UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on macroeconomics and corporate behavior Wang, Mingzhi 2014

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

Item Metadata


24-ubc_2015_february_wang_mingzhi.pdf [ 1.42MB ]
JSON: 24-1.0167065.json
JSON-LD: 24-1.0167065-ld.json
RDF/XML (Pretty): 24-1.0167065-rdf.xml
RDF/JSON: 24-1.0167065-rdf.json
Turtle: 24-1.0167065-turtle.txt
N-Triples: 24-1.0167065-rdf-ntriples.txt
Original Record: 24-1.0167065-source.json
Full Text

Full Text

Essays on Macroeconomics andCorporate BehaviorbyMingzhi WangB.Econ, Renmin University of China, 2006M.A., The University of British Columbia, 2007A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Economics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)December 2014c© Mingzhi Wang 2014AbstractThis dissertation focuses on the link between corporate behavior and macroe-conomic phenomena. It is comprised of three separate but related chapters:The first chapter asks whether increases in firms’ outsourcing can explainthe downward trend of the investment to GDP ratio of the US. I develop amodel with heterogeneous firms and a fixed cost to enter outsourcing. I testthe model’s implications using a novel dataset collected from US computermanufacturing firms’ annual reports and find that empirical facts are con-sistent with the model. I find outsourcing firms invest less while producemore. Also, the lessening effect of outsourcing on investment is increasingover time.In the second paper I first show that the divergent trends in the macroand micro volatility of output is nonexistent after the year 2000. Then Iargue that rapid technological innovation in the 80s and 90s contributes tothese divergent trends. I explain the driving force of the different patterns ofvolatility using a model in which firms’ investment has both long term andshort term gains. Higher long term gain leads more firms to invest in riskyprojects resulting in higher micro volatility. At the same time, more par-ticipation contributes to better diversification at the aggregate level, hencelower macro volatility.The third chapter investigates the stock market’s reaction to monetarypolicy shocks. I use the Romer and Romer(2005) measure of policy shocksand event study approach to perform this task. I find that stock marketreturns respond to monetary policy shocks. There is an asymmetry in thereaction in terms of boom and recession periods, small firms are more sen-sitive to shocks than large firms, highly leveraged firms are more sensitiveto shocks than those with less leverage. These findings shed new lights onwhether monetary authority should take into account the effects on financialmarket when making monetary policy changes.iiPrefaceThis dissertation is original, unpublished, independent work by the author,Mingzhi Wang.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . x1 Outsourcing and Corporate Investment . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Data and Measurement . . . . . . . . . . . . . . . . . . . . . 51.2.1 Measures of Outsourcing . . . . . . . . . . . . . . . . 51.2.2 Examples of Narrative Description of Outsourcing . . 61.2.3 Descriptive Statistics . . . . . . . . . . . . . . . . . . 71.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.3.1 Production Technology . . . . . . . . . . . . . . . . . 91.3.2 Optimal Outsourcing . . . . . . . . . . . . . . . . . . 101.3.3 Main Implications of the Model . . . . . . . . . . . . 111.3.4 Model Simulation . . . . . . . . . . . . . . . . . . . . 121.4 Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . 131.4.1 Outsourcing and Firm Size . . . . . . . . . . . . . . . 131.4.2 Investment, Capital and Sales . . . . . . . . . . . . . 131.4.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . 141.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.6 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . 172 Firm-level Risk-taking and Volatility Trends . . . . . . . . 342.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 34ivTable of Contents2.2 Firm-level Volatility and Macro Volatility . . . . . . . . . . . 372.2.1 Data and Measures . . . . . . . . . . . . . . . . . . . 372.2.2 Volatility in COMPUSTAT Firms . . . . . . . . . . . 372.2.3 Cohort and Number of New Firms . . . . . . . . . . . 382.2.4 Firm Size . . . . . . . . . . . . . . . . . . . . . . . . . 382.2.5 Industry . . . . . . . . . . . . . . . . . . . . . . . . . 382.2.6 R & D Intensity and Investment Behavior . . . . . . 392.2.7 Estimating the Time-varying Firm-level Volatility . . 392.2.8 Baseline Regression Results . . . . . . . . . . . . . . . 402.2.9 Correlation between Firm-level and GDP Growth . . 412.3 An Illustrative Model . . . . . . . . . . . . . . . . . . . . . . 422.3.1 Entrepreneurs and Risky Projects . . . . . . . . . . . 422.3.2 Optimal Risk Taking and Volatility . . . . . . . . . . 432.3.3 Comparative Statics . . . . . . . . . . . . . . . . . . . 442.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452.5 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . 473 Stock Market Reaction to Monetary Policy Shocks . . . . 603.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.2 Measures of Monetary Policy Shocks . . . . . . . . . . . . . . 633.2.1 Regression-based Measures . . . . . . . . . . . . . . . 633.2.2 Market-based Measures . . . . . . . . . . . . . . . . . 643.2.3 Narrative Measures . . . . . . . . . . . . . . . . . . . 663.3 Stock Market Reaction to Monetary Policy Shocks . . . . . . 673.3.1 Data and Methodology . . . . . . . . . . . . . . . . . 673.3.2 Baseline Regression . . . . . . . . . . . . . . . . . . . 683.3.3 Industry and Country Portfolios . . . . . . . . . . . . 693.3.4 Asymmetry: Controlling for Recession . . . . . . . . 703.3.5 Alternative Methods of Checking Asymmetry . . . . 713.3.6 Does the Fed Signal Its (Superior) Information? . . . 713.4 Monetary Policy Shocks and Cross-sectional Stock Returns . 723.4.1 Size, Book-to-market, and Market Leverage . . . . . . 733.4.2 Evidence from 25 Portfolios formed on Different Char-acteristics . . . . . . . . . . . . . . . . . . . . . . . . 753.4.3 Statistical Test of Significance . . . . . . . . . . . . . 763.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763.6 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . 78Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102vList of Tables1.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . 251.2 Baseline Parameter Values . . . . . . . . . . . . . . . . . 261.3 Linear Probability regression of Outsource on FirmSize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271.4 Main Results: Capital Per Employee . . . . . . . . . . 281.5 Main Results: Investment Per Employee . . . . . . . . 291.6 Main Results: Sales Per Employee . . . . . . . . . . . . 301.7 Investment: Controlling for Investment Opportunity 311.8 Investment: Controlling for Financial Constraints . . 321.9 Sales: Controlling for Productivity and Profitability . 332.1 Regression of Volatility on firm-level controls . . . . . 593.1 Summary Statistics: Monetary Policy Variables . . . . 783.2 Summary Statistics: Stock Returns . . . . . . . . . . . . 793.3 Response of CRSP Value-weighted Stock Return tothe Monetary Policy Shocks . . . . . . . . . . . . . . . . 803.4 Response of F-F 10-industry portfolios to the Mone-tary Policy Shocks . . . . . . . . . . . . . . . . . . . . . . . 813.5 Monthly Stock Return Reaction to US Shocks byCountry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.6 Response of CRSP Value-weighted Stock Return tothe Monetary Policy Shocks: Controlling for Recession 833.7 Monthly Stock Return Reaction to US Monetary Shocksby Country . . . . . . . . . . . . . . . . . . . . . . . . . . . 843.8 Response of Stock Returns to Positive and NegativeShocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853.9 Checking Sign Asymmetry using Interacting Dummies 863.10 Market Reaction and the Greenbook Forecast . . . . . 873.11 Responses of Size, Book-to-market and Market Lever-age Portfolios . . . . . . . . . . . . . . . . . . . . . . . . . . 88viList of Tables3.12 Responses of Size, Book-to-market and Market Lever-age Portfolios Controlling for Recession . . . . . . . . . 893.13 Responses of 25 Portfolios Formed on Size and Book-to-market . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.14 Responses of 25 Portfolios Formed on Size and Mar-ket Leverage . . . . . . . . . . . . . . . . . . . . . . . . . . 913.15 Wald Test of Difference in Reactions of Leverage Port-folios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92viiList of Figures1.1 Nonresidential Investment from BEA and COMPU-STAT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.2 Investment Rates by Industry in COMPUSTAT . . . 181.3 Investment Rates in Manufacturing . . . . . . . . . . . . 181.4 Sample Comparison . . . . . . . . . . . . . . . . . . . . . . 191.5 Time Trend of Outsourcing . . . . . . . . . . . . . . . . . 191.6 Time Trend of Average Oscore . . . . . . . . . . . . . . . 201.7 Time Trend of Investment Rate . . . . . . . . . . . . . . 201.8 Outsourcing by Size . . . . . . . . . . . . . . . . . . . . . . 211.9 Outsourcing by Age . . . . . . . . . . . . . . . . . . . . . . 221.10 Shocks and Outsourcing Trend . . . . . . . . . . . . . . . 231.11 Capital Stock and Investment per Labor . . . . . . . . 231.12 Sales per Labor . . . . . . . . . . . . . . . . . . . . . . . . 242.1 Volatility at Firm-level v.s. Volatility of Real GDP . . 472.2 Number of Firms by Cohort . . . . . . . . . . . . . . . . 482.3 Volatility by Cohort . . . . . . . . . . . . . . . . . . . . . . 482.4 Number of Firms by Size . . . . . . . . . . . . . . . . . . 492.5 Volatility by Size . . . . . . . . . . . . . . . . . . . . . . . . 492.6 Number of Firms by Sector . . . . . . . . . . . . . . . . . 502.7 Volatility by Sector . . . . . . . . . . . . . . . . . . . . . . 502.8 Investment Intensity by Sector . . . . . . . . . . . . . . . 512.9 Investment Intensity by Size . . . . . . . . . . . . . . . . 512.10 Investment Intensity by Cohort . . . . . . . . . . . . . . 522.11 R&D Intensity by Sector . . . . . . . . . . . . . . . . . . 522.12 R&D Intensity by Size . . . . . . . . . . . . . . . . . . . . 532.13 R&D Intensity by Cohort . . . . . . . . . . . . . . . . . . 532.14 Plot of Age Effects (Age Cuttoff = 15) . . . . . . . . . 542.15 Plot of Age Effects (Age Cuttoff = 20) . . . . . . . . . 542.16 Plot of Age Effects (Age Cuttoff = 30) . . . . . . . . . 55viiiList of Figures2.17 Plot of Estimates on Time Fixed Effects (Age Cuttoff= 15) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552.18 Plot of Estimates on Time Fixed Effects (Age Cuttoff= 20) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562.19 Plot of Estimates on Time Fixed Effects (Age Cuttoff= 30) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562.20 Firm-level Volatility after Controlling for Composi-tion and Capital Intensity . . . . . . . . . . . . . . . . . . 572.21 Rolling Correlation Coefficients . . . . . . . . . . . . . . 572.22 Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 582.23 Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . 582.24 Minimum Size Requirement . . . . . . . . . . . . . . . . 593.1 Seven-day Cumulated Stock Return after Policy Shock 933.2 Seven-day Cumulated Stock Return after Policy Shock:Recession v.s. Expansion . . . . . . . . . . . . . . . . . . 943.3 Monetary Shock Series and Stock Returns . . . . . . . 953.4 Responses of 10 Portfolios Formed on Size . . . . . . . 963.5 Responses of 10 Portfolios Formed on Book-to-marketEquity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 973.6 Responses of 10 Portfolios Formed on Market Leverage 983.7 25 Portfolios Formed on Size and Book-to-market Eq-uity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 993.8 25 Portfolios Formed on Size and Market Leverage . . 1003.9 25 Portfolios Formed on Size and Market Leverage:3D Version . . . . . . . . . . . . . . . . . . . . . . . . . . . 101ixAcknowledgementsFirst and foremost I want to thank my thesis advisor, Professor Paul Beaudry.He is a patient, knowledgeable, insightful and helpful person. It has beenmy honor to be his student. I appreciate all his contributions of time, ideasand funding to make my PhD journey productive. Moreover, his joy andenthusiasm in research has always been an inspiration to me. I feel verylucky to have such a role model for my academic career.I would also like to thank the rest of my thesis committee: ProfessorAmartya Lahiri and Professor Henry Siu for their encouragement, insightfulcomments and helpful advice.I thank Viktoria Hnatkovstka, Jesse Perla, Yaniv Yedid-Levi, CarolinPflueger, Giovanni Gallipoli, Michael Devereux, Hiro Kasahara and all otherseminar participants at UBC for their insightful discussions and suggestions.Last but not least, I want to thank my parents and my girlfriend for theirunderstanding and support to help me get through the most challengingperiod of my first 30 years of life.xChapter 1Outsourcing and CorporateInvestment1.1 IntroductionNon-residential investment in the United States(US) was falling over thepast 30 years.1 (Figure 1.1 shows the pattern of U.S. investment.) Non-residential investment as a share of GDP was 13 percent in 1980s, while itfalls to around 10 percent in 2010. This fall in investment was mainly drivenby the corporate sector. The total investment of US public firms fell from 8.1percent of GDP to 4.5 percent of GDP. The difference between the two series,as an approximation of the total investment made by privately held firms, isrelatively stable. Over the past years, the U.S. non-financial corporate sectorbecame a net lender vis-a-vis the rest of the economy by hoarding cash andreducing investment(Armenter & Hnatkovska, 2012). Asker, Farre-Mensa& Ljungqvist (2013) compares the investment of public firms to that ofobservably similar privately held firms using a novel panel dataset of privateU.S. firms, and find public firms invest significantly less than private firms.What is driving the change in investment? A natural question would bewhether this is driven by all firms investing less or by a compositional changewhereby industries with less investment are more important in the economy.Figure 1.2 shows this secular trend of declining investment is mainly drivenby the manufacturing sector. Total capital expenditures by COMPUSTATfirms declined from 4 percent of GDP in 1980 to 1.3 percent in 2010. Thedownward trend is also partly driven by transportation and communica-tion industries. However, these firms are highly regulated and are oftenexcluded from studies on firm behavior in private sector. Figure 1.3 showsthe median investment to asset ratio for three subsectors of manufacturing:nondurable, durable, and business equipment. For an average firm in each1During the same time, residential investment over GDP ratio has a general upwardtrend, especially from 1990 to 2005, but this can not compensate the drop in non-residential investment. In general, US investment over GDP ratio is falling.11.1. Introductionsector, investment rate declined over time.At the same time, there is an increasing trend of outsourcing, especiallyin the manufacturing sector. McMillan (1995) reviews the management lit-erature and concludes that during the 1980s and 1990s, firms subcontractedmore. Grossman and Helpman (2005), Shy and Stenbacka (2004) and Feen-stra (1998)(citing Tempest (1996) and Tisdale (1998) ), all show examples oranecdotes of how outsourcing is widespread in today’s business. There areheated political debates in the media and during the 2012 U.S. presidentialdebates on the consequences of the widespread and increasing trend of out-sourcing on the U.S. economy.2. The linkage between physical investmentand outsourcing is straightforward: In corporate annual reports, public firmsoften list “avoid large capital expenditure” as a reason for outsourcing. Itis therefore natural to investigate the quantitative impact of outsourcing onfirms’ investment behavior using firm-level data. However, empirical workon outsourcing at the plant or firm level for the US is limited by the avail-ability of data, with an exception of Kurtz (2006), which studies the foreignoutsourcing of intermediate materials between 1987-1992 using plant-leveldata. Most studies on the impact of international outsourcing use industrylevel data because the availability of firm-level data is limited, for example,Feenstra and Hanson (2004) and Feenstra (1998).This paper give an explanation of the observed data pattern in invest-ment by linking firms’ capital investment to their organizational choice be-tween integration and outsourcing. I develop a model with heterogeneousfirms with a fixed cost of finding an outsourcing partner and a variable costof outsourcing. Capital investment is associated to firms’ “make or buy”decision. The model has several implications as the marginal cost of out-sourcing declines. First, physical investment per employee declines due tofirms substituting internal production with outsourcing while output peremployee increases due to more firms switching to more cost efficient out-sourcing. Second, outsourcing increases both at the intensive and extensivemargins. With the existence of fixed costs, the model also predicts thatlarger firms are more likely to outsource.I test the main implications of the model using a novel dataset of US2In the media and presidential debates , “outsourcing” and “offshoring” are often usedinterchangeably. In business, outsourcing is the contracting out of a business process toa third-party, while offshoring often describes the relocation by a company of a businessprocess from one country to another. In the case of the U.S., these two are often correlatedsince the outsourcing partners of the firms are often located outside of the U.S. This paperfocuses on outsourcing (the boundary of the firm), when the information about offshoring(the boundary of manufacturing country) is unobservable.21.1. Introductionpublic firms in the computer manufacturing industry. The annual reports ofthese firms describe their organization of manufacturing activity. I combinetheir outsourcing choice information with financial data in Standard andPoor’s COMPUSTAT dataset to form a panel dataset of sales, investment,outsourcing and various firm characteristics. I find outsourcing firms investless. Also, the reduction in investment by these firms is more significantover time than for non-outsourcing firms. On the other hand, sales peremployee increase within outsourcing firms, and outsourcing firms’ salesincrease more than non-outsourcing firms over the sample period. The shareof outsourcing firms increases while the extent of outsourcing also increases.Larger firms are more likely to outsource, which is also consistent with themodel’s prediction.The first and foremost contribution of this paper is that I utilize publicinformation from U.S. corporations’ annual reports to gather informationon outsourcing. Both the decision and extent of outsourcing are measuredfrom the narrative record of the reports. This methodology can be extendedto all publicly traded manufacturing firms. To my knowledge, this is thefirst dataset on U.S. corporations’ manufacturing outsourcing decisions.This paper adds to the literature of the determination and effect of firms’organizational choice between integration and outsourcing. Existing theo-ries of outsourcing are centered around the transaction-cost economics ap-proach(TCE) or the property-rights theory(PRT).3 The general empiricalresult on investment is consistent with both TCE and PRT. However, theempirical facts shown in this paper have also placed more restriction to ex-isting theories of outsourcing: Outsourcing increases with firm size, and out-sourcing is increasing at both the extensive and intensive margins over time;outsourcing firms also invest less while producing more. The empirical sideof literature is not only interested in testing the theory of the determinants ofvertical integration4, but also address the consequences of firms’ integrationversus outsourcing choice by asking the question “do firm boundaries mat-ter?” The consequence of price, cost, profit, stock ratings and stock returnsare all investigated using different datasets and methodologies. However,none of these studies utilize the public information of publicly traded firms.This paper provides a new evidence of how outsourcing matters in terms offirm-level investment, output and firms’ size distribution.This paper also fits in the broad