Measuring the Monetary Policy Stance of the People s Bank of China: an Ordered Probit Analysis.

Transcription

1 Measuring the Monetary Policy Stance of the People s Bank of China: an Ordered Probit Analysis. Weibo Xiong 3 rd year PhD student. School of Management and Languages, Heriot-Watt University Supervisors: Professor David Cobham; Dr. Atanas Christev September 2009 Abstract: In order to capture how the People s Bank of China (PBC) reacts to economic situations, we develop a new policy stance index and examine it in an ordered probit model, following the approaches of Gerlach (2004), and He and Pauwels (2006). Then we examine the relationship between monetary policy stance and the PBC s own outlook of the economic situation as given in the Quarterly Monetary Policy Executive Report by the PBC from 200Q. The empirical results show that the policy index reacts mainly to output growth, the growth of broad money supply, and the accumulated level of the monetary policy stance. On the other hand, the empirical results suggest that the PBC s policy changes are less closely tied to CPI inflation and the nominal effective exchange rate. When the indicator variables are considered, it seems that only the PBC s own interpretations of money supply play a significant role in determining the PBC s monetary policy decisions. I am grateful to my supervisors, Professor David Cobham and Dr. Atanas Christev at Heriot-Watt University, as they provide plenty of supports and give many helpful comments and suggestions to this paper. The views expressed are those of the author, and no responsibility for the mistakes in this paper should be attributed to the two people mentioned above.

2 Table of Contents Abstract:... Table of Contents Introduction Literature Review Constructing a Measure for the PBC s Policy Stance: A Policy Change Index Identifying Policy Instruments in the Policy Change Index Identifying the Direction of the Policy Change Index Some Issues Related to the Policy Change Index A Historical Review based on the Policy Change Index The Policy Change Index in an Ordered Probit Model Model and Data Some Empirical Results Further Analysis of the Model Indicator Variables in the Ordered Probit Model Conclusions Reference Appendix A: Real Activity Indicator Appendix B: Inflation Indicator Appendix C: Money Indicator

3 . Introduction What Napoleon Bonaparte once said has been widely quoted: Let China sleep. For when China awakes it will shake the world. Indeed, there has been increasing attention to China s development from many other countries since the last couple of decades, especially after the reform in 978, and during and after the latest global economic recession in This is not only because of China s impressive domestic economic achievements, but also because of its growing influence outside on global political and economic life. As an important component of the framework of China s macroeconomic policy, the monetary policy operations of the People s Bank of China (PBC) have also become a great concern to the rest of the world. One example is the disagreements since the 990s over China s foreign exchange rate policy between China and its main trading partners such as the US and the EU, for whom the value of the RMB is a cause for concern over cheaper and more competitive Chinese products, and hence by whom the foreign exchange rate has been regarded as an important monetary policy issue. However, despite the growing importance of monetary policy in countries like China, the current literature on monetary policy is still centred on those developed countries which have an advanced central banking system or a well-functioning money market, like the US, the UK, or Germany. One possible reason for this bias is that the approaches, theories and conclusions developed and used for the institutional framework and markets of the developed countries may not be easily applicable to countries like China. Another possible explanation is that, as many of the emerging countries are still in a process of economic transition, data problems impose numerous restrictions on the application of modern econometric approaches. These problems include the lack of some necessary economic indicators at a high time frequency, or the limited observation period for some economic data. Despite disadvantages, we will focus on two fundamental questions in examining monetary policy in China. First, is there a suitable indicator that can be used to 3

4 appropriately measure the PBC s monetary policy stance since the 990s? To obtain such a measure while taking into account some implicit features of China s monetary policy organization, we look at different monetary policy instruments in different periods, as most current studies only see the link between the changes in a single policy instrument and the macroeconomic situation in considering monetary policy stance. The second question we consider is, once the policy index as a measure of the monetary policy stance is found, what is the connection between this index and the state of the economy? Instead of the standard OLS analyses, the main empirical results in this paper are obtained in the framework of an ordered probit model. In addition, to study the relationship between the monetary policy stance and the PBC s own interpretations of the state of the economy, we attempt to extract information on the PBC s outlook and assessment from statements in the PBC s Quarterly Monetary Policy Executive Report since its first issue in 200Q. It seems that this is the first time such an exercise has been done for China. We hope that our search for the answer to the second question may shed some light on China s monetary policy reactions, as well as helping in understanding and predicting the making of monetary policy by the authorities. The structure of this paper is as follows. Section 2 provides a review of the literature on the monetary policy stance, and the role of policy stance measures in examining monetary policy operations. Section 3 explains the construction of an index for measuring the change in the stance of monetary policy in China from 99Q to 2008Q4. Section 4 presents some empirical results on the relationship between the policy index and some main economic variables in an ordered probit model. It then examines the links between the policy index and the PBC s interpretations of the state of the economy between 200Q and 2008Q4 as represented by some narrative indicator variables. Section 5 gives a summary of the paper s findings. 4

