such limitations have been lamentednot only by financial economists such as Cochrane (2014), but have been central motivation of judicial in-tervention33. Cochrane (2014) discusses at length the complexity of deriving meaningful assessments ofregulatory counterfactuals in financial and banking regulation, question also discussed in Posner and Weyl(2013). Relative to the pessimistic assessment in Coates and John (2014) of the infeasibility of meaningfulcost-benefit analysis in financial and banking regulation34, our paper offers a more optimistic counterpoint,at least in terms of ex-post quantitative assessment35 along the specific dimension of market liquidity de-pletion. Related our study, Bessembinder et al. (2016) find that trade execution costs of corporate bondshave decreased over time, a finding consistent with ours. However, they interpret the decline in inventoryand the shift of dealers’ business model as a sign of liquidity deterioration induced by post-crisis regulation,while we find that the shift started before regulatory intervention, and does not seem to be associated withdeterioration in other commonly used liquidity measures. In other OTC markets, Loon and Zhang (2016)provide evidence that Dodd-Frank improves the liquidity in the CDS market through several reforms suchas public dissemination of transactions and central counterparty (CCP) clearing.A fourth literature touched by this paper revolves around the post-financial crisis policy responses. Mc-Carthy, Poole, and Rosenthal (2013) debate political distortions in post-crisis responses, an issue also ex-plored in Frieden (2015) and Mian, Sufi, and Trebbi (2014). More explicitly, Mian, Sufi and Trebbi (2010)focus on the legislative response to the financial crisis pre-dating the Dodd-Frank Act, while Kaiser (2013)offers an interesting and detailed discussion of the congressional evolution of the Dodd-Frank Act itself.Finally, the regulatory rulemaking of Dodd-Frank is fully explored from a systematic empirical perspectiveby Bertrand, Bombardini, and Trebbi (2015).The remainder of this paper is organized as follows. In Section 3.2 we discuss the main empiricalmeasures, the variables construction, and provide a descriptive analysis of our samples. In Section 3.3 wediscuss our econometric model and single breakpoint/multiple breakpoint testing in dynamic factor models.Our main empirical results on U.S. corporate bonds are reported in Section 3.4 and on Treasuries in Section3.5. Section 3.6 concludes.31Important theoretical contributions include Andrews (1983), Andrews and Ploberger (1994), Bai and Perron (1998), Stock andWatson (2002, 2011), and Chen, Dolado, and Gonzalo (2014).32See McConnell and Perez-Quiros (2000), Hansen (2001), and Stock and Watson (2011) for examples of applications.33Coates and John (2014) referring to Business Roundtable et al. v. SEC, 647 F. 3d 1144 (D.C. Cir. 2011), report that "One panelof the U.S. Court of Appeals for the District of Columbia Circuit, composed entirely of Republican-appointed judges, has held thatexisting law requires the SEC to quantify the costs and benefits of its proposed rules".34Specifically speaking about the Volcker Rule, Coates and John (2014, p.73): "Could the agencies go beyond conceptual CBAand conduct a reliable, precise, quantified CBA/FR? The short answer is no. There is simply no historical data on which anyonecould base a reliable estimate of the benefits of preventing banks from engaging in proprietary trading or investing in hedge andprivate equity funds."35See also Cochrane (2014)’s discussion of retrospective analysis of financial regulation.493.2. Data3.2 Data3.2.1 U.S. Corporate Bonds Sample DescriptionThe first main data set used for this paper is the Financial Industry Regulatory Authority’s (FINRA) TRACE.This data currently provides transaction-level information of approximately 99% of all secondary corporatebond market transactions. Our sample period is from April 1, 2005 to December 31, 2014, covering the2008-09 financial crisis and post-crisis regulatory interventions. We filter out erroneous trades followingDick-Nielsen, Feldhütter, and Lando (2012).We merge the cleaned TRACE transactions to bond characteristics provided by Mergent Fixed IncomeData. This data provides bond-level information such as issue date, issuance size, coupon rate, maturitydate, credit ratings, underwriter identity and roles. Following Dick-Nielsen, Feldhütter, and Lando (2012),we limit the sample to fixed-rate bonds that are not callable, convertible, putable, or have sinking fund pro-visions. We drop bonds issued more than 10 years ago, since these old bonds present very few transactions.Since our goal is to provide the most comprehensive coverage of U.S. corporate bond market, we keep bondswith semi-annual coupons because they are the most common bonds in the U.S.. The raw TRACE data con-tains 34,422 bonds. After applying the above filters, our final sample contains 18,632 semi-annual couponbonds36. Using the underwriting information from Mergent, we link each bond to its lead underwriters.We first construct the nine measures for each corporate bond in our sample. Then we calculate the equalweighted average by bond rating group (investment-grade v.s high-yield) and issue size (above $1 billion v.s.below $1 billion) for each underwriter, which we refer as disaggregate series37. Since smaller underwritersonly underwrite a limited number of bonds, this makes the underwriter-level measure of liquidity quitenoisy. Therefore, we keep the top 4 biggest underwriters, Bank of America (Merrill Lynch), JPMorganChase, Morgan Stanley and Goldman Sachs, and combine the rest into a residual “Others” group. We alsoconstruct aggregate liquidity measures for the whole corporate bond market.3.2.2 Corporate Bonds Liquidity Measures: ConstructionMarket liquidity is the degree to which investors can execute a given trade size within a given period oftime without moving the price against the trade. We use the following nine liquidity measures which arecommonly used in the literature to capture different aspects of liquidity (the easiness to trade, the pecuniarycost of trading, etc.). Previous literature has shown that these liquidity measures generally affect assetprices, indicating that investors indeed care about them38. All measures below are decreasing in the level ofliquidity39.1. Amihud measure. Amihud (2002) constructs an illiquidity measure based on the theoretical modelof Kyle (1985). We use a slightly modified version of this measure following Dick-Nielsen, Feldhütter, andLando (2012). The Amihud proxy measures the price impact of a trade per unit traded. For a given bond,define r j,i,t as the return and Q j,i,t as the trade size (in million $) of the j−th trade on day i in month t. Thedaily Amihud measure is the average of the absolute returns divided by the corresponding trade size within36This is different from Dick-Nielsen, Feldhütter, and Lando (2012), who keep the no-coupon bullet bonds. They cover 2,224bullet bonds and turn to focus on more liquid segment of the market.37We also experimented with value-weighted averages with similar results to the ones reported below.38Dick-Nielsen, Feldhütter, and Lando (2012) show that higher value of Amihud measure, Roll measure, IRC, Amihud vari-ability, and IRC variability are associated with significantly higher credit spreads of corporate bonds. However, the evidence ofturnover and zero-trading days is mixed.39Some measures (e.g. Amihud) require a minimum number of trades to compute. We keep all the observations even if someliquidity measures are missing in certain days because we want to have a comprehensive coverage of the entire bond universe. Tobe sure, measures such as zero-trading days and turnover can be computed for all bonds.503.2. Dataday i:Amihudi,t =1Ni,tNi,t∑j=1∣∣r j,i,t∣∣Q j,i,t(3.1)where Ni,t +1 is the number of trades recorded on day i. We exclude retail trades (i.e. trades below $100,000in volume), as they are unlikely to have price impact. At least two trades are required on a given day tocalculate the measure, and we define a monthly Amihud measure by taking the median of the daily measureswithin month t.2. Imputed round-trip cost (IRC). Feldhütter (2012) shows that if a bond that does not trade for dayssuddenly has two or three trades with the same volume within a short period of time (one day in our defini-tion), then such trades are likely part of a pre-matched arrangement in which a dealer has matched a buyerand a seller. These trades are defined as a set of imputed round-trip trades. The difference between highestand lowest price in a set of imputed round-trip trades is the bid-ask spread collected by the dealer, whichis a measure of liquidity of the bond. We follow this approach. Specifically, for a given bond, on each dayi we identify sets of imputed round-trip trades indexed by k. A set of imputed round-trip trades involvestwo or more transactions with the same trading volume. Define Pmaxk,i,t (resp. Pmink,i,t ) as the maximum (resp.minimum) price among all the transactions in the k-th set of round-trip trades for that bond on day i in montht. The imputed round-trip cost of k-th set of round-trip trade is defined as:IRCk,i,t =Pmaxk,i,t −Pmink,i,tPmink,i,t. (3.2)We define a monthly IRC measure by taking the mean of the IRC of each set of imputed round-trip tradeswithin month t, weighted by the number of transactions involved in each set of imputed round-trip trades.3. Roll measure. The intuition of the Roll measure is as follows: the transaction price tends to bouncebetween the bid and ask price, which causes consecutive trade returns to be negatively correlated. Undercertain assumptions as shown in Roll (1984), the Roll measure equals to the bid-ask spread. The Rollmeasure is defined as two times the square root of the negative covariance between two consecutive dailyreturns ri,t ,ri−1,t in month t. If the covariance is positive, the covariance is replaced with zero.Rollt = 2√−Cov(ri,t ,ri−1,t) (3.3)4. Non-block trades. A trade is defined as non-block trade if the trading volume is less than $5 millionfor investment-grade bonds, and $1 million for high-yield bonds. The frequency of non-block trades isdefined as the ratio between the number of non-block trades and the total number of trades in month t.5. Size (negative log). Lower liquidity is usually associated with smaller size of trade. We first take thenegative logarithm of the par value for each trade, then compute the monthly median for each security.6. Turnover (negative). The annualized turnover for month t is defined as the annualized trading volumedevided by the amount outstanding. In what follows we take the negative of turnover as proxy of illiquidity,for consistency with the other measures.7. Zero trading days. We define this measure as the ratio between days with zero trade and the numberof trading days in month t.8 . Variability of Amihud and 9. Variability of IRC. Investors not only care about the current levelof liquidity, but also the risk of future liquidity. Therefore, we create the standard deviations of the dailyAmihud measure and imputed round-trip costs in a month as measures of liquidity risk.3.2.3 U.S. Treasuries Sample DescriptionWe use the CRSP Treasury database to construct our liquidity measures for the U.S. Treasury market. Thedaily data file is used to construct the Roll measure, and the monthly data file is used to construct the513.2. Dataon-the-run premium.We restrict our analysis to the same period as our corporate bond sample, April 1, 2005 to December31, 2014. Our sample consists of Treasury bills, notes, and bonds that are noncallable, nonflowering, andwith no special tax treatment. We also drop observations with obvious pricing errors such as negative prices.Treasury securities with remaining maturity less than 30 days are also dropped because of potential liquidityproblems. After applying the filters, our final sample contains 1,124 bonds. In addition to bond prices,we obtain the total Treasury trading volume from Securities Industry and Financial Markets Association(SIFMA), and the total public debt outstanding from Bloomberg.The liquidity measures for U.S. Treasuries are the following:1. Yield curve fitting noise. Hu, Pan, and Wang (2013) proposes a market-wide liquidity measure byexploiting the connection between the amount of arbitrage capital in the market and observed “noise” inU.S. Treasury bonds—the shortage of arbitrage capital allows yields to deviate more freely from the curve,resulting in more noise in prices. They construct the noise measure by first fitting Treasury daily prices intoa smooth yield curve, and then calculate the mean squared errors40.2. On-the-run premium. On-the-run Treasury bond (latest issue) usually enjoys a price premium overold bonds with similar maturity. We follow Gurkaynak et al. (2007) to construct the liquidity premium asthe difference between the yield of this synthetic off-the-run bond and the on-the-run bond.3. Roll measure and 4. Turnover (negative). Roll measure and Turnover (negative) measure are con-structed similarly as in the case of corporate bonds.3.2.4 Summary Statistics and DescriptivesTable 3.1 reports the summary statistics of the aggregate-level liquidity measures of the U.S. corporate bondsfor the period April 2005 to December 2014. For a typical bond, there is no single trade on 74% of businessdays. The annualized turnover rate is only 29%41. In comparison, stocks in NYSE have a turnover ratioof 92% in December 201442. Among all the trades, only 4% are block trades, and the median trade size is$35,000.To get a quantitative assessment of the illiquidity, one can compare various trading cost measures tocredit spreads, the compensation for investors to bear the credit and liquidity risk of corporate bonds. Theaverage credit spread of a U.S. corporate bond over a Treasury bond is 2.20% over our sample period. Incomparison, the mean Amihud measure, which is based on the impact of $1 million dollar trade, is 1.29%,as reported in Table 3.1. This amounts to half of the average credit spread earned in a year. The averageIRC, which measures the cost charged by dealers in a round-trip trade, is 0.70%. This equals to a third ofthe average credit spread. The average Roll measure is 1.59%, which implies a bid-ask spread as large asthree-fourth of the average credit spread.Additionally, investors face high uncertainty in trading cost when executing their trades, as shown bya high time series variability of the Amihud and IRC measure. In synthesis, Table 3.1 shows that the U.S.corporate bond market is typically not particularly liquid. In this respect, the a priori concerns of publiccommentators of the effects of regulatory intervention on market liquidity were well placed.In Table 3.2 we report the monthly linear correlations for each pair of liquidity proxies, to show consis-tency across our nine different measures of liquidity. Correlations are typically positive and sizeable, withpartial exceptions of the Turnover (negative) measure43.40We obtain the measure from the authors’ website at http://www.mit.edu/~junpan/Noise_Measure.xlsx41The average of turnover across bonds is much lower than the aggregate turnover of the market (total trading volume divided bytotal bond outstanding). This is because most of the total trading volume comes from a small group of large size bonds.42See http://www.nyxdata.com/nysedata/asp/factbook/ for the historical trading volume of NYSE stocks.43In online appendix Table 1, we provide summary statistics of the 180 disaggregate series. In online appendix Figure 1, we plottime series of nine liquidity measures for each underwriter.523.3. Econometric Model3.3 Econometric ModelOur goal is to formally test for structural breaks in the market liquidity of fixed-income assets in the after-math of the financial crisis. If post-crisis financial regulation indeed generates adverse impacts on marketliquidity, we should be able to detect structural breaks towards lower liquidity in the period of regulatory in-tervention (July 2010-December 2014). We present here the econometric setup that we are going to employ.As anticipated in Section 3.2 we take both an aggregate-level and a disaggregate-level perspective inour analysis. Let us define the matrix Y of L aggregate liquidity measures observed for T periods. Y is ofdimension (T ×L). With the term "aggregate" liquidity measure we mean a measure of liquidity (such asthose listed in Subsection (3.2.2)) that aggregates all securities in a market irrespective of identity of theunderwriter, issue size, or credit rating. Although intuitive, this approach may mask heterogeneity in thedynamics of different types of securities. Therefore, to identify specific structural breaks that might ariseonly within particular classes of securities or only for bonds where markets are made by specific under-writers/banks, we will refer to disaggregate liquidity measures as the matrix X of N > L liquidity measuresobserved for T periods. X is of dimension (T ×N) where each column measures liquidity grouping bondsat the level of(identity of the underwriter× issue size× credit rating) (3.4)As a matter of accounting, recall that for our case we have L = 9 measures. With 4 major underwritersplus 1 for the residual Others, 2 types of issue sizes (small or large), 2 types of credit rating (high yield andinvestment grade), we have N = 180. Our sample covers T = 117 months.3.3.1 Multiple Breakpoint Tests for Liquidity LevelsOur first methodology studies the question of whether regulatory intervention has produced structural breaksin the level of liquidity, in either Y or X . We employ tests for multiple breakpoint estimation (Bai and Perron,1998, 2003). The underlying assumption of these tests is that the level of liquidity fluctuates around a stablemean in absence of structural changes. If regulation shifts the long-run mean towards a different level, thesetests will detect the dates when the changes occur. Although highly stylized, this analysis offers a first-pass examination of the potential dates around which liquidity depletion may have happened. More flexiblemodels allowing for more general types of breaks will be presented below.3.3.2 Single Breakpoint Testing for Dynamic Factor ModelsOur second and third methodologies employ a more innovative approach based on dynamic factor models(Stock and Watson, 2011; Chen, Dolado and Gonzalo, 2014) to capture breaks in the latent factor structure.This approach allows flexible forms of structural breaks, such as breaks in trends, in serial correlation, orin factor loadings. These methodologies are more recent and deserve a more complete discussion. We nowintroduce the basic notation, econometric setup, and follow the exposition in Chen, Dolado and Gonzalo(2014), to which we refer for a detailed discussion of the proofs and the Monte Carlo evidence of power andsize of the tests.Consider a set of N observed liquidity measures constructed as in Section 3.2 and observed for t =1, ...,T periods, say, at monthly frequency. The matrix of observed disaggregate variables44 X of dimension(T ×N) is expressed as function of r unobserved factors F of dimension (T × r), a matrix Λ of factorloadings of dimension (N× r), and a matrix of idiosyncratic errors ε of dimension (T ×N). As typical inthe literature, we have in period t:Xt = ΛF ′t + εt . (3.5)44For the dynamic factor model analysis let us indicate with an abuse of notation X as the matrix of first differenced and normal-ized liquidity measures, as indicated by Stock and Watson (2011).533.3. Econometric ModelThis formulation accommodates flexibly several possible latent structures: r static factors; or r˜ dynamicfactors and p = r/r˜−1 lags; or an arbitrary combination of static and dynamic factors and lags (Stock andWatson, 2011).Due to their flexibility in accommodating general dynamics across correlated time series, large factormodels have enjoyed substantial success in the macroeconomics and finance literature. Stock and Watson(2002) show that the latent factors are consistently estimable by principal component analysis (PCA), anapproach we follow here. PCA allows to estimate the r factors of X :Fˆt ≡[Fˆ1t , Fˆ2t , ...Fˆrt](3.6)by focusing on the first r largest eigenvalues of the matrix XX ′ in the case T ≤N (or of the matrix X ′X in thecase T > N) and selecting the (appropriately orthogonalized and normalized) corresponding eigenvectors.Following Chen, Dolado and Gonzalo (2014) we also define Fˆ−1t ≡[Fˆ2t , ...Fˆrt].The number of factors r has to be estimated, as the true number of factors is unknown. Let us indicatewith rˆ such estimated value over the full sample.To this goal we employ ten different estimators, some with better finite sample properties than oth-ers, with the aim of providing an exhaustive range of rˆ’s. Eight of the estimators we employ follow thepopular information criteria (IC) proposed by Bai and Ng (2002), including their preferred ICp1, ICp2,PCp1, and PCp2. IC estimators, however, can occasionally display in finite samples a somewhat undesir-able dependency on a specific parameter necessary to the estimation: the maximum number of admissi-ble factors in the model (typically indicated as kmax). This may lead to overestimation of the true num-ber of factors (Ahn and Horenstein, 2014). It is also the reason we additionally employ the recent ER(eigenvalue ratio) and GR (growth ratio) estimators of Ahn and Horenstein (2014), which do not sharethis drawback and, by focusing on the ratio of subsequent eigenvalues (or the ratio of their logs), alsohinge on the straightforward intuition of principal component analysis screeplots (i.e. a popular graphi-cal representation of the progressive explanatory power of each principal component ranked by size of itseigenvalue). We consider all number of factors between the minimum and the maximum of the estimated{ICp1, ICp2, ICp3,PCp1,PCp2,PCp3,AIC3,BIC3,ER,GR}, allowing for at least rˆ = 2 unobserved factors (anecessary condition for the statistical tests below).We now proceed in introducing structural breaks in (3.5) and focus initially on the methodology fortesting a single breakpoint, leaving multiple breakpoints to Section 3.3.3. It is relevant first to specifywhether one is interested in breaks in the factor loadings Λ or in the factors F . Let us begin by representinga single structural break in all factor loadings at date τ:Xt = ΛF ′t + εt t = 1, ...,τ (3.7)Xt = ΓF ′t + εt t = τ+1, ...,T (3.8)where Γ is the post-break matrix of factor loadings of dimension (N× r). An important insight of Chen,Dolado and Gonzalo (2014) is that (3.7)-(3.8) can be represented asXt = ΛF ′t +∆G′t + εt (3.9)where ∆= Γ−Λ measures the change in the loadings andGt = 0 t = 1, ...,τ (3.10)Gt = Ft t = τ+1, ...,T.The notation so far has focused on a single structural breakpoint for all r factors. At a given breakpoint,Chen, Dolado and Gonzalo (2014) distinguish between two types of breaks: small and large. Consider k2543.3. Econometric Modelsmall breaks, of the type discussed by Stock and Watson (2002, 2009). These are defined as local-to-zeroinstabilities in the factor loadings that asymptotically average out without affecting estimation and inferenceunder PCA. These are not the type of breaks we are interested in. In the context of large policy shifts, oneis most likely interested in big structural breaks, indicated as k1 = r− k2. The formal definition is given inChen, Dolado and Gonzalo (2014), but more importantly it is proven that under k1 big breaks in (3.9), Fˆtestimated by PCA delivers inconsistent estimates of the space of the original factors Ft . Instead, defining G1tthe partition of Gt corresponding to the large breaks only, the full sample PCA delivers consistent estimatesof the space of the new factors[Ft G1t]. Specifically, over the full sample the number of factors tends tobe overestimated by k1. Chen, Dolado and Gonzalo (2014) prove that a factor model with r unobservedfactors and with 0< k1 ≤ r big structural breaks in the factor loadings at time τ admits a representation with(asymptotically) r+k1 factors. Particularly, given an IC estimator in Bai and Ng (2002) rˆ and under generalassumptions, it is shown (Proposition 2, p.34):limN,T→∞P [rˆ = r+ k1] = 1. (3.11)An important remark at this point is to notice that if the break date τ were known, one could recovera consistent estimate of r by simply splitting the sample in a “before-breakpoint” and “after-breakpoint”subsamples and performing PCA and Bai and Ng (2002) or Ahn and Horenstein (2014) in either subsample.In either case,limN,T→∞P[rˆbe f ore = r]= 1 (3.12)limN,T→∞P [rˆa f ter = r] = 1.both rˆbe f ore and rˆa f ter typically lower than the full sample estimate rˆ.For the sake of generality, we take the exact breakpoint date τ as unknown. Although we explicitlyconsider the exact date of the finalization of the Volcker Rule in the difference-in differences matchingbelow, the possibility of anticipatory behavior or of delayed response for a policy intervention so sizeableand publicly debated would caution against a ‘known breakpoint’ approach. Hence, we do not impose suchrestriction here.Chen, Dolado and Gonzalo (2014) present a test for the null H0 : k1 = 0 versus the alternative of atleast one big break H1 : k1 > 0 based on detecting breaks in Fˆt estimated over the full sample by PCA.The implementation is straightforward. Define βˆ the estimated (rˆ−1)× 1 coefficient vector obtained byregressing Fˆ1t on Fˆ−1t and Sˆ its corresponding Newey-West HAC covariance matrix45. One can test forstructural breaks in β by focusing for the case of unknown breakpoint τ = Tpi with pi ∈ Π ≡ (pi0,1−pi0)and 0 < pi0 < 1 based on Andrews (1993) Sup-Wald statistic or Sup-LM statistic. Specifically, for given τ,and hence pi = τ/T , define βˆ1 (pi) the estimated (rˆ−1)×1 coefficient vector obtained by regressing Fˆ1t onFˆ−1t for t = 1, ...,τ and βˆ2 (pi) the estimated (rˆ−1)×1 coefficient vector obtained by regressing Fˆ1t on Fˆ−1tfor t = τ+1, ...,T the Sup-Wald statistic is:L ∗ (Π) = suppi∈ΠTpi(1−pi)(βˆ1 (pi)− βˆ2 (pi))′Sˆ−1(βˆ1 (pi)− βˆ2 (pi))(3.13)and the Sup-LM statistic is:L (Π) = suppi∈Π1pi(1−pi)(1√TTpi∑t=1Fˆ−1t Fˆ1t)′Sˆ−1(1√TTpi∑t=1Fˆ−1t Fˆ1t)(3.14)45Newey and West (1987). Sˆ is estimated over the full sample.553.3. Econometric ModelIn the analysis we will maintain a conservative pi0 = 0.3 which in our case is not overly restrictive as itallows a search for structural breaks between January 2008 and January 2012 covering the full financialcrisis, the full legislative debate on Dodd-Frank and large part of the regulatory rulemaking period for theVolcker Rule. We employ the critical values for the (3.13) and (3.14) statistics reported in Andrews (1993).To conclude this subsection, let us consider the matter of detecting a structural break in the factorsthemselves as opposed to a break in the factor loadings at τ . There are at least two different formulationsfor a break in the factors one should consider. First, the formulation discussed in Chen, Dolado and Gonzalo(2014) considers maintaining unvaried the loadings Λ, but changing the variance-covariance matrix of the roriginal factors:E[FtF ′t]= Σ t = 1, ...,τ (3.15)E[FtF ′t]= Ξ t = τ+1, ...,T (3.16)where Σ is the factor covariance before the break and Ξ after the break and both are (r× r). Given that theapproach above focused on testing breaks in the Fˆt PCA factors estimated over the full sample, it may notappear surprising that the Sup tests above (based on regressing Fˆ1t on Fˆ−1t ) will be naturally able to pick upbreaks of the type (3.15)-(3.16). In fact, the same regression approach described above will reject the nullof big breaks in presence of changes in factors.It is possible however to discriminate between breaks in loadings and breaks in factors by noticing thatin the case of breaks in factors:limN,T→∞P [rˆ = r] = limN,T→∞P[rˆbe f ore = r]= limN,T→∞P [rˆa f ter = r] = 1. (3.17)This implies that in the case of breaks in the factors typically rˆ estimated over the whole sample will beidentical as when estimated on subsamples either before or after the breakpoint. In the case of breaks in theloadings, instead, rˆ estimated over the full sample will be higher than when estimated on subsamples eitherbefore or after the breakpoint, as evident from the result in (3.11).A second formulation for a break is more drastic and entails a break in the number of factors r in (3.5),that is the addition or subtraction of specific factors in the model at date τ . Section 3.3.4 offers an applicationof this methodology to this formulation and shows how it can be incorporated in this setting.3.3.3 Multiple Breakpoint Testing for Dynamic Factor ModelsLet us now focus on multiple structural breaks M in factor loadings at unknown dates τ1,τ2, ...,τM. Thisstructure partitions the sample period of length T in M+1 intervals:Xt = ΛF ′t + εt t = 1, ...,τ1 (3.18)Xt = Γ1F ′t + εt t = τ1+1, ...,τ2...Xt = ΓMF ′t + εt t = τM +1, ...,Twhere Γm with m = 1, ...,M are the post first break matrices of factor loadings of dimension (N× r). Inthe context of multiple breakpoints, standard estimators in the literature include the ones proposed by Baiand Perron (1998, 2003), which we employ in combination to the regression approach delineated in Section3.3.2. Considering the regression of Fˆ1t on Fˆ−1t with the goal of detecting not one, but multiple breakpoints,we implement the recommended approach of Bai and Perron (1998, 2003).Consider for the interval t = τm+1, ...,τm+1 the regression of Fˆ1t on Fˆ−1t in this subsample and call theestimated coefficient βˆm. Notice that, like βˆ1 (pi) and βˆ2 (pi) in Section 3.3.2, βˆm depends on the breakpoint563.3. Econometric Modelparameters, pim = τm/T and pim+1 = τm+1/T . Given M, let us also define βˆ =(βˆ ′1, βˆ ′2, ..., βˆ ′M+1)′. Bai andPerron (1998) first consider the Sup-F type test of the null hypothesis of no structural break (M = 0) againstthe alternative hypothesis that there is a known number of breaks M = k :sup(pi1,...,pik)FT (pi1, ...,pik;r−1) (3.19)=1T(T − (k+1)(r−1)k(r−1))βˆ ′R′(RSˆR′)−1Rβˆwhere R is the matrix such that(Rβˆ)′=(βˆ ′1− βˆ ′2, ..., βˆ ′k− βˆ ′k+1)and Sˆ is now an estimated HAC variancecovariance matrix of βˆ 46.As the number of breaks is unknown, a second type of test is more useful: Bai and Perron (1998)consider a test of the null hypothesis of no structural break (M = 0) against the alternative hypothesis thatthere is an unknown number of breaks M = m with m ranging between 1 and m¯, which is given47. The testis referred to as the double maximum test and two different statistics are employed:UDmaxFT (m¯;r−1) = max1≤m≤m¯sup(pi1,...,pik)FT (pi1, ...,pik;r−1) (3.20)which is unweighted with respect of each break number, andWDmaxFT (m¯;r−1,a1, ...,am¯) = max1≤m≤m¯am sup(pi1,...,pik)FT (pi1, ...,pik;r−1) (3.21)which is a weighted version, where weights are defined such that the marginal p-values are equal acrossvalues of m 48.The final test proposed by Bai and Perron is a sequential test. One proceeds by testing ` breaks against`+1 breaks. The test is commonly labelled supFT (`+1|`) and intuitively is built as follows. Consider the`+1 intervals generated by the ` break points under the null hypothesis. Within each interval a separate testof the type sup(pi1)FT (pi1;r−1) is run, i.e. a test of the null hypothesis of no break versus the alternativehypothesis of 1 break. The test rejects the null hypothesis in favor of `+ 1 breaks if, relatively to the sumof squared residuals obtained under the ` breaks model obtained by regressing Fˆ1t on Fˆ−1t and aggregatedacross all intervals, there is one additional break that produces a sum of squared residuals sufficiently smallerunder the `+1 breaks model.Bai and Perron (2003) recommend to first obtain both the UDmax and WDmax to test whether at leastone break is detected in the entire sample, as these tests are more prompt in rejecting the null hypothesis inpresence of multiple but contiguous breaks (e.g. which would be the case for instance if there were a breakat the beginning of the crisis and one at its end). If at least one break is detected, then the sequential approachshould be employed. Specifically one should select M = m such that supFT (`+1|`) are insignificant for`≥ m. We follow this approach here.46In the tests we perform we apply a short trimming of 10%. The Bai and Perron requires a minimal admissible distanceexpressed as fraction of T among any pair of breakpoints τm and τm+1 and we set it to 10% of the sample length, in order to allowfor relatively close multiple breaks. In all the test we also allow the distribution of εt to vary across different intervals.47In the tests we perform we allow for a maximum of m¯= 5 total breakpoints (which, as shown below, will prove to be sufficientlyhigh and is also the value suggested in Bai and Perron, 2003).48Specifically a1 = 1 and am = c(r− 1,α,1)/c(r− 1,α,m), for m > 1, where α is the significance level of the test and c(r−1,α,m) is the asymptotic critical value of the corresponding Sup-F test for m breaks, which is reported by Bai and Perron (1998,2003).573.4. Results for Market Liquidity of U.S. Corporate Bonds3.3.4 Breaks in Trends and a Simulation ExampleWe first provide a simple example to illustrate the flexibility of the dynamic factor model to capture breaksin trends, which are a realistic type of structural break in our setting. Suppose the illiquidity measure, lt , isjointly driven by supply of liquidity, st , and demand for liquidity, dt . Suppose that post-crisis regulationslead to a upward trend with a constant drift γ in illiquidity from τ+1:lt =−αst +βdt + et t = 1, ...,τ (3.22)lt =−αst +βdt + γ(t− τ)+ et t = τ+1, ...,T (3.23)Taking the first difference of the above equation system gives:xt =−α f1t +β f2t +0 f3t + εt t = 1, ...,τ (3.24)xt =−α f1t +β f2t + γ f3t + εt t = τ+1, ...,T (3.25)Where xt = lt − lt−1 is the innovation in illiquidity, f1t = st − st−1 is the supply factor, f2t = dt − dt−1is the demand factor, f3t = 1 is the regulation factor, and εt = et − et−1 is the differenced measurementerrors. It is immediately obvious that the break in trend can be reformulated as a break in the loading on theregulation factor, which can be consistently estimated by our methodology, as shown in Section (3.3.2).We simulate a panel of 180 liquidity measures to illustrate the power of our tests49. A detailed discussionof simulation and the Monte Carlo evidence of power and size of the tests can be found in Chen, Dolado andGonzalo (2014). Figure 3.2 plots the simulated liquidity index, defined as the average of 180 standardizedsimulated liquidity measures. The blue solid line is the path with the structural break, and the green dottedline plots the counterfactual scenario where regulation has no effects by design. The star sign indicates thedate when the structural break happens. The difference between the two paths is the regulation-induced liq-uidity gap. We can see that the magnitude of liquidity deterioration is very small at the beginning comparedwith the normal fluctuations of liquidity, and builds up very slowly. We conduct our structural break testsdescribed in Section (3.3.2) and (3.3.3). The estimated break date is marked by the vertical dashed line.Despite of the small magnitude, both tests successful identify the date of the structural break.We also use the dynamic factor model to estimate the counterfactual path of liquidity assuming there isno structural break. We first use the observed data before the break to estimate the loadings. Specifically, weregress each of the 180 liquidity measures on the estimated factors. Then we predict the counterfactual pathof liquidity assuming the factor loadings in the post-break period are the same as the pre-break period. Thered dash line shows the estimated path. Our estimation accurately traces out the true counterfactual path.Such accuracy is obtained because the large cross-section dimension (N = 180) of our liquidity measurescompensate the relatively short time span for loading estimation (62 months).3.4 Results for Market Liquidity of U.S. Corporate BondsFor U.S. corporate bonds we present four different estimation strategies. We will begin by applying multiplebreakpoint tests in levels to measures of market liquidity provisions. Subsequently, we will focus on adynamic factor model and presents results of both single and multiple breakpoints in factor loadings, withthe understanding that also further testing for factor breaks is available. Finally, we will focus on difference-in-differences matching results.49To mimic our empirical application, we simulate 180 liquidity measures driven by two latent factors: a supply factor anda demand factor. The two factors follow AR(1) process with autocorrelation of 0.5, and cross-correlation of 0.5. The loadingparameters on the two latent factors are drawn from N(0,1). A structural break occurs in July 2010 where 180 liquidity measuresstart to load on a new regulation factor, which follows AR(1) process with autocorrelation of 0.5 and an upward drift of 0.1. Theloading parameters on the regulation factor follows N(0,0.2). The cross-correlation between regulation and supply and demandfactor is also 0.5.583.4. Results for Market Liquidity of U.S. Corporate Bonds3.4.1 Multiple Breakpoint Tests for Liquidity LevelsWe begin by studying break in levels of our main nine liquidity measures (or properly seven measures ofliquidity levels and two measures of liquidity risk) employing the Bai and Perron (1998, 2003) estimationapproach for multiple unknown breakpoints in the undifferenced and unstandardized time series. This sim-ple test serves as a visual examination on the potential dates around which liquidity depletion may haveoccurred.At the onset we do not separate bonds by underwriter, issue size, and credit rating. Rather we aggregateall bonds and plot their time series in Figure 3.350. The estimated means for each sub-period (red dashedline) are also reported, where the break dates (a shift in the red dashed line) are estimated by the Bai andPerron (1998-2003) approach and are breaks significant at the 5% confidence level51.Concerning the dating of the structural breaks, the estimators should pick up at least the drastic reductionin liquidity produced by the near collapse of the U.S. financial system in September 2008 and the subsequentbreak towards more normal market liquidity levels at the end of 2009. Any detection of subsequent structuralbreaks towards lower levels of liquidity over the periods 2010-2014 needs instead to be carefully examined,as potential telltale indication of liquidity depletion concurrent with (and possibly caused by) regulatoryintervention. The 2010-14 period covers important regulatory events such as the approval of Dodd-Frank,shutdowns of proprietary trading desks by major banks, Basel III, and the approval of the interim and thefinalized Volcker Rule.The double maximum tests indicate the presence of at least one structural break at 5% confidence levelin all nine proxies52. The sequential supFT (`+1|`) indicates three breakpoints for the IRC, IRC (standarddeviation), Roll measure, and Non-block trades; one for the Amihud, Amihud (standard deviation), andTurnover (negative), and four for Size (negative) and Zero trading. As clarified by Figure 3.3, the Bai-Perron approach indicates clearly breaks in liquidity around the financial crisis. None of the structuralbreaks towards lower liquidity happen during the period of regulatory intervention. Instead, breaks towardshigher liquidity are detected for seven out of nine liquidity measures.We further compare the estimated mean liquidity in the subperiods before and after the crisis. Withthe possible exceptions of turnover and non-block trades, most of the liquidity measures indicate higherliquidity levels at the end of the sample period comparing to the pre-crisis level: the price impact of largetransactions goes down (Amihud), bid-asked spreads tighten (IRC), the price reversal goes down (Roll), themedian trade size stays stable, and trading becomes more frequent (Zero trading). Turnover and block tradeare somewhat lower than the pre-crisis level, but the breaks occurred before or during the crisis, well beforeregulations came into place. In fact, using the aggregate bond turnover statistics from SIFMA we find thatcorporate bond turnover has been on a downward trend for more than ten years, and actually flattens outduring the post-crisis period53, suggesting factors other than post-crisis regulation may be the driving force.For example, an increasing share of corporate bond trading may have moved to bond ETFs which is notcaptured by TRACE data54. The reduction in the share of block trades may be driven by market structuretransition from over-the-counter market to electronic trading platforms where transactions are conductedpredominantly as non-block trades (Hendershott and Madhavan, 2015). Even the share of block trade seemsto be lower, the median trade size is similar to the pre-crisis level.While this is prima facie evidence against drastic reductions in liquidity following regulatory interven-50In the online appendix Figure 2, we create an aggregate liquidity index using the average z-score of 9 liquidity measures. Thisapproach helps to average out some noises from a particular liquidity measure but may lose some detailed information. We applythe same analysis on this liquidity index, and rearch similar conclusion if we examine each measure separately.51The estimated break dates are reported in online appendix Table 2.52In online appendix Table 3 and 4 we report the relevant statistics for the double maximum tests and the supFT (`+1|`) tests.53The result is reported in online appendix Figure 3.54In online appendix Figure 3, we adjust the turnover by adding the trading volume of corporate bond ETFs. The trading volumefrom ETFs accounts a non-trivial share of decline in turnover, especially for high yield bonds.593.4. Results for Market Liquidity of U.S. Corporate Bondstion, it is still possible that at the level of specific types of corporate bonds structural breaks may arise. InFigure 3.4 we present a graph tracing for each month the fraction of the 180 disaggregated market liquidityvariables that are described to have a statistically significant (at 5% confidence level) break in that monthand in what direction (i.e. towards lower liquidity -in blue- or higher liquidity -in red). The bulk of thestructural breaks toward lower liquidity happens in July and August 2008, right before Lehman Brothers’failure. As it appears clear in Figure 3.4, if anything, around subsequent periods of regulatory interventionthe disaggregate liquidity measures pointed systematically toward higher liquidity, not lower.To understand the source of the disaggregate-level structural breaks, Figure 3.5 shows the decompositionof break dates by underwriting bank55. We can see that the bankruptcy of Lehman Brothers in September2008 caused liquidity reductions for all underwriters. In comparison, the later recoveries are more heteroge-nous: bonds underwritten by JP Morgan and Goldman Sachs experienced earlier recovery in liquidity thanbonds of other underwriters. This is consistent with anecdotal evidence that these two banks had relativelystronger balance sheets throughout the crisis.The most important observation from this graph, however, is from the later period when banks startto shutdown their proprietary trading desks after the passage of the Dodd-Frank Act. Were proprietarytrading indispensable for market making, one would expected to see bank-specific liquidity reductions lineup with an announced trading desk shutdown by the same bank. This is hardly the case: no large bankspecific liquidity reduction is observed after 2010 (all the bank-specific frequencies of liquidity reductionare below 5% after 2010)56. On the contrary, many banks experienced liquidity increases around July 2012,in the midst of regulatory interventions. There appears to be no clear evidence that the shutting down ofproprietary trading desks was associated with an adverse impact on market liquidity.3.4.2 Single Breakpoint Tests for the Dynamic Factor ModelThis subsection shifts the attention to a dynamic factor model with the goal of assessing whether the under-lying structure of correlation and of latent dynamics of liquidity across different bond types displays salientbreaks during the period of crisis and post-crisis regulatory intervention. Comparing to the breakpoint testsfor liquidity levels in the previous subsection, this approach allows more realistic modeling of the liquidityprocesses and captures more flexible forms of breaks, including breaks in trends, in serial correlation, andin factor loadings.We discuss here the application of Chen, Dolado and Gonzalo (2014) using the 2005-14 monthly sampleand our full matrix X of N = 180 differenced and standardized time series. A first preliminary step requiresto estimate the number of factors over the full sample T = 117. According to our discussion in Section 3.3.2this approach will not deliver a consistent estimate of the number of true factors in (3.5), but rather the sumof the true factors r and the number of big breaks in these factor loadings k1. In online appendix Table 5 wereport the full set of estimates based on Bai and Ng (2002) and Ahn and Horenstein (2014). Here we imposea kmax= 10 and notice that the estimates for {ICp1, ICp2, ICp3,PCp1,PCp2,PCp3,AIC3,BIC3,ER,GR} rangefrom 3 to 10. Although this range is not particularly tight, this is of little effect for the interpretation of ourmain findings in Figure 3.6.Figure 3.6 reports the Sup-Wald and the Sup-LM test statistics of the full interval over which the un-known breakpoint is allowed to belong given a conservative pi0 = 0.3. Such sample restriction is due topower loss concerns for the Sup tests (Andrews, 1993). Our interval of search of breakpoints covers theperiod between January 2008 and January 2012. Figure 3.6 also reports the Andrews (1993) critical values55In online appendix Figure 4 and 5, we show the decomposition by types of bonds and measures of liquidity. The results areconsistent.56A gradual shutdown of the trading desk would not be a problem for our test, since the estimated break points will show upsometime after the announcement date. However, we see none of this lagged liqudity reduction.603.4. Results for Market Liquidity of U.S. Corporate Bondsabove which the structural break is significant at the 10% and 5% confidence. We perform the analysis forany possible number of factors in the range estimated in online appendix Table 5.As evident from Figure 3.6, the Sup tests systematically pick breaks in factor loadings (at 5% confidence)when we allow a number of estimated factors above 4. Typically the Sup statistic indicates the breakpointas occurring during the 2008-2009 recession or shortly after. This is informative because again such datingdoes not correspond to regulatory events of prominence, but rather corresponds to the financial crisis itself.In essence what the Chen, Dolado and Gonzalo (2014) methodology allows us to exclude is that a structuralbreak in the underlying factor structure of the disaggregate liquidity occurred around dates of post-crisisregulatory activity57.So far the methodology in this subsection has focused on a single breakpoint, a restriction that, given themultitude of potential shocks affecting the U.S. financial system during our period of analysis, one shouldfind unwarranted. We relax this restriction in the following subsection.3.4.3 Multiple Breakpoint Tests for the Dynamic Factor ModelThis subsection employs the Bai and Perron (1998, 2003) approach within the dynamic factor model, trans-posing the logic of Chen, Dolado and Gonzalo (2014) to the multiple breakpoint setting.Figure 3.7 reports the results. Each panel represents a different factor models ranging from rˆ = 2, ...,10estimated factors, employing the Bai and Perron (1998, 2003) preferred approach to the first rˆ PCA estimatedfactors of the matrix X of differenced and standardize disaggregate liquidity measures. The blue solid linerepresents the liquidity index, defined as the average of 180 standardized liquidity measures. The dashedvertical lines indicate estimated dates of breaks in dynamic factor model58. The double maximum testsindicates the presence of at least a structural break at the 5% confidence level in all nine dynamic factormodels. The sequential supFT (`+1|`) indicates at most two breakpoints for the models with rˆ = 2,3,4,5,all essentially coincident with the start and end of the recession and the financial crisis. As in the previoussection, such dating occurs well before regulatory events of prominence (the passage of the Dodd-FrankAct in July 2010, the announcement of the final rules of Basel III in July 2013, or the announcement of thefinalized Volcker Rule in January 2014), but rather appears to correspond to dynamics within the confinesof the financial crisis itself.With rˆ = 6,7,8,9,10, more breakpoints in the factor loadings appear. Notably, there are breaks inlate 2010 and 2011, which fall into the regulatory intervention period. To examine whether these breaksindicate deterioration or improvement in liquidity, we estimate the counterfactual path of liquidity assumingno structural breaks occur from the last estimated breakpoint onwards. Specifically, we first estimate thefactor loadings using the data in the interval immediately before the structural break. Then we predict thecounterfactual path of the average liquidity after the break assuming the factor loadings take the same valueas before. For the models which do not detect breaks during the regulatory intervention period (rˆ= 2,3,4,5),we use the break closest to the regulatory intervention to conduct the counterfactual analysis. We conductthis exercise for each of the 180 liquidity measures, and take the average to create a liquidity index.The red dashed line in Figure 3.7 is the estimated counterfactual path of liquidity in absence of thelast structural break. Comparing the observed and counterfactual path, we can tell that whether liquiditywould be lower or higher in absence of the breaks. Consistently across all the nine specifications, thesestructural breaks during or right before the regulatory intervention period lead to slightly higher liquidity(lower illiquidity as the figure shows) comparing the counterfactual path. This is consistent with Figure 3.4which shows the level of liquidity breaks towards higher liquidity, not lower around this time period. One57In online appendix Table 6, we report the number of factors before and after the break.58In online appendix Table 7, 8, 9, and 10 we report the estimated break dates, double maximum test statistics, the supFT (`+1|`)test statistics, and the number of factors in each subperiod.613.4. Results for Market Liquidity of U.S. Corporate Bondslikely explanation could be the ability of our model to pick up an increasing role for electronic trading andfor open-end mutual funds59.3.4.4 Difference-in-Differences Matching for Liquidity LevelsWe now present a more standard estimation strategy based on a difference-in-differences exercise augmentedby matching of corporate bonds based on pre-treatment covariates (Heckman, Ichimura, Smith, and Todd,1998; Smith and Todd, 2005). Here, for reason that will become clear in the construction of the test, wewill focus only on the finalization of the Volcker Rule in January 2014 as our treatment date. Given thelimitation in our “post” sample of just 12 months available, we will take a symmetric 12-month windowaround January 201460.We proceed as follows. First, we manually classify the top 40 underwriters into two groups –one coveredby Volcker Rule and the other not covered based on the revised finalized version of Volcker Rule61. Thenwe identify the set of bonds which have at least one underwriter not covered by the Volcker Rule, that is anon-banking entity for which proprietary trading is not restricted. This set of bonds is a useful benchmarkas at least one of the underwriters who typically make market on that bond is virtually unconstrained by themain regulatory restriction in the rule, and hence virtually free to provide liquidity services in case bankingentities were so impaired. For each of these 3,106 non-Volcker Rule bonds that are outstanding betweenJanuary 2013 and December 2014, we find a match among all the Volcker Rule bonds issues in the samemonth, matures in the same month, has the same credit rating (investment grade/high yield), and has arelative size difference less than 50% of the average size of the pair62.Table 3.3 reports the results for a difference-in-differences model for each of our nine liquidity proxieswhere the treatment is administered to the Volcker Rule bonds after January 2014 and each regressioncontrols for the reciprocal of issue age, the reciprocal of issue age squared, bond fixed effects, and monthfixed effects. Standard errors are two-way clustered at the bond and month level. In seven out of ninemeasures the treatment does not predict reductions in liquidity with a confidence level of 5%. Only forIRC and IRC (standard deviation) we find a statistically significant effect. This is not particularly worryingsince more than 80% observations have missing IRC and IRC (standard deviation) measure in the matchedsample63. Overall, there seems to be no robust evidence of liquidity depletion as consequence of the VolckerRule.The regression evidence is also supported by the graphical representation. In Figure 3.8 we show thetime series of the Volcker Rule bonds and non-Volcker Rule bonds around the time when the revised finalizedversion of the Rule was approved (the vertical line, January 2014). Both time series are normalized to takevalue of 0 at December 2013. Were evidence of liquidity depletion present in the data, one would expect tosee systematically higher levels of the blue line after the treatment, a sign of reduced liquidity or heightened59See also Dudley (2015). In online appendix Table 11, we show that some of the latent factors are indeed significantly correlatedwith innovations in bond mutual fund flows.60Relative to the analysis above, the approach of this subsection is more restrictive, as it focuses on a single regulatory dimensionand relies on a difference-in-differences type of identification, but it is also an approach much more familiar to applied econometri-cians. In addition, by relying on different identifying assumptions, complements nicely the macroeconometric estimation strategyabove.61See the the following document from Federal Register for details of the final rule: http://www.gpo.gov/fdsys/pkg/FR-2014-01-31/pdf/2013-31476.pdf62There are fewer non-Volcker Rule bonds so we start our matching with them. If more than one bond satisfies the above criteria,we keep the one with smallest relative size difference. Since the Volcker Rule bonds are significantly larger than non-Volcker Rulebonds, many observations are dropped due to the last criterion on relative size. We ended up with a matched sample of 350 pairs ofbonds.63This is because the non-Volcker Rule bonds are usually very small and thinly traded. Therefore, liquidity measures whichrequire a certain number of transactions of specific types are very noisy.623.4. Results for Market Liquidity of U.S. Corporate Bondsliquidity risk. This is hardly the case both in reporting unconditional time series as in Figure 3.8 or timeseries where bond and month fixed effects are conditioned out (not reported to save space).3.4.5 Comments on the Change of Market StructureWith systematic evidence supporting the absence of structural deterioration in corporate bond liquidity, wewill now conclude this section by going back to two of the most often cited evidences for liquidity depletion:the decline in dealer corporate bond inventories and the increase in agency trading.Figure 3.9 shows the amount of corporate bonds held by dealer banks as the percentage of total corporatebond outstanding. We apply the Bai and Perron (1998, 2003) approach to estimate break points to this series,and three lessons can be learned from this test.First, the estimation shows, as is obvious in observing the time series of the raw data, that the majorreductions in dealer inventories occurred at the onset of the financial crisis (September 2008), far aheadof the post-crisis financial regulation. Therefore, at a minimum, there are other important factors drivingthe reductions of the inventories unrelated to the post-crisis financial regulation. One potential factor is thedeleveraging of broker-dealers forced by rehypothecation lenders (Mitchell and Pulvino, 2012).Second, the abnormally high level of bond holdings in 2007 seems the result of a pre-crisis run-up ofrisk-taking, as shown by a series of breaks towards greater holding amounts between 2002 and 2007. Inthis light, the dramatic reduction during the crisis appears actually more a “getting back to normal”. In thissense, using the pre-crisis level as a baseline to calculate the change of inventory is somewhat misleading.Third, there are two minor breaks, one in August 2011 and the other in March 2013, that fall into theperiod of regulatory intervention. However, as our tests on market liquidity have systematically shown,no structural reductions in market liquidity occurred during this period. This seems to suggest that someof the holdings may be held by the proprietary trading desks for risk-taking purposes, exactly the kind ofactivities that the Volcker Rule restrains. This possibility has also being raised by informal discussions(Brainard, 2015)64. The fact that dealer banks rapidly reduced their bond holding during 2008-09 crisissuggests that they demanded rather than supplied liquidity at the time when liquidity was most needed65.Another possibility is that this data series from the Federal Reserve overstates pre-crisis inventories becauseit improperly includes non-agency MBS. Indeed, one of the post-crisis breaks corresponds to the date ofsurvey revision66.Another commonly cited evidence of liquidity depletion is the shift from principal-based transactions toagency-based transactions. Principal and agency transactions are two main types of trade that dealers mayconduct. Principal trading occurs when a dealer uses its own inventory to fill the order for the client. Thepurpose behind principal trading is for the dealer to create extra profits (over and above the commissioncharged) for its own portfolios through price appreciations and bid-ask spreads. Traditionally, large bankshave mainly focused on principal trading. Agency trading instead involves a dealer searching for the securitydemand by a client from other clients or dealers. It is an empirical question whether regulation has causedthe shift to agency trading, and it is also unclear that such shift of business model would lead to liquiditydeterioration.Figure 3.10 plots the fraction of agency transactions over time67. We apply the Bai and Perron (1998,2003) approach to estimate break points in the level of this series. Coincidently with the decline in bond64In a speech by Federal Reserve Governor Lael Brainard at Salzburg Global Forum on July 1, 2015, he also mentioned that"since not all broker-dealer inventories are used for market-making activities, the extent to which lower inventories are affectingliquidity is unclear."65We thank Albert Kyle for suggesting this point.66See "Revised survey of primary dealers sheds new light on inventories," The Credit Line, April 18, 2013.67TRACE does not disseminate the agency trade indicator. We create a proxy which equals to 1 if two or more transactions of thesame bond with the same volume and at the same price happen at the same time. See Dick-Nielsen (2009) for a detailed discussionfor measuring agency