literature of firms’ investment and saving3An comprehensive review of this literature is Lafontaine and Slade (2007)4In addition to Lafontaine and Slade (2007), Acemoglu et al. (2007) has a brief reviewof the empirical studies centered around testing TCE and PRT. Teixeira (2011) providesa general survey of the empirical literature of outsourcing and vertical integration.31.1. Introductionbehavior. This literature is mainly focused on the “puzzle” of firms hold-ing more cash than necessary for investment and operation needs. Existingexplanations in the literature include transaction motive, precautionary mo-tive, tax motive, and agency problem between shareholders and managers(Bates, Kahle & Stultz, 2009). While most papers in this literature provideexplanation from the firms’ saving perspective, i.e., why firms are hoardingcash, my work offers an explanation from the spending perspective: whyfirms do not invest as they did before. The model does not explain thecorporate saving behavior directly, however, my result explains firms’ in-vestment and saving choice and is consistent with the general view thatchanges in firms’ vertical supply relationships contributes to the “corporatesaving glut” 5.Both my theoretical and empirical results on the extensive margin (per-centage of firms) and intensive margin (extent) of outsourcing shed lighton a specific line of literature in international trade, which discusses thecontribution of extensive margin and intensive margin of trade. For exam-ple,Eaton, Kortum and Kramarz (2011) use firm-level data on exportingbehavior of French firms and show that the extensive and intensive marginof export both increase with the size of a destination market. My resultsfrom the model show that the extensive margin and intensive margin ofoutsourcing, is decreasing with the variable cost of outsourcing. Since out-sourcing may be highly correlated to imports, and outsourcing is increasingin firm size (which can be interpreted in a firm-specific market-size), it isconsistent with their result. Helpman, Melitz and Rubinstein (2008) showthe extensive margin of trade flow is important in estimating the effect oftrade frictions. My empirical results show that extensive margin of firms’soutsourcing is indeed increasing over time and makes the importance ofaddressing the extensive margin in such estimation increasingly important.The rest of the paper is organized as follows. Section 2 presents themethodology of creating the novel dataset from combining the Lexis Nexisdataset with the COMPUSTAT dataset. Section 3 shows the theoreticalframework and derives the main testable implications. Section 4 gives themain empirical results and robustness checks. Section 5 concludes.5For example, Gao (2013) shows how changes in inventory management contributes tocorporate cash hoarding with both theory and evidence.41.2. Data and Measurement1.2 Data and Measurement1.2.1 Measures of OutsourcingIn this section, I introduce a novel dataset of US computer manufacturingfirms. The dataset is constructed by combining the financial data in Stan-dard and Poor’s COMPUSTAT with the narrative information from theannual reports of each firm in the LexisNexis database. Companies with atleast 500 stockholders and/or $5 million in assets and/or big private debtplacements must file disclosure documents with the Securities and ExchangeCommission(SEC). Brokers, investment advisers and certain shareholdersmust make filing too. These 10-K annual reports can be retrieved throughthe LexisNexis dataset. Although COMPUSTAT is also based on compa-nies’ reports, it does not have information of firms’ outsourcing; therefore,existing empirical studies of US outsourcing at firm level is limited by theavailability of data. However, in firms’ annual reports, they often discusstheir outsourcing activity in several sections.Firstly they discuss the outsourcing decision in “Business” when theydiscuss their manufacturing activity and materials sourcing. Secondly, somediscuss this information in “Item 1A. Risk Factors”, or “Item 7. Manage-ment Discussion and Analysis” The detailedness of this information varies:from full disclosure of the outsourcing partner and share of product out-sourced, to only a mention of whether they outsource manufacturing. I findthese reports in the LexisNexis database and construct a dummy variableof outsourcing based on the description of the manufacturing activity. Ialso assign a “outsourcing score” based on the description of the amount orstages of manufacturing outsourced noted in the reports. This measure issubject to my subjective interpretation of the report.The specific industry studied in this paper is the computer manufac-turing industry, with Standard Industrial Classification (SIC) codes from3670 to 3677, over the period of 1987-2010. It is very common for thisindustry to outsource subassembly manufacturing or even the whole man-ufacturing manufacturing business to another firm. The generality of theoutsourcing pattern in computer manufacturing industry may be unclear,since outsourcing information for other industries are not widely reported.Firms not covered in LexisNexis are deleted from the sample. I assign theoutsourcing score based on both the amount of outsourcing and the extent ofoutsourcing. I divide the extent of outsourcing into the following categories:no manufacturing conducted internally, only final assembly conducted in-ternally, subassembly manufacturing conducted internally, key components51.2. Data and Measurementand circuit boards are manufactured internally, and most components aremanufactured internally. For example, if a firm indicates that it buys mate-rials and components to manufacture the final product, it will be assignedas a non-outsourcing firm and the outsourcing score will be 0. On the otherhand, if it states that only final assembly is conducted internally, it will beassigned as a outsourcing firm with a outsourcing score of 80%. If multipleproducts are sold in one company and the manufacturing stages conductedinternally are different for each products, I condider the share of total salesfor each product when assigning the value for outsourcing scores. Therefore,the outsourcing score is a assessment of the available narrative informationon both the stages and amount of outsourced manufacturing.1.2.2 Examples of Narrative Description of OutsourcingIn this section I show how I extract the information from narrative descrip-tions of outsourcing and subcontracting.Example 1. Cray Inc., 2004, “While the Company has designed all ofthe MTA system hardware components, it subcontracts the manufacture ofthese components, including integrated circuits,printed circuit boards, flexcircuits and power supplies, on a sole or limited source basis to third-partysuppliers. The Company’s strategy is to avoid the large capital commitmentand overhead associated with establishing manufacturing facilities and tomaintain the flexibility to adopt new technologies if and when they becomeavailable without the risk of equipment obsolescence. The Company per-forms final system integration and testing, and designs and maintains itsMTA system software internally. ”In this case, since the company only does system integration and test-ing done internally, and subcontracts the manufacture of key componentsto third-part suppliers, the outsourcing dummy has value 1 while the out-sourcing score is assigned as 90%.Example 2. Maxwell Technology Inc., 2002, “All of our manufacturingoperations are conducted in two production facilities located in San Diego,California, and Rossens, Switzerland. Over the past several years, we havemade substantial capital investments to outfit and expand our productionfacilities and incorporate the latest available mechanization and automationtechniques and processes. We have trained our manufacturing personnelin advanced operational techniques including demand-based manufacturing.We have also added advanced information technology infrastructure andhave implemented new business processes and systems to increase our man-ufacturing capacity and improve efficiency, planning and product quality.61.2. Data and MeasurementOur production facilities have been designed with flexible overhead powergrids and modular manufacturing cells and equipment that allow factoryoperations to be reconfigured rapidly at minimal expense.”Since all the manufacturing operations are conducted internally in thecompany’s own production facilities, it receives an outsourcing dummy ofvalue 0 and the outsourcing score is also 0. It is worth mentioning thatoutsourcing is concerned with the boundary of firms instead of a nation’sboundaries(offshoring). The production facility in Switzerland is still ownedby the company, so it is only offshoring instead of outsourcing.Example 3. Dell Inc., 1994, “The Company manufactures all of its desk-top and server personal computer systems at its Austin, Texas and Limer-ick, Ireland manufacturing facilities. The Company expects to begin con-struction in March 1995 of a 238,000 square foot combination office andmanufacturing facility on a nine-acre site in Penang, Malaysia, to meet theneeds of its expanding Asia-Pacific business. The Company contracts withQuanta Computer, Inc. and Sony Corporation to manufacture unconfiguredbase Latitude and Latitude XP notebook personal computers, respectively,which the Company custom configures for shipment to customers.”Here the company manufactures all desktops while subcontracting themanufacture of some notebook PC products. I assign an outsourcing dummyof value 1, while giving an outsourcing score of 20%.Example 4. Dell Inc., 2010, “Third parties manufacture the majority ofthe client products we sell under the Dell brand. We use contract manufac-turers and manufacturing outsourcing relationships as part of our strategyto enhance our variable cost structure and to achieve our goals of generat-ing cost efficiencies, delivering products faster, better serving our customers,and building a world-class supply chain. Our manufacturing facilities are lo-cated in Austin, Texas; Penang, Malaysia; Chengdu, China; Xiamen, China;Hortolandia, Brazil; Chennai, India; and Lodz, Poland. ”Now since the majority of products are manufactured by third parties,the outsourcing dummy still has a value of 1, while the outsourcing score isassigned as 80%.1.2.3 Descriptive StatisticsTable 1.1 gives the summary statistics for the whole sample. The wholesample contains 166 different firms with 1787 observations. Of these firms,about 48% are outsourcing firms and the average outsourcing score (Oscore)is 38%. Investment is deflated by the chained price index of non-residentialinvestment from the Bureau of Economic Analysis (BEA). Sales is deflated71.2. Data and Measurementby the chained price index of Computer and peripheral equipment from theBEA. Capital stock is constructed using the perpetual inventory methodwith the initial value set to the net property, plant and equipment anddepreciation rate of 20%.6 There is huge variation in firm size, with thesmallest firm has only 7 employees while the largest firm has 324,600 em-ployees. Measured by total book assests, the smallest firm has 0.57 millionwhile the largest observation has 124.5 billion assets. By deleting the firmsnot covered by LexisNexis, I lose 997 observations from the COMPUSTATdataset. As firms size become smaller, more observations are lost. Fig-ure 1.4 shows the general pattern that LexisNexis tends to cover the largerfirms from the COMPUSTAT dataset. I form 10 size groups based on theCOMPUSTAT data size decile cutoffs. So for the original COMPUSTATdata, each decile should have equal number of observations. After deletingmissing observations, as firm size becomes smaller, more observations aremissing in LexisNexis dataset.Firm-level outsourcing is increasing at both the extensive and intensivemargins. Figure 1.5 shows the time trend of the share of firms which areclassified as outsourcers in the sample. This number starts from around25% in 1987, and grew to 84% in 2010. The number of outsourcing firmsare increasing in the computer manufacturing industry. Figure 1.6 showsthe average Oscore over time for the whole sample, it starts from 15% in1987 and ends at 75% in 2010. In other words, an average public firm incomputer manufacturing industry did 85% of its manufacturing in house in1987, while in 2010 it only does 25% of all manufacturing within its ownfacilities. The investment rate in this sample is consistent with the generalpattern of the whole manufacturing sector. Figure 1.7 shows the averageinvestment (capital expenditure) over total asset ratio in the industry. Itfalls from 7.2% in late 1980s to only 2% in 2010. The falling was drasticfrom 1999 to 2003, due to the recession resulting from the crash of dotcombubble and 911 attacks.Figure 1.8 plots the unconditional mean of outsourcing and Ocsore ac-cording to different size percentiles. Here firm size is measured by the totalbook asset of the firm. Generally, outsourcing is widespread in every sizegroup. For the first eight groups, firms are more likely to outsource as firmsize increases while for the last two size bins, the shares of outsourcing firmsare lower. Similar data pattern is also found in the Oscore. Although the un-conditional relationship between firm size and outsourcing is not apparent,6The regression results are similar when using depreciation rate assumption of 15% to25%81.3. ModelI will test this relationship in Section 4 conditional on other characteristicsand show larger firms are indeed more likely to outsource. Figure 1.9 showsoutsourcing firm numbers and Oscore for different age groups. The generalpattern is that younger firms(newly financed publicly) are more likely tooutsource and outsource more intensively. In the model of Section 3, thispattern is not directly derived from the model. However, the model predictsthat new outsourcers contribute more to the average extent of outsourcing(Oscore in the data), which can be interpreted as firm age.1.3 ModelIn this section, I provide a model where firms are faced with technologyshocks and fixed and variable costs of outsourcing. A firms’s outsourcingchoice is introduced as a substitute for manufacturing the intermediate goodinternally. I derive the optimal outsourcing and show the implications of themodel using numerical methods.1.3.1 Production TechnologyThere are a set of firms indexed by i at time t, firms employ non-manufacturinglabor, Ln,i,t and combine it with a intermediate good Xi,t to produce thefinal good, the production function is given byYi,t = Ai,t[Lβn,i,tX1−βi,t]θwhere 0 < θ < 1 is the return to scale parameter to ensure there is asolution for this partial equilibrium problem. Firms can either outsourcethe manufacturing of Xi,t or produce it in house. Ai,t is the productivityof firm i. If produced in house (integration), the production technology isgiven byXi,t = si,tKαi,tL1−αm,i,twhere Ki,t is physical capital and Lm,i,t is manufacturing labor hired bythe firm, si,t is the productivity of internal production of the intermediategood. In the baseline case, si,t is the only heterogeneity in the cross sectionand it follows an AR(1) process log si,t = ρ log si,t+i,t, where i,t ∼ N (0, σ).Firms are faced with a fixed entry cost c to outsource. This cost can be in-terpreted as the transaction cost incurred from search effort as in Grossmanand Helpman (2002), or simply the cost of changing manufacturing organi-zations. Investment is defined as Ii,t = Ki,t − (1− δ)Ki,t−1.91.3. ModelIf a firm does not pay the fixed cost, it has to produce the good insidethe firm, therefore it maximizespiV,i,t = Ai,t[Lβn,i,t(si,tKαn,i,tL1−αm,i,t)1−β]θ− wnLn,i,t − wmLm,i,t − rKi,twhere the subscript V stands for vertical integration.If the firm decides to pay the fixed cost, an outsourcing firm maximizesprofitpiO,i,t = Ai,t[Lβn,i,t(Oi,t + si,tKαi,tL1−αm,i,t)1−β]θ−pt2O2i,t−wn,tLn,i,t−wm,tLm,i,t−rtKi,tHere firm i can buy the amount Oi of intermediate good from the out-sourcing partner. This is subject to a convex cost function C(Oi,t) =pt2O2i,t,a setup motivated by the transaction cost of outsourcing literature, whereoutsourcing is subject to hold up problems and other cost. Other interpre-tations and motivations of this cost include cost of communication, qualityassurance, supply chain management, and the cost disadvantage caused byoutsourcing from giving the supplier more bargaining power. In our model,the exogenous shock is coming from the change in p. This can be interpretedas the decline in information cost or transaction cost between trading part-ners.1.3.2 Optimal OutsourcingIn each period firm i observes the productivity (Ai,t, s(i, t)) and make itsdecision for whether to outsource or not, then they decide the amount ofintermediate goods outsourced and the optimal inputs (K,Ln,i,t, Lm,i,t) usedin production. From the first order conditions of the firm’s profit maximiza-tion problem, an outsourcing firm’s optimal amount of outsourcing is givenby:O?i =(wm1− α)(1−α) ( rα)α(sip)−1The intuition behind this expression is that as the productivity of in-ternal production of the intermediate good increases, firm i will outsourceless. Also if the marginal cost of outsourcing increases, firms will also reducethe amount of outsourcing. The optimal amount of outsourcing is inversely101.3. Modelrelated to the input prices for internal production. If the wage of manu-facturing labor or the cost of capital increases, firms will outsource more.7Firms decide whether to pay the fixed cost by comparing the optimalprofit of outsourcing and non-outsourcing. So it will enter outsourcing ifand only ifpi?O,i,t ≥ pi?V,i,t + c1.3.3 Main Implications of the ModelThe model has several implications when the cost parameter pt is decreasingover time. First, outsourcing will increase at both the extensive and intensivemargins. This is achieved by the existence of both the fixed and variablecosts of outsourcing in the model. This feature of the model has the sameflavor as models explaining the extensive margin and intensive margin ofinternational trade.Second, investment per labor will decrease while output per labor will in-crease. Moreover, the reducing effect of outsourcing on investment is higheras pt declines. In other words, the gap of investment per labor or capitalper labor between outsourcing firms and non-outsourcing firms is bigger asthe cost of outsourcing declines. Similarly, the difference in sales/outputper labor between outsourcing and non-outsourcing firms also increases astime goes by. The intuition behind this result is, as the variable cost of out-sourcing declines, firms shift internal manufacturing to more cost-effectiveoutsourcing, therefore, capital rented and manufacturing labor hired declinewhile non-manufacturing labor demand increases. Since capital per manu-facturing labor is constant in a Cobb-Douglas setup, capital per total laborwill decrease.The model also predicts larger firms are more likely to outsource. Thisfollows from how the fixed cost of outsourcing works in the model. Out-sourcing is not optimal until the gain from the outsourcing is larger thanthe fixed cost. This economy of scale feature predicts outsourcing firms tendto be larger.7During the 1987-2010 period, real wage rate and real interest rate both had a down-ward trend in general, which should have caused a decline in outsourcing based on mymodel. Then decrease in si or p must be large enough to dominate this effect. Sincesi is just introduced for heterogeneity purpose in my model, and there is no evidence ofdeclining manufacturing productivity, exogenous changes in p is the driving force for theincreased outsourcing.111.3. Model1.3.4 Model SimulationIn this section, I give a qualitative simulation of the baseline model. Thisframework can be easily extended to various functional forms to match thequantitative aspects in the data better. The baseline specification of themodel has closed form solutions for optimal outsourcing but is only enoughto show qualitative implications.Table 1.2 shows the parameter values in the baseline calibration. Mostof the values are taken directly from the literature. For some of the newparameters specific to the model in this paper, such as pt and si,t I choosethe ones that best match the data pattern of the outsourcing and Oscoretrends.The first two panels in Figure 1.10 show the two exogenous shocks in thebaseline model. I assume that pt falls from 0.7 to 0.2 over a 20-year period,and firms receive idiosyncratic internal productivity