UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on finance and growth in China He, Qichun 2007

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

Item Metadata

Download

Media
831-ubc_2007-317822.pdf [ 7.45MB ]
Metadata
JSON: 831-1.0100616.json
JSON-LD: 831-1.0100616-ld.json
RDF/XML (Pretty): 831-1.0100616-rdf.xml
RDF/JSON: 831-1.0100616-rdf.json
Turtle: 831-1.0100616-turtle.txt
N-Triples: 831-1.0100616-rdf-ntriples.txt
Original Record: 831-1.0100616-source.json
Full Text
831-1.0100616-fulltext.txt
Citation
831-1.0100616.ris

Full Text

ESSAYS ON FINANCE AND GROWTH IN CHINA by Qichun He B .Sc , B . A . , Peking University, 1999 M . A . , McMaster University, 2001 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F D O C T O R O F P H I L O S O P H Y in The Faculty of Graduate Studies (Economics) T H E U N I V E R S I T Y O F BRITISH C O L U M B I A August 2007 © Qichun He, 2007 A B S T R A C T This dissertation contributes to the finance-growth literature theoretically and empirically. The first chapter investigates whether financial deregulation causes economic growth and whether the effect happens through changes in the allocation of credit across sectors or changes in the savings an investment rate. This chapter uses the province-level financial deregulation experi-ence of China from 1981 to 1998 to study these questions. It addresses issues of endogeneity of the growth channels by instrumenting with a series of financial reform policies. It finds that financial deregulation causes economic growth, and the effect is largely through the reallocation of credit across sectors rather than changes in savings and investment rates. The second chapter studies the 'capital-lord-entrepreneur' problem in an endogenous growth model. There are two representative agents a household (the capital lord) financing R & D and an entrepreneur-inventor performing R & D . How do the contractual provisions on how house-holds and entrepreneurs share property rights on inventions when asymmetric information exists affect growth? The credit contract giving entrepreneurs a higher share of monopolistic profit from inventions (i.e. entrepreneur's inventive incentive (EII) ) elicits more entrepreneurs' effort, generating a "bigger cake", but it also decreases the share of households, resulting in a "house-hold's dissaving" effect. To ensure bounded growth, entrepreneur's effort increases as her share increases but at a decreasing rate, so the "bigger cake" effect is decreasing. The "household's dissaving" effect is increasing because the one additional share of cake given by households is bigger given effort is increasing. At the beginning, the "bigger cake" effect dominates, but be-yond a point, "household's dissaving" effect dominates. Therefore the balanced growth rate is an inverted-U function of EII. Whether agricultural resource abundance is a blessing for industrialization is an issue that generates controversy in the development literature and has not been well addressed in the resource-curse literature. Attempting to address this issue, I examine the effect of long-term yearly average rainfall, average and variance of temperature, and sunshine on quality-adjusted farmland per capita and economic growth in China from 1981 to 1998. I find initial quality-adjusted farmland per capita as instrumented by weather indicators has a significantly negative impact on subsequent growth rate, so farmland abundance is a curse for growth. This negative relationship holds true even after controlling for traditional growth factors, population density (i.e. total land per capita), and time effects. Moreover, the weather indicators affect growth only through the channel of initial quality-adjusted farmland per capita, while initial quality-adjusted farmland per capita lowers growth rates mainly through the channel of total factor productivity. i i T A B L E O F C O N T E N T S Abstract ii Table of Contents iii List of Tables .' iv List of Figures v Acknowledgements vi 1 Introduction 1 References : 5 2 Gradual Financial Reform, Sectoral Allocation of Credit, and Economic Growth: Evidence from China 7 2.1 Gradual Financial Reform, Sectoral Allocation of Credit, and Economic Growth: Evidence from China 7 2.2 Tables 32 2.3 Figures 43 2.4 References 44 3 Entrepreneur's Inventive Incentive and Growth: A n Inverted-U Relationship 48 3.1 Entrepreneur's Inventive Incentive and Growth: An Inverted-U Relationship 48 3.2 Figures 69 3.3 References 71 4 Farmland Curse in Industrialization: Evidence from China 74 4.1 Farmland Curse in Industrialization: Evidence from China 74 4.2 Tables 97 4.3 Figures 108 4.4 References 110 5 Concluding Chapter 113 References 116 Appendix 117 iii L I S T O F T A B L E S Table 1.1 32 Table 1.2 32 Table 1.3 33 Table 1.4 34 Table 1.5 34 Table 1.6 35 Table 1.7 36 Table 1.8 37 Table 1.9 38 Table 1.10 : 38 Table 1.11 39 Table 1.12 40 Table 1.13 40 Table 1.14 41 Table 1.15 42 Table 3.1 97 Table 3.2 97 Table 3.3 98 Table 3.4 99 Table 3.5 100 Table 3.6 : 101 Table 3.7 .'....102 Table 3.8 103 Table 3.9 103 Table 3.10 104 Table 3.11 105 Table 3.12 106 Table 3.13 , 106 Table 3.14 107 iv L I S T O F F I G U R E S Figure 1.1 43 Figure 1.2 43 Figure 2.1 69 Figure 2.2 69 Figure 2.3 69 Figure 2.4 70 Figure 2.5 70 Figure 3.1 108 Figure 3.2 108 Figure 3.3 ..109 Figure 3.4 109 v A C K N O W L E D G E M E N T S I offer my enduring gratitude to my supervisor, Professor Paul Beaudry, who has encouraged, supported and unselfishly instructed me to work on Economic Growth with an emphasis on the Chinese Economy. I have learned the strict and scientific attitude towards economic re-searches from Professor Paul Beaudry. And Professor Paul Beaudry's emphasis on empirical work made me rethink and appreciate the interactive relationship between empirical and theo-retical researches. I owe particular thanks to Professor Patrick Francois, whose strict yet friendly approach assists me enormously in finishing my theoretical chapter, and I thank Professor Ashok Kotwal for valuable discussions. I thank the faculty and my fellow students at the department of economics at U B C for valuable discussion. The comments from seminar participants at the Canadian Economics As-sociation 2005 and 2006 Annual Meetings, the Division of Social Science and the department of economics at the Hong Kong University of Science and Technology, the department of economics at the University of Manitoba, the department of economics at the Chinese University of Hong Kong, the 2005 and 2006 T A R G E T Student Workshops at U B C , and the macro-international lunches at U B C are gratefully acknowledged. Special thanks are owed to my wife, who has always believed in me and encouraged me to continue my research in economics even in those difficult search-match days. I am also grateful to my son, Richard Siyi He, who cheers me up with his innocence and smiles and has made my work and life more meaningful. v i "The task is not so much to see what no one yet has seen, but to think what nobody yet has thought about that which everyone sees." -Arthur Schopenhauer (1788~1860) 1 INTRODUCTION Unprecedentedly large volume of growth theory and analysis has been produced in the last decade, spurred by foundational break-through in techniques in theoretical economic growth, namely, incorporating non-competitive pricing behavior, especially monopolistic pricing behav-ior, into general equilibrium framework to explain long-run total factor productivity (TFP) growth. Romer (1990) and Aghion and Howitt (1992) are among the first to achieve the break-through in modeling the intentional R & D by entrepreneurs in driving long-run T F P growth, though economists have long understood and appreciated R & D as the driving force for long-run T F P growth. The endogenous growth models are generally termed as new growth models (NGMs) in the literature. This dissertation contributes to the literature related to N G M s by focusing on the Chinese economy. The Chinese economy is unique in several aspects. First of all, China is a large develop-ing country with huge provincial variations in economic performance. Unobserved cross-country characteristics may render the cross-country regressions undesirable. However, to address in-teresting empirical questions, cross-province regressions within China are subject to, to a lesser extent, the bias from unobserved cross-sectional characteristics as the provinces within China are relatively more homogeneous. Secondly, China has undertaken the opening and reform since 1978 in a gradual approach. This generates huge time variations, which could be changes in underlying economic, cultural, political and even legal institutions. This natural experiment provides an ideal platform for us to examine the role of economic deregulation, institutions, and geography in affecting the path and speed of economic growth. This thesis explores three aspects of the Chinese economy following its reform, namely, (1) the role of more economic freedom, which allows economic agents to make their own decisions, in raising the welfare of an agrarian economy as it moves from a central-planned economy into a free market one; (2) the role of economic institutions, which refers to the distribution of the bargaining power over innovation surplus among innovators and households who finance innovation, in promoting growth; (3) the role of geography, as reflected by weather indicators and agricultural land endowment, in the process of industrialization. First of all, empirically, following the emergence of N G M s , economists are interested in testing which type of growth models, neoclassical growth models (e.g. Solow 1956) versus N G M s (e.g. Romer 1990), is more consistent with the empirical data. Mankiw, Romer and Weil (1992) were the first to derive the empirical specifications based on Solow model, which is tested with empirical data and found to fit the data pretty well. Aghion and Howitt (1998) also derived 1 the empirical specifications based on N G M s , which are found to be consistent with and similar to those of Mankiw, Romer and Weil (1992). Based on the empirical specifications of Mankiw, Romer and Weil (1992), economists tested the role of finance, financial development and financial deregulation in explaining growth (e.g. King and Levine, 1993, Rajan and Zingales 1998, and Levine and Zervos 1998). This is a well-established literature. However, I noticed that there are some unaddressed important issues in the strand of literature, namely, does financial deregulation contribute to economic growth? What is the channel or mechanism through which financial deregulation impacts growth? I believed that the gradual financial deregulation experience of China at the provincial level may help to address the two questions. This is the motivation of the first chapter of my Ph.D. thesis. The theoretical part of the first chapter predicts the following. If government takes the lead in allocating bank credit, sectors that pay government higher tax possess priority in receiving bank credit, though they pay banks lower interest rates. Then inefficient sectoral allocation of credit occurs, but changes with the reforming of the financial system into one where market determines credit allocation. Hence financial deregulation affects sectoral allocation of credit, which in turn affects growth. To test for this, I examine using Chinese panel data the effects of province-level financial deregulation on sectoral allocation of credit and economic growth from 1981 to 1998. I find the following. First, sectoral allocation of credit (measure of financial development), instrumented by the gradually implemented financial reform policies, causes growth. Specifically, real G D P per work growth rates are higher for provinces with faster pace of reallocating loans from state-owned industries to state-owned commerce. Second, sectoral allocation of credit, rather than saving and investment rate, is the only significant channel through which financial deregulation affects growth. The results are confirmed with growth accounting regressions and are robust after controlling for standard neoclassical growth factors, and time and province effects. Secondly, most N G M s do not account for credit market imperfection (Aghion and Howitt 1998). The reason is that even though there are attempts to introduce financial imperfection into N G M s , there is little gain in terms of new economic insights (Aghion and Howitt 1998). As suggested by Aghion and Howitt (1998), a project that focuses on the financial and institu-tional aspects of R & D that involve credit-constraints on individual researchers would enrich the Schumpeterian approach to growth. Motivated by the suggestion of Aghion and Howitt (1998), I started on my thesis by studying the theoretical links between financial market imperfections and growth. However, if I only introduce asymmetric information into N G M s , I would get sim-ilar results as in previous works, meaning little gain in terms of new economic insights (Aghion and Howitt 1998). I further drew insights from Lewis (1955). According to Lewis (1955), there are three ultimate determinants of long-run growth: factor accumulation (Solow 1956), accumulation of knowledge (Romer 1990), and the effort of agents. The first two have been well recognized and studied in 2 the literature. However, how agents' effort affects long-run growth has not been well studied and recognized. In those endogenous growth models with a perfect credit market (e.g. Aghion and Howitt 1992), the role of credit contract is limited, and there is little room for effort (i.e. for entrepreneurs to work harder) to play a role in determining long-run growth. This contrasts with the belief of economists and governments that pro-entrepreneur policies can achieve higher growth. For effort to have room to play a role in determining growth, markets must be imperfect, otherwise the results would be the same as those in standard endogenous growth models (e.g. Romer, 1990, Aghion and Howitt 1992). Therefore, introducing both asymmetric information and entrepreneur's effort may enable us to gain new economic insights. Following the literature, I introduced heterogeneous agents to generate borrowers (entrepreneurs) and lenders (house-holds through financial intermediary). Then by incorporating incentive contracts1 to model the optimal effort level of different type of entrepreneurs, I find economic growth is an inverted-U function of entrepreneur's inventive incentive. A n increase in inventive incentive elicits more entrepreneurs' effort, generating a "bigger cake", but it also decreases the share that goes to households. The "bigger cake effect" dominates at the beginning; but beyond a point, the "smaller household's share effect" dominates, generating the inverted-U. The results are estab-lished under the five necessary and sufficient conditions, namely, the presence of (1) monopolistic competition, (2) two representative agents: a household and an entrepreneur, (3) hidden infor-mation: entrepreneurs' types are unknown to others, (4) hidden action: entrepreneurs' effort is unknown to others, (5) the second derivative of entrepreneur's effort with respect to his share is negative. This becomes the second chapter of my Ph.D. thesis. After finishing the first chapter, I became interested in explaining the huge difference in provincial growth performance of China. And I suspect that financial deregulation cannot wholly explain the huge difference, which pushes me to look into the resource-curse literature. Is agricultural resource abundance a blessing for industrialization? 2 This question, first of all, generates controversy in the development literature. One view is that higher agricultural productivity negatively affects industrialization. For instance, Matsuyama (1992) shows that higher agricultural productivity lowers growth in small open economies. Another view argues that agricultural advantage is good for industrialization (e.g. Rostow 1960 and Nurkse 1953). The third shows that the answer to the question is not unique (e.g. Yanagawa 1996). Secondly, this question has not been well addressed in the resource-curse literature. There is a large literature which examines whether resource abundance is a curse (e.g. Sachs and Warner 1995, Gylfason 2001, Atkinson and Hamilton 2003). However, most papers conclude that resource-curse is driven by mineral resources rather than by agricultural resources. The 1 Optimal incentive contracts in this thesis refer to debt contract which was modeled by Townsend (1979). 2 Agricultural resource abundance, in this paper, is a measure of higher agricultural productivity or agricultural advantage. 3 works on agricultural resources have shown that agricultural resources are insignificant for'either growth or income level (e.g. Papyrakis and Gerlach 2004, Ng 20063), however, most of those works fail to correct for the potential endogeneity problem of agricultural resource indicators. Attempting to contribute to solving the controversy in the development literature, this paper employs Chinese panel data to examine whether initial provincial agricultural resource abun-dance is good for provincial economic growth. Firstly, cross-province analyses within China are advantageous over cross-country regressions (see Papyrakis and Gerlach 2004). The results with China have wide implications for developing countries with a large agricultural sector. Secondly, unlike previous agriculture-resource literature,4 this paper is the first to construct quality-adjusted farmland per capita as the measure of agricultural resource abundance. Thirdly, the indicator of quality-adjusted farmland per capita makes it possible to correct its potential endogeneity problem by using different weather indicators as instruments. The instrumental variable method makes it credible to examine whether there is a causal relationship between agricultural resource endowment and economic growth. Lastly, using weather as instruments connects this paper with the geography/climate and growth literature. Climate as part of geographical characteristics can affect economic de-velopment, but there is a debate over the exact channel of causality. Some show that cli-mate/geography directly affects economic development (e.g. Sachs et al. 1998). Others focus on the channel and have identified it as past events/institution (e.g. Acemoglu, Johnson, and Robinson 2001, and Sokoloff and Engerman 2000).5 This paper contributes by exploring whether agricultural resource abundance is the main channel by which weather affects growth in China. The main empirical findings of the paper are as follows. First of all, estimated by TSLS regressions using different weather indicators as instruments, initial quality-adjusted farmland per capita exerts a significant negative causal effect on subsequent growth rate of real G D P per worker for the Chinese provinces at the 5% level. That is, abundance of quality-adjusted farmland per capita is a curse for real G D P per worker growth. This negative relationship holds true even after controlling for traditional neoclassical growth factors, population density (i.e. total land per capita) and time effects. Secondly, Sagan test accepts the null hypothesis that quality-adjusted farmland per capita is the main channel by which weather indicators have significant effects on growth at the 5% level. Lastly, this paper shows that total factor productivity is the dominant channel by which farmland endowment lowers growth rates. 'Some works differentiate between agricultural resource dependence and abundance (e.g. Ng 2006), while others do not (e.g. Papyrakis and Gerlach 2004). 4 In previous literature, agricultural resource abundance is measured as the value-added of agricultural sectors (farming, forestry, husbandry and fishery), while agricultural resource dependence is measured as value-added of agricultural sectors divided by G D P (e.g. Ng 2006). 5 Some works show that weather causes institution conflict (e.g. Miguel et al. 2003). Some argue that institutional conflict hurts growth (e.g. Easterly and Levine 1997 and Rodrik 1998). Together, they imply that climate affects growth through institutional conflict. 4 References [I] Acemoglu, Daron, Simon Johnson, and James A . Robinson. 2001. "The Colonial Origins of Comparative Development: an Empirical Investigation." American Economic Review 91 (December): 1369-401. [2] Aghion, Philippe, and Peter Howitt. 1998. Endogenous Growth Theory, Cambridge, Mass.: M I T Press. [3] Atkinson, G . and K . Hamilton. 2003. "Savings, Growth and the Resource Curse Hypothe-sis." World Development 31, No. 11, 1793-807. [4] Easterly, Will iam, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Divisions". Quarterly Journal of Economics 112(November): 1203-50. [5] Gylfason, Thorvaldur. 2001. "Natural Resources, Education, and Economic Development." European Economic Review 45: 847-59. [6] King, Robert G . , and Ross Levine. 1993. "Finance and Growth: Schumpeter Might Be Right." Quarterly Journal of Economics 108 (August): 717-37. [7] Levine Ross, and Sara Zervos. 1998. "Stock Markets, Banks, and Economic Growth." Amer-ican Economic Review 88 (June): 537-58. [8] Lewis, Wil l iam A . 1955. The Theory of Economic Growth, London: George Allen & Unwin Ltd . [9] Mankiw, Gregory N . , David Romer, and David N . Weil. 1992. " A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107 (May): 407-37. [10] Miguel, Edward, Shanker Satyanath, and Ernest Sergenti. 2003. "Economic Shocks and Civ i l Conflict: an Instrumental Variables Approach." University of California at Berkeley Working Paper. [II] Ng, Chi-Yung (Eric). 2006. "Is Natural Resource Abundance Beneficial or Detrimental to Output Level and Growth?" University of Western Ontario Working Paper. [12] Nurkse, Ragnar, 1953. Problems of Capital Formation in Underdeveloped Countries. Oxford University Press, Oxford. [13] Papyrakis, Elissaios, and Reyer Gerlach. 2004. "The Resource Curse Hypothesis and Its Transmission Channels." Journal of Comparative Economics 32: 181-93. [14] Rajan, Raghuram G. , and Luigi Zingales. 1998. "Financial Dependence and Growth." Amer-ican Economic Review 88 (June): 559-86. 5 [15] Rodrik, Dani. 1998. "Globalization, Social Conflict, and Economic Growth" (Prebisch Lec-ture), The World Economy 21(2), March. [16] Romer, Paul M . 1990. "Endogenous Technological Change." Journal of Political Economy 98 (October): S71-102 [17] Rostow, W . W . , 1960. The Stages of Economic Growth: A Non-communist Manifesto. Cam-bridge University Press, Cambridge. [18] Sachs, Jeffrey D. , and Andrew Warner. 1995. "Natural Resource Abundance and Economic Growth." NBER Working Paper 5398. [19] Sokoloff, Kenneth L . and Stanley L . Engerman. 2000. "Institutions, Factor Endowments, and Paths of Development in the New World." Journal of Economic Perspectives 14(3): 217-32. [20] Solow, Robert. 1956. " A Contribution to the Theory of Economic Growth." Quarterly Journal of Economics 70 (February): 65-94. [21] Townsend, Robert M . 1979. "Optimal Contracts and Competitive Markets with Costly State Verification." Journal of Economic Theory 21 (October): 265-93. [22] Yanagawa, Noriyuki. 1996. "Economic Development in a World with Many Countries." Journal of Development Economics 49: 271-88. 6 2 GRADUAL FINANCIAL REFORM, SECTORAL ALLOCA-TION OF CREDIT, AND ECONOMIC GROWTH: EVIDENCE FROM CHINA 2.1 Gradual Financial Reform, Sectoral Allocation of Credit and Economics Growth: Evidence from China 2.1.1 Introduction Many countries believe in the gospel of rapid industrialization as the path to economic prosperity. To achieve this, government usually takes the lead in allocating financial credit and favors certain sectors over others. The experience of many countries like South Korea, Japan and China revealed that the government involved itself in financial credit allocation in the process of economic development. In the last decades, many of these countries have begun reforming their old financial system into one where market determines the allocation of financial credit. A n important question often asked is whether financial reform contributes to economic growth? If so, what is the channel or mechanism through which financial reform affects growth? This paper uses the gradual financial reform experience of China to study these questions.6 Before 1978, the central government of China commanded that state-owned industries (SOIs) have priority in receiving bank credit. In return, SOIs paid government higher taxes to finance its spending. The heavy taxes made SOIs pay lower return to financial intermediaries than other sectors. After 1978, China began the gradual process of reforming the financial system into one where financial intermediaries are allowed more autonomy in allocating loans. Since then, China has achieved miraculous economic growth. 7 On the contrary, some governments are impatient in reforming their old financial system. Usually, 'shock therapy' is adopted, ending up with stagnant growth performance. Though there is a vast literature on financial development (intermediation) and economic growth, 8 this paper contributes to the finance-growth nexus literature by examining the gradual financial reform experience of China. First, the miraculous economic growth of China offers a perfect platform to study whether gradual financial reform is one of the underlying generating forces. Second, the Chinese experience provides a case to explore and identify the channel or mechanism (if there is any) through which gradual financial reform affects growth. The issue on growth channels has not been clearly identified by previous literature partly because the channels may be endogenous to the growth process. Gradual financial reform has the advantages over 6For a formal definition of gradual reform, see Dewatripoint and Roland (1992a, b). For China, gradual financial reform refers to a deregulation process that takes a long period of time and it involves a gradual implementation of piece-meal policies that include the provision of more autonomy to state banks to allocate their credits and the legal set up of new banks. 7China's average annual growth rate of real GDP per worker stands at 8% from 1981 to 1998. 8See Eschenbach (2004), and Demirguc-Kunt and Levine (2001) for a survey of the literature. 7 'shock therapy' in dealing w i t h the potential endogeneity problem encountered i n empir ical work . 9 A m o n g the potential channels by which gradual financial reform may affect growth, this paper focuses on two most probable ones. 1 0 The first is capi tal accumulation, that is, channeling more saving towards investment, either by reallocating wi th in the country or by mobi l i z ing formal savings . 1 1 The second involves changing the sectoral allocation of c red i t , 1 2 by which we mean the reallocation of bank loans among different sectors (such as agriculture and state-owned industry) based on the balance sheet of the provincial financial system (see Table 1.1). Th i s paper offers a in tu i t ion why sectoral al location of credit is the more important channel. Then , this paper uses Chinese panel data to dis t inct ly show that gradual financial reforms significantly affect growth only through sectoral al location of credit instead of other channels such as capi tal accumulation. T h e economic in tui t ion goes as fo l lows . 1 3 To analyze the two aforementioned channels, let us base our th ink ing on a one-sector A K model of endogenous growth including financial in te rmedia t ion . 1 4 A t the in i t i a l stage, government takes the lead i n a l locat ing financial credit through the o ld financial system to sectors possessing the same A K technology. Sectors that pay government higher taxes have prior i ty i n receiving financial credit. G i v e n the same A K technology, sectors subjected to higher taxes pay financial intermediaries lower interest rates. In a Ramsey economy wi th u t i l i ty max imiz ing households, i f households receive lower interest rate from financing projects, the balanced growth rate w i l l be lower. Hence, government tak ing lead in al locat ing financial credit results i n inefficient sectoral allocation of credit, which in tu rn lowers the balanced growth rate. The inefficient sectoral allocation of credit changes wi th the process of gradually reform-ing the o ld financial system to one where market determines the allocation of financial credit. G r a d u a l financial deregula t ion 1 5 means that the government gradually provides financial inter-mediaries more autonomy in allocating their credit. Since financial intermediaries earn lower return from sectors that pay higher taxes, financial intermediaries w i l l reallocate loans out of such sectors after financial deregulation, resulting in higher returns to financial intermediaries and higher growth of the economy. Thus, i n the reform process, sectoral al location of credit is an "Gradual financial reform policies can be instruments for the channels not only because they have time variations, but also because, cross-sectionally, 'shock therapy' cannot function as instruments since all provinces receive the same treatment. 1 0 I also consider other potential channels such as saving, export, F D I and human capital and it is shown that they do not provide significant channels by which financial reforms affect growth. u M c K i n n o n (1973) and Shaw (1973) highlighted that financial reform mobilizes saving. Higher level of saving transformed into greater capital formulation brings higher growth according to neoclassical growth models. Cross-country regressions find that national savings are affected by financial liberalization (Loayza, Schmidt-Hebbel and Serven 2000). 1 2 Sectoral allocation of credit refers to allocation of investment among different sectors, which is related to investment efficiency or T F P . Capital accumulation refers to investment volume. 1 1 The theoretical model in the first chapter is taken out. However, it is available upon request. M F o r the original one-sector model of endogenous growth, see Barro and Sala-i-Martin (1995). 1 5 I n this paper, I refer to China financial reform and financial deregulation interchangeably. 8 important determinant of economic growth. Gradual financial reform affects sectoral allocation of credit, and thus growth. In addition, in the Ramsey model, households' willingness to save is governed by parameters of time preference and consumption smoothing. Given higher interest rates, fixed time preference tends to increase household saving, while consumption smoothing tends to decrease household saving. Hence, the effect of financial reform on savings rate (thus investment rate in the model) is ambiguous and much smaller. Sectoral allocation of credit is the dominant channel through which gradual financial reform affects growth. To test for this, this paper employs Chinese panel data to determine the two aforemen-tioned channels via which financial deregulation affects growth. The novelty of my approach lies in its transparency about the channel via which financial reform affects growth. It im-proves on previous works that commonly employ financial reform indicators in the regression but fail to identify the channels of growth (see Jayaratne and Strahan, 1996). The cross-province regressions employed in the current paper are advantageous over the commonly reported cross-country regressions given that it is more natural to assume aggregate production function for cross-province analyses (see Papyrakis and Gerlach 2004). In addition, I take account of the possibility that both capital accumulation and sectoral allocation of credit may be endogenous to the growth process by noting that financial reform policies that are implemented during the gradual financial reform period in China can be used as an instrumental variable that overcomes the endogeneity problem. 1 6 In so doing, the approach that I use in the paper concurs with Levine's (2005) suggestions of considering the policy and legal aspects of financial development in testing the finance-growth nexus. In addition, the empirical study delivers new insights and policy implications when compared with cross-country regressions that use legal-origins as instruments for financial de-velopment (e.g. Levine, Loayza and Beck 2000). It has been highlighted in the literature that there are problems using legal origins as instruments for financial development: legal origins may work on growth through institution (Acemoglu, Johnson and Robinson, 2001), and they are potentially weak instruments (Coviello 2005). Specifically, I firstly examine whether financial reform affects capital accumulation and sec-toral allocation of credit by using data for China's 27 provinces from 1981 to 1998. Sectoral allocation of credit is classified into within the bank loans to traditional sectors (agriculture, state-owned industries and commerce), 1 7 how the loans are allocated among agriculture, state-owned industries and commerce (see Table 1.2).18 Financial reform refers to financial reform policy indicators which are constructed by using indicator variables and population as weights 1 ( iThe choice of financial reform policies as instruments may not be perfect because polices may also be endogenous. The validity of the instruments is discussed later. 1 7 Non-traditional sectors are private enterprises, foreign-related ones (foreign-owned ones, sino-foreign joint venture ones, and sino-foreign cooperative ones), and others (see Table 1.1). Non-traditional sectors can be viewed as non-stated-owned ones. 1 8For China, bank loans are important sources of firms' external financing (Kuijs 2005). 9 for financial reform policies. The empirical results show that financial reform policy indicators have significant effects on sectoral allocation of credit at the 5% level and insignificant effects on physical capital investment rate at the 10% level. Second, I examine whether sectoral allocation of credit affects growth. Sectoral allocation of credit may be endogenous to the growth process. Financial reform policy indicators are chosen as instruments to deal with the endogeneity problem. The results are as follows. First, T S L S regressions show that the exogenous financial reform policy component of sectoral allocation of credit causes real G D P per worker growth at the 5% level. Growth rates are higher for those provinces with a faster pace of reallocating bank loans out of SOIs. This paper controls the credit reallocation effect between traditional sectors and non-traditional sectors and shows that the growth-promoting effect mainly comes from the credit reallocation within traditional (state-owned) sectors. Secondly, Sargan tests cannot reject the null hypothesis that financial deregulation affects growth only through sectoral allocation of credit at the 10% level. The IV results hold in robustness checks. The results are robust after controlling for initial real G D P per worker, labor force growth, human and physical capital investment, fiscal expenditure, export, FDI , and time and province effects. The results are also robust to alternative measures of sectoral allocation of credit and different combinations of financial deregulation policy indicators. In addition, growth accounting regressions show that sectoral allocation of credit significantly affects total factor productivity growth at the 5% level. The estimated magnitudes of IV regressions are significant for sectoral allocation of credit. For example, for Heilongjiang and Ji l in (see Figure 1.2), controlling for the loans to agriculture, if Heilongjiang could lower its share of bank loans to SOIs to equal that of Jilin, then its annual average growth rate of real G D P per worker would have increased by 0.67% for 1981-1986, 1.88% for 1987-1992, and 1.66% for 1993-1998. The magnitudes of the effect of gradual financial reform on growth are also large. For example, for Liaoning and Jil in, controlling for other factors, if Jilin could have the same level of gradual financial reform policy index Bank (see Table 4) during the same period, this would have increased its annual average growth rate of real G D P per worker by 0.24% for 1981-1986, 0.96% for 1987-1992, and 3.04% for 1993-1998 by decreasing the share of loans to SOIs. This paper is related to the literature on financial development and economic growth. There are numerous evidence of the financial development and growth linkages that are based upon country-level data (King and Levine, 1993), firm (industry)-level data (Rajan and Zingales, 1998), single country's panel data (Jayaratne and Strahan, 1996), and stock and bond markets data (Levine and Zervos, 1998). This paper is also related to the literature on state-level banking deregulation and growth (e.g. Dehejia and Lleras-Muney 2003), on China's saving and investment (e.g. Kuijs 2005), on new growth models (e.g. Aghion and Howitt 1998), and on growth accounting (e.g. Chow 1993). The paper is organized as follows. Section 2.1.2 describes the data and measures of the variables. Section 2.1.3 tests the effects of financial reform on growth via the two channels and 10 only the channel with a significant effect on growth is examined in section 2.1.4. Section 2.1.5 checks the robustness of the empirical results. Section 2.1.6 concludes. 