5 2. Literature Review A good measure of the monetary policy stance should provide us with reasonable and straightforward information concerning the reactions of monetary policy. Bernanke and Mihov (995) provide a general review of the measures of monetary policy stance which have been widely used by other scholars. Initially the rate of growth of monetary aggregates, such as M, M2 or the monetary base, was used to measure the policy stance. But it was then argued that the link between the changes in these variables and monetary policy implementation is complicated, as one disadvantage of this measure is that the money growth rate also reflects other noisy information. For example, broad money M2 growth reflects information about money demand, changes in velocity, the central bank's special preferences in setting the short-term interest rates, or other factors which are not relevant to determining the monetary policy stance. Therefore, Friedman and Schwartz (963) and Romer and Romer (989) introduced a narrative approach to capture the monetary policy stance. They read the documents, statements and publications issued by the monetary authorities to extract information indicating when there was a change in the monetary policy stance. To improve the performance of the narrative approach, Boschen and Mills (99) extended the measure of policy stance to five categories, from strongly expansionary to strongly contractionary. As Bernanke and Mihov (995) point out, this narrative approach is often criticised on the grounds that it depends on the subjectivity of the researcher who reads the documents and measures the policy. At the same time, the policy stance obtained from this narrative approach cannot distinguish an exogenous component of a policy change from an endogenous one, although the exogenous part of monetary policy (also known as a policy shock) is important in identifying policy effects on the economy according to rational expectations theory. Another potential weakness of the narrative approach is that it might focus too heavily on the intentions of the monetary authorities, while their actual policy actions may diverge from their 5

6 intentions. So this approach may produce a misleading result in the analysis of monetary policy reactions. Another way to measure the monetary policy stance, apart from the approaches of using a single macro variable or subjective understanding of relevant documents, is to look at the monetary operational procedures of the monetary authorities, which can be called the index approach. Bernanke and Blinder (992) use Granger causality tests with US data from 959 to 989. They show that it is the Federal funds rate which is the best predictive variable compared to M, M2, and the Treasury bill rate. Bernanke and Blinder (992) also develop their index from the changes in the Federal funds overnight rate for commercial bank reserves. Since in the US legal reserve requirements on deposits provide the Federal Reserve with considerable direct leverage over the quantity of funds banks may obtain, Christiano and Eichenbaum (992) use the quantity of nonborrowed reserves as the measure of the policy stance. In their influential paper Measuring Monetary Policy, Bernanke and Mihov (995) develop a policy stance indicator which incorporates the total reserves, nonborrowed reserves and funds rate. They take a linear combination of those variables as the policy stance and examine its relationship with the state of the economy in a vector autoregressive model. In fact, policy rates such as the Federal funds rate and the repo rate are widely treated as measures of monetary policy stance in a great deal of literature on monetary policy rules, for example, the Taylor rule and its variations. Rather than considering a continuous change in interbank deposit rates, Gerlach (2002, 2006) measures the policy stance of the European Central Bank (ECB) in a discrete way. In his model, the monetary policy stance is measured by the changes in the repo rate, where a 0.25 percent rise (cut) of the repo rate is considered as a small rise (cut), and a 0.5 percent rise (cut) as a large rise (cut) one. Then the monetary policy stance can be coded as a series of discrete numbers from -2 to 2. In the current literature focusing on the empirical analysis of the PBC s monetary policy, one problem in identifying the policy stance is that it seems to be difficult to See Bofinger, 200 pp 86. 6

7 find a measure of the monetary policy stance for the PBC that can be used throughout the last couple of decades. Firstly, as many scholars have argued, the intermediate policy target, ether M or M2, may not be a good measure of the PBC's policy stance. For example, the changes in the broad money supply may reflect endogenously foreign capital inflows or export revenues. Since a part of the increased money has been sterilized by the PBC, the relationship between the growth of broad money after sterilization operations and the PBC s policy actions remains complicated and unclear, and critically depends on whether the central bank s sterilization operations are successful. In addition, He and Pauwels (2008) argue that even the targets for broad money announced by the PBC are not suitable as an indicator of the policy stance, as those targets are announced at an annual frequency and are not continuously adjusted to the state of the economy. Second, interest rates are not yet a good measure of monetary policy. Laurens and Maino (2007) refer to some factors which reduce the efficiency of the interest rate transmission channel as a monetary policy instrument, for example, the uncompleted restructuring of the banking system, and the insensitive response of firms and individuals to the price of capital. These factors mean the PBC cannot rely on the interest rate instrument to influence the level of economic activity or aggregate demand. In addition, short-term interbank rates in the money market are not necessarily a good measure of the policy stance largely because of the segmentation of the credit market. Hence, despite the huge interest rate reforms during the last two decades, the changes in interest rates alone are not capable of representing fully the change in the PBC's monetary policy actions. Due to the lack of an integrated measure of the policy stance, some recent empirical works which focus on China's monetary policy stance have had no choice but to look at a single variable, such as the growth rate of the broad money supply M2, or changes in some policy interest rate, and use it throughout the total sample period. So identifying the policy stance based on those indicators is likely to omit or misplace important information on actual policy actions, and then lead to misleading results when modelling the PBC's monetary reaction function. 7