shocks. The model isessentially static with only pt changing over time. The resulting share ofoutsourcing firms and the average outsourcing share of intermediate goodare shown in the lower two panels. Outsourcing firms increase over time,and the average outsourcing share (Oscore) also increases over time. Thissimple model can match the data pattern of secular trends of outsourcingat both the extensive margin and intensive margin. Comparing the aver-age outsourcing share between all outsourcing firms and the subgroup thatwas outsourcing from the beginning, the average outsourcing shares for thetotal outsourcing firms are higher. New outsourcers are outsourcing moreagressively.Figure 1.11 shows the dynamics of corresponding capital stock and in-vestment per labor. Due to the Cobb-Douglas specification, for non-outsourcingfirms, the capital labor ratio is always constant which is a benchmark tostudy the behavior of outsourcing firms. Outsourcers’ capital labor ratioand investment labor ratio decline over time. Also, new outsourcers haveless capital and investment per labor than existing ones. The gaps betweenoutsourcers and non-outsourcers increases as the declining rate of both cap-ital and investment also increase over time.Figure 1.12 shows average sales/output per labor for different groups offirms as time goes by. Outsourcers’ sales labor ratio increases over time,and the difference between outsourcers and non-outsourcers also increases.However, newcomers to outsourcing do not behave significantly differentlyin output per labor. For an outsourcing firm, when switching from non-outsourcing to outsourcing, sales increase. However, non-manufacturing la-bor hired also increase. Therefore the difference between outsourcers and121.4. Empirical Evidencenon-outsourcers is not widening dramatically like capital labor ratio.1.4 Empirical EvidenceIn this section, I test the main implications derived from the model: Largerfirms are more likely to outsource, the gap of investment per labor, cap-ital labor ratio and sales labor ratio between outsourcing firms and non-outsourcing firms all increase with time.1.4.1 Outsourcing and Firm SizeOne aspect of the model is due to the fixed cost of entering into outsourcing.Larger firms are more likely to outsource. I test whether larger firms aremore likely to outsource by using a linear probability model, as well as usinglogit and probit model for robustness checks.Table 1.3 shows the correlation of outsourcing and different measures offirm size using a linear probability regression. I present results both withand without fixed effects. These results all show that as firm size increases,outsourcing is more likely. Based on the results with firm fixed effects, aone percent increase in total asset leads to a 0.21 percent increase in theprobability of outsourcing. A one percent increase in sales leads to 0.05 per-cent increase in the probability of outsourcing. Using size group dummies,the estimated coefficients are increasing as size gets larger. Without firmfixed effects, these estimates are all significant at 1% level. With firm fixedeffects, the estimates on the large firm groups are statistically significant.These results prove that larger firms are indeed more likely to outsource.1.4.2 Investment, Capital and SalesThe model predicts the difference in investment labor ratio, capital laborratio, and sales labor ratio between outsourcing firms and non-outsourcingfirms grow larger as time passes. In other words, firms’ investment and out-put respond more to outsourcing over time. I test these model implicationsby running the following regression:Yit = α0 + α1Outsourceit + α2time+ β ×Outsourceit × time+X ′itγ + itwhere Yit is the dependent variable such as investment labor ratio, capitallabor ratio and sales labor ratio, α0 is the intercept, α1 is the coefficient onan outsourcing measure, α2 is the coefficient on a time trend. Xit contains131.4. Empirical Evidencefirm-level control variable to control for different firm characteristics. Themain coefficient of interest is β, which is the estimate on the interaction termof outsourcing and a time trend. This measures how dependent variablesrespond to the outsourcing decision in the data.Table 1.4 shows the main results of the regression using capital per em-ployee as the dependent variable. The estimates on β using the outsourcingdummy, oscore, with or without firm fixed effects are all negative and statis-tically significant. Based on the results of regression (3), using outsourcingdummy as the measure of outsourcing, an outsourcing firm’s capital stockis 538 dollars per employee per year less than a non-outsourcing firm. Theregression result in (4) using Oscore shows that a one percentage changein Oscore will make capital per employee decrease by 7.89 dollars per year.These results show that the response of capital investment to outsourcing isindeed varying over time.Table 1.5 show the main results of the regression using investment peremployee as dependent variable. Again, the estimates of β using the out-sourcing dummy, Oscore, with or without firm fixed effects are all negativeand statistically significant. Based on the results of regression (3), usingoutsourcing dummy as the measure of outsourcing, an outsourcing firm re-duces investment by 447 dollars per employee per year relative to a non-outsourcing firm. The regression result in (4) using oscore shows that onepercentage change in oscore will make investment per employee decrease7.26 dollars per year. The investment gap is increasing over time and mostof this can be attributed to the increased reaction to outsourcing.Table 1.6 shows the main results of the regression using sales per em-ployee as the dependent variable. The estimates of β using the outsourcingdummy, Oscore, with or without firm fixed effects are all positive and statis-tically significant. Based on the results of regression (3), using outsourcingdummy as the measure of outsourcing, an outsourcing firm’s investment is15,590 dollars per employee per year more than a non-outsourcing firm. Theregression result in (4) using Oscore shows that a one percentage change inOscore will make output per employee increase 179.35 dollars per year. Theinvestment gap is increasing over time and most of this can be attributedto the increased reaction to outsourcing.1.4.3 RobustnessI consider robustness checks in terms of several concerns in the baselineregression. For investment and capital regression, I consider controllingfor measures of investment opportunity and financial constraint. For sales141.4. Empirical Evidenceregression, I consider measures of productivity, and market power as controlvariables. I also check robustness in terms of controlling different measuresof firm size.Table 1.7 shows the results by considering different proxies for investmentopportunities in the literature. Firms’ investment may respond to invest-ment opportunities, such as a demand increase, which can be proxied bysales growth rate. Also, R&D expenses may be a measure of change in pro-ductivity and may be related to physical investment opportunity. Moreover,the market to book asset ratio may be a reflection of investment opportu-nity, since Tobin’s q is high. Using the market to book ratio, only 1,212observations are left in the sample since not every firm’s stock market priceis available. Column (1) and (4) use the sales growth rate, Column (2) and(5) use R&D to total asset, while Column (3) and (6) use the market to bookratio. I report the results both for controlling asset or sales as a measure offirm size. The estimates of these are all non-significantly different from zerowhen outsourcing and the interaction of outsourcing and time are controlledfor. The estimates of β are still negative and significant.Firms’ investment may also be limited by financial constraints. I considertwo measures of financial constraints here: debt ratio and interest expenserate. Debt ratio is measured by the ratio of total liability over total asset,and interest expense rate is measured by total interest expense over totalliability. Only 1,512 observations contain information of interest expense,since missing observations are dropped from the regression. Table 1.8 showsthe results of these regressions. Investment is negatively related to financialconstraints since the coefficients on debt ratio and interest expense rate areall negative. However, when firm fixed effects, outsourcing and the timetrend are controlled for, these estimates become not statistically differentfrom zero, although still negative. The estimates of β are still negative andsignificant, which shows the robustness of the relationship.Table 1.9 contains the results of the sales regression by controlling formeasures of productivity and profitability. As in the investment regressionanalysis, I consider R&D expenses as a proxy for productivity, while usingthe ratio of operating income before depreciation over sales as the measureof profitability. In the literature, this is also interpreted as a measure ofmarket power. The estimates on the interaction term are still positive andsignificant, both controlling for asset or sales as a measure of firm size. Onesurprising finding is that the coefficient on R&D to asset ratio is negative,which means the contribution of R&D to output per employee is negativeafter controlling for the effect of outsourcing and firm fixed effects.The results for capital per employee are similar to the investment regres-151.5. Conclusionsion. These results show that the baseline relationship between investment,capital, sales and outsourcing over time is robust to different specifications.1.5 ConclusionAlthough there are a number of theories about the determinants and effectsof outsourcing, this paper is centered around testing the model implicationsof a simple outsourcing model with a fixed cost and variable cost. To makethe model simple and the prediction more relevant to the data, we abstractaway from the interaction between the outsourcing firms and their tradepartners, and introduce search frictions, incomplete contracts and invest-ment specificity just as costs faced by outsourcing firms. This model hasseveral clear testable predictions.The empirical results of this paper show that in the US computer manu-facturing industry, outsourcing is a widespread phenomenon and over time,outsourcing is increasing at both the extensive and intensive margins. Largerfirms tend to outsource more while also having a lower investment laborrate. Over time, The investment gap between outsourcing firms and non-outsourcing firms is increasing while the difference in the output labor ratiois also increasing. Outsourcing firms increasingly invest less while produc-ing more in a per employee sense. Also, the approach to the data can beapplied to other industries and firm-level datasets to obtain information ofoutsourcing from narrative records.I test the implications using the panel regression and also report theresults controlling for firm fixed effects, I check the robustness of the resultsby using a variety of other specifications and controlling for different firm-level characteristics. The results in this paper provide a number of usefulempirical patterns for theories of outsourcing or vertical integration to con-front. The model in this paper suggests a simple way of examining thesepatterns qualitatively. Existing theories (like PRT, TCE) may explain whyinvestment is less in outsourcing firms, but these theories are not relevantin terms of predicting the extensive and intensive margin of outsourcing,the relationship between firm size and outsourcing, and the difference be-tween outsourcing firms and non-outsourcing firms in investment and outputdynamics. To explain these patterns quantitatively with a richer theory ap-pears to be interesting areas of future research.161.6. Tables and Figures1.6 Tables and FiguresFigure 1.1: Nonresidential Investment from BEA and COMPUS-TAT171.6. Tables and FiguresFigure 1.2: Investment Rates by Industry in COMPUSTATFigure 1.3: Investment Rates in Manufacturing181.6. Tables and FiguresFigure 1.4: Sample ComparisonFigure 1.5: Time Trend of Outsourcing191.6. Tables and FiguresFigure 1.6: Time Trend of Average OscoreFigure 1.7: Time Trend of Investment Rate201.6. Tables and FiguresFigure 1.8: Outsourcing by Size211.6. Tables and FiguresFigure 1.9: Outsourcing by Age221.6. Tables and Figures0 5 10 15 over Time0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2020406080100120140 Distribution of s0 5 10 15 Firm Ratio0 5 10 15 2000. ShareAverage Outsourcing Share  OutsourcersOutsourcers:period0Figure 1.10: Shocks and Outsourcing Trend0 2 4 6 8 10 12 14 16 18 Capital to Labor Ratio  AllOutsourcersNon−outsourcersOutsourcers:period0Figure 1.11: Capital Stock and Investment per Labor231.6. Tables and Figures0 2 4 6 8 10 12 14 16 18 200.720.7250.730.7350.740.7450.750.755TimeY/LAverage Sales to Labor Ratio  AllOutsourcersNon−outsourcersOutsourcers:period0Figure 1.12: Sales per Labor241.6. Tables and FiguresTable 1.1: Descriptive StatisticsVariable Mean Std Dev Min MaxOutsource 0.48 0.50 0 1Oscore 0.38 0.44 0 1Employees 7040 23742 7 324600I (mil) 0.89 3.06 0 41.72Sales(mil) 13.21 75.83 0.00014 1323.96I/Emp (thou) 13.07 14.68 0 217Sales/Emp (thou) 121.29 172.69 0.16 1474Capital/Emp (thou) 26.06 25.67 0.98 565.57Asset(mil) 2340 9240 0.57 124503Age(time after IPO) 10.95 8.57 1 49R&D/Asset 0.14 0.14 0 2.01Net Trade Credit -0.10 0.11 -0.56 1.09Liability/Asset 0.41 0.33 0.04 7.96Source: Author’s calculation using COMPUSTAT annual data for U.S. companiesand the narrative records in LexisNexis Academic database.Notes: Investment is deflated by chained price index of non-residential investmentfrom BEA. Sales is deflated by chained price index of Computer and peripheralequipment from BEA. Capital stock is constructed using perpetual inventorymethod with initial value set to the net property, plant and equipment and de-preciation rate of 20%. Net Trade Credit received is the difference of accountpayable and acount receivable over total sales. The sample contains 1787 obser-vations on 166 firms for computer and peripheral industry, SIC 3670-3677. Thesample period is 1987-2010.251.6.TablesandFiguresTable 1.2: Baseline Parameter ValuesParameter Description Valueα Intermediate good production function parameter 0.3 Clementi & Palazzo (2010)β Final good production function parameter 0.3 Clementi & Palazzo (2010)θ Return to scale 0.7 Clementi & Palazzo (2010)δ Depreciation 0.15 Bloom (2009)wm manufacturing labor wage rate 0.4 Clementi & Palazzo (2010)wn non-manufacturing labor wage rate 0.4 Clementi & Palazzo (2010)r capital rental rate 0.05 Clementi & Palazzo (2010)ρ autocorrelation of idiosyncratic productivity 1 Tian (2001)σ standard deviation of idiosyncratic shock 0.02 Tian (2001)c Fixed cost for Outsourcing 0.1261.6. Tables and FiguresTable 1.3: Linear Probability regression of Outsource on Firm SizeNo firm fixed effects With firm fixed effects(1) (2) (3) (4) (5) (6)Intercept -0.79** -0.71** -0.27** -1.23** -0.83** -0.68**(-11.97) (-7.09) (-4.38) (-15.27) (-9.74) (-9.92)log(Asset) 0.21** −− −− 0.14** −− −−(16.01) −− −− (9.93) −− −−log(Sale) −− 0.19** −− −− 0.05** −−−− (9.54) −− −− (3.56) −−sizebin 2 −− −− 0.17** −− −− 0.02−− −− (2.84) −− −− (0.51)sizebin 3 −− −− 0.18** −− −− 0.003−− −− (3.09) −− −− (0.1)sizebin 4 −− −− 0.23** −− −− 0.02−− −− (3.79) −− −− (0.60)sizebin 5 −− −− 0.35** −− −− 0.07*−− −− (5.68) −− −− (1.75)sizebin 6 −− −− 0.40** −− −− 0.05−− −− (6.22) −− −− (1.03)sizebin 7 −− −− 0.55** −− −− 0.09*−− −− (8.24) −− −− (1.82)sizebin 8 −− −− 0.71** −− −− 0.20**−− −− (10.18) −− −− (3.35)sizebin 9 −− −− 0.83** −− −− 0.25**−− −− (10.74) −− −− (3.64)sizebin 10 −− −− 1.12** −− −− 0.29**−− −− (11.55) −− −− (3.45)log(age) -0.05** -0.07** -0.06** -0.02** -0.04** -0.05**(-3.90) (-4.86) (-3.94) (-1.26) (-2.03) (-2.56)log(emp) -0.22** -0.20** -0.16** -0.17** -0.08** -0.07**(-15.34) (-9.29) (-11.45) (-10.23) (-4.54) (-4.48)Time 0.02** 0.03** 0.03** 0.02** 0.02** 0.02**(9.66) (12.45) (12.42) (7.21) (10.55) (10.58)Dependent variable is outsourcing dummy. Standard errors are clustered at firm level,t-statistics in parentheses. Regression (1) to (3) are pooled regressions, while (4) to (6)includes a full set of firm fixed effects.271.6. Tables and FiguresTable 1.4: Main Results: Capital Per EmployeeK/Emp (1) (2) (3) (4)Outsource 120.22** −− 120.93** −−(5.12) −− (3.13) −−Oscore −− 211.97** −− 349.29**−− (6.04) −− (3.01)Outsource × Time -4.72** −− -5.38** −−(-3.00) −− (-2.46) −−Oscore× Time −− -7.17** −− -7.89**−− (-3.67) −− (-2.00)Time 2.37** 0.82 -0.58 -0.62(2.44) (0.98) (-0.45) (-0.64log(Asset) 28.13** 31.03** 33.79** 32.23**(10.11) (11.16) (3.70) (5.59)Standard errors are clustered at firm level, t-statistics in parentheses. Regression(1) and (2) are pooled regressions, while (3) and (4) includes a full set of firmfixed effects.281.6. Tables and FiguresTable 1.5: Main Results: Investment Per EmployeeI/Emp (1) (2) (3) (4)Outsource 69.92** −− 73.32** −−(5.82) −− (3.31) −−Oscore −− 115.43** −− 138.51**−− (6.78) −− (3.94)Outsource × Time -3.48** −− -4.47** −−(-4.08) −− (-4.01) −−Oscore× Time −− -4.55** −− -7.26**−− (-4.53) −− (-4.00)Time 0.69 -0.28 -1.55** -1.89**(1.32) (-0.59) (-2.84) (-3.47)log(Asset) 19.21** 20.5** 32.57** 32.63**(11.57) (11.82) (8.58) (8.52)Standard errors are clustered at firm level, t-statistics in parentheses. Regression(1) and (2) are pooled regressions, while (3) and (4) includes a full set of firmfixed effects.291.6. Tables and FiguresTable 1.6: Main Results: Sales Per EmployeeSales/Emp (1) (2) (3) (4)Outsource -968.7** −− -1329** −−(-7.98) −− (-6.04) −−Oscore −− -1169.02** −− -1167.17**−− (9.38) −− (-4.02)Outsource × Time 114.26** −− 155.9** −−(9.98) −− (11.94) −−Oscore× Time −− 131.9** −− 179.35**−− (9.38) −− (11.42)Time 115.3** 117.34** 79.93** 81.66**(22.04) (22.59) (13.49) (14.32)log(Asset) 127.72** 146.26** 235.4** 238.7**(8.39) (9.20) (5.92) (6.15)Standard errors are clustered at firm level, t-statistics in parentheses. Regression(1) and (2) are pooled regressions, while (3) and (4) includes a full set of firmfixed effects.301.6. Tables and FiguresTable 1.7: Investment: Controlling for Investment OpportunityI/Emp (1) (2) (3) (4) (5) (6)Outsource 75.14** 83.13** 57.56* 84.29** 90.85** 72.57**(2.98) (3.75) (-1.82) (3.22) (4.05) (2.30)Outsource × Time -4.72** -5.10** -3.92** -4.75** -5.15** -4.48**(-3.42) (-4.38) (-2.38) (-3.45) (-4.38) (-2.69)sales growth 13.25 −− −− 9.20 −− −−(0.50) −− −− (0.36) −− −−R&D/Asset −− 68.46** −− −− 32.63* −−−− (3.20) −− −− (1.63) −−M/B ratio −− −− 0.74 −− −− 0.43−− −− (1.06) −− −− (0.67)Time -1.30** -1.92** -3.71** -0.49 -0.69 -2.60**(-2.07) (-3.48) (-4.19) (-0.74) (-1.07) (-2.81)log(Asset) 32.58** 35.39** 38.00** −− −− −−(7.82) (9.22) (7.02) −− −− −−log(Sales) −− −− −− 25.32** 25.33** 31.25**−− −− −− (4.61) (5.21) (5.00)Standard errors are clustered at firm level, t-statistics in parentheses. All regressions include a fullset of firm fixed effects.311.6. Tables and FiguresTable 1.8: Investment: Controlling for Financial ConstraintsI/Emp (1) (2) (3) (4)Outsource 74.27** 69.84** 83.84** 79.90**(3.32) (2.96) (3.68) (3.30)Outsource × Time -4.78** -3.98** -4.86** -3.96**(-4.05) (-3.26) (-4.11) (-3.24)Debt Ratio -6.07 −− -19.38 −−(-0.57) −− (-1.71) −−Interest Expense −− -166.30** −− -173.97*−− (-1.63) −− (-1.78)Time -1.48** -2.11** -0.45 -1.07(-2.61) (-3.41) (-0.67) (-1.43)log(Asset) 32.36** 36.25** −− −−(8.28) (8.11) −− −−log(Sales) −− −− 23.98** 25.82**−− −− (4.72) (4.33)Standard errors are clustered at firm level, t-statistics in parentheses. All regres-sions include a full set of firm fixed effects.321.6. Tables and FiguresTable 1.9: Sales: Controlling for Productivity and ProfitabilitySales/Emp (1) (2) (3) (4)Outsource -1248.11** -1357.44** -1176.36** -1263.76**(-6.30) (-6.41) (3.68) (-5.76)Outsource × Time 151.47** 155.84** 151.83** 157.23**(12.67) (11.89) (12.60) (11.94)R&D/Asset -468.23** −− -572.22** −−(-2.71) −− (-3.38) −−Profitability −− 0.38 −− -6.30**−− (0.31) −− (-2.99)Time 77.93** 79.99** 77.35** 77.08**(12.71) (13.44) (12.97) (12.81)log(Asset) 239.17** 235.06** −− −−(5.71) (5.88) −− −−log(Sales) −− −− 259.348** 276.11**−− −− (6.18) (6.33)Standard errors are clustered at firm level, t-statistics in parentheses. All regres-sions include a full set of firm fixed effects.33Chapter 2Firm-level Risk-taking andVolatility Trends2.1 IntroductionIn this chapter I study the relationship between aggregate output volatilityand firm-level volatility. Many researchers