EMPIRICAL STUDY OF MONEY DEMAND FUNCTIONS IN CHINA

EMPIRICAL STUDY OF MONEY DEMAND FUNCTIONS IN CHINA WANG WEIWEI (B. Econ., NKU) A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF SOCIAL SCIENCES DEPARTMENT OF ECONOMICS NATIONAL UNIVERSITY OF SINGAPORE 2006 ACKNOWLEDGEMENTS I would like to express my sincere gratitude to my supervisors, Associate Professor Shandre M. Thangavelu and Assistant Professor Tong Yueting, for their invaluable guidance, sustained encouragement, great patience and understanding, and continuous support over the period of this research without whom the work will not be achieved. My appreciation also goes to committee members for their support and advice. Particularly, I would like to thank Professor Basant K. Kapur for his insightful comments on my thesis and Assistant Professor Lee Jin for his cheerful assistance. I am extremely grateful to my beloved family members for their love and support throughout my Master course. The thesis is dedicated to them. Last but not least, I would also like to thank National University of Singapore for granting me the graduate level research scholarship. i TABLE OF CONTENTS ACKNOWLEDGEMENTS.............................................................................................. i TABLE OF CONTENTS ................................................................................................. ii SUMMARY .................................................................................................................... iv LIST OF TABLES .......................................................................................................... vi LIST OF FIGURES ....................................................................................................... vii 1. Introduction.................................................................................................................. 1 2. Overview of Financial Reforms in China .................................................................... 3 2.1 Financial Institutional Reforms........................................................................... 3 2.2 Monetary Policy Reforms ................................................................................... 5 2.3 Financial Markets Development and Financial Instruments Innovations........... 7 3. Literature Review....................................................................................................... 10 4. The Basic Model and Data......................................................................................... 17 4.1 The Model ......................................................................................................... 17 4.2 Data Description ............................................................................................... 19 5. Empirical Analysis ..................................................................................................... 23 5.1 Integration Properties of Data........................................................................... 23 5.1.1 The ADF Unit Root Tests........................................................................ 23 5.1.2 The Unit Root Tests for Structural Change............................................. 26 5.2 Estimation of Long-Run Money Demand Functions Using Johanson Procedure ................................................................................................................................. 31 ii 5.2.1 The Johanson Tests for M1 Money Demand Function ............................ 33 5.2.2 The Johanson Tests for M2 Money Demand Function ............................ 37 5.3 Short-Run Analysis of Money Demand Using Vector Error-Correction Model (VECM) .................................................................................................................. 43 5.3.1 Granger Causality Tests .......................................................................... 43 5.3.2 Impulse Response Analysis..................................................................... 46 5.3.3 Seemingly Unrelated Regression (SUR) Estimates of the VECM ......... 47 6. Conclusion and Policy Implications .......................................................................... 50 REFERENCES .............................................................................................................. 52 iii SUMMARY Since the late-1980s, the issue of the presence of a stable long-run function of money demand has drawn some attention in the Chinese context. A variety of empirical models have been developed to analyze the long-run behavior of money demand in China. To date, most of these studies concluded that there exists a stable long-run money demand relationship in China. However, there are still debates about the estimated long-run elasticities and the appropriate specification of the money demand functions. Moreover, most of the studies focused on the money demand in China during the pre-reform period and not the post-reform period beginning from 1979. The objective of this thesis is to investigate the behavior of money demand in China both in the long-run and short-run over the post-reform period. The empirical model is the log-linear specification of the standard money demand model with three monetary aggregates used: M0, M1 and M2. CPI was used as price measure, and the real gross domestic product (RGDP) as the transaction variable. For the opportunity costs of holding money, we considered two short interest rates and one long interest rate. The Johansen maximum likelihood procedure (Johansen and Juselius, 1990; Johansen 1991) was employed to test for the long-run equilibrium relationships among monetary aggregates, price, real income and the variables representing the opportunity cost of holding money balances. The tests for unit roots were first conducted in order to determine the integration properties of the time series under study. Our empirical results indicate that there exists a long-run money demand function for M2 over the iv sample period, and the short-term saving deposits rate is a good measure of the opportunity cost variable upon availability. The study also undertakes several tests on the M2 money demand function such as price homogeneity, unity of income elasticity and the weak exogeneity of each variable. The short-run analysis of the money demand was conducted on the basis of the long-run equilibrium relationship. A vector error-correction model (VECM) is constructed to capture the short-run dynamics and the adjustment of the long-run static disequilibrium. The Granger causality and impulse responses were used to investigate the dynamic relationship of the money demand function. The empirical results are consistent with the hypotheses of the exogenous money and long-run neutrality of money. Finally, we estimated the dynamic adjustment mechanism of the money demand using the seemingly unrelated regression (SUR) method and detect the important influence of monetary forces on the movement of inflation and output growth in China. Our results suggest that it is valid for the People’s Bank of China (PBC) to adopt M2 as its intermediate target. Moreover, it is of importance for the PBC to set proper monetary target and keep stable money growth. v LIST OF TABLES Table 2.1 Main Financing Channels of the Non-financial Institutions in China 9 Table 3.1 Results of Some Studies Based on the Methodologies of Cointegration 14 Table 5.1 A Procedure of ADF Tests 24 Table 5.2 Results of the ADF unit root tests 25 Table 5.3 Results of the Unit Root Tests for Structural Change 29 Table 5.4 Johansen Cointegration Test Results for M1 money demand function using i6 as the opportunity cost variable 33 Table 5.5 Johansen Cointegration Test Results for M1 money demand function using id as the opportunity cost variable 35 Table 5.6 Johansen Cointegration Test Results for M1 money demand function using i as the opportunity cost variable 36 Table 5.7 Johansen Cointegration Rank Test for M1 money demand function (without the interest rate) 36 Table 5.8 Johansen Cointegration Test Results for M2 money demand function using i6 as the opportunity cost variable 38 Table 5.9 Johansen Cointegration Test Results for M2 money demand function using id as the opportunity cost variable 41 Table 5.10 Johansen Cointegration Test Results for M2 money demand function using i as the opportunity cost variable 42 Table 5.11 Results of VEC Granger Causality Tests 45 Table 5.12 SUR Estimates of the VECM 48 vi LIST OF FIGURES Fig. 2.1 The Financial Deepening in China over the Reform Period 9 Fig. 4.1 Time Series Related to the Study of the Money Demand Functions 21 Fig. 5.1 Hodrick-Prescott (HP) Trend of the Series 27 Fig. 5.2 Response to Cholesky One S.D. impulse in M2 47 vii 1. Introduction The demand for money is of special importance in the formulation of a successful monetary policy, and a stable money demand function has long been perceived as a prerequisite for targeting monetary aggregates in the conduct of policy. Since the late 1980s, a number of studies have investigated the money demand function in China. Most of these studies concluded that there exists a stable long-run money demand relationship in China over different sample spans. Although some useful empirical results were obtained, there are still some drawbacks and unanswered questions in previous studies. First, there was divergence of the estimated parameters in the long-run money demand relationship, especially for the coefficient of the opportunity cost of holding money balances. These studies used different proxies for the opportunity cost of holding money, hence, there is still uncertainty concerning the appropriate measure of the opportunity cost variable and the appropriate specification of the money demand functions. Second, there are limited studies about the money demand in China during the post-reform period. In particular, the works on the post-reform period typically did not include the recent reforms since the early 1990s. China’s financial system has developed dramatically since the economic reforms in 1979. The reforms have brought significant changes in the economic structure and People’s Bank of China (PBC) relies increasingly on indirect monetary policy (monetary aggregates management) to operate macroeconomic management. Finally, these money demand functions presupposed the order of causality, and generally had little to articulate about the short-run 1 interrelationships among the variables (Zhang and Wan, 2004). In order to address some of the above issues, the study investigates the money demand function in China for post-economic reforms in China. Different proxies for the opportunity cost of holding money such as the short interest rates and the expected inflation rate were used in the study. The long-run elasticites were estimated in accordance with the cointegration tests. Furthermore, the causality tests and the impulse responses were undertaken and the short-run dynamic adjustment mechanism of the money demand was formulated on the basis of the long-run equilibrium function. The thesis is organized as follows. In next section, we introduce the post-economic reforms in China’s financial system since 1979. In section 3, we review the existing empirical studies on the money demand in China. Section 4 describes the basic empirical model and data. Section 5 examines the existence of the long-run money demand functions for currency in circulation, narrow money and broad money in China over the post-reform period. The last section concludes and discusses the policy implications of the model. 