China s Inflation and Exchange Rate

Transcription

1 Doctorate Dissertation China s Inflation and Exchange Rate International Graduate School of Social Sciences Yokohama National University Fang Guo September, 2014

2 Doctorate Dissertation China s Inflation and Exchange Rate International Graduate School of Social Sciences Yokohama National University Fang Guo September,

3 Acknowledgement First, please allow me taking this opportunity to express my sincere gratitude to Professor Hasebe, Professor Sato and the Advisory Committee for their great guidance and support during my doctorate study. Their guidance really provides me a lot of chances to improve myself, and I developed my academic skills, broadened my sights and am confident about the future. Second, my sincere thanks also go to my teachers, administrative staff, friends and all the people I met during my doctorate study. I feel very lucky because they are always so nice to help me and assist me when needed. Finally, my special thanks to my mother, without her always warm support, I do not think I can make this far. Her love makes me a better person. My sincere thanks also go to my grandmother. She is such a wise lady who taught me to be independent and strong. I can always feel their support even though I am far from home, and their love will make me advance further. 2

4 CONTENTS I. Introduction 1-8 II. Chapter 1. What Causes China s High Inflation? A Threshold Structural Vector Autoregression Analysis 9-28 III. Chapter 2. The Role of Inflation Regime in the Pricing-to-Market to China s Electronics Exports: Exponential Smooth-Transition Autoregressive Analysis IV. Chapter 3. The Asymmetric Analysis on Exchange Rate Pass-Through to China s Disaggregated Electronics Imports V. Conclusion VI. Reference

5 Introduction China has become the world second largest economy. The economic development in China gains a lot of attention from the world. Thus its inflation and exchange rate issues become the hottest topics in the academia field. This dissertation concentrates on the inflation and exchange rate issues of China. China s inflation oscillated greatly in the past decade, coinciding with several rounds of inflationary and deflationary periods. When the economy changes from an expansionary regime to a contractionary regime, the dynamic adjustments of inflation and exchange rate are likely to change. As Shen et al. (1999) demonstrate, the central bank will respond differently to changes in economic variables in high-inflationary and low-inflationary environment. Bussiere (2012) argues that exchange rate pass-through to trade prices sometimes shows nonlinearity and asymmetries. Many studies concentrate on the causes to China s inflation, however, this dissertation pays attention to asymmetric and nonlinear causes to China s inflation, aiming at the different causes to China s inflation in the high inflation regime and low inflation regime. Since the empirical results show the importance of inflation asymmetry and nonlinearity, the dissertation further investigates whether inflation asymmetry and nonlinearity have influence on China s electronics import and export. And empirical results show that the inflation asymmetry and nonlinearity also play very important roles in the electronics import and export, which demonstrates the asymmetric responses of electronic trade to the high and low inflation regime. Compared with the previous literature, first, the dissertation analyzes the causes to China s inflation by dividing the whole inflation period into low and high inflation regimes. By applying the threshold structural vector autoregression analysis, the results demonstrate that the causes to China s inflation are asymmetric and nonlinear in the high and low inflation regimes. Second, because of the importance of inflation asymmetry and nonlinearity, the dissertation utilizes the exponential smooth-transition autoregressive model to study the relationship between pricing-to-market and the inflation regime in highly disaggregated electronics exports in China. It is found that higher pricing-to-market behavior occurs in the high inflation regime for some subcategories of electronics exports, whereas there are no pricing-to-market behavior changes in the low and high inflation regimes for the other subcategories of electronics exports. Third, the dissertation employs the threshold nonrecursive structural vector autoregression analysis to explore the asymmetric effects of exchange rate pass-through to China s disaggregated electronics imports between the low and high 4