2.1.2 T h e D a t a 2.1.2.1 Empirical Specifications Given the two potential channels (investment rate and sectoral allocation of credit) through which financial reform may affect growth, this paper first examines whether financial reforms have any effect on the two channels. Unless financial reforms have significant effects on the channel itself, a channel cannot be the means through which financial reforms significantly affect growth. To examine this, I utilize the following empirical formulation Dep. = ao + ^^aj (Financial Reform) ^  + Other Control Variables (1) j where Dep. is the dependent variable: either investment rate or indicators of sectoral allocation of credit; (Financial Reform)^ s are the measures of financial deregulation policies, while the subscript j stands for different types of financial deregulation policies, which are described in detail below; other control variables are those generally viewed important for growth: for example, initial real G D P per worker, labor force growth, human capital investment rate, fiscal expenditure, F D I and export to G D P ratio. If the channel turns out to be significant, then the paper empirically assesses whether the channel has effects on growth. I utilize the following standard empirical specification, 1 9 which both neoclassical (see Mankiw, Romer and Weil 1992) and endogenous growth models (see Aghion and Howitt 1998) have used: g — (30 + m (Indicator'j) + Other Control Variables (2.1) i where g is growth rate of real G D P per worker; 2 0 ' 2 1 Indicator's are the measures of sectoral allocation of credit, which are described in detail below (see Table 1.2); the subscript i stands for sector i; other control variables are the same as in equation (1) that are viewed important for growth. This empirical formulation is consistent with cross-country (cross-region) regressions with conditional convergence. Thus, when it is necessary to control for other neoclassical factors to check the robustness of the results, the other neoclassical factors are simply added into the RHS of equations (1) and (2.1). A specific formulation of equation (2.1) without time and province effects is as follows U )The theoretical part abstracts from deriving the empirical specifications used here. For references on deriving the empirical specifications for endogenous growth models, see Aghion and Howitt (1998). 2 0 This measure can deal with the concern raised by Young (2003) that China's economic growth is related to its changing participation rate of its labor force. 2 1 A s in Young (2003), the GDP deflator used to calculate real GDP may not be perfectly measured. However, measurement error of dependent variable has no effect on the results. 11 9t = p0 + ^pi\n(Indimtc^i)t+(33ln^j + (3 4ln (S chool) t + (3b\n{Fiscal\ In ( y ) -ln(n + g + 6)t + 37 In (Export)t + Bs In (FDI)t + et (2.2) where In {j;)t_1 is the initial real G D P per worker, School is human capital, y is nominal investment rate, ln(n + g + 6) measures labor force growth, 2 2 Fiscal, FDI, and Export are fiscal expenditure, foreign direct investment and export to G D P ratio respectively. The channel of capital accumulation is measured as y. Hence measures of sectoral allocation of credit for the second channel need to be constructed, which is performed in section 2.1.2.2. The indicators of financial deregulation policies are constructed in section 2.1.2.3. The other variables in equation (2.2) are constructed in section 2.1.2.4. 2.1.2.2 Constructing Indicators of Sectoral Allocation of Credit The Chinese Academy of Sciences' Natural Resources Database provides detailed data on the balance sheet of national banks of China from 1980 to 1998. The data-set contains total loans, and detailed uses of loans for Chinese National banks at provincial level (see Table l . l ) . 2 3 The panel data on the national banking system from 1980 to 1998 are the most complete in providing measures on the sectoral allocation of credit. 2 4 Table 1 is the typical layout of the balance sheet of a provincial national banking system for the uses of funds. There is no overlap among the same level divisions, that is, the summation of the data of all sub-divisions is the datum for the upper-level division. For the 27 provinces' uses of funds between 1980 and 1998, I have complete data on short-term loans to industrial enterprises, to commercial enterprises, to agriculture, on middle-term and long-term loans, and on total loans. Given the availability of the data, this paper employs those indicators of sectoral allocation of credit listed in Table 1.2. In Table 1.2, SOE is the ratio of short-term loans allocated to sectors agriculture, state-owned industry and commerce in total loans. The average of the ratio for the 18 years is 74% (see Table 1.3). Thus, short-term loans to the three sectors consist of an even higher share in total short-term loans, which makes the sectoral allocation of short-term loans within agriculture, 2 2 This is like a restricted form regression on investment rate and labor force growth. See Mankiw, Romer and Weil (1992). 2 ! For some provinces, the data for some years are missing, in which case they are found in "A Collection of New China's 50-year Statistical Data". 2 4 For China, the term "financial system" appeared in provincial statistical yearbooks after 1997. Half of the provinces have data on financial systems since 1992. The other half have data on national banks only. 12 state-owned industry and commerce good indicators of sectoral allocation of credit. 2 5 SOE will be included in equation (2.2) to control for the total loans allocated to other sectors than the three ones. This paper reports the results with indicators A G R , IND, C O M , and SOE. The results with alternative measures of sectoral allocation of credit are reported in the robustness check section. China's provincial national banks consist of the dominant part of whole financial system (see Figure 1.1). Before the early 1980s, the financial system mainly consisted of national banks. This situation did not change greatly until early 1990s. The ratio of national bank loans over total loans in the financial system is approximately 90% or above for most provinces. Even for those well-developed coastal provinces, such as Tianjin, the ratio is above 90% in the 1980s and approximately 85% in the 1990s. Thus, the data on national banks for the 18 years can proxy for the whole financial system. The financial intermediation provided by other financial institutions is also important for China after its reform (see Naughton 1998, and Brandt and Zhu 2000). This paper uses only data for national banks because the data on national banks are the most consistent and complete. As shown later on, all the financial deregulation policies are measured. Financial deregulation will not only affect the credit reallocation among sectors for national banks, it will also affect the credit reallocation among sectors for other financial intermediaries. Therefore, using only data for the national banks actually under-estimates the combined credit reallocation effect among sectors of the whole financial system. Finally, G D P is the flow variable defined over a year, while the data on sectoral allocation of credit are stock variables, which are year-end data. I follow King and Levine (1995) by using the average of two consecutive years' stock variables as the flow variable for the latter year. Table 1.3 lists some of the final data. For example, the first three data points are for Beijing, whose average annual growth rate of real G D P per worker is 6.0% for 1981-86, 5.0% for 1987-92, and 9.5% for 1993-98. Its corresponding sectoral allocation of credit indicators, A G R and SOE, are listed in column 3 and 4. 2.1.2.3 Constructing Indicators of Financial Reform Policies 2.1.2.3.1 Describing and Finding Financial Deregulation Policies In 1978, China began to reform its financial system from one where the government takes the lead in allocating financial credit to one where market forces determine the allocation. The 2 5 Some facts can minimize a suspicion that the three sectors have not contributed much to the growth performance of China in the 18 years. Firstly, the FRS greatly improved the productivity growth of agriculture (see Lin 1992). Secondly, World Bank (1996) has estimated that the average annual growth rate of TFP in the state sector was 2.4 per cent per year from 1980 to 1988. Recently, Young (2003) found that the TFP growth of non-agriculture sectors is similar to that of four Asian dragons. Thus, it is not the case, as believed by some people, that the source of Chinese economic growth is mainly from the private sectors. 13 ongoing process is enacted by the central government of Ch ina 2 6 which gradually gives financial intermediaries more autonomy to allocate their loans. Each year, the central government chooses particular financial reforming policies which are performed only in some designated cities. After such policies mature, the central government may spread them to the whole province, further to several provinces, and finally to the whole country. The gradual financial reform of China has been studied by many. For example, Park and Sehrt (2001) use Chinese provincial data from 1991 to 1997 to test whether financial reform has increased the efficiency of intermediation by financial intermediaries. They find that policy lend-ing by state banks did not fall in their data sample. Naughton (1998) reviews the achievements and challenges of China's financial reform. Lardy (1998) also discusses in depth the achieve-ments and challenges of China's financial system and its reform. Shirk (2003) looks into the political logic behind the gradual financial reform of China. This paper attempts to quantitative evaluate the effect of China's financial reform on economic growth. First of all, the financial reform policies have to be quantitatively measured. The chapter "Fiscal, Finance, and Insurance" in the book "The Big Economic Events since China's Reform and Opening-up (1978-1998)"27 edited by the Institute of Economic Research, the China Academy of Social Sciences, documents the important financial reform policies which the central government carried out from 1978 to 1998. The attractiveness of the financial reforming policies in the book is its provision for authority, uniformity and objectivity. With such policies now located, subsection 2.1.2.3.2 divides them into different policy indicators and turns those policy indicators into yearly provincial indexes. 2.1.2.3.2 Constructing Financial Deregulation Policy indexes It is no easy task to divide financial reforming policies and convert them into policy indexes. The Chinese Economists Society organized an international symposium on Chinese financial reform at the University of Southern California in 1997. The symposium divided Chinese financial reform as follows: 1. Domestic Financial Deregulation (1) Reforms of the banking sector: a) Functions of commercial banks and policy banks; b) Regulations of banking institutions in China; c) Roles of foreign banks in enhancing competition and transferring technologies; d) Possibilities of more domestic private banks. (2) Non-bank Financial Institutions and Regulations: a) Insurance market; b) Non-bank deposit market; c) Non-bank deposit-taking institutions; d) Gray and black credit market for small loans. Central government stands for national government agencies, departments, councils, etc. " T h e title is "Zhongguo Gaige Kaifang Y i l a i Jingji Dashi Ji(1978-1998)" in Chinese. 14 2. Capital Market Development (1) on Equity and Bond market (2) on Foreign Exchange Market The symposium's division of the financial reforming policies gives a basis to divide the numerous financial reform policies documented in the previously mentioned book. The specific divisions of the policies are summarized in Table 1.4. There is no overlap among the policy indicators, that is, each policy is counted once. Fiscal reform policies in the same chapter are excluded. For example, in 1986, the People's Bank of China and the National System Reforms Council of China jointly gave notice that the cities of Guangzhou, Chongqing, Wuhan, Shenyang and Changzhou should conduct financial system reforms experiments; then in 1988, the State Council of China ratified Fujian Province to conduct the same financial system reforms. And I put this policy under the indicator Bank. The formula to construct the policy index for each policy indicator for a province in a particular year, using 1990 as an example, is as follows: Index = | Total Population of City i in 1990 j . 1 9 9 0 ^ . 1 9 9 0 \ 11 GX j \ i T ° t a l P ° P u l a t i o n o f t h e P r o v i n c e i n !990 c i V ) where i ? j 9 9 0 is an indicator variable that equals one if city i receives a financial reform policy j; 7p 9 9 0 is an indicator variable that equals one i f a financial reform policy j is conducted in the whole province. I add together all policies (the j's) in the same category defined in Table 1.5 in and before year 1990 for all the cities (the i's) within a province. For example, if an area received a financial reform policy in 1988, the policy would also have effect in 1990, which therefore has to be taken into account. Thus, the construction of this index is the same as that of building the index for the new policies in year 1990 and carrying the index to all the years after 1990. City's total population data are taken from Statistical Yearbook on China's Cities (1986-1999).28 For each indicator in Table 1.4, its index is built using the formula in equation (3). In the case of the indicator Bank, if a deregulation policy is conducted through one of the big four state banks of China , 2 9 a weight of \ is given to the policy. In the case of the indicator Newbank, a weight of ^ is given as there are already big four state banks in operation. No adjustments are made for the other three indicators. Table 1.5 lists the correlation among the built policy indexes, which states that the policy indexes are significantly correlated with one another. The built policy indexes have explicit variations across-province and across-time. Figure 1.2 plots the provincial policy index Bank, which shows that financial reform policies are unequally distributed across provinces: firstly, coastal provinces (those provinces on the right side of Figure 1.2) receive more banking reform policies; secondly, within coastal area, some provinces receive 2 8 I only have Statistical Yearbooks after 1986. The data of 1986 are used for 1981 to 1986. 2 9The big four state banks of China axe: Bank of China (BOC), the Industrial and Commercial Bank of China (ICBC), China Construction Bank, the Agricultural Bank of China (ABC). 15 more banking reform policies than other provinces do, for example, Canton and Fujian receive more policies than Jiangsu and Zhejiang do. 2.1.2.4 Data on Neoclassical Growth Factors In equation (2.2), In (j;)t_l is the initial real G D P per worker, which takes the value of the beginning year of each sub-period. 3 0 A l l other variables are six-year averages to eliminate business cycle phenomenon. School is measure of human capital and is measured as secondary school enrollment to the total number of workers according to Mankiw Romer and Weil (1992).3 1 Secondary school enrollment is the sum of student enrollment for middle schools (grade 7 to 9) and high schools (grade 10 to 12). Fiscal is fiscal expenditure to G D P ratio. For labor force growth measure: In (n + g + 6), I follow Klenow and Rodriguez-Clare (1997) and use 0.08 for (g + 6). Both F D I and Export are nominal values divided by nominal G D P . (^) is the nominal physical capital investment rate. There are problems with the various deflators of China (see Young 2003). The Chinese local statistical bureau tends to under report the deflators for investment relative to that of G D P , thus if one uses those deflators to measure real investment rate, some provinces would have unreasonably high real investment rate. 3 2 In this paper, I assume the deflators of investment and G D P grow at the same rate, 3 3 which likely produces a less severe problem for my paper. There are 31 provinces in China . 3 4 Before 1997, Chongqing was a city of Sichuan province, thus both of them are excluded from the sample for the following reason. There are no data for Chongqing before 1997; the data for Sichuan before 1997 contain those for Chongqing while those after 1997 do not. The same argument applies for Hainan and Guangdong. However, there is a complete set of data for Guangdong province, thus it is kept while Hainan is dropped. Tibet is excluded because there are no data on its agricultural loans. I only employ data from 1981 to 1998. For sectoral allocation of credit, the data after 1998 are not available, and the data for 1978 and 1979 are not complete. I follow King and Levine (1993) to obtain the deflator for the indicators of sectoral allocation of credit by using the average of two consecutive years' stock variables as the flow variable for the latter year. In other words, 1980 is excluded for the lack of data in 1979. In summary, my data comprises 27 provinces for 18 years (1981-1998). I take six-year averages for the Chinese panel data to avoid the influence from business cycle phenomenon, which produces three sub-periods: 1981-1986, 1987-1992, and 1993-1998. Each : i 0The dependent variable is in percentage, to calculate speed of convergence, A, the estimated coefficient on In ( £ ) ( 1 should be divided by 100. And for (16.2): = ^ . 3 1 The human capital measure that is mostly correlated with growth is the male secondary schooling by Barro and Lee (1993). Later, this paper shows that the results on sectoral allocation of credits hold true with only initial GDP as the control variable. Thus, School is not a perfect measure of human capital but will not affect the results of the paper. 3 2 Weeks and Yao (2003) produced a particularly insignificant coefficient for real investment rate. '"Some studies on Chinese provincial conditional convergence adopted this approach. 3 4 In China, out of the 31 provincial governments, four are municipalities and four are autonomous regions. For simplicity, I delegate the usage 'province' to all. 16 province has three data points. The data are gathered from various sources. I have used provincial statistical yearbooks and Statistical Yearbook of China for the data on real G D P per worker, secondary school enrollment rate, fiscal expenditure, physical capital investment rate, labor force growth, export and F D I . 3 5 The data on sectoral allocation of credit are taken from the China Academy of Sciences' Natural Resources Database and " A Collection of New China's 50-year Statistical Data" . 3 6 There are few data on sectoral allocation of credit missing after combining the two data sources, in which case I insert those missing data points using the linear average method. 3 7 Finally, I obtain Chinese cities' total population data from Statistical Yearbook on China's Cities (1986-1999). The panel data of China have been used by economists (e.g. Weeks and Yao 2003) to evaluate the conditional convergence of Chinese provinces. 2.1.3 An Empirical Assessment of the Two Channels Section 2.1.3.1 examines if financial reform affects investment rate, the first channel through which financial deregulation may affect growth. Section 2.1.3.2 examines the second channel of sectoral allocation of credit. If a channel is deemed to be significant, its effect on growth will be examined in section 2.1.4. In this paper, two groups of financial deregulation indicators are chosen to represent financial reform policies. The first group consists of indicators: Bank, Newbank and Stock. This group is connected with the literature on banking sector development and growth (e.g. King and Levine 1993) and on stock and bond market development and growth (e.g. Levine and Zervos 1998). The second group consists of Bank, Newbank, Stock and Nonbank, which presents a more thorough representation of financial system reforms. This is in line with Demirguc-Kunt and Levine (2001), in which the effect of non-bank financial sector development on growth is examined. The results with other combinations of indicators are discussed in the robustness check section. 2.1.3.1 Does Financial Reform Affect Investment Rate? In the literature, it is believed that financial reform usually works on growth through increasing national saving and investment. McKinnon (1973) and Shaw (1973) are among the first to highlight the role of financial sector in mobilizing savings which contributes to investment and growth. Higher level of saving results in greater capital accumulation, which brings higher growth according to neoclassical growth models (Solow 1956). Jbil i , Enders and Treichel (1997) show that financial reforms contributed to higher savings in three African countries. In addition, ; 15Qinghai province doesn't have any FDI for 1981-1986, and the datum from 1987-1992 is used. : i <Tn Chinese, the title is "Xinzhongguo 50 nian Tongji Ziliao Huibian". "For example, if the data on 1991 and 1992 are missing, I use the data on 1990 and 1993 (denoted by D I M O and D 1 9 9 3 ) , then impose ( D 1 9 9 0 + D i w - D w « ) for 1991 and [£>i 9 9 o + f (£"1993 - £1990)] for 1992. 17 based on cross-country regressions, Loayza, Schmidt-Hebbel and Serven (2000) find that national savings are affected by financial liberalization. For China, the national levels of saving and investment are roughly equal with domestic savings being the predominant source of financing investment (Kuijs 2005). Modigliani and Cao (2004) document evidence that China's household savings have risen partly due to the economic reforms since 1978. Kraay (2000) shows that China's national savings rate is around 16 percent higher than the international average.38 The higher savings rate in China is mainly driven by differences in the enterprise sector saving rate rather than households and government savings (Kuijs, 2005). I verify if financial deregulation works on growth through affecting the specific components of saving (fiscal saving, household savings, for example) according to the balance sheet of provincial financial system. The results show that this does not function as a channel. Hence, I only consider the impact of financial deregulation on provincial physical capital investment rate following the cross-country regressions approach undertaken by Loayza, Schmidt-Hebbel and Serven (2000). Empirical specifications in equation (15) with a set of controls are estimated, with the results reported in Table 1.6. In Table 1.6, column 1 reports the results for the first group of financial deregulation indica-tors, while column 2 reports the second group. It can be seen that the F-test statistics on either group of financial reform indicators are notably smaller, and their associated p-values are over 10%. There is evidence to suggest that financial reforms (be it banking sector and stock market reforms, or banking sector, non-bank sector and stock market reforms) do not have significant effects on physical capital investment rate at the 10% level. The results are robust having con-trolled for variables that are important for growth, and the time and province effects. Therefore, physical capital investment is ruled out as the significant channel via which gradual financial deregulation affects growth. The lack of evidence on the link between financial reform and capital investment seems to contradict the common belief that gradual financial reform may affect growth through increasing saving and investment. The results also imply that developing economies should focus their reform effort more on sectoral allocation of credit rather than on saving and investment. 2.1.3.2 Does Financial Reform Affect Sectoral Allocation of Credit? Here I examine if financial deregulation has a significant effect on sectoral allocation of credit. I estimate empirical specifications given in equation (1) using a set of control variables in equation (2.2). Referring to the results reported in Table 1.7, column 1 shows the results for ln(AGR) with the first group of financial deregulation indicators, and column 2 shows the results for ln(AGR) with the second group. The p-values of the F-test statistics on either group of financial reform indicators are close to 0% which suggest that either group of financial reform policy indicators has a very significant effect on ln(AGR) at the 1% significance level. 1 , 8 The national saving rate of China is on average 37 percent between 1978 and 1995, compared to an international average of 21 percent (see Kuijs 2005). 18 Similarly, in Table 1.7, column 1' reports the results for ln(IND) with the first group of financial policy indicators, while column 2' reports the results for ln(IND) with the second group. Although the F-test statistics on either group of financial reform indicators are smaller, the associated p-values of the F-test are still below the 5% significance level thereby suggesting that either group of financial reform policy indicators exerts a significant effect on ln(IND). Moreover, the impact of the Bank indicator on ln(IND) is defined by a U-shaped relationship as given by the negative linear component that is significant at the 5% level. The impact of stock market reform policies on ln(AGR) is also defined by an inverted U-shaped relationship which is equally significant at the 5% level. By and large, financial reform policies tend to non-linearly decrease ln(IND), the share of loans allocated to SOIs, while financial reform policies tend to increase ln (AGR) , the fraction of loans to allocated to agriculture. Section 2.1.3 concludes that financial reforms do not exert significant effects on physical capital investment rate at the 10% significance level, while they do have impacts on sectoral allocation of credit at the 5% significance level. These results are robust having controlled for other variables which are important for growth, as well as time and province effects. Nonetheless, it should be noted that the lack of statistical evidence of the impact of financial reforms on physical capital at the 10% significance level does not override the role of financial reforms in influencing the rate of physical capital investment. On balance, the results suggest that between the two potential channels that financial deregulation may affect growth, sectoral allocation of credit operates as a more important channel. Based on these results, section 2.1.4 proceeds by examining if sectoral allocation of credit significantly affects the growth rate of real G D P per worker. 2.1.4 Does Sectoral Allocation of Credit Affect Growth? This section examines if sectoral allocation of credit is an important determinant of economic growth given that financial reforms have significant effects on sectoral allocation of credit. 3 9 If it turns out that sectoral allocation of credit also affects growth, it can be inferred that financial reforms work on growth through affecting sectoral allocation of credit. Since financial reforms do not have a significant effect on the investment rate, it is included as a regressor in empirical specification equation (2.2). 4 0 2.1.4.1 OLS Estimation and Results The OLS regression results are reported in Table 1.8, column 2 without ln(SOE) and column 5 with ln(SOE) to control for the effect of loans allocated to those sectors other than agriculture, ! 9Many empirical works on financial development and economic growth conclude that financial de-velopment matters for growth at the early stage of development (e.g. Aghion et al 2005). Thus, here, the sectoral allocation of credit is an important determinant of economic is best interpreted as a transi-tory phenomenon, or a phenomenon at an early stage of development or in the gradual financial reform process. . 4 0 (2.2) compacts physical capital investment rate with labor force growth as a regressor, which is a restricted form as in Mankiw Romer and Weil (1992). 19 state-owned industry and commerce.4 1 The ln(COM) is excluded to avoid multi-collinearity. Both models produce similar results with /?, > 0, a positive and insignificant estimated coefficient on In(AGR); (32 < 0, a negative but insignificant estimated coefficient on In(IND), which says that after controlling for the loans to agriculture, the reallocation of loans from state-owned commerce to state-owned industry has insignificant effects on economic growth. The coefficients on ln(School) are positive and significant at the 5% level, and those on ln n+ff+a) a r e P o s i f ive with similar magnitude and significant at the 10% level. The coeffi-cients on In {j£)t-\ a r e negative and significant at the 5% level, which confirms the existence of conditional convergence. The speed of convergence, A, is 2.7% per year. 4 2 What should be noted is that, if sectoral allocation of credit is endogenous to the growth process, the simultaneity bias will make the estimated coefficient in OLS regression inconsistent. Using ln(IND) as an example, it can be shown that p l im P ° L S = P2 +  c o r r ( l n (IND), e) — — (4) °~\n(IND) where corrQ stands for correlation, and a > 0 stands for standard deviation. Given the asymp-totic property of the OLS regression, if corr (ln (IND), e) > 0 , 4 3 then the estimated coefficient from OLS regression will attenuate towards zero if /32 < 0. The endogeneity problem will be tested by using Hausman test in section 2.1.4.2 which deals with the endogeneity problem of sectoral allocation of credit. 2.1.4.2 IV Estimation and Results: Endogeneity of Sectoral Allocation of Credit To deal with the potential endogeneity problem of sectoral allocation of loans, I utilize the instrumental variable approach and locate instruments that affect growth only through changing sectoral allocation of credit. 2.1.4.2.1 Using Financial Reform Policies as Instrumental Variables In cross-country regressions, to deal with the endogeneity problem of financial development, the popular instrument used is the legal-origin of countries (e.g. Demirkunt and Levine 2001). Firstly, though in those TSLS regressions, the exogenous legal component of financial develop-ment causes growth, little policy suggestion is implied. It is unbelievable that a government wishes to achieve higher growth by changing its legal origin. In other words, the legal-origin ap-proach has solved the direction of causality between financial development and economic growth 4 'The coefficients on ln(SOE) are insignificantly negative, which tells that reallocating loans from traditional sectors to non-traditional (private) ones is only insignificantly good for growth. 4 2 Weeks and Yao(2003) found that A is around 2% per year using G M M method. 4 f This inequality is true given that as a result of growth in an economy, it will witness industrialization. 20 in a way that is not desirable. Secondly, in addition to the critics of legal origins as weak in-struments (Coviello 2005), it is hard to justify that legal origins work on growth only through the channel of financial development, and there is an equally compelling channel of institutions (Acemoglu, Johnson, and Robinson 2001) as discussed in Coviello (2005). Levine (2005) suggests that future work should look into the policy aspects of financial development in explaining growth to solve the direction of causality between the two, which suggests another IV strategy. However, in cross-country regressions, it is impossible to find a policy instrument. The reason is that countries are different, thus policies will vary from country to country, which makes it impossible to measure those policies and look into the policy aspects of financial development. While for the provinces within a country, it is appropriate and possible to do so. China's gradual financial deregulation experience offers a perfect platform to examine the policy aspects of financial development in explaining growth. Section 2.1.2.3 has divided financial deregulation policies and turned them into policy in-dexes, which have explicit policy variations across-province and across-time (see Figure 1.2). Before we instrument sectoral allocation of credit using the policy indexes to test if the policy component of sectoral allocation of credit is an important determinant of economic growth, the following section justifies the validity of those policy indexes as instruments. 2.1.4.2.2 The Validity of the Instruments It is important to verify the validity of using the constructed financial reform policies indicators discussed in section 2.1.2.3 as instruments. Table 1.5 shows that those policy indexes are highly correlated with one another. In addition, Table 1.9 shows that there are some degrees of corre-lation between sectoral allocation indicators and the various instruments. The only exception is that the Stock index is not significantly correlated with In(SOE). Nevertheless, these are simple correlations that do not take into account the effects of other control variables on sectoral credit allocation indicators. The results in Table 1.7 further show that these financial deregulation policy instruments have significant effects on sectoral credit allocation indicators and therefore suggest that these financial policies indicators are not weak instruments. Another necessary condition that the financial policies indicators must satisfy in order to be considered as valid instruments is that they must not be endogenous themselves. In other words, the logic for the government to conduct financial deregulation should not be due to the anticipation of future economic growth. To many, the logic behind gradual financial deregulation of China is like a black box. However, it is of vital importance to look deep into the gradual financial deregulation process to examine how the gradual financial deregulation is set-up, what determines the choice locations of the gradual financial deregulation, and how the gradual fi-nancial deregulation policies are implemented. In so doing, it helps us to see why the financial policies indicators may be exogenous to the growth process. There are previous literature on the gradual financial deregulation process of China (e.g. Naughton 1998 , Lardy 1998, Park and Sehrt 2001, and Shirk 2003). 21 In the case of China, the choice of the provinces in which financial reform policies are conducted is determined to a large extent by the provinces' geographical locations. Referring to Table 1.10, among the provinces with similar initial conditions (similar initial real G D P per worker), such as Guangdong, Fujian, Zhejiang and Jiangsu, the former two provinces have B A N K indexes that are twice as large as that of the latter two. It was noted that the Chinese central government implemented more financial reform policies in Guangdong and Fujian provinces, even though the growth performance of Jiangsu and Zhejiang provinces, in reality, surpassed that of the former for the 18 years of my sample period. The reason behind this is that both Guangdong and Fujian provinces are located close to Hong Kong, Taiwan and Macao, and the reform policies are strategically targeted at those provinces with superior geographical locations (see Figure 1.2). Geography is deemed as exogenous to the growth process in the literature (e.g. Gallup and Sachs 2001 and Acemoglu, Johnson, and Robinson 2001). 