2 2. Overview of Financial Reforms in China China’s financial system has developed dramatically since the economic reforms in 1979. Before the reforms, the People's Bank of China (PBC) functioned as a central bank and commercial banks. The financial sector was small in relation to the economy and played a small role in the economic activities of the centrally planned economy. As pointed out by Yu and Xie (1999), the PBC within the pre-reform period was in fact an accounting subsidiary of the Finance Ministry, expected to assist in the fulfillment of the state physical production plan, which did not have any independent monetary policy. The main role of the PBC in the sense of commercial banks was to fund additional working capital needs beyond the planned quotas on a temporary basis. Since 1979, China’s financial system has undergone substantially development. Shi (2001) summarized the major financial reforms in China over the past decades: 1) the reforms of financial institutions, which transform the monobank system into a diversified financial system comprising central bank, commercial bank and non-bank financial institutions; 2) the reforms of monetary policy framework, which substituted the indirect monetary aggregates management for the direct credit quota management; 3) the creation of new financial markets, such as money market and capital market, as well as the introduction of new financial instruments, such as treasure bond and stock. 2.1 Financial Institutional Reforms In the early years of the reforms, decentralization of the monobank system was a main task of the financial reforms. Up to 1994, the monobank system had been split 3 into a central bank- People's Bank of China (PBC) and four state-owned specialized banks- Agricultural Bank of China, Bank of China, People's Construction Bank of China, as well as Industrial and Commercial Bank of China. Each of the four state banks was to conduct commercial banking business in a specialized part of economy1. At the same time, two nonbank financial institutions (NBFIs) –People’s Insurance Corporation of China and China International Trust and Investment Corporation2 – were established. Since then, another group of state commercial banks and other commercial banks emerged, along with the emergence of many other NBFIs. In spite of the establishment of the PBC as an independent central bank, based on the arrangement of the two-tiered banking system, the PBC still kept issuing policy loans for government special priority projects and financing a large part of the government budget deficit in the following years (Shi, 2001). Since 1994, the situation has changed, and in March 1995, the Central Bank Law was enacted and the PBC was legally empowered with the role as an independent central bank whose main responsibilities comprise conduct of monetary policies and regulation of financial institutions. Another important reform to reinforce the regulation of the PBC was the reorganization of the PBC branches in 1999. Since then, the provincial level branches of the PBC were united into nine national branches to help prevent the intervention of the local government and to strengthen the PBC’s control over its branches. The four state-owned banks also underwent considerable administrative 1 Agricultural Bank of China was responsible for rural sector, Bank of China for international transactions, People's Construction Bank of China for fixed asset investment and Industrial and Commercial Bank of China for urban industrial and commercial. 2 People’s Insurance Corporation of China specialized in insurance business and China International Trust and Investment Corporation was in charge of joint ventures. 4 interventions by the state and local governments before 1994. In 1994, three policy-lending banks 3 were set up to make policy lending determined by the governments, which separated policy lending from commercial lending and facilitated the four state-owned banks to turn into financially independent commercial banks. Furthermore, in May 1995, the Commercial Bank Law was enacted following the Central Bank Law. The law prescribed the independence and responsibilities of the commercial banks and with a view of the financial safety and the stability of the financial system, explicitly separated the business of the commercial banking from the security industry. In December 1998, the Securities Law was promulgated to further establish the separation of the banking industry and the security industry, and a separated financial system was brought into effect (Shi, 2001). 2.2 Monetary Policy Reforms With the decentralization of the banking system in the early period of the reforms, a new system of monetary policy emerged. Under the new system, the deposit-loan balance at the branch level was set as a lever of monetary control to improve the efficiency of the banking system (Xu, 1998). In other words, the more deposits a bank branch attracted, the more it could lend and earn. As a result, the lending activities of the banks dramatically increased leading to high inflation in mid-1980s, as the PBC having little control over the money supply. Under such circumstances, a credit plan in the form of loan quotas was reinstalled 3 The three policy-lending banks are: State Development Bank of China for national infrastructure, Agricultural Development Bank of China for agricultural development, and Export-Import Bank of China for international transactions. 5 to strengthen monetary control. Before 1994, the credit plan was used as the principle instrument of monetary policy for the PBC (Xu, 1998; Yu and Xie, 1999; Brandt and Zhu, 2000). In each year, the state-owned banks were given loan quotas by the PBC, in accordance with the credit plan approved by the State Council. The loan quotas set ceilings on the entire amount each bank could lend with specified loan usage, in each province, for fixed asset investment or working capital, and to state-owned enterprises or non-state enterprises. (Brandt and Zhu, 2000). By employing the credit plan, the PBC effectively controlled the credit and monetary aggregate in the post-reform period. However, along with rapid progress of the competition, the direct monetary policy of the PBC had some major drawbacks. Presented by Yu and Xie (1999), the credit allocation is a pro-cyclical monetary policy that may accelerate economic volatility, since the loan quotas are set to meet the state credit plan and the economic growth target. Furthermore, with the formation of the diversified financial institutions, enterprises could obtain funds outside the state banks, which made the credit plan inadequate in monitoring the total financing activities. In addition, since a large portion of the total bank loans were given to the nonproductive state sector according to the credit plan, the policy caused serious inefficiency in resource allocation, and the non-performing debts placed a heavy burden on the banking system. Although the PBC did not formally abandon the credit plan until 1998, it decided to replace direct credit policy by indirect monetary aggregate management in 1993, and the quantity targeting actually began in 1994. Since then, a target range for the broad 6 money (M2) was set as the PBC’s major intermediate objective (Yu and Tsui, 2000). Currency in circulation, considered as the key monetary index in the pre-reform period, has gradually lost its importance in monetary policy management4. Along with the shift of policy target to a broader monetary aggregate, the PBC began to use various policy instruments such as interest rates, required reserve ratio and open market operations to affect money supply and macroeconomic activities. 2.3 Financial Markets Development and Financial Instruments Innovations Both the money markets and the capital markets in China emerged with the decentralization of the financial system. However, the markets had not experienced rapid development until the second decade of the reforms, which mainly consist of interbank market, repurchase agreements market, and the commercial paper market (Shi, 2001). The development of the money markets provides important channels and policy instruments for the PBC to conduct open market operations to influence the liquidity in the financial system. At the end of 1990 and in the spring of 1991, the Shanghai Securities Exchange and the Shenzhen Stock Exchange opened respectively, which marked a new era for the development of the capital markets (Shi, 2001). At the same time, over-the-counter exchange systems such as the Securities Trading Automated Quotations System (STAQS) and the National Electronic Trading System (NETS) were established. Many 4 In 2003 Monetary Policy Report, the PBC noted that M0 is now monitored mainly for the purpose of countering money-laundering and tax evasion. 7 financial institutions, such as securities companies and nonbank financial institutions, were set up to trade in the markets. Because of the strict limit on the issue of the corporate bonds, the government bonds become the main financial products traded in the bond market. Furthermore, the central government has increasingly financed its expenditures through the bond market since 1994, when the governments are prohibited from borrowing from the PBC. In early 1997, the interbank government bond market was set up to meet the needs of commercial banks for more secured investment instruments after the separation of the banking business from the securities industry. From then on the government has begun to issue the government bonds via the market. At the initial stage of the reforms, Chinese government favored the bank-centered financial model and the stock market had not taken much attention from the government before the East Asian financial crisis in 1997. East Asian financial crisis in 1997, however, has led to increased awareness of the importance of a healthy stock market for financial security and the government has taken many steps to improve the supervision and regulations over the stock market. The enactment of the Securities Law in December 1998 marked a new phase of the development of the stock market. Subsequently, a series of regulations were stipulated towards the behavior of each participant in the market. The stock market has increasingly developed and improved until now, in spite of the deficiencies and the immaturity of the market. To sum up, China has experienced substantial changes in the financial system over the post-reform period and the further development can be expected in near future as the commitments to the WTO. During the reforms, economic development is generally 8 accompanied by a gradual process of financial deepening. The ratio of various financial assets to GDP increases steadily with economic development (Fig. 2.1). Up to now, however, the four state-owned banks has not yet become the independent commercial agents responsible for its own profits or losses, and the PBC still exerts an influence on the credit allocation of the banks (Shi, 2001; Qin, et al., 2005). The capital markets are still underdeveloped and only play a limited role in the macroeconomic activities. The bank credit is still the dominant channel of financing in China, through which about 80% of the total amounts of the financing inject into the non-financial institutions in recent years (Table 2.1). 