8 In order to measure the monetary policy stance for China more accurately, He and Pauwels (2008) take the discrete changes in the reserve requirement ratio, lending and deposit rates, and magnitudes of open market operations together to measure the monetary policy stance for the PBC from 997 to In their policy index they assume that all the policy instruments share the same weight in determining the direction of the policy movement. In addition, some empirical results are provided in the paper, which suggest that the monetary policy index they construct from observed changes in monetary policy instruments can be well explained by the objectives of policy, especially the CPI inflation. On the other hand, Shen and Chen (995) assign different levels of priority to those policy instruments in the policy index they construct for the central bank of Taiwan. For example, the policy stance is firstly determined by the reserve requirement ratio; it is then determined by changes in the discount rate if no change in reserve requirement ratio is observed, and so on. By providing a set of sequential rules the importance of several policy instruments are distinguished in their policy index. 8

9 3. Constructing a Measure for the PBC s Policy Stance: A Policy Change Index Generally, a policy index should reflect the movements of some important monetary policy instruments which can be controlled directly and effectively by the central bank, and should have a clear and stable relationship with the state of the economy. In spite of the shortcomings of each single policy instrument as a measure of policy stance as mentioned earlier, it is important to note that the main monetary policy instruments the PBC depends on have been varying over time. For instance, the credit plan was used as the main monetary policy instrument during most of the 990s. But it was abandoned in 998 and replaced by central bank refinancing to the commercial banks. Then from 2003 on, the PBC's open market operations began to play an important role as the main monetary policy instrument. Therefore, the variation in monetary policy instruments are an important element to be dealt with in constructing a policy index, and this necessitates separating the entire period into several different subperiods based on shifts between policy instruments. Another key feature of the PBC's policy implementation is that the PBC appears to prefer a combination of different monetary policy instruments in some periods. For example, from 200 to 2002, both the central bank refinancing and open market operations via transactions of government bonds were used as important instruments, while the latter were aimed at sterilizing the increased liquidity caused by trade surpluses and foreign direct investments (FDI). Then during the reserve requirement ratio and open market operations were frequently used together by the PBC to absorb the excessive liquidity. Consequently, given the inherent limitations of each monetary policy instrument alone as a measure of the policy stance, as well as the characteristics of the PBC s monetary policy operations, it would be useful and appropriate to develop a measure of the monetary policy stance which is capable of capturing as much information on 9

10 the monetary policy actions as possible. One plausible way to do this is to consider the movements of all the important monetary policy instruments in measuring the PBC's policy stance. The details of constructing such an index are discussed in the rest of this section. 3. Identifying Policy Instruments in the Policy Change Index As mentioned earlier, the PBC has been relying on different policy instruments over time, so we start by identifying the set of main monetary policy instruments from 99Q to 2008Q4. Presumably, the changes in those policy instruments together can sufficiently signal the policy stance of the PBC within the periods when such policy actions were observed. The policy instruments considered in the policy index of different periods are reported in Table 3.. Table 3. Monetary policy instruments considered in the policy change index: 99Q-2008Q4. Period Main monetary policy instruments 99Q-997Q4. Credit plan for banks lending; 2. Various interest rates. 998Q-2000Q4. Central bank s refinancing to the commercial banks; 2. Various interest rates; 3. Reserve requirement ratio. 200Q-2002Q2. Central bank s refinancing to the commercial banks; 2. Various interest rates; 3. Open market operations (treasury bonds). 2002Q3-2008Q4. Open market operations (central bank bills); 2. Various interest rates; 3. Reserve requirement ratio. Before 998 the credit plan was still the PBC's main instrument for controlling credit and money supply. During that subperiod, the PBC could theoretically control the money supply by imposing credit controls on banks lending according to the 0

11 credit plan. However, relevant data and information on the credit plan is not available. Instead, we use the item claims of the banking system on non-government sectors on the balance sheet of the entire banking system in China to measure bank credit. Since bank financing was the only source of finance available to the public during that time, it is reasonable to suppose that changes in bank credit reflected the credit plan which was used as the main policy instrument during this period. It is often argued, however, that credit quotas under the credit plan were usually exceeded by State banks excessive credit creation, which made the credit plan less credible as a policy instrument. Due to strong economic growth and the very low cost of capital, the aggregate demand for capital by firms, farmers and government agencies was far beyond the credit quotas and the banks had to provide excess credit under political pressure from local governments. 2 So the credit plan alone may not be enough to measure the policy stance. Therefore, we incorporate another conventional policy tool in our policy index besides the credit plan: the changes in various interest rates, which were under a strict administrative control from the PBC. In order to eliminate the excess demand for capital, the interest rate tool was occasionally used by the PBC where an interest rate rise could increase the cost paid by borrowers. Although in fact the PBC depended less on the interest rate tools during that time, changes in interest rates could be treated as signals sent by the PBC on its policy intentions. For the same reason, changes in interest rates are considered in the policy index throughout the overall sample period. Secondly, after 998 when the credit plan was abandoned and before 2003 when open market operations came into regularly use the direct lending from the PBC to the commercial banks became a main policy instrument during the subperiod from 998Q to 2000Q4. In order to measure the magnitude of the direct lending by the PBC, we look at the changes in the item the PBC s claims on other depository corporations on the PBC s balance sheet. This item reflects the standing facilities where liquidities are provided by the PBC to the commercial banks and other depository institutions. It consists of two parts: the central bank refinancing and the 2 See Guo, 2002, pp 38.