4 4 Political location is another determinant of choosing the provinces to conduct financial reform policies. Arguably, the choice of political locations is influenced by policy makers. The Chinese culture is that policy makers tend to give preferable treatment or policies to their hometown or places where they have worked and have good political relationship with. Shirk (2003) has studied the political logic and process of financial reform in China since 1979. Shirk (2003, p. 6) argues: "The pattern of economic reform in China should be seen not merely as the trial-and-error attempts of Chinese leaders to find a formula that work...The real challenge of economic reforms was the political one. Every Chinese economist I've ever met believes that the path of reform reflects a political logic." Shirk (2003, ch. 8, p. 129) points out: "Every policy bore the marks of being hammered out in Chinese communist institutions: the C C P delegated specific economic policy decisions to government bureaucracies; party leaders used reform policies to compete for power by appealing to officials in the selectorate; bureaucrats articulated the interests of their economic sectors and geographical regions; regions as well as sectors sat around the table; industry had a stronger voice than agriculture, and heavy industry had a stronger voice than light industry; and decisions were made by consensus. The pace, sequencing, content, and form of industrial reform policies from 1979 on reflect this institutional context." Shirk (2003, ch. 8, p. 129) continues: "As one Chinese commentator observed, the actual pattern of economic reform did not reflect economic theories so much as it did 'the conflict of various kinds of interests, that is the conflict, coordination, and balancing of interests between various trades and industries, between urban and rural areas, between localities, and between localities and the central authorities' (Wang Depei 1991, 39)." Therefore, it can be seen that, the conflict and coordination of different geographical regions plays an important role in the gradual financial deregulation process. Government officials' preferences, bargaining, and coordination play an influential part in determining which cities or provinces to conduct financial deregulation policies. These reveal that those financial deregu-4 4 I t is commonly agreed that geography can have causal effects on economic growth, though there is a debate over the channel of causality. 22 lation policies are worked out based on the considerations of many geographical and political factors. Whether financial deregulation causes economic growth remains to be examined next. It is worth admitting that the choice of financial reform policies as instruments may not be perfect because polices may not be totally exogenous. However, the existence of many factors other than future growth considerations that drive the gradual financial deregulation process makes the financial deregulation policies subject to the potential endogeneity problem to a much lesser extent. Given this, using financial reform policies allows us to gain much insights about the direction of causality between financial deregulation and economic growth. Future work is needed to examine whether financial deregulation policy instruments are weak instruments by more tests besides the Sargan tests used in this paper. 2.1.4.2.3 Results for Two-Stage Least Square (TSLS) Regressions Here, I employ the two groups of financial reform indicators that are constructed in section 2.1.2.3 as two sets of instruments. The first set of instruments includes Bank, Newbank and Stock, and the second set includes Bank, Newbank, Stock and Non-bank. Firstly, the endogeneity of the sectoral allocation of credit indicators is tested using the Hausman test (Greene 2000). The results are reported in Table 1.8. For example, for IV-1, using the Wu (1973) method, the F-test statistic on ln(AGR) and ln(IND) with 2 and 41 degrees of freedom is 5.52, and the associated p-value is 0.01 which leads me to reject the null hypothesis that ln(AGR) and ln(IND) are exogenous at the 5% level. For IV-1 ' , the Hausman test also provides evidence about the endogeneity of sectoral allocation indicators ln(AGR), ln(IND) and ln(SOE) at the 5% significance level. The first-stage results of the TSLS regressions are almost identical to those in Table 1.7. The only difference is that investment rate together with labor force growth are treated as regressors. The second stage results are reported in Table 1.8. IV-1 displays the first group of instruments results and IV-2 reports the results of the second group. The difference between I V -1 and IV-1' is that the latter also includes ln(SOE) in the regression and instruments it with the instruments. When compared with the OLS results, the coefficients on In(IND) are still negative but become significant at the 5% level, while those on ln(AGR) remain insignificant. This provides evidence that after controlling for the loans to agriculture, the reallocation of loans from state-owned industry to state-owned commerce improves economic growth. Since ln(SOE) controls for the credit reallocation effect between traditional sectors and non-traditional sectors. As mentioned before, the non-traditional sectors can be viewed as relatively more non-stated-owned ones. The IV results that the coefficients on ln(SOE) are insignificant show evidence that the growth-promoting effect mainly comes from credit reallocation within traditional sectors. The Sargan over-identification test shows that the null hypothesis that financial deregulation policies affect growth only through sectoral allocation of credit can not be rejected at the 10% 23 significance level. In the regression with ln(SOE), the speed of convergence is 3.9% per year with the first group of instruments and 2.5% per year with the second group. The magnitude of IV regressions estimates are economically significant for ln(IND). For example, for the regressions that utilize the first set of instruments and controlling for the loans to agriculture, the results for the two nearby non-coastal provinces of Heilongjiang and Ji l in suggest that if Heilongjiang could lower its share of bank loans to SOIs to a level that is equal to that of Jil in, then its annual average growth rate of real G D P per worker would increase by 0.67% for 1981-1986, 1.88% for 1987-1992, and 1.66% for 1993-1998. The TSLS results confirms that sectoral allocation of credit is an important determinant of growth for China for the period 1981-1998. The exogenous financial reform component of sectoral credit allocation causes economic growth. 4 5 Section 2.1.3.1 shows that financial reforms do not operate through physical capital investment rate to spur growth. Here, the Sargan test fails to reject the null hypothesis that financial deregulation policies affect growth only through sectoral allocation of credit. Taken together, these suggest that financial reforms have a significantly positive effect on real G D P per worker growth by switching bank loans from SOIs to agriculture and state-owned commerce. 2.1.4.2.4 Discussions on the Empirical Results The results show that the growth promoting effect of financial deregulation mainly comes from the credit reallocation among state-owned sectors, specifically from state-owned commerce to state-owned industries. It may be suspected that all the state-owned sectors are inefficient, therefore, allocation of credit to any sector is highly inefficient. Brandt and Zhu (2000) show that there was a sharp drop in the proportion of output produced in the state sector because the state sector experienced considerably slower productivity growth. It is then further suspected that state sectors have not contributed much to the economic growth of China. First of all, World Bank (1996) has estimated that the average annual growth rate of T F P in the state sector of China was 2.4 per cent per year from 1980 to 1988. Young (2003) found that the T F P growth of non-agriculture sectors of China is similar to that of four Asian dragons, which is still impressive. Therefore, state sectors of China did contribute much to its economic growth. Secondly, this paper argues that by looking into the institutional background of China, especially the economic policies before the reform period, one can understand why even within state-owned sectors, there may be potential huge gains from the credit reallocation within those sectors. Before the economic reform, the Chinese government believed in the gospel of rapid indus-trialization as the path to economic prosperity. As a result, the Chinese government favored state-owned industries instead of state-owned commerce and agriculture in credit allocation. Shirk (2003, p. 26) shows: "In China, iron and steel and machine building, the backbone heavy 4 5 This paper does not examine the direction of causality from growth to sectoral allocation of credit. For the literature that proposes bi-directional causality between financial development and economic growth, see Calder6n and Liu (2004). 24 industries, were given priority, consuming more than one-third of total investment in industrial capital construction (Statistical Yearbook 1990, 168). During the period of 1949-78 the value of heavy industrial output multiplied 90.6 times, while agriculture and light industry rose only 2.4 times and 19.8 times, respectively (Dong Furen 1982)." Shirk (2003, p. 26) continues: "Mean-while, energy and transportation infrastructure...tends to be neglected in communist economies. One study...found that...capitalist countries had invested a higher share of capital and labor into infrastructure (transport, communication, housing, health, education, and culture) and therefore had a greater stock of infrastructure than socialist countries had (Ehrlich and Szilagyi 1980). Chinese planners made the same short-sighted choices." Therefore, it can be seen that, even within the traditional sectors such as agriculture, state-owned industries and state-owned commerce, state-owned industries have priority in receiving credit than other sectors. Therefore, the credit to and thus the growth of other sectors are relatively more suppressed comparing to state-owned industries. For the Chinese case, without those existing distortions on credit allocation due to government intervention, there would be no role for the financial deregulation to have impact on growth and it would make little sense to even conduct financial deregulation. In China, given that the financial reform process is a gradual one, those government-imposed constraints for credit allocation are gradual removed. It comes as no surprise that more credit would be allocated to those sectors other than state-owned industries for a relatively more balanced sectoral growth and a more healthy economy, even though there are problems and challenges with the gradual financial deregulation experience of China (see Lardy 1998). 2.1.5 Results on Robustness Analyses The results presented so far are not subject to robustness checks. I consider the robustness of my results to growth accounting regressions, alternative measures of sectoral allocation of credit indicators, the choice of right hand side (RHS) control variables, alternative combinations of financial reform policies indicators, and the treatment of investment rate, export, FDI and School as endogenous variables. 2.1.5.1 Growth Accounting Combined with TSLS Regressions The TSLS regressions show that sectoral allocation of credit have significant effects on economic growth even after controlling for standard neoclassical growth factors. This implies that sectoral allocation of credit affects growth through total factor productivity (TFP); financial reforms significantly affect T F P , and thus growth, through affecting sectoral allocation of credit instead of factor accumulation. The growth accounting literature offers an avenue to evaluate this implication. Following the standard literature of growth accounting, growth can decomposed into two parts: factor accumulation growth and total factor productivity growth ( T F P G ) . And I use the 25 following standard growth accounting formula: ( 5 ) where A stands for T F P , Y for total real GDP, K for total real capital stock, L for labor force, and a for the exponent of capital in the production function, which is assumed to be | . The data on real G D P and labor force are already available. If we have the data on capital stock, then the T F P G , ^ , can be calculated. 2 .1 .5 .1 .1 B u i l d i n g t h e P r o v i n c i a l C a p i t a l S t o c k The quality of the investment data of China and thus the quality of the data on capital stock of China are questioned most in the literature. However, as Chow (1993) argues, the Chinese data are actually intrinsically consistent with each other. Chow (1993) and Chow and L i n (2003) have produced the total real capital stock of China for the period of 1952 to 1998. Based on their method and the data on the whole capital stock of China, I build the provincial capital stock according to the formula in Chow (1993): where Ki is the non-land fixed capital plus inventory of province i, Landi is the provincial capital of land, Iitt+i is the real gross investment (including inventory investment) of province i, and 6 is the depreciation rate. Following Chow (1993), the depreciation rate equals 0.04. • I n i t i a l p r o v i n c i a l r e a l C a p i t a l S t o c k . Chow and L i n (2003) have produced the whole capital stock of China from 1952 to 1998. M y data sample begins from 1981, so I use their real capital stock data in 1981 for China to get the initial capital stock of the provinces of China. Following Biggeri (1999), I assume the same capital-output ratio for all the provinces which equals the national one. Given this, using the total real provincial G D P as weight, I get the initial real capital stock of all the provinces of China. • P r o v i n c i a l v a l u e o f L a n d . As in Chow (1993), the capital of land is assumed to be fixed here. Given land=720 for whole China, I use the provincial total farm land equivalent ( T F L E ) as weight to calculate the provincial capital of land. The T F L E can be found in 4 , 'Human capital is not included as Pissarides argues (2000): "Human capital was not used in the calculation of T F P growth, the idea being that human capital contributes to T F P growth; i.e. that in a more general analysis human capital should be used to explain the path of T F P growth." 4 7 As in He (2006), the T F L E is built as follows. The China Academy of Science's natural resource database has the data for 1985 on provincial farmland, forestry land, pasture land and fishery waters, and the data on the value-added of the four types of land. Using the value-added as weight, the latter three types of land areas are discounted into farmland equivalent and added up to get the T F L E . Kiit+i = (1-6) (Kiit - Landi) + h,t+\ + Landi (6) He (2006). 47 26 • Provincial real grOSS investment. The provincial real gross investment is built following Chow and Lin (2003). First, I use G D P deflator to deflate the sum of nominal provincial consumption and provincial gross capital formation to get real provincial final expenditures. 4 8 Second, I use provincial general consumer price index 4 9 to turn nominal provincial consumption into real provincial consumption. Last, subtracting real provincial consumption from real provincial domestic final expenditures yields real provincial gross investment (including inventory investment). Wi th all the necessary data, equation (6) yields the provincial real capital stock. Then the T F P G can be calculated based on equation (5). 2.1.5.1.2 TSLS Regressions with T F P G Here, I regress the T F P G on indicators of sectoral allocation of credit, ln(AGR) and ln(IND). Following the literature, the results with the RHS variables in (2.2) excluding physical capital investment rate and labor force growth, the variable In (^y n + g + < ^ , and those with all RHS variables including In (^y n + g + g ^ in (2.2) are reported Table 1.11. According to Table 1.11, the TSLS regressions show that sectoral allocation of credit has significant effects on T F P G ; financial reforms significantly affect T F P through affecting sectoral allocation of credit. These confirms the TSLS regression results in section 2.1.4.2.3. The Hausman test statistics suggest the endogeneity of the sectoral allocation of credit indicators: ln(IND) and ln(AGR). The Sargan test cannot reject the null hypothesis that sectoral allocation of credit is the only channel through which financial reforms have significant effects on T F P G . The coefficients on human capital measure, School, are always significant at the 5% level, which is consistent with the arguments of Pissarides (2000): "Human capital was not used in the calculation of T F P growth, the idea being that human capital contributes to T F P growth; i.e. that in a more general analysis human capital should be used to explain the path of T F P growth." 2.1.5.2 Invariance of Other Control Variables It is desirable to check if the results are affected by the inclusion or exclusion of other RHS control variables. The results with different sets of RHS control variables are shown in Table 1.12. The results using only initial real G D P per worker as the control variable are reported in ( l ) 1 ; the results with initial real G D P per worker, human capital and fiscal expenditure as other / l sThe provincial nominal consumption for 1981-83 cannot be found. Thus, I use the total national nominal consumption to get the provincial nominal consumption using 1984 provincial consumption values as weights. •""The provincial CPI for 1984 cannot be found, thus, the national general retail price index is used for all the provinces. 27 control variables are found in ( l ) 2 ; finally, the results with initial real G D P per worker, human capital, fiscal expenditure, investment rate and labor force growth as other control variables are found in ( l ) 3 . According to Table 1.12, we can see that the TSLS results on sectoral allocation of credit are invariant with the set of other control variables. Comparing to the TSLS results in Table 1.8, the coefficients on ln(IND) have similar magnitudes and remain significant at the 5% level regardless of various control variables. The Sargan test statistic is 0.09 only for IV-2 in ( l ) 1 , which indicators that financial reforms may affect growth through other channels. However, as more control variables are added, all the Sargan over-identification test statistics cannot reject the null hypothesis that financial reforms work on growth only through sectoral allocation of credit at the 10% level. 2.1.5.3 Using Other Measures of Sectoral Allocation of Credit The results are also checked with other measures of sectoral allocation of credit. Given the data availability, the other possible measures of sectoral allocation of credit are listed in model [2] and [3] in Table 1.13. For model [2], the average of the ratio of short-term loans allocated to agriculture, state-owned industry and commerce in total loans (i.e. SOE in Table 1.2) for 1981-1986 for some provinces is above 90% (see Table 1.3). Thus, the indicators of short-term loans allocated to state-owned commerce are removed to avoid the potential multi-collinearity problem for 1981-1986. The second stage results of TSLS regression with the alternative measures of sectoral alloca-tion of credit are reported in Table 1.14, where it can be seen that the main empirical findings of this paper are robust with different measures of sectoral allocation of credit. The coefficients on the alternative measures of the fraction of loans allocated to SOIs (ln(INDl) or ln(IND2)) remain significant at the 5% level and have a similar magnitude as those of ln(IND). The coeffi-cients on the alternative measures of the fraction of loans allocated to agriculture (In(AGRl) or ln(AGR2)) are insignificantly negative and also have a similar magnitude as those of ln (AGR) . These confirm that sectoral allocation of credit is an important determinant of growth, while financial reforms have significant effects on growth through affecting sectoral allocation of credit. The p-value of the Sargan test being much larger than 10% shows that the null hypothesis that financial reforms work on growth only through sectoral allocation of credit cannot be rejected. 2.1.5.4 Using Alternative Combinations of Financial Reform Policies Here the results are checked with different combinations of financial reform policies. First, the results hold true with Bank, Newbank, Resi-Bank, Stock and their square being instruments. Those instruments do not have significant effects on the investment rate (the P-value on those instruments is 0.18), while they have significant effects on sectoral allocation of credit (the p-value on those instruments is 0.00 for ln(AGR) and 0.04 for ln(IND), which are below 5%). The 28 second stage results of TSLS regressions are found in IV-3 of Table 1.15, the results being similar to those in Table 8. The probability of the Sargan-test is 0.39, so the null hypothesis that all banking sector and capital market (bond and stock) reforming policies have a significant effect on growth only through sectoral allocation of credit cannot be rejected. Secondly, the results hold true with all indicators of financial deregulation policies (Bank, Newbank, Resi-Bank, Stock, Nonbank and their square) being instruments. Those instruments have an insignificant effect on investment rate (the P-value on those instruments is 0.26), while they have a significant effect on sectoral allocation of credit (the P-value on those instruments is 0.00 for ln(AGR) and 0.07 for ln(IND)). The second stage results of TSLS regressions are found in IV-4 of Table 1.15, which are similar to the results in Table 1.8. The probability of the Sargan-test is 0.29, so all financial reforming policies (banking sector, non-banking sector, capital market) have significant effects on growth only through sectoral allocation of credit. This result is surprising given that Non-bank and Resi-bank are not particularly effective in affecting sectoral allocation of credit (it is natural that not all policies are effective ones, and adding bad instruments tends to decrease the effectiveness of TSLS regressions). 2.1.5.5 Other Robustness Checks and Issues First, the results hold true by treating investment rate as endogenous,50 and by treating export, F D I and School as endogenous.51 Second, the TSLS results on sectoral allocation of credit are not affected by the measurement errors of the Chinese data. The measurement errors on dependent variable of annual average growth rate of real G D P per worker, discussed in Young (2003), can be absorbed in the disturbance of the regression and ignored (Greene 2000). There are no incentives for the provincial statistical bureau to report incorrect sectoral allocation of credit. Thus, the indicators of sectoral allocation of credit do not have any intentional measurement problem, which evidences that the sectoral allocation of credit is measured accurately. Other statistical measurement errors are dealt with by the IV approach. Finally, Weeks and Yao (2003) assumed different aggregate production functions for coastal provinces and interior ones and confirmed this with the Chinese data. In this paper, the coastal dummy is excluded, but the empirical results are invariant to this exclusion. 5 0 T h a t is, the coefficient on ln(y) is insignificant and the results on ln(IND) and ln (AGR) are similar to that in Table 8. The probability of the Sargan test is above 10%. Even i fonly instrumenting l n ( y ) (without sectoral allocation indicators in the regression), the coefficient on l n ^ ) is still insignificant. Thus, financial deregulation does not significantly work on growth through increasing physical capital investment rate. s l T h i s means instrumenting export, F D I and School with financial reform indicators as well, and the coefficients on export, F D I and School are very insignificant. 29 2.1.6 Conclusions and Lessons for Developing Countries This paper addresses two important questions concerning financial development and economic growth. The first question investigates if financial reform contributes to economic growth. The second considers the channels via which financial reform affects growth. The empirical test of my hypotheses that is based on a set of Chinese panel data shows that financial reforms have significant effects on S A C but have no impact on investment rate. Unlike popular belief that investment rate is the channel through which financial reform affects growth, I find that S A C is an important determinant of growth for China for the period 1981-1998. This evidence is further supported by the two-stage least squares regression results which show that the exogenous financial reform policy component of S A C causes economic growth. In particular, I demonstrate that financial reforms have significantly positive effects on real G D P per worker growth rate by switching bank loans out of SOIs. And the growth-promoting effect mainly comes from credit reallocation within traditional sectors. I also provide evidence that apart from the sectoral credit allocation channel, there is no evidence that other channels through which financial reform influences growth in China. The results hold true in all robustness checks. The results of current study have important policy implications for developing countries. First and foremost, the empirical findings suggest that it may be productive for developing economies to focus their reform effort more on S A C than on saving and investment. Secondly, the results of this study make one ponder over a fundamental question in development economics: should developing countries favor the allocation of financial credit and resources to certain sectors over others (e.g. favoring industry over agriculture for rapid industrialization as in the B A W I plan 5 2 ) . The Chinese experience shows that favoring industry too much may yield inefficient S A C , and the gradual approach to introduce market forces is a good choice to eliminate existing inefficient S A C , even within traditional sectors that are generally state-owned ones. The results show that the growth promoting effect of financial deregulation mainly comes from the credit reallocation among state-owned sectors, specifically from state-owned commerce to state-owned industries. The institutional background of China helps to eliminate the suspicion that all the state-owned sectors are inefficient, therefore, allocation of credit to any sectors is highly inefficient. Before the economic reform, the Chinese government believed in rapid industrialization as the path to economic prosperity and favored state-owned industries instead of state-owned commerce and agriculture in credit allocation, which is common in many socialist countries as discussed in Shirk (2003, ch. 1). Therefore, the credit to and thus the growth of other sectors are relatively more suppressed comparing to state-owned industries. Given the existing distortions on credit allocation due to government intervention, it comes as no surprise that gradual financial deregulation which aims at introducing market forces in credit allocation 5 2 I n the 1930s, Mississippi's average annual per capita income was one-third of its national average, causing Governor White to propose a "Balance Agriculture with Industry (BAWI) plan" to attract more manufacturing. In the postwar era, the B A W I plan served as a model for those backward states with a large agricultural sector (Teaford 2002). 30 would generate that more credit would be allocated to those sectors other than state-owned industries for a relatively more balanced sectoral growth and a more healthy economy. Although there are problems and challenges with the gradual financial deregulation experience of China (see Naughton 1998 and Lardy 1998), the financial deregulation paid off for China. Lastly, though financial deregulation of China decreases the share of credit to SOIs, it is a challenging task given the soft budget-constraint (SBC) problem of SOIs (see Kornai Maskin and Roland 2003). The political logic of the financial reform of China studied by Shirk (2003) gave us some idea why the financial deregulation was able to promote growth through S A C . Furthermore, the Chinese government achieved the credit reallocation out of SOIs partly by credit expansion through the banking system to get around the SBC problem. The mechanism of such a policy requires future research. 31 2.2 Tables Table 1.1: T h e Balance Sheet of China ' s P rov inc ia l Na t iona l Banks Use of Funds Data Availability A l l Loans E = £ i + £ 2 + £ 3 Yes Short-term Loans E i No (state-owned) Industrial Enterprises i Yes (state-owned) Commercial Enterprise c Yes Agricul ture 0 3 A Yes Construction Enterprise No Urban Collective Enterprises No Individual Enterprises No Foreign-related Enterprises No Other No Mid-term and long-term Loans Yes Other Loans £ 3 No Others Table 1.2: Indicators of Sectoral Al loca t ion of Credi t Model Indicator Calculating Formula [1] A G R IND C O M S O E Short-term Loans to Agriculture A+C+I Short-term Loans to Industry, Commerce and Agriculture / Short-term Loans to Industry A+C+I Short-term Loans to Industry, Commerce and Agriculture C Short-term Loans to Commerce A+C+I Short-term Loans to Industry, Commerce and Agriculture A+I+C _ Short-term Loans to Industry, Commerce and Agriculture — Total Loans 5 3 After 1978, the family-contract responsibility system (FRS) was set up, which allows households to control their production and own the output subject to government taxation. For some provinces, state-owned agriculture still exists. 32 Table 1.3: Average Annual Growth Rate and Sectoral Allocation of Loans Province Annual Growth A G R S O E Province Annual Growth A G R S O E Beijing 0 6.0 2.2 78.1 Shandong 0 7.2 12.2 91.6 Beijing 0 5.0 4.1 66.4 Shandong 0 5.7 16.5 75.3 Beijing 0 9.5 4.0 57.3 Shandong 0 9.5 11.5 55.6 Tianj in c 5.6 0.8 86.9 Henan 5.9 5.8 91.5 Tianj in 0 4.2 2.2 76.4 Henan 3.8 5.8 80.4 Tianj in c 12.0 3.4 59.1 Henan 7.8 5.6 64.3 Hebei c 6.2 6.4 91.4 Hubei 7.5 7.8 59.5 Hebei c 5.6 5.3 78.9 Hubei 4.6 5.8 75.9 Hebei c 9.5 5.7 74.5 Hubei 10.2 3.8 64.0 Shanxi 7.7 10.8 88.5 Hunan 5.4 6.4 89.9 Shanxi 3.5 6.4 73.2 Hunan 3.4 5.6 78.4 Shanxi 7.8 5.6 58.9 Hunan 7.6 6.1 64.5 Inner Mongolia 7.5 12.9 84.6 Guangdong 0 7.7 8.8 82.9 Inner Mongolia 4.6 8.8 74.8 Guangdong 0 8.9 5.6 75.4 Inner Mongolia 8.1 4.4 62.5 Guangdong 0 9.0 6.1 61.8 Liaoning c 6.0 2.8 87.0 Guangxi 0 3.6 14.5 86.6 Liaoning c 4.3 3.8 76.8 Guangxi 0 5.2 17.6 70.5 Liaoning 0 8.2 3.9 67.7 Guangxi 0 6.9 14.0 50.9 J i l in 4.2 8.4 90.2 Guizhou 6.5 11.7 90.1 J i l in 2.6 7.4 82.0 Guizhou 2.4 12.7 73.5 J i l in 10.3 5.4 76.0 Guizhou 5.2 8.9 60.8 Heilongjiang 2.9 12.9 91.1 Yunnan 6.1 12.1 ' 88.2 Heilongjiang 3.7 9.4 83.5 Yunnan 5.1 9.9 75.5 Heilongjiang 4.9 4.6 72.8 Yunnan 6.8 4.4 69.5 Shanghai 0 6.3 0.2 84.6 Shaanxi 6.6 9.9 89.6 Shanghai 0 6.6 1.5 73.3 Shaanxi 4.3 6.9 71.6 Shanghai 0 11.7 1.1 57.5 Shaanxi 6.3 4.2 62.4 Jiangsu 0 7.9 6.8 90.1 Gansu 5.1 12.7 88.6 Jiangsu 0 7.9 3.2 73.8 Gansu 4.7 8.7 71.3 Jiangsu 0 11.0 3.7 63.6 Gansu 6.5 5.6 56.1 Zhejiang 0 8.2 4.6 84.9 Qinghai 6.5 9.2 84.4 Zhejiang 0 6.8 3.8 73.3 Qinghai 2.2 7.1 56.4 Zhejiang 0 11.0 3.2 63.0 Qinghai 5.8 3.3 48.6 Anhui 6.9 10.5 89.6 Ningxia 6.7 10.6 85.4 Anhui 2.0 7.0 76.8 Ningxia 3.2 7.0 60.9 . Anhui 9.6 4.5 66.8 Ningxia 5.0 4.1 56.7 Fujian 0 6.0 13.2 77.9 Xinjiang 8.7 11.9 86.0 Fujian 0 6.9 9.4 59.4 Xinjiang 7.2 13.3 71.1 Fujian 0 10.7 9.5 48.5 Xinjiang 6.3 7.4 65.9 Jiangxi 6.0 8.5 70.1 Jiangxi 5.2 9.2 75.6 Jiangxi 6.5 7.5 67.1 Note: all numbers are in percentage; c stands for coastal provinces 33 Table 1.4: Financial Reforming Policy Indicators Financial Deregulation Indicators Description Banking Sector Bank Banking sector general reforms and policies; Banking deregulation policies that might affect sectoral allocation of credit; Newbank The set-up of specific new banks; Resi-bank The remaining banking sector policies; Non-bank Nonbank Non-bank deposit-taking institutions; Insurance market; Capital Market Stock Policies associated with Equity, Stocks and Bonds. Table 1.5: Correlation among the Policy Indexes Bank Newbank Resi-bank Nonbank Stock Bank 1.0 Newbank 0.33*** 1.0 Resi-bank 0.48*** 0.90*** 1.0 Nonbank 0.64*** 0.85*** 0.92*** 1.0 Stock 0.55*** 0.62*** 0.74*** 1.0 *** indicates significant at the 0.01. level 34 Table 1.6: Does Financial Deregulation Affect Investment Rate? Indep. Var. Bank Bank2 Newbank Newbank2 Stock Stock2 Nonbank Nonbank2 Dependent Variable: l n (^ ) -0.03 (0.04) 0.01 (0.01) 0.54 (0.36) -0 .20 (0.26) 0.01 (0.07) -0.002 (0.007) -0.01 (0.05) 0.001 (0.02). 0.59 (0.42) -0.62 (0.48) -0.06 (0.10) -0.001 (0.01) -0.10 (0.11) 0.04 (0.04) Time F E Province F E F-Test (p-value) F-Test on indicators: P-value on indicators: R2 Observations: Yes Yes 22.10 (0.00) 1.88 0.11 0.96 81 Yes Yes 20.61 (0.00) 1.52 0.18 0.96 81 Other Indep. Variables: l n ( £ ) t - l , ln(School), ln(Fiscal), ln(n+g+<5), ln(Export), ln(FDI). ***Significant at 0.01 level, ** at 0.05 level, * at 0.10 level (Standard errors in parentheses) 35 Table 1.7: Does Financial Deregulation Affect Sectoral Allocation of credit? Dependent Variable as: In(AGR) \n(IND) Indep. Var. 1 2 1' 2' Bank 0.02 -0 .07 -0.14*** -0.13** (0.15) (0.16) (0.05) (0.06) Bank2 0.04 0.03 0.02 0.01 (0.05) (0.05) (0.02) (0.02) 0.14 -1.30 -0.26 -0 .33 Newbank (1.25) (1.36) (0.43) (0.49) Newbank2 -0.72 -1 .93 0.005 -0.71 (0.31) (0.56) (0.90) (1.56) Stock 0.98*** 0.60* -0.02 -0.15 (0.25) (0.31) (0.09) (0.11) -0.08*** -0.06** 0.002 0.01 Stock2 (0.02) (0.02) (0.01) (0.01) 0.02 -0.13 Nonbank (0.34) (0.12) 0.13 0.07 Nonbank2 (0.12) (0.04) Time F E Province F E F-Test (P-value) F-Test on indicators: P-value on indicators: R2 Observations: Yes Yes Yes Yes 18.42 (0.00) 20.30 (0.00) 8.10 7.96 0.00 0.00 0.95 0.96 81 81 Yes Yes 18.24 (0.00) 2.63 0.03 0.95 81 Yes Yes 17.94 (0.00) 2.42 0.03 0.95 81 Other Indep. Variables: l n ( £ ) t _ i , ln(School), ln(Fiscal), ln(n+g+<5), ln(Export), ln(FDI). ***Significant at 0.01 level, ** at 0.05 level, * at 0.10 level (Standard errors in parentheses) 36 Table 1.8: Regression between Growth and Sectoral Allocation of Credit Dep. Var.: Average Annual Growth Rate of Real G D P per worker Indep. Vari . OLS I V - 1 I V - 2 OLS I V - 1 ' I V - 2 ' ln(IND) -2 .64 -15.17** -10.92** -2.71 -15.18** -13.71* (2.33) (6.45) (4.60) (2.73) (6.88) (5.99) ln (AGR) 0.20 -1 .67 -0 .67 0.19 -1 .67 -1 .17 (0.65) (1.29) • (0.95) (0.69) (1.36) (1.19) -0.15 -0 .05 -9.02 ln(SOE) (2.90) (14.76) (11.35) -4.27** -3.46 -3.51 -4.25** -3.46 -2.34 (2.02) (2.73) (2.37) (2.08) (3.30) (2.91) 4.53** 8.89*** T 03*** 4.51** 8.87* 5.58* In(School) (1.98) (3.19) (2.50) (2.03) (4.62) (3.22) IT I \ 3.07* 1.86 1.92 3.08* 1.86 2.41 (1.58) (2.34) (2.02) (1.60) (2.54) (2.22) ln(Fiscal) 1.14 0.90 0.77 1.15 0.90 1.83 (1.74) (2.32) (2.02) (1.79) (2.91) (2.51) -0.16 -0.41 -0.20 -0 .17 -0.42 -0.86 In(Export) (0.62) (0.86) (0.74) (0.67) (1.33) (1.14) 0.03 -0.16 -0 .07 0.02 -0.16 -0.16 \n(FDI) (0.24) (0.33) (0.28) (0.25) (0.35) (0.32) Time F E Yes Yes Yes Yes Yes Yes Province F E Yes Yes Yes Yes Yes Yes Hausman Test F = 5.52 F = 4.08 F = 3.75 F = 3.2 Prob. of F = 0.01 0.02 0.02 0.03 Sargan Test (P-value) — 0.66 0.14 0.66 0.27 R2 0.83 0.71 0.78 0.83 0.71 0.76 Observations: 81 81 81 81 81 81 r V - l ' s Instruments: Bank, Newbank, Stock and their square I V - 2 ' s Instruments: Bank, Newbank, Stock, Nonbank and their square ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 37 Table 1.9: Correlation between Instruments and Sectoral Allocation of Credit Bank Newbank Resi-bank Nonbank Stock ln(SOE) -0.38*** -0.24** -0.22** -0.31*** -0.16 l n ( A G R ) ln(IND) -0.38*** -0.32*** -0.46*** -0.47*** -0.45*** 0.38*** 0.37*** 0.42*** 0.48*** 0.30*** ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level Table 1.10: Determinants of Financial Reform Policies Coastal Average of the Average Annual Growth Real G D P per worker Province Index B A N K Rate of Real G D P per worker in 1981 Shanghai 2.42 9.5 4470.6 Beijing 0.41 8.3 2582.5 Tianjin 1.45 8.2 2544.8 Guangdong 1.69 10.3 1057.4 Fujian 1.74 9.4 956.1 Zhejiang 0.99 10.2 931.1 Jiangsu 0.76 10.5 1114.3 Liaoning 0.82 7.6 1716.5 Hebei 0.13 8.3 895.0 Shandong 0.86 9.0 893.7 Guangxi 0.02 6.3 581.6 Note: B A N K Index is 18 year average; growth rates are in percentage 38 Table 1.11: Regression between T F P G and Sectoral Allocation of Credit Dep. Var.: Average Annual Growth Rate of Total Factor Productivity Indep. Vari . OLS I V - 1 I V - 2 I V - 1 I V - 2 -2.54 -12.65** -9.37** -13.78** -10.23** ln(IND) (2.11) (5.27) (3.87) (5.90) (4.31) 0.57 -0 .94 -0.14 -0.78 0.05 ln(AGR) (0.58) (1.11) (0.82) (1.18) (0.89) -1.39 -1.34 (2.14) (1.89) -5.48*** -5.13** -5.14** -4.56* -4.59** (1.79) (2.23) (2.01) (2.50) (2.22) 3.33* 6.82** 5.36** 6.91** 5.37** l n (School) (1.84) (2.81) (2.27) (2.92) (2.34) In(Fiscal) 0.21 -0 .35 -0.45 -0.20 -0 .30 (1.59) (2.05) (1.84) (2.12) (1.89) 0.04 -0 .10 -0.08 -0 .07 0.11 In(Export) (0.58) (0.75) (0.67) (0.79) (0.69) -0.01 -0 .17 -0.10 -0.15 -0.08 \n(FDI) (0.23) (0.29) (0.26) (0.30) (0.26) Time F E Yes Yes Yes Yes Yes Province F E Yes Yes Yes Yes Yes Hausman Test Prob. of F = 0.02 0.04 0.02 0.04 Sargan Test: (P-value) 0.89 0.42 0.76 0.29 B? 0.78 0.66 0.73 0.64 0.72 Observations: 81 81 81 81 81 I V — l ' s Instruments: Bank, Newbank, Stock and their square I V - 2 ' s Instruments: Bank, Newbank, Stock, Nonbank and their square ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 39 Table 1.12: Invariance of Other Cont ro l Variables in T S L S Regression Second-Stage: Dep. Var.: Aver. Annual Growth Rate of Real G D P per worker (D 1 ( I ) 2 ( I ) 3 ln (AGR) ln(IND) ln (AGR) ln(IND) ln(AGR) ln(IND) I V - 1 0.17 (0.93) -12.03** (5.28) -1.32 (1.22) -16.50** (6.21) -1.51 (1.22) -15.72** (6.63) Sargan Test: (P-value) R2 0.11 0.69 0.80 0.69 0.65 0.70 I V - 2 0.65 (0.74) -9.58** (4.08) -0 .27 (0.85) -11.61*** (4.22) -0.50 (0.88) -10.97** (4.48) Sargan Test: (P-value) R2 0.09 0.74 0.23 0.77 0.14 0.78 Time F E Prov. F E Yes Yes Yes Yes Yes Yes Observations: 81 81 81 IV— l ' s Instruments: Bank, Newbank, Stock and their square IV—2's Instruments: Bank, Newbank, Stock, Nonbank and their square ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) Table 1.13: Al ternat ive Indicators of Sectoral Al loca t ion of Credi t Model Indicator Calculating Formula A Short term Loans to Agriculture Total Loans / Short term Loans to Industry ^ — Total Loans A Short term Loans to Agriculture YJ—A Total Loans - Short, term Loans to Agriculture / Short term Loans to Industry ]T]—J Total Loans - Short term Loans to Industry [2] A G R 1 IND1 [3] A G R 2 IND2 40 Table 1.14: Regression with Alternative Indicators of Sectoral Allocation of Credit Dep. Var.: Aver. Annual Growth Rate of Real G D P per worker Time F E Yes Yes Yes Province F E Yes Yes Yes Observations: 81 81 81 Model OLS I V - 1 I V - 2 -1 .54 -13.50** -13.72** [2] ln( INDl) (2.31) (6.62) (6.46) 0.35 -1.56 -1.33 l n ( A G R l ) (0.66) (1.33) (1.25) Sargan Test (P-value) 0.28 0.32 R2 0.83 0.71 0.71 -1 .17 -9.62** -9.50** [3] ln(IND2) (1.66) (4.70) (4.54) 0.36 -1.56 -1.26 ln(AGR2) (0.64) (1.30) (1.21) Sargan Test (P-value) 0.27 0.25 R2 0.83 0.72 0.72 I V — l ' s Instruments: Bank, Newbank, Stock and their square IV—2's Instruments: Bank, Newbank, Stock, Nonbank and their square ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 41 Table 1.15: Regression with Alternative Indicators of Financial Reform Policies Dep. Var.: Aver. Annual Growth Rate of Real G D P per worker O L S I V - 3 I V - 4 -2.64 -11.85** -11.27** ln(IND) (2.33) (4.77) (4.47) 0.20 -0 .99 -0 .72 ln (AGR) (0.65) (0.99) (0.94) Time F E Yes Yes Yes Province F E Yes Yes Yes Sargan Test (P-value) 0.39 0.29 R2 0.83 0.77 0.78 Observations: 81 81 81 IV—3's Instruments: Bank, Newbank, Resi-Bank, Stock and their square rV - 4 ' s Instruments: Bank, Newbank, Stock, Nonbank, Resi-Bank and their square ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 42 2.3 Figures 70% 1 1 1 1 1 1 1 1 1 ' 1981 1983 1985 1987 1989 1991 1993 1995 1997 F i g . 