trades with the TRACE database.633.5. Results for Market Liquidity of U.S. Treasuriesinventory, we find that there is a secular increasing trend of agency-based transactions, and the bulk ofthe increase occurred before regulatory interventions. The timing casts doubts on the claim that the post-crisis regulation causes this change. Moreover, comparing to the time series of our liquidity measures inFigure 3.3, the increase in agency-based transactions does not line up with periods of liquidity reductions,suggesting the two are not necessarily equivalent68.A more interesting question is what explains the structural breaks towards higher liquidity levels duringthe regulatory intervention period. We suggest that post-crisis regulation, by encouraging competition inmarket-making, could be a contributing factor. The idea is that big banks used to enjoy a big fundingadvantage over non-bank entities in corporate bond market-making business due to explicit (e.g. depositinsurance) and implicit (e.g. too-big-too-fail status) subsidies from the government. The funding advantageof big banks generated an entry barrier for non-bank entities to compete in this capital intensive business.If post-crisis regulations by and large reduced the funding advantage of big banks, this might have led to alevel playing field for non-bank entities to compete. As a result, more players can now enter the market, andincreased competition should induce a downward pressure on the price of intermediacy.There is evidence consistent with this explanation. The average number of competing market-makerstrading a bond has increased by 40% from the period of July 2007-April 2009 to the period of May 2009-May 2014 (Bessembinder et al. 2016). Competition between trading venues has also intensified: bond ETFsand electronic trading platforms such as MarketAxess provide investors with cost-effective ways to tradecorporate bonds outside the OTC market dominated by big banks. Some market commentators also expressa similar view. For example, on July 26, 2015 The Wall Street Journal in an article titled "Overlookingthe Other Sources of Liquidity" reports "Missing from much of this debate, however, is recognition of theradical transformation that has taken place in many fixed-income markets as barriers to entry have fallenand new liquidity providers have stepped forward."69 With more non-bank entities entering the market-making business, overall liquidity supply may increase, and sources of liquidity supply may become morediversified. In this sense, post-crisis regulation might have actually made market liquidity more resilient.This perspective is often missing in the post-crisis policy debate and definitely requires further investigationbeyond the scope of this paper.3.5 Results for Market Liquidity of U.S. TreasuriesThis section extends our analysis to the U.S. Treasuries market. Much of the interest and the discussionpertinent to this market’s liquidity can be ascribed to the salience of events like the flash crash of October15th, 2014 when the yield of the U.S. 10-year note dropped by 34 basis points from 2.2% to 1.86% in theeight minutes between 9:33 and 9:45AM Eastern Time.In Table 3.4 we report the summary statistics for this asset class, including Noise, On-the-run premium,Roll measure (all expressed in basis points) and Turnover (negative) over the April 2005-December 2014sample70, again calculated at the monthly frequency. The correlations among these proxies are intuitivelypositive, with the exception of Turnover (negative), as reported in Table 3.5. The reason for this counterintu-itive negative correlation is given by the construction of the measure for the Treasuries. As the denominatorin the Turnover variable is the total stock of public debt outstanding, the explosion of U.S. sovereign debt as68Even if agency transactions may not directly impact liquidity, a legitimate concern is that it may bias the measurement ofliquidity. To address this concern, we drop all agency trades and repeat our tests exclusively on principal-based transactions. Westill find no systematic evidence of liquidity deterioration. The results are available upon request.69See also an industry discussion panel titled "Are There Structural Issues in the U.S. Bond Market?" organized by the BrookingsInstitute for discussion on regulation and competition in market-making business.70In the online appendix Figure 6, we extend our analysis to an earlier sample period from 1995 to 2005, which covers thecollapse of LTCM, a liquidity crisis much smaller in scale comparing to the 2008-09 financial crisis.643.6. Conclusionsconsequence of the automatic stabilizers and the 2009 Fiscal Stimulus appear to severely affect the qualityof this measure post 2009, an issue that will become clearer below.We employ the Bai and Perron (1998, 2003) approach to estimate breakpoints in the level of the four liq-uidity time series: Noise, On-the-run premium, Roll measure and Turnover (negative)71. The correspondingdouble maximum tests indicates the presence of at least one structural break at the 5% confidence level inall four proxies, with the exception of the UDmax for the Noise variable. However, for the same variableWDmax reject the null that there is no break. The sequential supFT (`+1|`) indicates three breakpointsfor the Noise and Roll measures, one for the On-the-run premium and four for the Turnover (negative).Figure 3.11 reports an informative visualization of when the breakpoints happen over time and in whichdirection the series breaks. For both the Noise and Roll measures this approach clearly captures the suddendeterioration of market liquidity around the 2008-09 financial crisis and a return to normality mid-2009.The Roll measures seems to suggest further liquidity amelioration in December 2011 (in fact close to therelease of the first Proposed Volcker Rule published in November 2011). The On-the-run premium exhibitsqualitatively very similar dynamics, as evident from the North-East panel in Figure 3.11, but our approachfails to pick up a structural break at the start of the crisis. The only proxy that seems to systematically breakin terms of lower liquidity levels for Treasuries is Turnover (negative) in October 2008. However, lookingat the components of this measure, this result appears mainly driven by two factors: 1. Treasury issuancedramatically increased after 2008. 2. The Federal Reserve balance sheet structurally increased, holding avery large portfolio of public debt due to the Quantitative Easing. Since the Fed typically is not activelytrading, the turnover should intuitively drop.3.6 ConclusionsThis paper complements, both methodologically and substantively, a rigorous retrospective analysis of post-crisis regulatory intervention in domestic financial markets. Such analysis has been surprisingly bare interms of systematic empirical evidence and it appears to be a necessary exercise in informing future legisla-tive and rulemaking activities aimed at improving financial markets stability (Cochrane, 2014).We specifically focus on the aftermath of the 2008-09 U.S. financial crisis and on the role played bythe Dodd-Frank Act of 2010 and Basel III as potential triggers of liquidity shortages driven by retrench-ment of financial institutions adversely affected by overreaching regulation. Several market participantshave claimed this assessment to be crucial in the context of an informed cost-benefit analysis of regulatoryintervention and rulemaking.We initially focus on a large set of liquidity proxies with emphasis on the U.S. corporate bond market (anasset class likely to be adversely affected by regulatory tightening through disruption of ordinary market-making activities) and with particular attention paid to different underwriters, credit ratings, and issue sizes.Our analysis is based on multiple estimation strategies, including standard breakpoint tests in levels,tests for structural breaks in dynamic factor models and difference-in-differences matching analysis. Reas-suringly, the data display no statistical evidence of substantial deterioration in market liquidity after 2010.The tests presented are powerful enough to pick structural breaks in the data -they clearly pinpoint the cri-sis itself as a liquidity breakpoint- yet they consistently show no significant liquidity deterioration in theperiod of regulatory intervention covering the approval of the Dodd-Frank Act and Basel III, shutdowns ofproprietary trading desks by major banks, or the proposal and finalization of the Volcker Rule. If anything,we detect evidence of liquidity improvement during periods of regulatory interventions, possibly due to theentry of non-banking participants. Evidence from the U.S. Treasuries market, by and large, confirms the71Given the small number of time series available for the analysis of liquidity of Treasuries we do not employ dynamic factormodel approaches in this Section. Online appendix Table 12, 13 and 14 report the estimated break dates, the double maximum teststatistics and supFT (`+1|`) test statistics respectively.653.6. Conclusionsabsence of liquidity deterioration.663.7.TablesandFigures3.7 Tables and FiguresFigure 3.1: Timeline of Crisis and Post-Crisis Regulatory Activities673.7. Tables and FiguresFigure 3.2: Simulated Illiquidity IndexNotes: This graph shows the average of 180 simulated liquidity measures over time. The blue solid linerepresents the average of 180 liquidity measures if regulation leads to a gradual deterioration in marketliquidity, while the green dotted line represents counterfactual scenario where regulation has no effects. Thedashed vertical line indicates the date of true and estimated structural break in the latent factor structure.The red dashed line is the estimated counterfactual path. The break date is estimated by Chen, Dolado, andGonzalo (2014) and Bai and Perron (2003) approach with 5 percent significance level. The sample period isfrom April 2005 to December 2014. The data frequency is monthly. The grey area indicates recession.683.7. Tables and FiguresFigure 3.3: Time Series of Liquidity of U.S. Corporate Bonds (Aggregate-level)Notes: This graph shows the time series of 9 aggregate-level liquidity measures of U.S. corporate bondmarket (blue line), and the estimated mean for each sub-period (red dashed line). The break dates (dateswith a shift in the level of the red dashed line) are estimated by the Bai and Perron (1998-2003) approachwith 5 percent significance level. The sample period is from April 2005 to December 2014. The datafrequency is monthly. The grey area indicates recession.693.7. Tables and FiguresFigure 3.4: Breaks in the Means of Liquidity (Disaggregate-level)Notes: This graph shows the frequency of break in the levels of 180 disaggregate-level liquidity measures forthe U.S. corporate bond market over time. The x-axis shows the dates and the y-axis shows the correspond-ing fraction of the 180 liquidity measures which have a break at each date. The break dates are estimatedusing the Bai and Perron (1998-2003) approach with 5 percent significance level. The solid vertical lineindicates the passage of Dodd-Frank Act (July, 2010). The sample period is from April 2005 to December2014. The data frequency is monthly. The grey area indicates recession.703.7. Tables and FiguresFigure 3.5: Breaks in the Means of Liquidity by Underwriter (Disaggregate-level)Notes: This graph shows the decomposition of break dates by underwriter. The x-axis shows the datesand the y-axis shows the corresponding fraction of the 36 (=9×2×2) liquidity measures of each underwriterwhich have a break at each date. The break dates are estimated using the Bai and Perron (1998-2003)approach with 5 percent significance level. The solid vertical line indicates the passage of Dodd-Frank Act(July, 2010). The sample period is from April 2005 to December 2014. The data frequency is monthly.713.7. Tables and FiguresFigure 3.6: Breaks in the Means of Liquidity by Underwriter (Disaggregate-level)Notes: This graph shows the test statistics of a single break in factor structure of 180 disaggregate-levelliquidity measures employing the Chen, Dolado, and Gonzalo (2014) approach. The sample period is fromApril 2005 to December 2014. The solid vertical line indicates the passage of Dodd-Frank Act (July, 2010).The data frequency is monthly. The grey area indicates recession.723.7. Tables and FiguresFigure 3.7: Liquidity Index of the U.S. Corporate Bond MarketNotes: This graph shows the average of 180 standardized liquidity measures of U.S. corporate bond market(blue solid line) and the estimated counterfactual path (red dashed line). The dashed vertical line indicatesthe dates of estimated structural breaks in the latent factor structure. The solid vertical line indicates thepassage of Dodd-Frank Act (July, 2010). The break dates are estimated by Chen, Dolado, and Gonzalo(2014) and Bai and Perron (2003) approach with 5 percent significance level. The sample period is fromApril 2005 to December 2014. The data frequency is monthly. The grey area indicates recession.733.7. Tables and FiguresFigure 3.8: Liquidity of Volcker Rule and Non-Volcker Rule Bonds (Matched Sample)Notes: This graph shows the time series of liquidity of Volcker Rule bonds and non-Volcker Rule bondsaround the time when revised finalized version of the Volcker Rule was approved (January 2014). A non-Volcker Rule bond is defined as a bond which at least one of the underwriters is not subject to the VolckerRule. A Volcker Rule bond is defined as a bond which all of the underwriters are subject to the Volcker Rule.Both time series are normalized to 0 in December 2013. The red vertical line indicates the date when therevised finalized version of the Volcker Rule was approved (2014m1). The sample period is from January2013 to December 2014. The data frequency is monthly.743.7. Tables and FiguresFigure 3.9: Primary Dealer Corporate Bond HoldingNotes: This graph shows the time series of the U.S. primary dealer corporate bond holding as the percentageof total corporate bond outstanding (blue line) and the estimated mean for each sub-period (red dashed line).The solid vertical line indicates the passage of Dodd-Frank Act (July, 2010). The break dates (dates with ashift in the level of the red dashed line) are estimated by the Bai and Perron (1998-2003) approach with 5percent significance level. The sample period is from January 2002 to December 2014. The data frequencyis monthly. The grey area indicates recession.753.7. Tables and FiguresFigure 3.10: Fraction of Agency TransactionsNotes: This graph shows the fraction of agency transactions (blue line), and the estimated mean for eachsub-period (red dashed line) over time. The break dates (dates with a shift in the level of the red dashedline) are estimated by the Bai and Perron (1998-2003) approach with 5 percent significance level. The solidvertical line indicates the passage of Dodd-Frank Act (July, 2010). The sample period is from April 2005 toDecember 2014. The data frequency is monthly. The grey area indicates recession.763.7. Tables and FiguresFigure 3.11: Time Series of Liquidity of the U.S. Treasury BondsNotes: This graph shows the time series of liquidity measures of U.S. Treasury market (blue line), and theestimated mean for each sub-period (red dashed line). The break dates (dates with a shift in the level ofthe red dashed line) are estimated by the Bai and Perron (1998-2003) approach with 5 percent significancelevel. The solid vertical line indicates the passage of Dodd-Frank Act (July, 2010). The sample period isfrom April 2005 to December 2014. The data frequency is monthly. The grey area indicates recession.77                      Measures N mean sd p10 p25 p50 p75 p90          Amihud 117 1.29 0.48 0.79 0.94 1.17 1.47 2.12 Amihud (sd) 117 1.57 0.48 1.10 1.19 1.43 1.83 2.35 IRC 117 0.70 0.24 0.40 0.49 0.70 0.82 1.08 IRC (sd) 117 0.61 0.18 0.42 0.47 0.58 0.67 0.88 Roll 117 1.59 0.54 0.96 1.23 1.52 1.81 2.41 Non-block trade 117 0.96 0.01 0.93 0.94 0.96 0.97 0.97 Size (negative) 117 -10.48 0.20 -10.71 -10.66 -10.51 -10.28 -10.18 Turnover (negative) 117 -0.29 0.05 -0.36 -0.32 -0.28 -0.25 -0.23 Zero-trading 117 0.74 0.03 0.71 0.72 0.74 0.76 0.79                    Notes: This table shows the summary statistics of 9 aggregate-level liquidity measures for the U.S. corporate bond market. The sample period is from April 2005 to December 2014. The data frequency is monthly. The unit of Amihud, Amihud (sd), IRC, IRC (sd), and Roll is percentage point. The unit of Non-block trade, Turnover (negative) and Zero-trading is 1.     Table 3.1: Summary Statistics of the U.S. Corporate Bond Liquidity (Aggregate-level)78               Amihud Amihud (sd) IRC IRC (sd) Roll Non-block trade Size (negative) Turnover (negative)          Amihud (sd) 0.98        IRC 0.88 0.84       IRC (sd) 0.91 0.88 0.98      Roll 0.93 0.93 0.96 0.97     Non-block trade 0.29 0.33 -0.15 -0.06 0.01    Size (negative) 0.75 0.76 0.51 0.51 0.58 0.65   Turnover (negative) -0.07 0.05 -0.28 -0.20 -0.10 0.27 0.00  Zero-trading 0.40 0.43 0.54 0.52 0.56 -0.43 0.03 0.37                    Notes: This table shows the correlations among 9 aggregate-level liquidity measures for the U.S. corporate bond market. The sample period is from April 2005 to December 2014. The data frequency is monthly.  Table 3.2: Correlation Table of the U.S. Corporate Bond Liquidity (Aggregate Level)79      (1) (2) (3) (4) (5) (6) (7) (8) (9)   Amihud Amihud (sd) IRC IRC (sd) Roll Non-block trade Size (negative) Turnover (negative) Zero-trading           Volcker*Post -0.231 -0.115 0.106* 0.113** 0.0177 0.00168 0.0145 -0.0160 -0.00202  [0.466] [0.359] [0.0573] [0.0474] [0.122] [0.00265] [0.0859] [0.0118] [0.00320]           Controls Y Y Y Y Y Y Y Y Y Time F.E. Y Y Y Y Y Y Y Y Y Bond F.E. Y Y Y Y Y Y Y Y Y           Observations 542 340 1807 1260 1992 2006 2006 11060 11060 Adjusted R-squared 0.458 0.398 0.238 0.325 0.245 0.312 0.524 0.353 0.879  Notes: This table shows the difference-in-difference regression of Volcker Rule bonds and non-Volcker Rule bonds around the time when revised finalized version of the Volcker Rule is approved (January 2014). A non-Volcker Rule bond is defined as a bond which at least one of the underwriters is not subject to the Volcker Rule. Each of the non-Volcker Rule bonds in our sample is matched to a Volcker Rule bond which issues in the same month, matures in the same month, has the same rating group (investment-grade/high-yield), and has a relative size difference less than 50 percent of the average size of the pair. The sample period is from January 2013 to December 2014. The data frequency is monthly. Control variables include the reciprocal of issue age, and the reciprocal of issue age squared. The standard errors are two-way clustered at the bond and month level. ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.  Table 3.3: Difference-in-Difference Regression80    Measure N mean sd p10 p25 p50 p75 p90          Noise 117 3.14 3.24 1.20 1.48 1.93 3.33 6.51 On the run premium 117 13.48 12.62 3.33 6.23 8.94 16.39 28.73 Roll 117 13.37 4.09 8.62 10.35 12.73 15.83 19.23 Turnover 117 -11.48 3.93 -17.64 -14.76 -9.79 -8.11 -7.39  Notes: This table shows the summary statistics of liquidity measures for the U.S. Treasury market. The sample period is from April 2005 to December 2014. The data frequency is monthly. The unit of Noise, On the run premium and Roll measure is basis point. The unit of Turnover (negative) is 1.     Table 3.4: Summary Statistics of the U.S. Treasury Liquidity81     Noise On the run premium Roll     On the run premium 0.90   Roll 0.62 0.72  Turnover 0.03 -0.08 -0.37  Notes: This table shows the correlations between liquidity measures for the U.S. Treasury market. The sample period is from April 2005 to December 2014. The data frequency is monthly.  Table 3.5: Correlation Table of the U.S. Treasury Liquidity82Chapter 4Factions in Nondemocracies: Theory andEvidence from the Chinese CommunistParty4.1 IntroductionThis paper presents a theoretical and empirical analysis of the internal organization of China’s political linch-pin: the Chinese Communist Party (CCP). As the regime party of the People’s Republic of China (PRC), theCCP is de jure and de facto the be-all and end-all of political activity in the second largest economy and themost populous country in the world today72. This motivates the interest of political economists in the CCP.The nontransparent and often informal nature of elite interaction within a country lacking competitiveelections and with a rich history of informal political jousting among factional leaders raises formidableobstacles to a rigorous politico-economic analysis. The economic literature on the internal organization (and,we will see, factional competition) at the highest levels of the Chinese government is limited73. Politicalscientists focused on China studies have been more attentive, but also often more qualitative and descriptive,at least until recently74.The CCP remains today “a secretive, selective organization of about 65 million members who havepositions of influence in all sectors of Chinese society...” (Nathan and Gilley, 2003 p.7)75. Operations ofthe Politburo and the highest echelons of the CCP have been often described as opaque at best (Pye, 1980;Dittmer, 1995; Shih, 2008). As reported in Nathan (2016): “Deng built a system of tacit norms by whichsenior leaders were limited to two terms in office, members of the Politburo Standing Committee dividedleadership roles among themselves, and the senior leader made decisions in consultation with other leadersand retired elders.”Within this context, intra-elite competition is extremely hard to assess. The CCP officially rejects fac-tional elite politics76, but scholars since Nathan (1973) have emphasized how the faction –intended aspatron-client clusters of mutually linked officials– represents the correct unit of analysis of elite politicsin China. Since Nathan (1973), evidence supporting this interpretation has also steadily accumulated (Pye,1981; Dittmer and Wu, 1995; Nathan and Gilley, 2003; Shih, 2004; Li, 2012; Li, 2013; Shih, 2016; Meyer,Shih, and Lee 2016). The present paper follows this line of inquiry, but with special attention paid to in-72And plays a crucial role in steering economic activity in the country. See Bai, Hsieh, and Song (2016).73The study of the political economy of China has several important exceptions, but often not precisely focused on national elitecompetition. Persico, Pueblita, and Silverman (2011) in their analysis of factional politics focus on the CCP, among their variouscase studies. Less relatedly, work such as Li and Zhou (2005) focuses on the promotion profiles of provincial leaders and so doesJia, Kudamatsu, and Seim (2015). Work by Lau, Qian, and Roland (2000) models the process of reform under Deng Xiaoping andthe reform era.74Descriptive discussion most pertinent to this paper includes Li (2012, 2013). Several quantitative exceptions are discussedin Shih (2016) with respect to scholarship in East Asian studies and political science, while less recent examples include Huang(2000), Shih (2004, 2007); Shih, Adolph and Liu (2012).75At the time of writing. By 2016 the CCP membership has grown to 88.76 million.76BBC, Monday January 5, 2015: “An editorial in Monday’s flagship newspaper, The People’s Daily, says cliques are akin toparasites and are ‘harmful for both the country and the people.’” http://www.bbc.com/news/blogs-china-blog-30685782834.1. Introductiondividual incentives, supplying an inherently economic model of behavior, where “lower-level officials [...]join factions in order to secure promotions and other regime goods from powerful patrons” (Shih, 2016,p.1) and where promotion dynamics throughout the party hierarchy are microfounded and characterized. Atheoretical contribution of this paper is in the formal model of factional interaction that we present.In our model factions operate within a given party hierarchy. On the one hand, the advantage of factionsis that they provide support to their members in obtaining promotions up the pyramid. On the other, factionsallow the allocation of that support to be decided by senior affiliates, with the possibility of junior membersbeing blocked by higher ranked cofactionals keen to avoid promoting colleagues who will compete withthem for future openings. A faction member, though potentially benefiting from cofactional support, hasto bide his time and wait for the seniors in his faction to allow that support to materialize. The seniorsmake this decision based on their own career objectives, so that a junior member’s ascendancy through thehierarchy is tethered to the rise of the relevant seniors above him. Unaffiliated (neutral) politicians faceno such restrictions, and this is why neutrals can also emerge in equilibrium. Though they do not enjoyfactional support, they are also not restricted in their capacity to contest openings higher up. The analysisof the costs and benefits of joining factions is complicated by the dependence of promotion opportunities onthe factional composition of every level of the hierarchy at any point in time. This determines what kind ofopenings may arise and who is in a position to block advancement at any level, a problem that we study indetail.Our theoretical results are important in matching empirical moments in terms of factional composition,promotion rates, and the effects of changes in the factional identity of the top leadership in China. Absenthard and verifiable information, we rely on the extant discussion of Chinese elite politics to identify aminimal set of factions within the CCP. Factions have historically emerged within the CCP through closepersonal connections with prominent patrons (e.g. in the cases of former General Secretary Jiang Zemin andhis successor, Hu Jintao) to mutually foster the career prospects of affiliated cadres, and do not necessarilyrepresent specific territorial or economic interest groups (Dittmer, 1995). As we discuss in Sections 4.2and 4.3, this paper will lever only the most obvious factional links identified within the CCP, links basedon affiliation to the Communist Youth League of China (related to General Secretary Hu Jintao) or to theso-called Shanghai Gang (affiliated most prominently with Jiang Zemin and bolstered by the special statusof Shanghai in Chinese politics).Scholars such as Shih, Shan, and Liu (2010), Shih, Adolph, and Liu (2012), Jia et al. (2015) haveexplored methodologies for the imputation of factional linkages based on place of birth, university ties, andshared career profiles77. While we also focus on systematic biographical information, we remain wary ofpotential mismeasurement in the identification of factional ties, as is likely for factional affiliation basedpurely on place of birth or shared career paths. An important reason for this wariness will be evident inour statistical analysis. Based on our factional definitions and within a proximate set of party officials ofalmost equivalent rank in the same office and area (e.g. the number 1 and number 2 highest ranked partymembers in a province), we show that members of a faction (let us call it B) are virtually never paired withmembers of the same faction B at the same office. On the contrary, they are paired with members of a rivalfaction (R) in excess to what would be predicted by random chance alone. For instance, if a province hasa B faction Party Secretary (ranked number 1), the Governor (his number 2) is likely to be an R, possibly aneutral official, but most definitely not a B faction member. Thus simply sharing part of their career pathsmay not be informative of factional affiliation for CCP elite officials. In fact, our evidence shows it maymislead completely.The statistical analysis of these systematic factional cross-patterns in top CCP positions is new to theliterature and will be discussed in Section 4.4. In addition to studying these cross-factional patterns, Section77Shih (2008, p.66) discusses issues of measurement with the premise that “Despite the centrality of factions in Chinese politics,they are extremely difficult to observe in a systematic manner, especially in such an opaque political system.”844.2. Institutional Background: the CCP4.4 reports statistically significant premia in terms of promotion rates and seat allocations to a leader’scofactionals. That factions may deliver advantages to their members is a necessary condition for our model’scoherence. But the existence of precisely estimated leadership premia points also in the direction of factionsboth being reasonably identified within our analysis and of operative relevance within the CCP.We formally explore and test for the presence of additional factions. This is possible within our settingthanks to the structural econometric approach we follow. We directly bring our model to the data, obtainestimates of the primitive parameters (such as leadership premia and parameters governing the contest func-tions for promotion) and formally test our mechanism against alternatives, including mechanisms basedon pure seniority or meritocracy. Our factional model displays excellent in-sample and out-of-sample fit.We show how the estimated leadership premia in the CCP are quantitatively substantial, but quite far fromwinner-take-all levels, and that the intra-faction competition among faction members operates as a de factoendogenous dampening mechanism in slowing factional growth.Our analysis includes several counterfactuals. We model possible institutional changes within the CCP,including the effect of increased leadership premia, which may indicate a break away from the “collec-tive leadership” design envisioned by Deng. We also study the role of the identity of the top leadership,the factional role of princelings, and we try to explicitly assess General Secretary Xi Jinping’s factionalaffiliation.Besides the politico-economic literature