6 inflation regimes. The empirical evidence demonstrates the reactions of disaggregated electronics imports to nominal effective exchange rate shocks are inflation-regime-dependent and asymmetric. In chapter 1, the published literature shows that monetary policy is widely affected by asymmetry. Morgan (1993) finds that asymmetry exists in a number of widely accepted economic models. Cover (1992) argues that positive and negative money-supply shocks have asymmetric effects on output. Such asymmetric effects are also asserted by DeLong and Summers (1988), who compare the effectiveness of different demand management policies in the USA and other industrial nations before and after World War II. They find evidence of asymmetry in business cycles. There was considerable improvement in the US economy after World War II thanks to the more stable financial system and improved macroeconomic policies. Beaudry and Koop (1993) find that allowing for the impulse response of gross national product to be asymmetric, the effect of negative innovations on gross national product are observed to be much less persistent than the effects of positive innovations. Thoma (1994) shows that money income causalities across sample periods are highly correlated with the level of real activity. When real activity declines, money income causality becomes stronger, and when real activity increases, the relationship becomes weaker. Kandil (1995) shows that asymmetric effects are expected to differentiate the cyclical impact of positive and negative demand shocks on output growth. In regards to asymmetric effects of macro-variables on inflation, the previous published literature further reveals that the economic growth and inflation nexus might not necessarily be linear and can be unstable. Friedman (1968) suggests that the changes in inflation rate might induce regime switching and asymmetry. Threshold models provide an intuitive way to capture the nonlinearity. Tong and Lim (1980) introduce a univariate threshold autoregressive model to demonstrate that this model exhibits nonlinearity. Tong (1983) also employs the threshold model, with inflation chosen as the threshold to separate the whole sample into low and high inflation regimes. In this framework, the money supply is assumed to be an exogenous variable. It is also assumed that the public and the central bank can influence the money supply. The money supply movements are not only influenced by the monetary authorities, but also by feedback from other economic forces. However, a limitation of this threshold model is the utilization of the univariate framework. Tsay (1998) extends the univariate threshold principles to the multivariate threshold frame. The studies by Leeper and Gordon (1992), Gordon and Leeper (1994), Christiano and Eichenbaum (1992), Christiano et al. (1994), Pagan and Robertson (1994), Sims and Zha (2003) and Leeper (1995) find that in a univariate framework, monetary policy shocks cannot be successfully identified. Therefore, the interpretation for the response of interest rate to monetary policy shock might be mistaken under a univariate framework. As a result, a multivariate framework, such as a VAR model, is best suited to studying such issues. The threshold autoregression approach has been applied to many studies. Shen and Chiang (1999) use a threshold VAR model in which data is 5

7 subdivided into low and high inflation regimes. They hypothesize that the interest rate responds inversely to increased monetary growth in a low inflation regime and positively to increased monetary growth in a high inflation regime. Charles (1999) incorporates nonlinearity in a VAR analysis including output, prices and money supply. He finds that monetary policy has an asymmetric effect on the macroeconomic variables. Balke (2000) employs threshold VAR analysis, with credit condition as a threshold variable for the USA, and reports that output responds more to monetary policy in a credit-rationed regime. Mandler (2010) uses a threshold VAR model to study regime dependence and the effects of systematic and nonsystematic monetary policy in the USA. The empirical evidence shows that a change in inflation dynamics is the most important source of a transition of the US economy from a high inflation state to a low inflation state, while a change in monetary policy reaction functions has very little effect. Bunzel and Enders (2010) also use the inflation threshold to estimate a nonlinear Taylor rule. Hwang and Wu (2011) find that there is an asymmetric and nonlinear relationship between inflation and economic growth in China; Liu and Pang (2011) point out that asymmetric effects exist between monetary policy and inflation. The reaction of inflation to various macro-variables varies and may show nonlinearity in both low and high inflation regimes. Most studies focus on the asymmetric effects of monetary policy in developed countries, rather than developing countries, such as China. Although standard recursive threshold VAR analysis has been applied in some studies, those studies do not extend this methodology to threshold nonrecursive structural VAR. A significant advantage of threshold nonrecursive structural VAR analysis is that it allows for both regime change and nonlinearities when analyzing the causes of China s high inflation. The inflation threshold allows for modeling nonlinearities empirically. The inflation threshold model separates a low inflation regime and a high inflation regime using different sets of model parameters. The inflation threshold separates sample data from 2001 to 2011 into two regimes: low and high inflation regimes. When inflation exceeds a given threshold value, the time point belongs to a high inflation regime. The time point belongs to a low inflation regime when inflation is lower than the given threshold value. Introducing an inflation threshold value allows division of the time sample into low and high inflation regimes. The inflation threshold enables the study of asymmetric effects in different inflation regimes. However, there is another question: How do we analyze the causes of China s inflation in low and high inflation regimes? In order to solve this problem, the present paper employs a nonrecursive structural VAR model. This paper extends the recursive structural VAR approach to a nonrecursive structural VAR approach (see Kim and Roubini, 2000; Kim, 2001), because a nonrecursive VAR can identify the coefficients of the contemporaneous relationships. Thus, by employing nonrecursive structural VAR analysis, the impulse responses of inflation to macro-variables can be captured in low and high inflation regimes. Therefore, threshold nonrecursive structural VAR analysis is a very important new method for studying the causes of 6