1 . 1 : P r o p o r t i o n o f N a t i o n a l Bank Loans t o Inner Mongolia the T o t a l Loans of the F i n a n c i a l System T i a n i i n Fig. 1.2. Provincial Distribution of Financial Reform Policy Index BANK (18 Year Average) 43 2.4 References [I] Acemoglu, Daron, Simon Johnson, and James A . Robinson. 2001. "The Colonial Origins of Comparative Development: an Empirical Investigation." American Economic Review 91 (December): 1369-401. [2] Aghion, Philippe, and Peter Howitt. 1998. Endogenous Growth Theory, Cambridge, Mass.: M I T Press. [3] Philippe Aghion, Peter Howitt and David Mayer-Foulkes. 2005. "The Effect of Financial Development on Convergence: Theory and Evidence." Quarterly Journal of Economics 120 (February): 173-222. [4] Barro, Robert J . , and Jong-Wha Lee. 1993. "International Comparison of Educational At-tainment." Journal of Monetary Economics 32 (December): 361-94. [5] Barro, Robert J . , and Xavier Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hi l l . [6] Benhabib, J . and M . Spiegel. 2000. "The Role of Financial Development in Growth and Investment." Journal of Economic Growth 5 (): 341-60. [7] Biggeri, Mario. 1999. "Determinants of Chinese Provincial Economic Growth in the Last Decade." Department of Economics, University of Forence. [8] Brandt, Loren, and Ziaodong Zhu. 2000. "Redistribution in a Decentralized Economy: Growth and Inflation in China Under Reform." Journal of Political Economy, Vol . 108 (April): 422-39. [9] Calderon, C , and L . L iu . 2004. "The Direction of Causality between Financial Development and Economic Growth." Working Paper, http://www.bcentral.cl/ Estu-dios/DTBC/doctrab.htm. [10] Chow, Gregory C. 1993. "Capital Formation and Economic Growth in China." Quarterly Journal of Economics 108 (August), 809-42. [II] Chow, Gregory C, and Anloh Lin . "Accounting for Economic Growth in Taiwan and Main-land China: A Comparative Analysis." EWP-GE/0306003. [12] Coviello, Decio. 2005. "Legal Origins, Financial Development and Growth: Revisiting the Evidence in the Case of Weak Identification." Universitat Pompeu Fabra Working Paper. 44 [13] Dehejia, R. and A . Lleras-Muney. 2003. "Why Does Financial Development Matter? The United States from 1900 to 1940." NBER Working Paper No. 9551. [14] Demirguc-Kunt, Asl i and Ross Levine. 2001. Financial Structure and Economic Growth: A cross-country comparison of Banks, Markets, and Development. Cambridge, Mass.: M I T Press. [15] Dewatripoint, M . , and Gerard Roland. 1992a. "The Virtues of Gradualism and Legitimacy in the Transition to a Market Economy." Economic Journal 102: 291-300. [16] . 1992b. "Economic Reform and Dynamic Political Constraints. " Review of Eco-nomic Studies 59 (October): 703-730. [17] Eschenbach, F . 2004. "Finance and Growth: A Survey of the Theoretical and Empirical Literature." Working Paper. [18] Gallup, John Luke and Jeffrey D. Sachs. 2001. "The Economic Burden of Malaria." Amer-ican Journal of Tropical Medicine and Hygiene 64(12): 85-96. [19] Greene, Wil l iam Ff. 2000. Econometric Analysis. New Jersey: Prentice Hall . [20] He, Qichun. 2006. "Land-curse in Industrialization: Evidence from China." University of British Columbia Working Paper. [21] Jayaratne, Jith, and Philip E . Strahan. 1996. "The Finance-Growth Nexus: Evidence from Bank Branch Deregulation." Quarterly Journal of Economics 111 (August): 639-70. [22] Jbil i , Abdelali, Klaus Enders, and Volker Treichel. 1997. "Financial Sector Reforms in Algeria, Morocco, and Tunisia: A Preliminary Assessment." IMF Working Paper No. 81. [23] Klenow, Peter J . , and Andres Rodriguez-Clare. 1997. "The Neoclassical Revival in Growth Economies: Has it Gone too Far?" NBER Macroeconomics Annual 12: 73-103. Cambridge, Mass.: M I T Press. [24] King, Robert G. , and Ross Levine. 1993. "Finance and Growth: Schumpeter Might Be Right." Quarterly Journal of Economics 108 (August): 717-37. [25] Kornai, Jdnos, Eric Maskin, and Gerard Roland. 2003. "Understanding the Soft Budget Constraint." Journal of Economic Literature 41 (December): 1095-136. [26] Kraay, Aart. 2000. "Household Saving in China." The World Bank Economic Review 14 (3): 545-70. [27] Kuijs, Louis. 2005. "Investment and Savings in China." World Bank Policy Research Work-ing Paper No. 3633. 45 [28] Lardy, Nicolas R. 1998. China's Unfinished Economic Revolution. Brookings Institution Press. [29] Levine, Ross. 2005. "Finance and Growth: Theory and Evidence." Handbook of Economic Growth. The Netherlands: Elsevier Science. [30] Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. "Financial Intermediation and Growth: Causality and Causes." Journal of Monetary Economics 46 (August): 31-77. [31] Levine Ross, and Sara Zervos. 1998. "Stock Markets, Banks, and Economic Growth." Amer-ican Economic Review 88 (June): 537-58. [32] Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Serven. 2000. "What Drives Saving across the World?" Review of Economics and Statistics L X X X I I (May): 165-81. [33] Ma , Xiaoying, and Ortolano, Leonard. 2000. Environmental Regulation in China: Institu-tions, Enforcement, and Compliance. Lanham: Rowman and Littlefield Publishers. [34] Mankiw, Gregory N . , David Romer, and David N . Weil. 1992. " A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107 (May): 407-37. [35] McKinnon, Ronald I. 1973. Money and Capital in Economic Development. Washington, D.C. : Brookings. [36] Modigliani, Franco and Shi Larry Cao. 2004. "The Chinese Saving Puzzle and the Life-Cycle Hypothesis." Journal of Economic Literature 42 (March): 145-70. [37] Naughton, Barry. 1998. "China's Financial Reform: Achievements and Challenges." B R I E Working Paper 112. [38] Papyrakis, E . and R. Gerlagh. 2004. "The Resource Curse Hypothesis and Its Transmission Channels." Journal of Comparative Economics, 32, 181-193. [39] Park, Albert, and Kaja Sehrt. 2001. "Tests of Financial Intermediation and Banking Reform in China." Journal of Comparative Economics 29 (4): 608-44. [40] Pissarides, Christopher. 2000. "Labor markets and economic growth in the M E N A region." LSE Working Paper. [41] Rajan, Raghuram G. , and Luigi Zingales. 1998. "Financial Dependence and Growth." Amer-ican Economic Review 88 (June): 559-86. [42] Shaw, Edward S. 1973. Financial Deepening in Economic Development. New York: Oxford University Press. 46 [43] Shirk, Susan L . 2003. The Political Logic of Economic Reform in China. Berkeley: Univer-sity of California Press. [44] Solow, Robert. 1956. " A Contribution to the Theory of Economic Growth." Quarterly Journal of Economics 70 (February): 65-94. [45] Teaford, Jon C. 2002. The Rise of the States: Evolution of American State Government. Baltimore: Johns Hopkins University Press. [46] Weeks, Melvyn and James Yudong Yao. 2003. "Provincial Conditional Income Convergence in China, 1953-1997: A Panel Data Approach." Econometric Reviews 22 (February): 59-77. [47] Williamson, Stephen D. 1987. "Financial Intermediation, Business Failures, and Real Busi-ness Cycles." Journal of Political Economy 95 (December): 1196-216. [48] World Bank. 1996. World Development Report 1996: from Plan to Market. New York: Oxford University Press. [49] Young, Alwyn. 1995. "The Tyranny of Numbers: Confronting the Statistical Realities of the East Asia Growth Experience." Quarterly Journal of Economics 110 (August): 641-80. [50] Young, Alwyn. 2003. "Gold into Base Metals: Productivity Growth in the People's Republic of China during the Reform Period." Journal of Political Economy 111 (December): 1220-61. [51] A Collection of New China's 50-year Statistical Data [Xinzhongguo 50 nian Tongji Ziliao Huibian]. 1999. China Statistics Press. [52] Big Economic Events Since China's Reform and Opening-up (1978-1998) [Zhongguo Gaige Kaifang Yilai Dashiji (1978-1998)]. Institute of Economic Research, the China Academy of Social Sciences. 2000. The Economic Science Publishing. [53] China Statistical Yearbook [Zhongguo Tongji Nianjian]. Beijing: China Statistical Press 1978-1999 (annual). [54] Natural Resources Database [China Academy of Social Sciences]. Beijing, China. [55] Statistical Yearbook on China's Cities [Zhongguo Chengshi Tongji Nianjian], Beijing: China Construction and Consulting and Service Center on Chinese Statistical Information 1986-1989 (annual). [56] Statistical Yearbook on China's Cities [Zhongguo Chengshi Tongji Nianjian]. Beijing: China Statistics 1990-1999 (annual). 47 3 ENTREPRENEUR'S INVENTIVE INCENTIVE A N D GROWTH: A N INVERTED-U RELATIONSHIP 5 4 3.1 Entrepreneur's Inventive Incentive and Growth: An Inverted-U Relationship 3.1.1 Introduction Models that formally show accumulation of knowledge as the de terminant 5 5 of to ta l factor pro-duct iv i ty ( T F P ) growth were not developed un t i l the early 1990s. 5 6 Those models that involve accumulat ion of knowledge are widely referred to as new growth models ( N G M s ) . M a n y N G M s (e.g. Romer 1990 and A g h i o n and Howi t t 1992) regard the role of inventions by entrepreneurs as crucial in dr iv ing T F P growth, hence the natural impl ica t ion would be that higher entrepre-neur's inventive incentive would generate higher balanced growth rate. T h e n pro-entrepreneur policies should be good for economic growth because they give entrepreneurs more incentive to innovate. T h i s paper explores the relationship between entrepreneur's inventive incentive and balanced growth rate by incorporat ing entrepreneurs' effort and financial imperfection into N G M s . In so doing, this paper also expounds on Lewis (1955)' th i rd determinant of long-run growth — the effort of agen t s . 5 7 ' 5 8 B y incorporat ing entrepreneurs' effort and credit market imperfection into N G M s , this pa-per finds: firstly, balanced growth rate is an inverted-U function of entrepreneurs' inventive incentive (measured as entrepreneur's bargaining share of monopolist ic profit from innovation, referred to as entrepreneur's bargaining share henceforth), which has not been found by previous works; second, entrepreneur's bargaining share, through affecting entrepreneur's effort is a de-terminant of long-run growth, which offers insights into the aforementioned Lewis ' view. Three features besides monopolist ic competi t ion are necessary and sufficient to ensure the inverted-U result. F i rs t ly , there are two representative agents (entrepreneur and household) rather than 5 4 This chapter has benefited from the comments from three anonymous referees of the Journal of Political Economy. 5 5 Strictly speaking, the determinant should be viewed as a channel because it is endogenous to the growth process. This paper uses determinant just to be consistent with previous literature. 5 8 There are other very early attempts to generate endogenous long-run growth (e.g. Shell 1967, Sheshinski 1967). But there are limitations to those models (see Grossman and Helpman 1991). "According to Lewis (1955), there are three ultimate determinants of long-run growth: factor accu-mulation (modeled by Solow 1956), accumulation of knowledge, and the effort of agents. One reason for the lack of theoretical work that expounds on Lewis' view of the effort of agents as a determinant of economic growth is that the meaning of the effort of agents by Lewis is open to interpretation. To minimize confusion, this paper focuses on the effort of entrepreneurs. 5 8 There are many works studying entrepreneur's effort. For example, Aghion and Tirole (1994) in-troduced entrepreneur's effort when studying the organization of R & D in an incomplete contract frame-work. But the relationship between entrepreneur's inventive incentive (and thus entrepreneur's effort) and growth has not been explored by previous literature. 48 one (household) as in N G M s . Explicitly, there is 'separation' between R & D financier (house-holds and entrepreneurs) and its carrier (entrepreneurs). Secondly, entrepreneurs' effort is their private information (hidden action). Thirdly, entrepreneurs' type is hidden information. Given these, an increase in the bargaining share of entrepreneurs elicits more effort from them, gen-erating a "bigger cake", but it also decreases the share that goes to R & D financiers. For R & D financiers (households and entrepreneurs), the bigger cake effect dominates at the beginning; but beyond a point, the "decreasing share effect" dominates, generating the inverted-U. Standard N G M s assume that there is only one representative agent. Furthermore, in N G M s , there is a perfect credit market (e.g. Aghion and Howitt 1992). Given these, the role of credit contract is limited, and there is no rich dynamics if we introduce entrepreneurs' effort into N G M s . Given the perfect credit market, the one representative agent who also works as entre-preneurs definitely has the highest inventive incentive and contributes the highest effort level. The unchanging inventive incentive and thus the fixed effort level of entrepreneurs yields only one balanced growth rate as in standard N G M s . Introducing 'separation' alone into N G M s yields only one effort level of entrepreneurs. Given the separation between R & D financier (households) and R & D carrier (entrepreneurs), there is "smaller household's share effect". However, when there is no asymmetric information, entrepre-neurs' effort is fixed. That is, entrepreneur's effort does not respond to entrepreneur's inventive incentive (i.e. entrepreneur's bargaining share), hence there is no "bigger cake effect" and growth is decreasing in entrepreneurs' bargaining share. How do we go about introducing only asymmetric information into NGMs? In the real world, credit markets are full of imperfections. Paulson, Townsend and Karaivanov (2006) show empirical evidence that there are credit market imperfections with Thailand data. If entrepreneurs are credit-constrained, and must borrow from financial intermediaries to finance their Research and Development ( R & D ) , 5 9 asymmetric information between entrepreneurs and financial intermediaries may affect the achievement of creative destruction, the source of growth in N G M s (e.g. Aghion and Howitt 1992). Asymmetry information in the credit market can come in various forms. There can be ex-ante hidden information which includes the different types of entrepreneurs. Wang and Williamson (1998) have shown that in the presence of ex-ante hidden information, a screening equilibrium with debt contract exists. Bernanke and Gertler (1989) and Williamson (1987) have introduced ex-ante asymmetric information in financial markets into real business cycle models. There can also be ex-post hidden information which involves the output of entrepreneurs (e.g. Townsend 1979 6 0). However, most N G M s do not account for credit market imperfection (Aghion 5 9 As Schumpeter (1961) argued, entrepreneurs can have their own money, but that is not necessary for them to become entrepreneurs. Hence this paper assumes that entrepreneurs have to borrow from financial intermediaries to finance their R&D later. This is in line with Schumpeter's emphasizing the role of credit in economic growth. ( i U Townsend (1979) shows that if financial intermediaries have to use a CSV (costly state verification) technology to examine the returns of projects in cases of bankruptcy, the optimal contract is a debt 49 and Howitt, 1998). The reason is that even though there are attempts to introduce imperfect credit markets into N G M s (e.g. King and Levine 1993), there is little gain in terms of new economic insights (Aghion and Howitt 1998). Hence, Aghion and Howitt (1998) believe that future research should focus more on the financial and institutional aspects of R & D that involve credit-constraints on individual researcher and that this direction of research would enrich the Schumpeterian approach to growth. This paper argues that introducing asymmetric information alone cannot generate richer economic insights. If entrepreneurs' effort is also introduced into N G M s , the extent of new economic insights is contingent on the types of asymmetric information that are introduced in the model. First, for effort to be introduced, the model cannot only have hidden information (i.e. entrepreneurs' type is private information), otherwise, the type can be inferred from the observed effort. In addition to entrepreneur's effort, if hidden action (i.e., entrepreneurs' effort is private information) is solely introduced into N G M s , the model can only yield one effort level of entrepreneurs (thus one growth rate). As for the case when entrepreneur's effort, hidden information and action are introduced into N G M s , growth rate is increasing in entrepreneurs' bargaining share. This is because, without the separation between R & D financier and its carrier (that is, without two representative agents), no "smaller household's share effect" would exist. Higher entrepreneur's bargaining share (i.e. inventive incentive) elicits more entrepreneur's ef-fort, generating the "bigger cake effect" and therefore growth rate is increasing in entrepreneurs' share. Hence introducing asymmetric information alone cannot generate the inverted-U. Based on the foregoing discussion, apart from the assumptions of monopolistic competi-tion, the necessary and sufficient conditions that would generate the inverted-U result are the existence of: (1) hidden action, that is, entrepreneurs' effort is their private information and cannot be observed by others, (2) hidden information, that is, entrepreneurs' type is their private information, (3) two representative agents comprising households who finance R & D , and entre-preneurs who not only finance R & D by saving through financial intermediaries but also borrow from financial intermediaries to innovate, and (4) the second derivative of entrepreneur's effort with respect to his share is negative. Introducing any one of the aforementioned five conditions will complicate the model tremendously, thus making it almost impossible to consider a model that includes all the five conditions. Nevertheless, I developed a model that is considerably less complex than any of the models that would involve those assumptions, and the model deliv-ers reasonable economic insights regarding the inverted-U relationship between entrepreneurs' inventive incentive (their bargaining share) and growth. In this paper, I introduce those aforementioned necessary and sufficient conditions into N G M s . In addition, I focus on debt contract 6 1 given that, in practice, debt contract is an contract. 8 1 A project must have a fixed set-up cost, c(0 < c < 1), while its return is randomly distributed on [0,1]. A debt contract states a payback rule as follows: Financial intermediary pays c, then if the ex post return is in the interval [D, 1] with (0 < D < 1), the entrepreneur needs only to pay back D. If the realization of return is in the interval [0,D], the entrepreneur announces bankruptcy and the financial 50 important arrangement between entrepreneurs and financial intermediaries. F ina l ly , Nash bargaining is used to study how the dis t r ibut ion of the monopolist ic profit from innovation between financial intermediaries and entrepreneurs affects g r o w t h . 6 3 The working of the model is as follows. Entrepreneurs sign debt contracts w i t h financial intermediaries to obtain credit for R & D . In the presence of both hidden information and action, there is an incentive for some entrepreneurs to signal their type. Specifically, entrepreneurs w i t h good projects w i l l signal themselves by renegotiating the terms of contract w i t h financial intermediar ies . 6 4 U p o n receiving the signal, financial intermediaries w i l l commit to changing the terms of the contract in the hope that this w i l l stimulate the entrepreneurs by contr ibut ing more effort into R & D . The result of the renegotiation of the contract is that there w i l l be higher return to bo th financial intermediaries and entrepreneurs. A separating signaling equi l ibr ium, therefore, emerges. The increase in entrepreneur's contr ibution of effort for R & D is contingent on how much bargaining power they have in d iv id ing the monopolistic profit from innovation wi th financial intermediaries. T h e bargaining power is determined by Nash bargaining. The outcomes of the model are as follows. Increasing entrepreneurs' bargaining share of the profit from innovation has two conflicting effects. O n the one hand, a higher share stimulates entrepreneurs to contribute more effort into R & D , which increases the expected profit from innovation (a bigger cake effect). O n the other hand, a higher share for entrepreneurs means lower share for households, which makes households less wi l l ing to save to finance innovation (a dissaving effect). A t the beginning, the bigger cake effect dominates the dissaving effect thereby increasing the growth rates. However, beyond a certain point, the lower share effect dominates and this decreases the growth rates. It can be seen that the balanced growth rate is an inverted-U function of entrepreneurs' bargaining power. The higher the share of entrepreneurs, the more effort is elicited from them, hence the balanced growth rate is also an inverted-U function of entrepreneurs' effort, which offer insights into Lewis (1955)'s view of the th i rd determinant of g r o w t h . 6 5 D i s t r ibu t ing a higher share of surplus from innovation to its creators — entrepreneurs — w i l l monotonical ly increase the income gap between entrepreneurs and households, as well as between good and bad entrepreneurs. 6 6 The presence of asymmetric information coupled w i t h debt contracts makes the decentral-intermediaxy takes over the project. Debt contract is actually a fixed interest rate rule. Here the interest rate is defined as (1 + r) — ^ . 6 2 T h i s is pointed out in Wang and Williamson (1998). For the optimality of debt contract, please see Wang and Williamson (1998), Williamson (1987), Bernanke and Gertler (1989). i i : i Nash bargaining is not a necessary assumption for the inverted-U results. However it helps to simplify the model. M F o r the literature on signaling and renegotiation in contractual relationships, see Beaudry and Poitevin (1993). C i 5 Just as expanding varieties or quality ladders are source of growth in N G M s though they are en-dogenously determined, effort is viewed as a determinant of growth even effort is also endogenous. Effort wil l be affected by other primitive parameters such as bargaining share of entrepreneurs, just as qualities ladders axe affected by patent or product market competition (see Aghion et al 2005). f i f iFor reference on the scarce literature on the interaction between the distribution of bargaining power among different groups of people and economic growth, see Mookherjee and Ray (2002). 51 ized growth rate deviate further from that in a social planner's problem. The reason is that, entrepreneurs bear all of the marginal cost of effort, while receiving only part of its marginal product, so less effort will be forthcoming from them (Stiglitz and Weiss 1981). Though entre-preneurs save through financial intermediaries, they will not take into account the effect of their effort on the whole economy (the balanced growth rate) when choosing their optimal effort due to large market effect. This information asymmetry problem cannot be solved by government intervention. But government can intervene by affecting the bargaining power of entrepreneurs. The paper is organized as follows. Section 3.1.2 describes and solves the model to establish the inverted-U function between the balanced growth rate and the bargaining power of entre-preneurs (also their effort level). Section 3.1.3 evaluates different combinations of assumptions to identify the necessary and sufficient assumptions for the inverted-U results. In doing so, it enables one to understand the economic intuitions more clearly. Section 3.1.4 concludes. To provide a framework to prove that effort is a determinant of long-run growth, I extend the endogenous growth model with expanding varieties to include financial imperfections. 6 7 The economy consists of a final goods sector, an intermediate goods sector, financial intermediaries and households. A l l investments are made through financial intermediaries, which absorb sav-ings from households. Each intermediate good represents an innovation by an entrepreneur. Entrepreneurs are credit-constrained and must borrow from financial intermediaries to finance innovation. Each innovation is a project. Entrepreneurs cannot finance their projects with their own money, but they can put deposits in financial intermediaries to earn dividends. Thus, only external financing is considered here. Intermediate goods are inputs of the final good sector. Similar to Borensztein, De Gregorio and Lee (1998), a final good firm produces a single consumption good using aggregate production function in equation (1). where N the number of innovations, Xj the amount of intermediate good j, Aj the productivity of intermediate good j, L the labor force. Each j is supplied by a monopolistic entrepreneur. The innovation cost (the cost of R & D ) of each intermediate good is a fixed amount, n. Each intermediate good j has a random productivity Aj with all A'-s being distributed on interval [0, A], where A > 0. Each innovation ends up with probability density function (PDF) fg yA, ej with probability A, and fb (A) with probability (1 — A), where 0 < A < 1, and e (> 0) stands for the effort level 6 7 For the original model of endogenous growth models with expanding varieties, see Barro and Salami-Mart in (1995). 3.1.2 The Model ( i ) 52 contributed by entrepreneurs, which is a parameter in the density function. I call those projects with P D F fg (^A, ej "good" projects, whose owners are therefore "good" entrepreneurs; and I call those projects with (^A^j "bad" projects, whose owners are "bad" entrepreneurs. For simplicity, I assume the P D F of bad projects cannot be affected by the effort of bad entrepreneurs.68 Good projects' cumulative distribution function (CDF) first-order-stochastically dominates that of bad projects, that is, for x € [0, A], Fg (x, e) < Fi, (x). This ensures that good projects have higher mean value of A (higher expected profit). Further, I assume that Fe(x,e) < 0 and Fee (x, e) > 0, which ensure that good projects' distribution function with a higher effort first-order-stochastically dominates that with a lower effort.6 9 Given these, the higher the share of monopoly profit from innovation that goes to entrepreneurs (denoted as 3), the more effort will be forthcoming from entrepreneurs, that is, ^§ > 0 (see lemma 6). I assume that g f < 0: a higher share of monopoly profit to entrepreneurs will elicit more effort from them, but at a decreasing rate. 7 0 The economy consists of a fixed number of people, denoted by L. The assigning of occupation is through a lottery. Those who lose the lottery become workers who have unit labor endowment and supply it to final good firms. Those who win the lottery become entrepreneurs. People cannot change their occupations. For entrepreneurs, some end up with "good" projects, while the rest end up with "bad" projects. Each entrepreneur has a reservation utility, u, for each period. And I assume infinite horizon and agents live forever. I assume the objective functions of entrepreneurs and workers are the same: 7 1 f°° otic-N-hie)]1-0 -1 , Max jQ e-"I fU} dt (2) where h (e) is the cost associated with effort, whose cost is additive to consumption C; N is the number of innovations. I assume that h (0) = h' (0) = 0 and h' (e) > 0, h" (e) > 0. Workers maximize the case of h (0) = 0 with respect to C. The maximization problem for all entrepreneurs is the same. The utility function of entrepreneurs consists of two parts: the utility from consumption, and the disutility of effort. The consumption of entrepreneurs comes from the profit of innovation, n(e,8), and interest income. For entrepreneurs, using U (e.B) to denote ir(e,3) — h(e), I assume U (e.B) is concave in e if Fg (x,e) first-order-stochastically h 8The result of the paper does also hold with the assumption that PDF of bad projects can be affected by the effort of bad-entrepreneurs, that is, the PDF of bad projects is: (A, e^j. For this case, the single crossing property: Fg>e (^A, > Fb>e (A, ej I S needed. B 9That is, given Fe (x, e) < 0, if e' > e, then Fg (x, e') < Fg {x, e) for V x 6 [0, A]. 7 0 This condition is sufficient to ensure that balanced growth rate is concave in fj. But it is not a necessary condition. If this assumption does not hold, then balanced growth rate is not bounded in equilibrium. This assumption is made to be consistent with the roughly constant balanced growth rates in steady state economies such as the USA. 7 1 There is separation between R&D financier (i.e. the households) and R&D carrier (i.e. the entre-preneurs) in this model, which differs from the one representative agent model in standard endogenous growth literature. This separation helps to derive the results of this paper. 53 dominates Fb (x) and Fg (x,e) is very elastic in e. The model has two representative agents: a worker (household) and an entrepreneur. This allows us to examine how the distribution of the monopoly profit from innovation between households (the R & D financier) and credit-constrained entrepreneurs (the R & D carrier) affects growth. Entrepreneurs have to sign a basic debt contract 7 2 with financial intermediaries to get credit for their R & D . After entrepreneurs spend their credit, nature determines the types of their projects, which is hidden information. Those entrepreneurs who have good projects will spend <r in signaling to renegotiate the terms of their contracts. Receiving the signal, financial interme-diaries will agree to change the terms of contract to encourage entrepreneurs to contribute more effort into R & D , ending up with a higher return both to financial intermediaries and to entre-preneurs. The new contract is unattractive to entrepreneurs with bad projects, so a separating signaling equilibrium exists. Entrepreneurs contribute more effort into R & D depending on how much bargaining power they have in dividing the monopolistic profit from innovation with financial intermediaries. Entrepreneurs and financial intermediaries use costless Nash bargaining to share the monopolistic profit from innovation, and entrepreneurs' share is given as 3. In this paper, 3 is assumed to exogenously given but can be changed by government intervention. Financial intermediaries and/or entrepreneurs cannot change 3 alone or together. In the end, the ex post realization of A is known to everyone. Entrepreneurs pay back their loans in accordance with the contracts. The timing of the model is summarized as follows: 1. People make their savings decisions, and if they save, they must save through financial intermediaries. Then the lottery for patents to the intermediate goods is announced. Those win the lottery are entrepreneurs. 2. A l l entrepreneurs receive credit from financial intermediaries by signing a basic debt con-tract. After they spend their credit on R & D , nature determines the types of their projects. Then entrepreneurs determine how much effort to contribute. The type and effort of en-trepreneurs cannot be observed by financial intermediaries. 3. If projects turn out to be good, their owners have incentives to signal themselves to rene-gotiate with financial intermediaries. Where there is a separating equilibrium, 7 3 based on the chosen effort, good entrepreneurs determine how much to spend in signaling. Receiv-ing the signal, financial intermediaries agree to share the difference in the profits between good projects and bad ones with entrepreneurs according to costless Nash bargaining. The share of good entrepreneurs is given as 3. The owners of bad projects stick to the basic debt contract. 4. A is realized and the profit from innovation is distributed in accordance with the contracts. 7 2 See Section 2.3 for the definition of a basic debt contract. 7 3 A separating signaling equilibrium with hidden information and hidden action is used here. 54 The remaining of Section 3.1.2 proceeds as follows. Subsection 3.1.2.1 solves the behavior of firms. 3.1.2.2 solves the debt contracts. 3.1.2.3 solves the behavior of financial intermediaries. 3.1.2.4 solves the behavior of households and entrepreneurs and the general equilibrium of the model. 3.1.2.5 studies how balanced growth rate is affected by the bargaining power and effort of entrepreneurs. 3.1.2.6 discusses Pareto optimality and policy suggestions. 