have noticed that the reductionin gdp volatility, referred to as the great moderation, came to an end withthe Great recession.8 Gone with the moderation, is also the volatility diver-gence between macro measures of volatility and micro measures of volatility.During the period when GDP volatility was decreasing, it has been docu-mented that volatility in sales growth at the firm level, employment growthat the firm level, as well at individual level stock returns for publicly tradedfirms was increasing. This pattern has been termed as the volatility diver-gence. However, the so called “volatility divergence” is not robust when weinclude recent data. This paper first shows that the “volatility divergence”in the macro and micro level had come to an end as early as the year 2000:the volatility of the growth rates of Real GDP and median volatility of salestend to move together after 2000. Volatility divergence was also nonexis-tent before the Great Moderation. Therefore, the actual divergence onlyhappened during the late 1980s and through the 1990s. The main goal ofthis paper is to explain the temporary volatility divergence and subsequentconvergence.How to explain the relationship between the micro and macro volatil-ity? Based on observed volatility divergence in the past, some researchershave suggested it is very possible that the diverging trend in macro andmicro volatility is caused by a common force: the “creative destruction”process. This includes deregulation and technology innovation which makefirms invest more in idiosyncratic R&D and replace the leader in that in-8For the documentation and possible causes of the Great Moderation, please referto Blanchard and Simon (2001) McConnell and Perez-Quiros (2000), Stock and Watson(2003) and Jaimovich and Siu (2009). For recent discussion of the end of the GreatModeration, please see Clark (2009)342.1. Introductiondustry faster, hence firm’s idiosyncratic volatility increase. But this processalso make firms more heterogeneous and less correlated with each other orless correlated with aggregate shocks. Therefore, the aggregate volatilitydecreases. Comin and Phillipon (2006) find positive relation between R&Dand firm-level volatility. Also, Chun, Kim, Lee and Morck (2008) finds firm-specific stock return volatility is higher in industries that invest more ininformation technology, and Chun, Kim, Lee and Morck (2004) argues thatincreasing use of information technology can explain higher stock returnand sales growth volatility. Comin and Mulani (2009) build an endogenousgrowth model and point to the R&D subsidy as the driving force of thischange.In this paper, I also show several related data patterns: after 2000,Number of IPOs per year was much smaller than during the late 1980s andthrough the 1990s. Considering new listed firms are generally more volatile,this may cause the volatility reduction in publicly traded firms. Also, firm-level volatility in the 1980s is in fact slightly decreasing if one excludes thenew firms which come to the sample in the 1980s. Over time, more volatilefirms go public, get financed, research more and invest more. Motivated bythese facts, I develop a simple model to provide a possible explanation ofwhat we see during a technology boom like the 1990s, offering a story forboth the volatility divergence and the subsequent convergence: I considerthe uncertainty in future productivity gains of a technology as the drivingforce of firms taking risk. The building blocks of the model are: firmschoose the allocation of investment into risky and riskless projects, with riskyprojects having a short-run idiosyncratic term and long-run common risk.Idiosyncratic uncertainty in the short-run is not totally diversifiable becauseof endogenous market incompleteness. I model the uncertainty in technologyinnovation as a process which contains two parts: idiosyncratic risk andgeneral productivity growth uncertainty. Using a model with endogenousmarket incompleteness, I show that uncertainty about the future permanentchange growth rate which the technology innovation can bring will causean increase in the expected value of the risky investment. This leads tomore investment in the risky project compared to the risk-free productiontechnology. Pastor and Veronesi (2006) show that this similar mechanismmay have generated the bubble-like behavior in the NASDAQ firm stockprices. I show that this may also cause an increase in individual risk-taking,and will increase the firm-level volatility while making market more completeand better diversified, hence reducing the aggregate volatility. However, asthe uncertain effect of the new technology realizes, firm-level risk-takingdrops while aggregate volatility increases. When the participation is low352.1. Introductionenough, aggregate volatility and firm-level volatility move together.My work is consistent with the general view that technology change andinnovation can explain firm-level volatility in the literature. If technologyinnovation is the main driving force behind the “volatility divergence” after1980s, then the subsequent convergence may be an indicator of a slowdownof the innovation process. Median firm-level R&D intensity was growingdramatically during the 1990s and this was caused by firms from the 1990cohort, then it holds almost constant during the first decade of the 21stcentury. Also, at the macro level, non-residential investment normalized byGDP rose dramatically during 1990s, dropping during the dot-com crash,and increasing steadily after the 2000, but it never returned to the peakin the 1990s before the 2007 crisis. These findings are consistent with theview that the investment opportunities in the non-residential sector are few,which in combination with the low risk premium during the period, causeda boom in the residential investment. Moreover, there is no evidence thatR&D projects are idiosyncratic for firms, or individual risk taking is totallynot correlated with each other. An alternative way of thinking R&D is thatas a proxy for firm’s risk taking behavior as Bargeron, Lehn and Zutter(2010) and Li, Griffin, Yue and Zhao (2013).Moreover, my work can also help reconcile the debate of whether the ob-served firm-level volatility trend is just a selection bias. Davis, Haltiwanger,Jarmin and Miranda (2005) notice that there is a “cohort effect” in thevolatility of the COMPUSTAT firms. New listings tend to be more volatile.They also argue that the increase of volatility in the publicly traded firmsmay be a result of sample selection: More young and volatile firms go public,which make publicly traded firms more volatile and privately held firms lessvolatile. However, they did not give a clear explanation of why this selec-tion happened. Also, the employment share in publicly traded firms is stableover time, which put the selection hypothesis in question. Since high risk,high return projects can not be financed by debt contracts,(Brown, Fazzari,Petersen (2009) ) it is natural for firms to use public equity financing forsuch investments.The rest of the paper is organized as follows. Section 2 presents theempirical results and shows the volatility divergence was only a temporaryphenomenon. Section 3 shows the theoretical framework which incorporatesthe empirical facts. Section 4 concludes.362.2. Firm-level Volatility and Macro Volatility2.2 Firm-level Volatility and Macro VolatilityIn this section, I show that the “volatility divergence” of aggregate outputand the firm-level output is not in the data since 2000. Also, for the periodin which there is a positive time trend, the increase in the firm-level volatilityis largely driven by new firms added to the sample. This is consistent withthe view that volatility on the firm level is driven by entry and exits of theinvestment in risky assets, especially innovation-driven risk taking. Duringthe tech boom starting from the 1980s to the crash of “dot-com bubble”, thecorrelation of individual firm’s growth to GDP growth dropped significantlyand then come back to the level before the boom.2.2.1 Data and MeasuresFor firm-level data, I use the COMPUSTAT firms, starting from 1970 to2009. The main measure I use for firm-level activity is the annual salesgrowth rate, which is the same as Comin and Philippon(2005). However,to investigate data patterns after the year 2000, I use a 5-year, instead of a10-year, rolling standard deviation9 of the sales growth rate, that isσit =[15+2∑τ=−2(st+τ,i − s¯t,i)2] 12where s¯t is the average growth rate between t− 2 and t+ 2. To measurethe representative firm’s volatility, I use the median of the 5-year rollingvolatility of sales growth rates:σft = mediani {σit}For macro variables, I use the five-year rolling standard deviation of RealGDP growth rate.σat =[15+2∑τ=−2(γt+τ − γ¯t)2] 122.2.2 Volatility in COMPUSTAT FirmsFigure 2.1 shows the median volatility of COMPUSTAT firms’ real salesgrowth and the aggregate volatility of real GDP growth. The “divergence”9The purpose here is to capture volatility in shorter terms, which shows both volatilitytrends and cycles. Constrained by sample size after year 2000, using 5-year rolling standarddeviation also shows the volatility dynamics after 2000 with more observations.372.2. Firm-level Volatility and Macro Volatilityonly happens after the mid-1980s and the two measures start to move inthe same direction around the 2000. What can really be called a “diver-gent” trend period is from the mid-1980s to the late 1990s, which is thetechnology boom period. During this period , there was “the Great Mod-eration” as noted by the literature. Also, starting from 1990, firm-levelvolatility increased significantly until 2000. After 2000, both volatility mea-sures dropped and both increased during the 2007 crisis.2.2.3 Cohort and Number of New FirmsFigure 2.2 shows the number of firms by each cohort. During the 1990s,there were a large number of IPOs, and the number is larger than previousdecades. The relationship between the large increase in firm-level volatilityand the increase in number of IPOs is unclear, however, as several studiessuggested10 , new issuing firms are generally riskier.Figure 2.3 shows that younger cohorts indeed often have higher volatility.Increase in firm-level volatility in the 1980s is mostly driven by new firmsadded to the sample in the 1980s. Only during the 1990s, we see a steadyincrease in volatility in every cohort. But the question is whether this iscaused by selection or something else. Especially, whether this selection isexogenous. This raises the question of why firms go public.2.2.4 Firm SizeSmall firms are generally more volatile. Figure 2.4 shows the volatility ofdifferent firm sizes. Here I measure firm size by number of employees. Thecutoffs for each group are constant over time. Also, during this period, wehave more small firms that went public. The number of firms in the smallertwo size groups increase a lot especially during the 1990s. Figure 2.5 showsthis pattern.2.2.5 IndustryIn this section I show the volatility in different sectors of the economy.Sectors are defined according to the SIC code of the firm. Figure 2.6 andFigure 2.7 show number of firms and volatility in different sectors. Withineach sector, the cohort effect is still evident. That is, new comers often have10For example, Fama and French (2004) show that new lists are riskier, Brown andKapadia (2007) also argue that the increase in idiosyncratic risk measured by stock returnsis mostly driven by new listings382.2. Firm-level Volatility and Macro Volatilityhigher volatility. In sectors like consumer non-durables and durables, theupward trend in volatility is almost totally caused by the new firms that enterthe sample. The only sector in which every cohort increase its volatility inthe 1990s is Business Equipment. This sector includes computers, softwareand electronics, which contribute to the technology boom. This is also amain driving force of the volatility increase in COMPUSTAT firms.2.2.6 R & D Intensity and Investment BehaviorWhat causes the volatility of the firms? Existing studies have pointedto product market competition, selection and technology innovation. Itis worth noting that, firm’s investment behavior should be endogenous tothe riskiness of the projects. Although we do not observe firm’s invest-ment opportunity, one can observe the investment decisions. More R&Dand more intensive capital expenditure generally can be a proxy of firm-level risk taking. I explore the relationship between risky investment andfirm-level volatility in this section. I measure R&D intensity using the ratioof R&D over sales, and investment intensity using capital expenditure overnet property, plant and equipment.Firms with high R&D intensity tend to be more volatile. Also, youngercohorts, smaller firms and the business equipment sector are more R&D in-tensive. The increase in volatility in 1990s is caused by both more R&D byexisting firms and new entrants being more R&D intensive firms. However,R&D only happens in certain industries. For a lot of industries, we do nothave any R&D expenditure observations. Another good proxy for risk takingis investment over capital ratio, which is measured by capital expenditureover net property, plant and equipment. This gives us more observationsand also in different sectors. Again, smaller firms, younger firms and busi-ness equipment firms have a higher investment to capital ratio. Figure 2.8 -Figure 2.10 show the pattern of R&D intensity of different size groups, sec-tors and cohorts. Figure 2.11 - Figure 2.13 show the pattern of investmentintensity.2.2.7 Estimating the Time-varying Firm-level VolatilitySince including a full set of time, age and cohort dummy variables will causemulticollinearity, I assume that firms’ volatility is constant after certain age.(measured by listing years)392.2. Firm-level Volatility and Macro Volatilityσit = β1sizeit + Denteryear,i · γ1 + Dage,i,t · γ2 + Dindustry · γ3+ Dt · γ4 + itIn terms of the upper bound of the age effect on volatility, I check whetherthe assumption of different cutoffs affects the result. Figure 2.14 to Fig-ure 2.16 show the estimated age effects using cutoffs of 15 years, 20 yearsand 30 years. Volatility is lower as firms grow older, and this age effect ismost apparent in the first 12 years. As firms mature, the volatility becomeslower while the marginal age effects also become less.Figure 2.17 to Figure 2.19 show the estimated coefficients on time dum-mies using different assumptions of the age effect cutoff. Controlling forother effects, firm-level volatility has a general upward trend before the year2000. However, the volatility cycle seems more apparent, and it is differentfrom the general business cycle measured by GDP. The result is robust todifferent specifications of the age cutoffs.2.2.8 Baseline Regression ResultsTo get a sense of the volatility trend after controlling composition, I run thefollowing regression:σit = β1risktakingit + β2sizeit + β3ageit+Dentryyear,i · γ1 + Dt · γ2 + Dindustry · γ3 + itThe results are presented in Table 2.1. I use both employment and totalassets as the measures of size. I use R&D intensity and investment intensityto measure firm-level risk taking. The coefficients on the time dummiesfrom the baseline regression are plotted in Figure 2.20. After controlling forinvestment intensity, we have a downward trend in the firm-level volatility.This shows that the upward sloping volatility trend is mainly driven byindividual risk taking behavior. The downward sloping trend in Figure 2.20reflects other factors in the literature such as better inventory management,better policy, smaller aggregate shocks or age composition may play animportant role in the downward trend of macro volatility. However, if therewas no change in risk taking behavior, we should observe firm-level volatilitydecrease over time as well.402.2. Firm-level Volatility and Macro Volatility2.2.9 Correlation between Firm-level and GDP GrowthWhat makes the average firm volatility and aggregate volatility diverge orconverge? Let us assume the following reduced-form determination of afirm’s sales:sit = aitµit + bittwhere sit is firm i’s sales growth rate over time period t, µit is an id-iosyncratic shock which only affects firm i’s growth rates, t is an aggregateshock which affects every firm in the economy and bit is firm i’s sensitivityto the aggregate shock at time t. Then the firm-level volatility in the salesgrowth rate is given byσ2s = a2itσ2µ + b2itσ2Aggregate growth rate is a weighted average of all firms’ growth rates:yt =∑iωitsit =∑iωitaitµit +∑iωitbittand hence, the covariance of an average firm’s sales growth with theaggregate growth is given by:Cov(sit, yt) = ωitaitσ2µ + bitσ2Taking a weighted average:∑iωitCov(sit, yt) =∑iω2itaitσ2µ + (∑iωitbit)σ2In case of a large enough sample, the first term of the above equationwill asymptotically go to 0, and this measure will be an indicator of averageof bit over time.Also, the aggregate variance isV ar(yt) =∑iω2ita2itσ2µ + (∑iωitbit)2σ2Therefore, changes in the average of bit will also change the aggregate volatil-ity; However, with a sufficiently large sample, changes in ait will not affectaggregate volatility.Figure 2.21 shows that the median and mean correlation of COMPUS-TAT firms’ sales growth rate with the Real GDP growth rate, with the 25th412.3. An Illustrative Modelpercentile and the 75th percentile plotted. During the 1990s, firms’ corre-lation with the aggregate had a downward trend, which could have beencaused by a drop in the propagation of the shocks if we assume the secondmoment of aggregate shocks are stable over this period. This suggests thattechnology innovation and adoption may cause individual firms to take morerisk while reducing the correlation with the aggregate shocks.2.3 An Illustrative ModelIn this section I present a model with several features: First, firms’ riskyinvestment contain an short-term idiosyncratic part and a long-term gen-eral productivity gain. Firms make decisions according to an expected dis-counted cash flow approach. Second, firms are restricted by a minimum sizerequirement when investing in risky projects as per Acemoglu and Zilibotti(1997), therefore the idiosyncratic risk can not be fully diversified. This willgenerate limited participation in risky investment11 in the first period andcause aggregate volatility.2.3.1 Entrepreneurs and Risky ProjectsTime is discrete. In each period a continuum of entrepreneurs are born withan endowment Kj = K¯, where entrepreneurs are indexed by j ∈ [0, 1], andthey have a risky investment plan. The projects are infinitely lived andentrepreneurs only live for two periods. They sell the project to the nextgeneration and consume. The project’s payoff is as follows:At time t,YT (j) ={Rt(j)Fjt + f(φjt) if T = tA(1 + g˜)T−tFjt + f(φjt) if T > twhere f(•) is a risklesss production technology with f ′(•) > 0 and f ′′(•) < 0,F (jt) is the investment in the risky project, φ is the investment in the risklessproduction and the risky project has a random growth rate of g˜ in the future.The uncertainty of the growth rate make the value of risky project higher,which is similar to the idea behind Pastor and Veronesi (2006), althoughPastor and Veronesi (2006) talks about the asset price bubble. Another wayof thinking this is that higher asset price implies lower cost of equity, whichcan also induce more investment. Here the uncertainty is in the growth rate11Since the firm index is also the state space, considering a portfolio of all firms, limitedparticipation means incomplete market.422.3. An Illustrative Modeland I do not introduce irreversible investment, therefore the uncertainty donot reduce investment as in the real option literature.The realization of the random variable R(j) is a set of firms on [0, 1],an interval I with measure κ, so that the firm’s pay off at the end of timeperiod t is given byRt(j) ={R if j ∈ I0 if j /∈ IThe exact timing of the events is illustrated in Figure 2.22.The firm index j is also the index of the state of the world in the venturingperiod t. If κ → 0, each firm represents a single Arrow security. Then fora fraction of firms n that invest in the risky asset, aggregate return will beR with probability n and 0 with probability (1− n). The realization of theuncertainty is illustrated in Figure 2.23.To have aggregate uncertainty in the first period, I adopt the minimumsize assumption:Fjt ≥D1− γ (j − γ) if Fjt 6= 0The relationship between the minimum size requirement and the optimalinvestment in risky asset is illustrated