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 m2/gdp household saving deposits/gdp Fig.2.1 The Financial Deepening in China over the Reform Period Table 2.1 Main Financing Channels of the Non-financial Institutions in China 2005 Billion Yuan Weight (%) Loans Stocks Treasury Bonds Corporate Bonds Total Amounts 24617 1884 2996 2010 31507 78.1 6.0 9.5 6.4 100.0 2004 Billion Yuan Weight (%) 24066 1504 3126 327 29023 82.9 5.2 10.8 1.1 100.0 Source: Monetary Policy Report, 2005 Quarter 4 (the PBC, 2006) 9 3. Literature Review The studies of money demand functions in China started since Chow (1987) pioneered an econometric exercise to apply the quantity theory of money to China using the partial adjustment model on annual data over the sample period of 1952-1983. He concluded that the quantity theory is valid in China and the estimated model provides “a reasonable first approximation" to explain money demand in China. Chow’s work brought much attention to the study of money demand and monetary policy in China in the following decades. For instance, Portes and Santorum (1987) and Chan et al. (1991) extended Chow’s study by establishing more general specification of money demand functions. Feltenstein and Farhadian (1987) estimated a money supply and real balance demand function of a planned economy of China with price control. Ma (1993) estimated the household money demand equation by the non-linear instrumental variables method. A main flaw of these studies is that these partial adjustment models “fail on theoretical and econometric grounds” (Hafer and Kutan, 1994). Some literatures on money demand functions employed the autoregressive distributed lag (ADL) model and the error correction model (ECM) to estimate the long-run relationship using quarterly data over the post-reform period (Burton and Ha, 1990; Qin, 1994; Girardin, 1996). All of these studies concluded that there exists a long-run equilibrium relationship among real balances, real income, some opportunity cost measures, and/or some institutional variables as proxy of the characteristics of the economic transition in China. These models were typically obtained using Hendry’s 10 general-to-specific methodology (Hendry, 1980), which in fact assumeed that a long-run relationship underlied the short-run dynamic models before estimating the long-run parameters. Moreover, with the small samples used, whether these equilibrium relationships exist and stable is questioned. With the development and prevalence of the econometric technique of cointegration tests, many recent works applied various methods of cointegration to investigate the long-run money demand in China. Hafer and Kutan (1994) employed the Johansen procedure (Johansen and Juselius, 1990) to estimate the long-run money demand equation. He used two aggregate price indices, the retail price index (RPI) and national income deflator (NID), and the results of the max-eigenvalue tests displayed that NID seemed a favorable price index in the sense that a single cointegrating vector among M0 (M2), NID, real national income and the one-year saving deposits rate. Another key conclusion was that M2 was a preferred aggregate measure since its price elasticity was unity, unlike M0 with the long-fun price elasticity significantly larger than one. Chen (1997) applied the Johansen procedure to test the possible cointegration relationships between the real monetary balances (RM0, RM2 and RM3) and real national income. The measurement of the opportunity cost variable he employed was the retail price inflation rate, which was I(0) hence only explained the short-run dynamics of the real money demand. He found out that a cointegration relationship existed for M0 and M2, and the results of the stability test provided by Hansen (1992) indicated that the cointegration relationship was stable through the sample period. 11 Arize and Malindretos (2000) used the max-eigenvalue tests proposed by Johansen to estimate the presence of cointegration relationship among real monetary balances (RM0, RM2 and RM3), real income, the one-year saving deposit rate and two measures of inflation variability. They found a single long-run equilibrium relationship among the variables for each measure of money. Both the Johansen procedure and FM-OLS method by Phillips and Hansen (1990) were applied to estimate the cointegrating coefficients. Furthermore, the evidence from Hansen’s stability tests indicated that only the money demand equation for M2 was stable, so M2 was a better measure of monetary target. They also estimated an ECM for each monetary aggregate and showed that inflation variability had positive effect on money demand in both the short run and long run. Other researchers paid some attention to China’s money demand during the post-reform period. Huang (1994) employed both of the Engle and Granger (E-G) method (Engle and Granger, 1987) and the Johansen procedure to test the existence of cointegrations using quarterly data. Since one cointegration relationship was found among M2, real gross national products (GNP), consumer retail price (RPI) and the real interest rate on one-year saving deposit, he constructed an error correction model (ECM) based on the estimated cointegration relationship by the E-G procedure. Deng and Liu (1999) used the Engle and Granger cointegration tests and ECM to specify the money demand function among real monetary balances (RM1, RM2), real gross domestic product (RGDP), the expectation of retail price inflation rate and the real interest rate on three-year time deposit rate using quarterly data. They found that 12 the expected inflation π te (= π t −1 ) was I(1) hence involved in the cointegration, 1 + π t −1 where π t was the inflation rate compared to the same period last year. One cointegrating equation was estimated for both measures of monetary aggregates. Moreover, after the estimation using ECM for both RM1 and RM2, they constructed a nonlinear ECM with an error-correction term obtained from E-G procedure to formulate the money demand function for RM2 and obtained a parsimonious model. Both the whole sample period of 1952-2000 and the post-reform period of 1982:1-2000:4 were used in Gu’s dissertation (2004). He employed both the maximum likelihood method by Johansen and the dynamic OLS method by Stock and Watson (1993) in estimation of the long-run relationship among RM0 (RM2), RPI and the interest rate on one-year saving deposit. The empirical results indicated that for both narrow money and broad money, there existed long-run money demand functions over the whole period. In sharp contrast to the picture from 1952 through 2000, there existed a long-run money demand function only for broad money holdings over the post-reform period. The Gregory and Hansen tests (Gregory and Hansen, 1996) suggested there was a structural break in 1966 in broad money demand function for the whole sample. Comparisons of the estimated cointegrating coefficients from these studies are listed in Table 3.1. In general, the estimated income elasticities were relatively high for all the monetary aggregates using different data, sample periods and/or estimation methods. Moreover, the income elasticities for broad money were all larger than those for currency and narrow money. Theoretically, real money demand increases proportionally to an increase in real income. However, the evidence from developing 13 countries indicated that it is a fairly common phenomenon that the long-run income elasticities tend to exceed unity, especially for broad money (Aghevil et. al, 1979; Bordo and Jonung, 1981; Ghatak, 1995). Table 3.1 Results of Some Studies Based on the Methodologies of Cointegration Authors Period Money Real Income Interest Rate Price Inflation Hafer &. Kutan (1994) 1952-88 M0 M2 1.13 1.33 0.13 0.15 2.48 1.52 ─ ─ Chen (1997) 1951-91 RM0 RM2 1.50/1.44 1.93/1.79 ─ ─ ─ ─ ─ ─ Arize &. Malindretos (2000) 1952-94 RM0 RM2 RM3 1.33/1.20 1.51/1.63 1.13/1.19 -0.17/0.10 ─ -0.32*/-0.18* ─ ─ ─ ─ ─ ─ Huang (1994) Deng &. Liu (1999) 79-90 (Quarter) 80:1-94:4 M2 1.56 -0.29 2.12 ─ RM1 RM2 1.29 1.80 -0.12 0.97 ─ ─ -0.34 ─ Gu (2004) 1952-00 RM0 RM2 RM0 RM2 1.32/1.34 1.54/1.56 ─ 1.51/1.49 0.44*/0.22* 0.41*/-0.002 ─ -0.15*/-0.16* ─ ─ ─ ─ ─ ─ ─ ─ 82:1-00:4 Notes: The figures in the last four columns are estimated values of different elasticities. * denotes that the interest elasticity is significant. In particular, Deng &. Liu (1999) did not report the significance of the interest elasticities. In contrast to the results of the income elasticities, the estimated elasticities of the interest rates appeared quite different among the studies, both in the significance and signs. It seems that there is still uncertainty about whether the interest rates are good proxy for the opportunity costs of holding money. Some authors used various measures 14 of the expectation of inflation rate as the opportunity costs, and discrepancies of their results were observed. Chen (1997), Arize and Malindretos (2000) and Gu (2004) used the retail price inflation rate, and found the inflation was stationary hence ruled it out from the cointegration relationship. On the other hand, Deng and Liu (1999) used ( π t −1 ) as the specification of the expected inflation, which was I(1) hence involved 1 + π t −1 in the cointegration, where π t was the inflation rate compared to the same period last year. As suggested by Calvo and Leiderman (1992) and Easterly et. al (1995), the inflation rate did not represent the true inflation cost of holding money in discrete time, while the appropriate measure of the capital loss due to inflation should be ( πt ). 1+ πt Since the authors used different measures of the expectation of inflation, it is not odd that they draw different conclusions about the money demand functions. A surprising result from existing works was that the price elasticity of M0 was significantly greater than unity (Hafer and Kutan, 1994). A similar result for M2 was obtained by Huang (1994). As we know, the demand for money is in fact a demand for real balances. Therefore, high price elasticity brings us a doubt whether the long-run relationships among money and other variables reflect the true money demand relation. According to our analysis given above, there are still some questions about the estimation of the long-run money demand in China. The reforms have brought significant changes in the economic structure and the PBC relies increasingly on indirect monetary policies to operate macroeconomic management. Therefore, separate consideration of China’s post-reform period is necessary, although some empirical 15 works did not find a significant structural change in China prior to and after the reforms. However, most previous works estimated money demand over the whole sample, including both pre and post reform period; the works on the post-reform period basically did not include the recent reforms since the early 1990s. Our thesis is to investigate the money demand function in China for various monetary aggregates using the updated quarterly data to contain the new changes in the economy. 16 4. The Basic Model and Data 4.1 The Model Our departure point is the standard model of money demand, which comprises a transaction variable measuring the volume of real economic activity, for instance, real national income, and one and more measures for the opportunity cost of holding money, such as the short interest rate of some kind or the expected inflation rate. Theory suggests that real money demand is unchanged when price level rises since individuals hold money for real purchasing power. Thus, nominal money demand changes in proportion to the change in price level. Moreover, real money demand depends positively on real income and the associated volume of transactions, and negatively on the opportunity cost of holding money. The former embodies “transaction demands” for money in the income version, and in the latter argument, money is treated as one among a number of alternative assets to hold by individuals, the demand for which relies on its opportunity cost. Therefore, the demand for money balances can be written as: M d / P = L(Y , i ) , L y > 0, Li < 0 (4.1) where Md is the nominal money demand, and P is the general price level. L represents the liquidity function relating real money balances to real income (Y) and the opportunity cost variable (i); Ly and Li are the sensitivity of real demand for money to the change in real income and the change in the opportunity cost, respectively. According to the theory of money demand above, an empirical specification of 17 equation (4.1) is to be estimated in the log-linear form: mt = β 0 + β1 pt + β 2 yt + β 3it + ε t (4.2) where mt is the nominal money supply, pt is the general price level, yt is real income, it is the opportunity cost variable, β0 is a constant, and εt is random error term. All the variables are expressed in logarithms. In my study, I assume that the money market is in equilibrium so that money demand equates money supply. A key assumption of the model is that the random sequence {εt} must be a stationary process, because any deviation from the long-run equilibrium relationship of the demand for money must necessarily be temporary in nature. Although all the variables of money supply, price level, real income and the opportunity cost can be non-stationary, following I(1) process, a linear combination of these non-stationary variables can be stationary based on the concept of cointegration (Engle and Granger, 1987), which