3.1.2.1 The Behavior of Final Good Firms and Intermediate Good Firms A final good firm produces a single final good using production function in equation (1). It maximizes its profit by taking as given the wage rate, the prices of intermediate goods, and the ex post realization of Aj. The demand for the j - t h intermediate is obtained from the F O C A n innovation transforms one unit of final good into one unit of an intermediate good. Normalizing final goods' price as one, the unit cost to intermediate good firms is also one. After invention is done, intermediate good firm j takes as given the demand from the final good firm 3.1.2.2 Debt Contracts with Nash Bargaining Before any R & D is conducted, the ex ante types of projects are unknown to all parties, so entrepreneurs must sign a basic debt contract with financial intermediaries to get credit for their R & D . Basic debt contract is signed according to a project with P D F fb (^A^J, and it states a payback rule as follows. Financial intermediary pays the cost of R & D , r\. Then if the ex post A is in the interval [A,, A] with 0 < Db < A, entrepreneur pays back Db • I; If the realization of A is in the interval [0, Db], entrepreneur announces bankruptcy and financial intermediary takes over the project without additional cost. After entrepreneurs spend the credit on R & D , nature determines the types of their projects, which are their private information. Entrepreneurs with good projects have incentives to signal their type so as to renegotiate the payback rule with financial intermediaries. Good entre-preneurs choose their optimal effort and then spend <j in signaling. Observing the <j, financial intermediaries agree to change the payback rule, and the new debt contract has a similar struc-ture to that of the basic debt contract, with Db being replaced by Dg. The new debt contract is set so that good entrepreneurs and financial intermediaries will share the difference of the expected profits between the two types of projects according to a Nash Bargaining rule, with entrepreneurs' share being 3. In a separating signaling equilibrium, there are two debt contracts offered by financial in-termediaries. The basic debt contract is for bad entrepreneurs, who have no incentive to signal their type. The second debt contract is for good entrepreneurs who have to spend c, which is a function of their effort e, in signaling to secure the new contract. to maximize its profit, which determines the price mark-up: Pj = ^ , and the monopoly profit TXJ = AjL — l ) aJ1^ = / • Aj, where I is constant and equal to L (-^p) a 1 - ^ -55 Given a Nash bargaining rule (3), the separating signaling equilibrium with debt contracts can be solved by backward induction. In 3.1.2.2.1, I solve the separating signaling equilibrium, taking the effort of entrepreneurs as given. Here, two debt contracts, that is, Db, Dg and <; are solved. Then I determine the effort levels of entrepreneurs in 3.1.2.2.2. After effort is solved, the separating signaling equilibrium is solved. 3.1.2.2.1 Solve the Separating Signaling Equilibrium Taking Effort as Given Step 1. Solve the Basic Debt Contract In a separating signaling equilibrium, bad entrepreneurs spend nothing in signaling (i.e. they do not signal themselves) and stick to the basic debt contract as follows: f I • (A-Db) ; if l e [Db,A] Entrepreneur has to pay back: < \ / _ 1 0 -L ifAe[0,Db] ™ . , . . • • f l-Db ; ifAe[Db;A] Financial intermediary receives: < ~ ~ \ I-A ; ifAE[0,Db] where Db needs to be solved. Due to the structure of the model, each agent's objective is simply to maximize each period's utility. Bad entrepreneurs must accept the terms of the basic debt contract, a take-it-or-leave-it offer by financial intermediary. The financial intermediary's prob-lem is to maximize its expected profit, subject to bad entrepreneur's participation constraint: a bad entrepreneur earns at least as much as his reservation utility u. Solving the financial intermediary's problem in equation (3) produces lemma 1. Max Rb = / c-xfb{x)dx + l-Db[l-Fb(Dbj\ (3) Db JO .t. irb = / I • (x — Db) fb (x) dx > u JDb Lemma 1. Z>* is solved by equation (4), and financial intermediary's expected profit from financing bad projects is given by equation (5). Proof: A t optimality, the constraint in equation (3) binds: 7T*, = u. Integrating the constraint by parts solves the optimal £)*. as in equation (4). Rewriting financial intermediary's objective function, integrating by parts, and using equation (4), financial intermediary's expected profit from financing bad projects is given in equation (5). Q.E.D. D*b- /o Fb(x)dx = Eb(A)-j (4) R*b = l-Eb(A)-u (5) Lemma 1 states the following. Since the P D F of bad projects cannot be affected by bad en-trepreneurs' effort, the financial intermediary always pays a bad entrepreneur his reservation 56 utility. Then the financial intermediary' expected profit is the total expected profit from a bad project, lEb (^)> l e s s the bad entrepreneur's reservation utility, u. Step 2. Solve the Debt Contract for Entrepreneurs with Good Projects To renegotiate with financial intermediaries to get this new contract, entrepreneurs with good projects have to spend <j in signaling. The new debt contract gives: v f , , f • J l - U - D g ) - , ; ifAe[D9,A] Entrepreneur s profit is : < \ / _ I ; ifAe[0,Dg] Financial intermediary receives:/ 1 ' ? ' 5 \ l - A ; if A £ [0, D g ] where I3 9 and c need to be determined. The difference between this debt contract and the basic one is that, good entrepreneurs must signal their type each period to renegotiate Dg with a financial intermediary according to the Nash bargaining rule. Note that, good entrepreneurs bear all of the cost of signaling. Good entrepreneurs solve their effort independently from their consumption, which will be proved later (see proof of proposition 1). Here, good entrepreneurs' objective is: Max Ug = I I • (x — Dg) fg (x, e)dx — h (e) — subject to: [1] D* is determined by Nash bargaining with the entrepreneurs' share 0 used to divide the difference of profits between good projects and bad ones; [2] Bad entrepreneurs' incentive constraint, which makes sure the contract of {c*,D*} is not better than {0,D*}; [3] Good entrepreneurs' incentive constraint, which makes sure the contract of {c*,D* (e*)} is not worse than {0,-D*. (e)}; [4] Good entrepreneurs earn at least as much from contract {0, (e)}; [5] Financial intermediaries' incentive constraint, that is the financial intermediary gets at least as much from contract {c*,D* (e*)} as from contract {0,D*. (e)}. Given good entrepreneurs' problem, the financial intermediary gets expected profit: Rg = l / xfg(x,e)dx + l-Dg[l-Fg(Dg,e)] Jo Now solving good entrepreneurs' maximization problem proceeds as follows. Constraints [1], [4] and [5] are considered first. 57 Constraints [4] and [5] are determined by the case of 3 == 0, which means that financial intermedi-aries offer the same basic debt contract £)* for both types of entrepreneurs. Good entrepreneurs will not signal themselves and contribute an optimal amount of effort, denoted as e, into their projects. And e is solved as: fA = arg maxUg = / I • (x — Dl) fg (x, e) dx — h (e) e JDT (6) Given e and 7J*, good entrepreneur and financial intermediaries receive the following, which are their reservation utilities in Nash Bargaining: Ug = [ l-(x-D*b)fg(x,e)dx-h(e) (7) .ID; Rg = I (DI - jjb Fg (x,e)dx^j (8) However, if 3 > 0, then good entrepreneurs have incentives to signal themselves. After spending C in signal, a good entrepreneur earns at least as much as Ug. Since good entrepreneurs bear the signaling cost, 7 4 the signaling cost is not involved in the Nash bargaining. For any given 3 > 0 in Nash bargaining, the numbers in equations (7) and (8) are the reservation prices of financial intermediaries and good entrepreneurs, who bargain over the profit from good projects less the sum of their reservation prices. The sum of the two parties' reservation utilities is n (e) = lEg ^A, —h (e), which is the total profit from good projects less the cost of good entrepreneur's effort in the case of 3 = 0. Note that the reservation prices are already determined. U s i n g c o n s t r a i n t s [4] a n d [5] t o s o l v e t h e N a s h B a r g a i n i n g g i v e n 3 £ (0,1]: 1-/3 Max(ug-Ug) (Rg-Rg The solution is: ug — Ug = ^Rg — Rg^j, where ug denotes the utility of good entrepreneurs before deducting the signaling cost. Since good entrepreneurs bear all of the signaling cost, their final utility is (ug — c). Given (ug — Ug^ + ^Rg — Rg^j = U (e) — n (e) , where n (e) = cEg ^ A , ej — h (e), and 11(e) = Ug + Rg, I have: ug-Ug = B[U(e)-U(e)\ (9) R9-R9 = ( l - / 3 ) [ n ( e ) - f i ( e ) ] (10) 7 4 T h e results axe the same if we let financial intermediaries share the signaling cost, <;, that is, to let entrepreneurs and financial intermediaries bargain over / ^ Eg ^A, e^j — Et, ~ 58 The Nash bargaining solution in equations (9) and (10) says that besides their reservation prices, good entrepreneurs receive 8 of the share of the difference in total surplus IT (e) — II (e). while financial intermediaries receive (1 — 8) of the share. If II (e) > II (e), receiving the signal q, financial intermediaries agree to change the payback rule by changing the basic debt contract into a new debt contract (change into D*) for good entrepreneurs. And financial intermediaries receive higher return: Rg > Rg. L e m m a 2. In a separating equilibrium, constraint [1] is solved, that is, Dg is solved as a function of effort. Further, Dg can be solved independent of signal And D* is a decreasing dD* function of 8, that is, < 0. Proof: Rg = I (og - jf' Fg (x, e) dx") =(1-8) [n (e) - ft (e)] + Rg =» I (D9 - f°9 Fg (x, e) dx^j =(1-8) [n (e) - fi (e)] + RG (11) From equation (11). the optimal Dg can be solved as a function of e only, independent of signal q. Thus constraint [1] is solved. The reaction function of Dg as a function of e will be used later on to pin down the optimal effort level e*. And taking the partial derivative of Dg with respect to 8 will deliver ^ < 0. Q.E.D. N o w it is t ime to check i f constraints [2] and [3] are satisfied. Tr'b = / l-(x-Dg)fb(x)dx-q<ir*b=u [2] JDg Ug(e) = / l(x-Dl)fg(x,e)dx-h(e)<Ug(e*) [3] •'Dt where in constraint [3], e* = arg max!7 s = fD I (x — Dg) fg (x, e)dx — h (e) — q; e is determined e ' 9 by equation (6), and q (e) = 0. L e m m a 3. For a separating equilibrium to exist, I must have D* < D^, that is, the interest rate ( ^ ) above which to announce bankruptcy is lower for good entrepreneurs than for bad entrepreneurs. Proof: The informed good entrepreneurs must spend some of their profits to signal them-selves. Given effort, good entrepreneurs' utility is a decreasing function of the signal cost as in equation (12). Good entrepreneurs' utility is higher if the signal is smaller. ,A Max :Ug= l(x- Dg) fg (x, e)dx-h (e) - q (12) Then, from the profit function of bad entrepreneurs, I have: Trb = A - D - l I Fb(x)dx=*^ = -[1-Fb (D)} < 0. 59 Thus the higher D is, the lower will be the profit of bad entrepreneurs. If Dg > Z)*, then ir'b = fp I (x — Dg) fb (x) dx < 7T*,, and then good entrepreneurs will not spend anything in signaling, <j = 0, and no separating equilibrium exists. For a separating equilibrium to exist, therefore, I must have D* < D£. In this case, good entrepreneurs have to make the signal so large that it will be unprofitable for bad entrepreneurs to pretend they are good ones. Q.E.D. Combining lemma 2 and 3 produces figure 2.1. L e m m a 4: The optimal c is to make constraint [2] bind, and it is positive. Proof: From the proof of lemma 3, good entrepreneurs' utility is higher if the signal is smaller. Now I must use constraint [2] (the bad entrepreneurs' IC) to pin down the signal. B y observing constraint [2], I get the higher the signal, the lower the profit for a bad entrepreneur to mimic a good entrepreneur. Thus, the optimal signal will be that just makes constraint [2] bind. The minimum signal that makes constraint [2] bind is: (D*b-D*g)- fD" Fb(x)dx JD* (13) From lemma 3, D* > D*, thus Lemma 4 states that optima] (D*b - D*g) - Qg Fb (x) dxj > 0, thus, c* > 0. Q.E .D. signal c* makes the basic debt contract (D*) and spending c* to get the new debt contract (D*^ indifferent for bad entrepreneurs; in this case bad entrepreneurs choose the basic debt contract. Signals lower than c* do not separate the types of entrepreneurs, and signals higher than c* give good entrepreneurs lower utility. Thus good entrepreneurs will always spend just <;* in signaling. L e m m a 5. Constraint [3] is satisfied if constraint [2] holds with equality. Proof: I have: Ug (e) = I (x — D*) fg (x, e) dx — h (e) — < Ug (e*), which is ensured by: e* = argmaxt/g = (x — D*) fg (x,e) dx — h(e) — q*. Thus constraint [3], that is, Ug (e*) > Ug (e), is satisfied if Ug (e) > Ug (e). And Ug (e) > Ug (e) if and only if: 7T, rA /-A (e) = / / (x - DD fg (x, e) dx < TT9 (e) = / / (x - D*) fg (x, e) dx -JDI JD* The above can be simplified as: JD*„ Fg (x, e) dx Given that Fg (x.0) < Fb (x) for Vx G [0,A], and Fe (x,e) < 0, I have: Fg (x,e) < Fb (x) for Ve > 0. This together with (13) makes sure the above equation is satisfied. Thus constraint [3] is satisfied if constraint [2] holds with equality. Q.E.D. Lemma 5 reveals that given the optimal signal c*, the basic debt contract (D*) and spending 60 q* to get a new debt contract (D*) are indifferent to bad entrepreneurs, and good entrepreneurs will always find it optimal to signal themselves by spending c* to renegotiate their payback rule. In accordance with the above Nash Bargaining solution, financial intermediary will agree to change D * into D* for good entrepreneurs. 3.1.2.2.2 Solve the Opt imal Effort e* and e*. Taking into account the debt contracts solved in 1, entrepreneurs choose their optimal effort to maximize their utility, which produces lemma 7. Lemma 6. e* and e*, can be solved and have the following properties, e*, = 0. If Fg (x,e) first-order-stochastically dominates Fb (x) and Fg (x, e) is very elastic in e, then e* > e for V/3 > 0, which produces: II (e*) > 11(e). Then constraints [4] and [5] are satisfied and a separating signaling equilibrium exists. And optimal effort e* is an increasing function of 8: Proof: See the Appendix. Lemma 6 ensures the existence of a separating signaling equilibrium. After efforts are solved, lemma 2 solves D*, lemma 1 solves D*, and lemma 4 pins down q*. And the separating signaling equilibrium with two debt contracts is solved. 7 5 To recap, after entrepreneurs sign a basic debt contract and spend the credit in R & D , they know the types of their projects. If the C D F of good projects first-order-stochastically dominates that of bad ones, and the C D F of good projects is very elastic in entrepreneurs' effort, then good entrepreneurs find it optimal to signal themselves to renegotiate the pay-back rule with financial intermediaries in accordance with a Nash Bargaining rule. And financial intermediaries agree to offer good entrepreneurs a new debt contract which yields higher returns to both parties by stimulating good entrepreneurs to work harder. And the cost of signaling is just high enough to make bad entrepreneurs stick to the basic debt contract. A separating signaling equilibrium emerges. Figures 2.2 and 2.3 illustrate the separating signaling equilibrium and lemma 6. 3.1.2.3 The Interest Rate I assume Bertrand competition for financial intermediaries, or simply the zero-profit condition. Financial intermediaries represent those who save (households). Given the existence of a sepa-rating signaling equilibrium, the expected (average) interest rate is: E(R)_ _^ _ ( 1 - A ) i % + A [ f l g + (1 - / ? ) [n(e*)- I I (e ) V 7 5 F r o m the assumptions, proofs and arguments, it can be seen that the signaling equilibrium is the optimal equilibrium. 61 Given the interest rate, solving the households' and entrepreneurs' problem will produce the balanced growth rate of the model, which is performed in 3.1.2.4. 3.1.2.4 The Behavior of Households and the General Equilibrium of the Model P r o p o s i t i o n 1. The model has a balanced growth path and a steady state. Proof: See the Appendix. In steady state: ^ — jf- — %l — Ti = Y ^  where C denotes the consumption of workers, Cg the consumption of good entrepreneurs, C\, the consumption of bad entrepreneurs, N the number of varieties, and Y the total output. On a balanced growth path, different groups of people share the same growth rate of their consumption, though their consumption levels are different. A worker's consumption grows at rate ^ . A good entrepreneur's income from savings grows at rate JJ, and so does his income from innovation, n*XN. A bad entrepreneur earns dividends from financial intermediaries and profit from innovation, (1 — A) Nu, both of which grow at rate jf. Thus the consumption of all entrepreneurs grows at rate ^ given that their saving rate is constant facing a constant interest rate. From the appendix, the balanced growth rate of the economy (all consumptions) is: c 1 _(l-\)Rt + x[R9 + (l-d)(u(e*)-U(e) P c = i < ' - » - 1—er^ <14> Existing endogenous growth theories do not model entrepreneurs' effort, in other words, entre-preneurs' effort does not respond to their share, 3. In those models, first of all, the balanced growth rate will be a monotonically decreasing function of 3, which cannot not be used to eval-uate existing pro-entrepreneur policies; secondly, the balanced growth rate is not a function of entrepreneurs' effort, which does not allow us to examine how entrepreneurs' effort affects the balanced growth rate. In this model, equation (14) shows that, by incorporating entrepreneurs' effort which responds to 3 as shown in figure 2.3, the balanced growth rate will not necessarily be a monotone function of 3, because entrepreneurs' effort, which responds to B, also affects the balanced growth rate. 3.1.2.5 The Balanced Growth Rate, Entrepreneurs' Bargaining Power and Effort P r o p o s i t i o n 2. If the conditions in lemma 6 are satisfied, firstly, the balanced growth rate is an inverted-U function of 3 (the bargaining share of good entrepreneurs) and there exists a unique 3* 6 (0,1) that maximizes the balanced growth rate; second, the balanced growth rate is also an inverted-U function of good entrepreneurs' effort. Proof: See the Appendix. 62 The balanced growth rate is in equation (14). Increasing the bargaining power of good entrepreneurs over the surplus from innovation (an increase in 8) has two conflicting effects on balanced growth rates. On the one hand, a higher 8 increases good entrepreneurs' incentive to innovate by stimulating them to contribute more effort in R & D , increasing the expected profit from innovation, the II (e*). This "bigger cake effect" pushes up the growth rate. On the other hand, a higher 8 means lower share goes to households (lower (1 — 8)), which makes them less willing to save to finance innovation. The "smaller households' share effect" decreases the growth rate. To see this clearly, taking derivative of balanced growth rate with respect to 8 produces: dg 08 X_ 9ri II ( e * ) - 1 1 ( e ) +(1-/3) a n (e*) de de dfi (15) In equation (15), the "bigger cake effect" is (1 — 8) d U ^ ^ ^jf, and "smaller households' share effect" is II (e*) — 11(e) . From the proof in the appendix, we have: dg, _ A fdn(e)l_\ d e > Q QP*P=p,~* On \ de '7 dB % = _ A d8l0=1 ev II ( e * ) - 1 1 ( e ) < 0 (17) When 8 = 0, [II ( e * ^ ) - II (e)j = II (e) - II (e) = 0, so the "smaller households' share effect" is zero. A n d equation (16) states, when 8 = 0, the "bigger cake effect" (9n^ ^ jp) is larger than zero. Thus, at the beginning, as 8 increases, good entrepreneurs will work harder to make the cake larger (dUQ^ ^ 5§ > 0), a n d this "bigger cake effect" dominates the "smaller households' share effect". The balanced growth rate, therefore, is increasing as 8 increases. As 8 continues to increase, say the extreme case (1 — (3) = 0, then all the increase of cake goes to entrepreneurs, the "smaller households' share effect" is dominated, giving households the least incentive to finance R & D , and yielding the lowest growth rate. Thus, the balanced growth rate is an inverted-U function of 8, and equations (16) and (17) ensure there is a unique 8* E (0,1) that maximizes the balanced growth rate (see figure 2.4). Lemma 6 delivers that good entrepreneurs' optimal effort (e*) is an increasing function of 8. Thus, balanced growth rate is also an inverted-U function of the optimal effort of good entrepreneurs. This makes us ponder over the aforementioned Lewis' view: the effort of agents is the third determinant of long-run growth. 3.1.2.6 Pareto Optimality and Policy Implications Proposition 3. In this model with asymmetric information, the balanced growth rate is not the highest in a decentralized economy. Proof: I define ep = arg max II (e), where II (e) = lEg (^A, e^j — h (e). And ep can be enforced in the case of perfect information, where financial intermediaries observe entrepreneurs' effort 63 and types. T h e n financial intermediaries choose the tangent point of their iso-profit curve and entrepreneurs' indifference curve, which gives entrepreneurs their reservation uti l i ty, gives financial intermediaries the highest expected profit, and yields the highest balanced growth rate i n a decentralized economy (see figure 2.2). However, w i th information asymmetry, good entrepreneurs' op t imal effort (e*) is different from ep since e* ^ a rgmax n ( e ) . T h e reason e is that good entrepreneurs bear al l of the marginal cost of effort, h! (e), while receiving only 8 W (e*) — II (e) , or, 3 share of the marginal product of effort, so less effort is forthcoming from them, which lowers the growth rate i n a decentralized economy. Q . E . D . Thus , besides the monopoly pr ic ing dis tor t ion i n endogenous growth models w i t h expanding var ie t ies , 7 6 there are distortions in this model due to information asymmetry, which makes the decentralized growth rate further deviate from that i n a social planner's problem. Though good entrepreneurs save through financial intermediaries, when they choose their effort to maximize their share of the output, they w i l l not take into account the effect of their effort on the whole economy due to a large market effect. If government does not have information advantages over private agents, government inter-vention cannot solve the distort ion from information asymmetry. However, there are other ways government can increase the balanced growth rate. Accord ing to proposit ion 2, government can intervene by affecting 3, the Nash Barga in ing power of entrepreneurs. W h e n 3 £ [0, 3*}, an increase i n 3 w i l l give both financial intermediaries and good entrepreneurs higher return, so such an increase in 3 w i l l meet no obstacle . 7 7 In this case, government can increase banking sector compet i t ion. W h e n 8 £ \J3*,1], an decrease in 8 w i l l give higher return to financial i n -termediaries and lower return to good entrepreneurs, so such a decrease in 8 (redistribution) w i l l be hard. In this case, government can increase the entry barrier to the banking sector, or increase compet i t ion i n the industr ia l sector. 3.1.3 Necessary and Siifficient Assumptions for the inverted-U Results Here I formally shows that three features (hidden information, hidden action and two-representative agents) together w i t h monopolist ic competi t ion and J^f < 0 are necessary and sufficient to ensure the inverted-U result. Introducing 'separation' and/or hidden action alone into N G M s yields either one effort level of entrepreneurs (thus one growth rate) or growth rate is monotone in entrepreneurs' bargaining share (inventive incentive), which is shown i n 3.1.3.2, 3.1.3.3 and 3.1.3.4. If there is only hidden information, effort cannot be introduced; otherwise type can be 7 6 For the monopoly pricing distortions in endogenous growth models with expanding varieties, see Barro and Sala-i-Martin (1995). 7 7 It looks like that the financial intermediaries wil l automatically give up bargaining power in this case. However, as pointed out in footnote 6, debt contract is virtually a fixed interest rate rule. If interest rate is controlled by the monetary authority or the government, then the Nash bargaining power is actually controlled by the the monetary authority or the government. In this paper, f3 is assumed to exogenously given but can be changed by government intervention. 64 inferred from observed effort. If there are only hidden information and hidden action, growth rate is increasing in entrepreneurs' bargaining share, which is shown in 3.1.3.4. 3.1.3.1 One Representative Agent (No Separation) and No Asymmetric Information This is the simplest case, which corresponds to a standard N G M . If effort is introduced into N G M s without introducing asymmetric information and two representative agents, then there is only one type of projects, the good projects. Then entrepreneurs only contribute the optimal (and highest effort) into R & D , that is, ep = argmax 11(e) (see section 3.1.2.6). Then the e balanced growth rate is: 1 1 fU(eP) \ 9=e{r-p) = e\—-p) Hence, entrepreneurs always choose the effort level, ep, and the balanced growth rate is fixed. Introducing only entrepreneurs' effort into standard NGMs, there is little gain in terms of new economic insights. 3.1.3.2 Two Representative Agents and No Asymmetric Information If there is no asymmetric information, the bad projects are always known to financial interme-diaries. And since they always give financial intermediaries lower return, thus they are always not financed. For simplicity, they are dropped from the model, so the model has only one type of projects, the good projects, whose return are affected by entrepreneurs' effort. As there is no asymmetric information, entrepreneurs' effort can be observed by financial intermediaries. Financial intermediaries will always force entrepreneurs to contribute the effort level, ep as in 3.1.3.1. In figure 3.5, the blue lines are the indifference curves of entrepreneur; the green lines are the iso-profit curves of financial intermediary; and the red line is the reaction curve (IC curve) of entrepreneur's effort as a function of D (See Appendix for the proof). The perfect information case allows financial intermediary to choose the debt contract as the tangent point between entrepreneur's indifference curve I: U = Ti and financial intermediary's iso-profit curve 1, the point (Dp,ep). Then they divide the maximized total surplus from innovation according to a Nash bargaining rule with the share of entrepreneurs as 3. Now the reservation utilities are zero and u, for financial intermediary and entrepreneurs respectively. Thus, the balanced growth rate is: (l-3)(U(ep)-u) p 9 6n 6 Hence, we have: 65 Now, we only have the "smaller households' share effect", so higher share of entrepreneurs yields lower balanced growth. This reason is straightforward. As the total surplus from innovation is fixed, higher share of the surplus taken away by entrepreneurs means lower share left for R & D financiers (both households and entrepreneurs). As R & D financiers receives lower return from financing R & D , balanced growth rate is lower. In this framework, there is no room for pro-entrepreneur policies to exist. 3.1.3.3 Two Representative Agents, Hidden Action and No Hidden Infor-mation Now suppose in 3.1.3.2, the effort of entrepreneurs is unobservable by financial intermediary. Then the contract pair (Dp, ep) can't be enforced. The reason is that, given Dp, entrepreneur chooses his optimal effort according to equation (A- l ) . The resulting contract pair will be the cross point of the red IC curve and entrepreneur's blue indifference curve II, the point A shown in figure 3.5. Given point A , the resulting iso-profit curve for financial intermediary will be the dashed green line 3. And most important, at point A , the slopes of the IC curve and iso-profit curve are different. Thus, point A is not an equilibrium. To determine the equilibrium, it is necessary to compare the slope of the IC curve in equation (A-2) and the slope of the iso-profit curve for financial intermediary in (A-4). If the red IC curve is steeper than the iso-profit curve 3 at point A(the effort is very in-elastic with respect to D), then the cross point, (D*,e*), of the red line (IC) and the blue line of the indifference curve I: U = u of entrepreneur (PC) gives the optimal debt contract between financial intermediary and entrepreneur. The reason is that effort is very in-elastic with respect to D, financial intermediary will find it optimal to increase D to induce lower effort level from entrepreneur, and the increase in financial intermediary's profit due to higher D dominates that of from the decrease from lower e, D increases until the participation constraint (PC) is binding. As in figure 1, the points (D*,e*), and (Dp,ep), both lie on the same indifference curve that gives entrepreneur his reservation utility u. If the IC curve is flatter than the iso-profit curve at point A (the effort is very in-elastic with respect to D), then the equilibrium point will be to the right of point A, where IC curve is tangent to iso-profit curve. In that case, entrepreneur will earn higher utility than u, for the reason that effort is very in-elastic with respect to D, financial intermediary will find it optimal to lower D to induce higher effort level from entrepreneur, and the increase in financial intermediary's profit due to higher effort dominates that of from the decrease from lower D. In this case, the participation constraint (PC) is not binding. However, in either case, the IC curve is to the left of the perfect information point (Dp,ep), financial intermediary always receives lower expected profit than that in the perfect information case, Rp. 66 As to the relationship between entrepreneurs' bargaining share and growth, it is obvious that only two contract points are possible in this case, thus there is only one effort level, thus only one growth rate. 3.1.3.4 One Representative Agent (No Separation) and Asymmetric In-formation In this case, everything is same as in the original model in section 3.1.2. The only difference is deriving the balanced growth rate. Since now there is only one representative agent, in getting the market interest rate, I should use the sum of the surplus from innovation for both financial intermediaries and entrepreneurs instead of just the surplus of financial intermediaries as in section 3.1.2. The reason is straightforward, everything in the economy belongs to the only representative agent now. Hence the balanced growth rate is not as that in equation (14), but the one as follows: _ ( 1 - A ) (R*b + u) + \U(e*) _ p 9 ~ 6n 9 Taking derivative of balanced growth rate with respect to 8 produces: dg_ _ A dU (e*) de dp ~ 9^ de dp Now we only have the "bigger cake effect" left. Thus increasing the bargaining share of en-trepreneurs elicits more effort from them, which is unambiguously good for growth in a one representative agent framework. Thus, no separation between R & D financier and its carrier, that is, no two representative agents, no "smaller households' share effect". 3.1.3.5 Necessary and Sufficient Assumptions for the Inverted-U Result First of all, monopolistic competition as in N G M s is necessary since the model and thus en-dogenous growth is based on standard N G M s . Besides this, two representative agents, the R & D carrier (entrepreneurs) and its financier (households, entrepreneurs as savers, and their delegate financial intermediaries), makes sure the existence of "smaller households' share effect". Hidden action together with hidden information ensures the existence of "bigger cake effect". The rea-son is that, when information is asymmetric, the effort level of entrepreneurs is distorted. And there is institutional arrangements which can elicit more effort from entrepreneurs, generating a "bigger cake". Hidden action alone cannot generate the "bigger cake effect" as shown in 3.1.3.3. The economic reason is that, when is no hidden information (types), the financial intermediary has no way to elicit more effort from entrepreneurs given the incentive constraint of entrepreneurs has to be satisfied. Adding hidden information (type) gives financial intermediaries one more 67 instrument (by offering two types of contracts) to elicit more effort level of good entrepreneurs since they no longer need to worry about the incentive constraint of good entrepreneurs. The different contract deals with the incentive problem. Hidden information (type) alone is not interesting because we cannot introduce entrepreneurs' effort into N G M s , otherwise the types can be inferred based on the observed effort level of entrepreneurs which is not hidden action. 3.1.4 Conclusions and Policy Implications The model introduces micro-foundations and contract frictions into N G M s and shows the exis-tence of an inverted-U relationship between entrepreneurs' inventive incentive measured as their bargaining share and balanced growth. The results are new and have never been previously documented in the literature. Not only does the model enrich the Schumpeterian approach to growth (Aghion and Howitt 1998), it also shows that entrepreneurs' bargaining share is an im-portant determinant of long-run growth. While the inclusion of informational and contractual frictions complicates the model, I show that they are the necessary and sufficient conditions in es-tablishing the inverted-U result for balanced growth rate. Most importantly, these assumptions are standard and realistic. 68 3.2 Figures Fig. 2.1. Good entrepreneurs' Dg and Their Bargaining Share 3 Fig. 2.2. Entrepreneurs' Indifference Curves and Equilibrium Debt Contracts Fig. 2.3. Good entrepreneurs' Optimal Effort and Their Bargaining Share 3. 69 g ' (g \p* ) g * 0 p* 1 Fig. 2.4. The Balanced Growth Rate and Good Entrepreneurs' Bargaining Share 3. Fig. 2.5: Optimal Debt Contract between Entrepreneur and Financial Intermediary(IC steeper than iso-profit curve 3 at point A) 70 3.3 References [I] Aghion Philippe, Nick Bloom, Richard Blundell, Rachel Griffith, and Peter Howitt. 2005. "Competition and Innovation: A n Inverted-U Relationship," Quarterly Journal of Economics 120 (May): 701-728. [2] Aghion, Philippe, and Peter Howitt. 1992. " A Model of Growth through Creative Destruc-tion." Econometrica 60 (March): 323-51. [3] . 1998. Endogenous Growth Theory, Cambridge: M I T Press. [4] Aghion, Philippe, and Jean Tirole. 1994. "The Management of Innovation." Quarterly Jour-nal of Economics 109 (November): 1185-1209. [5] Barro, Robert J . , and Xavier Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hi l l . [6] Beaudry, Paul, and Michael Poitevin. 1993. "Signaling and Renegotiation in Contractual Relationships." Econometrica 61 (July): 745-82. [7] Bernanke, Ben S., and Gertler, Mark, 1989. "Agency Costs, Net Worth, and Business Fluc-tuations." American Economic Review 79 (March): 14-31. [8] Berthelemy, Jean-Claude, and Aristomene Varoudakis. 1996. "Economic Growth, Conver-gence Clubs, and the Role of Financial Development." Oxford Economic Papers 48 (April): 300-28. [9] Borensztein, Eduardo R., Jose De Gregorio, and Jong-Wha Lee. 1998. "How does Foreign Direct Investment Affect Economic Growth?" Journal of International Economics 45 (June): 115-35. [10] Grossman, Gene M . , and Elhanan Helpman. 1991. Innovation and Growth in the Global Economy. Cambridge: M I T Press. [II] He, Qichun. 2006. "Gradual Financial Reform, Sectoral Allocation of Credit and Economic Growth: Evidence from China." University of British Columbia Working Paper. [12] Klenow, Peter J . , and Andres Rodriguez-Clare. 1997. "The Neoclassical Revival in Growth Economies: Has it Gone too Far?" NBER Macroeconomics Annual 12: 73-103. Cambridge, Mass.: M I T Press. [13] King, Robert G. , and Ross Levine. 1993. "Finance, Entrepreneurship and Growth: Theory and Evidence." Journal of Monetary Economics 32: 513-42. 71 [14] Lewis, Wil l iam A . 1955. The Theory of Economic Growth, London: George Allen & Unwin Ltd . [15] Mankiw, Gregory N . , David Romer, and David N . Weil. 1992. " A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107 (May): 407-37. [16] Mas-Colell, Andreu, Michael D. Whinston, and Jerry Green. 1995. Microeconomic Theory. New York: Oxford University Press. [17] Mookherjee, Dilip, and Debraj Ray. 2002. "Contractual Structure and Wealth Accumula-tion." American Economic Review 92 (September): 818-49. [18] Paulson, Anna L. , Robert M . Townsend, and Alexander Karaivanov. 2006. "Distinguishing Limited Liability from Moral Hazard in a Model of Entrepreneurship." Journal of Political Economy 114 (February): 100-44. [19] Romer, Paul M . 1990. "Endogenous Technological Change." Journal of Political Economy 98 (October): S71-102 [20] Schumpeter, Joseph A . 1961. The Theory of Economic Development. New York: Oxford University Press. [21] Shell, Kar l . 1967. " A Model of Inventive Activity and Capital Accumulation." In K . Shell (ed.), Essays on the Theory of Optimal Economic Growth. Cambridge: M I T Press. [22] Sheshinski, Eitan. 1967. "Optimal Accumulation with Learning by Doing." In K . Shell (ed.), Essays on the Theory of Optimal Economic Growth. Cambridge: M I T Press. [23] Solow, Robert. 1956. "A Contribution to the Theory of Economic Growth." Quarterly Journal of Economics 70 (February): 65-94. [24] Stiglitz, Joseph E. , and Andrew Weiss. 1981. "Credit Rationing in Markets with Imperfect Information." American Economic Review 71 (June): 393-410. [25] Townsend, Robert M . 1979. "Optimal Contracts and Competitive Markets with Costly State Verification." Journal of Economic Theory 21 (October): 265-93. [26] Wang, Cheng, and Stephen D. Williamson. 