12 discount facilities, of which the refinancing volume and the discount rate can be determined by the PBC. At the same time, as the reserve requirement ratio was introduced in 998, the changes in this ratio are also considered in the construction of the policy index. So in the second row of Table 3., three policy instruments are identified for the subperiod 998Q-2000Q4: the PBC's refinancing, the reserve requirement ratio, and the interest rates. Then from 200 open market operations (OMOs), which are considered a more market-based policy instrument, were frequently used as one of the PBC's monetary policy instruments. It is noteworthy, however, that the open market operations are different before and after 2002Q2. Initially, the PBC s OMOs were implemented by selling and buying treasury bonds. In September 2002, the outstanding repo contracts signed between June and September in 2002 were converted into central bank bills. From 2002Q3, the PBC began to issue central bank bills as the main OMOs instrument, which is used for sterilization operations related to interventions in the foreign exchange market. From 2002Q3 to the end of the sample period, central bank bills transactions, the interest rate changes and the reserve requirement ratio are examined to determine the direction of the monetary policy movements. 3.2 Identifying the Direction of the Policy Change Index Following most of the literature which uses the index approach to measure monetary policy stance (Bernanke and Blinder, 992; Christiano and Eichenbaum, 992; Gerlach, 2002, 2006; He and Pauwels, 2008), we define the monetary policy stance in terms of a triple-choice set, where r t denotes the policy index; the value -, 0, and indicate an expansionary change, no change, and a contractionary change in the policy stance respectively. r t -, if there is an expansionary change 0, if there is no change (3.), if there is a contractionary change One problem is that it is difficult to identify the magnitudes of some policy 2

13 4-quarter % change instrument changes. Usually, the policy stance can be identified from the direction of changes in the price-based or ratio-based policy instruments, for example, an interest rate raise or cut, or changes in the reserve requirement ratio. However, there is no agreement on the standards by which changes in quantity-based policy instruments should be identified as a tighter or a looser or an unchanged policy stance. Hence we have to establish some criteria or thresholds for these changes, which are necessarily subjective and artificial. When the claims of the banking system on other non-government is used to measure the changes in the credit plan during 99Q-997Q4, we look at its four-quarter percentage change, that is, the credit provided by the banking system of this quarter relative to that in the same quarter of the previous year. This eliminates any seasonality effect. Then by looking at the general trend in the series of this percentage change, some remarkable changes in credit growth can be observed, which are taken to indicate the monetary policy actions of the PBC. For example, from 992Q4 to 993Q, credit growth increased sharply from to percent; while from 993Q3 to 993Q4, it jumped from to percent. Hence the significant increase in this number suggests that monetary policy was loosened by the PBC. On the other hand, from 994Q3 to 994Q4 credit growth fell rapidly from 3.39 to 2.2 percent, indicating a shrinking of the credit supply by the PBC through the commercial banks. More details can be seen in Figure 3. below Figure 3. Claims of the banking system on other non-government sector: 99Q-997Q4. 3

14 In fact, due to the tremendous public demand for credit it is reasonable to infer that the credit supply was easy to expand for the commercial banks, but hard to reduce for the PBC. So in 993Q2 and Q3 the rapid credit growth which follows the growth in 993Q is regarded as driven by endogenous factors, and the change in the policy stance is identified as unchanged. In other cases, if the percentage change of the credit growth varies in a relatively small range, usually less than five percent, that quarter is classified as having no policy change. The same approach is used to identify changes in the policy stance from the central bank's refinancing, as measured by claims on other depository corporations of the PBC, for the second and third subperiods in Table 3.. Another quantitative policy instrument, the open market operations by the PBC, is treated in a different way. We adopt the criterion by He and Pauwels (2008) that if a net injection (withdrawal) of central bank bills is larger than 00 billion Yuan (the unit of Chinese currency), the policy stance is then identified as looser (tighter). He and Pauwels (2008) justify this on the grounds that this amount is roughly equivalent to an average change of the reserve requirement ratio of 0.25 percent at that time. They also argue that such a criterion is a reasonable way to balance the risk of over-identifying and under-identifying when changes in open market operations truly signal changes in policy stance. However, in their policy index they do not consider injections or withdrawals through treasury bond transactions in the years before 2002 when central bank bills came to be frequently issued. In order to do that, we introduce another arbitrary criterion for open market operations before 2002, that a expansionary policy stance is identified if the incremental volume of the treasury bonds held by the PBC is larger than 50 billion Yuan. The reason we choose this criterion is that the typical range of the net change in the PBC s claims on the government from 200Q to 2002Q2 is roughly one half of that of the net change in central bank bills issued in the first four quarters since central bank bills were issued in 2002Q3, so the threshold is accordingly cut by a half. Table show the changes in the various policy instruments and the policy change index is recorded in the last column of each table. 4