on Chinese elite politics mentioned above, this paper speaks tothe literature on the internal organization of autocratic regimes. Francois, Rainer, and Trebbi (2015, 2016)discuss at length the importance of its connection to the expanding literature on the political economy ofdevelopment. Most related to our work (and one of the first rigorous analyses of factional politics within theeconomic literature) is Persico, Rodriguez-Pueblita, and Silverman (2011), who present a theoretical modelof endogenous factional growth and link it qualitatively to evidence from factional local politics in Mexicowithin the Institutional Revolutionary Party.78From a theoretical perspective, Dewan and Squintani (2015) model endogenous faction formation (anissue we address in our setting as well, when characterizing the decision of party members to join a faction).The authors develop a model where incentives for faction formation are ideological rather than economic (asin our setting and in Persico, Rodriguez-Pueblita and Silverman, 2011) and show how within their frame-work factions may serve welfare-enhancing purposes, limiting extremists within the party by tying them tomoderate faction leaders. Factions are also shown to facilitate information sharing and party effectivenessin their model.The remainder of this paper is organized as follows. In Section 4.2 we provide a brief institutionaloverview of the CCP. In Section 4.3 we discuss our data, operationalize factions, and provide a descriptiveanalysis of our samples. Section 4.4 produces a set of stylized facts, some novel, useful to frame andguide the theoretical analysis. In Section 4.5 we discuss our theoretical setup and Section 4.6 develops ourestimator. Our main empirical results are reported in Section 4.7. Section 4.8 presents our counterfactualexercises. Section 4.9 concludes.4.2 Institutional Background: the CCPThis section presents a brief institutional overview of the internal organization of the CCP in the reform era.It is in no way exhaustive, but only of assistance to the reader unfamiliar with Chinese politics in framing78See also Belloni and Beller (1978). Persico et al. (2011) also point out to the relevance of factional politics well beyondMexico’s camarillas or the CCP, with references to studies of factionalism within the Japanese legislature (Cox et al., 1999, 2000)and the Italian parliament (Zuckerman, 1975; Kato and Mershon, 2006; Ceron, 2015; and Laver and Giannetti 2004). Factionsin Australian politics are discussed in McAllister (1991). The US urban party machine factional structure, such as in the case ofTammany Hall, are subject of an entire and even earlier literature. See Myers (1917).854.2. Institutional Background: the CCPthe analysis that follows79.In 2016 the Chinese Communist Party, with its 88.8 million members, is one of the largest politicalparties worldwide and one of the most enduring (founded in 1921). The CCP organization is stronglyhierarchical in nature and the party reflects one-to-one the organization of the Chinese state, as typical in thearchitecture of Leninist regimes.The top of the CCP hierarchy is shared by the figures of the General Secretary of the CCP and thesecond ranked member of the CCP, which respectively assume the roles of President and Premier of theState Council of the PRC. Both leaders belong in turn to the Politburo Standing Committee (PBSC), formedby the other 5 members and which represents the set of the highest ranked politicians in China. The PBSCis an expression of the 25-member Politburo (PB), the executive body of the Central Committee of theChinese Communist Party. The Central Committee (CC) is de jure the highest political body in the CCP andcurrently consists of 205 full members and a set of 171 Alternate Central Committee (AC) members in juniorstanding relative to the full members (and without voting rights). All members of the CC and AC are rankedhierarchically. The CC and AC are elected during National Congresses of the CCP and the interim plenarysessions fill retirements or deaths, granting promotions (and occasionally administer demotions). Typically,CC members include ministerial-level officials and provincial ranking officials, including Provincial PartySecretaries (the highest CCP post in a Province) and Governor (the second ranked). It is important to noticethat Provinces tend to display a political architecture that mimics the national government and the nationalparty structure. Provincial leaders operate in the context of local party committees and local party congressesare held typically every five years. The CCP maintains a pyramidal structure, branching all the way down tothe village level and the Village Party Branch Secretary.While not all layers of the Chinese political hierarchy present nodes mapping into a diarchic struc-ture, most do, typically separating party roles and administrative roles. Examples of diarchic arrangementsinclude the presidency and premiership as the two highest ranking members of the Politburo Standing Com-mittee; the PRC Presidency (President and Vice President); the State Council (Premier and Executive VicePremier); and the top dyads at the provincial level (Provincial Party Secretary and Governor)80. We willoccasionally refer to such pairs of positions as position 1 and 2.The opportunity of entering the ranks of the CCP is closely guarded and party membership typicallyguarantees access and career opportunities beyond those available to common citizens81. For this reason, anelaborate recruitment process typically operates through the selection of successful university students andthrough family and work connections.Membership of the Communist Youth League of China (CYLC), an ancillary organization to the CCPresponsible for the youth (members are typically between 4 and 28 years of age), has traditionally operatedas an entry point in the CCP. As discussed in Li (2012, 2013), individuals with a background in the CYLCare often referred to as members of the tuanpai (i.e. Youth League [faction]) and tend to originate, althoughby no means exclusively, from the less prosperous (“red”) regions82. Li (2012) associates with the CYLC“populist” policies close to the rural poor and recent migrants to cities, as opposed to the policies preferredby more “elitist” groups comprised by CCP cadres close to former General Secretary Jiang Zemin and agroup of party officials connected to the Shanghai municipal administration. Indeed, the economic and79See also Chapter 1 in Nathan and Gilley (2003) for a less brief overview. For a comprehensive discussion of elite politics inChina see references in Shih (2016).80See Li (2014) for a discussion and examples. Other instances include the CMC (chairman and executive vice chairman),the CCP Secretariat, the NPC and CPPCC (chairman and executive vice chariman), the Supreme People’s Court. Assuming thepresence of such dyads across the whole hierarchy should be simply read as allowing for the presence of a close substitute in theparty hierarchy for any member.81The Organization Department of the CCP Central Committee on June 30th, 2016 in an official release indi-cated that 22 million Chinese residents had applied in 2015 and less than 4.5% of the applications were accepted.http://news.xinhuanet.com/english/2016-06/30/c_135478976.htm82Prominent members include current Premier Li Keqiang and former General Secretary and President of the PRC Hu Jintao.864.3. Datapolitical role of Shanghai cannot be emphasized enough in CCP internal interactions, to the point that theterm Shanghai Bang (Gang) has been often employed to identify the patronage cluster close to Jiang and tothe economic interests of the coastal (blue) provinces (Li, 2002).Whether additional factional groups besides the CYLC and the Shanghai Gang may be present withinthe CCP is unclear and disputed even among scholars of Chinese elite politics. For instance, some observerspoint at the anomaly of the exceptionally rapid careers of sons and daughters of prominent party officialsand revolutionary veterans under Mao, often referred to as “princelings”. The analysis below will discussthis specific group of CCP members in detail.4.3 DataWe combine two biographical databases of Chinese politicians. The first data source is China Vitae, whichcollects biographical information on more than 4,494 Chinese elites in government, politics, the military,education, business, and the media since 1992. Information provided by China Vitae includes gender, yearof birth, place of birth, ethnicity, colleges attended, and career trajectory. Information in China Vitae comesfrom Chinese and English language web sites in China that are supported by or affiliated with the Chinesegovernment.Our second data source is a biographical database of CC members developed by Shih, Shan, and Liu(2008), and further updated by Lu and Ma (2015). This database contains all CC and AC members from thefirst Party Congress in 1921 to the eighteenth Party Congress in 2012. This data also provides biographicalinformation and career trajectories similar to China Vitae. We focus our analysis on the period of 1956to 2014, which starts from the first Party Congress since the founding of People’s Republic of China (8thParty Congress in 1956) and ends with the most recent Central Committee (18th Party Congress in 2012),covering a total number of 1,853 individuals.We combine these two data sources to construct our estimation samples. Whenever there is incon-sistency between the two data sources, (e.g. multiple politicians in the same position in the same year),we manually check with a third source, typically official websites affiliated with the Chinese government(e.g. www.xinhuanet.com; cpc.people.com.cn). We also collect provincial population and GDP data fromChina Data Online. The anti-corruption data originates from ChinaFile and China’s Central Commissionfor Discipline Inspection (CCDI) website.Following the literature on Chinese politics (Bo, 2008; Li, 2013a; Li, 2013b), we construct four affili-ation indicators for the full sample of politicians: CYLC, Shanghai Gang, but also Military and Princelingstatus. A politician is classified as from the CYLC if he/she has held provincial and national level positionsin CYLC. A politician is classified as from the Shanghai Gang if he/she has held official positions in theShanghai municipal party apparatus, municipal government, municipal People’s Congress, and municipalPeople’s Political Consultative Conference. This again underlies the exceptionality of the Shanghai politicalmachine. A politician is classified as from the Military if he/she served as military personnel in the Rev-olutionary Era (1921-1949), or has participated in the volunteer armies to Korea or Vietnam, or served asmilitary personnel for more than half of its career after the founding of People’s Republic of China. Therestriction on the minimum time of military experience is to rule out civilian officials who work as the partysecretary of a military region for a short period of time (e.g. Hu Jintao as the First Secretary of GuizhouMilitary District from 1985 to 1988), or civilian officials chair the Central Military Commission (e.g. JiangZemin as the chairman of the Central Military Commission from 1990 to 2005). A politician is classified asa Princeling if he/she is from a prominent political family, the so called “red aristocracy” (prominent exam-ples include General Secretary Xi Jinping and disgraced former governor of Liaoning Bo Xilai). These fouraffiliations are not mutually exclusive (for example, Xi Jinping is both a princeling and an affiliated of theShanghai Gang according to our definition) and not all party members in our sample are affiliated. In fact,874.4. CCP Factional Politics: Reduced Form Resultswe allow for politicians in our sample to also be unaffiliated (neutral, indicated as N).Theoretically one could consider CYLC, Shanghai Gang, military, and princelings alternative politicalfactions. In Section 4.4 we show however than only two of these groups, CYLC and Shanghai Gang, trulyexhibit the features of political factions within the CCP. Formal statistical tests will be also developed andbrought in support of this thesis. To distinguish, we will refer to princelings and military as “groups” andCYLC and Shanghai Gang as “factions”.The military is virtually a parallel structure with limited political control, while the princelings as agroup are extremely heterogeneous and appear to operate as a set of neutral and independently powerfulactors (in fact, often times in deep rivalry among themselves, such is the case of Bo Xilai and Xi Jinping).While we will keep track of all types of affiliations in the analysis that follows, we emphasize here thatour theoretical and empirical design will separate CYLC and Shanghai Gang faction members from allother political actors, including the military and princelings, which we will deem “neutral”. Because ofthe traditional coloring associated with these two established factions, we will also occasionally refer to theCYLC as the Red faction, R, and to the Shanghai Gang as the Blue faction, B.Table 4.1 provides summary statistics of demographics and careers of 4,494 politicians who held im-portant positions in government, politics, the military, education, business, and the media in China since1992. The unit of observation is a position-individual pair. We classify the organizations into 12 categories:party apparatus, government, military, People’s Congress, Chinese People’s Political Consultative Confer-ence (CPPCC), court, procuratorate, CYLC, business, media, education, and an unclassified category. Theaverage duration of each position is about 4 years, and the age of starting each position varies from the early30s (CYLC) to the late 50s (People’s Congress). Individuals who hold these positions are predominatelymale, which reflects the large gender imbalance at the top levels of government and business in China83.Ethnicity is predominately Han, reflective of the ethnic composition in the Chinese population. The last fourcolumns provide the frequency of the various affiliations in each type of organization. CYLC members tendto work in the party apparatus and media instead of the government system.84 The Shanghai Gang is moreevenly distributed across all types of organizations. Princelings are more likely to have experience in themilitary, but are less likely to work in the legal system (court and procuratorate), potentially due to the factthat the power of the judiciary is relatively muted in China.We then turn our focus to a subset of elites, the members of Central Committees of the CCP. This is agroup of around 400 people who comprise the CCP top leaders. Table 4.2 provides the demographics and thefactional affiliation by sessions of the Central Committees. Similarly to the larger sample of elites, the CCmembers are predominantly male, in their mid-50s and mostly Han. Over the past 60 years, more membershold college or even post-graduate degrees. However, only 10 percent of them studied or worked abroad.More than 10 percent of them have worked as personal secretaries (Mishu) of prominent politicians, illus-trating the importance of personal ties in Chinese politics. Conditioning on entering the Central Committee,around 20 percent of them are promoted into higher level in the four levels of the Central Committee, andaround 50 percent will retire in next CC session. In terms of factional affiliation, CYLC, Shanghai Gang,and princelings each account for around 5 percent to 10 percent of members. The military has experienceda large downward trend, dropping from 56 percent in the 8th Central Committee to less than 20 percent inrecent years.4.4 CCP Factional Politics: Reduced Form ResultsThis section presents a set of facts on factional politics in China, the most important of which are novel, to8325.1% of CCP members were women in 2016.84This is consistent with the anecdotal discussion of Hoffmann and Enright (2008) that CYLC leaders often have experience innon-economic fields, such as party organization and propaganda884.4. CCP Factional Politics: Reduced Form Resultsthe best of our knowledge. These stylized facts are going to inform and motivate the theoretical analysis thatfollows.i) National Political Actors. We begin by arguing qualitatively that the factional affiliations we posit(CYLC and Shanghai Gang) share properties that make them bona fide large national players within theCCP and are not merely political actors representing local constituencies.In Figure 4.1 and Table 4.3 we describe the geographic distribution of members affiliated with theCYLC and the Shanghai Gang in provincial roles. As is evident, the representation across provinces is fairlybroad and not limited to a particular local area, despite a small positive correlation between the presence ofShanghai Gang and the average GDP per capita of the province. On the other hand, individuals associatedwith princelings and the military group are distributed more unevenly: princelings are more likely to holdpositions in rich costal areas – possibly due to their privileged status — while military members are moreconcentrated in poorer western provinces and places with strategic importance (e.g. Fujian, which neighborsTaiwan).ii) Cross-Factional Mix. Useful to the understanding of factional dynamics within the CCP is the studyof the peculiar factional mix which we observe when sampling the diarchic nodes pervading Chinese insti-tutional design. These are pairs of positions of similar rank and operating in close institutional proximityto each other. Table 4.4 reports formal statistical tests of the factional composition of virtually all top twoleadership posts in post-Deng China. In particular, we ask: given the factional affiliation of a politiciansitting in one of the top two leadership positions of a national or provincial organ, what is the likelihood thatthe other position will be held by a cofactional member? It turns out it is extremely low.Table 4.4 shows panel regressions of the factional affiliation of the number 1 official on the number2 official’s affiliation at the same node. The variables CY LC1 and Shanghai1 (respectively, CY LC2 andShanghai2) are dummies which equal 1 if the number 1 official (respectively, number 2) is from that factionand 0 otherwise. We will also refer to such factions through the abbreviations R,B. The sample periodis from 1992 to 2014. Columns 1-4 include all positions, and Columns 5-6 break down to provincial andnational level positions. The provincial positions include 31 provincial and municipal units (secretary andgovernor)85. The national positions include the Politburo Standing Committee (two highest ranking mem-bers), PRC presidency (President and Vice President), the State Council (Premier and Executive Vice Pre-mier), Central Military Committee (Chairman and Executive Vice Chairman), CCP Secretariat (two highestranking secretaries), NPC (Chairman and Executive Vice Chairman), CPPCC (Chairman and Executive ViceChairman), the Supreme People’s Court (President and Executive Vice President).Taking the top two leadership positions in any CCP (or PRC) organ, position 2 being filled by a R(respectively, a B) politician predicts negatively and significantly the likelihood of position 1 being filled byan R (respectively, a B) politician. The estimated negative coefficients indicate a statistically robust lowerlikelihood of same-faction pairs (R,R) or (B,B) relative to what would happen in case of pairings formingrandomly between B,R,N. Interestingly, the evidence for princelings is much weaker, in line with furtherevidence below showing their lack of behavior as an organized faction. In Table 4.5 we further show thatthere is also a statistically precise excess likelihood of matching pairs in the form (R,B) and (B,R) relativeto possible pairings with neutrals, N.The presence of cross-factional pairs exceeds significantly what would emerge by random chance alone.To the best of our knowledge these facts on systematic cross-matching within Chinese elite politics are new.An implication of this evidence is that methodologies imputing factional affiliation based solely on sharedprofessional paths may be highly deceptive, as discussed in the Introduction.iii) Leadership Premia. A crucial feature of any theoretical model of factional politics is the ability offactions to deliver resources to their members. This seems a necessary condition that our factional definitionshould satisfy, a conceptual underpinning that we must be able to verify in the CCP data in order to justify85Shanghai Municipality is excluded in the regression sample of Shanghai Gang.894.4. CCP Factional Politics: Reduced Form Resultsour approach.We will do this in what is possibly the starkest way: estimating premia in factional seat assignment andpromotion rates of cofactionals of the country leader (i.e. the PRC President and General Secretary of theCCP). Again, we are not aware of any systematic analysis of this type for the CYLC and Shanghai Gang.Table 4.6 shows a panel regression of promotion and retirement dummies on the factional affiliation ofCentral Committee members interacted with the faction of the General Secretary. The sample includes allmembers of the 8th to the 18th Central Committees (Politburo Standing Committee members are excludedfrom the promotion regression). Promotion is equal to 1 if a Central Committee member moves up in therank defined by the four levels of Central Committee (1 PBSC, 2 PB, 3 CC, and 4 AC).As is clear from the reduced form regressions, an R (respectively, a B) politician has substantially higherlikelihood of promotion when an R (respectively, a B) leader is in power. On average CYLC and ShanghaiGang members exhibit promotion rates higher by 10 percentage points relative to neutral members (ex-cluding military and princelings), as reported in Appendix Table 3. However, this result masks substantialheterogeneity. While CYLC and Shanghai Gang members’ promotion rates hover around 4 percentagepoints higher than neutrals in times where the leadership is not from an individual faction, having a co-factional leader adds 20.6 percentage points to CYLC and 19.3 to Shanghai Gang, inducing a substantial,highly significant, leadership premium in the speed at which leader’s cofactionals are promoted. Figure 4.2provides a vivid visualization of the leadership premia in promotion rates.We also perform an analysis looking at allocations of crucial posts to factional members. The dependentvariables include: the share of official positions allocated to a faction constructed following the schemeof Bo (2010) and weighted by value (we will refer to it as “power score”); the share of seats of AlternateCentral Committee members (AC); of the full Central Committee (CC); of the Politburo members (PB); andof the Politburo Standing Committee members (PBSC). These effects are reported in Table 4.7. Leadershippremia are statistically significant, between 4 percentage points higher in terms of power score shares for theCYLC and around 2 percentage points for the Shanghai Gang. These estimates are not trivial, but quite farfrom winner-take-all levels. The leadership premia in the power score can be easily observed in the simpletime series plots of Figure 4.3.iv) Anti-Corruption Campaign. As in the allocation of rewards to cofactionals through leadership pre-mia, we would also expect evidence of factional bias in the administration of punishment. We have limitedsystematic evidence in this respect, but it interestingly points in a direction consistent with the limited lead-ership premia discussed at point iii).This novel evidence comes from the factional analysis of the CCP members hit by President Xi Jinping’santi-corruption campaign (initiated in 2012 and still ongoing as of 2016). A remarkable factional balanceseems to be present in the administration of punishment, when looking at the detailed resumes of the so-called “tigers”, a code name for high-ranking party members affected by the purge86. Table 4.8 shows thatboth CYLC87 and Shanghai Gang cadres appear represented in the purged sample88 and, importantly, bothfactions are represented in shares proportional to their overall representation in the upper echelons of theCCP, and not statistically significantly higher or lower. The reader may however notice a lower, but notsignificant, representation of Shanghai Gang members, the faction most likely to be associated with Xi (if86As opposed to low-level politicians, “flies”, involved in petty corruption. Tigers directly hit by the anti-corruption purge haveincluded retired PBSC member Zhou Yongkang and retired PB member Xu Caiou.87Links to the CYLC were evident in official news releases by The People’s Daily which explicitly singled out specificsubsets of this faction, particularly "The Shanxi Gang", officials linked to Ling Jihua, a disgraced protegé of Hu Jintao.http://www.bbc.com/news/blogs-china-blog-3068578288We build a corruption dummy indicator for whether a political/military official is listed in the public anticorruption database ofthe Central Commission for Discipline Inspection and from ChinaFile. Table 4.8 shows the cross-section regression of corruptiondummy on faction affiliation of an official. The sample includes all the individuals covered by China Vitae who have not retired inthe year of 2007, the year of 17th party Congress. We dropped military personnel from the sample as the coverage of this group isrelatively limited in China Vitae.904.5. Modelat all –see Section 4.8). These results appear also completely consistent with an independent analysis of theanti-corruption campaign presented in Lu and Lorentzen (2016).v) Post-Deng era. Finally, we provide brief empirical justification for our focus on the post-Deng era.Mao Zedong and Deng Xiaoping have been often characterized as political “strong men” by many observers,as their legendary careers in the revolutionary era won them ultimate control over the military. In contrast,subsequent leaders, Jiang Zemin, Hu Jintao, and Xi Jinping, appear categorically different: civilian officerswho rose through the party hierarchy relying on their ability and connections. This structural break is evidentin the data.Underlying the symbolic retirement of Deng in 1989, we document structural changes in the wholespectrum of political elites. Figure 4.4 shows the share of power score by factions or groups in the CentralCommittees of the CCP. Post-Deng China witnesses a significant decline in the influence of the militarygroup, and a rise in factions such as CYLC and Shanghai Gang. Figure 4.5 breaks down the power scoreby four constituencies of the Central Committee: state organs, party apparatus, military, and regional gov-ernments. The pre-Deng era was ridden with volatile shifts across constituencies, with the most salientexample being the Cultural Revolution between 1966 and 1976, during which state organs and party appa-ratus were virtually paralyzed. In contrast, the post-Deng era witnessed the stabilization of power sharesfor each constituency. Despite the lack of political reform often alleged by outside observers, the above evi-dence suggests that Chinese politics evolved to a new phase in which political strongmen became replacedby factional politics after Deng89. This is the period we focus on.4.5 ModelHaving produced a series of statistical regularities pointing in the direction of a systematic role for factionalaffiliation in the organization of the CCP (and the Chinese state more in general), we now proceed withthe construction of a formal theory useful to understanding the incentive structure driving the data in thepost-Deng era.4.5.1 The Hierarchy of PositionsThere is a L level hierarchy of leadership positions, ordered from the highest level 1, to the bottom, L. Eachlevel, `, of the hierarchy has a M(`)/2 leadership nodes. Each leadership node has a pair of leadershippositions. The two positions at each node are ordered (position 1 and position 2). The hierarchy is brokenup into regions, each of which nests a higher number of smaller regions below it. Level 1, the top level, hasone node and hence two positions; M(1) = 2. It is the paramount leadership node for the country as a whole(currently, President Xi Jinping and Premier Li Keqiang). Level 2, the second layer in the hierarchy, hasM(2)>M(1) positions divided up into M(2)/2 , and so on, with the number of positions strictly increasingdown to level L. The nodes at the lowest level are the “entry” leadership positions, corresponding to the firststep in a political life that we model.Time is continuous. Each individual politician “dies” (or exogenously retires) with an instantaneousprobability, δ , which also acts as the instantaneous discount rate. Upon a politician’s demise, his or herposition opens up for replacement. A politician’s position also opens up when promoted to a position above,freeing the current spot. We assume that the flow utility from being in office is increasing in the positionwithin the hierarchy. Denote by u(`) the instantaneous utility generated at any position at level `, with` ∈ {1, . . . ,L}, so that u(`)> u(`+1). Positions within a level are ranked, but the utility flow difference is89Appendix Figure 3 shows additional evidence that age limits on Politburo members are strictly and systematically enforced inthe post-Deng era, again another sign of break toward institutional regularization.914.5. Modelsmall. Position 1 at a node at any level ` is preferred to position 2, but to reduce complexity, simply refer toeach as identically generating a flow of u(`).90Politicians cannot leapfrog levels of the hierarchy. An opening for either leadership position at a node inlevel ` is filled by applicants from the level immediately below, level `+1. The only exception is positionsat entry L (where there is no lower position). Though levels cannot be jumped, positions within a