1998. "Debt Contracts and Financial Interme-diation with Costly Screening." Canadian Journal of Economics 31 (August): 573-95. [27] Williamson, Stephen D. 1987. "Financial Intermediation, Business Failures, and Real Busi-ness Cycles." Journal of Political Economy 95 (December): 1196-216. [28] Young, Alwyn. 2003. "Gold into Base Metals: Productivity Growth in the People's Republic of China during the Reform Period." Journal of Political Economy 111 (December): 1220-61. 72 [29] Big Economic Events Since China's Reform and Opening-up (1978-1998) [Zhongguo Gaige Kaifang Yilai Dashiji (1978-1998)]. Institute of Economic Research, the China Academy of Social Sciences. 2000. The Economic Science Publishing. [30] China Statistical Yearbook [Zhongguo Tongji Nianjian]. Beijing: China Statistical Press 1978-1999 (annual). [31] Statistical Yearbook on China's Cities ]Zhongguo Chengshi Tongji Nianjian]. Beijing: China Construction and Consulting and Service Center on Chinese Statistical Information 1986-1989 (annual). [32] Statistical Yearbook on China's Cities [Zhongguo Chengshi Tongji Nianjian]. Beijing: China Statistics 1990-1999 (annual). 73 4 FARMLAND CURSE IN INDUSTRIALIZATION: EVIDENCE FROM CHINA 7 8 4.1 Farmland Curse in Industrialization: Evidence from China 4.1.1 Introduction Is agricultural resource abundance a blessing for industrialization? 7 9 This question, first of all, generates controversy in the development literature. One view is that higher agricultural productivity negatively affects industrialization. For instance, Matsuyama (1992) shows that higher agricultural productivity lowers growth in small open economies. Another view argues that agricultural advantage is good for industrialization (e.g. Rostow 1960 and Nurkse 1953). The third shows that the answer to the question is not unique (e.g. Yanagawa 1996). Secondly, this question has not been well addressed in the resource-curse literature. There is a large literature which examines whether resource abundance is a curse (e.g. Sachs and Warner 1995, Gylfason 2001, Atkinson and Hamilton 2003). However, most papers conclude that resource-curse is driven by mineral resources rather than by agricultural resources. The works on agricultural resources have shown that agricultural resources are insignificant for either growth or income level (e.g. Papyrakis and Gerlach 2004, Ng 2006 8 0), however, most of those works fail to correct for the potential endogeneity problem of agricultural resource indicators. Attempting to contribute to solving the controversy in the development literature, this paper employs Chinese panel data to examine whether initial provincial agricultural resource abun-dance is good for provincial economic growth. Firstly, cross-province analyses within China are advantageous over cross-country regressions (see Papyrakis and Gerlach 2004). The results with China have wide implications for developing countries with a large agricultural sector. Secondly, unlike previous agriculture-resource literature, 8 1 this paper is the first to construct quality-adjusted farmland per capita as the measure of agricultural resource abundance. Thirdly, the indicator of quality-adjusted farmland per capita makes it possible to correct its potential endo-geneity problem by using different weather indicators as instruments. The instrumental variable method makes it credible to examine whether there is a causal relationship between agricultural resource endowment and economic growth. 7 8 A version of this chapter has been submitted for reviewing. 7 0 Agricultural resource abundance, in this paper, is a measure of higher agricultural productivity or agricultural advantage. 8 0 Some works differentiate between agricultural resource dependence and abundance (e.g. Ng 2006), while others do not (e.g. Papyrakis and Gerlach 2004). 8 1 In previous literature, agricultural resource abundance is measured as the value-added of agricultural sectors (farming, forestry, husbandry and fishery), while agricultural resource dependence is measured as value-added of agricultural sectors divided by GDP (e.g. Ng 2006). 74 Lastly, using weather as instruments connects this paper with the geography/climate and growth literature. Climate as part of geographical characteristics can affect economic de-velopment, but there is a debate over the exact channel of causality. Some show that cli-mate/geography directly affects economic development (e.g. Sachs et al. 1998). Others fo-cus on the channel and have identified it as past events/institution (e.g. Acemoglu, Johnson, and Robinson 2001, and Sokoloff and Engerman 2000). 8 2 This paper contributes by exploring whether agricultural resource abundance is the main channel by which weather affects growth in China. The main empirical findings of the paper are as follows. First of all, estimated by TSLS regressions using different weather indicators as instruments, initial quality-adjusted farmland per capita exerts a significant negative causal effect on subsequent growth rate of real G D P per worker for the Chinese provinces at the 5% level. That is, abundance of quality-adjusted farmland per capita is a curse for real G D P per worker growth. Secondly, Sagan test accepts the null hypothesis that quality-adjusted farmland per capita is the main channel by which weather indicators have significant effects on growth at the 5% level. Lastly, this paper shows that total factor productivity is the dominant channel by which farmland endowment lowers growth rates. This paper starts off by using a theoretical model to illustrate why agricultural resource abundance can be detrimental for industrialization. The engine of growth is assumed to be learning-by-doing in manufacturing. The model argues the following. Ceteris paribus, if a small open economy initially has higher farmland per capita, then it will have comparative advantage in agriculture thus resulting in a smaller share of labor employed by manufacturing than the rest of the world. Moreover, the degree of comparative advantage of the small open economy increases over time thus resulting in its manufacturing employment share to drop gradually. Given the engine of growth is learning-by-doing in manufacturing, the growth rate of the economy will be always lower compared with those economies that have lower initial farmland per capita. The empirical part of this paper employs Chinese panel data to examine whether agricultural resource abundance has negative effects on growth. Firstly, quality-adjusted farmland per capita is constructed as measures of agricultural resource abundance. 8 3 Two measures of farmland are employed, namely, farmland which is the area of farmland and agriculture land which sums over farmland, forestry land, pasture land, and fishery waters using unit value-added as weight. Quality-adjusted farmland is measured as the area of farmland multiplied by the quality of 82Some works show that weather causes institution conflict (e.g. Miguel et al. 2003). Some argue that institutional conflict hurts growth (e.g. Easterly and Levine 1997 and Rodrik 1998). Together, they imply that climate affects growth through institutional conflict. 8 3 See Table 1. First of all, quality-adjusted farmland per capita takes count of the quality of farmland. Farmland per capita without adjusting for the quality of land is subjected to measurement and interpre-tation problem, which is discussed in section 4. Secondly, quality-adjusted farmland per capita is not total land of a province divided by population. Sachs and Warner (1995) use total land area per person as an indicator of resource abundance without correcting its potential endogeneity problem. My robustness check show that there is no total land-abundance curse. 75 farmland. The quality of farmland is measured as the average grain yield per hectare. Secondly, the quality-adjusted farmland per capita may be endogenous to the growth process caused by omitted variable bias. The attractiveness of Chinese panel data is that I can find weather indicators such as long-run yearly average rainfall, temperature, temperature variation, and hours of sunshine as instruments for quality-adjusted farmland per capita. Sachs and Warner (2001) deal with the omitted variable bias of mineral resource-abundance by including geography/climate variable directly in the cross-country regression. It makes sense in the case of mineral resources because it takes millions of years to form them so that they are not affected by current climate variables. However, weather indicators are plausible instruments for quality-adjusted farmland per capita for the following reasons. Firstly, weather indicators are exogenous to the growth process. Secondly, weather indicators affect quality-adjusted farmland per capita. 8 5 China's agriculture sector lacks agricultural infrastructure (e.g. complete irrigation system), so agricultural production is partially rain-fed and affected by temperature. Different combinations of rainfall, temperature and sunshine not only determine whether a piece of land is used as farmland, pasture land or forestry land, but they also affect the quality of land. Statisti-cally, weather indicators are significantly correlated with quality-adjusted farmland per capita. Thirdly, whether weather indicators affect growth only through quality-adjusted farmland per capita can be examined by Sargan test. Specifically, the empirical analysis employs data for China's 27 provinces from 1981 to 1998 to examine whether initial provincial quality-adjusted farmland per capita affects subsequent growth of provincial real G D P per worker. The results are as follows. Based on OLS regressions, initial quality-adjusted farmland per capita has negative but insignificant effects on growth at the 10% level. When long-term yearly average rainfall, temperature and its variance, and hours of sunshine are employed as instruments, the TSLS regressions show that the exogenous weather component of initial quality-adjusted farmland per capita has significantly negative effects on real G D P per worker growth at the 5% level. Based on the instrumental variable approach, the direction of causality goes from quality-adjusted farmland per capita to economic growth. Higher initial quality-adjusted farmland per capita causes lower growth rates for Chinese provinces, that is, quality-adjusted farmland abundance is a curse. Based on Sargan test, initial quality-adjusted farmland per capita cannot by be rejected as the only channel by which weather indicators significantly affect growth. The quality-adjusted farmland abundance curse holds true even after controlling for initial G D P per worker, investment rate, human capital, labor force growth, fiscal expenditure, export, FDI , and population density (the inverse of total land per capita). 8 6 8 4 Results with indicators in previous literature are presented and discussed in section 4. 8 5 Weather is an important factor for agricultural production even for Israel which has an advanced agricultural sector (see Fishelson 1974). 8 6 For comparison, this paper, based on OLS regressions, also shows the following. The ratio of initial value-added of agricultural sectors to GDP has no significant relationship with subsequent income level or growth rates. Initial value-added of agricultural sectors exerts a positive effect on subsequent income level and economic growth. 76 The estimated magnitudes of IV regressions are significant for the quality-adjusted farmland per capita. For example, for the two coastal provinces of Liaoning and Guangdong, controlling for other possible growth factors such as labor force growth, human and physical capital investment, fiscal expenditure, export, and FDI and time effects, and assuming that Liaoning could lower its initial quality-adjusted farmland per capita to that of Guangdong, then its average annual growth rate of real G D P per worker would have increased by 2.48% and its level of real G D P per worker could have increased by 55% for the 18 years of 1981-1998. The effect of quality-adjusted farmland-abundance on growth is decomposed into different transmission channels. 8 7 For China, the total factor productivity growth is the dominant channel through which farmland-abundance decreases growth. Human capital is the other important channel, however, farmland-abundance promotes the accumulation of human capital in China . 8 8 This growth-promoting effect is dominated by the growth-retarding effect of farmland-abundance that works by lowering total factor productivity. The paper is organized as follows. Section 4.1.2 describes and solves a simple theoretical model. Section 4.1.3 describes the data and defines the various measures of the variables. Section 4.1.4 examines whether farmland abundance is a curse and whether it is the sole channel through which weather indicators affect growth. Section 4.1.5 decomposes the effect of farmland-resource on growth into different transmission channels. Section 4.1.6 concludes. 4.1.2 A Simple Two-Sector Model of Endogenous Growth To provide a framework to ponder over the effect of agricultural resources on economic growth for the Chinese provinces, I employ a simple two-sector model of endogenous growth in a small open economy based on Matsuyama (1992) to include farmland as one input of agricultural production. 8 9 The reason to use a small open economy model is that section 3 of the paper will test the theoretical predictions of the model with the panel data of China in which the provinces are trading with each other. A representative province is a small open economy which consists of two sectors: agriculture and manufacturing. The size of population is constant and equal to L . Labor is assumed to be immobile across economies (provinces).9 0 Manufacturing employs labor only, while agriculture uses both labor and farmland for production. The production functions are as follows: 87See Papyrakis and Gerlach (2004) for the transmission channels in the US. 8 8 Sachs and Warner (1997) find little evidence that resource-abundance (share of primary products in GDP) is related to human or physical capital accumulation in their cross-country work. Papyrakis and Gerlach (2004) find that agricultural resource dependence is bad for human capital accumulation in USA. 8 9 For the original model, see Matsuyama (1992). Farmland per capita as the indicator of agricultural resource in this paper is equivalent to agricultural productivity in Matsuyama (1992). 9 0 It is worth noting that, there are also labor migration among the provinces in China, which may affect the per capita farmland measure. However, the indicators used are 1985 data. In 1985, the labor migration is negligible. 77 YtM = M t F { n t ) YtA = T 0 G ( l - n t ) (1) (2) where nt stands for the fraction of labor employed in manufacturing, To for per capi ta farmland endowment. I assume: F(0) = G(0) = 0; F ' , G ' > 0; F " , G " < 0. I assume per capi ta farmland endowment, To, is constant over t ime. Due to the learning-by-doing effects which are generated by and external to ind iv idua l manufacturing firms, manufacturing product iv i ty Mt is endogenous and evolves over t ime according t o 9 1 Mt = S Y t M . ( 3 ) Compet i t ion i n labor market imp l i e s 9 2 TG' [(1 - nt)\ = P t M t F ' (nt) (4) where Pt is the relative price of the manufacturing good. A l l consumers i n the economy have the same non-homothetic preferences which follows / • C O U = / e-<* [3ln (cf - -y) + In ( c f ) ] dt (5) Jo where stands for the consumption of agricultural good for each consumer, cf1 for consumption of manufacturing good, 7 for the subsistence level of consumption for each consumer , 9 3 and p for the t ime preference. I further assume T 0 G (1) > 7 L > 0. (6) Equa t ion (6) ascertains that the agricultural sector can produce more than the subsistence level of agricul tural goods. 4.1.2.1 The Equilibrium of the World Economy The whole country represents the world economy. The provinces, excluding the representative province, consist of the rest of the world. The rest of the world differs from the representative 9 1 As in Matsuyama (1992), in this model, the engine of growth is learning-by-doing in the manufac-turing sector. A n d the learning-by-doing does not spill over across provinces. 9 2 The labor market in this model is competitive as in Matsuyama (1992). However, as Matsuyama (1992) argued, the wage-gaps between manufacturing and agricultural and the labor migration from agriculture to manufacturing may be substantial in reality, they would not affect the predictions of the model much and are assumed away for simplicity. 9 : 1 The existence of 7 makes preferences are non-homothetic and the income elasticity of demand for the agricultural good is less than unitary. 78 province only in that their per capita farmland endowment is given by T*. The equilibrium of the world economy mimics that of the representative closed economy. I assume all consumers have more than the subsistence level of consumption for agricultural goods. Solving the consumer's problem in equation (5) and after aggregating gives Ct = iL*A-QPtC{*. (7) In the world economy, the goods market clearing condition is given by Cf - = YtA and C t M = YtM. Combining them with equations (4) and (7) yields G(l-nt)-0G>(l-nt)p^ = ^ . (8) A l l variables are exogenous except nt. It can be shown that equation (8) has a unique solution in (0.1) for nt- nt = n* = n (T*). The fraction of labor employed in manufacturing for the world is constant. According to equation (3), the world manufacturing output grows at a constant rate 6F (ra*). 4.1.2.2 Per capita Farmland Endowment and Economic Growth By taking the ratios of each side of equation (4) and that of the rest of world so as to remove the relative price of manufacturing good, equation (4) becomes F'(nt) _ TQM* F>(n*) G' [(1 - m)} T*MT G' [(1 - n*)\ where the variables with a star superscript denote the world. At the initial period, t = 0, ceteris paribus, it is obvious that n 0 | n* if and only if ^ g | | . (10) At the initial point in time, the rest of the world differs from the representative province only in their per capita farmland endowment. The productivity in manufacturing is assumed to be the same at the initial point in time, that is MQ = Mo. Hence, a country has comparative advantage in agriculture if it has higher per capita farmland endowment. In other words, a country (province) has a smaller share of labor employed by manufacturing than the rest of the world if it has higher per capita farmland endowment compared with the rest of the world. As in Matsuyama (1992), differentiating equation (9) gives G"[(l-nt)} F"[(l-nt)] G> [(1 - nt)] + F' [(1 - nt)] J nt = 6[F(n*)-F(nt)]. (11) Equation (11) shows that if nt > n*, then nt will rise over time; nt < n*, then nt will decrease over time. Therefore, equation (10) states that if a country (province) initially has higher 79 farmland per capita, then it w i l l have comparative advantage i n agriculture. Accord ing to equation (11), the dynamics of the model shows that the in i t i a l comparative advantage due to higher farmland per capi ta of the smal l open economy (province) w i l l intensify over time, and the manufacturing employment share of the economy (province) w i l l drop. G i v e n the engine of growth is learning-by-doing in manufacturing, the growth rate of the economy w i l l be even lower when compared wi th those economies that have lower in i t ia l per capi ta farmland. In other words, the productive agricultural sector due to higher in i t ia l farmland per capi ta squeezes out labor from manufacturing, so the economy w i l l de-industrialize over t ime (see M a t s u y a m a 1992). 4.1.3 T h e D a t a 4.1.3.1 Empirical Specifications Using da ta on the Chinese provinces from 1981 to 1998 to examine whether agricul tural resource is a blessing for economic growth, I uti l ize the following standard empir ical specification, based on the neoclassical growth models (see M a n k i w , Romer and W e i l 1992) : 9 4 g i t = 30 + P1 ln (Agricul ture Resource)^ + (Other Cont ro l Var iables)^ (12) where gu is growth rate of real G D P per worker for province i at period £ ; 9 5 (Agricul ture Resource)^ is the measure of provincial agricultural resource abundance, which is described and constructed i n 3.2; other control variables are those commonly viewed as important factors for growth, namely, in i t i a l real G D P per worker, labor force growth, human capi tal investment rate, fiscal expenditure, F D I and export to G D P ratio, and coastal dummy. This empir ical formulation is consistent w i t h cross-country (cross-region) regressions w i th condit ional convergence. In addi-t ion, the regression is attractive in that one can easily check the robustness of the results to other growth factors by including them as regressors i n equation (12). A specific formulation of equation (12) without t ime effects is as follows: git = 30 + 3Y l n (Agricul ture Resource)^ + / 3 2 l n ( y ) + 83 l n (School)it + B± l n (Fiscal)it + \ L J i,t-l + 86 In (Export)it + 37 l n (FDI)it + 88(Coast) + eu. (13) i n U InC 'n + g + 6)i M To examine whether agricultural resource is a blessing for income level, the empirical specification is similar with income level being the dependent variable (see Mankiw, Romer and Weil 1992). 9 5 This measure can deal with the concern raised by Young (2003) that China's economic growth is related to its changing participation rate of its labor force. As in Young (2003), the G D P deflator used to calculate real G D P may not be perfectly measured. However, measurement error of dependent variable has no effect on the results. 80 4.1.3.2 Measuring Agricultural Resources As mentioned in the introduction, this paper proposes quality-adjusted farmland per capita as a preferred measure of agricultural resource abundance,9 6 the reason is that quality-adjusted farmland per capita represents both agricultural resource endowment and geographical char-acteristics. Quality-adjusted farmland is measured as the area of farmland multiplied by the quality of farmland. Quality-adjusted farmland per capita is quality-adjusted farmland divided by total population. The data on the initial provincial area of the different types of agricultural land (farmland, forestry land, pasture land, and fishery waters) are from the China Academy of Sciences' Natural Resources Database ( C A S - N R D ) . 9 7 I use the provincial average grain yield per hectare in 1983 from the China Agriculture Yearbook 9 8 as the measure of the quality of farmland. 9 9 Given the data, this paper presents two measures of quality-adjusted farmland per capita (see Table 3.1). The first is denoted as Q F A R M L A N D . For each province, Q F A R M L A N D is calculated as the product of the area of its farmland in 1985 and its average grain yield per hectare in 1983 divided by its total population in 1985. The second is Q A G R - L A N D . For each province, Q A G R - L A N D is calculated as the product of the area of its agricultural land (which sums the areas of farmland, forestry land, pasture land, and fishery waters using their unit value-added as weight) in 1985 and its average grain yield per hectare in 1983 divided by its total population in 1985. I take six-year average of the Chinese cross-province time series data (from 1981 to 1998) to get rid of business cycle phenomena. I use the data on farmland per capita of year 1985 for all three sub-periods: 1981-1986, 1987-1992, and 1993-1998. " It is worth noting that, farmland per capita without adjusting the quality of farmland ( F A R M L A N D and A G R - L A N D in Table 3.1) may be subjected to potential measurement and interpretation problem. The difference between F A R M L A N D and Q F A R M L A N D will be dis-cussed further in section 4.1.4.2 which deal with the validity of weather indicators as instruments for F A R M L A N D and Q F A R M L A N D . The quality-adjusted farmland per capita indicators have explicit variations across-province. Figure 3.1 plots the provincial quality-adjusted farmland per capita indicator, Q F A R M L A N D , which shows huge provincial variations. Notice that the coastal provinces (those on the right hand side of figure 3.1) tend to have lower quality-adjusted farmland per capita. Southern provinces tend to have lower quality-adjusted farmland per capita. Since this paper uses quality adjusted farmland, the provincial variations in quality-adjusted farmland per capita partially 9 6 This paper believes that a measure of agricultural abundance is superior to one of agricultural dependence in answering whether agricultural resource is a blessing for industrialization. 9 7 The database has data in 1985 as the earliest. The data before 1985 axe not available. 9 8 The Chinese title is "Zhongguo Nong Ye Nian Jian". "Farmland can be used to produce economic crops. However, the average economic crop yield per hectare is not available in the China Agriculture Yearbook. Furthermore, the differences in average grain yield per hectare includes those from different uses of inputs such as fertilizer. Therefore, the grain yield per hectare is an imperfect measure of the quality of farmland. 81 come from variations in population density. This point will be discussed further in section 4.1.4.2. For comparison, this paper also presents the results with indicators used in previous literature (see Table 3.1). The data on the value-added of agriculture, forestry, husbandry, and fishery and their indexes are from the C A S - N R D . I calculate the real value-added of farming, forestry, husbandry, and fishery with the base year being 1978, the same year for real G D P . I add up the four real value-added to get the measure of agricultural resource abundance, denoted as A B U N D . Dividing A B U N D by real G D P gives the measure of agricultural resource dependence, D E P E N D . For A B U N D and D E P E N D , the values of 1981 are taken for the period 1981-1986, 1987 for 1987-1992, and 1993 for 1993-1998. This paper will argue that these two indicators are also subjected to potential endogeneity problem. Therefore, the OLS results are biased. However, it is hard to find valid instruments to deal with the potential endogeneity problem. China's agriculture is an important part of the whole economy. As in figure 3.2, from 1981 to 1998, for a non-coastal undeveloped province, Inner Mongolia, its agricultural real value-added consists of about 40% of real GDP. For Beijing, its agricultural real value-added is around 20% of real G D P . Figure 3.3 plots average annual growth rate of real G D P per worker against quality-adjusted farmland per capita, which suggests the higher quality-adjusted farmland per capita a province begins with, the lower growth rate of real G D P per worker it would have for the next 18 years. 4.1.3.3 Measuring Weather The quality-adjusted farmland per capita indicator may be endogenous due to measurement error and omitted variable bias. The attractiveness of Chinese Panel data is that I can utilize weather as instrument to deal with the endogeneity problem. The endogeneity of quality-adjusted farm-land per capita will be tested and the validity of weather as instrument will be discussed in section 4.1.4.2. Here, I attempt to find some weather indicators from the C A S - N R D . From the C A S - N R D , I use the provincial rainfall in 1997 and its difference from long-term yearly average to calculate the provincial long-term yearly average rainfall, denoted by R A I N -F A L L . The C A S - N R D also provides temperature and sunshine, data for the period of 1951-1980 1 0 0 for around 600 weather observatories (stations) across China. On average, each province has around 20 weather observatories scattering the province. Each of the 600 weather observa-tories has data on yearly and monthly temperature average, "temperature yearly difference" 1 0 1, and yearly average hours of sunshine. Altogether there are 600x15=9000 data points. Based on the data, I calculate the average temperature of the observatories within one province as the long-term yearly average temperature, denoted by T E M P E R . I calculate the 1 0 0Some data on temperature are between another year in 1950s and 1980. 1 0 1 "Temperature yearly difference" is the difference between the highest and lowest monthly average temperature, which measures the fluctuations of temperature. 82 average "temperature yearly difference" of the observatories within one province as the long-term average "temperature yearly difference", denoted by T E M P D I F F . I calculate the monthly average temperature variance for each weather observatory and then average over all the ob-servatories within one province to get the long-term yearly temperature variance, denoted as T E M P V A R . The last weather indicator is SUNSHINE, which is calculated as the yearly average hours of sunshine of the observatories within one province. The correlation among the weather indicators are listed in Table 3.2. According to Table 3.2, the correlations between the weather indicators are significant at the 1% level. Therefore, the weather indicators are significantly correlated with one another. 4.1.3.4 Measuring Al l Other Variables Real G D P per worker for province i is measured as its real G D P divided by its labor force (total number of workers). Real G D P is calculated with 1978 being the base year. In equation (13), In i s t n e initial real G D P per worker, which takes the value of the beginning year of each sub-period. 1 0 2 (School) is measure of human capital and is measured as secondary school enrollment to the total number of workers according to Mankiw Romer and Weil (1992) . 1 0 3 Secondary school enrollment is the sum of student enrollment for middle schools (grade 7 to 9) and high schools (grade 10 to 12). (Fiscal) is fiscal expenditure to G D P ratio. For labor force growth measure: ln(n + g + 6), I follow Klenow and Rodriguez-Clare (1997) to use 0.08 for (g + 6). Both FDI and Export are nominal values divided by nominal G D P . (y ) is the nominal physical capital investment rate. There are problems with the various deflators of China (see Young 2003). The Chinese local statistical bureau tends to downwardly report the deflators for investment relative to that of GDP, thus if one uses those deflators to measure real investment rate, some provinces would have unreasonably high real investment rate. 1 0 4 In this paper, I assume the deflators of investment and G D P grow at the same rate, 1 0 5 which likely produces a less severe problem for my paper. There are 31 provinces in C h i n a . 1 0 0 Before 1997, Chongqing was a city of Sichuan province. Both of them are excluded from the sample because Chongqing does not have data before 1997 and the data for Sichuan before 1997 contain those of Chongqing while those after 1997 do not. It is similar for Hainan and Guangdong. There are a complete set of data for Guangdong 1 0 2 T h e dependent variable is in percentage, to calculate speed of convergence, A, the estimated coefficient on In ( £ ) t l should be divided by 100. And for (13): = 1 0 3 The human capital measure that is mostly correlated with growth is the male secondary schooling by Baxro and Lee (1993). Later, this paper shows that the results on land per capita hold true with only initial G D P as the control variable. Thus, School is not a perfect measure of human capital but wil l not affect the results of the paper. 1 0 4 Weeks and Yao (2003) produced a particularly insignificant coefficient for real investment rate. 1 0 5 Some studies on Chinese provincial conditional convergence adopted this approach. 1 0 6 I n China, out of the 31 provincial governments, four are municipalities and four are autonomous regions. For simplicity, I delegate the usage 'province' to all. 83 province, thus it is kept while Hainan is dropped. Tibet is also excluded because there are many missing data on other control variables. I employ data from 1981 to 1998. Thus, in the sample, I have data on 27 provinces for 18 years (1981-1998). I take six-year averages for the Chinese panel data to avoid the influence from business cycle phenomena, which produces three sub-periods: 1981-1986, 1987-1992, 1993-1998. Each province has three data points. Thus, there are 27 provinces with three time periods, which gives 81 observations. The data are gathered from various sources. I have used provincial statistical yearbooks and Statistical Yearbook of China for the data on real G D P per worker, secondary school enrollment rate, fiscal expenditure, physical capital investment rate, labor force growth, export and F D I . 1 0 7 The panel data of China have been used by economists (e.g. Weeks and Yao 2003) to evaluate the conditional convergence of Chinese provinces. Weeks and Yao (2003) assumed different aggregate production functions for coastal provinces and interior ones and confirmed this with the Chinese data. The coastal dummy is included in this paper, though the results are not sensitive to this inclusion. Table 3.3 lists the average annual growth rate of real G D P per worker and initial farmland per capita. The first three data points are for Beijing, whose average annual growth rate of real G D P per worker is 6.0% for 1981-86, 5.0% for 1987-92, and 9.5% for 1993-98. Its corresponding initial endowment of quality-adjusted farmland per capita Q F A R M L A N D is in column 3. 4.1.4 Estimations and Results This section examines whether agricultural resource abundance significantly matters for eco-nomic growth using empirical specifications in (13). In this paper, the indicators of agricultural resource abundance are l n ( Q F A R M L A N D ) , l n ( Q A G R - L A N D ) , ln ( F A R M L A N D ) , and ln(AGR-L A N D ) . The difference between l n ( Q F A R M L A N D ) and l n ( F A R M L A N D ) and the associated measurement and interpretation problem will be discussed in 4.2.1. This paper mainly focuses on reporting the results with ln ( Q F A R M L A N D ) . The results with all the other three indicators are checked, and the results are very similar to those with l n ( Q F A R M L A N D ) . The results with the other three indicators are reported as much as possible, while the results for those unreported indicators are available upon request. 4.1.4.1 OLS Results The OLS results for ln ( Q F A R M L A N D ) , l n ( Q A G R - L A N D ) , and l n ( A G R - L A N D ) are presented in Table 3.4. The results show that the coefficients on l n ( Q F A R M L A N D ) , l n ( Q A G R - L A N D ) , and l n ( A G R - L A N D ) are insignificantly negative at the 10% level. The coefficients on In n+lQ+g) 7 Qinghai province doesn't have any F D I for 1981-1986, and the datum from 1987-1992 is used. 84 are positive and significant at the 1% level. The coefficients on ln(School) are positive as ex-pected, but they are significant at the 5% level only in some cases. The coefficients on In ( j ) t _ 1 are negative, which shows the existence of conditional convergence.1 0 8 However, the agricultural resource measures of farmland per capita may be endogenous for the following reasons. First of all, the data on farmland is from year 1985, which is not from year 1981 in which the data are not available. Secondly, there may be measurement error associated with farmland per capita. Thirdly, the most important reason is potential omitted variable bias. There may be important variables omitted from the regression. For the omitted variable bias, suppose that a correctly specified regression model would be: g = 30 + f3x l n ( Q F A R M L A N D ) + 62X + Other Control Variables (14) Then, if variable X is omitted from the regression, then the estimated coefficient on l n ( Q F A R M L A N D ) would be inconsistent in OLS regressions: -OLS t Cov (ln (QFARMLAND), X) P h m ^ = / ? 1 + Var (ln (AGR-LAND)) ( 1 5 ) Given the asymptotic property of the OLS regression, if Cov (ln (QFARMLAND), X) ^ 0, then the estimated coefficient from OLS regression will be biased. The endogeneity of l n ( Q F A R M L A N D ) will be tested by Hausman test in section 4.1.4.2, in which the endogeneity problem of ln ( Q F A R M L A N D ) is dealt with by the instrumental variable approach. 4.1.4.2 Endogeneity and Omitted Variable Bias of Farmland per capita To test and deal with the potential endogeneity problem of quality-adjusted farmland per capita due to omitted variable bias, I utilize the instrumental variable (IV) approach and locate instru-ments that affect growth only through affecting quality-adjusted farmland per capita. 