in Figure 2.24.Therefore, as j increases, minimum risky investment required is in-creased. Thus for firm j, the optimal investment problem is:maxFjt,φjtE[Rt(j)Fjt] + β∞∑T=t+1A[βδ(1 + g˜)]TFjt + f(φjt)s.t. Fjt + φjt = KFjt ≥D1− γ (j − γ) if Fjt 6= 02.3.2 Optimal Risk Taking and VolatilityThen the optimality condition is:E[Rt(j)] + βE[A1βδ − 1− g˜]= f ′(K − Fjt)orκR+ βE[A1βδ − 1− g˜]= f ′(K − F ?jt)432.3. An Illustrative ModelNow that the optimal investment for each firm j is determined, theequilibrium number of participating firms or measure n? is given byF ?jt =D1− γ (n? − γ)Then aggregate output in period t is∫ n?0[Rt(j) + f(φjt)] dG(j)and conditional output variance of participating and optimal investmentF ∗V ar(Yj) = κ(1− κ)R2(F ∗)2which is increasing in risky investment, F ∗.Aggregate output volatility for a given investment in F ∗, is given asV ar(Yt) = V ar(∫ 10Yt(j)dj)= n∗(1− n∗)κ2R2(F ∗)22.3.3 Comparative StaticsThere are three propositions we can deduce from this expression:Proposition 1 All else being equal, an increase in the mean future produc-tivity growth rate, g˜, will result in an increase in risky investment and anincrease in participation in risky projects; that is, both F ?jt and n? increase.This proposition directly follows the optimality condition. Optimal in-vestment is increasing in expected growth rate. Equilibrium participationrate can be calculated by substituting the optimal investment back into theminimum size condition. Since n? is increasing in F ?jt , n? is also increasingin expected growth rate.Proposition 2 A mean-preserving spread in the future productivity growthrate g˜ will result in more risky investment and more participation in riskyprojects; that is, both F ?jt and n? increase.This proposition is resulted from using Jensen’s inequality on the opti-mality condition. SinceA1βδ − 1− g˜is a convex function in g˜, E[A1βδ − 1− g˜]is increasing in the uncertainty of g˜. Therefore, both F ?jt and n? are increas-ing in the uncertainty of g˜.442.4. ConclusionProposition 3 Under certain parameter assumptions, aggregate volatilityat venturing period t is decreasing in the participation rate n?, that is,∂V ar(Yt)∂n∗< 0.The intuition is that since there is convexity in the valuation in terms offuture productivity growth rate and the firm is required to invest in advance,uncertainty in productivity growth will cause more risk-taking. This alsoleads to more participation and makes the market more complete, therebyreduce the aggregate volatility of output in the short run.In order to have individual risk taking cause aggregate moderation∂V ar(Yt)∂n∗< 0we need:n∗ >3 + 2γ +√16γ2 − 4γ + 98When γ = 0, the condition degenerates ton∗ >34The technology boom during 1980s and 1990s can be viewed as a periodwhen new ideas and projects brought higher but more uncertain returns tolong-term investment. More firms invested in risky and promising technol-ogy, hence a higher n∗. When the participation rate exceeds the threshold,aggregate output volatility becomes smaller while individual volatility in-creases. This phenomenon is possible when both the expectation of growthrate and uncertainty of growth rate are large. At the end of the technol-ogy boom, or after the year 2000, aggregate output volatility and individualvolatility returned to positive correlation.2.4 ConclusionThis paper found that the divergent trends in publicly traded firms andmacro volatility have come to an end after year 2000. Before 2000, morerisky firms go public, research more and invest more. I link this to the risktaking behavior and the new investment opportunity during the technol-ogy revolution period. More micro-level risk taking makes the market morecomplete hence less aggregate volatility. This mechanism also predicts thatwhen firms invest more into the risky project at the intensive margin, more452.4. Conclusionfirms would participate in the risky project at the extensive margin. Ag-gregate output volatility should move in the opposite direction of the microlevel volatility. I conjecture that better risk sharing and diversification canbe a mechanism which contributes to the Great Moderation, especially inthe 1990s.462.5. Tables and Figures2.5 Tables and FiguresFigure 2.1: Volatility at Firm-level v.s. Volatility of Real GDP472.5. Tables and FiguresFigure 2.2: Number of Firms by CohortFigure 2.3: Volatility by Cohort482.5. Tables and FiguresFigure 2.4: Number of Firms by SizeFigure 2.5: Volatility by Size492.5. Tables and FiguresFigure 2.6: Number of Firms by SectorFigure 2.7: Volatility by Sector502.5. Tables and FiguresFigure 2.8: Investment Intensity by SectorFigure 2.9: Investment Intensity by Size512.5. Tables and FiguresFigure 2.10: Investment Intensity by CohortFigure 2.11: R&D Intensity by Sector522.5. Tables and FiguresFigure 2.12: R&D Intensity by SizeFigure 2.13: R&D Intensity by Cohort532.5. Tables and FiguresFigure 2.14: Plot of Age Effects (Age Cuttoff = 15)Figure 2.15: Plot of Age Effects (Age Cuttoff = 20)542.5. Tables and FiguresFigure 2.16: Plot of Age Effects (Age Cuttoff = 30)Figure 2.17: Plot of Estimates on Time Fixed Effects (Age Cuttoff= 15)552.5. Tables and FiguresFigure 2.18: Plot of Estimates on Time Fixed Effects (Age Cuttoff= 20)Figure 2.19: Plot of Estimates on Time Fixed Effects (Age Cuttoff= 30)562.5. Tables and FiguresFigure 2.20: Firm-level Volatility after Controlling for Compositionand Capital IntensityFigure 2.21: Rolling Correlation Coefficients572.5. Tables and Figurest t + 1 t + 2Rj realized jt consume and dieEntrepreneurs jtborn with Kjt get f(φjt) andrealized RjtFjtg˜t realized, jt sell theclaim on the marketji determinesinvestmentPorject t startpaying dividendperiod t period t + 1Figure 2.22: Timing0011n∗n∗Realization of I with measure κwhen I ∈ [0, n∗]Realization of I with measure κwhen I /∈ [0, n∗]Figure 2.23: Uncertainty582.5. Tables and Figures0 1FjjF ∗jγFjt = D1−γ (j−γ)n∗Figure 2.24: Minimum Size RequirementTable 2.1: Regression of Volatility on firm-level controlsOnly IK Only R&D BothIK 0.04** 0.06** - - 0.023 0.023s.e. (0.01) (0.01) - - (0.013) (0.018)log( R&DSales ) - - 2.59** 2.87** 2.59** 2.88**s.e. - - (0.15) (0.15) (0.15) (0.16)log(emp) -2.45** - -2.21** - -2.18** -s.e. (0.05) - (0.05) - (0.045) -log(asset) - -1.99** - -1.69** - -1.66**s.e. - (0.04) - (0.04) - (0.04)R¯2 0.29 0.26 0.32 0.29 0.33 0.29N.obs 141,368 150,149 57,467 58,469 56,839 5779659Chapter 3Stock Market Reaction toMonetary Policy Shocks3.1 IntroductionHow the stock market reacts to monetary shocks is of interest for bothacademia and practical business. Recently, much effort has been devotedto understanding the asset market mechanism of propagating the monetarypolicy shock. The empirical evidence, however, is not as clear as people thinkit should be. One complication of the issue is that to answer the question,the researcher first needs a measure of the monetary “policy”, which canbe actions or official statements or even a speech given by Ben Bernanke.Second, one needs a measure of policy shock which does “shock” the market,instead of being anticipated. Hence not only a measure matters, but alsothe econometric methods used can change the results. The other problemis that monetary policy and the stock market may simultaneously react tosome other shocks in the economy, such as the news about inflation pressureor unemployment. This kind of endogenous problem may be not be solvedwhile using long-frequency data at yearly or quarterly levels.This paper investigates the issue by looking at the stock market reactionto monetary policy shocks using an event-study approach. I focus on themarket response only around the Federal Open Market Committee (FOMC)announcement dates and look at the stock market reaction to the FOMCmonetary policy statement in one-day to seven-day time window as well asusing monthly data. As a measure of monetary policy shocks, I use Romerand Romer (2004)’s measure, which combines FOMC statements and ac-tions, and the information available to FOMC around the meeting dates.Therefore it is a good measure of monetary policy shock, especially for ear-lier sample period in which financial futures data is not available. The stockmarket return is measured by CRSP value-weighted market return, which isstandard in the finance literature as market return. By using an event-studytype of analysis, I can solve the endogeneity problem: first, the new informa-603.1. Introductiontion which affects both monetary policy and the stock market return is verylittle in the narrow time-window. Second, since the FOMC meeting datesand most corporate financial reporting dates are exogenously determinedin advance, it is very unlikely that new important information happens tobe revealed at the meeting dates, at least not systematically, although wecannot rule out the possibility that some firms may select to announce newsaround the meeting dates. Monetary policy and stock markets react to thenews simultaneously, although this may happen in an inter-meeting federalfund rates change. Therefore, focusing only on the FOMC meetings allowsme to investigate the market reaction to monetary policy shocks free of theendogeneity problem.The results in Section 3 of the paper show that the 2-day response ofthe stock market to contractionary monetary policy shocks is strong andsignificant, while the reaction to the expected federal funds rate change issmall and not significant. The relationship between the stock market returnand monetary policy shocks can be illustrated as in Figure 3.1. It showsthat a scatter plot of 7-day stock market returns after the FOMC meeting,where the red dotted line is the fitted regression line. The results are broadlyconsistent with other studies that have examined the link between monetarypolicy and the stock market. Thorbecke (1997) is one of the first to look atthe link between monetary policy and the stock market and the result of hispaper is using a monetary policy shock measured from an identified VAR,which is conventional in the literature of monetary economics. Bernanke andKuttner (2005) is the closest to my research, but they use monetary policysurprises identified from the financial futures market, which is popular inrecent research on the link between financial markets and monetary policy.However, all previous works claim that there does not exist any signifi-cant asymmetry in the reactions. These authors address this issue by differ-entiating the shocks by positive shocks and negative shocks. In this paper, Idefine asymmetry as different reactions during recessions and booms. whencontrolling for recession, I find large and significant asymmetry between thereactions. Figure 3.2 illustrates the result by differentiating data between re-cession and expansion. It is clear that the slopes of the two regression linesare very different. This also explains why Bernanke and Kuttner (2005)find very large and symmetric shocks in the reactions. Their sample startsonly from 1989 yet the short sample contains two large recession periods:the early nineties recession and the 2001 recession after September 11th.Moreover, they not only use policy changes on FOMC meetings but alsointer-meeting changes, and there are usually more inter-meeting changesduring recessions than during booms. Therefore their sample has more ob-613.1. Introductionservations during recessions compared to this paper, hence larger estimatesand possibly finds no asymmetry.Given the large and asymmetric reactions of stock market reactions tomonetary policy shocks, the interpretation is twofold. From a macroeco-nomic perspective, asset prices or credit market frictions can be a importantpropagation mechanism of macroeconomic shocks; From an asset pricingperspective, the uncertainty in the monetary policy actions is a risk sourcewhich should be (or perhaps has been) priced in the market. This litera-ture of the former implication has been developed by Bernanke and Gertler(1989), Kiyotaki and Moore (1997), and Bernanke, Gertler and Gilchrist(1996). In these types of models, credit markets are characterized by asym-metric information and agency problems, therefore the Modigliani-Millertheorem no longer applies. Hence monetary policy is transmitted throughthe firm’s balance sheet: because the cost of borrowing (or external finance)is countercyclical and the borrowers “net worth”(defined as the borrower’sliquid assets plus collateral value of illiquid assets less outstanding obli-gations) is procyclical, this creates a “financial accelerator” of investment,spending and production. One of the implications of these models, at themicro level, is that the sensitivity of a firm’s value to monetary policy shocksdepends on the capital structure of the firm. Also, accessiblity to credit maycause a friction in the economy and propagates aggregate shocks. The risksource interpretation, is not stressed much in the literature except workssuch as Piazzesi (2005) and Gallmeyer, Hollifield and Zin (2005), but thesemodels all apply to the term structure of interest rates.I address these issues by testing the cross-sectional stock returns’ reac-tion to monetary shocks. I use portfolios formed on different size, book-to-market equity ratio and market leverage. Small size firms have beenfound to be more credit constrained than large firms(Gertler and Gilchrist(1994)); high book-to-market ration is also related to financial distress; andmarket leverage is a good measure of capital structure, allowing me to testthe implication of the financial accelerator in the asset market. On theother hand, since Fama and French(1992), others have tried to explain the“size effect”(small stocks earn higher returns than big stocks) and “valueeffect”(high book-to-market ratio stocks earn higher returns than low book-to-market ratio stocks). If the stock returns of firms with different sizes andbook-to-market ratios react differently to monetary policy shocks, monetarypolicy may itself be a risk source behind these risk “factors”.Section 4 shows that the result is encouraging. The evidence from themonthly data shows that stock return of small size firms react more to themonetary policy shock than large firms, that high book-to-market equity623.2. Measures of Monetary Policy Shocksratio firms react more to the shock than lower ones, and that high marketleverage ratio firms react more to the shock than lower ones. These are allevidence of credit market friction as a propagation mechanism of macroe-conomic shocks, and monetary policy is itself an important macroeconomicrisk source which should be taken into account in building asset pricingmodels. Monetary policy shocks as a macroeconomic risk, is priced in thestock market and it is at least partly accountable for the “size premium”and “value premium” in the finance literature.The rest of the paper is organized as follows: Section 2 reviews the mea-sures of monetary shocks used in the literature of investigating the financialmarket’s response to policy shocks and discusses their advantages and dis-advantages. Section 3 presents the data and the main results of the stockmarket’s reaction to monetary policy shocks. Section 4 shows responsesof portfolios formed on firm characteristics which may be related to creditconstraint or financial market frictions in the economy. Section 5 concludes.3.2 Measures of Monetary Policy ShocksThe accuracy of estimates of the effects of monetary policy depends cru-cially on the validity of the measure of monetary policy that is used. Theliterature of investigating the effects of monetary policy shocks on assetmarket prices often measure monetary policy shocks in three methods. Thefirst involves identifying monetary policy shocks from vector auto-regressionmodels (VAR) which is conventional and widely used in the macro literature.The second uses federal funds futures data to gauge market expectation ofmonetary policy actions. This method gains popularity in recent researcherswho look at the link between monetary policy and market response. Thethird one involves interpretation of FOMC’s announcement and narrativerecord, examples of this kind of measures are Boschen and Mills (1995),Lucca and Trebbi (2009), and also Romer and Romer (2004).3.2.1 Regression-based MeasuresThis approach involves first making enough identifying assumptions of theFed’s feedback rule.12 In so doing, one not only needs to make assump-tions about which economic variables the Fed considers when making policydecisions, but also assume the interaction mechanism between monetary12For a more complete review of this methodology, see Christiano, Eichenbaum andEvans (1999)633.2. Measures of Monetary Policy Shocksshocks and these macroeconomic variables. One common assumption is re-cursiveness assumption, which assumes that time t variables in the Fed’sinformation set do not respond to time t realizations of the monetary pol-icy shock. Then this and the assumption of linearity of the Fed’s feedbackrule can naturally justify obtaining the fitted residuals from a simple regres-sion of policy instruments on the variables in the Fed’s information set. Inpractice, this is often done in a identified structural VAR.Although the above approach is widely used in the literature, it obvi-ously has some flaws. One is the validity of the recursiveness assumption.Although some authors have models that are consistent with the assump-tion,13 other authors such as like Bernanke (1986), Sims (1986),Sims andZha (1995) have abandoned the assumption and use other identification as-sumptions. However, the main problem here is not only the recursivenessassumption, but also the likelihood of endogenous movements of the federalfunds rate, which has become the standard indicator of monetary actionsin studies of the effects of monetary policy, moves a great deal from day today for reasons unrelated to monetary policy. Also in non-federal funds ratetargeting eras, the rate is moving endogenously with other economic vari-ables. Ignoring this may lead to biased estimates. Last but not the least, theconventional measures almost surely contain anticipatory movements sincemovements in federal funds rate target series are often responses to infor-mation about future economic developments as the Fed devotes a lot of itsresources to forecasting. Therefore, using these measures sometimes leadsto puzzling results such as the “price puzzle”, which is the conclusion thata contractionary shock will cause price level to go up.On the subject of the stock market’s reaction, the VAR-based measuresare not widely used, except by Thorbecke (1997). Recent studies mostly usemarket-based measures derived from federal funds futures. The next sectionwill discuss this measure in detail.3.2.2 Market-based MeasuresThis measure does not rely on any specific assumption on the monetarypolicy rules. Instead, it relies on the expectation hypothesis of the termstructure in the financial market. Expectations of financial market agentsare not directly observable, but some financial products traded in the marketcan be natural market-based proxies for those expectations. An incompletelist of these are: Kuttner (2001) and Faust, Swanson and Wright (2004)13for example, Christiano, Eichenbaum and Evans (1997b) and Rotemberg and Wood-ford (1997)643.2. Measures of Monetary Policy Shocksuse the current month federal funds futures contract; Bomfim (2002) andPoole and Rasche (2000) use the month-ahead federal funds futures con-tract; Cochrane and Piazzessi(2002) use the one-month eurodollar depositrate; and Ellingsen and Soderstrom (1999) use the three-month Treasury bill.Gurkaynak, Sack and Swanson (2007) compare these market-based measuresand find that expectations derived from federal funds futures dominate othermarket-based measures of monetary policy expectations at horizons out sev-eral months. Based on their findings, it appears that the best measure ofshocks to the immediate policy setting would be based on federal funds fu-tures rates. Some of these applications to the stock market reactions areBernanke and Kuttner (2005), using federal funds futures and Rigobon andSack(2002), using three-month eurodollar futures rates.The advantage of expectations derived from Federal funds futures con-tracts is obvious. However, there are also some apparent flaws in this mea-sure: first, we are left with a very small sample since the Federal funds fu-tures start trading from 1989. If we want to look at policy actions associatedwith FOMC meetings, the sample size is even smaller. Second, monetarypolicy, contains not only “action” but also announcements