level can.Leaders can move from one level in the hierarchy to the next without having to progress through all thepositions at their level. For example, a leader at position 2 in level `+ 1 can be promoted to position 1 inlevel ` without having to first go through the intervening positions.All eligible leaders from lower positions can apply for openings. It costs an arbitrarily small amount todo so. So, if there is an opening in any node of level `, then all leaders from level `+ 1 will apply. Thewinning applicant is said to be promoted up a step in the hierarchy.4.5.2 FactionsThere are two factions, denoted B (Blue) and R (Red), and the remaining individuals are neutrals, denotedN.91 Factions exist to create promotion opportunities for their members and are organized in a hierarchy.A faction can support one and only one member applying for a single position. A faction member notsupported by his faction for an opening cannot win promotion against a supported member.92 For the timebeing, let us assume factions randomly choose whom to support amongst their eligible candidates.93When a faction holds the paramount position, the effect of promotion support is enhanced, thus increas-ing the chances of the paramount leader’s faction’s candidate winning promotion vis-à-vis the other factioncandidate and neutrals.Factions write binding “contracts” with their members determining and restricting how factional supportwill be allocated. One can never quit a faction and the contract is a quid-pro-quo. On the way up theleadership hierarchy, the faction member will be helped in obtaining positions through the support from thefaction infrastructure. If the paramount leader is from his faction, he will receive additional support. If heeventually becomes the paramount leader, the faction member will then offer the same support to the juniorsthat will follow him in return. This specific characterization of a faction aims at capturing in a stylizedfashion the essential patron-client nature of such an organization, as emphasized in Nathan (1973).Factions are organized geographically (for the sake of exposition and, to a certain extent, realism), in away that mimics the allocation of power positions within the country. The most senior faction member is theindividual with the highest leadership position in the hierarchy. Any faction member occupying a leadershipposition at level ` is senior to a faction member at level `′ > `. Faction members are designated by theirregion. A member who has a position at the top of the government is in the region of the whole country,but a member holding a position at the top of a provincial government is a member of that province andis parallel in faction seniority to a member holding a similar position in another province. This person hasfactional seniority over all individuals below him in the leadership hierarchy within his province. So, if amember of faction B is the provincial leader in province a, he has factional seniority over any member offaction B who is a village leader in province a. He does not have factional precedence over a village leaderin province c, or any other B member who is not in a.90Formally: position 1 generates ε → 0+ extra utility relative to a position 2 at all ` in the same node.91The presence of more than two factions is easily incorporated. Here, we maintain this assumption only for expositionalpurposes and in line with the empirical analysis that follows.92Or against a neutral.93We will relax this assumption below when we introduce a role for meritocracy and seniority in promotions. If a faction doesnot support a member, he could, in principle, quit the faction and contest positions as a neutral. We do not allow this, implicitlyassuming that the costs of doing this are prohibitive – factions are like the mafia: able to severely punish people who do not fulfilcommitments.924.5. ModelVetoesFactions exist to facilitate their members’ rise through the leadership hierarchy. This requires having bothas many members as possible and ensuring that members attain promotions. Each of these dimensionsincreases the probability of the faction being “powerful”, i.e. attaining the paramount leadership. But giventhat factional support for a contested position can only be given to a single faction member, an individualmay have personal incentives that run counter to his faction’s objectives. For instance, a member mayhave an incentive to block the rise of cofactionals who could dilute his own factional support in futurecompetitions for promotion. Factions guard against this by allowing for a seniority veto in allocating supportfor promotions. Support can be given if, and only if, no faction member within the region of the openingand senior to the candidate requesting support blocks it. Thus, when an opening arises in a region, eachcofactional at equivalent or higher levels of seniority to the opening in that region can veto the provision ofsupport. Vetoes importantly allow for individuals to block the rise of a member from the same faction whowould directly compete with them for factional support in a future opening94.Such localized blocking of cofactional members will be very important in determining the shape offactional allocations throughout the hierarchy and the distribution of individuals across factions. The vetoensures that a faction member never has to support someone in his region from his own faction that willdirectly compete with him for subsequent promotions. At the same time, since the veto is regional, itdoes not provide so much blocking power that a high up faction member can freeze the advancement ofanyone below him anywhere in the country. Providing these limited vetoes is the faction’s way of balancingcareer incentives while lessening the costs of intra-faction rivalry, so that sufficient faction members in thehierarchy have a good chance of attaining the paramount leadership.4.5.3 The State VariableIn principle, promotion probabilities at each point in time for each politician in the hierarchy will depend ontheir faction and the distribution of faction members across all other positions in the hierarchy. Hence, wewill need to define the full distribution of positions by faction as the state variable of the system. Denote thisby St at instant t. The state space is thus a ∑L`=1 M(`)` dimensional space, with each dimension taking oneof three values B,R,N. The state does not change if no position opens up. However, each time an openinghappens at a level `, then one individual will be promoted from `+1 to ` to fill the open position, creating anopening at `+1 leading to one promotion from `+2, and so on, until the bottom of the hierarchy L, where anew politician enters and chooses his faction. Thus a single opening will lead to a cascade or, what we call,a “chain” of promotions. We assume that these chains occur instantaneously, and if at least one individualmoving in a chain replaces an individual from a different faction, then St changes.94Vetoes can be exercised for a promotion anywhere below in the hierarchy - as long as within one’s region of pertinence.However, a politician at ` has no interest in vetoing any co-faction member below `+1. He can always veto them if, and when, theyget to `+1. If vetoes cost even an arbitrarily small amount, they will not be exercised for promotions up to any level lower than `,else they may be wasted (a politician may be promoted to `−1, potential rivals may retire, etc.). The single exception is where allpoliticians below are cofaction members. According to the model, a politician above would never let this happen and would havevetoed the rise of one of these cofactionals to avoid such a situation and ensure there is at least one individual below who can bepromoted to his accompanying position not from his faction. Reassuringly, this is observed at all levels and for all periods in thedata.934.5. Model4.5.4 Paramount Leadership and ContestsIn a competition for promotion with one member supported from each faction and one neutral politician, theprobability of winning promotion, for a I faction member is given by the following contest function, W (I):W (I) =iβ +ρ+η, (4.1)where i = β , i f I = B;i = ρ, i f I = R;i = η , i f I = N.β ,ρ, andη are parameters determining the strength of faction members in the contest function. Since afaction can only support a single member, the relative value of faction membership for a single politician,compared with being a neutral, depends on both the size of these parameters and the endogenous number ofeligible candidates from that faction. Additionally, having the paramount leadership position in your factionhelps getting a promotion for the faction’s supported candidate. If the paramount leader comes from factionB, we allow β l > β , and if from faction R we allow ρ l > ρ , thus incorporating leadership premia in themodel.95Neutrals contesting a position operate as a somewhat disorganized faction. The overall likelihood of aposition going to a neutral is unaffected by the number of neutrals contesting a position, provided there isat least one. Their total contest weight function is η . This treats neutrals symmetrically to factions and canbe thought of as a proportional diluting of the neutral support in the same way a faction’s support would bediluted were they to forward multiple candidates instead of one.Promotions and Factional DistributionsThe hierarchical structure of positions within the party is taken as given and constant over time.Promotions arise to fill openings occasioned by a death/retirement or other promotions. As alreadyexplained, a single death can have many knock-on effects. At level 1, the instantaneous probability of anopening arising at any position is δ . Since this is the highest level we observe, only death/retirement removesthe top leader. However, the instantaneous probability of an opening arising at a post at level 2 comprises thedeath hazard δ , plus the probability that there was an opening at level 1 and the individual at that level 2 postascended to level 1 to fill it. This probability of promotion can, in principle, depend on both the factionalaffiliation of the individual at the post at level 2 and the faction of the individual at the post partnering theopening at level 1. Similarly, the instantaneous probability of an opening at a post at level 3 is δ plus theprobability that the individual at the post at level 3 ascended to an opening at level 2 in the hierarchy, andso on. In the estimation Section 4.6 that follows these knock on promotions, or promotion chains, will beexplicitly computed.Let pJtI (`) denote the probability that an I faction member at level ` gets promoted to an opening pairedwith a J at level `− 1 at instant t, for I,J = B,R, or N. Let It(`) denote the number of positions held byfaction I at level `, at time t for I = B,R, or N. By definition M (`) = Rt (`)+Bt (`)+Nt (`). Since theinstantaneous arrival rate of death is δ at any position, there are, in expectation, It (`)δ deaths arriving at aposition paired with an I at level ` each t, and M(`)δ at level ` in general at each instant.Let δ tI (`) denote the instantaneous arrival rate of promotions for an I politician at level `. Let δptI (`)denote the instantaneous arrival rate of a promotion for a politician sharing a node with an I politician at95We allow for the possibility of no factional advantage, which might be especially likely at low levels of the hierarchy wherethe reach of the paramount leader could be muted. Note that it is also the case that a neutral’s ascension to the paramount positiondoes not advantage neutrals down the hierarchy.944.5. Modellevel ` at time t. Consider first the simplest case, which is a promotion from level 2 to the top of the hierarchy`= 1. Since there are, in expectation, It(1)δ openings arriving for a position paired with an I due to a death,and since at level 1 there is no other way for an opening to arise, the instantaneous arrival of promotion fora I from level 2 is:96δ tI (2) (4.2)= Rt (1)δ × pRtI (2)+Nt (1)δ × pNtI (2)+Bt (1)δ × pBtI (2).We can now similarly compute the arrival of promotions from level 3 to level 2. Intuitively, the possi-bility of these arises when either a leader at level 2 dies, or is himself promoted to level 1, which in turndepends on a death at level 1, as specified in equation (4.2). Using these, we can compute the instantaneousarrival of promotions for an I from level 3 at t as depending on the probability of a position paired with anI being promoted or dying. The instantaneous death arrival of such an individual is δ , the probability of thepaired partner being promoted is δ ptI (2) for each of the factions I at level 2 at time t, hence:δ tI (3)= Rt(2)(δ +δ ptR (2))× pRtI (3)+Nt(2)(δ +δ ptN (2))× pNtI (3)+Bt(2)(δ +δ ptB (2))× pBtI (3).Similarly, continuing down the hierarchy, we have for any level ` > 2:δ tI (`) (4.3)= Rt(`−1)(δ +δ ptR (`−1))× pRtI (`)+Nt(`−1)(δ +δ ptN (`−1))× pNtI (`)+Bt(`−1)(δ +δ ptB (`−1))× pBtI (`).(4.3) explicitly shows that the arrival rate of I promotions at level ` depends not only on the distributionat level `− 1, i.e. on Bt(`− 1),Rt(`− 1),Nt(`− 1), but also, through each of the δ ptI (`− 1), on Bt(`−2),Rt(`−2),Nt(`−2). A convenient feature of our model specification is that higher levels of the hierarchyenter recursively, allowing the computation of arrival rates for all I factions all the way down the hierarchy.Let us now consider the explicit form of the pJtI (`) using the contest function (4.1). We begin by assum-ing that the other faction J 6= I will support one of its members for the position as well.In this case pJtI (`) is determined as follows:pJtI (`) ≡1It (`)×ii+k+k′ i f Kt(`), K′t(`)> 0,ii+k i f Kt(`)> 0, K′t(`) = 0,ii+k′ i f Kt(`) = 0, K′t(`)> 0,1 i f Kt(`), K′t(`) = 0,(4.4)where {It , i} = {Bt ,β},{Rt ,ρ} or {Nt ,η};{Kt ,k} 6= {It , i},{K′t ,k′};{K′t ,k′} 6= {It , i}.Note that J (the faction of the politician that the opening at level `−1 is paired with) does not enter directlyinto the probability of winning a promotion contest. But this is because specification (4.4) assumes that if96This expression uses the fact that in continuous time simultaneous hazards do not arrive. That is, we put zero weight on theprobability of a death opening occurring at the same instant in two positions.954.5. Modelmembers of another faction are present, one of them will always be supported in the contest for the position.As we now demonstrate, this will not always be the case, which will in fact simplify the expression aboveconsiderably:Proposition 1. i) A politician from faction J at level ` will veto the support of a cofactional member ascend-ing to his level from `+1 at t if there are members of both I 6= J and neutrals, N, at level `.ii) If there are no members of faction I at level `, a politician from faction J 6= I at level ` will vetoa member of his own faction from `+ 1 at t if the number of cofactional members at level ` is such thatJt(`)< j+ηi where i = β , j = ρ if Jt = R and i = ρ , j = β if Jt = B.Proof. All proofs are in Appendix.If both R,B types are represented at a politician’s level, he will gain by vetoing the ascension of acompetitor from his own faction, as this increases the probability that his faction will support him for asubsequent opening at the level above. However, if all other factions are not already present, then he facesa trade-off. By vetoing a cofactional’s promotion the party member still improves his chances of gainingfactional support. But he also increases the chance that a member of a rival faction, which was not alreadypresent, gains entry to the group of competitors. This lowers the chances of him winning promotion condi-tional upon receiving the support of his faction. The sufficient condition in the statement of Proposition 1ensures that the former effect dominates the latter. From now on, we proceed under the assumption that thesufficient condition for vetoes holds, so that we continue to see them throughout our observations. We willverify that this is indeed the case in the data, so we do not dwell on weaker necessary conditions for vetoesto hold further.Following Proposition 1, vetoes generate a large amount of structure to the pattern of openings – meaningno two cofactional members will ever be paired at the same node. We have already verified in Section 4.4that this is, in fact, a systematic feature of the data. Moreover, the prospects of promotion at any node dependnot only on the distribution of openings immediately above, but also on the distribution of openings furtherup, as these determine the chances that a politician immediately above will himself be promoted. Promotionchances at all levels are affected by the full distribution of positions above. We can compute this explicitlyusing the recursive structure of the δ tI (`) terms and our results on vetoes.Proposition 2. The instantaneous arrival rate of promotions at each level of the hierarchy is as follows.Let ItB = 1, iff Bt(`)> 0 and ItB = 0, otherwise; ItR = 1, iff Rt(`)> 0 and ItR = 0, otherwise; ItN = 1, iffNt(`)> 0 and ItN = 0, otherwise.For an N member:δ tN(`) =ηNt(`)×Rt(`−1)(δ +δ ptR (`−1))ItBβ +η+Nt(`−1)(δ +δ ptN (`−1))ItBβ +ItRρ+η+Bt(`−1)(δ +δ ptB (`−1))ItRρ+η .For a B member:δ tB(`) =βBt(`)×Rt(`−1)(δ +δ ptR (`−1))β +ItNη+Nt(`−1)(δ +δ ptN (`−1))β +ItRρ+ItNη .964.5. ModelFor an R member:δ tR(`) =ρRt(`)×Bt(`−1)(δ +δ ptB (`−1))ρ+ItNη+Nt(`−1)(δ +δ ptN (`−1))ItBβ +ρ+ItNη .For each one of these expressions we can see the negative dependence on the prevalence of one’s ownfaction members. Take for example the last expression for R. The greater the number of other R’s at level `at t, the more diluted is an R’s support (i.e. the lower the probability that any given R member will be chosenby the faction as the one to be supported), as per Rt(`) in the denominator. Further, the more frequent theR’s at level `−1 the harder it is to get an opening for which an R at ` will not be vetoed (e.g. at the extremeif Rt(`− 1) = Mt(`− 1), then δ tR(`) = 0). This is true for all levels of the hierarchy from the recursion ofthese equations.The proposition highlights the possible down side of factional affiliation. Though factions have the po-tential to provide support for promotions such support is decided by cofaction members sitting above onein the hierarchy. They will never let a junior member contest with them for their own future promotions so,in a sense, the rise of the junior is tethered to, and thus depends upon, the rise of his cofactional seniors. Ifthey do not rise, then, not only do they not generate the extra support that comes from the paramount lead-ership, they actively block their own juniors from ascending in their place. In other words, the propositionformalizes a form of natural pecking order in the factional structure, a feature that appears realistic in largeorganizations.Finally, note that each statement of δ tI in Proposition 2 ignores the effect of a faction’s holding of theparamount leadership on promotion (i.e. il). Effectively δ tI is written for the case of an N in paramountleadership. In the Appendix we state the full set of δ tI conditional upon paramount leadership affiliation.4.5.5 EntryEntry into the hierarchy of political positions occurs only at the lowest level, L. An entering politician atinstant t decides which faction to join when starting his politician career, or to contest as a neutral, and basesthis decision on the discounted expected utility he will receive via each one of the options. He maximizeshis discounted expected utility stream:V t =ˆ ∞te−δ sυsdswhere υ t is the instantaneous utility at t. We formally consider this decision here. Recall that u(`) denotesthe politician per instant payoff to holding a position at level `∈ {1,L} in the hierarchy. So that if a politicianholds a position at ` at instant t then υ t = u(`). Define the corresponding value function for a politician oftype I = B,R,N at level ` at instant t by, V tI (`). This is related to the promotion probabilities, δ tI (`) , via theBellman equation:δEtV tI (`) = u(`)+δ tI (`)Et[V tI (`−1)−V tI (`)](4.5)The expectations operator appears in the expression because the value of being a type I politician at `depends on the instantaneous probability of being promoted to level V tI (`− 1). Though this is known atinstant t, via δ tI (`) , the value of being at this higher level in turn depends on the evolution of δ tI (`) . Theevolution of these δ tI (`) promotion probabilities themselves depend on the state of the system, St , which ischanging continuously in a stochastic manner due to deaths, openings, and promotions occurring throughtime via the contest function (4.1).974.5. ModelThe entering politician at t chooses the faction with the highest expected utility stream:supI∈{B,R,N}{EtV tB(L),EtV tR(L),EtV tN(L)}. (4.6)After entry, since a politician is fixed in his faction from then on, his choices are simple. He will apply forall promotions to which he is eligible, and he will veto according to Proposition 1. We consider the moredifficult problem of the initial entry decision (4.6) now.4.5.6 Equilibrium BehaviorEntering politicians will choose to enter the faction (or remain neutral) yielding the highest expected utility,which implies choosing the faction guaranteeing in expectation the fastest progression through the hierar-chy. The most immediately relevant information for the agent will be the arrival of promotions if he/sheregisters as a I politician from level L to L− 1, but one cannot specify, a priori, the relative weight an en-tering politician puts on the chances of being promoted at higher levels of the hierarchy compared to lowerlevels. Perhaps politicians care little about regional promotions, that occur early in their career, but greatlyabout promotions from the province to the central government. Conversely, politicians may put substantialvalue on their immediate entry prospects. Note that, indirectly at least, the relative performance of fac-tions at higher levels already enters into a politician’s evaluation of promotion at the lowest level, L, sinceopenings immediately above depend negatively on the frequency of cofactional politicians all the way upthe hierarchy; as discussed above after Proposition 2. At any point in time this valuation will depend onthe full distribution of positions higher than the politician, that is on St , the high-dimensionality state spaceof the system. Without mapping the full form of expected hierarchy evolution, it is not possible to com-pute the value function V tI (`) analytically. However, it is possible to establish a sufficient condition underwhich optimal entry ensures that along any time path all factional types and neutrals will be observed inequilibrium:Proposition 3. With M(`) large enough for all `, any equilibrium necessarily involves politicians in factionsB,R, and N.Intuitively, with sufficiently many openings at all levels of the hierarchy, the value of entering via afaction (or as a neutral) that is not already present will eventually outweigh even the largest parametricdisadvantages of that faction (or being a neutral). That is, for example, even if β  ρ (so ceteris paribus it isbetter to enter as an R than a B), if there are sufficiently many positions in the hierarchy, a large number of Rmembers and Proposition 1 will imply that the expected promotion rate will be faster if entering as a (rare)B member over entering as (one of the many) R. Thus, though we are not able to fully characterize optimalentry in an equilibrium, the sufficient condition of the proposition ensures that any equilibrium distributionof positions that we do observe will feature both factions and neutrals.4.5.7 From Model to DataOpenings in the hierarchy occur at any point in time via the functions in Proposition 2. Other than throughthe effect of time on the changing distribution of factions across the hierarchy St , which the model explicitlyaccounts for, the process leading to openings occurs independently of time (conditional on St)97.Treating openings this way amounts to assuming that openings are independent events caused by ex-ogenous factors, each triggering a chain of knock on effects. This assumption may be violated at the timeof Chinese Communist Party Congresses, when there appears to be a large number of shuffles at different97In what follows below we will dispense with the time index t for the empirical analysis.984.5. Modellevels of the hierarchy observed in a way that seems simultaneous, not sequential. Indeed, for the most part,the data is observed at low frequency, i.e. at each CCP Congress T,T +1, ... This implies that the promotionchains that our model postulates are not fully observable, so simulation methods will be necessary to linktwo subsequent ST ,ST+1.To operationalize the model in our specific empirical setting, we will assume that the simultaneity ob-served in exits and promotions reflects a particular structure, as follows.First, we purge all individuals from all positions that we observe leaving the data in between snapshotsT,T + 1, ... That is, all individuals who are no longer present between times T and T + 1 are assumed tohave retired at some point between two Congresses.Second, openings are filled through a sequence of promotion chains. Each chain starts with the highestranked exit in the sample and selects politicians to fill in the knock-on openings sequentially. This continuesuntil all the exits and promotions between ST and ST+1 are accounted for and all positions have been filled.Because there are many sets of promotion chains that can rationalize the observed openings in the data,Section 4.6 shows how simulation methods can be used to transparently address this issue in practice.Third, for positions for which there is no explicit dyadic structure in the data, we draw at random apaired politician from the set of potential matches at the level at which the promotion occurs.4.5.8 Discussion of the ModelBefore moving to the estimation of the model, we offer here a brief discussion of an alternative modelingchoice and justify our specific line of reasoning empirically.Perhaps the best alternative to our individual career concerns model is a model that views the allocationof positions as the outcome of factional bargaining. In such a model the faction, as opposed to the individualpolitician, is the decision maker, and factions negotiate with each other over the allocation of positions in thehierarchy. Negotiations would favour the faction holding the paramount leadership position, and could thuseasily exhibit the patterns of increased representation at all levels with leadership of a faction. The relativeoverall balancing could also be supported as an equilibrium outcome that ensures peace. If a leader comesfrom faction B he is not willing to completely expropriate faction R because he fears dissent from R. Dissentin extreme cases could take the form of revolt that would destabilize not just his own position, but, in thelimit, the overall hold of the party. So positions could be still allocated to the other faction, as a price forpeace. Reciprocally, the other faction might show similar restraint if it ascended to the paramount position.Anticipating this, the current leader would have further incentives to be moderate and inclusive in allocatingpositions.Problems arise for this alternative story when the actual distribution of positions — and not just theiroverall number — is scrutinized further, as done in Section 4.4. There we observed that a pronounced patternin the data was the omission of (B,B) and (R,R) pairs at leadership nodes. R members are more likely tobe accompanied by B members than by N members and much less likely to be accompanied by anotherR — which is extremely rare. Why? One explanation consistent with factional negotiations is that eachfaction fears that the other faction may gain control of the node. If a B is in place, placing an R alongsidehim ensures that the B members do not gain permanent control of the node. But this sort of concern doesnot seem likely, as there does not appear to be evidence of such permanent nodal control in the data. Weobserve shuffling of cadres occurring regularly for Provinces for instance. There does not seem to be lockin of factions to posts. B members are replaced by R members at a node with the R subsequently replacedby another B. This additional evidence is available by the authors upon request.But shuffling could itself be the strategy that factions employ to ensure that control does not get held toostrongly. We may see B members replaced by R members in order to ensure that the B members do not holdthe position at the node too strongly. But if this is the case, a further puzzle arises. A process of shuffling —though able to easily explain (B,R) nodes — would not especially favour these. We should also regularly994.6. Maximum Simulated Likelihood Estimationsee (B,B) nodes and (R,R) nodes, which are then replaced by (R,R) and (B,B) nodes immediately after.If shuffling is used to avoid entrenchment, then there is no reason that intertemporal sharing of the nodesshould not be sufficient to achieve this. There should be no particular reason to see the proliferation of (B,R)nodes that we observe in the data.This seems to be the single most difficult fact to explain with a factional balancing model. Our individualcareer-concerns model explains this directly. In our view, the model we develop has a further advantagerelative to a model that treats the factions as bargaining parties in that it sets the individual as the decisionmaker in a microfounded way.4.6 Maximum Simulated Likelihood EstimationThis Section describes our estimation methodology. Define Y the observed data on career outcomes (i.e.promotions, exits, etc.) between two Congresses T and T +1 and X the observed data on the hierarchy plusa set of individual characteristics (i.e. X includes factions and position within the hierarchy/level ST , plusindividual covariates).We define k as a set of promotion chains, so that k = {k(1),k(2), ...