implies that the four variables mt, pt, yt and it can be cointegrated. A lack of cointegration implies no long-run equilibrium among the four variables. Hence, if the model (4.2) is meaningful, a cointegration relationship of money demand can be specified. Various monetary measures can be employed to test which one is preferable in determining the long-run equilibrium relationship. If no cointegration is tested by using a particular measure of monetary aggregate, it implies that there exist permanent deviations among variables so that such a measure of monetary aggregate can not be considered as a viable policy tool for the PBC. 18 4.2 Data Description Before the empirical analysis, the data given as quarterly unadjusted observations will be briefly described. The sample period is 1986:1 to 2004:4 due to the availability of quarterly data in China. Three monetary aggregate data are used: M0, M1 and M2. M0 is currency in circulation. Narrow money (M1) equals the sum of currency in circulation (M0) and demand deposits. Broad money (M2) comprises narrow money (M1) and quasi-money, which includes time deposits, saving deposits as well as other deposits. The data of money are obtained from IFS. Two alternative price data can be considered as the price measure, consumer price index (CPI) and retail price index (RPI). Because China has adopted CPI as the major indicator of inflation in place of RPI, and CPI is commonly used as aggregate price level in a country, CPI is used in examining money demand equation. CPI is obtained from IFS and 1986 is the base year. For transaction variable data, the real gross domestic product (RGDP) is employed. Since CPI is used as price measure, we use CPI as the deflator of nominal GDP for consistency. However, the nominal GDP data, obtained from China Quarterly Gross Domestic Product Estimates and updated from National Bureau of Statistics of China (NBS), only covers the period 1992:1-2004:4 on a cumulative basis. In order to obtain GDP data over the whole sample period, the quarterly nominal GDP before 1992 are first generated from the annual data. A simple procedure is used here. Because China’s nominal GDP shows strong seasonality and relatively stable year-on-year growth rate 19 for each quarter compared to the annual growth rate5, we backtrack the quarterly nominal GDP from the year of 1992 presuming the same year-on-year growth rate of each quarterly as the annual growth rate. For the opportunity costs of holding money, we consider two sets of measures: the nominal interest rate and the expected inflation rate. For nominal interest rate, various short interest rates available are tested, such as the interest rate on 6-month saving deposits (i6) and the interest rate on demand deposits (id). The one-year saving deposits rate (i) is also used for comparison. Using these nominal interest rates is not ideal since the official interest rates are often regulated by the government and vary very infrequently in China. Furthermore, these measures are not suitable as the opportunity cost of holding broad money aggregate (M2) since these interest rate variables are in fact some kind of own return rate for M2. However, these variables are the only ones we can use due to the unavailability of the data of market interest rates. For the expected inflation rate (πe), we use ( πt ) as a proxy, where π is the year-to-year inflation rate. 1+ πt The data of the nominal interest rates are obtained from IFS and the PBC. Fig. 4.1 plots all the series under study. All the variables are log transformed except for the interest rates and the expected inflation. It can be seen that monetary aggregates and RGDP appear to exhibit decidedly upward trends and marked seasonal patterns are exhibited especially for M0, CPI and RGDP. In addition, the interest rates are fixed at certain levels for most of the period. 5 From 1992-2004, the average deviations of the year-on-year growth rate for each quarter from the annual growth rate are 0.008, -0.003, 0.006 and 0.003, respectively. The standard deviations are 0.020, 0.020, 0.019 and 0.027, respectively, which are mild considering the average annual growth rate of 15% over this period. 20 13 6.0 12 5.8 11 5.6 5.4 10 5.2 9 5.0 8 4.8 7 4.6 6 86 90 88 92 96 94 98 02 04 4.4 86 88 90 92 94 M2 M1 M0 00 96 98 00 02 04 02 04 CPI 9.6 .12 9.2 .10 8.8 .08 8.4 .06 8.0 .04 7.6 .02 7.2 .00 86 88 90 92 94 96 98 00 02 04 86 88 90 92 94 I6 RGDP 96 Id 98 00 I .20 .16 .12 .08 .04 .00 -.04 86 88 90 92 94 96 98 00 02 04 πe Fig. 4.1 Time Series Related to the Study of the Money Demand Functions Note that CPI series show a distinctive upward trending behavior before 1996, and fluctuate around horizon afterwards. That is due to the fact that monetary authorities allowed rapid growth in domestic credit to foster economic growth before the 21 mid-1990s, while generally kept inflation at a relatively high level (even at an inflation level of two digits). High inflation seriously damaged economic and financial stability. In order to tackle the problem of high inflation, PBC implemented tight monetary policies which sharply brought down the inflation rate from the peek. Sharp disinflation caused concerns about deflation in the following years, and inflation rate was kept close to zero. From 2003, inflation started to rise and moderate inflation has been maintained until now. 22 5. Empirical Analysis This section presents unit root tests using different methods and then Johansen maximum likelihood procedure is applied to test for cointegration among monetary aggregates (M0, M1, M2), price (CPI), real income (RGDP) and some measures of the opportunity cost of holding money balances (i6, id, i or π e ). Moreover, the Granger causality tests and the impulse response analysis are used to investigate the interrelationships among the cointegrated variables in the system. Further, we estimate the dynamic adjustment mechanism of money demand based on a vector error-correction model (VECM). 5.1 Integration Properties of Data Since non-stationarity is suspected to the series plotted in Fig. 4.1, unit root tests should be conducted for the variables of interest to determine their orders of integration before modeling money demand equation. 5.1.1 The ADF Unit Root Tests It is necessary to determine the order of integration for each series, therefore both level and first difference of the series are tested for unit roots in case of I(2) processes. The Augmented Dickey-Fuller (ADF) tests (Dickey and Fuller, 1979) are examined for the presence of a unit root based on the following regression: p 3 i =1 i =1 ∆yt = a0 + γyt −1 + a2 t + ∑ β i ∆yt −i + ∑ φi Dit +ut (5.1) 23 where y is the variable to be tested, α0 is a constant, t is the time trend, p is the lag order of autoregression and ut is a stationary random error term. In order to account for the seasonality in the quarterly data, the centered seasonal dummy variables {D1}, {D2} and {D3} are considered in the autoregression equations. The null hypothesis of a unit root is rejected if γ is negative at conventional significances. Table 5.1 A Procedure of ADF Tests Model 1: ∆yt = a0 + γyt −1 + a 2 t + p ∑ β ∆y i i =1 3 t −i + ∑ φi Dit +ut i =1 Hypothesis Test Statistic If not rejected If rejected γ=0 γ =a2= 0 γ=0 ττ φ3 N(0,1) Proceed Model 2 Unit root No unit root Proceed No unit root Model 2: ∆y t = a 0 + γy t −1 + p 3 i =1 i =1 ∑ β i ∆yt −i + ∑ φi Dit +ut Hypothesis Test Statistic If not rejected If rejected γ=0 a0=γ= 0 γ=0 τµ φ1 N(0,1) Proceed Model 1 Unit root No unit root Proceed No unit root Model 3: ∆y t = γy t −1 + p 3 i =1 i =1 ∑ β i ∆yt −i + ∑ φi Dit +ut Hypothesis Test Statistic If not rejected If rejected γ=0 τ Unit root No unit root Source: Enders (2004) Since the presence of the additional estimated parameters reduces degrees of freedom and the power of the test (Enders, 2004) the form of the regression is critical 24 for obtaining the appropriate testing results. Doldado, et al. (1990) suggested a procedure to test for a unit root when the form of the data-generating process is unknown. Here a modification of the procedure provided by Enders (2004) is followed to select appropriate set of regressors when doing the ADF tests. The procedure is shown in Table 5.1. The maximum lag-length pmax = 11 is chosen according to the Bartlett criteria and the lag-order p of the univariate autoregression equations is selected based on Akaike information criterion (AIC) and Schwarz information criterion (SC). The Ljung-Box Q-test is conducted to make sure that the residuals appear to be white noise. Table 5.2 Results of the ADF unit root tests Level First Difference Series M0 M1 M2 CPI RGDP i6 id i πe p ττ φ3 τµ φ1 τ p τµ 8 0 3 1 4 1 0 1 5 -1.344 -1.069 -0.127 -1.365 -2.501 ─ ─ ─ ─ 2.370 1.593 1.282 2.896 3.499 ─ ─ ─ ─ -1.963 -1.520 -1.610 -2.376 0.568 -1.123 -0.715 -0.708 -2.318 2.960 72.619** 5.890* 4.494 4.008 0.784 0.986 0.480 2.893 1.097 ─ ─ 1.561 2.838 -0.989 -1.404 -0.925 -2.043* 1 0 2 0 3 0 0 0 3 -4.010** -7.311** -3.406* -2.903* -3.029* -6.258** -7.290** -6.494** -4.839** Notes: Sample period is 1986:1-2004:4. *, **: significant at 5%, 1% level, respectively. The test regression equation for M2 does not include the centered seasonal dummies since no seasonality is examined in the test. The regression for i6, id, i and πe only includes a constant due to the fact that the series do not show a decidedly upward trend and obvious seasonality over the sample period. We do not consider the time trend in the regressions for first differenced series. Critical values for the τ statistics are taken from MacKinnon (1991), and those for the φ statistics are taken from Dickey and Fuller (1981). The results of the ADF tests are reported in Table 5.2. We first test for the presence 25 of double unit roots in the series. Results indicate that the τµ statistics are all significant at 5% level. Thus we reject the null hypothesis of double unit roots and conclude that the differenced series of M0, M1, M2, CPI, RGDP, the interest rates and πe are all stationary. For the level series, the null hypothesis of a unit root is rejected at 5% significant level for monetary aggregates, CPI, RGDP and the interest rates according to the testing procedure shown in Table 5.2, implying that these series are all integrated of order 1, or I(1). However, for the series of πe, the τ statistic is significant at 5% level, implying that no unit root is detected as suggested by the ADF tests and we can conclude that the sequences πe is stationary. Therefore, we exclude it from the cointegration tests since no cointegration relationship exists among I(0) and I(1) series. 5.1.2 The Unit Root Tests for Structural Change Before turning to cointegration tests, the unit root tests for structural change are conducted since structural change is suspected for variables such as CPI according to above analysis. Perron (1989) argued that standard tests for a unit root against trend stationary can not reject the null hypothesis if the true data generating process is stationary around a trend function with a one-time break. We test the hypothesis of a unit root against the alternative of trend stationarity with an endogenous structural break point (Zivot and Andrews, 1992; Perron, 1997). In order to identify the long-term trend component of the series, Hodrick-Prescott (HP) filter is employed and we can observe the possible changes in time trends of the series from Fig. 5.1. It seems that the slope of the trend function has decreased for 26 monetary aggregates and CPI series since the mid-1990s. This phenomenon is consistent with the policy effects of tight monetary policies which caused a slowdown in the growth rate of monetary aggregates and that of CPI. However, we do not detect the marked change in growth of real GDP over the span of 1986-2004; hence we do not reject a unit root for real GDP data based on the result of the ADF tests if the growth rate of real GDP does not change over the sample period. The series of the interest rates are frequently invariant therefore excluded from the tests. The tests for πe series are also omitted given that it is perceived as being stationary based on the results of the ADF tests. 