15 Table 3.2 Monetary policy index: 99Q-997Q4. Time Credit growth % Interest rate change Policy change index 99Q ( ) 0 99Q ( ) discount rate (-0.72); deposit rate (-.08); loan rate (-0.72) - 99Q ( ) 0 99Q ( ) 0 992Q ( ) 0 992Q ( ) 0 992Q3 2.8 ( ) 0 992Q ( ) 0 993Q ( ) - 993Q ( ) discount rate (+.26); deposit rate (+.08); loan rate (+0.72) 993Q ( ) discount rate (+.62); deposit rate (+.8); loan rate (+.62) 993Q ( ) - 994Q 26.8 ( ) 994Q ( ) 0 994Q ( ) 0 994Q4 2.2 ( ) 995Q ( ) discount rate (+0.8) 995Q ( ) 995Q ( ) discount rate (+0.8); loan rate (+.08); relending rate (+0.8) 995Q ( ) 0 996Q ( ) 0 996Q ( ) discount rate (-.44); deposit rate (-.8); loan rate (-.08); relending rate (-0.54) - 996Q ( ) deposit rate (-.8); loan rate (-0.9); relending rate (-0.36) - 996Q ( ) discount rate (+0.36); relending rate (+0.8) 997Q ( ) 0 997Q ( ) 0 997Q ( ) 0 997Q ( ) deposit rate (-.26); loan rate (-.44) - Note:.,, denote contractionary, no change and expansionary policy actions respectively. 2 In the second column, the central bank refinancing is measured by the 4-quarter percentage change in the item the PBC s claims on other depository corporations on the PBC's balance sheet. 5

16 Table 3.3 Monetary policy index: 998Q-2000Q4. Time Central bank refinancing % Interest rate change Reverse requirement ratio change Policy change index 998Q 2.83 ( ) discount rate (-0.36); loan rate (-0.72) Q2.97 ( ) relending rate (-0.8) - 998Q ( ) rediscount rate (-.7); relending rate (-4.23); discount rate (-3.78); deposit rate 0 (-0.63); loan rate (-0.99) 998Q ( ) rediscount rate (-0.36);relending rate (-0.63); discount rate (-0.63); deposit rate -2 - (-0.63); loan rate (-0.54) 999Q ( ) 999Q2-2.2 ( ) discount rate (-.35); deposit rate (-.35); loan rate (-0.54); relending rate (-.35) 0 999Q ( ) rediscount rate (-.8) - 999Q ( ) Q.7 ( ) Q ( ) 2000Q ( ) 2000Q ( ) Note:.,, denote contractionary, no change and expansionary policy actions respectively. 2. In the second column, the central bank refinancing is measured by a 4-quarter percentage change in the item the PBC's claims on other depository corporations on the PBC s balance sheet. 6

17 Table 3.4 Monetary policy index: 200Q-2002Q2. Time Central bank refinancing % Interest rate change OMOs (in billion) Policy change index 200Q ( ) 4.83 ( ) 0 200Q ( ).0 ( ) 0 200Q ( ) ( ) 0 200Q ( ) rediscount rate (+0.8) 0.98 ( ) 2002Q ( ) relending rate (-0.54); discount rate (-0.54); deposit rate (-0.27); loan rate ( ) - (-0.54) 2002Q ( ) ( ) Note:.,, denote contractionary, no change and expansionary policy actions respectively. 2. In the second column, the central bank refinancing is measured by a 4-quarter percentage change in the item the PBC s claims on other depository corporations on the PBC's balance sheet. 3. In the fourth column, the open market operation is measured by changes in the item claims on government on the PBC s balance sheet. 7

18 Table 3.5 Monetary policy index: 2002Q3-2008Q4. Time OMOs (in billion) Interest rate change Reverse requirement ratio change Policy change index 2002Q ( ) 2002Q ( ) Q ( ) Q ( ) 2003Q ( ) 2003Q ( ) Q ( ) discount rate (+0.63) Q ( ) relending rate (+0.63) 2004Q ( ) Q ( ) loan rate (+0.27) 2005Q ( ) 2005Q ( ) rediscount rate (+0.27) 2005Q ( ) Q ( ) 2006Q ( ) 2006Q ( ) loan rate (+0.27) Q ( ) deposit rate (+0.09); loan rate (+0.27) Q ( ) Q ( ) deposit rate (+0.8); loan rate (+0.27) 2007Q ( ) deposit rate (+0.09); loan rate (+0.8) Q ( ) deposit rate (+0.8); loan rate (+0.72) 2007Q ( ) deposit rate (+0.45); loan rate (+0.8) Q ( ) 2008Q ( ) Q ( ) loan rate (-0.8); deposit rate (0.54) 2008Q4-3.6 ( ) rediscount rate (-2.52); deposit rate (-.89); loan rate (-2.6); relending rate (-.35) Note:.,, denote contractionary, no change and expansionary policy actions respectively. 2. In the second column, the open market operation is measured by changes in the item bond issue on the PBC's balance sheet. 8