}, where each chain k(c) of politi-cians (say, s0,s,s′,and s′′) is simply a set of politicians each belonging to different, but adjacent hierarchicallevels `, whose promotions were triggered by the exit of the highest ranking one of the chain (e.g. whens0 dies or retires at ` = 1, s is promoted from ` = 2 to ` = 1, then s′ replaces s at level ` = 2, and then s′′replaces s′ at `= 3).A chain starts from an opening at level `−1 and involves promotions from ` all the way down to L.98We impose that each politician promoted in the data belongs to exactly one chain and that each changebetween ST and ST+1 is part of at least one chain k(c). (A politician promoted by two levels between T andT + 1 will need to belong to two separate promotion chains.) Let C = #(k) be the number of promotionchains in set k.The unconditional likelihood of observing Y given X is:f (Y |X) = Ek [ f (Y |X ,k)] .Define Yk(c) as the set of career outcomes pertinent to the individuals involved in promotion chain numberc of k. Because the structure of the political hierarchy will change once a promotion chain is realized (i.e.the interim S will change), positions within the hierarchy/level and factional affiliations at all levels Xk(c)need to be modified after each chain k(c) is realized.The conditional likelihood upon the realization of a set of promotion chains k happening over time isgiven by:f (Y |X ,k) =C∏c=1f (Yk(c)|Xk(c),k).The likelihood contribution f (Yk(c)|Xk(c),k) of a chain k(c) of promotions initiated at `− 1 involvescomputing the conditional promotion probabilities of all individuals involved in k(c) at the various levels,down to L. A promotion from level ` to level `−1 to be paired to a politician K = R,B,N is a random eventdistributed over a discrete support formed of M(`) points (individual politicians), B(`) of which occurringwith probability pKB (`), R(`) occurring with probability pKR (`), and N(`) occurring with probability pKN (`).(We omit time indexes as they are unnecessary here.)98Plus a new entry at the lowest level, which we do not model, as per our discussion of Proposition 3. The entry choice is notnecessary for estimation and all parameters are identified without its addition.1004.7. CCP Factional Politics: Structural ResultsGiven the independence of the promotion events across levels, the construction of this likelihood isstraightforward. Let I` be the faction of the individual belonging to k(c) at level ` and J`−1 be the faction ofthe individual with which s/he is paired when promoted to level `−1:f (Yk(c)|Xk(c),k) = δL∏l=`pJl−1Il (l) .Going back to the example above of a chain of politicians s0,s,s′,and s′′ belonging to factions N, R, and Brespectively, and assuming they all happen to get paired with N-type politicians, the likelihood contributionof this chain is:f (Yk(c)|Xk(c),k) = δ × pNN (2)× pNR (3)× pNB (4)where each probability pJ`−1I` (`) is computed based on Xk(c), ordered from the top promotion to the level Lpromotion, as imposed by the sequential nature of the promotions comprised in each chain.The Maximum Simulated Likelihood (MSL), for given number of simulated sets of promotion chainsRK ,99 is:f (Y |X) = 1RKRK∑r=1C∏c=1f (Ykr(c)|Xkr(c),kr).This is the estimator that we employ.4.7 CCP Factional Politics: Structural ResultsThis section presents MSL estimates of the model and sample fit assessments. The sample includes all themembers of the 14th-18th Central Committees in the post-Deng era. The simulation procedure in Section4.6 was first implemented in a series of Montecarlo simulations and successfully probed for: i) identificationof the structural parameters; ii) sensitivity to misspecification in the number of factions; and iii) sensitivityto misspecification in the contest function we use100.We begin our analysis with the most parametrically parsimonious model possible, one where we nor-malize η = 1 and the two faction parameters {β ,ρ} are estimated on top of a single leadership premiumλ , defined as λ = β l/β = ρ l/ρ . The MSL results for this model are reported in Column 1 of Table 4.9.The estimated contest function parameters are 0.045 and 0.029 for CYLC and Shanghai Gang respectively,which are close to the average share of seats in the Central Committee. The estimated leadership premium λis 2.553, implying that a faction candidate is more than twice as likely to be promoted when the paramountleader is from the same faction. The magnitude of the leadership premium is consistent with the reducedform evidence in Table 4.6. All parameters driving the promotion process across factions are preciselyestimated.Because it may seem restrictive to assume a common contest function across all levels of the CCP topechelons (which include heterogeneous layers in both size and jurisdiction, such as the top CCP positionsand the PBSC, PB, CC, AC), Column 2 in Table 4.9 allows for level-specific parameters {βk,ρk}k=H,Lfor the PB and higher versus CC and lower. The parameter estimates show that faction affiliation helpssignificantly more at higher levels than that at lower levels within the CCP: the estimated contest functionparameters reach 0.162 and 0.193 at the PB and higher for CYLC and Shanghai Gang relative to CC andAC levels of 0.041 and 0.022.99We employ 100 simulated chains sets for each CCP National Party Congress.100All simulation results are available upon request.1014.7. CCP Factional Politics: Structural ResultsOne may also wonder whether the leadership premium differs across factions. Column 3 explores thispossibility by allowing for faction-specific leadership premia {λR,λB}. The parameter estimates show thattwo factions have very similar premia (both are between 2 and 3). The improvement of log-likelihood isnegligible, indicating that the two factions operate in a similar fashion. This result is also consistent withthe reduced-form evidence in Table 4.6.Column 4 in Table 4.9 combines both level-specific parameters {βk,ρk}k=H,L and faction-specific lead-ership premia {λR,λB}. We conduct Likelihood Ratio (LR) tests for model 1, 2 and 3 against model 4(numbering indicates the Column of reference). LR tests reject model 1 and 3, which impose a constantcontest function across levels, against model 4, but do not reject model 2, which imposes a constant leader-ship premium across factions. In the following analysis, we will thus use the more parsimonious model 2 asour benchmark and refer to it as the baseline faction model.Figure 4.6 provides a visual representation of the factions’ seat shares by level as predicted by the model.The five bars represent the five levels of the Central Committee (the top two CCP positions, PBSC, PB, CC,and AC). The blue, white, and red parts represent the seat shares of the Shanghai Gang, Neutral, and CYLCrespectively. The left panel is the data, while the right are the predictions of our baseline faction model.Our baseline faction model successfully replicates the distribution of factions across different levels of thehierarchy: faction members are relatively scarce in the lower levels, but become increasingly concentrated inthe higher ones. This is related to the increasing contest function parameters estimated above, which implyan increasing advantage of factional affiliation as one progresses up the hierarchy. Notice that our model alsocaptures the inertia of the factional composition of the various levels over time evident in the data thanks tothe slow percolation of factional members up the hierarchy. The intuition is that promotions and retirementsoccur gradually over time. It takes time for a faction leader to grow his inner circle from the bottom of thehierarchy up. Interestingly, such dynamics can function as checks and balances on an incoming paramountleader. When a new leader first assumes power, he is likely to be surrounded by members from rival factions.There is also anecdotal evidence in line with this finding: Jiang himself once described his first few yearsas the General Secretary “as standing on the brink of a deep ravine, or walking on thin ice”101. Bo (2004)also suggests that the Shanghai Gang continued to exert strong influence in the first term of Hu Jintao. Thisfinding will be particularly useful in understanding the upcoming second term of General Secretary Xi,expected by many observers to gain greater clout relative to his first term in office102.Our faction model also provides insights for the dynamics of power transition between factions. Figure4.7 plots the aggregate share of promotions of each faction over time103. The share of promotions is definedas the ratio between the number of promotions for a faction and the total number of promotions. Again,the fit of the model is good. Figure 4.7 points also to a more subtle implication of our model: there areno discontinuous drops in the share of promotions of the paramount leader’s faction right after he retires.When Jiang Zemin retired after the 15th Party Congress, a large share of the Shanghai Gang continued to bepromoted to the 16th Central Committee. The pattern was repeated at Hu Jintao’s transition to Xi Jinpingat the 18th Party Congress. In reality there is uncertainty over the precise point at which the influenceof the incoming paramount leader eclipses that of the departing incumbent and this influences promotionrates. Scholars have suggested that Deng retained considerable influence well after formal retirement in1989; Jiang maintained informal and formal military oversight after stepping down as General Secretary.A retiring paramount leader may continue to shape the composition of the next Central Committee. Suchintricate dynamics are captured by our simulation approach that draws different paramount leader transitiondates across multiple simulations, smoothing out sharp discontinuities around the official power transitiondate.101See Kuhn (2005).102The 19th Congress is currently scheduled in the Fall of 2017.103A more detailed breakdown by level of the Central Committee can be found in Table 4.11.1024.7. CCP Factional Politics: Structural Results4.7.1 Adding Individual CovariatesSo far we have assumed that faction members are selected to challenge a post randomly within a factionand level (modulo vetoes, of course). We can easily add individual characteristics, Z, to the within-factionselection process as well. Consider each row of the matrix Zs to be a vector of characteristics for politicians. Define qI,s (`) as the probability that s of faction I is selected as the candidate of this faction at level `,also define AI (`) as the set of the members of faction I at level `. We assume a within-faction selectionprobability of the logistic form104:qI,s (`)≡ exp(γZs)∑s′∈AI(`) exp(γZs′).Therefore, the probability of winning promotion can be rewritten as qI,s (`)×W (I). Notice that ourbaseline faction model is nested in this formula by setting coefficients of individual characteristics, γ , to0. In this case we get back our random within-faction selection probability, qI,s (`) ≡ 1I(`) . We refer to theabove model as the faction model with individual characteristics.The parameter estimates are reported in the Column 2 of Table 4.10. Comparing with the baselinefaction model in Column 1, we see a reasonable improvement in model fit measured by log-likelihood. Atthe same time, however, we observe little change in the estimates of the parameters for the contest functionand the leadership premium, suggesting that these parameters are indeed more related to the technology offactions than to individual covariates omitted in the baseline model. Examining the estimated coefficients ofindividual characteristics, we find that being a princeling or a male increases the probability of promotion,while having a graduate degree or being an ethnic minority hurts. The effect of age is non-linear: it has apositive effect at first, but eventually negatively affects promotion chances, in line with previously observedhard age limits enforced within the CCP.4.7.2 Alternative ModelsGiven our main specifications, we are equipped for both in-sample and out-of-sample fit analysis of ourstructural model. It is useful in this respect also to present some alternative benchmarks to which we cancompare our model’s performance. First, we can use as the simplest alternative a model based on randompromotion. This is done by setting:p(`)≡ 1M (`).Second, we implement a pure seniority-based promotion mechanism, setting for politician s:ps(`)≡ ϕ(ages)M (`) ,with ϕ(.) a (third order) polynomial in age105.Figure 4.8 provides the scatter plots of model predicted shares of promotions by Party Congress and bylevel of the CCP against the data106. Our models (baseline faction and faction with individual characteristics)handily outperform both the random and the seniority models: the predicted shares by the faction models line104Since our data only includes the top 5 levels of the party hierarchy (President/Premier, PBSC, PB, CC, AC), individual char-acteristics of the potential candidates eligible for promotion to AC are not always observable to us. As a result, we assumewithin-faction selection is random below the AC level.105For the seniority and random model, we calibrate the probability of entering AC using the average share of each faction in theCentral Committee.106We combine level 1, 2 and 3 because there are two few observations in the first two levels.1034.8. Counterfactuals and Model Analysisup with the data nicely along the 45 degree line, whereas the shares predicted by random and pure seniority-based promotion models appear completely flat. This result is independent of which Party Congress weconsider. More concretely, with only five structural parameters our baseline faction model reduces the meansquared errors of the predicted shares of promotion by more than 80 percent comparing to the randommodel, and more than 70 comparing to the seniority model. The right panel of Table 4.10 conducts formalspecification tests. The Vuong statistics reject the random and the pure seniority-based promotion modelsagainst the faction model with individual characteristics107.What about political meritocracy? Because the CCP promotion model is by many referred to as a strictlymeritocratic mechanism (Li and Zhou, 2005; Bell, 2015) and there is substantial debate as to whethersystematic assessment of cadres based on economic performance plays a role in the CCP, we test our modelagainst this third “purely meritocratic” mechanism. First, in order to find a suitable measure of performance,we need to restrict our analysis of promotions to provincial leaders in the Central Committee. We associatethese prominent provincial CCP cadres with the economic performance (in terms of real GDP growth overtheir tenure) of their Province of service –precisely as in Li and Zhou (2005) and Jia et al. (2015)– and usethis as a (admittedly rough) proxy for overall performance. Graphical evidence of the performance of ourmodel is reported in Figure 4.9108. In the appropriate subset of promotions (i.e. those for which performancemetrics are available), our model performs better than the purely meritocratic model in terms of sum of meansquared errors, which is reduced by 35 percent.We further examine the out-of-sample fit of our model. Specifically, we re-estimate the model usingonly the 14th-17th Central Committees and predict the shares of promotion of each faction at 18th CentralCommittee. We compare the model predicted share of promotion with the actual data in the scatter plotin Figure 4.10. Our faction model again outperforms random and seniority-based models in terms of out-of-sample fit. The reduction in mean squared errors is 77 percent and 69 percent comparing our baselinefaction model to the random model and seniority model respectively, reassuring us of its robustness.4.8 Counterfactuals and Model AnalysisWithin our econometric framework we can explore a set of counterfactual exercises and present an additionalquantitative analysis of several questions relevant to the study of Chinese political economy.4.8.1 Forgoing Collective LeadershipWe begin by exploring an historical counterfactual on leadership premia in the post-Deng era. Our modelexplicitly recognizes such premia, but a wealth of anecdotal discussion in Chinese politics (and the empiricalevidence of Section 4.4) suggests them to have been curtailed in the post-Deng era. This peculiarity ofthe post-Deng Chinese system, the emergence of so-called “collective leadership”, has been frequentlyrecognized in the literature. It is often indicated as the main structural break from the strongman politicalequilibria thought to have prevailed under Mao Zedong109 and the paramount leadership of Deng Xiaoping(Tsou, 1995; Fewsmith, 2001; Shambaugh, 2008). This exercise is also useful in perspective of the present,latent changes in Chinese politics. Scholars like Nathan (2016) suggest President Xi may be “overturning107The pure seniority-based model outperforms the baseline faction model in terms of log-likelihood. However, this is driven bythe fact that only 10% of the politicians have factional affiliation. After we include individual characteristics in the factional model,the pure seniority-based model is easily rejected by the Vuong test.108In this scatter plot, we do not break down the share of promotion by level because of the small number of observations in thesubset of provincial leaders.109“During the Maoist era, factions were ideologically as well as personally defined, and remained fiercely loyal in what couldbecome a winner-take-all game.” Dittmer (2004, p.18)1044.8. Counterfactuals and Model AnalysisDeng’s system”, as he “has taken the chairmanship of the most important seven of the twenty-two leadingsmall groups that guide policy in specific areas” and “tightened direct control over the military”.Here, we will ramp up the limited role played by leadership premia in factional representation in Chinaand present a counterfactual of what would have happened under heightened winner-take-all type factionalcompetition. We run the model with twice as high a leadership premium λ .Results are reported in Figure 4.11. A more detailed breakdown by level can be found in the third panelof Table 4.11. The counterfactual is implemented by simulating for each Congress T the share of promotionof each faction to the following Congress T + 1. Under the Jiang Zemin (Shanghai Gang) presidency,openings in the Politburo and the Central Committee are filled with more of the top leader’s cofactionals.Under the CYLC leadership of Hu Jintao, numbers would have been comparable, swinging in the oppositedirection with more tuanpai members promoted. The magnitude of the increase in the shares of promotions,however, is less than the increase in the leadership premium. The dampening effect emerges from thefactional veto mechanism detailed in Proposition 1. As members of a faction become crowded at a certainlevel `, new promotions from the same faction are more likely to be blocked by their cofactionals out of theirown career concerns. Indeed, as shown in Table 4.11, the dampening effect is stronger in higher levels ofthe hierarchy where faction members are more concentrated (Figure 4.6). Therefore, individual incentivesin intra-faction competition counterweigh the ability of a paramount leader to grow his own faction.4.8.2 Li Keqiang PresidencyA second counterfactual we perform involves the choice of leadership ratified by the 2012 18th PartyCongress. This is the event that brought Xi Jinping to the PRC Presidency. Nathan and Gilley (2003)present compelling documentary evidence that already ten years before the formal power transition Xi Jin-ping and Li Keqiang, the current PRC Premier, belonged to a select few with potential accreditation to theparamount post. Bo Xilai was also part of this highly selected group.It is possible for us to study a counterfactual Li presidency. Figure 4.12 reports the aggregate shareof promotion, and a more detailed breakdown by level can be found in the fourth panel of Table 4.11.Interestingly, given the estimated leadership premia, the promotion at PB level would have had a verylimited increase in CYLC representation (Li’s faction). More radical shifts would have been recorded in thepromotion at the CC and AC though. Again this is a result of the slow percolation of factional representationinduced by our model, compounded with the already high CYLC representation at the upper levels of theCCP at the end of Hu’s last term in office.4.8.3 Are Princelings a Faction?The reader will notice that the analysis above posits factional affiliation of president Xi Jinping as a memberof Shanghai Gang. This is in itself a matter of debate among scholars interested in Chinese elite politics.For instance, Li (2013) in his bi-factional representation of the Chinese top tiers defines Xi as a princelingassociated with Jiang’s camp (Shanghai Gang). In fact, Xi spent only seven months in any official role inShanghai, but Jiang’s substantial influence on Xi has been noted by many. Other researchers have pointedto President Xi as the leader of a new faction of his own, mostly with roots in Shaanxi, where Xi was born,and in Zhejiang Province, where he served as Party Secretary from 2002 to 2007110. Our model allows aformal statistical analysis of some of these questions.110Some recurring affiliated politicians include current PBSC member and anti-corruption czar Wang Qishan, and potential PBSCfuture members such as Li Zhanshu, director of the CC General Office, and Politburo member Zhao Leji. Shih (2016) estimates,based on shared career experience, that less than 6 percent of current CC members have past ties with President Xi. This shouldhowever not be confused with a truly factional organization of the President’s inner circle for which hard evidence is not available.1054.8. Counterfactuals and Model AnalysisWe begin by investigating whether our postulate of the princelings not behaving as a unified faction iswarranted by the data. To assess this formally we implement Vuong specification tests between our baselinemodel and one where princeling status is coded as membership in faction P, with a specific parameter piregulating an expanded contest function of the type (4.1):W (P) =piβ +pi+ρ+η.We also specify a faction-specific leadership premium, λp = pi l/pi , which regulates the differential pro-motion probability when the paramount leader is from the princelings (e.g. Xi in the 18th Party Congress).Results are reported in Table 4.12. The Vuong test indicates that the model where princelings are consid-ered to be neutrals is preferred over one where princelings are treated as a separate faction. More importantly,the estimated leadership premium within the model imposing princelings as a faction, λp, is estimated tobe less than 1. This means that, as princeling Xi reached the paramount position, other princelings did notappear to enjoy a higher premium in promotions. This finding prima facie violates one of the crucial featuresof factional politics – delivering resources to members of the faction once the faction leader is in power –and appears in stark contrast to what we have already observed for the broadly accepted factions, CYLCand Shanghai Gang, where we estimate λ > 1. In brief, the evidence rejects the hypothesis that princelingsoperate as a unified faction.4.8.4 Is President Xi Jinping Affiliated to the Shanghai Gang?Our structural approach allows also to produce formal tests for the analysis of factional affiliation of the topleadership. The case of Xi Jinping is emblematic because of both his strong ties to the CCP elite throughfamily connections and his repeated rejection of intra-party factional politics (e.g. “cabals and cliques”mentioned in official transcripts on People’s Daily, May 3rd, 2016111).To this goal, we re-estimate the model assuming that Xi is an unaffiliated neutral, and compare thealternative model against our baseline specification where Xi is a Shanghai Gang member. The Vuong testshows that Xi is slightly more likely to be a Shanghai Gang member, although the statistical evidence isinconclusive. Our tests do not have enough power in this specific instance. Fortunately, such ambiguityis likely to be resolved after the 2017 19th Party Congress, which will unveil a wealth of data on newpromotions within the CCP.4.8.5 An Out-of-sample Forecast for the 2017 19th Party CongressTo conclude our quantitative exercises we employ our model to forecast the 19th Party Congress in 2017.Although admittedly speculative, to the best of our knowledge this is probably one of the very few rig-orous quantitative environments allowing for an exercise of this kind. The model incorporates individualcharacteristics in this analysis to obtain more accurate forecasts112.The top panel of Table 4.13 shows that share of promotions by level of the Central Committee. Underthe assumption that Xi is in fact a Shanghai Gang member, the Shanghai faction is expected to enjoy a highershare of promotions in the Politburo than the CYLC faction due to leadership premia. In contrast, promotionsat lower levels are expected to be more comparable between the two factions due to the dampening effectsstemming from vetoes.Since there is still unresolved ambiguity regarding Xi’s factional affiliation, we also conduct a forecastassuming Xi is a neutral in the bottom panel of Table 4.13. In this case, the Shanghai Gang would appear tolose its advantage in promotion for all the levels of the Central Committee.111Available at http://en.people.cn/n3/2016/0503/c90000-9052676.html112For individuals who newly enter AC at the 19th Party Congress whose characteristics are not readily available, we randomlydraw the characteristics from the sample of the new entries of 18th Party Congress.1064.9. Conclusions4.9 ConclusionsThis paper contributes to an emerging literature on the political economy of economic development byfocusing on elite organization in a nondemocracy. We specifically focus on modern China and on the internalorganization of the Chinese Communist Party. The CCP, much like historical Leninist parties in Socialistcountries, represents the linchpin of national politics and understanding its inner workings is central to anypolitical economic analysis of the PRC.We present a model of internal organization of this single-party regime, where explicit factional dynam-ics within the party enrich a problem of career concerns of political cadres. The model offers a series ofnovel insights on the role of factions in these regimes in a fully microfounded setting. Alternative modelingchoices are also discussed.The model is validated empirically employing a rich data set on the career profiles of top CCP members.In reduced form, a set of previously unexplored systematic empirical regularities in Chinese elite politicsare probed and discussed. The extent of the 2012-2016 anti-corruption purge in shaping Chinese factionalpolitics is also analyzed. In our structural estimation, we explore important counterfactuals pertinent to theChinese historical case and use the model to answer a series of questions relevant to the political economy ofthe CCP. We hope that this framework may also prove useful to the understanding of the latent institutionalshifts occurring within the CCP under General Secretary Xi.In future research we hope to extend our analysis to the 2017 19th Party Congress. This will allowprecision on all dimensions concerning the Xi Presidency.Besides our application to Chinese politics, we plan to focus on similarly complex nondemocratic envi-ronments –the example of Russia comes to mind– where our model of hierarchical party organization maybe to a certain extent transposable.1074.10. Proofs4.10 ProofsFor the proofs’ notation we exclude time indexes unless necessary.Proof of Proposition 1. Part (i). Suppose that, I(`) > 0, where I 6= J, and N(`) > 0. Consider thedecision by a faction-J politician in a node at ` of whether to veto a cofactional’s support for promotion tohis node. With the promotion of a same-faction member from J to the politician’s node, let J∗(`) denote thetotal number of faction J members that would be present at level `. Then, using equation (4.3) and (4.4),the promotion hazard parameter for this J politician at level ` (if the other faction K also vetoes co-factionmembers) if he does not veto becomes:δJ(`) =jJ∗(`)(I(`−1)(δ +δ pI (`−1))j+η+N(`−1)(δ +δ pN(`−1))j+ i+η),with j = β and i= ρ or viceversa. If instead, the politician vetoes his cofactional, and a member of the otherfaction (or a neutral) ascends to his node, his promotion hazard becomes:δJ(`) =jJ∗(`)−1(I(`−1)(δ +δ pI (`−1))j+η+N(`−1)(δ +δ pN(`−1))j+ i+η),which is strictly greater.If the other faction does not veto its members the respective expressions become:δJ(`) =jJ∗(`)(I(`−1)(δ +δ pI (`−1))j+ i+η+N(`−1)(δ +δ pN(`−1))j+ i+η)andδJ(`) =jJ∗(`)−1(I(`−1)(δ +δ pI (`−1))j+ i+η+N(`−1)(δ +δ pN(`−1))j+ i+η).And the latter hazard is clearly higher again. This proves part a).Part (ii). Suppose that, I(`) = 0, and N(`)> 0. Suppose further that, with the promotion of a co-factionmember to J’s node there will be J∗(`) members of J’s faction at level `, then the hazard parameter forpromotion of this J politician is:δJ(`) =jJ∗(`)(I(`−1)(δ +δ pI (`−1))j+η+N(`−1)(δ +δ pN(`−1))j+η). (4.7)If an N member instead ascends to his node, then the J member’s promotion hazard is:δJ(`) =jJ∗(`)−1(I(`−1)(δ +δ pI (`−1))j+η+N(`−1)(δ +δ pN(`−1))j+η),which exceeds (4.7), so he is clearly better off vetoing his own faction member.However, if an I 6= J,N ascends to his node, then the J member’s promotion hazard becomes:113δJ(`) =jJ∗(`)−1(I(`−1)(δ +δ pI (`−1))j+η+N(`−1)(δ +δ pN(`−1))i+ j+η). (4.8)113Assuming that I’s also veto. If they don’t the sufficient condition is slightly altered, but qualitatively identical.1084.10. ProofsSince this I will contest `− 1 level openings (the second expression above), this lowers the chances of theJ begin promoted to one of those. Assuming I factionals also veto, an I ascending to J’s node lowers thechances of a J promotion the most if N(`−1) = M(`−1). So a sufficient condition for J to exercise a vetoassumes all higher positions are filled by members that are neutral, N. Under this assumption expression(4.8) becomes:δJ(`) =jJ∗(`)−1(M(`−1)(δ +δ pN(`−1))i+ j+η),and expression (4.7) becomes:δJ(`) =jJ∗(`)(M(`−1)(δ +δ pN(`−1))j+η).So J will veto a J coming from level `+1 provided that:jJ∗(`)(M(`−1)(δ +δ pN(`−1))j+η)<jJ∗(`)−1(M(`−1)(δ +δ pN(`−1))i+ j+η)⇒ J∗(`)−1J∗(`)<j+ηi+ j+η⇒ J(`)< j+ηiwhere we use that J∗(`)−1 = J(`). Proof of Proposition 2. Let us define the indicator functions IB = 1, iff B(`)> 0 and IB = 0, otherwise; IN =1, iff N(`)> 0 and IN = 0, otherwise; IR = 1, iff R(`)> 0 and IR = 0, otherwise.Start with a neutral (N), who is at level ` in the hierarchy. δI(`− 1) is determined from the hierarchyabove:δN(`) = R(`−1)(δ +δ pR (`−1))pRN(`)+N(`−1)(δ +δ pN(`−1)) pNN(`)+B(`−1)(δ +δ pB (`−1)) pBN(`).Consider further that, differently from (4.4) where pRN(`) = η/(IBβ +η+ IRρ), now pRN(`) = η/(IBβ +η)because in Proposition 1 each R(`−1) is proven to veto any R possibly competing against N. For a similarreason, it holds that pBN(`) = η/(IRρ+η).We then have:δN(`)=ηN(`)(R(`−1)(δ +δ pR (`−1))IBβ +η+N(`−1)(δ +δ pN(`−1))IBβ +IRρ+η+B(`−1)(δ +δ pB (`−1))IRρ+η).Similarly, for a faction B member this is given by:δB(`) = R(`−1)(δ +δ pR (`−1))pRB(`) (4.9)+N(`−1)(δ +δ pN(`−1)) pNB (`)+B(`−1)(δ +δ pB (`−1)) pBB(`)=βB(`)(R(`−1)(δ +δ pR (`−1))β +INη+N(`−1)(δ +δ pN(`−1))β +IRρ+INη),1094.10. Proofswhere the last line uses the fact that vetoing from Proposition 1 implies pBB(`)= 0,while pRB(`)= β/(β + INη)and pNB (`) = β/(β +IRρ+INη).Finally, for a faction R member this is:δR(`) = R(`−1)(δ +δ pR (`−1))pRR(`) (4.10)+N(`−1)(δ +δ pN(`−1)) pNR (`)+B(`−1)(δ +δ pB (`−1)) pBR(`)=ρR(`)(B(`−1)(δ +δ pB (`−1))ρ+INη+N(`−1)(δ +δ pN(`−1))IBβ +ρ+INη),where the last line uses the fact that our vetoing results in Proposition 1 imply pRR(`) = 0, while pBR(`) =ρ/(ρ+INη) and pNR (`) = ρ/(ρ+IBβ +INη). Full Listing of δI(`) conditional on paramount leadershipFor an N. If an N is paramount leader:δN(`) =ηN(`)(R(`−1)(δ +δ pR (`−1))IBβ +η+N(`−1)(δ +δ pN(`−1))IBβ +IRρ+η+B(`−1)(δ +δ pB (`−1))IRρ+η).If an R is paramount leader:δN(`) =ηN(`)(R(`−1)(δ +δ pR (`−1))IBβ +η+N(`−1)(δ +δ pN(`−1))IBβ +IRρ l +η+B(`−1)(δ +δ pB (`−1))IRρ l +η).If a B is paramount leader:δN(`) =ηN(`)(R(`−1)(δ +δ pR (`−1))IBβ l +η+N(`−1)(δ +δ pN(`−1))IBβ l +IRρ+η+B(`−1)(δ +δ pB (`−1))IRρ+η).where IB = 1, i f f B(`)> 0 and IB = 0, otherwise; IR = 1, i f f R(`)> 0 and IR = 0, otherwise.For faction B member. If an N is paramount leader:δB(`) =βB(`)(R(`−1)(δ +δ pR (`−1))β +INη+N(`−1)(δ +δ pN(`−1))β +IRρ+INη).If an R is paramount leader:δB(`) =βB(`)(R(`−1)(δ +δ pR (`−1))β +INη+N(`−1)(δ +δ pN(`−1))β +IRρ l +INη).If a B is paramount leader:δB(`) =β lB(`)(R(`−1)(δ +δ pR (`−1))β l +INη+N(`−1)(δ +δ pN(`−1))β l +IRρ+INη).where IN = 1, i f f N(`)> 0 and IN = 0, otherwise; IR = 1, i f f R(`)> 0 and IR = 0, otherwise.1104.10. ProofsFor a faction R member. If an N is paramount leader:δR(`) =ρR(`)(B(`−1)(δ +δ pB (`−1))ρ+INη+N(`−1)(δ +δ pN(`−1))IBβ +ρ+INη).If an R is paramount leader:δR(`) =ρ lR(`)(B(`−1)(δ +δ pB (`−1))ρ l +INη+N(`−1)(δ +δ pN(`−1))IBβ +ρ l +INη).If a B is paramount leader:δR(`) =ρR(`)(B(`−1)(δ +δ pB (`−1))ρ+INη+N(`−1)(δ +δ pN(`−1))IBβ l +ρ+INη).where IN = 1, i f f N(`)> 0 and IN = 0, otherwise; IB = 1, i f f B(`)> 0 and IB = 0, otherwise.Proof of Proposition 3. We first demonstrate that, if the system is stationary, so that V tI (`) = VI(`) andδ tI (`) = δI (`) ∀I, `, then δI (`)> δJ (`) implies VI(`)>VJ(`). So, (4.6) is solved by the I such that δI (`) issup{δB (`) ,δR (`) ,δN (`)} .The stationary analog of equation (4.5) where V tI (`) =VI(`) and δ tI (`) = δI (`) ∀I, ` is:δVI(`) = u(`)+δI (`) [VI(`−1)−VI(`)] ,which implies:VI(`) =u(`)+δI (`)VI(`−1)δ +δI (`)andVI(`−1) = u(`−1)+δI (`−1)VI(`−2)δ +δI (`−1) .By repeated substitution:VI(`) =u(`)δ +δI (`)+δI (`)u(`−1)(δ +δI (`))(δ +δI (`−1)) +δI (`)δI (`−1)u(`−2)(δ +δI (`))(δ +δI (`−1))(δ +δI (`−2)) +. . .