13 5.8 12 5.6 11 5.4 10 5.2 9 5.0 8 4.8 The HP Trend of M2 The HP Trend of M1 The HP Trend of M0 7 The HP Trend of CPI 4.6 6 4.4 86 88 90 92 94 96 98 00 02 04 86 88 90 92 94 96 98 00 02 04 9.6 9.2 8.8 8.4 8.0 The HP Trend of RGDP 7.6 86 88 90 92 94 96 98 00 02 04 Fig. 5.1 Hodrick-Prescott (HP) Trend of the Series Notes: Hodrick and Prescott (1997) suggested that the smoothing parameter (λ) used in HP filter is 1600 for quarterly data. Here we follow their suggestion. 27 We use the “changing growth” model (Model B) provided by Zivot and Andrews (1992), which tests for a unit root against the alternative of stationarity with changed trends at some unknown point. The model has the following ADF type augmented regression: p 3 i =1 i =1 y t = µ + βt + γDTt * (λ ) + αy t −1 + ∑ ci ∆y t −i + ∑ φi Dit +et where DTt * (λ ) = t − Tλ if t > Tλ , and 0 otherwise. λ = (5.2) TB , where TB denotes the T break point at which the change in trend occurs; λ is the break fraction, ranging from (p+2)/T to (T-1)/T. According to Perron (1997), there are two methods to select TB and correspondingly λ endogenously. First, TB is selected to minimize the t statistics for testing α=1. The minimized t statistic on α is defined as tα* = minTB∈( p +1,T ) tαˆ (TB , p) . The implication is to choose the breakpoint that gives the least support for the null hypothesis of a unit root. Second, TB is chosen to minimize the t statistics on the change in slope. The t- statistic on α correspondingly is denoted by tα* ,γ = tαˆ (TB∗ , p) , where TB∗ is such that tγ* = tγˆ (TB* ) = minTB∈( p+1,T ) tγˆ (TB , p) . This procedure allows for a prior assumption on the direction of the changing growth, i.e. the sign of γ. If not any a prior assumption can be imposed, the break date is selected to maximize the absolute value of tγˆ , and the corresponding t statistic on α is denoted by tα* , γ . The lag length is determined by using sequential testing of the t-statistics of the last lags in the estimated autoregression. The value of i is chosen as the lag order such that the t-statistic on cp is significant at 10% level in absolute value based on the asymptotic normal distribution, and the t-statistic on ci for i > p is less than the critical 28 value of 10% significance level. This procedure is denoted as “t-sig” by Perron (1997). Also, make sure that no significant serial correlation appears in the residual series. Table 5.3 Results of the Unit Root Tests for Structural Change Series TB∗ p M0 97:1 8 M1 97:1 3 96:3 3 96:1 3 95:3 3 CPI 96:4 2 RGDP 88:4 4 88:3 4 M2 βˆ 0.0287 [4.67] 0.0099 [ 3.08] 0.0098 [ 3.08] 0.0109 [ 4.06] 0.0102 [ 4.04] 0.0029 [ 3.46] -0.0016 [-0.49] -0.0036 [-0.90] λˆ -0.0140 [-4.54] -0.0032 [-3.08] -0.0031 [-3.10] -0.0051 [-4.45] -0.0049 [-4.47] -0.0027 [-3.35] 0.0070 [1.95] 0.0089 [2.02] αˆ tα* tα* ,γ tα* , γ -4.79** -4.79** 0.4583 -4.79* 0.8179 -3.02 ─ ─ 0.8226 ─ -3.01 -3.01 0.8419 -4.01 ─ ─ 0.8527 ─ -3.96 -3.96 0.8897 -3.81 -3.81 -3.81 -0.2571 -3.15 ─ ─ -0.2506 ─ ─ -3.145 Notes: Sample period is 1986:1-2004:4. *, **: significant at 10%, 5% level, respectively. The test regression equation for M2 does not include the centered seasonal dummies since no seasonality is examined in the test. Figures in brackets are the t-statistics. Critical values for the t-statistics on α in the last three columns are taken from Perron (1997). The empirical results in Table 5.3 show that the unit root hypothesis can not be rejected at the 5% or 10% significance level, for narrow and broad money, as well as CPI, according to the statistics of tα* and tα* ,γ . However, tα* statistic is significant at 10% level and tα* ,γ statistic at 5% level, which implies that M0 appears to be stationary with changing trends prior to and after an endogenous break point. Here tα* ,γ statistic is used instead of tα* , γ . Since it is observed of a slowdown in the growth rate of the 29 three monetary aggregates and CPI series, a priori imposition of negative sign is set on the term of change in slope ( DTt * (λ ) ). Hence, tα* ,γ statistic is used to select TB endogenously, which is a better method since using tα* ,γ allows tests with greater power (Perron, 1997). Note also that the similar results in favor of a unit root can be obtained when using the tα* , γ statistics on M0, M1, M2 and CPI series. For real GDP data, no change of trend is observed so tα* , γ statistic is used together with tα* . Results indicate that no unit root exists according to both statistics. A comment is made about TB∗ selected according to the two methods. The dates of TB∗ range from 1995 to 1997, associated with the period of tight monetary policies implemented since the mid-1990s. It is interesting to take a look at the estimated change in the slope of the trend. The estimated decreasing rates of growth are: M0, 48.8%; M1, 32.3%; M2, 46.8%; CPI, 93.1%. These figures are indeed quite large and suggest that the tight monetary policies have caused very important changes in some macroeconomic indices in China. On the other hand, no significant change in real GDP growth rate is examined. Although no critical values are provided for comparison, the estimates of trend and those of change in trend seem not significant. Anyway, the focus here is not on the consistent estimates of the break date and the change in slope, but on the presence of a unit root. According to the results of the unit root tests, we can conclude that both of narrow and broad money, CPI, real GDP as well as the interest rates series contain a unit root, while M0 and πe are stationary and therefore excluded from the cointegration tests in the following part. We now have two measures of monetary aggregates and the interest 30 rates are the only measures of the opportunity cost variable in the analysis of cointegration. 5.2 Estimation of Long-Run Money Demand Functions Using Johanson Procedure Many methods to estimate the cointegrating vectors have been developed since the end of 1980s. Among them, the maximum likelihood procedure suggested by Johansen is one of the most widely used methods for cointegration estimation. Phillips (1991) argued that the best way to proceed in the estimation of cointegrated systems is full system estimation by maximum likelihood (such as the Johansen procedure) with all prior knowledge about the presence of unit roots. Using Monte Carlo methodology, Gonzalo (1994) demonstrated that the Johansen procedure can clearly perform better than alternative estimators in the finite sample, which is consistent with its asymptotical properties, even when the errors are non-normal distributed and when the dynamics are unknown. A similar conclusion was obtained by Stock and Watson (1993). I use Johansen maximum likelihood procedure (Johansen and Juselius, 1990; Johansen, 1991) to test whether M1 (M2), real GDP and the interest rate are jointly cointegrated. Johansen developed the methodology of VAR-based cointegration tests. Consider a VAR of order p: xt = µ + A1 xt −1 + A2 xt − 2 + ... + Ap xt − p + ΦDt + ε t (5.3) where xt is a n-vector of non-stationary I(1) variables, µ is a n-vector of deterministic trends, and Dt are the centered seasonal dummy variables. The VECM representation of the model can be rewritten as: 31 ∆ x t = µ + Π x t −1 + p p i =1 j =i +1 p −1 ∑π i =1 i ∆ xt −i + Φ D t + ε t (5.4) where Π = −( I n − ∑ Ai ) , π i = − ∑ A j for j = 1, …, p-1. If the matrix Π has reduced rank r < n, then there exist n × r matrices α and β each with rank r such that Π = αβ ′ and β ′xt is I(0). β ′xt represents the r cointegrating equations and α is the matrix of the speed of adjustment coefficients. Johansen proposed two likelihood ratio (LR) test statistics—the trace statistic (QTrace) and the maximal-eigenvalue statistic (QMax)—to test for the presence of the cointegrating relationship and to determine the cointegration rank (r) in the system. Cheung and Lai (1993) argued that Johansen tests suffer from the finite sample distortion toward greater cointegration rank when asymptotic critical values are used. To correct for the finite sample bias, Reinsel and Ahn (1992) suggested adjusting the LR test statistics by a scaling factor of (T-np)/T and comparing them with their asymptotic critical values, where T, n, and p denote the effective sample size, the dimension of the model and the lag length, respectively. We calculate the modified version of the LR test statistics as suggested by these authors. The maximum lag-length pmax = 8 is chosen assuming that the effects of impulses will vanish after two years. The lag-order p of the unrestricted VAR is determined using the Sims modified LR statistic (Sims 1980). The multivariate autocorrelation LM test is performed on the residuals after the cointegration tests to ensure a white noise process of each residual series. 32 5.2.1 The Johanson Tests for M1 Money Demand Function 5.2.1.2 Cointegration test for M1, CPI, RGDP and the interest rate on 6-month saving deposits (i6) We first estimate whether there exists any cointegrating relationship among M1, CPI, RGDP and i6. Results are presented in Panel A of Table 5.4. Both the trace and max-eigenvalue tests reject the presence of a cointegrating equation at 5% significant level, while a single cointegrating vector is not rejected at 10% level according to the trace test. Table 5.4 Johansen Cointegration Test Results for M1 money demand function using i6 as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.3584 0.1613 0.1403 0.0603 41.65 19.46 10.67 3.11 47.21 29.68 15.41 3.76 22.19 8.79 7.56 3.11 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M1 CPI RGDP i6 Cointegrating vector ( βˆ ) 1.000 -0.448 -2.040 -4.703 Adjustment vector ( αˆ ) -0.071 -0.041 0.021 0.044 Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. r denotes the cointegration rank. p = 6 is selected based on the Sims LR statistics. The adjusted QTrace and QMax test statistics are reported. The results are based on the assumption of the presence of a linear deterministic trend in the data. Critical values (C.V.) for QTrace and QMax statistics are taken from Osterwald-Lenum (1992). 33 If r = 1 is assumed, the estimates of β and α are given in Panel B of Table 5.4. The results indicate that M1 is positively related to CPI and real GDP. However, the cointegrating coefficient of i6 shows a negative sign, which is unreasonable since narrow money should have a negative relation with its opportunity cost variable proxied by the interest rate on 6-month time deposits. Thus, i6 is not a correct measure of the opportunity cost of holding narrow money. Furthermore, the adjustment coefficient of CPI is negative, implying that CPI will deviate permanently from the long-run relationship. All in all, we conclude that no cointegrating relationship exists among the variables. 5.2.1.2 Cointegration test for M1, CPI, RGDP and the interest rate on demand deposits (id) We now turn to the use of the interest rate on demand deposits as the opportunity cost variable for M1 money demand function. The results are shown in Table 5.5. Both trace and max-eigenvalue tests indicate 1 cointegrating equation among the variables at 5% significant level, and there is no evidence in favor of more than one cointegrating vectors. The cointegrating coefficients are all of expected signs, in particular, the coefficient of id has a negative sign since demand deposits are main component of narrow money. However, the adjustment coefficient of CPI still has a negative sign, indicating that the cointegrated system is unstable. If a zero restriction is imposed on the parameter β in id, the null hypothesis can not be rejected at conventional significance levels, so the interest rate does not belong to the cointegrating space. 34 Table 5.5 Johansen Cointegration Test Results for M1 money demand function using id as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.5050 0.3527 0.1368 0.0440 59.86** 28.21 8.64 2.02 47.21 29.68 15.41 3.76 31.63* 19.57 6.62 2.02 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M1 CPI RGDP id Cointegrating vector ( βˆ ) 1.000 -0.190 -2.182 -9.545 Adjustment vector ( αˆ ) -0.044 -0.077 0.009 0.008 Zero interest rate semi-elasticity (β4 = 0) χ2(1) = 0.881 (0.35) Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p = 7 is selected based on the Sims LR statistics. The figure in parenthesis is p-value. 