or statements.This is especially true during or around FOMC meetings. The market seemsnot only react to the action, but also react to the statements of the Fed.Some researchers (e.g. Romer and Romer (2000)) also suggest that agents inthe stock market do so because they rationally deduce information about theeconomy(they assume the Fed has a better forecast of the economy),or theypredict the future policy path from the statements. Both of these will makethe market-based measure inaccurate, especially around FOMC meetings,when the policy “actions” and “words” come together. Another point isthat expectations measured from the future market rely on the expectationhypothesis, which is constantly rejected in the finance literature. Last butnot least, Federal funds future contracts are themselves risky assets tradedon the market and they have risk premium. If the risk premium is constantover time then this generally does not cause any severe problem. However, asPiazzesi and Swanson (2008) points out, since the risk premium is positive,time-varying, and varies systematically over the business cycle, it mattersin the future-based measure of monetary policy shocks. These futures ratestend to overpredict in recessions and underpredict in booms. The findingsalso suggest that monetary policy shocks may not be accurately measuredby the difference between the federal funds rate target and an ex ante mar-ket expectation based on federal funds futures. Another noteworthy findingpresented in Christiano, Eichenbaum and Evans (1999) is that there is noevidence to support the notion that inference is sensitive to incorporating653.2. Measures of Monetary Policy Shocksfederal funds market data into the analysis of the real effects of the monetaryshocks on macroeconomic variables.3.2.3 Narrative MeasuresWhen the Fed conducts its monetary policy with federal funds rate target-ing, it also announces a FOMC statement. This statement contains theFed’s understanding of the economic conditions, and their view of futureoutput, employment and inflation. Sometimes it also signals future actionsin the statement. Therefore, ignoring these statements does not seem tobe prudent, especially after policy makers have reached the consensus ofthe importance of ”expectation management”. However, to translate thesenarrative documents into policy shock measures is certainly not an easytask.Boschen and Mills (1995) is one of the earliest works addressing this is-sue. Based on their reading of the FOMC minutes, Boschen and Mills ratemonetary policy on a discrete scale {−2,−1, 0, 1, 2}, where -2 denotes verytight and +2 denotes very loose. This measure is very subjective and thevalidity is questionable. Recent studies such as Lucca and Trebbi (2009) uselinguistic score to measure the hawkishness and dovishness of the FOMCstatement. These are the pure narrative approaches of analyzing the mone-tary policy.On the other hand, Romer and Romer (2004)’s measure seems more con-ventional and give us an alternative way of investigating the issue. It com-bines the narrative statement of FOMC and the FOMC’s internal forecast togenerate a “new” measure of policy shock. Since it is derived from regressingthe intended funds rate change to the internal forecast, this shock series isallegedly free of endogenous and anticipatory movements. The sample goesfrom 1963 to 1996 and focuses only on the FOMC meetings. Therefore, thismeasure is a good compensation for the market-based measure. To someextent, Romer and Romer’s measure successfully takes into account the “fu-ture policy path”, which is suggested in the literature when it is related tomarket-based measures. Moreover, since it is derived from the Fed’s infor-mation set, it gives us a chance to check whether the policy surprises signalthe superior information that FOMC have.This paper is the first to use Romer and Romer (2004) “a new measure ofmonetary policy shocks” to investigate stock market reactions. Comparingthe results with other relevant studies may add to the knowledge of bothunderstanding the shock measures and the monetary shocks’ effects. Afterall, the literature has not reached a consensus as to which measure is better.663.3. Stock Market Reaction to Monetary Policy Shocks3.3 Stock Market Reaction to Monetary PolicyShocksIn this section, I will discuss the data and methodology I use and presentthe main results. The 2-day response of the stock market to contractionarymonetary policy shock is strong and significant, while the reaction to theexpected federal funds rate change is not significant. However, as we makethe window wider, both responses get larger and significant. The reasonthe market reacts slowly to “expected change” in the policy is not clear.One explanation is that since the expectation measure is fitted from theFed’s information set, the market may gradually react to the informationcontent in this variable. I then investigate this issue by considering thequestion ”Is there a statistical relationship between the market’s reactionand the Fed’s information?” and I do not find such a relationship. I thentest for asymmetries. Although the results for positive shocks and negativeshocks show no evidence for asymmetry, there exists a large and significantasymmetry in the reactions between recession and boom. Next, look at theresponses by different industries and different countries and find the resultis still significant.3.3.1 Data and MethodologyFor stock returns, I use the CRSP value-weighted market return which isa standard measure in the finance literature as a proxy for “the marketreturn”. In order to do the event-study type of regression, I use daily datawith a focus on data around the time of FOMC meetings. As a measureof monetary shocks, I use Romer and Romer (2004), the sample goes from1963 to 1996 and contains all FOMC meetings. To obtain this measure,they use two dataset which is unconventional in the literature: the first isthe intended fund rate change instead of the actual change and the other isthe Greenbook forecast, which is the internal forecast used around the theFOMC meeting and it is literally the information set of the FOMC meetingmembers.Table 3.2 shows the summary statistics of stock returns. For all timewindows and all stock returns, the stock returns are higher on average duringrecessions than during expansions. Also, during recession, the returns aremore volatile. This is consistent with the literature which says risk premiumsare high during recessions and low during expansions. Moreover, high book-to-market equity ratio firms earn a higher return than low book-to-marketfirms. Also, small stocks are more volatile than large stocks.673.3. Stock Market Reaction to Monetary Policy ShocksThen, the new measure of the the monetary policy shock is obtained byrunning the following regression:∆ffm = α+ βffbm +2∑i=−1γi∆˜ymi +2∑i=−1λi(∆˜ymi − ∆˜ym−1,i)+2∑i=−1ϕip˜imi +2∑i=−1θi(p˜imi − p˜im−1,i) + ρu˜m0 + mwhere ∆ffm is change in the intended funds rate around FOMC meetingsm; ffbm is the level of the intended funds rate before any changes associatedwith meetings m; p˜i, ∆˜y, u˜ are respectively the forecasts of inflation, realoutput growth, and the unemployment rate, and i subscripts is the horizonof the forecast. The “new shock” series is hence derived from taking theresiduals from the regression.Table 3.1 reports the mean and standard deviation of these policy vari-ables fitted using Romer and Romer (2004) data. Figure 3.3 is a plot offour time series at monthly frequency: the shock series of Romer and Romer(2004), the CRSP value-weighted market return, the average difference of re-turns between three highest book-to-market ratio portfolios and three lowestbook-to-market portfolios(High minus Low or HML), the average differencein returns between the three small firms portfolios and three large firms port-folios (Small minus Large or SML), with shaded area indicating the recessionperiods. It seems all series becomes more volatile during recessions.My first investigation is to regress the CRSP value-weighted market re-turn after the FOMC meeting on the Romer and Romer’s shock series.Before 1994, the FOMC’s decisions were conducted in open market oper-ations the day after the meeting, so market participants realize the policychange one day later. After 1994, the Fed explicitly announces the targetrate change in their FOMC statement, so the market can react to it pre-cisely on the day of the FOMC meeting. Thus, the two-day response periodsconsisted of the two days after the meeting before 1994, and consisted of theday of and the following day of the meeting after 1994.3.3.2 Baseline RegressionTable 3.3 presents the results from the basline regression equation:Rm(i) = α+ β1Expected change+ β2Surprise+ t683.3. Stock Market Reaction to Monetary Policy Shockswhere Rm(i) is the cumulated CRSP value-weighted return for i days,Expected change is expected policy change fitted from the Fed’s Green-book forecast, and Surprise is the policy shock obtained from regressing theintended funds rate change on the Greenbook forecast variables. Includingexpected policy changes can test the hypothesis that the financial marketparticipants have the same information set as the Fed.From the results we can see that the 2-day response of the stock marketto contractionary monetary policy shocks is strong and significant, whilethe reaction to the expected federal funds rate changes is not significant.However, as we make the window wider, both responses get larger and sig-nificant. If market is efficient, we should expect the stock market do notreact to expected changes no matter how wide the widow is. The reason themarket reacts slowly to “expected change” in the policy is not clear. Oneexplanation is that since the expectation measure is fitted from the Fed’s in-formation set, the market may gradually react to the information containedin this variable. I will address this issue later. An interesting result is themarket return reaction is much slower and smaller than that suggested byBK, who finds that 25 base point surprise can cause one-day market returnreaction of 1 percent.3.3.3 Industry and Country PortfoliosBesides equity indexes, we can also examine the response of disaggregatedindexes. I use Fama-French industry portfolios to do this check. Table 3.4shows the results. Using 2-day data, the most responsive industry is thetelecommunication industries, some industries are not very responsive andthe estimates are not significant at all. For 7-day data, however, the re-sponses are almost all large and significant except the manufacturing indus-try.The above results have shown that US stock returns respond to monetarypolicy shocks, both in the aggregate and from different industries. Anotherinteresting aspect would be looking at how foreign stock markets react to theUS shock. I use country portfolios from the Kenneth French Data Libraryto investigate this question. Table 3.5 reports the result. Generally, otherG7 countries also respond to US policy shocks, with UK, Germany, France,and Japan having significant estimates. It is also interesting to examinethat Canadian stock market’s response is relatively small and not signifi-cantly different from zero. This seemingly counter-intuitive effect is worthnoting, especially given the trade relationship and local integration of USand Canada.693.3. Stock Market Reaction to Monetary Policy Shocks3.3.4 Asymmetry: Controlling for RecessionMost researchers to date do not find asymmetry in stock markets’ reactionto monetary policy shocks.14 However, the asymmetry issue is often investi-gated by differentiating positive shocks and negative shocks. Inspired by thetime-varying risk premium literature, I reinvestigate the question by lookingat responses in recession periods and non-recession periods. In other words,I run the regression of the following type:Rm(i) = α+β1Expected change+β2Surprise+DReces×Surprise+DReces+twhere DReces is a dummy variable with value 1 if the period is in re-cession and 0 otherwise. I define recession as the period starting from theNBER definition of recession to the peak of unemployment rates. I also usethe standard NBER definition to check the robustness and find the result issimilar. Table 3.6 reports the result from this regression. Contrary to con-ventional wisdom, there exists very significant asymmetry in the responses.Not surprisingly, the dummy variable has negative estimates, which is con-sistent with our intuition that the stock return should be low in recession.One important fact is that after controlling for recession and the interac-tion of the reaction and recession, the estimates of the reaction to monetarysurprises are not significantly different from zero. However, the interactionterm has large and significant estimates. It seems the market reacts moreto monetary policy shocks in recession than in boom. Why is the stockmarket more sensitive to monetary policy shocks during recessions? Onereason may be during recessions, firms’ credit constraint is tighter. Mon-etary policy shocks are transmitted through the credit friction, thereforethe response is higher. Although further interpretation requires structuralmodels and sophisticated theory, one can conclude that the market reactsto monetary policy shocks differently in different economic conditions.Table 3.7 shows the stock market’s reactions to the shock in other coun-tries controlling for recession. Again, after controlling for recession, theabsolute value of the coefficients on surprise changes become smaller andinsignificant, whereas the estimates of the interaction term have large andsignificant values. For Hong Kong, Singapore and Canada the coefficient onsurprise even become positive. However, Canada has a large coefficient onthe interaction term, which suggests that Canadian stock markets react tothe shock dramatically during recession periods.14For example, Thorbecke (1997) and Bernanke and Kuttner (2005)).703.3. Stock Market Reaction to Monetary Policy Shocks3.3.5 Alternative Methods of Checking AsymmetryThe asymmetry that has been studied in the literature often addresses thefact that contractionary shocks and expansionary shocks have different ef-fets. Many empirical studies have found evidence of nonlinear and asymmet-ric effects of monetary policy shocks on inflation and output. 15 By usingthe traditional definition of asymmetry, I find that there exists asymmetricreactions for contractionary and expansionary shocks in the stock market insome industries, but the asymmetry is not significant. Moreover, as opposedto conventionally identified asymmetry in macroeconomic variables, expan-sionary shocks seem to have a larger impact on the stock market. FromTable 3.8, it is worth noting that expansionary shocks have much biggercoefficients than contractionary shocks, especially for a two-day window.Therefore, I use an interaction dummy variable to check whether this kindof asymmetry is significant in Table 3.9. However, the coefficients on theinteraction term are not significant for most portfolios.The above result is consistent with findings in Bernanke and Kuttner(2005). They also reject the hypothesis that the market reacts to mone-tary policy shocks asymmetrically, using the definition of asymmetry whichdifferentiates positive and negative shocks. However, the point estimatesalso display different magnitudes, though they are not statistically differentfrom each other. The reason using both their future-based measure and themeasure used here results in no statistically significant asymmetry but stilldifferent point estimates is the definition of asymmetry. As discussed in theprevious subsection, one can find large asymmetric responses by differenti-ating between recessions and booms. Bernanke and Kuttner (2005) only usedata from 1989 to 2003, which contains two large recession periods, resultingin dramatically large responses to shocks. Moreover, they use not only theFOMC meeting observations, but also inter-meeting changes, and there areusually more inter-meeting changes during a recession than during a boombecause of the Fed’s policy response to the economy. Therefore, their datahave a lot of observations in recession, which gives relatively large estimatesof the reaction.3.3.6 Does the Fed Signal Its (Superior) Information?The reaction to monetary policy shocks can be interpreted in different ways.One is that monetary policy affects the risk free rate and the risk premium.15see Cover(1992); Macklem et al.(1996); Ravn and Sola(2004) and the referencestherein.713.4. Monetary Policy Shocks and Cross-sectional Stock ReturnsBernanke and Kuttner (2005) use a predicting VAR to test this hypothesisand find that the reaction to tight money is not due to the effect on the realinterest rate, but through the expected future excess returns. However, evenif this is true, the interpretation of the response is still ambiguous. Tightmoney could affect the riskiness of the stocks, or it can affect the willingnessof investors to bear risk.Another popular argument regarding the market’s reaction to policy isthat the Fed has superior information about the economy and agents inthe financial market often deduce this kind of information content from thepolicy actions. Romer and Romer (2000) find that the Fed’s forecasts ofinflation and output dominate those of the private sector. I address thisissue by examining whether there is a statistical relationship between thereaction and the information in the Greenbook forecast. From Table 3.10,the conclusion is that there is no such relation in a narrow window. Thisresult is consistent with Faust et al.(2004), which investigate whether privateforecast producers change their forecast after the policy shock. However,as we allow the market respond to the shock and extract information for alonger time, I find the market seems to react to the Fed’s information such asthe unemployment rate forecast and the change in long-term output forecast.However, this is equivalent to controlling for recessions since unemploymentrate is a important indicator of recession.3.4 Monetary Policy Shocks and Cross-sectionalStock ReturnsAlthough it is very clear the stock market reacts to monetary policy shocks,the reason or mechanism of the reaction is not clear. Recent literature aboutthe credit frictions have addressed the lending and borrowing role of mon-etary policy transmission. For example, Gertler and Gilchrist (1993) findthat there is a large difference in response of credit flows to small versus largeborrowers after a tightening monetary policy action. Gertler and Gilchrist(1994) also find that small firms account for disproportionate share of themanufacturing decline after a tightening monetary policy action. These find-ings can explain the “size premium” in the empirical asset pricing literature,which suggests small firms constantly earn higher return than large firms.This section explores the link between monetary policy shock and thecross-sectional stock returns by looking at the reactions of stocks with dif-fering characteristics. I consider three credit-related firm characteristics:size (measured by market equity), book-to-market equity ratio, and market723.4. Monetary Policy Shocks and Cross-sectional Stock Returnsleverage (measured by the ratio of total market value of debt over the to-tal market value of assets). Small firms are more credit constrained thanlarge firms, as the above works and the literature of credit rationing havesuggested. Book-to-market ratio is also related to financial distress andChoi(2008) finds the book-to-market ratio is highly positively correlatedwith a firm’s leverage ratio. Also, high leverage firms are faced with highinterest payments as the cost of capital and therefore they could be sensitiveto monetary policy shocks.Since the seminal work of Fama and French (1992), the “size effect” and“book-to-market effect” have been studied a lot in the finance literature.Small firms constantly earn higher average returns than large firms, highbook-to-market (value) firms earn higher average returns than low book-to-market (growth) firms. Although these two are treated as “risk factors”,the risk sources behind these factors are not clear. In fact, the literaturehas not converged to a convincing theoretical explanation of these effects.Some authors use behavioral or psychological explanations16, while othersuse rational explanations which view small stocks and value stocks as funda-mentally riskier17. Many researchers who support the rational explanationshave realized that these risk factors are related to macroeconomic risks.Therefore, one goal of this section is to identify one possible risk source: theprocess of monetary policy decisions. The hypothesis is: monetary shocksare transmitted through the credit market and different firms with differentexposure to this risk would react to the shock differently.Using monthly data, I find small stocks react more to the shocks thanlarger firms, value stocks react more to the shocks than growth stocks, andfirms with high market leverage ratio react more than lower leveraged firms.All these facts are evidence that monetary policy shocks are transmittedthrough the credit channel and monetary policy as a macroeconomic risk ispriced in the stock market.3.4.1 Size, Book-to-market, and Market LeverageIn order to test the reaction of the cross-sectional stock returns to the mon-etary policy shocks, I use the portfolios formed on size and book-to-marketfrom the Ken French data library. I also use portfolios formed on marketleverage from the data library for Chen and Zhang(2009). Table 3.11 reportsthe result of regressing the portfolio returns on the expected and surprisemonetary policy changes. The first column shows results from 10 portfolios16For example, Lakonishok et al.