+δI (`)δI (`−1) · · ·δI (2)u(1)(δ +δI (`))(δ +δI (`−1)) · · ·(δ +δI (1)) .This reduces to:VI(`) =u(`)δ +δI (`)+`−1∑j=1u( j)× ∏k=`−1k= j δI (k+1)∏k=`k= j (δ +δI (k)).1114.10. ProofsSince flow payoffs are higher the higher the politician is in the hierarchy, i.e. u(`− 1) > u(`) ∀`,then necessarily increasing the rate of promotion improves valuations, dVI(`)dδI(`) > 0 ∀`. This implies thatδI (`)> δJ (`) ensures VI(`)>VJ(`).The proof proceeds next by establishing sufficient conditions for three parts. (i) The existence of neutralsgiven factions exist; (ii). The existence of a single faction given neutrals exist; (iii) The existence of a secondfaction, given neutrals and a first faction already exist.In each part, a sufficient condition is provided for δI(`) > δJ 6=I(`) and δK 6=I(`) at a single level, `. Thesufficient condition established in each case is thus required to hold at all ` in order to ensure that an enteringpolitician prefers entry as a type I.Part (i). We establish a sufficient condition for there to be neutrals. Suppose, on the contrary, that thereexist no N members. Necessarily, due to Proposition 1, without N’s, all nodes will be filled by both a Band an R. Thus, under the supposition, the hierarchy remains stationary, so that, from the result above, it issufficient to compute only the stationary δI (`) for each ` to determine the optimal I.Assume, without loss of generality, that the paramount leadership position is held by a B. Consider level` in the hierarchy. Necessarily the promotion hazard for an N at level ` is given by:δN(`) = η(R(`−1)(δ +δ pR (`−1))IBβ l +η+B(`−1)(δ +δ pB (`−1))IRρ+η).Due to optimal vetoes at each node, it must be that R(`− 1) = B(`− 1) = M(`− 1)/2 and IB = IR = 1.The relationship between δ pR (`− 1) and δ pB (`− 1) is ambiguous. So consider both cases separately. First,assume that δ pR (`− 1) ≤ δ pB (`− 1), which will imply, due to the symmetry of the posited hierarchy, thatδ pR (`)≤ δ pB (`) too. Then, substituting for IB, IR, R(`−1) and B(`−1) yields:δN(`) = η(M(`−1)/2×(δ +δ pR (`−1))β l +η+M(`−1)/2×(δ +δ pB (`−1))ρ+η).Since δ pR (`−1)≤ δ pB (`−1) then:δN(`)≥ ηM(`−1)/2×(δ +δ pR (`−1))( 1β l +η+1ρ+η),and assuming, for now, that β l > ρ implies:δN(`)≥ ηM(`−1)×(δ +δ pR (`−1))( 1β l +η). (4.11)Now consider δB(`):δB(`) =2β lM(`)(M(`−1)/2×(δ +δ pR (`−1))β l)=1M(`)(M(`−1)× (δ +δ pR (`−1))) .Then δN(`)> δB(`) if:ηM(`−1)× (δ +δ pR (`−1))( 1β L+η)>1M(`)(M(`−1)(δ +δ pR (`−1))) ,1124.10. Proofswhich rearranges to:ηβ l +η>1M(`). (4.12)Now consider δR(`):δR(`) =2ρM(`)(M(`−1)/2×(δ +δ pB (`−1))ρ)=1M(`)(M(`−1)× (δ +δ pB (`−1))) .Since, by supposition, δ pR (`−1)≤ δ pB (`−1) it is possible to define Z ≥ 1 such that δ+δpB (`)δ+δ pR (`)≡ Z. Note thatZ is invariant with respect to M(`). To see why, note that with a symmetric hierarchy in which each node isfilled by a B and R pair we have: δ pR (`) = δB(`) and δpB (`) = δR(`). Thus, using equations (4.9) and (4.10)and the fact that in such a hierarchy R(`) = B(`) = M(`)/2, we haveδR(`) =(M(`−1)× (δ +δR(`−1))M(`))andδB(`) =(M(`−1)× (δ +δB(`−1))M(`)).So the ratio δpB (`)δ pR (`)= δR(`)δB(`) =M(`−1)× (δ+δR(`−1))M(`)M(`−1)× (δ+δB(`−1))M(`)= δ+δR(`−1)δ+δB(`−1) , which is clearly independent of M(`). Using thenotation Z we then have:δR(`) =1M(`)(M(`−1)× (δ +δ pR (`−1))Z) .Then δN(`)> δR(`) if:ηM(`−1)× (δ +δ pR (`−1))( 1β L+η)>1M(`)(M(`−1)(δ +δ pR (`−1))Z) ,which rearranges to:ηβ l +η>ZM(`). (4.13)which again holds for M(`) large enough at all `. So for sufficiently large M(`), neutrals will be the preferredentering type, thus contradicting the maintained assumption that neutrals are not in the hierarchy. Assuming,alternatively, that β l ≤ ρ , instead of using the inequality in (4.11) we now have:δN(`)≥ ηM(`−1)×(δ +δ pR (`−1))( 1ρ+η),which, by following the same procedure as above, yields the analog to (4.12) as a sufficient condition forδN(`)> δB(`), namely:ηρ+η>1M(`). (4.14)This again holds for sufficiently high M(`), and again will hold for sufficiently high M(`) for the R entrantssubject to the scaling by factor Z. Again, entering politicians will choose to be neutral.1134.10. ProofsNow suppose the alternative relationship between δ pR (`−1) and δ pB (`−1), that is: δ pR (`−1)> δ pB (`−1),and again first posit that β l > ρ . Then let us use these two inequalities and substitute for IB, IR, R(`−1) andB(`−1) exactly as we did above. Equation (4.11) now yields:δN(`) = η(M(`−1)/2×(δ +δ pR (`−1))β l +η+M(`−1)/2×(δ +δ pB (`−1))ρ+η)> ηM(`−1)/2× (δ +δ pB (`−1))( 1β l +η + 1ρ+η)> ηM(`−1)× (δ +δ pB (`−1))( 1β l +η).Now δR(`) is given by:δR(`) =2ρM(`)(M(`−1)/2×(δ +δ pB (`−1))ρ)=1M(`)(M(`−1)× (δ +δ pB (`−1))) .Then δN(`)> δR(`) if:ηM(`−1)× (δ +δ pB (`−1))( 1β L+η)>1M(`)(M(`−1)(δ +δ pB (`−1))) .A sufficient condition for this is:ηβ l +η>1M(`).This again holds for M(`) high enough.Now δB(`) is given by:δB(`) =2βM(`)(M(`−1)/2×(δ +δ pR (`−1))β)=1M(`)(M(`−1)× (δ +δ pR (`−1))) .Since, by supposition it is now the case that, δ pB (`−1) ≤ δ pR (`−1) it is possible to define K ≥ 1 such thatδ+δ pR (`)δ+δ pB (`)≡ K. Similarly to the above, K is invariant with respect to M(`). Substituting for K we have:δB(`) =1M(`)(M(`−1)× (δ +δ pB (`−1))K) .Then δN(`)> δB(`) if:ηM(`−1)× (δ +δ pB (`−1))( 1β L+η)>1M(`)(M(`−1)(δ +δ pB (`−1))K) .A sufficient condition for this is:ηβ l +η>KM(`).This again holds for M(`) high enough. So new entrants will prefer to enter as neutrals over either factionfor M(`) large enough.1144.10. ProofsThe analogous procedure under the alternative assumption β l ≤ ρ yields a sufficient condition exactlyas in (4.14):ηρ+η>1M(`).Part (ii). We now establish a sufficient condition for there to exist at least a single faction. Suppose thatall positions in the hierarchy are held by a neutral. Consider an entrant choosing to also be a neutral. In thatcase under the supposition, the system is again stationary and we have:δN(`) =N(`−1)M(`)(δ +δ pN(`−1)).But by entering as a B member the entrant would have:δB(`) = βN(`−1)(δ +δ pN(`−1)β +η).These rearrange to imply that δB(`)> δN(`) provided that M(`)> β+ηβ . The analogous sufficient conditionfor an R entrant is M(`)> ρ+ηρ . This proves part (ii).Part (iii). We establish a sufficient condition for two factions to exist. We proceed as above, by demon-strating a contradiction. If there is only one faction present, without loss of generality let it be B, and theother politicians are N, for sufficiently high M(`), δR(`)> δB(`) or δN(`), so that an entering politician willchoose to enter as an R.As previously, with only N and B in the hierarchy we have:δN(`) =ηM(`)−B(`)(N(`−1)(δ +δ pN(`−1))β +η+B(`−1)(δ +δ pB (`−1))η), (4.15)δB(`) =βB(`)(N(`−1)(δ +δ pN(`−1))β +η).Either δN(`)> δB(`), so that a new entrant would prefer to enter as an N over a B, or the converse, in whichcase he would choose to enter as a B over an N. Suppose first that δN(`)> δB(`) and consider the promotionhazard for a single entering R:δR(`) = ρ(B(`−1)(δ +δ pB (`−1))ρ+η+N(`−1)(δ +δ pN(`−1))ρ+β +η).If δN(`)> δB(`) for an increase in M(`), then necessarily the term M(`)−B(`) in expression (4.15) increaseswith M(`), since an extra politician would enter as an N instead of a B. But since δR(`) above is independentof M(`), there exists an M(`) sufficiently high so that δR(`)> δN(`), and an entering politician would insteadchoose to be an R over being an N, contradicting the posited non-existence of R members in equilibrium.Alternatively, suppose that δN(`)≤ δB(`), then, for an increase in M(`) necessarily the term M(`)−N(`)increases with M(`), as a politician would choose to enter as a B over being an N. Now consider thepromotion hazard for a B:δB(`) =βM(`)−N(`)(N(`−1)(δ +δ pN(`−1))β +η).Again, since δR(`) is independent of M(`), there exists an M(`) high enough so that δR(`) > δB(`), whichimplies that a new entrant will choose to enter as an R member, again contradicting the posited non-existenceof R members. 1154.11. Tables and Figures4.11 Tables and FiguresFigure 4.1: Geographic Distribution of Factions or Groups (1956-2014)Notes: This graph shows the geographic distribution of factions or groups across provinces (municipalities)over the period of 1956 to 2014. The color scale represents the average share of faction or group in aprovince (municipality).1164.11. Tables and FiguresFigure 4.2: Leadership Premium in Promotion Rates of Each Faction or GroupNotes: This graph shows the leadership premium in promotion rates of each faction over the rest of membersin the Central Committee over time. The leadership premium in promotion rates is defined as the regressioncoefficients of promotion dummy on faction or group affiliation. The regression is repeated for each sessionof Central Committee. The capped spikes indicate the standard errors of the estimates. The shaded areaindicates that the General Secretary of CCP is from the same faction or group.1174.11. Tables and FiguresFigure 4.3: Leadership Premium in Power Score of Each Faction or GroupNotes: This graph shows the share of power score of each faction or group in the Central Committee overtime. The power score is constructed following the scheme of Bo (2010). The shaded area indicates that theGeneral Secretary of CCP is from the same faction or group.1184.11. Tables and FiguresFigure 4.4: Power Score of Each Faction or Group in the Central CommitteeNotes: This graph shows the share of power score of each faction or group in the Central Committee overtime. The power score is constructed following the scheme of Bo (2010). The vertical line indicates the yearof 1990, the first time when a civilian, Jiang Zemin, took over the Central Military Committee. The powerscore is normalized to zero in 1990. The upper panel shows the whole sample period from 1956 to 2012,the lower panel shows the post-Deng period from 1990 to 2012.1194.11. Tables and FiguresFigure 4.5: Power Score of Each Constituency in the Central CommitteeNotes: This graph shows the share of power score for each constituency in the Central Committee over time.The power score is constructed following the scheme of Bo (2010). The vertical line indicates the year of1990, the first time when a civilian, Jiang Zemin, took over the Central Military Committee. The powerscore is normalized to zero in 1990. The upper panel shows the whole sample period from 1956 to 2012,the lower panel shows the post-Deng period from 1990 to 2012.1204.11. Tables and FiguresFigure 4.6: Seat Shares at Each Level of the Central CommitteeNotes: This graph shows seat shares at each level of the Central Committee predicted by the baseline factionmodel and in the data. Each of the five bars represents the top two CCP positions, PBSC, PB, CC, andAC, from the top down, respectively. The blue/white/red bar represents the Shanghai Gang/Neutral/CYCL.The model is estimated using the 14th to 18th Central Committees and the results are averaged over 100simulations for each Party Congress.1214.11. Tables and FiguresFigure 4.7: Aggregate Share of Promotions over TimeNotes: This graph shows the time series plot of the share of promotions of each faction over time. The shareof promotions is defined as the ratio between the number of promotions of a faction and the total numberof promotions to this level. The share of promotions is predicted by the baseline faction model estimatedusing the 14th to 18th Central Committees and the results are averaged over 100 simulations for each PartyCongress.1224.11. Tables and FiguresFigure 4.8: Model Fit (In Sample)Notes: This graph shows the scatter plot of the model predicted share of promotions of each faction againstthe data. The share of promotions is defined as the ratio between the number of promotions of a faction andthe total number of promotions to this level. The blue/red dot represents Shanghai Gang/CYLC. Each dot isa share of a faction at a given level of a given Party Congress. The estimation sample includes the 14th to18th Central Committees and the results are averaged over 100 simulations for each Party Congress.1234.11. Tables and FiguresFigure 4.9: Meritocracy (In Sample)Notes: This graph shows the scatter plot of the model predicted share of promotions of each faction againstthe data. The share of promotions is defined as the ratio between the number of promotions of a faction andthe total number of promotions. The blue/red dot represents Shanghai Gang/CYLC. Each dot is a share of afaction at a given level of a given Party Congress. The estimation sample includes the 14th to 18th CentralCommittees and the results are averaged over 100 simulations for each Party Congress.1244.11. Tables and FiguresFigure 4.10: Model Fit (Out of Sample)Notes: This graph shows the scatter plot of the model predicted share of promotions of each faction againstthe data. The share of promotions is defined as the ratio between the number of promotions of a faction andthe total number of promotions to this level. The blue/red dot represents Shanghai Gang/CYLC. Each dot isa share of a faction at a given level of the 18th party congress. The estimation sample includes the 14th to17th Central Committees and the results are averaged over 100 simulations for each Party Congress.1254.11. Tables and FiguresFigure 4.11: Counterfactual Aggregate Share of Promotions over Time (Leadership Premium × 2)Notes: These graphs show the time series plot of the share of promotions of each faction over time. The shareof promotions is defined as the ratio between the number of promotions of a faction and the total number ofpromotions to this level. The counterfactual simulations are conducted by doubling the leadership premiumof the baseline faction model and the results are averaged over 100 simulations for each Party Congress.1264.11. Tables and FiguresFigure 4.12: Counterfactual Aggregate Share of Promotions over Time (Li Keqiang Presidency)Notes: These graphs show the time series plot of the share of promotions of each faction over time. Theshare of promotions is defined as the ratio between the number of promotions of a faction and the totalnumber of promotions to this level. The counterfactual simulations are conducted by assuming Li Keqiangbecame the president in the 18th Party Congress and the results are averaged over 100 simulations for eachParty Congress.127  Organizations N Duration Age Gender Ethnicity CYLC Shanghai Military Princelings Party Apparatus 10543 4.47 47.97 0.93 0.10 0.13 0.04 0.05 0.03   (2.98) (10.17) (0.26) (0.3) (0.34) (0.2) (0.21) (0.18) Government 7099 3.98 46.88 0.93 0.05 0.07 0.03 0.01 0.04   (12.54) (9.71) (0.25) (0.22) (0.25) (0.17) (0.09) (0.19) Military 2091 4.07 44.10 0.99 0.02 0.01 0.02 0.91 0.06   (3.86) (14.98) (0.1) (0.15) (0.12) (0.13) (0.29) (0.23) People's Congress 1696 5.34 56.43 0.89 0.13 0.07 0.03 0.04 0.02   (3.23) (8.94) (0.31) (0.34) (0.26) (0.17) (0.2) (0.14) CPPCC 1413 6.95 59.72 0.90 0.12 0.06 0.04 0.01 0.02   (52.9) (9.91) (0.29) (0.33) (0.23) (0.18) (0.12) (0.14) Court 213 4.46 37.16 0.97 0.22 0.08 0.02 0.00 0.00   (3.87) (8.59) (0.17) (0.41) (0.26) (0.14) (0) (0) Procuratorate 53 5.34 46.51 0.92 0.10 0.11 0.08 0.00 0.00   (3.16) (11.89) (0.27) (0.31) (0.32) (0.27) (0) (0) CYLC 521 3.85 31.91 0.87 0.12 0.77 0.01 0.00 0.03   (2.88) (7.14) (0.33) (0.32) (0.42) (0.11) (0.06) (0.16) Business 4590 4.27 42.99 0.93 0.04 0.04 0.04 0.01 0.02   (4.62) (13.66) (0.25) (0.2) (0.19) (0.19) (0.09) (0.14) Media 500 4.72 40.74 0.98 0.06 0.10 0.04 0.01 0.03   (4.61) (11.73) (0.15) (0.24) (0.3) (0.2) (0.12) (0.18) Education 3781 2.88 34.04 0.92 0.03 0.04 0.03 0.02 0.02   (4.23) (12.38) (0.26) (0.16) (0.21) (0.17) (0.14) (0.13) Unclassified 3558 3.94 40.00 0.90 0.07 0.06 0.03 0.03 0.03     (4.39) (13.58) (0.3) (0.26) (0.24) (0.17) (0.16) (0.16) Notes: This table shows summary statistics of demographics and career paths of 4,494 elites who hold import positions in government, politics, the military, education, business, and media in China since 1992. The unit of observation is position-individual pair. We report means and standard deviation, in parentheses below.  N is the number of observations in each type of organization. Duration is the length of tenure in the position. Age is the age when an individual first started the job. Gender equals 1 if an individual is male, 0 otherwise. Ethnicity equals 1 if a member is from an ethnic minority, 0 otherwise. CYLC/Shanghai/Military/Princelings equals 1 if an individual is from CYLC/Shanghai/Military/Princelings faction/group, 0 otherwise. The data source for this table is China Vitae.   Table 4.1: Summary Statistics of Elites in China128  C.C. Year N Gender Age College Graduate Mishu Ethnicity Abroad Promotion Retirement CYLC Shanghai Military Princelings 8 1956 173 0.95 51.23 0.40 0.00 0.34 0.03 0.07 0.14 0.51 0.03 0.05 0.56 0.02    (0.21) (6.97) (0.49) (0) (0.48) (0.17) (0.25) (0.35) (0.5) (0.18) (0.21) (0.5) (0.13) 9 1969 278 0.92 51.82 0.33 0.01 0.09 0.03 0.07 0.11 0.26 0.01 0.05 0.56 0.01    (0.26) (11.71) (0.47) (0.1) (0.29) (0.16) (0.25) (0.31) (0.44) (0.1) (0.23) (0.5) (0.08) 10 1973 318 0.88 52.36 0.36 0.01 0.09 0.02 0.08 0.09 0.42 0.02 0.07 0.44 0.01    (0.32) (12.94) (0.48) (0.1) (0.29) (0.15) (0.27) (0.29) (0.49) (0.12) (0.25) (0.5) (0.11) 11 1977 331 0.90 56.88 0.38 0.01 0.06 0.02 0.08 0.07 0.62 0.03 0.06 0.44 0.02    (0.3) (11.95) (0.49) (0.09) (0.24) (0.14) (0.27) (0.26) (0.49) (0.17) (0.23) (0.5) (0.12) 12 1982 344 0.93 59.61 0.58 0.04 0.12 0.04 0.09 0.13 0.62 0.08 0.06 0.33 0.02    (0.25) (8.08) (0.49) (0.2) (0.32) (0.2) (0.29) (0.34) (0.49) (0.26) (0.23) (0.47) (0.15) 13 1987 285 0.93 56.11 0.79 0.09 0.15 0.05 0.11 0.15 0.40 0.05 0.06 0.21 0.05    (0.26) (6.55) (0.41) (0.28) (0.36) (0.22) (0.32) (0.36) (0.49) (0.22) (0.23) (0.41) (0.22) 14 1992 318 0.92 56.87 0.88 0.11 0.13 0.07 0.10 0.18 0.45 0.03 0.03 0.21 0.04    (0.26) (6.18) (0.32) (0.32) (0.34) (0.25) (0.31) (0.38) (0.5) (0.16) (0.18) (0.41) (0.19) 15 1997 343 0.93 56.51 0.95 0.17 0.11 0.06 0.11 0.18 0.48 0.02 0.04 0.20 0.05    (0.26) (5.72) (0.22) (0.38) (0.31) (0.23) (0.31) (0.39) (0.5) (0.15) (0.19) (0.4) (0.22) 16 2002 356 0.93 56.05 0.98 0.31 0.07 0.07 0.10 0.16 0.47 0.05 0.06 0.19 0.05    (0.26) (5.33) (0.15) (0.46) (0.26) (0.26) (0.3) (0.36) (0.5) (0.22) (0.23) (0.39) (0.23) 17 2007 366 0.90 56.15 0.87 0.52 0.10 0.07 0.11 0.21 0.48 0.07 0.04 0.17 0.06    (0.29) (5.68) (0.34) (0.5) (0.29) (0.25) (0.31) (0.41) (0.5) (0.26) (0.2) (0.37) (0.23) 18 2012 373 0.91 56.50 0.87 0.68 0.17 0.07 0.10 0.00 0.00 0.09 0.05 0.17 0.05       (0.29) (4.73) (0.34) (0.47) (0.38) (0.26) (0.3) (0) (0) (0.29) (0.21) (0.37) (0.23) Notes: This table shows summary statistics of the members of the 8th -18th Central Committees. We report the mean and the standard deviation, below in parentheses.  Gender equals 1 if a member is male, 0 otherwise. College equals 1 if a member has a college degree, 0 otherwise. Graduate equals 1 if a member has a post-graduate degree, 0 otherwise. Abroad equals 1 if a member has studied or worked abroad, 0 otherwise. Mishu equals 1 if a member has been worked as a personal secretary of prominent politicians, 0 otherwise. Ethnicity equals 1 if a member is an ethnic minority, 0 otherwise. Promotion equals to 1 if a member will be promoted in the next session of Central Committee, 0 otherwise. Retirement equals to 1 if a member will retire after the current session of Central Committee, 0 otherwise. CYLC/Shanghai/Military/Princelings equals 1 if a member is from CYLC/Shanghai/Military/Princelings faction/group, 0 otherwise.   Table 4.2: Summary Statistics of Central Committee Members129 Dependent Variable: Average Share of Faction or Group  (1) (2) (3) (4)  Shanghai CYLC Military Princelings      GDP per capita 0.644** -0.652 -2.141*** 1.517***  [0.265] [0.623] [0.741] [0.319]      Constant 1.705*** 7.309*** 19.97*** 0.693*  [0.533] [0.915] [1.875] [0.374]      Observations 30 31 31 31 Adjusted R-squared 0.040 0.011 0.053 0.396  Notes: This table shows the cross-section regressions of the share of each faction in provinces (municipalities) on the average provincial (municipal) GDP per capita over the period of 1956-2014. The share of a faction in a province is defined as the ratio of the number of faction members who have worked in this province (municipality) over the total number of central committee members who have worked in the same place during their careers.  Robust standard errors are reported in the bracket.  ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.     Table 4.3: Geographical Distribution of Factions and Groups130      (1) (2) (3) (4) (5) (6)  All All All All Provincial National Dependent Variable:   CYLC1    CYLC2 -0.139** -0.185*** -0.189** -0.245*** -0.136* -0.499**  [0.0568] [0.0594] [0.0755] [0.0723] [0.0693] [0.143] Year F.E. N Y N Y Y Y Position F.E. N N Y Y Y Y        Observations 794 794 794 794 648 145 Adjusted R-squared 0.016 0.070 0.193 0.254 0.242 0.180        Dependent Variable:   Shanghai1    Shanghai2 -0.105*** -0.132*** -0.353* -0.378** -0.0319 -0.802*  [0.0319] [0.0346] [0.180] [0.175] [0.0466] [0.341] Year F.E. N Y N Y Y Y Position F.E. N N Y Y Y Y        Observations 773 773 773 773 627 145 Adjusted R-squared 0.006 0.011 0.382 0.392 0.187 0.278        Dependent Variable:   Princelings1   Princelings2 -0.0535 -0.0595 -0.132** -0.134** -0.155* -0.0411  [0.0505] [0.0523] [0.0571] [0.0545] [0.0806] [0.114] Year F.E. N Y N Y Y Y Position F.E. N N Y Y Y Y        Observations 794 794 794 794 648 145 Adjusted R-squared 0.001 0.020 0.133 0.154 0.202 0.227  Notes: This table shows panel regressions of the factional affiliation of the number 1 official on the number 2 official in the same political office. The top/middle/bottom panel shows results for CYLC/Shanghai/princelings respectively. Variable CYLC1 (CYLC2) is a dummy which equals to 1 if number 1 (2) official is from the CYLC faction. Shanghai1, Shanghai2, Princelings1 and Princelings2 and defined similarly. Column 1-4 include all positions, and Column 5-6 break down to provincial and national level positions. The provincial positions include 31 provincial and municipal units (secretary and governor). The position in Shanghai Municipality is excluded in the regression sample for Shanghai Gang. The national positions include Politburo Standing Committee (two highest ranking members), PRC presidency (President and Vice President), the State Council (Premier and Executive Vice premier), Central Military Committee (Chairman and Executive Vice Chairman), CCP Secretariat (two highest ranking secretaries), NPC (Chairman and Executive Vice Chairman), CPPCC (Chairman and Executive Vice Chairman), the Supreme People’s Court (President and Executive Vice President). Standard errors are clustered at both position unit and year level. ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.  Table 4.4: Factional Mix131         (1) (2) (3) (4) (5) (6)   All All All All Provincial National Dependent Variable: Shanghai1     CYLC2 0.164** 0.166* 0.187** 0.196** 0.00169 0.779***  [0.0810] [0.0818] [0.0747] [0.0761] [0.0241] [0.169] Year F.E. N Y N Y Y Y Position F.E. N N Y Y Y Y        Observations 773 773 773 773 627 145 Adjusted R-squared 0.046 0.047 0.376 0.381 0.186 0.489        Dependent Variable: CYCL1            Shanghai2 0.368* 0.315 0.396* 0.338* 0.101 0.758*  [0.195] [0.197] [0.207] [0.197] [0.207] [0.357] Year F.E. N Y N Y Y Y Position F.E. N N Y Y Y Y        Observations 773 773 773 773 627 145 Adjusted R-squared 0.043 0.069 0.207 0.239 0.232 0.200  Notes: This table shows panel regressions of the factional affiliation of the number 1 official on the number 2 official in the same political office. Variable Shanghai1 (Shanghai2) is a dummy which equals to 1 if number 1 (2) official is from the Shanghai faction. Variable CYLC1 (CYLC2) is a dummy which equals to 1 if number 1 (2) official is from the CYLC faction. The sample period is from 1992 to 2014. Column 1-4 include all positions, and Column 5-6 break down to provincial and national level positions. The provincial positions include 31 provincial and municipal units (secretary and governor) excluding Shanghai Municipality. The national positions include Politburo Standing Committee (two highest ranking members), PRC presidency (President and Vice President), the State Council (Premier and Executive Vice premier), Central Military Committee (Chairman and Executive Vice Chairman), CCP Secretariat (two highest ranking secretaries), NPC (Chairman and Executive Vice Chairman), CPPCC (Chairman and Executive Vice Chairman), the Supreme People’s Court (President and Executive Vice President). Standard errors are clustered at both position unit and year level. ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.   Table 4.5: Factional Mix (Shanghai vs. CYCL)132   Promotion  Retirement  (1) (2)  (1) (2)             CYLC 0.0397 0.0299  -0.111** -0.132***  [0.0450] [0.0456]  [0.0439] [0.0430]       CYLC*CYLC Secretary 0.206** 0.242**  -0.0797 -0.101  [0.0943] [0.0959]  [0.0818] [0.0836]       Shanghai 0.0144 0.0281  -0.0353 -0.0614  [0.0371] [0.0373]  [0.0493] [0.0498]       Shanghai*Shanghai Secretary 0.193*** 0.170**  -0.0394 0.0212  [0.0717] [0.0727]  [0.0724] [0.0737]       Princelings 0.0294 0.0368  -0.120** -0.106**  [0.0471] [0.0468]  [0.0489] [0.0484]       Princelings*Princelings Secretary 0.0158 -0.0125  -0.0161 -0.0772  [0.101] [0.103]  [0.112] [0.116]       Military -0.0414** -0.0392**  0.0229 0.0160  [0.0185] [0.0190]  [0.0280] [0.0287]       Military*Military Secretary -0.0239 -0.0313  -0.109*** -0.0465  [0.0207] [0.0262]  [0.0324] [0.0392]       Controls Y Y  Y Y Year F.E. N Y  N Y       P-value (CYLC*CYLC Secretary=Shanghai*Shanghai Secretary) 0.8275 0.5902  0.7131 0.283 Observations 2998 2998  3113 3113 Adjusted R-squared 0.066 0.068  0.121 0.155         Notes: This table shows panel regressions of promotion and retirement indicators on the faction or group affiliation of Central Committee members interacting with the affiliation of the General Secretary. The sample includes all the members of the 8th to 18th Central Committees, except Politburo Standing Committee members are excluded from the promotion regressions. Promotion is a dummy which equals to 1 if a Central Committee member moves up in the rank defined by the four levels of Central Committee (1 PBSC, 2 PB, 3 CC, and 4 AC), 0 otherwise. Retirement is a dummy which equals to 1 if a Central Committee member retires from the Central Committee, 0 otherwise.  