5.2.1.3 Cointegration test for M1, CPI, RGDP and the interest rate on one-year saving deposits (i) If the interest rate on one-year saving deposits (i) is used, both trace and max-eigenvalue tests indicate 1 cointegrating equation among the variables at 10% significant level, but accept the null hypothesis of r = 0 at 5% level. If we assume there exists one cointegrating vector, the cointegrating coefficient of i has no expected sign, so it is not a correct measure of the opportunity cost of holding narrow money, either. 35 Table 5.6 Johansen Cointegration Test Results for M1 money demand function using i as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.3955 0.1779 0.1432 0.0461 45.05 19.88 10.08 2.36 47.21 29.68 15.41 3.76 25.15 9.79 7.72 2.36 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M1 CPI RGDP i Cointegrating vector ( βˆ ) 1.000 -0.678 -1.796 -1.318 Adjustment vector ( αˆ ) -0.080 -0.037 0.086 0.072 Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p = 6 is selected based on the Sims LR statistics. The figure in parenthesis is p-value. Since these interest rates are not good proxies of the opportunity cost variable for M1, we may exclude the opportunity cost variables from the cointegration system and investigate the presence of a cointegrating relationship among M1, CPI and real GDP. The results are shown in Table 5.7. Table 5.7 Johansen Cointegration Rank Test for M1 money demand function (without the interest rate) Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 0.2805 0.1527 0.0323 29.03 10.92 1.81 29.68 15.41 3.76 18.10 9.11 1.81 20.97 14.07 3.76 Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p = 6 is selected based on the Sims LR statistics. We can see that both the trace and max-eigenvalue tests reject the presence of a 36 cointegrating equation at 5% significant level, while a single cointegrating vector is not rejected at 10% level according to the trace test. Taking one thing with another, we tend to reject the presence of a long-run equilibrium money demand equation for narrow money over the sample period. 5.2.2 The Johanson Tests for M2 Money Demand Function 5.2.2.1 Cointegration test for M2, CPI, RGDP and the interest rate on 6-month time deposits (i6) Table 5.8 reports the results of cointegration test for M2 money demand function using i6 as the opportunity cost variable. The null hypothesis of no cointegrating vector is rejected at 1% level based on both the trace and the max-eigenvalue tests. Furthermore, both tests indicate a unique long run equilibrium relationship among the variables since no evidence supports more than one cointegrating vectors. As shown in Panel B of Table 5.8, the estimators of the cointegrating coefficients are all of the expected signs. Note that the estimated cointegrating parameter in i6 is negative, implying a positive relationship between M2 and the interest rate in the long run. This demonstrates our expectation that the interest rate on saving deposits represents an own return of holding broad money. Therefore, the cointegrating relationship in fact can be interpreted as a model of portfolio money demand which regards the primary role of money as a form of wealth. The estimated adjustment coefficients also display correct signs except for that of CPI. However, the estimated α in CPI is quite close to zero, so it is very likely that CPI is weakly exogenous hence does not respond to the deviations 37 from the long-run equilibrium. Table 5.8 Johansen Cointegration Test Results for M2 money demand function using i6 as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.5221 0.2868 0.1845 0.0296 65.52** 28.60 11.70 1.50 47.21 29.68 15.41 3.76 36.92** 16.90 10.20 1.50 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M2 CPI RGDP i6 Cointegrating vector ( βˆ ) 1.000 -0.667 -2.392 -16.111 Adjustment vector ( αˆ ) -0.025 -0.004 0.074 0.028 C: Hypothesis Testing Null LR Statistics Zero interest rate semi-elasticity (β4 = 0) χ2(1) = 16.435 (0.00) Price homogeneity (β1 = -β2) χ2(1) = 2.415 (0.12) Unity income elasticity (β1 = -β2 = -β3) χ2(2) = 13.132 (0.00) Weak exogeneity of M2 (α1 = 0) χ2(1) = 0.849 (0.36) Weak exogeneity of CPI (α2 = 0) χ2(1) = 0.165 (0.68) Weak exogeneity of RGDP (α3 = 0) χ2(1) =7.450 (0.01) Weak exogeneity of i6 (α4 = 0) χ2(1) = 22.476 (0.00) Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p = 6 is selected based on the Sims LR statistics. The figures in parenthesis are p-values. We now consider several important hypotheses about the money demand equation. Results are presented in Panel C of Table 5.8. One test of interest is whether the interest rate does belong to the cointegrating equation. The null hypothesis of zero interest rate 38 semi-elasticity can be rejected at 5% significance level. We thus include i6 in the long-run money demand function. By imposing restrictions on the cointegrating vector, we can examine the hypotheses of price homogeneity and unity income elasticity. The results from Panel C demonstrate the validity of price homogeneity hypothesis but reject unity income elasticity in the money demand equation. The restricted cointegrating equation reported below: M 2 = 11.545 + CPI + 1.936 RGDP + 7.603i6 (5.5) The estimated income elasticity is significantly greater than unity, consistent with the findings of previous studies of China and developing countries. Some plausible explanations for the high income elasticity are presented by previous studies. A prevailing explanation is that in China, there has been a monetization process6 going on since the reforms (Feltenstein and Farhadian, 1987; Blejer et. al, 1991; Yi, 1991, Hafer and Kutan, 1994). The economic reforms led to greater reliance on markets to allocate resources and there is greater need for money as the medium of economic transactions associated with the rapid economic growth. Under this explanation, some researchers argued that augmented with some institutional factors specific to China’s situation which could account for the monetization process, the standard money demand function could perform satisfactorily with unity income elasticity (Qin, 1994; Girardin, 1996). On the other hand, bank deposits are still the main interest-bearing financial asset for the majority of households in China. Due to the underdevelopment of the financial 6 As pointed out by Yi (1991), monetization refers to the process in which a rising proportion of economic activities is conducted by money. 39 markets and the limited financial instruments, households have to mainly rely on deposits as the stock of wealth and the substitution of deposits and other financial assets is not relevant for most households. The estimated income elasticity thus can not reflect the true transaction demand of money and the excess money functions as stock of value in the absence of alternative assets. The weak exogeneity of the individual variable in the system is tested by restricting α i = 0 , i=1, 2, 3, 4. If zero restriction on the adjustment coefficient of a variable is not rejected, it means that the variable has no feedback on the past deviation from the long-run money demand relationship. According to the LR test statistics, there is no strong evidence for restricting the adjustment coefficients of real GDP and the interest rate to zero, so we reject the hypotheses that real GDP and the interest rate are weakly exogenous. However, the results show strong support for weak exogeneity of M2 and CPI. Because two variables in the system seem weakly exogenous, we can not construct an error-correction model with money as the only dependent variable as many previous studies did. 5.2.2.2 Cointegration test for M2, CPI, RGDP and the interest rate on demand deposits (id) We also test for the interest rate on demand deposits as the opportunity cost variable for M2 money demand function. As shown in Table 5.9, the trace test does not reject the presence of a cointegrating equation at 5% significant level, while a single cointegrating vector is rejected at 10% level according to the max-eigen test. If r = 1 is set, the cointegrating coefficients are all of expected signs. But just like in the M1 40 money demand function using id, the parameter β in id is insignificant at conventional levels, implying the interest rate does not belong to the cointegrating space. Table 5.9 Johansen Cointegration Test Results for M2 money demand function using id as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.4077 0.2613 0.1931 0.0373 53.96* 27.77 12.63 1.90 47.21 29.68 15.41 3.76 26.19 15.14 10.73 1.90 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M2 CPI RGDP id Cointegrating vector ( βˆ ) 1.000 -1.033 -1.864 -16.804 Adjustment vector ( αˆ ) -0.005 -0.024 0.138 0.008 Zero interest rate semi-elasticity (β4 = 0) χ2(1) = 1.362 (0.24) Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p=6 is selected based on the Sims LR statistics. The figure in parenthesis is p-value. 5.2.2.3 Cointegration test for M2, CPI, RGDP and the interest rate on one-year saving deposits (i) Table 5.10 reports the results of cointegration tests using i as the opportunity cost variable, which are similar to those using i6. A unique long run equilibrium relationship among the variables is estimated. There is evidence of the validity of price homogeneity hypothesis, but the hypothesis of unity income elasticity is rejected. The restricted cointegrating equation is: 41 M 2 = 11.673 + CPI + 1.958RGDP + 5.262i (5.6) Moreover, the results support the weak exogeneity of M2 and CPI, but reject the hypothesis for RGDP and i. Table 5.10 Johansen Cointegration Test Results for M2 money demand function using i as the opportunity cost variable A: Coinegration Rank Test Null Eigenvalue QTrace 5% C.V. QMax 5% C.V. r=0 r≤1 r≤2 r≤3 0.5066 0.2659 0.1794 0.0105 61.17** 25.86 10.41 0.53 47.21 29.68 15.41 3.76 35.31** 15.45 9.88 0.53 27.07 20.97 14.07 3.76 B: Cointegrating Parameter Estimates M2 CPI RGDP i Cointegrating vector ( βˆ ) 1.000 -0.928 -2.055 -6.354 Adjustment vector ( αˆ ) -0.034 0.007 0.129 0.046 C: Hypothesis Testing Null LR Statistics Zero interest rate semi-elasticity (β4 = 0) χ2(1) = 12.500 (0.00) Price homogeneity (β1 = -β2) χ2(1) = 0.246 (0.62) Unity income elasticity (β1 = -β2 = -β3) χ2(2) = 13.969 (0.00) Weak exogeneity of M2 (α1 = 0) χ2(1) = 0.601 (0.44) Weak exogeneity of CPI (α2 = 0) χ2(1) = 0.173 (0.68) Weak exogeneity of RGDP (α3 = 0) χ2(1) =8.860 (0.00) Weak exogeneity of i6 (α4 = 0) χ2(1) = 19.425 (0.00) Notes: Sample period is 1986:1-2004:4. *, **: significant at 5% and 1% level, respectively. p=6 is selected based on the Sims LR statistics. The figures in parenthesis are p-values. Comparing the magnitudes of the estimates across eq. (5.5) and eq. (5.6) reveals 42 that the magnitude of the semi-elasticity of i is slightly lower than that of i6, however both estimates are large compared to other elasticities in the equations. Note that similar coefficients on real GDP are obtained from these two equations. Therefore, the estimated long-run equilibrium relationship of money demand is reliable. Based on our analysis above, one long-run equilibrium relationship of broad money demand is specified, and the 6-month saving deposits rate is adopted as an appropriate measure of the opportunity cost of holding broad money. We can go further to analyze the short-run dynamics of the money demand function in the following part. 5.3 Short-Run Analysis of Money Demand Using Vector Error-Correction Model (VECM) Having estimated the error-correction term (the long-run relationship) for M2 money demand function based on the Johansen procedure, we then constructed a VECM to capture the short-run dynamics and the adjustment of the long-run static disequilibrium. An issue of interest is whether causality flows from money to price and real income, or goes from price and real income to money in the money demand function. Because both M2 and CPI are weakly exogenous, the answer is not so obvious and Granger causality tests are adopted to investigate the causality in the money demand function. 