4.1.4.2.1 The Validity of Weather Indicators as Instrumental Variables I use weather indicators as instruments for quality-adjusted farmland per capita to isolate the exogenous endowment effect to see whether this effect is significant for growth. Here I use four weather indicators: Rainfall, Temper, Tempdiff and Sunshine as instruments given that both Tempdiff and Tempvar measure the fluctuations of temperature. The results with Tempvar are reported in robustness check. Weather indicators are plausible instruments for quality-adjusted farmland per capita for the following reasons. First of all, the long-run weather indicators are exogenous to the growth process. Secondly, those weather indicators are significantly correlated with one another (see Table 3.2). Thirdly, weather indicators affect quality-adjusted farmland per capita for the Chinese provinces. China is not heavily industrialized and largely relies on agriculture (see figure 3.2). In many parts of 1 0 8 Weeks and Yao(2003) found that the speed of convergence is around 2% per year using G M M method. 85 China, agriculture has no extensive irrigation systems and agricultural production is partially rain-fed. Furthermore, in China, different combinations of long-run yearly average rainfall, temperature, temperature variation and hours of sunshine decide the types, quantity and quality of the land. Different combinations of rainfall, temperature and sunshine determine whether a piece of land is best to use as farmland, pasture land, or fishery waters. Better combinations of rainfall, temperature and sunshine make more land suitable for agricultural use and increase the quality of land for agricultural production. The correlation between the weather instruments and farmland per capita is shown in Table 3.5. Table 3.5 shows that weather indicators are significantly (at least at the 5% level) and highly correlated with l n ( Q F A R M L A N D ) , ln(QAGR-L A N D ) , and ln ( F A R M L A N D ) . Lastly, the validity of the IV strategy rests on the assumption that the instruments are uncorrelated with the omitted variables. If this is true, then the instruments can be used to isolate variation in the variable of interest that is uncorrelated with the omitted variables. The exogeneity of the instruments will be tested by Sargan test which will tell whether quality-adjusted farmland per capita is the only channel through which weather indicators affect growth. According to Table 3.5, the correlation between weather indicators Rainfall and Temper and l n ( Q F A R M L A N D ) is negative. The following elaborates on this. On the one hand, better weather conditions yield higher quantity and quality of farmland. In this paper, the quality of farmland is measured as average grain yield per hectare, denoted as Quality. The correlation between ln(Quality) and weather indicators in Table 3.5 confirms that higher level of rainfall and temperature is positively correlated with better quality of farmland. On the other hand, better weather conditions increase population density. Population density, denoted as P O P D E N , is total population of a province divided by its total area. According to Table 3.5, higher level of rainfall or temperature is significarftly correlated with higher level of population density. Due to normalization, l n ( Q F A R M L A N D ) is measured as total quality-adjusted farmland per capita. As weather gets better, if population increases more than the quantity and quality of farmland do, then Rainfall and l n ( Q F A R M L A N D ) are negatively correlated as illustrated in figure 3.4. Could it be possible that the increase in population is solely caused by the increase in the quantity and quality of farmland due to better weather? If it is true that the increase in population is proportional to that in the quantity and quality of farmland given better weather, l n ( Q F A R M L A N D ) which measures the quantity and quality of farmland per capita should have no correlation with weather indicators; l n ( F A R M L A N D ) would be negatively cor-related with weather indicators because it overlooks the increase in the quality of farmland. Then ln ( F A R M L A N D ) would not be a good measure of farmland abundance, instead it would measure the inverse of farmland quality. The data show that the correlation between Rainfall and Temper and l n ( Q F A R M L A N D ) is large and significantly negative (see Table 3.6). Therefore, it cannot be true that increase in population is solely caused by the increase in the quantity and quality of farmland due to better weather. This paper argues that weather affects population through non-agricultural channels 86 as well as the farmland channel. For instance, people are fed by agricultural output, but they also prefer to live in places with nicer weather because they enjoy utility directly from nicer weather. Furthermore, the fertility rate could be higher under nicer weather conditions. There-fore, weather indicators are negatively correlated with quality-adjusted farmland per capita, which is consistent with the fact in Table 3.5. This paper argues that this negative correlation comes from the fact that better weather conditions increase population more than they increase the quantity and quality of farmland. This is the reason why this paper constructs l n ( Q F A R M L A N D ) which takes count of the quality of farmland as the agricultural resource abundance measure. For l n ( F A R M L A N D ) , it is hard to differentiate whether the negative correlation between weather indicators and ln ( F A R M L A N D ) comes from the overlooking of farmland quality or the increase in population due to non-agricultural reasons. The two possibilities could lead to contradictory interpretations. However, l n ( Q F A R M L A N D ) is not subjected to this measurement and interpretation problem. Therefore, the following of the paper focuses on reporting the results with l n ( Q F A R M L A N D ) and l n ( Q A G R - L A N D ) . The results with l n ( F A R M L A N D ) and l n ( A G R - L A N D ) are reported in robustness check. Now, it is time to use the instruments to deal with the endogeneity problem of the farmland per capita indicators by the most efficient regression: TSLS. The instrumental variable method makes it credible to examine whether there is a causal relationship between natural endowment of agricultural resources and economic growth. 4.1.4.2.2 TSLS Regression The first and second stage results of the TSLS regressions are reported in Table 3.6, 1 and 1' displaying the results for l n ( Q F A R M L A N D ) and 2 and 2' containing those for l n ( Q A G R - L A N D ) . From the first-stage results for l n ( Q F A R M L A N D ) , the p-value of Hausman test is below 5%. Therefore, the null hypothesis of the exogeneity of l n ( Q F A R M L A N D ) can be rejected at the 5% level, which suggests the existence of omitted variable bias. The F-test statistics on the joint sig-nificance of the four instruments on l n ( Q F A R M L A N D ) is above 10, and the p-value of the F-test is below 1%, which evidences that the instruments have significant effects on l n ( Q F A R M L A N D ) . For l n ( Q A G R - L A N D ) , the null hypothesis of the exogeneity of ln ( F A R M L A N D ) can be rejected at the 10% level. The p-value of Hausman test for l n ( Q A G R - L A N D ) is only below 10% be-cause this paper uses average grain yield per hectare as the quality measure for A G R - L A N D . A G R - L A N D consists of four types of agricultural land, namely, farmland, forestry land, pasture land, and fishery waters. Therefore, The results on l n ( Q A G R - L A N D ) are compromised because the average grain yield per hectare may not be a good quality measure for all four types of agricultural land. The second-stage results for l n ( Q F A R M L A N D ) show that TSLS regression improves the results for other important control variables for the Chinese panel data, comparing with OLS 87 regression. The coefficients on ln In n + g + s ^ J , ln(School) are significant at the 5% level with expected sign. The coefficient on l n ( Q F A R M L A N D ) in TSLS regression with time effects is still negative but becomes significant the 5% level. The magnitude of the estimated coefficient on l n ( Q F A R M L A N D ) in TSLS regression is about twice as large as that of OLS regression. The results with l n ( Q A G R - L A N D ) are very similar. Based on the p-value of Sargan test, at the 10% level, we cannot reject the null hypothesis that quality-adjusted farmland per capita is the only channel through which weather indicators have significant effects on growth. This says that the instruments are not correlated with other omitted variables, which justifies the validity of weather indicators as instruments for endogenous quality-adjusted farmland per capita. These provide evidence that higher initial natural endowment of quality-adjusted farmland per capita is significantly bad for subsequent economic growth at the 5% level. The exoge-nous weather component of quality-adjusted farmland per capita causes lower economic growth. Farmland abundance (higher quality-adjusted farmland per capita) is a curse for economic growth. The estimated magnitude of IV regression is significant for indicator l n ( Q F A R M L A N D ) . For example, for the two coastal provinces of Liaoning and Guangdong, controlling for other possible growth factors such as labor force growth, human and physical capital investment, fiscal expenditure, export, FDI, and time effects, if Liaoning could lower its initial quality-adjusted farmland per capita, l n ( Q F A R M L A N D ) , to equal that of Guangdong, then its annual average growth rate of real G D P per worker would have increased by 2.48% for the 18 years of 1981-1998, and its level of real G D P per worker could have increased by 55% for the 18 years. 4.1.4.3 The Robustness Check of the TSLS Results The following conduct some robustness checks on the TSLS results one at a time. 4.1.4.3.1 Without Controlling for Time Effects First of all, the results are checked without controlling for fixed time effects. The results are reported in Table 3.7. From Table 3.7, the F-test statistics on the instruments in the first-stage regressions are well above 10 with p-values being smaller than 5%, which means the instruments are valid and significant ones. The Hausman tests suggest the endogeneity of both indicators at the 5% level. The coefficients on agricultural land-abundance indicators are still significant at the 5% level. Hence, time-effects do not affect the result that agricultural land-abundance is a curse for growth. 4.1.4.3.2 Invariance of Other Control Variables It is desirable to check if the results are affected by the inclusion or exclusion of other RHS control variables. The results with different sets of RHS control variables are shown in Table 3.8. 88 The results using only initial real G D P per worker as the control variable are reported in ( l ) 1 ; the results with initial real G D P per worker, investment rate, and labor force growth as other control variables are found in ( l ) 2 ; the results with initial real G D P per worker, investment rate, labor force growth, human capital, and fiscal expenditure as other control variables are found in ( l ) 3 ; the results with initial real G D P per worker, investment rate, labor force growth, human capital, fiscal expenditure, and coastal dummy as other control variables are found in ( l ) 4 ; finally, the results with all the control variables in ( l ) 4 plus population density in 1985 (same year as farmland per capita) are reported in ( l ) 5 . Obviously the TSLS results on quality-adjusted farmland per capita l n ( Q F A R M L A N D ) are invariant with the set of other control variables. The coefficients on l n ( Q F A R M L A N D ) have similar magnitudes and remain significant at the 5% level regardless of various control variables. The Sargan over-identification test statistics are below 10% level for ( l ) 1 and ( l ) 2 , suggesting weather may affect growth through other omitted variables. However, as more control variables are added, all the Sargan test statistics are above the 10% level, which says that we cannot reject the null hypothesis that weather indicators work on growth only through l n ( Q F A R M L A N D ) . 4.1.4.3.3 Is There a Total Land per capita Curse or Population Density Blessing? I have established the farmland-abundance curse result. It is natural to ask whether the result holds true if I use the indicator of total land area of a province divided by its population, denoted as total land per capita . 1 0 9 The provincial areas of the Chinese provinces are fixed, and total land per capita is more of a resource abundance indicator than of an agricultural resource abundance indicator. Furthermore, the reciprocal of total land per capita is population density. Therefore, a regression with ln (total land per capita) is the same as one with ln (population density), with the coefficient on ln(total land per capita) being the additive inverse of ln(population density). Given this, the regressions with population density allows us to examine two issues, namely, is there a total land per capita curse, and is population density rather than farmland per capita the channel by which weather affects growth as illustrated in figure 3.4. Three groups of regressions are conduct. In the first group, I replace farmland per capita by total land per capita and do the regressions again. The coefficients on ln(total land per capita) in both OLS and IV regressions are not significant at the 10% level. In the second group, I include both l n ( Q F A R M L A N D ) and ln (POPDEN) in TSLS regressions by using the four weather indicators as instruments. In the third group, I conduct TSLS regressions for l n ( Q F A R M L A N D ) by including ln (POPDEN) as another regressor (see ( l ) 5 in Table 3.8). For the last two groups, the coefficients on l n ( Q F A R M L A N D ) remain significant at the 5% level, while those on ln (POPDEN) are always insignificant at the 10% level. Controlling total land per capita does not change the results on farmland per capita. 1 0 9 Total land per capita has been used in previous works (e.g Sachs and Warner 1997 and Stijns 2000). 89 Therefore, in the case of China, the total area of a province divided by its population, or its reciprocal population density has no significant effects on economic growth. There is no total land abundance curse or population density blessing, while there is evidence of farmland abundance curse. The non-existence of total land per capita curse result makes sense because it is hard to believe that large countries such as Russia, USA, Canada, and Australia would grow slower than small ones such as Germany and Singapore because the former ones have higher country area per capita. In other words, we should not expect any "scale" effects of total land per capita or its reciprocal, population density. Furthermore, even if weather indicators are highly correlated with population density, l n (POPDEN) , in Table 3.5, the IV results show that population density per se cannot function as the channel by which weather has significant effects on growth. 4.1.4.3.4 Using Alternative Measures of Instruments Here the results are checked with different combination of instruments. Here I choose Rain-fall, Temper, Tempvar, and Sunshine as instruments. The results are reported in Table 3.9. Obviously, the results are very similar to those in Table 3.6. 4.1.4.3.5 Results with ln(FARMLAND) and ln(AGR-LAND) Given the discussion on the measurement problem of ln ( F A R M L A N D ) , the negative correla-tion between weather indicators and l n ( F A R M L A N D ) may come from both the overlooking of farmland quality and the increase in population due to non-agricultural reasons. That is why the magnitude of correlation between ln ( F A R M L A N D ) and weather is larger than that between l n ( Q F A R M L A N D ) and weather (see Table 3.5). The two possibilities could lead to contradic-tory interpretations. However, if the effect of weather on l n ( F A R M L A N D ) through the channel of increasing population due to non-agricultural reasons dominates that through the channel of increasing farmland quality, the results on l n ( F A R M L A N D ) and l n ( A G R - L A N D ) can still be used to examine whether farmland per capita is a curse for growth. The first and second stage results of the TSLS regressions are reported in Table 3.10, 1 and 1' displaying the results for l n ( A G R - L A N D ) and 2 and 2' containing the results for l n ( F A R M L A N D ) . The results are very similar to those on l n ( Q A G R - L A N D ) and l n ( Q F A R M L A N D ) reported in Table 3.6. The other robustness check is weather affects growth through farmland quality ln(Quality). The coefficients on ln(Quality) in OLS and IV regressions are insignificant at the 10% level, suggesting farmland quality per se cannot function as the channel weather affects growth and there is no farmland quality blessing. Last but not least, the TSLS results on farmland per capita are not affected by the measurement errors of the Chinese data. The measurement errors on dependent variable of annual average growth rate of real G D P per worker, discussed in Young (2003), can be absorbed in the disturbance of the regression and ignored (Greene 2000). The IV approach can also deal with the statistical measurement errors on farmland per capita. 90 4.1.4.4 Results with Indicators Used in Previous Literature for Compari-son Here the results with indicators in previous literature, A B U N D and D E P E N D , are presented in Table 3.11. Their effects on the level of real G D P per worker have similar patterns to those in Table 3.11. According to Table 3.11, OLS regressions produce a positive coefficient on In(ABUND), which is insignificant without time effects at the 10% level and significant at the 1% level with time effects.The coefficients on In(DEPEND) are always insignificant at the 10% level. These two indicators are also endogenous to the growth process and subjected to potential omitted variable bias, so the estimated coefficients from OLS regressions are biased and incon-sistent. Given that the OLS results are biased, the OLS regressions with these two indicators cannot answer whether agricultural resource abundance (dependence) is a blessing for growth. It is hard to find exogenous instruments to deal with the potential endogeneity problem of the two indicators. Only some of the weather indicators constructed in this paper are significantly correlated with these two indicators at the 1% level. Even by instrumenting these two indicators with the four weather indicators, IV regressions show that the coefficients on the two indica-tors are insignificant at the 5% level. Therefore, the potential endogeneity problem for the two indicators encountered in empirical works is hard to overcome. This paper argues that quality-adjusted farmland per capita is superior to A B U N D in mea-suring agricultural resource abundance. Furthermore, quality-adjusted farmland per capita makes it possible to use weather indicators to overcome its endogeneity problem, thus solving the direction of causality between agricultural resource abundance and growth. This paper does check the result with a new measure of agricultural resource abundance, the real value-added to agriculture ( A B U N D ) divided by agricultural employment, denoted as A B U N D 0 1 . A B U N D 0 1 measures the real output per agricultural worker. The OLS results on ln (ABUNDOl) shown in Table 3.11 show that the coefficients on ln(ABUNDOl) are significantly negative at the 5% level, which evidences that agricultural abundance measured as real agricultural output per agricultural worker is bad for growth. A B U N D 0 1 may also be endogenous, however, only some of the weather indicators are significantly correlated with A B U N D 0 1 . Therefore, its potential endogeneity problem cannot be solved. IV regressions which are not better than OLS regressions show that the coefficients on ln(ABUNDOl) are insignificant at the 10% level. 4.1.5 Decomposing the Effect of Farmland-abundance on Growth In section 4.1.4,1 have identified the significant negative effect of farmland abundance on growth even after controlling for possible traditional growth factors. The Sargan test cannot reject the null hypothesis that farmland abundance is the only channel through which weather indicators have significant effects on growth. This section attempts to identify the channel through which 91 farmland abundance works on growth, which is important for us to understand further the farmland abundance curse. In the literature, the civil conflict channel is deemed as one important channel for resource-abundance to negatively affect growth. For example, the cross-country analysis of Miguel, Satyanath, and Sergenti (2003) shows that weather component of growth cause civil conflict in Sub-Saharan African countries. For the provincial analysis within China, the standard factor accumulation and total factor productivity (TFP) rather than civil conflict are deemed as the potential channels through which resource-abundance works on growth. 4.1.5.1 Farmland-resources Work on Growth through TFP Growth 4.1.5.1.1 Building T F P Growth by Growth Accounting Since standard factor accumulation indicators are included as regressors, the significant coeffi-cients on farmland per capita in the TSLS regressions in section 4 imply that the agriculture resource abundance indicator works on growth through affecting T F P . The growth accounting literature offers an avenue to evaluate this implication. Following the standard growth ac-counting literature, I decompose growth into two parts: factor accumulation and total factor productivity growth (TFPG) by using the following growth accounting formula: 1 1 0 A Y K . . L 3 = _ - ^ _ ( l _ „ ) I (16) where A stands for T F P , and Y for total real GDP, and K for total real capital stock, L for labor force, and a for the exponent of capital in the production function, which is assumed to be g. The data on real G D P and labor force are already available as in section 3. If I have the data on capital stock, then the T F P G , ^ , can be calculated. The quality of the investment data of China and thus the quality of the data on capital stock of China are questioned most in the literature. However, as Chow (1993) argues, the Chinese data series are actually intrinsically consistent with one another. Chow (1993) and Chow and Lin (2003) have produced the total real capital stock of China for the period of 1952 to 1998. Based on their method and the data on the whole capital stock of China, I build the provincial capital stock according to the formula in Chow (1993): Ki,t+i = (1 - 8) (Ki>t - Landi) + Iijt+\ + Landi (17) where K is the non-land fixed capital plus inventory of province i, Land is the capital of land, Iitt+i is the real gross investment (including inventory investment) of province i , and 6 is the depreciation rate. Following Chow (1993), the depreciation rate equals 0.04. ""Human capital is not included as Pissarides argues (2000): "Human capital was not used in the calculation of TFP growth, the idea being that human capital contributes to T F P growth; i.e. that in a more general analysis human capital should be used to explain the path of T F P growth." 92 • Initial provincial real Capital Stock. Chow and L i n (2003) have produced the whole capital stock of China from 1952 to 1998. And my data sample begins from 1981. Thus, I use their real capital stock data in 1981 for China to get the initial capital stock of the provinces of China. Following Biggeri (1999), I assume the same capital-output ratio for all the provinces which equals the national one. Given this, using the total real provincial G D P as weight, I get the initial real capital stock of all the provinces of China. • Provincial value of Land. As in Chow (1993), the capital of land is assumed to be fixed here. Given land=720 for whole China, I use the provincial total agricultural land, the A G R - L A N D (see Table 1), as weight to calculate the provincial capital of land. • Provincial real grOSS investment. The provincial real gross investment is built following Chow and L in (2003). First, using G D P deflator to deflate the sum of nominal provincial consumption and provincial gross capital formation yields real domestic final expenditures. 1 1 1 Second, I use provincial general consumer price index (CPI) to turn nominal provincial consumption into real provincial consumption. 1 1 2 Last, subtracting real provincial consumption from real provincial domestic final expenditures gives the real provincial gross investment (including inventory investment). Now I can build the provincial capital stock according to equation (17). Then equation (16) produces the T F P G for the provinces. 4.1.5.1.2 TSLS Regressions with T F P G To confirm the results in section 4 that T F P is the channel through which farmland abun-dance works on growth, I regress the T F P G on l n ( Q F A R M L A N D ) with initial real G D P per worker, School, In n+g+6Fiscal, and fixed time effects being the control variables. The TSLS results show that the exogenous weather component of farmland abundance has signif-icantly negative effects on T F P G at the 5% level. The TSLS results of regression T F P G on l n ( Q F A R M L A N D ) with only fixed time effects are reported in Table 3.12, column 5-1, which shows that l n ( Q F A R M L A N D ) has a significantly negative effect on T F P G . 4.1.5.2 Other Channels and Relative Importance Papyrakis and Gerlach (2004) conduct a cross-state analysis for USA. When they control for other traditional factor accumulation variables such as investment, school, and openness, their agriculture resource indicators are no longer significant. They conclude that the factor accu-mulation variables are the intermediate channels through which agriculture resource works on 1 1 1 The provincial nominal consumption for 1981-83 cannot be found. I use the total national nominal consumption to get the provincial nominal consumption using 1984 provincial consumption as weight. 1 1 2 T h e provincial CPI of 1984 cannot be found. I use national general retail price index for all provinces. 93 growth. They have decomposed the effect of agricultural resource on growth into those compo-nents of investment, school, openness, R & D , and corruption. 1 1 3 As in Papyrakis and Gerlach (2004), I also examine whether farmland abundance works on growth through other channels such as investment and school. The TSLS results are reported in Table 3.12. The exogenous weather component of farmland abundance has significant and negative effects on Export and F D I . Exogenous weather component of farmland abundance has significantly positive effects on School and Fiscal, which is contrary to the results of Papyrakis and Gerlach (2004) and confirms that the results within the USA states are different from those within the provinces of China. The relative importance of the transmission channels are reported in Table 3.13. The al-gebra can be found in Papyrakis and Gerlach (2004). According to Table 3.13, T F P G is the dominant channel through which farmland abundance lowers growth for China. The traditional human capital indicator School is an important channel, but the sign of its effect is contrary to those found in Papyrakis and Gerlach (2004) for USA. The farmland-abundance indicator of ln ( Q F A R M L A N D ) is actually good for the accumulation of human capital. This is possible because higher quality-adjusted farmland per capita means higher agricultural resource abun-dance. Higher agricultural resource abundance makes farmer wealthier, and they will invest in their children's education. However, this growth-promoting effect is dominated by the growth-retarding effect of farmland-abundance in lowering total factor productivity, which generates the farmland abundance curse phenomenon in China. 4.1.5.3 Is the Theoretical Model Credible? Sections 4.1.5.1 and 4.1.5.2 confirm that farmland-abundance lowers growth mainly through T F P G . The theoretical model in section 4.1.2 can explain the farmland abundance curse. The theoretical model predicts that, a province with higher farmland per capita has comparative ad-vantage in agriculture. Its fraction of people employed by manufacturing is smaller, which results in smaller learning-by-doing in manufacturing and thus lower economic growth. Here, some of the model's assumptions are tested with the Chinese data. Here I test the following. Farmland per capita should be positively related to the fraction of people employed by agriculture. From the C A S - N R D , I get the data on the total employment, the agriculture employment of the 27 provinces in 1985 (given that farmland per capita data are 1985 data). The fraction of people employed by agriculture, denoted by Agr-labor, is measured as the total agricultural employment divided by total employment. The correlation between farmland per capita and Agr-labor and the OLS regression of Agr-labor on farmland per capita are found in Table 3.14. According to Table 3.14, farmland per capita is positively correlated with Agr-labor, which is significantly at the 1% level; the coefficient on farmland per capita in the OLS regression is 1 1 3 I n Papyrakis and Gerlach (2004), they regress the traditional factors on their resource-abundance indicator plus constant one at a time. Since the results with initial real G D P are the same, they mainly report the results without initial real G D P . I follow the same approach here. 94 significantly positive at the 1% level. This confirms that farmland per capita is positively related to the fraction of people employed by agriculture. Wi th Chinese data, this paper finds that farmland abundance is a curse for growth and farmland abundance adversely affects growth mainly through lowering T F P . To explain the findings, I rely on the theoretical model in section 4.1.2, some assumptions of which are supported by the Chinese data. However, further research is needed for evaluating the theoretical model and identifying whether there are better explanations for the farmland-curse phenomenon of China. 4.1.5.4 The Tension between the Theoretical Model and Empirics The theoretical model, which is based on Matsuyama (1992), is useful in explaining why higher farmland per capita brings lower economic growth. However, the theoretical model has difficulty in explaining the whole empirical facts with the Chinese panel data. The most important difficulty lies in explaining the positive effect of weather on economic growth. Let us take rainfall as an example. Wi th the Chinese data, rainfall per se has significant and positive effect on real G D P per worker growth. It is commonly accepted that rainfall is good for the quality and quantity of farmland. This means that higher level of rainfall brings higher farmland per capita. Based on the theoretical model, higher farmland per capita is bad for economic growth. Combining these two together yields that higher level of rainfall is bad for growth through increasing farmland per capita. This contradicts with the fact that rainfall has a significant and positive effect on economic growth. The theoretical model which is based on Matsuyama (1992) cannot reconcile the prediction of a negative effect of weather on growth with the stylized fact of China. The reason is straight-forward. In the theoretical model, population is fixed and does not respond to the change in weather. Given fixed population, better weather brings higher quality and quantity of farm-land, which gives higher farmland per capita. Therefore, weather has to have a negative effect on growth in the model. However, if according to Malthusian views, population responds to the change in weather only through the agricultural channel, then the theoretical model would predict zero correlation between weather and farmland per capita and no effect of weather on G D P growth. In either way, the theoretical model based on Matsuyama (1992) has difficulty in generating a positive effect of weather on growth. In reality, population is not fixed. Weather may affect population through an agricultural channel and non-agricultural ones. To explain the empirical regularities of the Chinese data, we need to relax the fixed population assumption and adopt a non-Malthusian view. For example, a theoretical model that takes into account the population's response to the change in weather such as fertility rates or migration would solve the tension between theory and empirics and help to explain the stylized facts and empirical regularities of the Chinese panel data. This needs future research exploration. 95 4.1.6 Conclusions and Lessons for Developing Countries In this paper, I examine the effect of agricultural resource abundance on economic growth for the Chinese provinces from 1981 to 1998 and find that higher quality-adjusted farmland per capita, which is an indicator of agricultural resource abundance, is a curse for industrialization. The result is robust even after controlling for traditional growth factors and time effects and correcting for potential omitted variable bias. These contribute to solving the debate over the role of agricultural advantage and agricultural resource abundance in affecting industrialization in the development and resource-curse literature. Empirical tests further document evidence that farmland-abundance is the only channel through which weather has significant effects on growth. This finding that geographical characteristics (weather indicators) affect growth through natural endowment (farmland abundance) in the absence of institutional conflict in China advances our understanding of the literature on natural resources, geography and growth. This paper also decomposes the effect of farmland abundance on growth into different trans-mission channels. Though farmland-resource abundance promotes the accumulation of human capital, this growth-promoting effect is dominated by the growth-retarding effect of farmland-abundance that works by lowering total factor productivity. A n intuitive explanation for the ob-served negative relationship between farmland abundance and growth is as follows. As farmland becomes more abundant for a small open economy where the source of growth is learning-by-doing in manufacturing, it will have comparative advantage in agriculture, which absorbs labor into agricultural sectors. Because there is lesser labor in manufacturing, there is lower learning-by-doing and the T F P growth in manufacturing retards. As a whole, the economy experiences lower growth rate. Some features of the theoretical explanation are supported by the empirical stylized fact that farmland per capita is positively related to the share of labor employed by agriculture. However, further work is needed to examine the theoretical explanation and to understand further the farmland abundance curse. The empirical findings have profound implications for developing economies. A common feature of many developing countries is that they usually have a large agricultural sector. There is little possibility that some farmland lies idle in developing countries because developing countries need to cultivate all possible farmland to feed the population as China did back in the 1980s. If there are provinces or states within those developing countries, those provinces with higher farmland per capita may witness slower industrialization given that farmland abundance is a curse. Moreover, developing countries usually specialize in agricultural production after trading with the rest of world. Hence trade may hurt the economic growth of those developing countries with rich farmland endowment that yields them comparative advantage in agriculture in the world trade system. This needs future research explorations. 