19 3.3 Some Issues Related to the Policy Change Index As more than one policy instrument is considered in the policy change index, it is inevitable that the policy instruments sometimes move in opposite directions to each other. Such cases occurred six times in the overall sample period: 998Q3, 998Q4, 999Q2, 200Q4, 2007Q4and 2008Q3. In 998Q3 and 999Q2, while the PBC reduced central bank refinancing by 5.97 percent and 2.2 percent respectively, it also cut various types of interest rate, which had not happened before. So we treat these two cases as no change in the policy stance. In 998Q4, there was a small decrease in central bank refinancing, but several interest rates continued to fall, while the reserve requirement ratio was cut by two percent in 998Q4, which is seen as a strong signal of loosening monetary policy. So the policy stance in 998Q4 is identified as an expansionary one. Similar cases occurred in 2007Q4 and 2008Q3 but in the opposite direction. In 200Q, the largest cut in central bank refinancing and the largest OMOs through buying treasury bonds were observed at the same time, so we treat these signals as mutually offsetting and classify this quarter as one of no change in the policy stance. Boschen and Mills (995) argue that an important standard in choosing a measure of policy stance is that such a measure should be unaffected by anything other than specific shifts in monetary policy. In many empirical studies of monetary policy using VAR approaches, the component of policy that responds to the movements in the economy and then can be captured by certain policy rules is excluded, leaving only the component of policy shocks to form the measure of the policy stance. But the policy change index we construct for China is more like an overall measure of the change in the policy stance, which contains both the endogenous components of the policy and the policy innovations (or policy shocks). In other word, the PBC s reactions to output, broad money supply or inflation may also change this policy change index. However, it is important to realize that currently in China, the monetary policy reaction function is far from clear to private agents due to a lack of transparency in the policy making process, of a clear policy target to which the authorities are committed, and of a well functioning communication channel between the PBC and the public. So the presumption that private agents know the policy rules and use them to form their 9

20 expectations about future policy does not seem to hold in the case of China. In fact, the changes in monetary policy actions by the PBC tend to be discretionary, so that they are not easy to predict. It may therefore be assumed that the proportion of the monetary policy actions that can be anticipated by the private agents is quite limited, and so the policy stance measured by our policy change index may reflect the policy movements by the authorities in a more useful way. 3.4 A Historical Review based on the Policy Change Index By consolidating the policy change index in the last columns of Tables 3.2-5, we plot the policy change index in Figure 3.2 for the whole sample period from 99Q to 2008Q4. It can be seen that the policy change index captures most of the important changes in the stance of the PBC s monetary policy Figure 3.2 Policy change index from 99Q to 2008Q4. In January 99, a new round of foreign trade reform was announced which ended all fiscal subsidies on export losses. 3 So the expansionary policy change represented by the interest rate cut in the second quarter may be understood as a measure to support this reform by reducing the financial cost of export enterprises. In , an investment fever was stimulated by the central government with the announcement of an annual GDP growth target of 8-9 percent GDP as compared to 3 See Liew and Wu (2007), pp

21 the original 6 percent in the First Plenary Session of the Eighth National People s Congress in March 993. Our policy index captures this change in the monetary policy stance in 993Q, while a remarkable percent increase in the supply of banks credit is observed. As many new projects were financed, the total investment in the first half of 993 was double the amount in the same period of the previous year. 4 On the other hand, the CPI inflation rose sharply to 3.9 percent in the second quarter, and the central government and the PBC decided quickly to fight the inflation and financial disorders. Interest rates were increased, and administrative controls including credit controls, business restrictions by local governments and foreign exchange rate controls, were imposed by the State Planning Commission in what was known as the famous The Sixteen Regulations in In the policy change index series, the measure of the policy stance is marked by in the second and third quarter of 993 due to the increase in various interest rates, such as the discount rate, deposit rate and loan rate. In January 994, on the unification of the RMB official rate and market rate, there was a devaluation of the RMB. A substantial increase in domestic money supply was observed and an inflation expectation occurred. Liew and Wu (2007, pp 96) find that the ratio of foreign exchange reserves to the reserve money 6 in 994 was more than doubled compared to the 993 ratio after the unification. The rapid increase in money supply may then explain an observed tightening in the policy stance in the first quarter of 994 in terms of a significant reduction on the growth of credit supplied by the banking system. After the Asian Financial Crisis in 997, China faced a continuously weak domestic demand as well as currency appreciation. In order to stimulate domestic demand, monetary policy played an active role, as the policy change index was expansionary in seven out of ten quarters from the last quarter of 997 to the first quarter of As in 994, it is believed that the contractionary policy stance of 2002 was due to a strong financial inflow caused by an increasing interest rate spread between China and the US, and the expectation that the RMB would be revalued. 7 In order to curb the surge in liquidity the PBC began to issue central bank bills for sterilization 4 See Guo (2002), pp See Guo (2002), pp On the balance sheet of the PBC, reserve money includes all the currency issued, and the deposits of all the financial corporations. 7 See Eichengreen (2004). 2

22 purposes from the third quarter of It is also interesting to note that in the first and last quarter in 2003 the amount of central bank bills issued declined sharply by and billion Yuan respectively. Therefore the policy change index is coded as - which indicates an expansionary policy stance. This decline in the amount of central bank bills was likely due to a maturity of the central bank bills while the PBC decided not to issue new central bank bills. Such expansionary policy may be associated with the outstanding central bank bills which increased quite rapidly. In 2007 a signal of an overheating economy emerged with CPI inflation rising sharply to 3.2 percent in comparison to.5 in 2006, and GDP growth rose to 2.2 percent in the first half of In 2007Q3 the PBC declared a switch of the monetary policy stance from a neutral one to moderately tight, and later to a tight one. Various monetary policy instruments were used, including ten rises in the reserve requirement ratio amounting to a total of 5.5 percent, and rises in the one year deposit and loan rates by.62 and.35 percent respectively. According to the policy change index, such tight policy could be observed from 2005Q4 when the PBC increased its issue of central bank bills significantly. In 2008Q3, due to the negative impacts of the international financial crisis and domestic natural disasters, the PBC announced a moderately loose monetary policy to stimulate economic growth. This policy decision explains the drop in the policy change index at the end of 2008 in Figure 3.2. To sum up, the policy index developed in this section appears to capture well the important monetary policy changes through the sample period. An empirical analysis of our policy change index is given in the next section in the context of an ordered probit model. 22