5.3.1 Granger Causality Tests Granger causality (Granger, 1969) measures whether current and past values of a variable can help to forecast future values of any other variables in the system. In a 43 cointegrated system, a variable yt does not Granger cause another variable zt if zt has no response to the discrepancy from long-run equilibrium and ∆zt does not respond to the changes in lags of ∆yt in the system. Causal links from the policy instruments (or the intermediate targets) to the final targets are essential for the central bank to conduct effective monetary policies. In particular, a necessary condition for feasible monetary policies is Granger causality from the instruments (or the intermediate targets) to the final targets. On the other hand, the past values of the final target variables may also have some effects on the current choice of the instruments and/or the targeting values of the intermediate target variables. Typically, policy analysis assumes no direct Granger causality from the final targets to the instruments and the intermediate target variables. This subsection conducts the pairwise Granger causality tests on the four-variable VECM including the error-correction term obtained from the Johansen procedure. The M2 money demand function comprises two final target variables (CPI and RGDP)7 and two instrument variables (M2 and i6). Through the causality tests, the policy effects using the alternative instrument are compared, which has important implications for policy analysis. Table 5.11 reports the results of the χ2 statistics for the joint significance of each of the other lagged dependent variables and the statistics for joint significance of all other lagged dependent variables in the VECM (the row of “All”). Results suggest that M2 can Granger cause price and real income, but past and current changes in nominal income have no effect on future money stock. Therefore, the causality goes from 7 As announced by the PBC, the objective of monetary policy in China is to maintain price stability and thereby promote economic growth. We therefore treat the price and real income as the final targets of the PBC. 44 money to price and real income, and not vice versa. Furthermore, since money does not feedback the discrepancy from the long-run equilibrium, either, we can conclude that money is an exogenous variable in the system. Our evidence supports the monetarist proposition of the exogeneity of the money supply, implying that it is that the monetary authorities determine the monetary aggregate. Table 5.11 Results of VEC Granger Causality Tests Dependent variable Zero Restriction LR Statistics p-value ∆M2 ∆CPI ∆RGDP ∆i6 All 2.728 7.304 2.944 10.971 0.74 0.20 0.71 0.75 ∆CPI ∆M2 ∆RGDP ∆i6 All 21.547 28.000 6.663 63.467 0.00 0.00 0.25 0.00 ∆RGDP ∆M2 ∆CPI ∆i6 All 26.929 37.446 7.724 64.030 0.00 0.00 0.17 0.00 ∆i6 ∆M2 ∆CPI ∆RGDP All 7.239 32.045 14.546 68.619 0.20 0.00 0.01 0.00 Notes: Sample period is 1986:1-2004:4. p = 5 is selected following the lag order of the cointegration tests. All refers to all other lagged dependent variables in the system. The lagged variables that are tested for exclusion are only those first differenced. The lagged level terms in the cointegrating equation (the error-correction term) is not tested. Some other causal relations among the variables can also be detected. For example, the causality runs from price and real income to the interest rate rather than the reverse. Although movements in M2 do not directly cause the interest rate, but could still 45 influence it indirectly through interactions with real income and price. Hence, indirect causality runs from M2 to the interest rate via real income and price. These findings conform to the fact that interest rate, which are infrequently adjusted to adapt to the changes in the economic environment, have not yet a very effective policy instrument until now. 5.3.2 Impulse Response Analysis Impulse response analysis can be a useful tool to investigate the interrelationships among the cointegrated variables in the system. Cholesky decomposition method requires an ordering of the variables in the VAR and responses may change dramatically if one changes the ordering of the variables. According to above analysis of the causal relations between the variables, we use the order M2, CPI, RGDP, and i6. In discussing the impulse responses one-standard deviation (SD) shock on each variable is given. Fig. 5.2 displays the dynamic responses for the four-variable VEC to an orthogonalized impulse in M2. A positive money shock has a lagged influence on price and leads to a permanent increase in price. Moreover, a transitory money shock on real income is not permanent but an obvious positive influence in the short-run can be observed. This is consistent with the hypothesis of long-run neutrality of money, and the empirical results suggest that money does have some temporary impact on the real economy in China. The response of the interest rate to a money shock is lagged and it actually reflects the needs of the PBC to adjust various interest rates to the whole 46 economy over a time span. Response of M2 to M2 Response of CPI to M2 .044 .05 .040 .04 .036 .03 .032 .02 .028 .01 .024 .00 5 10 15 20 5 10 15 20 Response of i6 to M2 Response of RGDP to M2 .012 .008 .006 .008 .004 .004 .002 .000 .000 -.004 -.002 5 10 15 20 5 10 15 20 Fig. 5.2 Response to Cholesky One S.D. impulse in M2 5.3.3 Seemingly Unrelated Regression (SUR) Estimates of the VECM We can simulate the short-run dynamics based on the discussion of Ganger causality and impulse responses. The seemingly unrelated regression (SUR) method, also known as the multivariate regression, or Zellner’s method, is used to estimate the parameters of the system, accounting for contemporaneous correlation in the errors across equations in the VECM. Moreover, the method can provide efficient estimates of the VEC coefficients if different regressors and lag lengths are included across equations. 47 Table 5.12 SUR Estimates of the VECM Dependent Variable Coefficient ∆CPI ∆RGDP ∆i6 Constant -0.015 [-5.28] -0.033 [-4.09] -0.010 [-7.97] vt-1 ─ 0.100 [3.95] 0.034 [7.02] ∆M2t-1 0.131 [2.83] ─ ─ ∆M2t-3 ─ 0.480 [5.06] ─ ∆M2t-4 0.183 [3.98] ─ ─ ∆M2t-5 ─ ─ ─ ∆CPIt-1 0.681 [7.52] -0.892 [-5.23] 0.137 [4.01] ∆CPIt-2 0.331 [3.02] ─ ─ ∆CPIt-3 -0.267 [-2.81] ─ 0.172 [3.66] ∆CPIt-4 ─ 0.587 [2.67] 0.119 [2.55] ∆CPIt-5 ─ 0.866 [3.90] ─ ∆RGDPt-1 0.124 [2.77] -0.318 [-3.22] 0.059 [7.68] ∆RGDPt-2 ─ ─ 0.026 [4.99] ∆RGDPt-3 0.125 [14.19] ─ 0.019 [3.42] ∆RGDPt-4 ─ 0.790 [12.76] ─ ∆RGDPt-5 -0.143 [-3.27] 0.472 [5.20] ─ D1t ─ -0.146 [-3.94] ─ D2t -0.047 [-3.16] ─ ─ D3t -0.058 [-9.92] -0.092 [-5.53] ─ R2 0.892 0.992 0.455 DW 1.927 2.283 1.926 Ljung-Box (8) 5.324 (0.72) 3.862 (0.87) 9.162 (0.33) Ljung-Box (16) 13.733 (0.62) 7.211 (0.97) 17.882 (0.33) Jarque-Bera 2.527 (0.28) 1.353 (0.51) 15.982 (0.00) Notes: Sample period is 1986:1-2004:4. vt-1 denotes the lagged long-run error obtained from the cointegration equation (5.5). Dit, i = 1,2,3 are the centered seasonal dummies. The figures in brackets are the t-statistics, and those in parenthesis are p-values. 48 The results are shown in Table 5.12. Due to the exogeneity of M2, only three equations are estimated in the system. All the insignificant regressors and lags are eliminated to yield a parsimonious specification. For the RGDP and the interest rate equations, the coefficients of the error-correction term are both significant at 5% level with the expected positive sign, indicating that real GDP and the interest rate will respond to a previous error and adjust to restore equilibrium. The estimated adjustment coefficients are similar to those obtained from the cointegration tests. The adjustment coefficient of the CPI equation significantly differs from zero, implying that CPI is weakly exogenous. The residuals of each equation pass the autocorrelation tests as suggested by the Ljung-Box statistics. The Jarque-Bera normality tests indicate the residual terms are normal distributed with 5% significant level except for the interest rate. These diagnostic tests suggest that the residuals of the three-equation ECM are well behaved. The results from our dynamic model indicate that money forces can exert a perceptible and predictable influence on the movement of inflation and output growth in China. In a word, M2 can be a viable policy variable in the monetary decision and it is valid for the PBC to adopt M2 as its intermediate target at present. 49 6. Conclusion and Policy Implications This thesis studies the money demand function in China for various monetary aggregates (M0, M1, M2) over the post-reform period 1986-2004. The Johansen maximum likelihood procedure was employed to test for the long-run equilibrium relationships among monetary aggregates, the CPI, real GDP and some measures of the opportunity cost of holding money balances. Furthermore, we estimated the dynamic adjustment mechanism of money demand based on a vector error-correction model (VECM). Before doing the cointegration tests, the unit root tests were conducted and the results show that M0 is stationary hence excluded from the cointegration analysis. For the opportunity cost of holding money, two sets of measures were considered: the nominal interest rate measure; the expected inflation measure πe (= πt ), where π is 1+ πt the year-to-year inflation rate. Since the expected inflation is stationary, the interest rate measure was used as a proxy for the opportunity cost variable in the long-run money demand function. The results of the Johansen procedure indicate that there exists a long-run money demand function for broad money with the opportunity cost variable proxied by the 6-month saving deposits interest rate over the sample period. Tests on the hypotheses of the weak exogeneity, price homogeneity and unit income elasticity were conducted and accordingly we proposed the long-run money demand function for M2. The short-run dynamics of the money demand function were specified on the basis 50 of the long-run equilibrium analysis. The Granger causality tests and the impulse response analysis were undertaken to investigate the interrelationships among the cointegrated variables in the system. Our evidence suggests that the causality flows from money to price and real income, and supports the hypotheses of the exogeneity and long-run neutrality of money. Further, a vector error-correction model (VECM) was constructed and the seemingly unrelated regression (SUR) method was used to catch the dynamic adjustment mechanism in money demand. Our empirical results have some important implications to the conduct of monetary policy in China. As suggested by the empirical results, it is feasible for the PBC to use M2 as its intermediate policy target due to the presence of the stable long-run money demand equation for broad money. Moreover, it is of importance for the PBC to set proper monetary target and keep stable money growth since an important influence of money forces on the movement of inflation and output growth in China. Finally, there exists some limitation in our analysis. Due to only annual CPI data available before 1986, our sample period is 1986:1 to 2004:4. A few dozens of observations may result in lack of power for the tests employed in the paper. Future research may include more data points to further investigate the stability of money demand functions in China. 