Robust standard errors are reported in brackets. ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively. Table 4.6: Leadership Premia in Promotion and Retirement133                 CYLC      Shanghai    (1) (2) (3) (4) (5)  (1) (2) (3) (4) (5)   Power score AC seats CC seats PB seats PBSC seats Power score AC seats CC seats PB seats PBSC seats             Secretary 0.0420*** 0.0233** 0.0340*** 0.0525 0.0955*  0.0105*** -0.00802 -0.0103* 0.0786*** 0.195***  [0.00876] [0.0102] [0.00995] [0.0327] [0.0555]  [0.00382] [0.00739] [0.00587] [0.0206] [0.0398]             Observations 59 59 59 59 59  59 59 59 59 59 Adjusted R-squared 0.390 0.212 0.274 0.072 0.100  0.139 0.032 0.076 0.382 0.394                            Military      Princelings   (1) (2) (3) (4) (5)  (1) (2) (3) (4) (5)   Power score AC seats CC seats PB seats PBSC seats Power score AC seats CC seats PB seats PBSC seats             Secretary 0.274*** 0.259*** 0.231*** 0.410*** 0.476***  0.0516*** 0.0178*** 0.0271*** 0.169*** 0.361***  [0.0695] [0.0724] [0.0623] [0.0813] [0.0821]  [0.00784] [0.00484] [0.00744] [0.0234] [0.0243]             Observations 59 59 59 59 59  59 59 59 59 59 Adjusted R-squared 0.533 0.541 0.485 0.583 0.558  0.165 0.046 0.044 0.179 0.465              Notes: This table shows regressions of the power scores and seat shares of each faction or group on the affiliation of the General Secretary. The dependent variables are the power score (Score), the share of Alternate Central Committee members (AC), the share of full Central Committee members (CC), the share of Politburo members (PB), and the share of Politburo Standing Committee members (PBSC). The independent variable Secretary is a dummy which equals to 1 if the General Secretary is from the same faction, 0 otherwise. The top left panel (column 1-5) reports the results for the CYLC faction. The top right panel (column 6-10) reports the results for the Shanghai faction. The bottom left panel (column 1-5) reports the results for the Military group, the bottom right panel (column 6-10) reports the results for the Princeling group. The sample period is from 1956 to 2014. Newey-West standard errors with 5 lags are reported in brackets. ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.    Table 4.7: Leadership Premia in Power Score and Seat Shares134     Dependent Variable Corruption     CYLC 0.0272  [0.0237]   Shanghai -0.0229  [0.0237]   Princelings 0.0189  [0.0321]   Gender 0.0139  [0.0167]   Ethnicity  -0.0191  [0.0168]   AC -0.0350**  [0.0136]   CC -0.00920  [0.0129]   PB 0.0125  [0.0407]   PBSC 0.0328  [0.0583]   age -0.00596***  [0.000649]   Observations 2240 Adjusted R-squared 0.032  Notes: This table shows the cross-sectional regression of a corruption dummy on the faction or group affiliation of an official. Corruption is defined as 1 if the official is investigated or prosecuted according to ChinaFile and the China’s Central Commission for Discipline Inspection (CCDI) website, and 0 otherwise. The sample includes all the individuals except military personnel covered by China Vitae who have not retired in the year of 2007, the year of 17th Party Congress. Robust standard errors are reported in brackets.  ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively.    Table 4.8: Anticorruption and Factional Affiliation135   (1)   (2)   (3)   (4)   0.045***    0.162**   0.042***    0.153**  [0.008]  [0.063]  [0.009]  [0.062]   0.029***    0.193***   0.033***    0.241**  [0.006]  [0.068]  [0.010]  [0.103]   2.553***    0.041***    2.791***    0.037***  [0.511]  [0.007]  [0.720]  [0.008]      0.022***    2.178***    0.027***    [0.005]  [0.758]  [0.009]     2.526***      2.898***    [0.514]    [0.761]          1.956***        [0.699]         Log-LL -2766  -2747  -2766  -2746 Log-LLR -19.305  -0.378  -19.142  - P-value 0.000   0.385   0.000   -  Notes: This table shows the parameter estimates of the faction model for different specifications. The sample includes all the members of the 14th to 18th Central Committees. Standard errors are reported in brackets.  ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively. The bottom panel shows log-likelihood, log-likelihood ratio, and p-value of the log-likelihood ratio tests for each specification against model (4) as the alternative hypothesis. The estimator employs 100 simulations for each Party Congress.      Table 4.9: Parameter Estimates of the Faction Model136   Baseline Faction Model   Faction with Individual Characteristics   Random   Seniority    0.162**    0.174**   entry 0.043   entry 0.043  [0.063]  [0.069]        0.193***    0.201***   entry 0.043   entry 0.043  [0.068]  [0.072]        0.041***    0.043***   Age1 0.464***  [0.007]  [0.008]    [0.105]    0.022***    0.023***   Age2 -1.213***  [0.005]  [0.005]    [0.127]   2.526***   2.390***   Age3 -0.428***  [0.514]  [0.531]    [0.050]   Princeling 0.413**        [0.202]       Military 0.129        [0.122]       College -0.152        [0.164]       Graduate -0.222*        [0.119]       Minority -0.813***        [0.208]       Gender 0.926***        [0.237]       Age1 0.361***        [0.109]       Age2 -1.201***        [0.136]       Age3 -0.421***        [0.055]             Log-LL -2747  -2617 -2763  -2660 Log-LLR -129.976  -  -  - P-value 0.000  -  -  - Vuong -  -  -13.429  -7.026 P-value  -   -  0.000   0.000  Notes: This table shows the parameter estimates of four alternative models of CCP promotion dynamics. The sample includes all the members of the 14th to 18th Central Committees. The probability of entry for seniority and random model is calibrated using the mean faction shares in the sample. Standard errors are reported in brackets.  ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively. The estimator employs 100 simulations for each Party Congress. The bottom panel shows log-likelihood, log-likelihood ratio, p-value of the log-likelihood ratio tests, Vuong test statistics, and the p-value of the Vuong tests for each model against the model “faction with individual characteristics” column as the alternative hypothesis.  Table 4.10: Parameter Estimates of Alternative Models137   Data  Baseline Faction Model  Counterfactual  (Leadership Premium×2)  Counterfactual  (Li Keqiang Presidency)   B N R  B N R  B N R  B N R 14th                  PB 18.18% 81.82% 0.00%  22.41% 68.25% 9.33%  33.76% 59.30% 6.95%  22.41% 68.25% 9.33%  CC 2.80% 96.26% 0.93%  3.48% 93.19% 3.33%  6.09% 90.59% 3.32%  3.48% 93.19% 3.33%  AC 2.83% 96.23% 0.94%  4.01% 92.65% 3.34%  6.60% 90.24% 3.16%  4.01% 92.65% 3.34% 15th                   PB 22.73% 68.18% 9.09%  19.42% 67.28% 13.30%  32.43% 53.14% 14.43%  19.42% 67.28% 13.30%  CC 5.61% 89.72% 4.67%  3.40% 92.77% 3.83%  6.16% 89.29% 4.55%  3.40% 92.77% 3.83%  AC 6.19% 85.84% 7.96%  3.85% 92.40% 3.75%  5.60% 89.51% 4.89%  3.85% 92.40% 3.75% 16th                   PB 9.09% 63.64% 27.27%  11.50% 69.25% 19.25%  9.96% 61.59% 28.45%  11.50% 69.25% 19.25%  CC 1.94% 85.44% 12.62%  1.68% 91.83% 6.49%  1.81% 86.49% 11.70%  1.68% 91.83% 6.49%  AC 2.44% 94.31% 3.25%  2.34% 90.61% 7.05%  2.27% 86.57% 11.16%  2.34% 90.61% 7.05% 17th                   PB 22.73% 63.64% 13.64%  16.27% 63.70% 20.03%  16.70% 54.83% 28.47%  14.28% 63.40% 22.32%  CC 4.31% 90.52% 5.17%  2.31% 90.94% 6.75%  3.27% 85.34% 11.39%  1.82% 90.54% 7.64%  AC 3.17% 88.10% 8.73%  2.53% 90.52% 6.95%  3.02% 86.51% 10.47%  1.90% 90.46% 7.64% 18th                Notes: This table shows the share of promotions of each faction by level of the Central Committee in the data and predicted by the different models. The share of promotions is defined as the ratio between the number of promotions of a faction and the total number of promotions to this level.  The sample includes all the members of the 14th to 18th Central Committees. The first panel shows the share of promotions of each faction in the data. The second panel shows the prediction by the baseline faction model. The third panel shows the counterfactual prediction in which the leadership premium is doubled comparing to the baseline faction model. The last panel shows the counterfactual prediction in which Li Keqiang becomes President in the 18th Party Congress.  Table 4.11: Share of promotion of Each Faction by Level of the Central Committee138   Baseline Faction Model   Princelings as Faction   Xi as Neutral    0.162**    0.178**    0.164**  [0.063]  [0.074]  [0.064]    0.193***    0.153**    0.195***  [0.068]  [0.067]  [0.069]    0.041***    0.364***    0.044***  [0.007]  [0.124]  [0.008]    0.022***    0.050***    0.027***  [0.005]  [0.009]  [0.006]   2.526***    0.027***   2.150***  [0.514]  [0.006]  [0.437]      0.059***      [0.010]       1.876***      [0.394]       0.564      [0.358]         Log LL -2747 -2866  -2748 Vuong -  -15.850  -0.197 P-value  -  0.000  0.422  Notes: This table shows the parameter estimates of three models of CCP promotion dynamics. The sample includes all the members of the 14th to 18th Central Committees. Standard errors are reported in brackets. The estimator employs 100 simulations for each Party Congress.  ***,**,* indicates 1 percent, 5 percent, and 10 percent significance level respectively. The bottom panel shows log-likelihood, Vuong test statistics, and the p-value of the Vuong tests for each model against the baseline faction model as the alternative hypothesis.   Table 4.12: Tests of Xi’s Factional Affiliation139 Xi as Shanghai Gang  B N R PB 24.18% 66.37% 9.45% CC 3.84% 92.72% 3.44% AC 4.35% 91.52% 4.13%     Xi as Neutral PB 14.12% 75.53% 10.35% CC 2.20% 94.03% 3.77% AC 2.28% 93.77% 3.95%  Notes: This table shows the aggregate share of promotions of each faction at each level of the Central Committee in the 19th Central Committee predicted by the faction model with individual characteristics. The share of promotions is defined as the ratio between the number of promotions of a faction and the total number of promotions to this level. The sample used to estimate the parameters includes all the members of the 14th to 18th Central Committees. The forecast employs 100 simulations for this Party Congress.  Table 4.13: Out-of-sample Forecast of 19th Central Committee140Chapter 5ConclusionThis thesis studies theoretical mechanisms and empirical consequences of government actions on financialmarket in order to better understand the organization of the financial sector and the inner working of thegovernments.This first essay documents a new channel of monetary policy transmission through the shadow bankingsystem. Analyzing U.S. money supply data from 1987 to 2012, I find that shadow bank deposits expandsignificantly when the Federal Reserve tightens monetary policy. This channel partially offsets the reductionof commercial bank deposits and dampens the impact of monetary tightening. I construct a structural modelof bank competition and show that this new channel is a result of deposit competition between commercialand shadow banks in a market with heterogeneous depositors. Facing more yield-sensitive clientele, shadowbanks pass through more rate hikes to depositors during periods of monetary tightening, thereby poachingdeposits from commercial banks. Fitting my model to institution-level data from both commercial banks andmoney market funds, I show that the shadow bank channel reduces the impact of monetary policy on moneysupply by 40 percent. My results suggest a cautious stance towards the use of monetary tightening as a toolfor promoting financial stability, because monetary tightening may unintentionally drive more deposits intothe uninsured shadow banking sector, thereby amplifying the risk of bank runs.The second essay examines the effects of the post-crisis financial regulations, encompassing the Dodd-Frank Act and Basel III, on market liquidity of the U.S. fixed income market. We estimate structural breaksin a large panel of liquidity measures of corporate and Treasury bonds. Our methodology does not requirea priori knowledge of the exact timing of breaks, can capture not only sudden jumps but also breaks inslow-moving trends, and displays excellent power properties in presence of confounding factors. Againstthe popular claim that post-crisis regulations hurt liquidity, we find no evidence of liquidity deteriorationduring periods of regulatory intervention. Instead, breaks towards higher liquidity are often identified. Wediscuss their connection to the post-crisis rise in agency trading.The third essay investigates theoretically and empirically the factional arrangements and dynamicswithin the Chinese Communist Party (CCP), the governing political party of the People’s Republic of China.Our empirical analysis ranges from the end of the Deng Xiaoping era to the current Xi Jinping presidencyand covers the appointments of both national and provincial officials. We present a set of new empirical reg-ularities within the CCP and a theoretical framework suited to model factional politics within single-partyregimes.Future work In the future work, I plan to extend my thesis in several dimensions. For instance, the firstessay studies monetary transmission in a partial equilibrium setting. It would be very interesting to extendthe model to a general equilibrium framework where I can study the interaction between market power andmonetary policy on firm investments and aggregate output. The second direction is to apply the structural IOframework to study other financial industry such as mutual funds. Previous literature on mutual fund usuallyassumes a representative investor which has perfectly elastic demand for mutual fund investment. A directresult of such assumption is that the representative investor will only care about abnormal returns generatedby mutual fund managers. However, in practice, mutual fund investors could have very heterogeneouspreference over factor loadings of different mutual funds. This may generate market segmentations anddifferent competitive environment for different mutual funds. It would be very interesting to explore theimplications of such framework.141BibliographyAcharya, Viral V., Philipp Schnabl, and Gustavo Suarez, 2013, Securitization without risk transfer, Journalof Financial Economics 107, 515–536.Adams, Robert M, Kenneth P Brevoors, and Elizabeth K Kiser, 2007, Who competes with whom? the caseof depository institutions, Journal of Industrial Economics 141–167.Adrian, Tobias, and Hyun Song Shin, 2010, Liquidity and leverage, Journal of Financial Intermediation 19,418–437.Ahn, Seung C, and Alex R Horenstein, 2013, Eigenvalue ratio test for the number of factors, Econometrica81, 1203–1227.Ajello, Andrea, Thomas Laubach, David López-Salido, and Taisuke Nakata, 2015, Financial stability andoptimal interest-rate policy, Working Paper .Amato, Jeffery D, and Eli M Remolona, 2003, The credit spread puzzle, BIS Quarterly Review, December .Amihud, Yakov, and Haim Mendelson, 1980, Dealership market: Market-making with inventory, Journal ofFinancial Economics 8, 31–53.Amihud, Yakov, and Haim Mendelson, 1986, Asset pricing and the bid-ask spread, Journal of financialEconomics 17, 223–249.Andrews, Donald WK, 1994, Empirical process methods in econometrics, Handbook of econometrics 4,2247–2294.Aragon, George O, and Philip E Strahan, 2012, Hedge funds as liquidity providers: Evidence from thelehman bankruptcy, Journal of Financial Economics 103, 570–587.Bai, Chong-En, Chang-Tai Hsieh, and Zheng Michael Song, 2014, Crony capitalism with chinese charac-teristics, University of Chicago, working paper .Bai, Jushan, and Serena Ng, 2002, Determining the number of factors in approximate factor models, Econo-metrica 70, 191–221.Bai, Jushan, and Pierre Perron, 1998, Estimating and testing linear models with multiple structural changes,Econometrica 47–78.Bai, Jushan, and Pierre Perron, 2003, Computation and analysis of multiple structural change models, Jour-nal of applied econometrics 18, 1–22.Bao, Jack, Jun Pan, and Jiang Wang, 2011, The illiquidity of corporate bonds, The Journal of Finance 66,911–946.Barr, Michael S, 2012, Financial crisis and the path of reform, the, Yale J. on Reg. 29, 91.142BibliographyBernanke, Ben S, 2011, The effects of the great recession on central bank doctrine and practice, Keynoteaddress at the Federal Reserve Bank of Boston 56th Economic Conference Long Term Effects of the GreatRecession 1, 8–19.Bernanke, Ben S, and Alan S Blinder, 1988, Credit, money, and aggregate demand, The American EconomicReview 435–439.Berry, Steven, 1994, Estimating discrete-choice models of product differentiation, The RAND Journal ofEconomics 242–262.Berry, Steven, James Levinsohn, and Ariel Pakes, 1995, Automobile prices in market equilibrium, Econo-metrica: Journal of the Econometric Society 841–890.Bertrand, Marianne, Matilde Bombardini, and Francesco Trebbi, 2015, Lobbying the Dodd-Frank Act of ,mimeo UBC.Bo, Zhiyue, 2004, The 16th central committee of the chinese communist party: formal institutions andfactional groups, Journal of Contemporary China 13, 223–256.Borio, Claudio, and Haibin Zhu, 2012, Capital regulation, risk-taking and monetary policy: a missing linkin the transmission mechanism?, Journal of Financial Stability 8, 236–251.Brunnermeier, Markus K, and Lasse Heje Pedersen, 2009, Market liquidity and funding liquidity, Review ofFinancial studies 22, 2201–2238.Carey, Mark, Mitch Post, and Steven A. Sharpe, 1998, Does corporate lending by banks and finance compa-nies differ? evidence on specialization in private debt contracting, The Journal of Finance 53, 845–878.Chamberlain, Gary, 1982, Multivariate regression models for panel data, Journal of Econometrics 18, 5–46.Chen, Liang, Juan J Dolado, and Jesús Gonzalo, 2014, Detecting big structural breaks in large factor models,Journal of Econometrics 180, 30–48.Chen, Ye, Hongbin Li, and Li-An Zhou, 2005, Relative performance evaluation and the turnover of provin-cial leaders in china, Economics Letters 88, 421–425.Coates, IV, and C John, 2014, Cost-benefit analysis of financial regulation: Case studies and implications,Yale Law Journal, Forthcoming .Cochrane, John H, 2014, Challenges for cost-benefit analysis of financial regulation, The Journal of LegalStudies 43, S63–S105.Comerton-Forde, Carole, Terrence Hendershott, Charles M Jones, Pamela C Moulton, and Mark SSeasholes, 2008, Time variation in liquidity: The role of market maker inventories and revenues .Cox, Gary W, Frances McCall Rosenbluth, and Michael F Thies, 1999, Electoral reform and the fate offactions: The case of japan’s liberal democratic party, British Journal of Political Science 29, 33–56.Dang, Tri Vi, Gary Gorton, and Bengt Holmström, 2016, The information sensitivity of a security, WorkingPaper .Di Maggio, Marco, and Marcin Kacperczyk, 2016, The unintended consequences of the zero lower boundpolicy, Working Paper .143BibliographyDick, Astrid A., 2008, Demand estimation and consumer welfare in the banking industry, Journal of Bankingand Finance 32, 1661–1676.Dick-Nielsen, Jens, Peter Feldhütter, and David Lando, 2012, Corporate bond liquidity before and after theonset of the subprime crisis, Journal of Financial Economics 103, 471–492.Dittmer, Lowell, 2003, Leadership change and chinese political development, The China Quarterly 176,903–925.Drechsler, Itamar, Alexi Savov, and Philipp Schnabl, 2016, The deposits channel of monetary policy, Work-ing Paper .Driscoll, John C, and Ruth A Judson, 2009, Sticky deposit rates: Data and implications for models of priceadjustment, Technical report, Working Paper.Dudley, W., 2015, Regulation and liquidity provision, Remarks at the SIFMA Liquidity Forum, NYC,9/30/2015 .Duffie, Darrell, 2012, Market making under the proposed volcker rule, Rock Center for Corporate Gover-nance at Stanford University Working Paper .Edwards, Amy K, Lawrence E Harris, and Michael S Piwowar, 2007, Corporate bond market transactioncosts and transparency, The Journal of Finance 62, 1421–1451.Egan, Mark, Ali Hortaçsu, and Gregor Matvos, 2015, Deposit competition and financial fragility: Evidencefrom the us banking sector, Fama-Miller Working Paper 15–33.Feldhütter, Peter, 2012, The same bond at different prices: identifying search frictions and selling pressures,Review of Financial Studies 25, 1155–1206.Francois, Patrick, Ilia Rainer, and Francesco Trebbi, 2014, The dictator’s inner circle, Technical report,National Bureau of Economic Research.Francois, Patrick, Ilia Rainer, and Francesco Trebbi, 2015, How is power shared in africa?, Econometrica83, 465–503.Frieden, Jeffry, 2015, The political economy of adjustment and rebalancing, Journal of International Moneyand Finance 52, 4–14.Garman, Mark B, 1976, Market microstructure, Journal of financial Economics 3, 257–275.Giannetti, Daniela, and Michael Laver, 2005, Party cohesion, party factions and legislative party discipline initaly, in Joint Workshop Sessions of the European Consortium for Political Research: Granada, Citeseer.Glushkov, Denys, 2007, Corporate bonds and ratings: Overview and research issues, in WRDS User Meet-ing, Wharton School, University of Pennsylvania.Goldstein, Michael A, and Edith S Hotchkiss, 2007, Dealer behavior and the trading of newly issued corpo-rate bonds, in Afa 2009 san francisco meetings paper.Gorton, Gary, and Andrew Metrick, 2010, Regulating the shadow banking system, Brookings Papers onEconomic Activity 2010, 261–297.Gorton, Gary, and Andrew Metrick, 2012, Securitized banking and the run on repo, Journal of FinancialEconomics 104, 425–451.144BibliographyGoyenko, Ruslan, Avanidhar Subrahmanyam, and Andrey Ukhov, 2011, The term structure of bond marketliquidity and its implications for expected bond returns, Journal of Financial and Quantitative Analysis46, 111–139.Gromb, Denis, and Dimitri Vayanos, 2002, Equilibrium and welfare in markets with financially constrainedarbitrageurs, Journal of financial Economics 66, 361–407.Gürkaynak, Refet S, Brian Sack, and Jonathan H Wright, 2007, The us treasury yield curve: 1961 to thepresent, Journal of Monetary Economics 54, 2291–2304.Gustafsson, Elisabeth, 2014 .Hanson, Samuel Gregory, David Stuart Scharfstein, and Aditya Vikram Sunderam, 2015, An evaluation ofmoney market fund reform proposals, Working Paper .Heckman, James, Hidehiko Ichimura, Jeffrey Smith, and Petra Todd, 1998, Characterizing selection biasusing experimental data, Technical report, National bureau of economic research.Heckman, James J, Hidehiko Ichimura, and Petra E Todd, 1997, Matching as an econometric evaluationestimator: Evidence from evaluating a job training programme, The review of economic studies 64, 605–654.Ho, Katherine, and Joy Ishii, 2011, Location and competition in retail banking, International Journal ofIndustrial Organization 29, 537–546.Ho, Thomas, and Hans R Stoll, 1981, Optimal dealer pricing under transactions and return uncertainty,Journal of Financial economics 9, 47–73.Hu, Grace Xing, Jun Pan, and Jiang Wang, 2013, Noise as information for illiquidity, The Journal of Finance68, 2341–2382.Huang, Jing, 2006, Factionalism in Chinese communist politics (Cambridge University Press).Iyer, Rajkamal, and Manju Puri, 2012, Understanding bank runs: The importance of depositor-bank rela-tionships and networks, American Economic Review 102, 1414–45.Jia, Ruixue, Masayuki Kudamatsu, and David Seim, 2015, Political selection in china: The complementaryroles of connections and performance, Journal of the European Economic Association 13, 631–668.Kacperczyk, Marcin, and Philipp Schnabl, 2013, How safe are money market funds?, The Quarterly Journalof Economics 128, 1073–1122.Kaiser, R. G., ????, How america’s essential institution works, and how it doesn’s .Kashyap, Anil K, and Jeremy C Stein, 1995, The impact of monetary policy on bank balance sheets, inCarnegie-Rochester Conference Series on Public Policy, volume 42, 151–195, Elsevier.Kashyap, Anil K., and Jeremy C. Stein, 2000, What do a million observations on banks say about thetransmission of monetary policy?, American Economic Review 90, 407–428.Krishnamurthy, Arvind, 2002, The bond/old-bond spread, Journal of Financial Economics 66, 463–506.Krishnamurthy, Arvind, and Annette Vissing-Jorgensen, 2015, The impact of treasury supply on financialsector lending and stability, Journal of Financial Economics 118, 571–600.145BibliographyKyle, Albert S, 1985, Continuous auctions and insider trading, Econometrica: Journal of the EconometricSociety 1315–1335.Li, Cheng, 2002, The shanghai gang: Force for stability or cause for conflict, China Leadership Monitor 1,1–18.Li, Cheng, 2012, The battle for china top nine leadership posts, The Washington Quarterly 35, 131–145.Li, Cheng, 2013, A biographical and factional analysis of the post-2012 politburo, China Leadership Moni-tor 41, 1–17.Li, Hongbin, and Li-An Zhou, 2005, Political turnover and economic performance: the incentive role ofpersonnel control in china, Journal of Public Economics 89, 1743–1762.Longstaff, Francis A., 2004, The flight-to-liquidity premium in u.s. treasury bond prices, Journal of Business77, 511–526.Longstaff, Francis A, Sanjay Mithal, and Eric Neis, 2005, Corporate yield spreads: Default risk or liquidity?new evidence from the credit default swap market, The Journal of Finance 60, 2213–2253.McCabe, Patrick E, Marco Cipriani, Michael Holscher, and Antoine Martin, 2013, The minimum balanceat risk: A proposal to mitigate the systemic risks posed by money market funds, Brookings Papers onEconomic Activity 2013, 211–278.McCarty, Nolan, Keith T Poole, and Howard Rosenthal, 2013, Political Bubbles: Financial Crises and theFailure of American Democracy (Princeton University Press).Mian, Atif, Amir Sufi, and Francesco Trebbi, 2008, The political economy of the us mortgage default crisis,Technical report, National Bureau of Economic Research.Mian, Atif, Amir Sufi, and Francesco Trebbi, 2014, Resolving debt overhang: Political constraints in theaftermath of financial crises, American Economic Journal: Macroeconomics 6, 1–28.Moreira, Alan, and Alexi Savov, 2016, The macroeconomics of shadow banking, National Bureau of Eco-nomic Research Working Paper .Nelson, Benjamin, Gabor Pinter, and Konstantinos Theodoridis, 2015, Do contractionary monetary policyshocks expand shadow banking?, Bank of England Working Paper .Nevo, Aviv, 2000, A practitioner’s guide to estimation of random-coefficients logit models of demand,Journal of Economics & Management Strategy 9, 513–548.Nevo, Aviv, 2001, Measuring market power in the ready-to-eat cereal industry, Econometrica 69, 307–342.Newey, Whitney K, and Kenneth D West, 1986, A simple, positive semi-definite, heteroskedasticity andautocorrelationconsistent covariance matrix.Parlatore, Cecilia, 2016, Fragility in money market funds: Sponsor support and regulation, Journal of Fi-nancial Economics .Persico, Nicola, Jose CR Pueblita, and Dan Silverman, 2011, Factions and political competition, Journal ofPolitical Economy 119, 242–288.Posner, Eric A, and E Glen Weyl, 2013, Benefit-cost analysis for financial regulation, American EconomicReview 103.146BibliographyPozsar, Zoltan, Tobias Adrian, Adam B Ashcraft, and Haley Boesky, 2010, Shadow banking, Working Paper.Radecki, Lawrence J., 1998, The expanding geographic reach of retail banking markets, Economic PolicyReview 15–34.Reynaert, Mathias, and Frank Verboven, 2014, Improving the performance of random coefficients demandmodels: The role of optimal instruments, Journal of Econometrics 179, 83–98.Roll, Richard, 1984, A simple implicit measure of the effective bid-ask spread in an efficient market., Jour-nal of Finance 39, 1127–39.Scharfstein, David, and Adi Sunderam, 2016, Market power in mortgage lending and the transmission ofmonetary policy, Working Paper .Schmidt, Lawrence DW, Allan G Timmermann, and Russ Wermers, 2016, Runs on money market mutualfunds, Working Paper .Shih, Victor, 2004, Factions matter: personal networks and the distribution of bank loans in china, Journalof Contemporary China 13, 3–19.Shih, Victor, Christopher Adolph, and Mingxing Liu, 2012, Getting ahead in the communist party: explain-ing the advancement of central committee members in china, American Political Science Review 106,166–187.Smets, Frank, 2013, Financial stability and monetary policy: How closely interlinked?, Sveriges RiksbankEconomic Review 3, 121–160.Stein, Jeremy C, 2012, Monetary policy as financial stability regulation, The Quarterly Journal of Eco-nomics 127, 57–95.Stein, Jeremy C, 2013, Overheating in credit markets: origins, measurement, and policy responses, speech,February 7.Stock, James H, and Mark Watson, 2009, Forecasting in dynamic factor models subject to structural insta-bility, The Methodology and Practice of Econometrics. A Festschrift in Honour of David F. Hendry 173,205.Stock, James H, and Mark W Watson, 2011, Dynamic factor models, Oxford Handbook of Economic Fore-casting 1, 35–59.Sunderam, Adi, 2015, Money creation and the shadow banking system, Review of Financial Studies 28,939–977.Teles, Pedro, and Ruilin Zhou, 2005, A stable money demand: Looking for the right monetary aggregate,Economic Perspectives 29, 50.Thakor, Anjan, 2012, The economic consequences of the volcker rule, Report by the US Chamber Centerfor Capital Market Competitiveness .Williams, John C., 2014, Financial stability and monetary policy: happy marriage or untenable union?,FRBSF Economic Letter .Yellen, Janet, 2014, Monetary policy and financial stability, The 2014 Michel Camdessus Central BankingLecture .147


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items