(1994) and DeBondt and Thaler (1987)17For example, Fama and French(1993), Petkova and Zhang(2005)733.4. Monetary Policy Shocks and Cross-sectional Stock Returnsformed on size. The smallest size firms have twice as large a reaction as thelargest firms. Value firms have nearly three times larger reactions to theshock than growth firms. Similarly, firms with a high leverage ratio reactmore. The most leveraged firms have nine times as large a coefficient as theleast leveraged firms, and still twice as large as the second lowest leverageportfolio. Another interesting aspect is that all these portfolios’ reactionsto the expected part of the monetary policy change are similar, which sug-gest that the policy surprise alone is related to the difference between thereturns. This is strong evidence the size effect and value effect are relatedto monetary policy risk.Figure 3.4 illustrates the result by plotting the responses of 10 portfoliosformed on size. The blue bar is the 95% confidence interval. There is a largedifference in the reactions betwen the small firms and large firms. Similarly,Figure 3.5 and Figure 3.6 plot the reactions of the book-to-market portfoliosand market leverage portfolios.The finding about the 10 size portfolios here is consistent with Guo(2004) who uses Cook and Hahn (1989a) and Poole and Rasche (2000) datato investigate the size and book-to-market portfolios. However, that paperfailed to find the different responses in the book-to-market portfolios, whichI emphasize as being important. The discrepancy may be caused by thedifferent measures used. Cook and Hahn (1989a) use the change in thetarget funds rate, which contains an obvious anticipated term which canbias the estimates downward. Poole and Rasche (2000) use federal fundsfutures data, which is subject to the time varying risk premium critique.Moreover, Guo (2004) only compares the data from 1974-1979 and 1988-2000 and I am using a larger sample. I also use monthly data to allowthe market to fully incorporate all information from the rate changes. Guo(2004) also claims that small stocks do not react more to the policy in the1990s since business condition is generally better than in the 1970s. I concurwith the result. By using only the data from 1984-1996, the “size effect”in the reaction almost disappear. However, the book-to-market effect andmarket leverage effect still remains in the sample after 1984.Table 3.12 shows the result of controlling for recessions. My resultsshow a difference, but there is no difference in the coefficients of the inter-action term. In other words, the difference is caused by different reactioninstead of the asymmetry. Even in good times, small firms may react morestrongly to the shock than large firms. However, in book-to-market andmarket leverage portfolios, the effects come from both the coefficient of thesurprise and the interaction term of surprise and recession. Perez-Quirosand Timmermann (2000) also find that expected returns on small stocks743.4. Monetary Policy Shocks and Cross-sectional Stock Returnsare more strongly affected by credit market conditions than returns on bigstocks during economic recessions, but not during economic expansions. Myresults are consistent with their finding. As I showed in Section 2, the stockmarket generally reacts only mildly to the monetary policy shock duringeconomic expansions. Also, the similar pattern of the reactions of portfoliosformed on book-to-market and market leverage confirm Choi (2008)’s resultthat book-to-market is generally a noisy measure of the leverage ratio. Thisresult is generally consistent with the theory that imperfect credit marketsare an important mechanism for the propagation of economic fluctuations,especially monetary policy shocks.3.4.2 Evidence from 25 Portfolios formed on DifferentCharacteristicsThe results in the previous section have shown firms with different sizes,book-to-market ratios and market leverage react considerably differently tomonetary policy shocks. It is natural to check whether this relation holdswithin each portfolio. To address this issue, I use 25 portfolios formed on sizeand book-to-market ratios from Ken French’s data library and 25 portfoliosformed on size and leverage from Lu Zhang’s data library.Table 3.13 reports the responses of 25 portfolios formed on size andbook-to-market ratios. To conserve space, I only report the whole sampleregression and the coefficients on surprise monetary policy change. Withineach size portfolio, the reaction is larger for high book-to-market stocks.Similarly, with each book-to-market quintile, the reaction is larger for smallfirms. The large value firms have coefficients which are three times as largeas those for the small growth firms. The result is illustrated in Figure 3.7.Table 3.14 reports the result from the 25 portfolios formed on size andmarket leverage and the corresponding result is illustrated in Figure 3.8.Again, within each size, reactions are larger for high leveraged firms. Withineach market leverage quintile, smaller firms react more strongly to largerfirms. Small, high leveraged portfolios have five times larger coefficients thanthe large, low leveraged firms. Therefore, the results are again consistentwith the theory. Figure 3.9 is a 3-dimensional version of Figure 3.8 to showthat how different size and leverage portfolios react to the monetary policyshocks.753.5. Conclusion3.4.3 Statistical Test of SignificanceFor 10 portfolios formed on each firm characteristics, it is possible to checkwhether the coefficients are significantly different from each other using aWald test. For the ith and jth portfolio, one can run a seemingly uncor-related regression by treating both equations as a system regression. Thena Wald test can be used to test whether the coefficients on surprises aresignificantly different.Table 3.15 reports the test statistics for the market leverage portfolios.We can see the lowest leverage portfolio’s reaction is significantly differentfrom other portfolios. Moreover, as the leverage difference becomes larger,the p-value becomes smaller, therefore we have more confidence to rejectthe null hypothesis that they are the same. The results for size and book-to-market portfolios are similar but the p-values are very high. However, asthe difference grows, the p-values also becomes lower, hence more confidenceto reject.3.5 ConclusionIn this paper I documented a state-dependent reaction of the stock mar-ket to the monetary policy shocks. I use the Romer and Romer (2004)“new measure” of policy shocks, which is relatively free of anticipatory andendogenous problems. There exists significantly large asymmetry betweenthe reactions of stock returns to monetary policy shocks during 1969-1996period. The stock market reacts more strongly to monetary policy duringrecession than during economic expansion. These results are robust in differ-ent industries, different firm characteristics and even from all G7 countries’stock market.Moreover, using cross-sectional returns, I find the reaction pattern is con-sistent with the theory that imperfect credit markets are important in prop-agating macroeconomic shocks. Small firms react more strongly to shocksthan large firms, value firms’ reactions are larger than those of growth firms,and high leveraged firms are more susceptible to monetary policy shocks thanlow leveraged firms. Since small firms are more credit-constrained than largefirms, and book-to-market equity ratio and market leverage are indicatorsof the firms capital structure which can be affected by credit market fluctu-ations, all the findings are supportive of the theory that external finance isan important channel in the transmission of monetary policy shocks.A future research topic is to explain the very different reactions of thestock market during economic recessions and expansions. Especially, it is763.5. Conclusioninteresting to note during a US recession, major industrial stock markets allaround the world react to the Fed’s monetary policy shocks more stronglythan during economic expansion. One possible explanation is since risk pre-mium is time-varying and monetary policy shock is a macroeconomic riskthat is priced in the market, the stock price can react to the shock differ-ently during good times and bad times. Another explanation is that duringrecession, the investors become more risk averse due to the consumption andwealth uncertainty, and therefore the reaction of stock price is larger. How-ever, without an asset pricing model which incorporates monetary policy,the direct answer to this question is not clear. Some works incorporate themonetary policy rule in bond pricing18, however, to my knowledge, thereis no model which explicitly includes monetary policy risk into the equitymarket.From the results of cross-sectional returns, one can conclude that the sizeeffect and value effect are all related to monetary policy. The mechanismbehind this is that partly because of the availability of less public informationabout small firms, these firms typically have to rely more on information-intensive sources of credit such as bank loans. If worsening credit marketconditions work through firms’ balance sheet, then their effect is likely tobe strongest for small firms in a recession. Since book-to-market ratio andleverage ratio often go hand-in-hand, a contractionary shock will increasethe firm’s cost of capital especially the interest rate payment from the debt,hence will further reduce profits for more leveraged firms. Monetary policy,with its goal of stabilizing output and inflation, should take into accountthese credit market frictions as the transmission mechanism of the policychange.18see Piazzesi (2005), Gallmeyer, Hollifield and Zin (2005)773.6. Tables and Figures3.6 Tables and FiguresTable 3.1: Summary Statistics: Monetary Policy VariablesThis table reports the summary statistics of monetary policy changes. ∆ff is theintended funds rate change in Romer and Romer(2005). The expected policy actions arethe fitted value of regressing intended funds rate change on the Greenbook forecastvariables. Surprise changes are the residuals from the regression.Data by Meeting ∆ff Expected Surprise Sample Sizemean -0.025 -0.025 0 263s.d. 0.45 0.24 0.38Data by Month ∆ff Expected Surprise Sample Sizemean -0.0022 -0.0022 0 336s.d. 0.7222 0.6531 0.3357783.6.TablesandFiguresTable 3.2: Summary Statistics: Stock ReturnsThis table reports the summary statistics of cummulated stock returns of different days after the FOMC meeting date, as well as themonthly return. Market return is CRSP value-weighted stock return. Returns of small size and large size firms are the first and lastdecile of Fama-French 10 portforlios formed on size. Returns of low book-to-market equity and high book-to-market equity firms areare the first and last decile of Fama-French 10 portforlios formed on book-to-market equity. The data is taken from Kenneth Frenchdata library. “W” means whole sample, “R” stands for only during recession, and “N” stands for non-recession period. There are 63recession observations in daily data and 80 in monthly dataWindow 1-day 2-day 3-day 7-day MonthlySample W R N W R N W R N W R N W R NMarket 0.12 0.20 0.10 0.26 0.34 0.23 0.34 0.45 0.30 0.49 0.58 0.46 0.49 -0.17 0.69s.d. 0.80 1.14 0.67 1.25 1.82 1.01 1.62 2.36 1.30 2.54 3.68 2.06 6.09 8.08 5.32Small 0.08 0.17 0.06 0.22 0.32 0.19 0.35 0.42 0.33 0.41 0.23 0.47 0.93 0.32 1.12s.d. 0.70 1.03 0.56 1.20 1.85 0.92 1.61 2.40 1.27 3.01 4.38 2.44 6.16 8.12 5.41Large 0.12 0.18 0.11 0.23 0.32 0.20 0.29 0.43 0.25 0.48 0.64 0.42 0.99 0.65 1.10s.d. 0.89 1.23 0.76 1.32 1.86 1.10 1.69 2.44 1.38 2.46 3.49 2.04 4.44 5.80 3.93Low B/M 0.15 0.20 0.13 0.28 0.30 0.28 0.35 0.40 0.33 0.54 0.68 0.49 0.93 0.47 1.07s.d. 1.03 1.50 0.84 1.60 2.38 1.27 2.04 3.01 1.63 2.99 4.37 2.40 5.05 6.62 4.45High B/M 0.16 0.27 0.13 0.27 0.39 0.24 0.41 0.61 0.35 0.54 0.70 0.49 1.36 1.12 1.43s.d. 0.91 1.12 0.83 1.33 1.72 1.18 1.73 2.20 1.55 2.87 3.97 2.44 5.05 6.43 4.54793.6. Tables and FiguresTable 3.3: Response of CRSP Value-weighted Stock Return to theMonetary Policy ShocksThis table reports the result from regressionRm(i) = α+ β1Expected change+ β2Surprise+ t, where Rm(i) is the cumulatedCRSP value-weighted return for i days, Expectedchange is expected policy change fittedfrom the Fed’s Greenbook forecast, and Surprise is the policy shock obtained fromregressing intended funds rate change on the Greenbook forecast variables. The wholesample contains 263 FOMC meetings during 1963-1996 and 336 months in monthly data.*means significant at 1% level. All standard errors heteroskedasticity-robust.Whole Sample Without OutliersWindow Constant Expected Surprise Constant Expected Surprise1-day 0.125* 0.002 -0.116 0.105* -0.008 -0.126s.e. (0.049) (0.221) (0.093) (0.045) (0.232) (0.131)2-day 0.252* -0.129 -0.332* 0.224* -0.244 -0.364s.e. (0.076) (0.332) (0.151) (0.070) (0.347) (0.212)3-day 0.327* -0.366 -0.473* 0.288* -0.399 -0.516*s.e. (0.098) (0.409) (0.186) (0.091) (0.432) (0.264)4-day 0.241* -0.510 -0.854* 0.198* -0.595 -0.996*s.e. (0.121) (0.536) (0.294) (0.111) (0.563) (0.393)5-day 0.329* -0.929 -0.961* 0.287* -1.035 -1.145*s.e. (0.129) (0.583) (0.331) (0.121) (0.606) (0.434)6-day 0.398* -1.185 -0.925* 0.351* -1.278 -1.082*s.e. (0.143) (0.644) (0.354) (0.135) (0.676) (0.480)7-day 0.451* -1.435* -1.100* 0.416* -1.611* -1.426*s.e. (0.152) (0.681) (0.451) (0.145) (0.706) (0.552)Monthly 0.4839* -1.3798* -2.4127* 0.4439* -1.5053* -3.0656*s.e. (0.3269) (0.5341) (1.1268) (0.3267) (0.5730) (1.4345)803.6. Tables and FiguresTable 3.4: Response of F-F 10-industry portfolios to the MonetaryPolicy ShocksThis table reports the result from regression like the one in Table 2, but using 10Fama-French industry portfolios.The whole sample contains 263 FOMC meetings during1963-1996 and 336 months in monthly data. *means significant at 1% level. All standarderrors are heteroskedasticity-robust.2-day 7-dayIndustry Constant Expected Surprise Constant Expected SurpriseNoDur 0.184* 0.146 -0.375* 0.468* -1.394* -1.335*s.e. (0.072) (0.313) (0.185) (0.146) (0.676) (0.465)Durbl 0.217* -0.412 -0.288 0.445* -1.697* -1.639*s.e. (0.091) (0.467) (0.262) (0.168) (0.826) (0.575)Manuf 0.254* -0.203 -0.335 0.288* -0.399 -0.516s.e. (0.078) (0.376) (0.231) (0.149) (0.722) (0.500)Enrgy 0.282* -0.252 -0.625* 0.580* -1.441 -1.686*s.e. (0.094) (0.477) (0.311) (0.161) (0.843) (0.701)HiTec 0.270* -0.016 -0.430 0.577* -1.502 -1.679*s.e. (0.109) (0.498) (0.333) (0.190) (0.875) (0.662)Telcm 0.098 -0.039 -0.925* 0.351* -1.278* -1.082*s.e. (0.086) (0.416) (0.237) (0.135) (0.621) (0.486)Shops 0.237* -0.470 -0.359 0.446* -1.669* -1.333*s.e. (0.088) (0.449) (0.252) (0.167) (0.796) (0.541)Hlth 0.202* -0.491 -0.269 0.496* -1.683* -1.251*s.e. (0.088) (0.420) (0.250) (0.159) (0.740) (0.532)Utils 0.116* -0.360 -0.245 0.328* -1.681* -1.122*s.e. (0.052) (0.220) (0.142) (0.125) (0.634) (0.431)Other 0.246* -0.313 -0.250 0.422* -1.640* -1.316*s.e. (0.075) (0.363) (0.211) (0.153) (0.729) (0.501)813.6. Tables and FiguresTable 3.5: Monthly Stock Return Reaction to US Shocks by Coun-tryThis table reports the result from regression like the one in Table 2, but using countryportfolios from UK, Italy, German, France, Japan, Hong Kong, Singapore and Canada.The whole sample contains 240 months during 1975-1996. *means significant at 1%level. All standard errors are heteroskedasticity-robust .UK ITA GER FRA JAP HK SGP CANCons 1.67* 1.31* 1.29* 1.51* 1.29* 1.99* 1.45* 1.01*s.e. (0.36) (0.49) (0.35) (0.42) (0.43) (0.58) (0.45) (0.32)Expe -0.48 -0.74 -0.80* -1.30* -0.18 0.54 -1.05 -1.6894*s.e. (0.65) (0.54) (0.43) (0.56) (0.44) (0.88) (0.57) (0.63)Surp -2.78* -1.59 -2.82* -3.76* -1.96* -0.64 -1.51 -1.6425s.e. (0.95) (1.54) (0.96) (1.31) (0.64) (2.15) (1.22) (1.30)R2 0.03 0.01 0.04 0.06 0.01 0.002 0.02 0.06R¯2 0.02 -0.003 0.03 0.04 -0.002 -0.01 0.004 0.05823.6. Tables and FiguresTable 3.6: Response of CRSP Value-weighted Stock Return to theMonetary Policy Shocks: Controlling for RecessionThis table reports the result from regressionRm(i) = α+ β1Expected change+ β2Surprise+DReces × Surprise+DReces + t,where Rm(i) is the cumulated CRSP value-weighted return for i days, Expectedchangeis expected policy change fitted from the Fed’s Greenbook forecast, and Surprise is thepolicy shock obtained from regressing intended funds rate change on the Greenbookforecast variables. The whole sample contains 263 FOMC meetings during 1963-1996and 336 months in monthly data. *means significant at 1% level. All standard errors areheteroskedasticity-robust .Window Constant Expected Surprise DReces × Surprise DReces2-day 0.2476* -0.2864 -0.2580 -0.4177 -0.13324s.e. (0.0730) (0.3482) (0.2412) (0.4552) (0.2104)3-day 0.3258* -0.4807 -0.3985 -0.4952 -0.1998s.e. (0.0937) (0.4313) (0.2772) (0.5724) (0.2672)7-day 0.5215* -1.6689* -0.5119 -3.3006* -0.6741s.e. (0.1466) (0.7055) (0.5191) (0.8584) (0.4117)Monthly 0.8229* -1.6286* -1.5066 -4.8664* -1.8020*s.e. (0.3351) (0.6041) (1.4351) (2.7823) (0.9125)833.6. Tables and FiguresTable 3.7: Monthly Stock Return Reaction to US Monetary Shocksby CountryThis table reports the result from regression like the one in Table 5, but using countryportfolios from UK, Italy, German, France, Japan, Hong Kong, Singapore and Canada.The whole sample contains 240 months during 1975-1996. *means significant at 1%level. All standard errors are heteroskedasticity-robust.UK ITA GER FRA JAP HK SGP CANCons 1.59* 1.51* 1.54* 1.63* 1.80* 2.28* 1.65* 1.27*s.e. (0.42) (0.5) (0.38) (0.45) (0.44) (0.59) (0.48) 0.34Expe -1.24 -1.39* -1.34* -1.74* -0.79 -0.25 -2.04* -2.05*s.e. (0.91) (0.57) (0.51) (0.63) (0.47) (0.80) (0.63) (0.64)Surp -0.98 -0.75 -1.20 -3.38 -1.58 3.68 1.09 1.00s.e. (1.39) (2.56) (1.26) (2.26) (1.14) (2.79) (1.49) (1.21)D×Surp -4.04 -2.37 -3.59* -1.99 -1.43 -9.94* -5.90* -5.58*s.e. (2.34) (2.95) (1.59) (2.55) (1.39) (3.24) (2.00) (1.91)DRces 0.59 -2.56* -1.76 -1.33 -2.52* -1.27 -0.99 -1.96*s.e. (1.32) (1.10) (0.94) (1.08) (1.19) (1.48) (1.32) (1.00)R2 0.05 0.03 0.07 0.07 0.03 0.04 0.05 0.11R¯2 0.02 0.02 0.05 0.05 0.02 0.02 0.02 0.09843.6. Tables and FiguresTable 3.8: Response of Stock Returns to Positive and NegativeShocksThis table reports the result from regression by dividing the sample according to thesign of the shocks. The sample contains 261 FOMC meetings during 1963-1996. Allstandard errors are heteroskedasticity-robust.2-day 7-dayPortfolio Positive Negative Positive NegativeMarket -0.3308 -1.3810 -2.0514 -2.9628s.e. (0.2745) (0.5114) (1.0325) (1.0039)NoDur -0.3795 -1.2615 -2.2208 -2.6609s.e. (0.1886) (0.5145) (0.8001) (0.9995)Durbl -0.3262 -1.6236 -2.4665 -3.5679s.e. (0.3223) (0.7112) (1.0319) (1.0956)Manuf -0.2720 -1.4006 -2.2092 -2.8593s.e. (0.2919) (0.6010) (0.8831) (1.0220)Enrgy -0.7671 -1.5161 -2.5592 -2.2986s.e. (0.5249) (0.6184) (1.3508) (1.1882)HiTec -0.3317 -1.9427 -2.3109 -3.4682s.e. (0.3648) (0.8707) (1.1144) (1.3986)Telcm -0.5422 -0.5353 -1.7605 -2.0258s.e. (0.2440) (0.7103) (0.8922) (1.0594)Shops -0.0842 -2.0351 -2.0675 -3.6228s.e. (0.2752) (0.7179) (0.8742) (1.2555)Hlth -0.3105 -1.3334 -1.9715 -2.3168s.e. (0.2793) (0.6472) (0.9287) (1.1117)Utils -0.3362 -0.6876 -2.0088 -1.4247s.e. (0.1875) (0.3408) (0.8139) (0.7864)Other -0.0263 -1.4293 -2.0796 -2.7818s.e. (0.2326) (0.5226) (0.9067) ( 1.0139)853.6. Tables and FiguresTable 3.9: Checking Sign Asymmetry using Interacting DummiesThe Coefficients reported in this table are obtained from regressing the 2-day and 7-dayreturns of each portfolio on the expected policy change, policy shock and an interactionterm of policy shock with positive shock dummy. Policy shocks are measured usingRomer and Romer(2005) method. The sample contains 261 FOMC meetings during1963-1996. All standard errors are heteroskedasticity-robust.2-day 7-dayPortfolio Shock Shock × positive Shock Shock × positiveMarket -0.9115 0.8612 -1.6882 0.4125s.e. (0.4794) (0.5806) (0.8797) (1.4117)NoDur -0.8214 0.7032 -1.2701 -0.1008s.e. (0.4817) (0.5474) (0.8921) (1.2590)Durbl -0.9375 1.0216 -1.9265 0.4529s.e. (0.6279) (0.7784) (0.9722) (1.4766)Manuf -0.9245 0.9271 -1.4702 0.1033s.e. (0.5523) (0.6679) (0.9081) (1.3377)Enrgy -0.9613 0.5296 -1.2792 -0.6403s.e. (0.5957) (0.8127) (1.0826) ( 1.7971)HiTec -1.2805 1.3390 -2.0682 0.6118s.e. (0.8022) (0.9601) (1.2522) ( 1.7825)Telcm -0.3766 -0.0462 -1.1700 -0.0633s.e. (0.6178) (0.7131) (0.9290) ( 1.3421)Shops -1.4479 1.7142 -1.8921 0.8795s.e. (0.6263) (0.7409) (1.0739) (1.4682)Hlth -0.7804 0.8055 -1.2047 -0.0728s.e. (0.6186) (0.7073) (0.9865) (1.4110)Utils -0.4029 0.2485 -0.5231 -0.9419s.e. (0.3167) (0.3826) ( 0.6979) (1.1007)Other -1.0404 1.2433 -1.4232 0.1693s.e. (0.4760) (0.5574) (0.8615) (1.3065)863.6. Tables and FiguresTable 3.10: Market Reaction and the Greenbook ForecastThis table reports the result from regression of CRSP value-weighted stock return on themonetary policy shock as measured using Romer and Romer(2005) method, controllingfor the Greenbook forecast during the FOMC meetings. The sample contains 261 FOMCmeetings during 1963-1996. All standard errors are heteroskedasticity-robust.2-day 7-daylevel change level changeConstant -0.058 -0.091 0.245 -0.356s.e. (0.392) (0.409) (0.799) (0.791)Inflation forecast0-quarter -0.035 -0.130s.e. (0.068) (0.141)1-quarter 0.104 0.023s.e. (0.126) (0.300)2-quarter -0.111 -0.021s.e. (0.137) (0.312)Output forecast0-quarter -0.010 -0.059s.e. (0.028) (0.063)1-quarter -0.032 -0.075s.e. (0.065) (0.152)2-quarter -0.046 -0.109s.e. (0.070) (0.158)Change in Inflation forecast0-quarter -0.106 -0.138s.e. (0.124) (0.285)1-quarter -0.067 0.160s.e. (0.185) (0.427)2-quarter -0.201 0.468s.e. (0.257) (0.454)Change in Output forecast0-quarter 0.042 -0.140s.e. (0.057) (0.121)1-quarter -0.051 -0.260s.e. (0.112) (0.222)2-quarter 0.177 0.557s.e. (0.111) (0.267)Unemployment forecast 0.111 0.053 0.232 0.122s.e. (0.066) (0.063) (0.124) (0.119)Policy Surprise -0.387 -0.319 -1.409 -1.311s.e. (0.220) (0.211) (0.564) (0.539)873.6. Tables and FiguresTable 3.11: Responses of Size, Book-to-market and Market Lever-age PortfoliosThis table reports the result from regression of 10 portfolios formed on size, 10 portfoliosformed on book-to-market equity ratio and 10 portfolios formed on market leverage onthe expected and surprise monetary policy changes. The number is increasing in size,book to market and leverage. The data are of monthly frequency. The whole samplecontains 336 months from 1969-1996. All standard errors are heteroskedasticity-robust.Size(Small-Large) B/M(Growth-Value) Leverage(Low-High)Portfolio Cons Expe Surp Cons Expe Surp Cons Expe Surp1 0.90 -1.10 -2.33 0.83 -1.03 -0.73 0.84 -0.92 -0.31s.e. 0.34 0.53 1.18 0.29 0.45 1.06 0.30 0.49 1.092 0.96 -1.26 -2.55 1.03 -1.07 -1.43 0.97 -1.19 -1.42s.e. 0.33 0.54 1.16 0.27 0.42 0.92 0.27 0.41 0.843 1.09 -1.31 -2.44 1.06 -1.22 -1.79 1.00 -1.20 -1.62s.e. 0.32 0.51 1.15 0.27 0.41 1.05 0.26 0.40 0.964 1.10 -1.42 -2.49 1.05 -1.17 -1.28 1.10 -1.20 -1.59s.e. 0.31 0.48 1.10 0.26 0.47 1.16 0.26 0.44 1.135 1.14 -1.36 -2.56 1.00 -1.12 -1.26 1.06 -0.97 -1.56s.e. 0.30 0.47 1.03 0.24 0.41 1.23 0.24 0.41 1.026 1.08 -1.33 -2.51 1.12 -0.94 -1.85 1.11 -1.02 -1.77s.e. 0.29 0.46 1.06 0.24 0.40 0.97 0.25 0.43 0.977 1.10 -1.47 -2.35 1.14 -1.30 -2.38 1.19 -1.02 -2.50s.e. 0.28 0.46 1.09 0.23 0.44 0.78 0.23 0.38 0.758 1.05 -1.39 -2.41 1.16 -0.97 -2.57 1.12 -1.29 -2.78s.e. 0.27 0.45 0.95 0.23 0.39 0.86 0.24 0.43 0.759 1.01 -1.24 -1.68 1.31 -1.19 -2.81 1.22 -1.51 -2.96s.e. 0.25 0.43 0.92 0.25 0.38 0.80 0.28 0.49 0.8210 0.97 -1.01 -1.10 1.41 -1.52 -2.68 1.18 -1.17 -2.80s.e. 0.23 0.39 0.97 0.30 0.51 0.94 0.30 0.45 0.79883.6. Tables and FiguresTable 3.12: Responses of Size, Book-to-market and Market Lever-age Portfolios Controlling for RecessionThis table reports the result from regression of 10 portfolios formed on size, 10 portfoliosformed on book-to-market equity ratio and 10 portfolios formed on market leverage onthe expected and surprise monetary policy changes, and controlling for recession. Thenumber is increasing in size, book to market and leverage. The data are of monthlyfrequency. The whole sample contains 336 months from 1969-1996. All standard errorsare heteroskedasticity-robust.Size(Small-Large) B/M(Growth-Value) Leverage(Low-High)Portfolio Surp D×Surp DReces Surp D×Surp DReces Surp D×Surp DReces1 -0.57 -4.09 -1.73 1.36 -4.62 -1.31 1.75 -4.52 -1.16s.e. 1.58 2.07 0.95 1.29 1.59 0.79 1.30 1.60 0.832 -0.35 -4.97 -1.73 0.40 -4.09 -1.31 0.46 -4.12 -1.10s.e. 1.45 1.88 0.90 1.25 1.50 0.71 1.02 1.32 0.713 -0.36 -4.77 -1.83 0.41 -4.85 -1.35 0.54 -4.73 -1.23s.e. 1.50 1.91 0.87 1.42 1.74 0.69 1.25 1.63 0.684 -0.58 -4.39 -1.68 1.18 -5.46 -1.63 0.47 -4.59 -1.42s.e. 1.46 1.84 0.84 1.61 1.86 0.65 1.62 1.87 0.645 -0.44 -4.76 -1.60 0.93 -4.79 -1.27 0.32 -4.25 -1.48s.e. 1.22 1.68 0.80 1.80 1.96 0.60 1.50 1.64 0.606 -0.39 -4.75 -1.52 0.01 -4.13 -1.24 0.33 -4.61 -1.21s.e. 1.34 1.63 0.76 1.33 1.48 0.61 1.24 1.51 0.627 -0.18 -4.85 -1.55 -0.76 -3.55 -0.93 -0.98 -3.43 -1.20s.e. 1.46 1.76 0.74 0.87 1.17 0.61 0.98 1.19 0.578 -0.34 -4.57 -1.31 -0.59 -4.23 -0.85 -1.12 -3.67 -1.05s.e. 1.19 1.46 0.71 1.06 1.41 0.59 0.96 1.32 0.669 0.35 -4.47 -1.25 -1.28 -3.52 -1.35 -1.35 -3.70 -1.40s.e. 1.20 1.45 0.66 1.06 1.44 0.65 1.01 1.41 0.7510 0.83 -4.24 -1.17 -1.74 -2.25 -1.13 -1.01 -3.95 -1.12s.e. 1.30 1.48 0.60 1.14 1.87 0.82 0.97 1.38 0.83893.6. Tables and FiguresTable 3.13: Responses of 25 Portfolios Formed on Size and Book-to-marketThis table reports the result from regression of 25 portfolios formed on size andbook-to-market equity ratio on the expected and surprise monetary policy changes. Thenumber is increasing in size and book to market. The data are of monthly frequency.The whole sample contains 336 months from 1969-1996. All standard errors areheteroskedasticity-robust.Portfolio Growth 2 3 4 ValueSmall -1.97 -2.50 -2.71 -2.69 -2.90s.e. 1.63 1.19 1.10 0.88 0.982 -1.86 -2.59 -2.96 -2.75 -3.03s.e. 1.46 1.09 1.00 0.99 0.973 -2.10 -2.54 -2.91 -2.81 -2.86s.e. 1.27 1.02 0.91 0.87 0.984 -1.61 -1.99 -2.65 -3.15 -3.30s.e. 1.20 1.29 0.85 0.79 0.90Large -0.80 -1.28 -1.01 -2.19 -2.22s.e. 0.88 1.08 1.26 0.82 0.75903.6. Tables and FiguresTable 3.14: Responses of 25 Portfolios Formed on Size and MarketLeverageThis table reports the result from regression of 25 portfolios formed on size and marketleverage on the expected and surprise monetary policy changes. The number is increasingin size and market leverage. The data are of monthly frequency. The whole samplecontains 336 months from 1969-1996. All standard errors are heteroskedasticity-robust.Portfolio Low 2 3 4 HighSmall -2.09 -2.52 -2.78 -2.92 -3.43s.e. 1.55 1.21 1.05 0.96 1.062 -1.97 -2.64 -2.75 -3.28 -3.26s.e. 1.39 1.12 1.10 1.03 0.933 -2.17 -2.69 -2.69 -2.94 -3.13s.e. 1.21 1.05 0.89 0.86 0.984 -1.73 -1.99 -2.47 -2.98 -3.38s.e. 1.15 1.09 0.92 0.80 0.89Large -0.63 -1.37 -1.32 -2.33 -2.44s.e. 0.85 1.06 0.99 0.70 0.71913.6. Tables and FiguresTable 3.15: Wald Test of Difference in Reactions of Leverage Port-foliosThis table reports the test statistics of the difference in the reactions of leverageportfolios, I run a seemingly unrelated regression for each portfolio pair i and j, thentest the difference of the coefficients on the surprise term by Wald test. Each numberbelow the test statistic is the p-value of the test. The whole sample contains 336 monthsfrom 1969-1996. All standard errors are heteroskedasticity-robust. * denotes significantat 10% levelDecile 1 2 3 4 5 6 7 8 9 101(Lowest) 0.64 0.81 0.66 0.69 1.00 2.74* 3.48* 3.78* 3.43*p.v. 0.42 0.37 0.42 0.40 0.32 0.10 0.06 0.05 0.062 0.02 0.02 0.01 0.08 0.93 1.47 1.74 1.45p.v. 0.87 0.90 0.91 0.78 0.33 0.23 0.19 0.233 0.00 0.00 0.01 0.53 0.91 1.14 0.91p.v. 0.99 0.97 0.91 0.47 0.34 0.29 0.344 0.00 0.01 0.45 0.77 0.97 0.77p.v. 0.98 0.90 0.50 0.38 0.33 0.385 0.02 0.55 0.92 1.14 0.92p.v. 0.88 0.46 0.34 0.28 0.346 0.36 0.68 0.89 0.69p.v. 0.55 0.41 0.35 0.417 0.07 0.17 0.08p.v. 0.79 0.68 0.788 0.03 0.00p.v. 0.87 0.999 0.02p.v. 0.8910(Highest)923.6. Tables and FiguresFigure 3.1: Seven-day Cumulated Stock Return after Policy Shock933.6. Tables and FiguresFigure 3.2: Seven-day Cumulated Stock Return after Policy Shock:Recession v.s. Expansion943.6. Tables and FiguresFigure 3.3: Monetary Shock Series and Stock Returns953.6. Tables and FiguresFigure 3.4: Responses of 10 Portfolios Formed on Size963.6. Tables and FiguresFigure 3.5: Responses of 10 Portfolios Formed on Book-to-marketEquity973.6. Tables and FiguresFigure 3.6: Responses of 10 Portfolios Formed on Market Leverage983.6. Tables and FiguresFigure 3.7: 25 Portfolios Formed on Size and Book-to-market Eq-uity993.6. Tables and FiguresFigure 3.8: 25 Portfolios Formed on Size and Market Leverage1003.6. Tables and FiguresFigure 3.9: 25 Portfolios Formed on Size and Market Leverage: 3DVersion101BibliographyAcemoglu, Daron, Aghion, Philippe, Griffith, Rachel, and Zilibotti, Fabrizio, 2010, Verti-cal Integration and Technology: Theory and Evidence, Journal of the European Eco-nomic Association, MIT Press, vol. 8(5), pages 989-1033, 09Armenter, Roc and Hnatkovska, Viktoria V., 2012, The Macroeconomics of Firms’ Sav-ings. FRB of Philadelphia Working Paper No. 12-1.Asker, John and Farre-Mensa, Joan and Ljungqvist, Alexander, 2013, Corporate Invest-ment and Stock Market Listing: A Puzzle? ECGI - Finance Working Paper .Bates, T. W., Kahle, K. M. and Stultz, R. M.,2009 , Why Do U.S. Firms Hold So MuchMore Cash than They Used To?. The Journal of Finance, 64: 1985-2021.Nicholas Bloom, 2009, The Impact of Uncertainty Shocks, Econometrica, EconometricSociety, Econometric Society, vol. 77(3), pages 623-685, 05.Gian Luca Clementi and Dino Palazzo, 2010, Entry, Exit, Firm Dynamics, and AggregateFluctuations, Working Paper Series, The Rimini Centre for Economic Analysis 27-10,The Rimini Centre for Economic Analysis.Eaton, Jonathan, Kortum, Samuel and Kramarz, Francis, 2008, An Anatomy of Inter-national Trade : Evidence from French Firms, Working Papers 2008-29, Centre deRecherche en Economie et Statistique.Feenstra, Robert C., Integration of Trade and Disintegration of Production in the GlobalEconomy, The Journal of Economic Perspectives , Vol. 12, No. 4 (Autumn, 1998), pp.31-50.Feenstra, Robert C. and Hanson,Gordon H, 2004, Intermediaries in Entrepot Trade: HongKong Re-Exports of Chinese Goods, Journal of Economics and Management Strategy,Wiley Blackwell, vol. 13(1), pages 3-35, 03Gao, Xiaodan, 2013, Corporate Cash Hoarding: The Role of Just-in-Time Adoption,working paper.Grossman, Gene M. and Helpman, Elhanan, 2002, Integration versus Outsourcing in In-dustry Equilibrium, The Quarterly Journal of Economics (2002) 117 (1): 85-120Grossman, Gene M. and Helpman, Elhanan, 2005, Outsourcing in a Global Economy,Review of Economic Studies, 72, 135-159.102BibliographyHelpman, Elhanan, Melitz, Marc and Rubinstein, Yona . 2008, Estimating trade flows:trading partners and trading volumes, Quarterly Journal of Economics 123, no. 2: 441-487Kur, Christopher J., 2006, Outstanding Outsourcers: A Firm- and Plant-Level Analysisof Production Sharing, Federal Reserve Board staff working papers in the Finance andEconomics Discussion Series.Lafontaine, Francine, and Slade, Margaret, 2007, Vertical Integration and Firm Bound-aries: The Evidence, Journal of Economic Literature, 45(3): 629-685.McMillan , John, 1995, Reorganizing Vertical Supply Relationships, in H.Siebert, ed.,Trends in Business Organization: Do Participation and Cooperation Increase Compet-itivenesss(Tu¨bingen:J.c.b.Mohr,1995)Shy, Oz and Stenbacka, Rune, 2005 Partial Outsourcing, Monitoring Cost and Mar-ket Structure, 2005, The Canadian Journal of Economics / Revue canadienned’Economique , Vol. 38, No. 4 (Nov., 2005), pp. 1173-1190Tempest, Rone, Barbie and the World Economy, Los Angeles Times, September 22, 1996,A1 and A12.Tisdale, Sallie, Shoe and Tell, The New Republic, September 12, 1994, 10-11.Tian, Can (2011): Technology choice and endogenous productivity dispersion over thebusiness cycles, working paper.Teixeira, Joao C.A., 2011, Outsourcing and Vertical Integration: a Survey of EmpiricalLiterature, working paper.Acemoglu, Daron and Zilibotti, Fabrizio, 1997, Was Prometheus Unbound by Chance?Risk, Diversification, and Growth, Journal of Political Economy, University of ChicagoPress, University of Chicago Press, vol. 105(4), pages 709-51, August.Blanchard, Olivier, and Simon, John, 2001, The Long and Large Decline in U.S. OutputVolatility, Brookings Papers on Economic Activity, no. 1.Fama, Eugene F. and French, Kenneth R., 2004, New lists: Fundamentals and SurvivalRates, Journal of Financial Economics, Elsevier, vol. 73(2), pages 229-269, August.Brown, Gregory and Kapadia, Nishad, 2007, Firm-specific Risk and Equity Market Devel-opment, Journal of Financial Economics, Volume 84, Issue 2, May 2007, Pages 35838Comin, Diego and Mulani, Sunil, 2009, A theory of growth and volatility at the aggregateand firm level, Journal of Monetary Economics, 56(2009),1023-1042Comin, Diego and Philippon, Thomas, 2006, The Rise in Firm-Level Volatility: Causesand Consequences, NBER Macroeconomics Annual 2005, Volume 2Chun, Hyunbaen, Kim, Jung-Wook, Lee, Jason and Morck, Randall, 2004, Patterns of Co-movement: The Role of Information Technology in the U.S. Economy, NBER WorkingPapers 10937, National Bureau of Economic Research, Inc.103BibliographyChun, Hyunbae ,Kim, Jung-Wook , Morck, Randall and Yeung, Bernard, 2008, Cre-ative destruction and firm-specific performance heterogeneity, Journal of Financial Eco-nomics, Elsevier, vol. 89(1), pages 109-135, July.Bargeron, Leonce L., Lehn, Kenneth M. and Zutter, Chad J., Sarbanes-Oxley and corpo-rate risk-taking, Journal of Accounting and Economics, 2010, vol. 49, issue 1-2, pages34-52.Li, Kai, Griffin, Dale, Yue, Heng and Zhao, Longkai, 2013, How Does Culture InfluenceCorporate Risk-Taking? Journal of Corporate Finance 23, 1-22.Davis, Steven J., Haltiwanger, John C., Jarmin, Ron and Miranda, Javier, Volatility andDispersion in Business Growth Rates: Publicly Traded versus Privately Held Firms,NBER Macroeconomics Annual 2006, Volume 21Brown, James R., Fazzari, Steven M. and Petersen, Bruce C., 2009, Financing Innova-tion and Growth: Cash Flow, External Equity, and the 1990s R&D Boom, Journal ofFinance, American Finance Association, vol. 64(1), pages 151-185, 02.McConnell, Margaret M. and Perez-Quiros, Gabriel, 2000, Output Fluctuations in theUnited States: What Has Changed since the Early 1980’s?, American Economic Review,American Economic Association, vol. 90(5), pages 1464-1476, December.Jaimovich, Nir and Henry E. Siu, 2009, The Young, the Old, and the Restless: Demo-graphics and Business Cycle Volatility, American Economic Review, American Eco-nomic Association, American Economic Association, vol. 99(3), pages 804-26, June.Pastor, Lubos and Veronesi, Pietro, 2006, Was there a Nasdaq bubble in the late 1990s?Journal of Financial Economics, Elsevier, vol. 81(1), pages 61-100, July.Clark, Todd E., 2009, Is the Great Moderation over? an empirical analysis, EconomicReview, Federal Reserve Bank of Kansas City, issue Q IV, pages 5-42.Stock, James H., and Mark W. Watson. 2003, Has the Business Cycle Changed and Why?,NBER Chapters, in: NBER Macroeconomics Annual 2002, Volume 17, pages 159-230National Bureau of Economic Research, Inc.Bernanke, B. and Blinder, A., 1992. The federal funds rate and the channels of monetarytransmission. American Economic Review.Bernanke, B. and Gertler, M., 1989. Agency costs, net worth, and business fluctuations.American Economic Review.Bernanke, B. and Gertler, M., 1995. Inside the black box: The credit channel of monetarypolicy transmission. Journal of Economic Perspectives.Bernanke, B., Gertler, M. and Gilchrist, S., 1996. The financial accelerator and the flightto quality. Review of Economics and Statistics.Ben S. Bernanke & Kenneth N. Kuttner, 2005. ”What Explains the Stock Market’s Re-action to Federal Reserve Policy?,” Journal of Finance, American Finance Association,vol. 60(3), pages 1221-1257, 06.104BibliographyBoschen, John, and Leonard Mills, ”The Effects of Countercyclical Policy on Money andInterest Rates: An Evaluation of Evidence from FOMC Documents,” working paper91-20, Federal Reserve Bank of Philadelphia, 1995.Christiano, Lawrence J. & Eichenbaum, Martin & Evans, Charles L., 1999. ”Monetarypolicy shocks: What have we learned and to what end?,” Handbook of Macroeconomics,in: J. B. Taylor & M. Woodford (ed.), Handbook of Macroeconomics, edition 1, volume1, chapter 2, pages 65-148 Elsevier.Cook, T. and Hahn, T., 1989. The effect of changes in the federal funds rate target onmarket interest rates in the 1970s. Journal of Monetary Economics 24.Gallmeyer, Michael F., Hollifield, Burton and Zin, Stanley E., 2005. ”Taylor rules, Mc-Callum rules and the term structure of interest rates,” Journal of Monetary Economics,Elsevier, vol. 52(5), pages 921-950, July.Gertler, Mark and Gilchrist, Simon, 1993. ” The Role of Credit Market Imperfectionsin the Monetary Transmission Mechanism: Arguments and Evidence,” ScandinavianJournal of Economics, Wiley Blackwell, vol. 95(1), pages 43-64.Gertler, Mark and Gilchrist, Simon, 1994. ”Monetary Policy, Business Cycles, and theBehavior of Small Manufacturing Firms,” The Quarterly Journal of Economics, MITPress, vol. 109(2), pages 309-40, May.Guo, Hui, 2004. ”Stock prices, firm size, and changes in the federal funds rate target,”The Quarterly Review of Economics and Finance, Elsevier, vol. 44(4), pages 487-507,September.Gurkaynak, Refet S. & Sack, Brian T. & Swanson, Eric P., 2005. ”Do Actions SpeakLouder Than Words? The Response of Asset Prices to Monetary Policy Actions andStatements,” International Journal of Central Banking, International Journal of CentralBanking, vol. 1(1), May.Gurkaynak, Refet S. & Sack, Brian T. & Swanson, Eric P., 2007. ”Market-Based Mea-sures of Monetary Policy Expectations,” Journal of Business & Economic Statistics,American Statistical Association, vol. 25, pages 201-212, April.Jon Faust & Eric Swanson & Jonathan Wright, 2004. ”Do Federal Reserve Policy SurprisesReveal Superior Information about the Economy?,” Contributions to Macroeconomics,Berkeley Electronic Press, vol. 4(1), pages 1246-1246.Fama, E. and French, K., 1995. Size and book-to-market factors in earnings and returns.Journal of Finance 50.Kiyotaki, Nobuhiro and Moore, John (1997), ”Credit Cycles”, Journal of Political Econ-omy 105 (2): 211C248Kuttner, Kenneth N., 2001. ”Monetary policy surprises and interest rates: Evidence fromthe Fed funds futures market,” Journal of Monetary Economics, Elsevier, vol. 47(3),pages 523-544, June.105BibliographyMonika Piazzesi, 2005. ”Bond Yields and the Federal Reserve,” Journal of Political Econ-omy, University of Chicago Press, vol. 113(2), pages 311-344, April.Piazzesi, Monika & Swanson, Eric T., 2008. ”Futures prices as risk-adjusted forecasts ofmonetary policy,” Journal of Monetary Economics, Elsevier, vol. 55(4), pages 677-691,May.Roberto Rigobon & Brian Sack, 2003. ”Measuring The Reaction Of Monetary Policy ToThe Stock Market,” The Quarterly Journal of Economics, MIT Press, vol. 118(2), pages639-669, May.Rigobon, Roberto & Sack, Brian, 2004. ”The impact of monetary policy on asset prices,”Journal of Monetary Economics, Elsevier, vol. 51(8), pages 1553-1575, November.Romer, C. & Romer, D., 2000. Federal reserve information and the behavior of interestrates. American Economic Review 90.Christina D. & David H. Romer, 2004. ”A New Measure of Monetary Shocks: Derivationand Implications,” American Economic Review, American Economic Association, vol.94(4), pages 1055-1084, September.Thorbecke, W., 1997. On stock market returns and monetary policy. Journal of Finance52.David O. Lucca & Francesco Trebbi, 2009. ”Measuring Central Bank Communication: AnAutomated Approach with Application to FOMC Statements,” NBER Working Papers15367, National Bureau of Economic Research, Inc.106


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