96 4.2 Tables Table 3.1: Indicators of agricultural resources M y Indicators Calculating Formula Quantity and Quality Indicators Q F A R M L A N D = T ° t o T o \ a ? P o p u l a ^ o I n l l a n d x ( q u a l i t y o f farmland =grain yield per hectare) Q A G R - L A N D = A G R - L A N D x (quality of farmland) Where Quantity Indicators F A R M L A N D = A G R - L A N D = Total Area of Farmland Total Population Total Area of Farmland x ( l Total Population Indicators in previous literature (for comparison) Value-added of Forestry, Husbandry and Fishery > Value-added of Farming ' A B U N D D E P E N D =Real Value-added of Farming, Forestry, Husbandry, Fishery _ Real Value-added of Farming, Forestry, Husbandry, Fishery Real GDP Table 3.2: Correlation among the weather indicators Rainfall Temper Tempdiff Sunshine Tempvar Rainfall 1.0 Temper 0.88*** 1.0 Tempdiff -0.73*** -0.82*** 1.0 Sunshine -0.87*** -0.78*** 0.72*** 1.0 Tempvar -0.76*** -0.82*** 0.98*** 0.72*** 1.0 *** indicates significant at the 0.01 level 97 Table 3.3: Average annual growth rate and in i t i a l quality-adjusted farmland per capi ta Annual Annual Province Growth l n ( Q F A R M L A N D ) Province Growth l n ( Q F A R M L A N D ) Beijing 0 6.0 5.8 Shandong 0 7.2 6.6 Beijing c 5.0 5.8 Shandong 0 5.7 6.6 Beijing 0 9.5 5.8 Shandong 0 9.5 6.6 Tianj in c 5.6 5.7 Henan 5.9 6.5 Tianj in c 4.2 5.7 Henan 3.8 6.5 Tianj in c 12.0 5.7 Henan 7.8 6.5 Hebei c 6.2 6.6 Hubei 7.5 6.6 Hebei c 5.6 6.6 Hubei 4.6 6.6 Hebei c 9.5 6.6 Hubei 10.2 6.6 Shanxi 7.7 6.8 Hunan 5.4 6.6 Shanxi 3.5 6.8 Hunan 3.4 6.6 Shanxi 7.8 6.8 Hunan 7.6 6.6 Inner Mongolia 7.5 6.9 Guangdong 0 7.7 6.0 Inner Mongolia 4.6 6.9 Guangdong 0 8.9 6.0 Inner Mongolia 8.1 6.9 Guangdong 0 9.0 6.0 Liaoning c 6.0 7.0 Guangxi 0 3.6 6.4 Liaoning 0 4.3 7.0 Guangxi 0 5.2 6.4 Liaoning 0 8.2 7.0 Guangxi 0 6.9 6.4 Ji l in 4.2 7.6 Guizhou 6.5 6.6 J i l in 2.6 7.6 Guizhou 2.4 6.6 J i l in 10.3 7.6 Guizhou 5.2 6.6 Heilongjiang 2.9 7.3 Yunnan 6.1 6.6 Heilongjiang 3.7 7.3 Yunnan 5.1 6.6 Heilongjiang 4.9 7.3 Yunnan 6.8 6.6 Shanghai 0 6.3 5.5 Shaanxi 6.6 6.8 Shanghai 0 6.6 5.5 Shaanxi 4.3 6.8 Shanghai 0 11.7 5.5 Shaanxi 6.3 6.8 Jiangsu 0 7.9 6.7 Gansu 5.1 6.8 Jiangsu 0 7.9 6.7 Gansu 4.7 6.8 Jiangsu 0 11.0 6.7 Gansu 6.5 6.8 Zhejiang 0 8.2 6.3 Qinghai 6.5 6.6 Zhejiang 0 6.8 6.3 Qinghai 2.2 6.6 Zhejiang 0 11.0 6.3 Qinghai 5.8 6.6 Anhui 6.9 6.6 Ningxia 6.7 7.2 Anhui 2.0 6.6 Ningxia 3.2 7.2 Anhui 9.6 6.6 Ningxia 5.0 7.2 Fujian 0 6.0 6.2 Xinjiang 8.7 7.1 Fujian 0 6.9 6.2 Xinjiang 7.2 7.1 Fujian 0 10.7 6.2 Xinjiang 6.3 7.1 Jiangxi 6.0 6.4 Jiangxi 5.2 6.4 Jiangxi 6.5 6.4 Note: growth rates are in percentage; c stands for coastal provinces 98 Table 3.4: O L S regressions between growth rates and farmland per capita Dep. Var.: Aver. Annual Growth Rate of real G D P per worker Indep. Vari . 1 V 2 2' 3 3' -1.02 -0 .87 l n ( Q F A R M L A N D ) (0.62) (0.59) -0 .77 -0.70 l n ( Q A G R - L A N D ) (0.65) (0.62) -0.81 -0.88 l n ( A G R - L A N D ) (0.54) (0.53) -0 .87 -1.34** -0.68 -1.20* -0 .67 -1.17* (0.77) (0.64) (0.76) (0.63) (0.75) (0.62) l n (School) 2.29** 1.35 1.96* 1.06 2.10** 1.40 (0.99) (1.05) (1.00) (1.09) (1.03) (1.06) l n (rn+g+s) 3.54*** 3.27*** 3.41*** 3.24*** 3.65*** 3.24*** (1.09) (1.05) (1.10) (1.09) (1.10) (1.05) In(Fiscal) -1.95*** -1.44** -1.80** -1.32** -1.51** -0.95 (0.69) (0.59) (0.70) (0.60) (0.72) (0.66) m(Export) -0.66 0.39 -0.70 0.39 -0 .73 0.32 (0.51) (0.47) (0.51) (0.46) (0.51) (0.46) In(FDI) 0.19 0.05 0.19 0.07 0.17 0.01 (0.15) (0.16) (0.16) (0.17) (0.16) (0.17) 1.34* 0.69 1.53** 0.76 1.42** 0.78 Coast Dummy (0.69) (0.66) (0.68) (0.66) (0.68) (0.65) Time F E No Yes No Yes No Yes . R2 0.45 0.64 0.44 0.63 0.45 0.64 Observations: 81 81 81 81 81 81 ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 99 Table 3.5: Correlations l n ( Q F A R M L A N D ) l n ( Q A G R - L A N D ) Rainfall —0.45*** —0.42*** Temper —0.62*** —0.64*** Tempdiff 0.52*** 0.49*** Sunshine 0.30*** 0.28** l n ( F A R M L A N D ) ln(Quality) l n ( P O P D E N ) Q QIJ*** Q yQ*** Q _^g*** -0.76*** 0.62*** 0.62*** 0.57*** -0.39*** -0.26** 0.52*** -0.61*** -0.41*** ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level 100 Table 3.6: TSLS regressions between growth and farmland per capita Indep. Vaxi. First Stage Dep. Vari . as: Second Stage Dep. Vari . l n ( Q F A R M L A N D ) l n ( Q A G R - L A N D ) Average Growth Rate 1' 2' l n ( Q F A R M L A N D ) l n ( Q A G R - L A N D ) Sunshine Temper Tempdiff Rainfall I" In (School) In(Fiscal) In(Export) In(FDI) Coast Dummy Time F E F-Test on Instruments (P-value) Hausman Test (P-value) Sargan Test (P-Value) R2 Observations: -2.31** (0.95) ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 101 -2.06** (1.00) -0.0003** -0.0003** (0.0001) (0.0001) -0.04*** -0.06*** (0.01) (0.01) 0.03** 0.02 (0.01) (0.01) 0.00 0.0002 (0.0002) (0.0002) -0.65*** -0.47*** -1.63** -1.29* (0.12) (0.12) (0.68) (0.66) 0.44** 0.41** 2.89** 2.30* (0.18) (0.17) (1.32) (1.24) 0.001 -0.06 2.71** 2.52** (0.19) (0.19) (1.13) (1.19) -0 .02 0.10 -1.42** -1.08* (0.12) (0.11) (0.61) (0.64) 0.16* 0.12 0.38 0.36 (0.08) (0.08) (0.48) (0.48) -0.02 -0.02 -0 .05 -0 .03 (0.03) (0.03) (0.18) (0.18) -0 .07 0.04 0.57 0.76 (0.11) (0.11) (0.69) (0.68) Yes Yes Yes Yes 11.85 11.56 (0.00) (0.00) 0.04 0.07 0.62 0.29 0.79 0.76 0.61 0.61 81 81 81 81 Table 3.7: T S L S regressions between growth and farmland per capi ta Second-Stage Dep. Var.: Aver. Annual Growth Rate of Real G D P per worker Indep. Vari . 1' 2' -2.71*** l n ( Q F A R M L A N D ) (0.95) l n ( Q A G R - L A N D ) -2.26** (0.98) -1.34 (0.83) -0.88 (0.79) l n (School) 3.75*** 3.03** (1.29) (1.21) 3.66*** 3.31*** (1.15) (1.14) In(Fiscal) -2.11*** -1.70** (0.73) (0.72) In(Export) -0 .52 -0 .62 (0.54) (0.53) In(FDI) 0.21 0.21 (0.16) (0.16) Coast Dummy 0.58 1.01 (0.79) (0.75) Time F E No No F-Test on Instruments (P-value) 15.41 (0.00) 15.67 (0.00) Hausman Test (P-value) 0.01 0.03 Sargan Test (P-Value) 0.27 0.07 R2 0.40 0.40 Instruments: Rainfall, Temper, Tempdiff, Sunshine; Observations: 81. ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 102 Table 3.8: Invariance of other control variables i n T S L S regressions Second-Stage Dep. Vax.: Aver. Annual Growth Rate of Real G D P per worker Of ( i ) 2 ( i ) 3 ( i ) 4 ( i ) 5 l n ( Q F A R M L A N D ) -1.80*** (0.52) -1.83*** (0.51) -2.95*** (0.88) -2.42*** (0.90) 2 y-j^ *** (1.00) ln(Population Density) 0.41 (0.29) Time F E Yes Yes Yes Yes Yes Sargan Test:(P-value) 0.06 0.01 0.41 0.57 0.63 R2 0.50 0.52 0.55 0.60 0.59 Observations: 81 81 81 81 81 Instruments: Rainfall, Temper, Tempdiff, Sunshine ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) Table 3.9: T S L S regressions w i t h alternative weather indicators Second-Stage Dep. Var.: Aver. Annual Growth Rate of Real G D P per worker l n ( Q F A R M L A N D ) -2.61*** -2.21** (0.93) (0.92) Time F E No Yes Hausman Test (P-value) 0.01 0.04 Sargan Test (P-value) 0.34 0.56 R2 0.40 0.61 Observations: 81 81 Instruments: Rainfall, Temper, Tempvar, Sunshine ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 103 Table 3.10: TSLS regressions between growth and farmland per capita First Stage Dep. Vari . as: Second Stage Dep. Vari . : l n ( A G R - L A N D ) l n ( F A R M L A N D ) Average Growth Rate Indep. Vari . 1 2 1' 2' l n ( F A R M L A N D ) -1.54** (0.62) l n ( A G R - L A N D ) -1.73** (0.71) Temper -0.03* (0.01) -0.04*** (0.01) TempdifF 0.03*** (0.01) 0.02* (0.01) Sunshine 0.0000 (0.0001) 0.0001 (0.0001) Rainfall -0.0002 (0.0002) -0.0001 (0.0002) 0 70*** (0.11) -0.51*** (0.11) -1.40** (0.64) -1.18* (0.64) In(School) 0.30* (0.17) 0.27 (0.17) 2.38** (1.13) 2.34** (1.14) -0.17 (0.18) -0.23 (0.18) 3.18*** (1.05) 2.89*** (1.08) In(Fiscal) 0.30*** (0.11) 0.42*** (o.n) -0.83 (0.64) -0.47 (0.72) In(Export) 0.12 (0.08) 0.08 (0.07) 0.28 (0.47) 0.25 (0.47) In(FDI) -0.01 (0.03) -0.01 (0.03) -0.07 (0.18) -0.09 (0.18) Coast Dummy -0.21 (0.10) -0.10 (0.10) 0.66 (0.66) 0.80 (0.67) Time F E Yes Yes Yes Yes F-Test on Instruments 28.95 23.78 (P-value) (0.00) (0.00) Hausman Test (P-value) 0.08 0.06 Sargan Test (P-Value) 0.55 0.52 R2 0.90 0.89 0.63 0.63 Observations: 81 81 81 81 ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 104 Table 3.11: OLS regressions between growth rate and agricultural resources Dep. Var.: Aver. Annual Growth Rate of real G D P per worker Indep. Vari . 1 1' 2 2' 3 3' l n ( A B U N D ) l n ( D E P E N D ) ln (ABUNDOl) In(School) In(Fiscal) \n(Export) In(FDI) Coast Dummy Time F E R2 Observations: 0.48 0.94** (0.38) (0.37) -1 .22 -0 .42 (0.81) (0.71) -1.42** -1.51** (0.63) (0.60) -0.10 -0 .36 -1 .92 -1.60 0.44 -0 .32 (0.84) (0.69) (1.16) (0.98) (0.86) (0.73) 1.24 -0.26 2.14** 0.73 2.29** 1.07 (0.96) (0.84) (1.06) (0.97) (1.01) (0.85) 3.31*** 4.17*** 3.80*** 3.64*** 2.72*** 3.35*** (1.11) (1.02) (1.12) (1.04) (1.12) (1.02) -0.84 0.28 -2.06*** -1.51** -1.63** -1.39** (1.05) (0.89) (0.70) (0.60) (0.68) (0.58) -0.83 0.35 -0 .79 0.35 -0 .95 0.21 (0.52) (0.45) (0.51) (0.48) (0.51) (0.46) 0.05 0.03 0.28 0.14 0.19 0.20 (0.19) (0.16) (0.17) (0.16) (0.15) (0.16) 2.14*** 0.76 1.78*** 0.83 2.24*** 0.90 (0.69) (0.64) (0.64) (0.68) (0.65) (0.65) No Yes No Yes No Yes 0.44 0.66 0.45 0.63 0.47 0.65 81 81 81 81 81 81 •••Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 105 Table 3.12: Transmission channels, second stage results T F P G In(School) In(Fiscal) In(Export) In(FDI) 5-1 5-2 5-3 5-4 5-5 5-6 17.04*** -0.48 1.36** -0.15*** 8.52*** 17.32*** Constant (3.32) (0.49) (0.56) (0.82) (1.61) (3.06) -1.85*** 0.41*** 0.04 0.38*** -0.94*** -2.51*** l n ( Q F A R M L A N D ) (0.50) (0.07) (0.08) (0.12) (0.24) (0.46) Time F E Yes Yes Yes Yes Yes Yes R2 0.42 0.28 0.44 0.17 • 0.42 0.70 Obs: 81 81 81 81 81 81 Instruments: Temper, Tempdiff, Rainfall, Sunshine ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) Table 3.13: Relat ive importance of transmission channels Coefficient Coefficient Total Relative Transmission Channels (from Table 6) (from Table 12) Contribution Contribution T F P G 1 1 4 -2.31 129% In(School) 2.89 0.41 1.18 - 6 6 % ln(y^+s) 2.71 0.04 0.11 - 6 % In(Fiscal) -1.42 0.38 -0.54 30% ln(Export) 0.38 -0.94 -0.36 20% \n(FDI) -0 .05 -2.51 0.13 - 7 % Tota l -1.79 100% 1 1 4 I treat the coefficient on l n ( Q F A R M L A N D ) in Table 7 as the effect of l n ( Q F A R M L A N D ) on growth through the T F P channel. 106 Table 3.14: Farmland per capi ta and agricultural employment share Dep. Var.: ln(share of agricultural employment in total employment) Correlation OLS regression l n ( Q F A R M L A N D ) 0.48*** Constant 0.45*** (0.09) -3.56** (0.60) R2 0.23 Observations: 81 81 ***Significant at the 0.01 level, ** at the 0.05 level, * at the 0.10 level (Standard errors in parentheses) 107 4.3 Figures t j " T a n t o n Liaoning /' Shandong Jiangsu ^.Shanghai Zhejiang Fujian ? 'Taiwan (J Hainan • Farmland per capita in 1985 • 6.81 to 7.56 (6) • 6.63 to 6.81 (6) M 6.47 to 6.63 (7) 5.54 to 6.47 (8) • Fig. 3. 1 Farmland Resource Abundance i n China, 1985. 108 ^ — : TT ; — I :—n 1—; 5.5 6 6.5 7 7.5 In(QFARMLAND) Fig. 3.3: Average annual growth rate of real G D P per worker and farmland per capita. Farmland Quantity of Farmland: F Quality of Farmland: Q tr-Weather Indicator: W Farmland per capita: FQ P V A Population: P Figure 3.4: Negative correlation between weather indicators and farmland per capita 109 4.4 References [I] Acemoglu, Daron, Simon Johnson, and James A . Robinson. 2001. "The Colonial Origins of Comparative Development: an Empirical Investigation." American Economic Review 91 (December): 1369-401. [2] Atkinson, G . and K . Hamilton. 2003. "Savings, Growth and the Resource Curse Hypothesis." World Development 31, No. 11, 1793-807. [3] Barro, Robert J . , and Jong-Wha Lee. 1993. "International Comparison of Educational At-tainment." Journal of Monetary Economics 32 (December): 361-94. [4] Biggeri, Mario. 1999. "Determinants of Chinese Provincial Economic Growth in the Last Decade." Department of Economics, University of Florence Working Paper. [5] Chow, Gregory C. 1993. "Capital Formation and Economic Growth in China." Quarterly Journal of Economics 108 (August), 809-42. [6] Chow, Gregory C, and Anloh Lin . "Accounting for Economic Growth in Taiwan and Main-land China: A Comparative Analysis." EWP-GE/0306003. [7] Easterly, Will iam, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Divisions". Quarterly Journal of Economics 112(November): 1203-50. [8] Fishelson, Gideon. 1974. "Relative Shares of Labor and Capital in Agriculture: A Subarid Area Israel, 1952-1969." The Review of Economics and Statistics 56 (August): 348-52. [9] Greene, Wil l iam H . 2000. Econometric Analysis. New Jersey: Prentice Hall . [10] Gylfason, Thorvaldur. 2001. "Natural Resources, Education, and Economic Development." European Economic Review 45: 847-59. [II] Klenow, Peter J . , and Andres Rodriguez-Clare. 1997. "The Neoclassical Revival in Growth Economies: Has it Gone too Far?" NBER Macroeconomics Annual 12: 73-103. Cambridge, Mass.: M I T Press. [12] Mankiw, Gregory N . , David Romer, and David N . Weil. 1992. " A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107 (May): 407-37. [13] Matsuyama, Kiminori. 1992. "Agricultural Productivity, Comparative Advantage and Eco-nomic Growth." N B E R Working Paper 3606. 110 [14] Mehlum, Halvor, Kar l Moene, and Ragnar Torvik. 2006. "Institutions and the Resource Curse". Economic Journal 116 (January), 1-20. [15] Miguel, Edward, Shanker Satyanath, and Ernest Sergenti. 2003. "Economic Shocks and Civ i l Conflict: an Instrumental Variables Approach." University of California at Berkeley Working Paper. [16] Ng, Chi-Yung (Eric). 2006. "Is Natural Resource Abundance Beneficial or Detrimental to Output Level and Growth?" University of Western Ontario Working Paper. [17] Nurkse, Ragnar, 1953. Problems of Capital Formation in Underdeveloped Countries. Oxford University Press, Oxford. [18] Papyrakis, Elissaios, and Reyer Gerlach. 2004. "The Resource Curse Hypothesis and Its Transmission Channels." Journal of Comparative Economics 32: 181-93. [19] Pissarides, Christopher. 2000. "Labor Markets and Economic Growth in the M E N A Re-gion." London School of Economics Working Paper. [20] Rodrik, Dani. 1998. "Globalization, Social Conflict, and Economic Growth" (Prebisch Lec-ture), The World Economy 21(2), March. [21] Rostow, W.W. , 1960. The Stages of Economic Growth: A Non-communist Manifesto. Cam-bridge University Press, Cambridge. [22] Sachs, Jeffrey D., John Luke Gallup, and Andrew D. Mellinger. 1998. "Geography and Economic Development." in Pleskovic, Boris and Joseph E . Stiglitz (eds.), Annual World Bank Conference on Development Economics (April), The World Bank: Washington, D C . [23] Sachs, Jeffrey D., and Andrew Warner. 1995. "Natural Resource Abundance and Economic Growth." NBER Working Paper 5398. [24] Sachs, Jeffrey D. , and Andrew Warner. 1997. "Natural Resource Abundance and Economic Growth." Center for International Development and Harvard Institute for International De-velopment, Harvard University, Cambridge M A . [25] Sachs, Jeffrey D. , and Andrew Warner. 2001. "Natural Resources and Economic Develop-ment: The Curse of Natural Resources." European Economic Review 45: 827-38. [26] Sokoloff, Kenneth L . and Stanley L . Engerman. 2000. "Institutions, Factor Endowments, and Paths of Development in the New World." Journal of Economic Perspectives 14(3): 217-32. [27] Stijns, Jean-Philippe C. 2000. "Natural Resource Abundance and Economic Growth Revis-ited." University of California at Berkeley Working Paper. I l l [28] Weeks, Melvyn, and James Yudong Yao. 2003. "Provincial Conditional Income Convergence in China, 1953-1997: A Panel Data Approach." Econometric Reviews 22 (February): 59-77. [29] Yanagawa, Noriyuki. 1996. "Economic Development in a World with Many Countries." Journal of Development Economics 49: 271-88. [30] Young, Alwyn. 2003. "Gold into Base Metals: Productivity Growth in the People's Republic of China during the Reform Period." Journal of Political Economy 111 (December): 1220-61. [31] China Agricultural Yearbook [Zhongguo Nongye Nianjian]. Beijing: Agriculture Press 1984. [32] China Statistical Yearbook [Zhongguo Tongji Nianjian]. Beijing: China Statistical Press 1978-1999 (annual). [33] Natural Resources Database [China Academy of Social Sciences]. Beijing, China. 112 5 CONCLUDING CHAPTER This dissertation contributes to the finance-growth literature theoretically and empirically by taking theoretical model to the Chinese data. The Chinese data are unique in several aspects. First of all, China is a large developing country with around 30 provinces which have huge variations in terms of economic performance. This huge provincial difference allows us to look into some interesting empirical questions without worrying about the huge unobserved country characteristics, which may render the cross-country regressions undesirable, as the provincials are relatively more homogenous. Secondly, China has undertaken the opening and reform since 1978 in a gradual approach. This generates huge time variations for the provinces. The time variations could be changes in underlying economic, cultural, political and even legal institutions. This natural experiment provides a platform for us to deeply examine the role of economic freedom, economic institutions, and even geography in affecting the path and speed of economic growth. This thesis explores three aspects of the Chinese economy following its reform, namely, (1) the role of financial deregulation, which allows individuals or economic agents to make their own decisions, in raising the welfare of an agrarian economy as it moves from a central-planned economy into a free market one; (2) the role of economic institutions, which refers to the distri-bution of the bargaining power over innovation surplus among innovators and households who finance innovation; (3) the role of geography, as reflected by weather indicators and agricultural land endowment, in the process of industrialization. I summarize each chapter as follows. Does financial deregulation cause growth? What is the mechanism through which financial deregulation affects growth? The first, chapter uses the province-level financial deregulation experience of China from 1981 to 1998 to study these questions and finds the following. First, sectoral allocation of credit (measure of financial development), instrumented by the gradually implemented financial reform policies, causes growth. Specifically, real G D P per work growth rates are higher for provinces with faster pace of reallocating loans from state-owned industries to state-owned commerce. Second, sectoral allocation of credit, rather than saving and investment rate, is the only significant channel through which financial deregulation affects growth. The results are confirmed with growth accounting regressions and are robust after controlling for standard neoclassical growth factors, and time and province effects. The results of current study have important policy implications for developing countries. First and foremost, the empirical findings suggest that it may be productive for developing economies to focus their reform effort more on S A C than on saving and investment. Secondly, the results of this study make one ponder over a fundamental question in development economics: should developing countries favor the allocation of financial credit and resources to certain sectors over others. The Chinese experience shows that favoring industry too much may yield inefficient S A C , and the gradual approach to introduce market forces is a good choice to eliminate existing inefficient S A C , even within traditional sectors that are generally state-owned ones. The results show that the growth promoting effect of financial deregulation mainly comes 113 from the credit reallocation among state-owned sectors, specifically from state-owned commerce to state-owned industries. The institutional background of China helps to eliminate the suspicion that all the state-owned sectors are inefficient, therefore, allocation of credit to any sectors is highly inefficient. Before the economic reform, the Chinese government believed in rapid industrialization as the path to economic prosperity and favored state-owned industries instead of state-owned commerce and agriculture in credit allocation, which is common in many socialist countries as discussed in Shirk (2003, ch. 1). Therefore, the credit to and thus the growth of other sectors are relatively more suppressed comparing to state-owned industries. Given the existing distortions on credit allocation due to government intervention, it comes as no surprise that gradual financial deregulation which aims at introducing market forces in credit allocation would generate that more credit would be allocated to those sectors other than state-owned industries for a relatively more balanced sectoral growth and a more healthy economy. Although there are problems and challenges with the gradual financial deregulation experience of China (see Naughton 1998 and Lardy 1998), the financial deregulation paid off for China. Lastly, though financial deregulation of China decreases the share of credit to SOIs, it is a challenging task given the soft budget-constraint (SBC) problem of SOIs (see Kornai Maskin and Roland 2003). The political logic of the financial reform of China studied by Shirk (2003) gave us some idea why the financial deregulation was able to promote growth through S A C . Furthermore, the Chinese government achieved the credit reallocation out of SOIs partly by credit expansion through the banking system to get around the SBC problem. The mechanism of such a policy requires future research. The second chapter studies the 'capital-lord-entrepreneur' problem in an endogenous growth model. There are two representative agents a household (the capital lord) financing R & D and an entrepreneur-inventor performing R & D . How do the contractual provisions on how house-holds and entrepreneurs share property rights on inventions when asymmetric information exists affect growth? The credit contract giving entrepreneurs a higher share of monopolistic profit from inventions (i.e. entrepreneur's inventive incentive (EII) ) elicits more entrepreneurs' effort, generating a "bigger cake", but it also decreases the share of households, resulting in a "house-hold's dissaving" effect. To ensure bounded growth, entrepreneur's effort increases as her share increases but at a decreasing rate, so the "bigger cake" effect is decreasing. The "household's dissaving" effect is increasing because the one additional share of cake given by households is bigger given effort is increasing. At the beginning, the "bigger cake" effect dominates, but be-yond a point, "household's dissaving" effect dominates. Therefore the balanced growth rate is an inverted-U function of EII. The results are new and have never been previously documented in the literature. Not only does the model enrich the Schumpeterian approach to growth (Aghion and Howitt 1998), it also shows that entrepreneurs' bargaining share is an important determinant of long-run growth. While the inclusion of informational and contractual frictions complicates the model, I show that they are the necessary and sufficient conditions in establishing the inverted-U result for 114 balanced growth rate. Most importantly, these assumptions are standard and realistic. Whether agricultural resource abundance is a blessing for industrialization has not been well addressed in the resource-curse literature. The third chapter examines the effect of agricultural resource abundance on economic growth for the Chinese provinces from 1981 to 1998 and find that higher quality-adjusted farmland per capita, which is an indicator of agricultural resource abundance, is a curse for industrialization. The result is robust even after controlling for tradi-tional growth factors and time effects and correcting for potential omitted variable bias. These contribute to solving the debate over the role of agricultural advantage and agricultural resource abundance in affecting industrialization in the development and resource-curse literature. Em-pirical tests further document evidence that farmland-abundance is the only channel through which weather has significant effects on growth. This finding that geographical characteristics (weather indicators) affect growth through natural endowment (farmland abundance) in the ab-sence of institutional conflict in China advances our understanding of the literature on natural resources, geography and growth. This paper also decomposes the effect of farmland abundance on growth into different trans-mission channels. Though farmland-resource abundance promotes the accumulation of human capital, this growth-promoting effect is dominated by the growth-retarding effect of farmland-abundance that works by lowering total factor productivity. A n intuitive explanation for the ob-served negative relationship between farmland abundance and growth is as follows. As farmland becomes more abundant for a small open economy where the source of growth is learning-by-doing in manufacturing, it will have comparative advantage in agriculture, which absorbs labor into agricultural sectors. Because there is lesser labor in manufacturing, there is lower learning-by-doing and the T F P growth in manufacturing retards. As a whole, the economy experiences lower growth rate. Some features of the theoretical explanation are supported by the empirical stylized fact that farmland per capita is positively related to the share of labor employed by agriculture. However, further work is needed to examine the theoretical explanation which is based on Matsuyama (1992) and to understand further the farmland abundance curse. The empirical findings have profound implications for developing economies. A common feature of many developing countries is that they usually have a large agricultural sector. There is little possibility that some farmland lies idle in developing countries because developing countries need to cultivate all possible farmland to feed the population as China did back in the 1980s. If there are provinces or states within those developing countries, those provinces with higher farmland per capita may witness slower industrialization given that farmland abundance is a curse. Moreover, developing countries usually specialize in agricultural production after trading with the rest of world. Hence trade may hurt the economic growth of those developing countries with rich farmland endowment that yields them comparative advantage in agriculture in the world trade system. This needs future research explorations. 115 References [1] Aghion, Philippe, and Peter Howitt. 1998. Endogenous Growth Theory, Cambridge, Mass.: M I T Press. [2] Kornai, Janos, Eric Maskin, and Gerard Roland. 2003. "Understanding the Soft Budget Constraint." Journal of Economic Literature 41 (December): 1095-136. [3] Lin , Justin Yifu, and Zhiyun L i . 2005. "Policy Burden, Moral Hazard and Soft Budget Constraint." CCER Working Paper. [4] Matsuyama, Kiminori. 1992. "Agricultural Productivity, Comparative Advantage and Eco-nomic Growth." N B E R Working Paper 3606. 116 APPENDIX Appendix: Proofs Proof of Lemma 6 With ug (e,/3), Dg (e,3) and c* (e,0) known to good entrepreneurs, they choose their optimal effort e* by (see proof of proposition 1): MoxU(e,B)=ug{e,B)-q*(e,8). (A l ) I simplify q* (e,0) first. Combining equations (1) and (8), I get: ( fDb fD*9 \ r i j Fb (x) dx- j Fg (x, e) dx J - (1 - 0) [n (e) - ff (e)J + B*b - Rg (A2) Plugging equation (A2) into equation (9) to simplify c* (e, 0) as: ?* = / (x) dx-Fg (x, e)] dx^ - (1 - 3) [n (e) - ff (e)] + R*b - Rg (A3) Substituting ug (e,3) using equation (6) and <;* (e.B) in equation (A3) into equation ( A l ) , and using LT (e) = Ug + Rg, I get: Max :U(e,B)=U(e)-l^j " [Fb (x) dx-Fg (x, e)] dx^j - R*b Using 11(e) = I • Eg (^A,— h(e), and equation (5): Rt = I • Eb ^A^j —u,l get: [e +u Max U (e, 3) = / ^Eg (A, e) - Eb ( A ) - j f * [F6 (x) d x - F f l (x, e)] dx j - /i (e - [F6 (x) d x - F s (x, e)] dx j — /i (e) + « (A4) The latter equality using Eg (A, ej - Eb (^AJ — f'^ [Fb (x) dx—Fg (x, e)] dx. Now, the objective of good entrepreneur will be maximizing U (e, 0) by choosing e and taking into account the reaction of D* as a function of e as in equation (11). 117 Max U(e,B) = l(^j ^ [Fb (x) dx-Fg (x, e)] dx j - h ( ±. l(l)g- j f 9 F s (x, e) da^ = (1 - /3) n (e) - S (e e) + u + Ra F O C : ^ = I ^ £ - F g , e (x, e) - [F b (A,) -Fg (Dg, e)] ^ - ft' (c) From equation (11): d D g (,/p4 - F g , e e) dx - h> (e)) - fQD° -Fg,e (x, e) dx ~de~ ~ l-Fg(Dg,e) Given 8 = 0, then the solution corresponds to e = e, and D* = £)* with: rA (11) (A5) (A6) dUg de ?,/j=o = lijD, ~F9,e (x> e)dx\-h'{e) = 0 (A7) To see how the equilibrium values of e* and D* change with 8, I can proceed as follows. First, evaluating ^ in equation (A5) using equation (A6) at 8 = 0 and e produces: dU, de |e=e,/3=0 Fb {Dg) -Fg(Dg,e) l-Fg{Dg,e) -Fg<e (x, e) dx — h! (e) (A8) Given an infinitely small change in 8, good entrepreneurs will choose a different e by max-imizing their utility subject to equation (11), that is, they will consider the change of Dg. From lemma 1, Dg is a decreasing function of 8, so the infinitely small change in 8 causes Dg to drop from Db. Given equation (A7), the decrease of Dg will cause ^\e=e,/3=o > 0 since j'£ —Fgte (x, e) dx > — Fg>e (x, e) dx given Dg < D* and Fgfi (x, e) < 0. Now, good entrepre-neurs have to adjust their effort. Given U (e,8) is assumed to be concave in effort e if Fg (x,e) first-order-stochastically dominates Fb(x) and Fg (x,e) is very elastic in e, optimization given ^ | e _ g ^ = 0 > 0 requires good entrepreneurs to increase their effort from e to e*. This proves that optimal effort e* is an increasing function of 8: > 0. A direct way of showing this is as follows: total differentiating equation (A8) with respect to e and 8, rearranging, and I can show that > 0 ii Fg (x,e) first-order-stochastically dominates Fb (x) and Fg (x,e) is very elastic in e. Since U (e, 8) is assumed to concave in effort e, IT (e) is also concave in effort e, given 118 S=e , /3=0 > 0, then: c m , de \e=e,P=0 — ' i-A / -FBA-x, e) dx h'(e) > — | e = g ) / 3 = 0 > 0 which gives that II (e*) > 11(e). Then according to Nash Bargaining solution in equations (9) and (10), if II (e*) > 11(e), then R* > Rg, and U* > Ug. Thus, constraints [4] and [5] are satisfied. Given that all the constraints [1] to [5] are satisfied, a separating signaling equilibrium exists. Substituting equation (A6) into equation (A5), and simplifying, I get: d U [Fb (Dg) - Fg (Dg, e)] 3 J0A -Fgte (x, e) dx + (1 - Fb (Dg)) j £ -F9ie (x, e) dx de l-Fg(Dg,e) (1 -6) \Fb(Dg)-Fg(Dg,e)) l-Fg(Dg,e) ti (e) (A9) Now combining equation (A9) and equation (11) produces a two equations-two unknowns system, which delivers the equilibrium values of (e*,D*). With IT* =u and /_)* are known to bad entrepreneurs, they choose their optimal effort level according to Max U (e) =u — h (e), which generates e*, = 0. Q.E.D. Proof of Proposition 1 The model has two representative agents: a worker (household) and an entrepreneur. Here I prove that the growth rate of consumption for a worker is same as that for an entrepreneur. Then the proof of the existence of a balanced growth path on which the growth rate of different groups of people's consumption is the same as that of output and varieties will be the same as that in Barro and Sala-i-Martin (1995), which is omitted here. A typical worker maximizes the present discounted value of his consumption stream: Max • c Jo \ - e • s.t. C + (1 - (f>)rjN =wL + (l- <f>)rnN where (1 — </>) is the share of R & D cost covered by the savings of workers in financial inter-mediaries, which is fixed on a balanced growth path. On a balanced growth path, house-holds (workers) will finance (1 — 0 ) of the R & D cost through a financial intermediary, and receive (1 — (j)) of the profits earned by the financial intermediary. Solving Hamiltonian: H = g - p t C 1 ^ - ! _|_ _\ ^wL+(i -a) r ^ j V ~ c 7 ^ gives the growth rate of workers' consumption: 119 C I c = e{r-p) = (1 - A) Rt + A [Rg + (1 - 0) (n(e*) - fi(e))" 6»77 (A10) I assume the objective functions of entrepreneurs are the same: Max e. _pt[c-N-h(e)} dt s.t. fax / _ c + = A"7r (e) + r</ynN where <f> is the share of R & D cost covered by the savings of entrepreneurs in financial intermedi-aries. On a balanced growth path, entrepreneurs will finance (f> share of the R & D cost through a financial intermediary, and receive </> of the profits of financial intermediary. To solve for the effort level of entrepreneurs, it is only necessary to solve the period-by-period maximization problem of entrepreneurs. To confirm this, I substitute out c in the objective function with the budget constraint to get: Max / e-pty- ^ dt e,c J0 1-9 hr (e) — h(e) + rarjN — avN <S> Max / e~pt L N1-oo -1 e /„ 1-9 1 — dt Thus, the optimal effort e can be solved independently of consumption. The optimal effort can be solved by maximizing (7(e) = ir (e) — h(e). And this says that, when entrepreneurs make effort choices, they won't take into account their effort's effect on the interest rate of the financial intermediaries. Given this property, the optimal effort choice of good entrepreneurs will be positive at e*, and that of bad entrepreneurs will be zero as in lemma 6. After solving the optimal effort level, I can show that entrepreneur's problem is similar to that of a household (worker). Defining new variables c = c — N • h (e*), 7r (e*) = Nir (e*) — N • h (e*), I rewrite the entrepreneurs' problem as: Max / e~pt— —dt c Jo 1-9 s.t. c + (f>nN = Nn(e*) +r<jyqN Then solving the Hamiltonian will give us: | = ^ (r — p), which is the same as that for workers. Thus, c and N will also grow at the rate. Given that e* is constant, c = c + N • h (e*) and N will also grow at the rate. Thus, the model has a steady state and a balanced growth path on 120 which the consumption of workers and entrepreneurs, final output and N grow at the same rate (see Barro and Sala-i-Martin 1995). Proof of Proposition 2 First of all, if the conditions in lemma 6 are satisfied, then lemma 6 proves that there exists a separating signaling equilibrium, in which the balanced growth rate is given in equation (A10). Taking the derivative of balanced growth rate with respect to 3 produces: 33 6n II ( e * ) - 1 1 ( e ) +(1-3) dU (e*) de de dp where II (e*) = cEg (A, e * ) - h (e*) = I (A - Fg (x, e*) dx^j - h (e*). Taking derivative of ( A l l ) again with respect to 3, I have: ( A H ) dd2 dU (e*) de de dp d2U(e*) fde\2 dli (e*) d2e de2 \dj3) + de d32 (A12) Given that Ug = I ^A — D — f£ Fg (x,e*) dxj — h(e*) is concave in e, then II (e*) must be concave in e, since II (e*) is a the case with D = 0 for Ug. It is obvious that |g > ^ r l e = 0 since e is maximizing Ug. Given the concavity of TI(e*), and | | > 0 (from lemma 6), and ^ | (from assumption), I have < 0 (balanced growth rate is an inverted-U function of 3) if dUQe ^ > 0. This is obvious since good entrepreneurs always equate their marginal cost of effort, h! (e*) with their share (3) of the marginal benefit of effort, so too little effort will be forthcoming from good entrepreneurs even if they take into account the reaction of Dg with respect to effort into account. dUa When 3 have: 0, then e* = e, II (e*) = II (e). Given dtl(e) de e Qe le 0, from equation ( A l l ) , I dg, dg_, ddlp=1 x_ / a n (e) On V de A r. de * > 0 On n (e*) - n (e) < 0 (A13) (A14) Equations (A13) and (A14) ensure that balanced growth rate has a maximum with 3* 6 (0,1). Then from lemma 6, the optimal effort of good entrepreneurs (e*) is an increasing function of 3, so the balanced growth rate is also an inverted-U related to the optimal effort of good entrepreneurs. Q.E.D. Justifying Figure 5. 121 Now the problem becomes standard as in any textbook (see Mas-Colell, Whinston, Green 1995). The problem of financial intermediary becomes: MaxR = I / xf(x,e')dx + l-D[l-F(D,e*)] D Jo s.t. U(e) = [ l(x-D)f(x,e*)dx-h(e*)>u (PC) JD ID ar g m&xU (e) = f l(x-D)f (x, e)dx-h (e) (IC) e .ID The problem can be solved in two steps. The first step is to consider the (IC) constraint by integrating by parts: e* = arg maxC/ (e) = l^A — D — J F(x, e) dx^j - h (e) Now taking the first-order condition of U (e) with respect to e delivers: U'(e) = -l(^l Fe(x,e)dxSj -h'(e) = 0 Thus, it is easy to show that: U'(e)\e=0 = -l^J F (x, 0) dx^j — h! (0) > 0 U" (e) = Fee (x, e) dx^j - h" (e) < 0 Given the assumption that Fe (x) < 0, h! (0) = 0; Fee > 0, and h" (e) > 0. Thus, U (e) is concave in e with U' (0) > 0. Thus, for any given D, there is a unique e* (e* > 0) that maximizes U (e). Thus the indifference curve of entrepreneurs are shaped as in figure 5, the blue lines. The F O C U' (e) = 0 says that entrepreneur always chooses the zenith point of his indifference curve given any D to determine his optimal effort. The IC constraint consists of the set of those zenith points of entrepreneur's indifference curve in the (D, e) space shown in figure 5. Total differentiating U' (e) to get: de lFe(D,e) dD -U"(e) which is ensured by Fe (x) < 0 and U" (e) < 0. Hence the red IC curve in figure 5 has negative slope. Rewriting the expected profit of financial intermediary by integrating by parts gives: R = l(l>-j F(x,e*)dx^j 122 So slope of the iso-profit curve is negative: de _ 1 -F(D,e) dD" j0DFe(x,e) 123 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0100616/manifest

Comment

Related Items