51 REFERENCES Aghevli, Bijan B., Mohsin S. Khan, P. R. Narveker and Brock K. Short (1979), “Monetary Policy in Selected Asian Countries.” IMF Staff Papers 26, 775-824. Arize, A. C. and John Malindretos (2000), “Does Inflation Variability Affect the Demand for Money in China? Evidence from Error-Correction Models.” International Economic Journal 14 (1), 47-60. Boughton, James M. (1992), “International Comparisons of Money Demand: a Review Essay.” Open Economies Review 3, 323-343. Blejer, Mario, David Burton, Steven Dunaway and Gyorgy Szapary (1991), “China: Economic Reform and Macroeconomic Management.” IMF Occasional Paper 76. Brandt, Loren and Zhu, Xiaodong (2000), “Redistribution in a Decentralized Economy: Growth and Inflation in China under Reform.” Journal of Political Economy 108 (2), 422-439. Burton, David and Jiming Ha (1990), “Economic Reform and the Demand for Money in China.” IMF Working Paper 90/42. Calvo, Guillermo and Leo Leiderman (1992), “Optimal Inflation Tax under Precommitment: Theory and Evidence.” American Economic Review 82 (2), 179-195. Chen, Baizhu (1997), “Long-run Money Demand and Inflation in China.” Journal of Macroeconomics 19(2), 609-617. Cheung, Yin-Wong and Kon S. Lai (1993), “Finite-Sample Sizes of Johansen’s Likelihood Ratio Tests for Cointegration.” Oxford Bulletin of economics and 52 statistics 55 (3), 313–328. Chow, Gregory (1987), “Money and Price Level Determination in China.” Journal of Comparative Economics 11, 319-333. Deng, Shuhui and Liu, Bin (1999). “Modelling and Forecasting the Money Demand in China: Cointegration and Nonlinear Analysis.” Annals of Operations Research 87, 177-189. Dickey, David A., and Wayne A. Fuller (1979), “Distribution of the Estimators for Autoregressive Time Series with a Unit Root.” Journal of the American Statistical Association (74), 427-31. Dickey, David A. and Wayne A. Fuller (1981), “Likelihood Ratio Statistics for Autoregressive Times Series with a Unit Root.” Econometrica 49, 1057-1072. Doldado, Juan, Jenkinson Tim and Simon Sosvilla-Rivero (1990), “Cointegration and Unit Roots.” Journal of Economic Surveys 4, 249-273. Easterly, William R. Paolo Mauro and Klaus Schmidt-Hebbel (1995), “Money Demand and Seigniorage-Maximizing Inflaition.” Journal of Money, Credit and Banking 27 (2), 583-603. Enders, Walter (2004). Applied Econometric Time Series (Second ed.). New York: Wiley. Engle, Robert F., and Clive W. J. Granger, “Cointergration and Error-Correction: Representation, Estimation and Testing.” Econometrica 55 (2), 251-276. Goldfeld, Stephen M. and Daniel E. Sichel (1990), “The Demand for Money.” Chapter 8 in B. M. Friedman and R.H. Hahn (eds.), Handbook of Monetary Economics. 53 North Holland. Gregory, Allan and Bruce Hansen (1996), “Residual-Based Tests for Cointegration in Models with Regime Shifts.” Journal of Econometrics 70, 99-126. M.L. Luke Chan, Wang, Cheng and Richard Deaves (1991), “Money Demand in China Revisited: Some New Empirical Evidence.” Journal of Asian Economics 2 (1), 137-144. Feltenstein, Andrew and Ziba, Farhadian (1987), Fiscal Policy, Monetary Targets, and the Price Level in a Centrally Planned Economy: An Application to the Case of China.” Journal of Money, Credit and Banking 19 (2), 137-156. Ghatak, Subarata (1995), Monetary Economics in Developing Countries (Second ed.). New York: St.Martin's Press. Girardin, Eric (1996), “Is There a Long Run Demand for Currency in China?” Economics of Planning 29, 169-184. Gonzalo, Jesus (1994), “Five Alternative Methods of Estimating Long-Run Equilibrium Relationships.” Journal of Econometrics 60, 203-233. Granger, C. W. J. (1969), “Investigating Causal Relations by Econometric Models and Cross-Spectral Methods.” Econometrica 37(3), 424–438. Gu, Chunhui (2004), “Empirical Analysis of Money Demand in China: A Cointegration Approach.” PhD Dissertation, American University. Hafer, R. W. and A. M. Kutan (1994), “Economic Reforms and Long-run Money Demand in China: Implications for monetary policy.” Southern Economic Journal 60(4), 936-945. 54 Johansen, Soren and Katarina Juselius (1990), “Maximum Likelihood Estimation and Inference on Cointegration-with Applications to the Demand for Money.” Oxford Bulletin of economics and statistics 52(2), 169-210. Johansen, Soren (1991), “Estimation and Hypothesis Testing of Cointergration Vectors in Gaussian Vector Autoregressive Models.” Econometrica 59(6), 1551-1580. Hansen, B. E. (1992), “Tests for Parameter Instability in Regressions with I(1) Processes.” Journal of Business and Economic Statistics 10, 321-335. Hendry, D. F. (1980), “Predictive Failure and Econometric Modeling in Macroeconomics: The Transactions Demand for Money.” In Economic Modeling. London: Heinemann Education Books. Hodrick, R. J. and E. C. Prescott (1997), “Postwar U.S. Business Cycles: An Empirical Investigation.” Journal of Money, Credit, and Banking 29, 1-16. Huang, Guobao (1994), “Money Demand in China in the Reform Period: An Error Correction model.” Applied Economics 26, 713-719. Ma, Guonan (1993), “Macroeconomic Disequilibrium, Structural Changes, and the Household Savings and Money Demand in China.” Journal of Development Economics 41, 115-136. MacKinnon, J. G. (1991), “Critical Values for Cointegration Tests.” Chapter 13 in R. F. Engle and C. W. J. Granger (eds.), Long-run Economic Relationships: Readings in Cointegration. Oxford University Press. Osterwald-Lenum, Michael (1992), “A Note with Quantiles of the Asymptotic Distribution of the Maximum Likelihood Cointegration Rank Test Statistics.” 55 Oxford Bulletin of Economics and Statistics 54, 461-472. People’ Bank of China (2004), “Monetary Policy Report, 2003”. People’ Bank of China (2006), “Monetary Policy Report, 2005 Quarter 4”. Perron, P. (1989), “The Great Crash, the Oil Price Shock and the Unit Root Hypothesis.” Econometrica 57, 1361-1401. Perron, P. (1997), “Further Evidence on Breaking Trend Functions in Macroeconomic Variables.” Journal of Econometrics 80, 355-385. Philips, P. C. B. and B. E. Hansen (1990), “Statistical Inference in Instrumental Variables Regression With I(1) Processes.” The Review of Economic Studies, 99-125. Philips, P. C. B. (1991), “Optimal Inference in Cointegrated System.” Econometrica 59, 283-306. Portes, Richard and Anita Santorum (1987), “Money and the Consumption Goods Markets in China.” Journal of Comparative Economics 11, 354-371. Qin, Duo (1994), “Money Demand in China: The Effect of Economic Reform.” Journal of AsianEconomics 5 (2), 253-271. Qin, Duo; Pilipinas, Quising; He, Xinhua and Liu, Shiguo (2005), “Modelling Monetary Transmission and Policy in China.” Journal of Policy Modelling 27, 157-175. Reinsel, Gregory C. and Sung K. Ahn (1992), “Vector Autoregressive Models with Unit Roots and Reduced Rank Structure: Estimation, Likelihood Ratio Test, and Forecasting.” Journal of Time Series Analysis 13(4), 353–375. 56 Shi, Jianhuai (2001), “Financial Innovations in China, 1990-2000.” China Center for EconomicResearch Working Paper Series (E2001006). Sims, C.A (1980), “Macroeconomics and Reality.” Econometrica 48, 1-49. Smithin, John (2003), “Controversies in Monetary Economics, Revised Edition”, Edward Elgar. Stock, James H. and Mark W. Watson (1993), “A Simple Estimator of Cointergrating Vectors in Higher Order Intergrated Systems.” Econometrica 61(4), 783-820. Xu, Yingfeng (1998), “Money Demand in China: A Disaggregate Approach.” Journal of Comparative Economics 26, 544-564. Yi, Gang (1991), “The Monetization Process in China during the Economic Reform.” China Economic Review 2(1), 75-95. Yu, Qiao and Xie, Ping (1999), “Money Aggregates Management: Problems and Prospects in China’s Economic Transition.” Contemporary Economic Policy 17 (1), 33-43. Yu, Qiao and Albert Tsui (2000), “Monetary Services and Money Demand in China.” China Economic Review 11, 134-148. Zhang, Yin and Wan, Guanghua (2004), “Output and Price Fluctuations in China’s Reform Years: What Role did Money Play?” World Institute for Development Economics Research Research Paper 2004/56. Zivot, Eric and Donald W. K. Andrews (1992), “Further Evidence on the Great Crash, the Oil-Price Shock, and the Unit-Root Hypothesis.” Journal of Business and Economic Statistics 10(3), 251-272. 57 [...]... Bank of China, Bank of China, People's Construction Bank of China, as well as Industrial and Commercial Bank of China Each of the four state banks was to conduct commercial banking business in a specialized part of economy1 At the same time, two nonbank financial institutions (NBFIs) –People’s Insurance Corporation of China and China International Trust and Investment Corporation2 – were established Since... asset investment and Industrial and Commercial Bank of China for urban industrial and commercial 2 People’s Insurance Corporation of China specialized in insurance business and China International Trust and Investment Corporation was in charge of joint ventures 4 interventions by the state and local governments before 1994 In 1994, three policy-lending banks 3 were set up to make policy lending determined... The studies of money demand functions in China started since Chow (1987) pioneered an econometric exercise to apply the quantity theory of money to China using the partial adjustment model on annual data over the sample period of 1952-1983 He concluded that the quantity theory is valid in China and the estimated model provides “a reasonable first approximation" to explain money demand in China Chow’s... brought much attention to the study of money demand and monetary policy in China in the following decades For instance, Portes and Santorum (1987) and Chan et al (1991) extended Chow’s study by establishing more general specification of money demand functions Feltenstein and Farhadian (1987) estimated a money supply and real balance demand function of a planned economy of China with price control Ma (1993)... creation of new financial markets, such as money market and capital market, as well as the introduction of new financial instruments, such as treasure bond and stock 2.1 Financial Institutional Reforms In the early years of the reforms, decentralization of the monobank system was a main task of the financial reforms Up to 1994, the monobank system had been split 3 into a central bank- People's Bank of China. .. role in the macroeconomic activities The bank credit is still the dominant channel of financing in China, through which about 80% of the total amounts of the financing inject into the non-financial institutions in recent years (Table 2.1) 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 m2/gdp household saving deposits/gdp Fig.2.1 The Financial... power Thus, nominal money demand changes in proportion to the change in price level Moreover, real money demand depends positively on real income and the associated volume of transactions, and negatively on the opportunity cost of holding money The former embodies “transaction demands” for money in the income version, and in the latter argument, money is treated as one among a number of alternative... rate on demand deposits (id) The one-year saving deposits rate (i) is also used for comparison Using these nominal interest rates is not ideal since the official interest rates are often regulated by the government and vary very infrequently in China Furthermore, these measures are not suitable as the opportunity cost of holding broad money aggregate (M2) since these interest rate variables are in fact... money demand, which comprises a transaction variable measuring the volume of real economic activity, for instance, real national income, and one and more measures for the opportunity cost of holding money, such as the short interest rate of some kind or the expected inflation rate Theory suggests that real money demand is unchanged when price level rises since individuals hold money for real purchasing...2 Overview of Financial Reforms in China China’s financial system has developed dramatically since the economic reforms in 1979 Before the reforms, the People's Bank of China (PBC) functioned as a central bank and commercial banks The financial sector was small in relation to the economy and played a small role in the economic activities of the centrally planned economy As pointed out by Yu and ... Bank of China, Bank of China, People's Construction Bank of China, as well as Industrial and Commercial Bank of China Each of the four state banks was to conduct commercial banking business in. .. of China for urban industrial and commercial People’s Insurance Corporation of China specialized in insurance business and China International Trust and Investment Corporation was in charge of. .. the money demand in China during the pre-reform period and not the post-reform period beginning from 1979 The objective of this thesis is to investigate the behavior of money demand in China