23 4. The Policy Change Index in an Ordered Probit Model The qualitative policy change index constructed in Section 3 cannot be applied through conventional estimation methods, because the index is discrete, in a certain order, and has limited values. Moreover, the spacing of the policy index cannot be assumed to be uniform, for example, the difference between the index of - and 0, can not be treated as the same as that between 0 and. However, these problems can be solved by using an ordered probit approach. A model which contains a variable referring to some qualitative outcome is often known as a qualitative response (QR) model, or qualitative choice model. One characteristic of this model is that the dependent variable is a discrete number of mutually exclusive values, which is quite different from the conventional OLS regressions. It is widely used in the econometric analysis of social science to model, for example, the decision to vote in an election, labour force participation, and consumers satisfaction levels of shopping experiences, and so on. If the qualitative outcome is ordered, and the model is estimated by a probit method, it is known as an ordered probit model Model and Data Our analysis closely follows Gerlach (2006) who estimates an empirical reaction function for the European Central Bank from 999 to Consider that the PBC s preferred monetary policy is determined by the function: where r x' () * t i, t it * r t is a latent variable representing a preferred monetary policy change by the PBC; 9 xit, is the observed macroeconomic variables with a one period lag; is the parameter vector; it is the residual term. Equation () indicates that the PBC s preferred policy stance can be specified as a linear function of certain macroeconomic variables. The role of this preferred policy stance is similar to the desired target policy rate in Taylor-typed monetary reaction functions, where the policy rate is believed to 8 The ordered model can also be estimated by the logic methods, in which case it is known as the ordered logit model. See Borooah (200), pp2. 9 Conventionally, r indicates some kind of interest rate in many empirical studies of monetary policy, for example, repo rate. To be consistent with other studies, we still use r to represent the authorities' monetary policy in China. 23

24 react to the changes in the economic environment xit,. The reason that the lagged term xit, is considered in our model is due to certain lags in the monetary policy making process which may make the effects of monetary policy uncertain. Bofinger (200, pp 74) shows different types of lags in the reaction of the central bank. Firstly, the data on the current cyclical situation is often available with a one to three month delay, which is known as the information lag. Secondly, once the data is available, the central bank still needs a longer observation period to necessarily identify the economic condition, for example, whether a decline in GDP growth of one quarter means a recession is coming. Third, there is a decision lag in the policymaking structure in which policies are produced. Liew and Wu (2007, pp 44-46) provide a detailed example on the process of the exchange rate policy is formed in China, which can be shown in a three-stage diagram: Policy inputs Input processing and evaluation of policy alternatives Final assessment of policy alternatives and decision making In the first stage, information, analyses and even policy recommendations are gathered from relevant ministries, research institutions and informal consultants. In the second stage, all the policy inputs are evaluated by some key policy organs, 0 and there is bargaining among those policy organs before the final decision is formed. Finally, the authorities assess all the policy alternatives from stage two and then make their final policy decision which is going to be implemented. Hence in our model, all the regressors are incorporated in the model with a one quarter lag, as we assume that the preferred monetary policy change decision is made based on the information of the previous quarter. We use quarterly data because many Chinese economic data are unavailable on a monthly frequency. However, the preferred policy change of the PBC is usually difficult to observe. In addition, the actual monetary policy remains unchanged in some periods even 0 The policy organs, as Liew and Wu (2007, pp 43) write in their book, are those sectors involved in policy making, such as the State Development and Planning Commission (SDPC), the Ministry of Finance (MOF), the Ministry of Foreign Trade and Economic Cooperation (MOFTEC) and the People's Bank of China (PBC). 24

25 though there is a change in macroeconomic conditions. To explore the relationship between the discrete monetary policy change and the continuous economic situation via an unobservable, preferred policy change, we define that r it * if rit 0 if r * if 2 rit * it 2 (2) It is worth noting that the numbers of -, 0, representing the policy stance are totally arbitrary. and 2 are unknown cutoff points which define the ranges of the latent variable, or the preferred policy stance, * r t. In other words, given the ordered choice of the policy change index, the authorities are forced to choose the category of the index that most closely represents their own intentions for monetary policy. At the same time, we believe that the set of macroeconomic variables in vector x can explain the policy decision made by the PBC. So, under the assumption of normality, we have: Pr( r x,, ) ( x' ) t i, t i, t Pr( r 0 x,, ) ( x' ) ( x' ) t i, t 2 i, t i, t Pr( r x,, ) ( x' ) t i, t 2 i, t (3) where Φ is a normal distribution cumulative function. To estimate, and 2, a maximum likelihood function is considered: N t i, t t (4) t j l(, ) log(pr( r j x,, )) ( r j) where ( r ) t j is an indicator function which takes the value if rt jis true, and 0 if rt j.by maximizing the log-likelihood function the estimated, and 2 are then obtained. It is important, however, to realize that in an ordered probit model, the sign and magnitude of the single coefficient cannot be used directly in interpreting the partial effects of the probability, which are determined by: 25

26 Pr( rt ) ( x' i ) ( i ) xi xi Pr( r 0) ( x ' ) ( x' ) t 2 i i ( )( i ) xi xi xi Pr( rt ) ( 2 x' i ) ( i ) xi xi (5) In (5) the partial effect of a change in any variable depends not only on the parameter of this variable, but also on all the other parameters in. Moreover, the direction of change in the probability of an expansionary policy change ( Pr( rt ) ) is opposite to the sign of the corresponding parameter ; and Pr( r ) changes in the same direction as the sign of i. However, the effect on the change in the probability of an unchanged policy is ambiguous, since it may be either positive or negative, depending on i. i Hence, the interpretation of the parameters in an ordered probit model is quite different from that in a classic OLS model, and must be undertaken carefully. In our model, the four-quarter percentage change in the industrial production index, CPI inflation and seasonally adjusted M2 supply are adopted as the regressors to represent real output, inflation and aggregate money supply respectively, and all the data are at a quarterly frequency. In addition, due to the important role of the foreign exchange rate in the framework of the PBC s monetary policy, the nominal effective exchange rate (NEER) is incorporated to reflect the PBC's foreign exchange policy. 2 In addition, we include the lagged accumulated policy change: this is the sum of the policy changes from the start of the index up to the previous period, and indicates whether the level of policy is relatively loose or tight. The series for M2 (after seasonal adjusted) and NEER are measured in logs, which approximately represent the percentage changes of M2 and NEER. All the data were taken from DataStream and IFS. The econometric software used is the ordered probit package in Eviews 6. t A more comprehensive graph example about this ambiguous direction of change in the probability is provided by Greene (2008), pp The nominal effective exchange rate (NEER) is a weighted average of nominal exchange rates of home with respect to a set of foreign currencies with the weight for each foreign country equal to its share in trade. A higher NEER means that the purchasing power of the currency is higher; if NEER decreases, the domestic currency depreciates against its trading partners. 26

27 4.2 Some Empirical Results. Firstly, the basic parameter estimates and their statistical significance in the ordered probit models are reported in column number () of Table 4. Table 4. Ordered Probit Model Estimates: 99Q-2008Q4. Model Industrial Production Index 0.67*** 0.20*** 0.205*** 0.60*** (0.062) (0.059) (0.05) (0.05) CPI inflation * (0.046) (0.029) (0.035) M2 supply.5*** 0.792***.93*** (0.423) (0.272) (0.36) NEER ** Accumulated policy change (.709) (.42) (.249) -0.24** ** -0.26** (0.054) (0.034) (0.048) (0.053) * *** (0.254) (6.359) (6.323) (3.442) *** (0.266) (6.343) (6.323) (3.487) Pseudo-R squared Pro (LR statistic) Note:. In all the models, the dependent variable is the policy change index. 2. The estimation is given by Eviews The values in brackets are standard errors; ***, ** and * indicate significance at %, 5%, and 0% respectively. 27

28 Intuitively, the signs of the estimated parameters describe the direction of change in the probability of the policy change index increasing when the corresponding independent variable changes. In column (), we can see that the signs of all the parameters are what we expect. The parameter on output is positive and significant. Thus it can be inferred that faster economic growth increases the probability of a contractionary monetary policy intervention by the PBC. The parameter on CPI is positive but insignificant, which means it is unclear how monetary policy reacts to inflation. The parameter on M2 growth is positive and highly significant, which suggests that faster money growth leads the PBC to tighten monetary policy. The parameter on the nominal effective exchange rate is negative, so depreciation may be associated with a tightening of monetary policy. But this effect is quite insignificant. The parameter on the accumulated policy change is negative and significant at the 5 percent level, which indicates that the probability of a tightening of policy decreases if the policy stance in the previous quarter is already tight. The Pseudo R-squared, which is used as a goodness-of-fit measure like the R-squared in an OLS regression, has the value of 0.66, and all the parameters are jointly significant at the percent level. From the empirical results it can be seen, firstly, that the highly significant role of output is consistent with the general understanding about the PBC's monetary policy making. According to the Central Bank Law in 995, to promote an economic growth acts as one of the two main ultimate policy goals. While the second goal, to stabilize the value of the RMB currency, appears to be designed mainly for assisting the first goal according to the statements in the Central bank Law. Secondly, it is interesting that CPI inflation is insignificant in regression 4.. This result may suggest that the PBC does not react significantly to the level of inflation, but it would be inappropriate to conclude this before more detailed information on how the PBC forms its concerns over inflation condition has been examined. One explanation of the insignificant role of CPI may be that rather than the current level of price indices, the PBC is more concerned by the potential inflation pressures in the future, and such inflation expectations are likely to be driven by current output growth. Figure 4. plots IPI growth and CPI inflation from 99Q to 2008Q4. From the graph it can be seen that IPI growth and CPI inflation move in a similar way. More importantly, IPI growth exhibits a clear lead in comparison to CPI inflation, with an approximate three to five quarters lead over CPI inflation. Consequently, it is 28