8 China s high inflation. Chapter 1 uses threshold nonrecursive structural VAR analysis to investigate whether macroeconomic variables have asymmetric effects on China s inflation in low and high inflation regimes. In particular, this study pays attention to the asymmetric monetary effects in low and high inflation regimes, and attempts to capture the different impacts on China s inflation in low and high inflation regimes. The empirical evidence extends the literature in four ways. First, the reactions of inflation to various shocks are inflation-regime-dependent and asymmetric. Second, adjusting the domestic interest rate in China may be an effective way to control inflation in a high inflation regime, but not in a low inflation regime. The results are consistent with findings of the previous study (See Van Der, 2004) that when inflation breaches its target level, adjustment in interest rates is required to move in the same direction as the expected inflation. Third, oil prices might not be as important as generally considered in pushing China s high inflation. Oil price increases may increase inflation in a high inflation regime, whereas there is no such obvious effect in a low inflation regime. The most intuitive explanation is that in a high inflation regime oil price shocks are indicative of the increased scarcity of energy. In a high inflation regime, the speed of adjustment of inflationary expectations is rapid. Thus, the oil price increase induces firms to invest in more energy-efficient capital because of the inflationary expectation in the future. Finally, a nominal effective exchange rate appreciation might be effective in controlling domestic inflation in both regimes. This is because a stronger RMB may very well curtail Chinese domestic inflation. Chinese consumers will find cheaper foreign goods in their markets and, in return, this will put downward pressure on similar domestic goods. Chinese importers of foreign raw materials and intermediate components will enjoy lower relative prices. Furthermore, prices on imported petroleum and petroleum products will also be reduced. In chapter 2, a large number of theoretical studies focus on the relationship between exchange rate pass-through and inflation regime. It is also generally pointed out that exchange rate pass-through effect has been declining in recent years (see e.g. Goldberg and Knetter, 1997, Campa and Goldberg, 2005). Taylor (2000) argues that the decline in exchange rate pass-through implies an increasing difficulty for firms to pass through the exchange rate changes to importers in low and stable inflation environment. With respect to the relationship of exchange rate pass-through and inflation regime, empirical studies are conducted by utilizing cross-county evidence, including the analysis by Calvo and Reinhart (2002), Choudri and Hakura (2006), Parsons and Sato (2008) and Devereux and Yetman (2010). Nonlinear time series models like the exponential smooth-transition autoregressive model 1 have been employed in many studies, including Michael et al. (1997), Taylor and Peel (2000), Taylor et al. (2001), Kilian and Taylor (2003), Junttila and Kprhonen (2011), Shintani et al. (2013). Chapter 2 is the first to employ U-shaped inflation transition function in exponential 1 See Granger & Teräsvirta, 1993; Teräsvirta, 1994, 1998 and Tong,

9 smooth-transition autoregressive model to study the relationship between pricing-to-market and the inflation regime in China s highly disaggregated electronics export categories. The empirical evidence shows three things. First, for durable consumer goods (COD) and intermediate goods for parts and accessories of capital goods (INA), complete pricing-to-market is chosen in a high inflation regime, whereas incomplete pricing-to-market is chosen in a low inflation regime. This is because materials and wage will increase the production costs in a high inflation regime. China s electronics exports firms have a lack of complete industry chain and are weak at capabilities in advanced technology. Thus, a large number of Chinese electronics exports companies are crowded in mid-to-low end technological electronics exports and must compete on price. When the exchange rate appreciates, the exporters have to adjust their export prices in term of RMB to maintain the foreign market share. High pricing-to-market behavior also happens in a high inflation regime for processed intermediate goods (INP) and nondurable consumer goods (CON) for the same reason. Second, the inflation regimes have no impact on pricing-to-market behavior for capital goods (CAG), semi-durable consumer goods (COS) and intermediate goods for parts and accessories of transport equipment (INT). And third, China s domestic PPI and the foreign IPI have limited explanatory power to explain the pricing-to-market behavior in most highly disaggregated electronics categories. In chapter 3, many theoretical studies focus on the relationship between exchange rate pass-through and inflation regimes. It is generally pointed out that exchange rate pass-through effect has been declining in recent years (see e.g. Goldberg and Knetter, 1997, Campa and Goldberg, 2005). Taylor (2000) argues that there is a wide spread and on-going decline in exchange rate pass-through. This is because under the environment of low and stable inflation, it becomes more difficult to pass through the exchange rate changes to the importers. With respect to the relationship between exchange rate pass-through and inflation regime, empirical studies are conducted by utilizing the cross-county evidence, including the analysis by Calvo and Reinhart (2002), Choudri and Hakura (2006), Parsons and Sato (2008) and Devereux and Yetman (2010). The previous literature further shows that the threshold vector autoregression analysis has its advantages to analyze asymmetric effects between the low and high inflation regimes. As suggested by Friedman (1968), the inflation rate might induce regime switching and asymmetry. Therefore threshold models provide an intuitive way to capture the nonlinearity. Tong and Lim (1980) introduce a univariate threshold autoregressive model to demonstrate this model exhibits asymmetries. Tong (1983) also employs the threshold model. Inflation is chosen as the threshold to separate the whole sample into low and high inflation regimes. In this framework, the money supply is assumed to be exogenous variable. It is also assumed that the public and central bank can influence the money supply. The paper demonstrates that money supply movements are not only influenced by the monetary authorities, but also by the feedbacks from other economic forces. But the limitation of this threshold model is the utilization of the univariate framework. Tsay (1998) 8

10 develops the univariate threshold principles to the multivariate threshold frame. The studies by Leeper and Gordon (1992), Gordon and Leeper (1994), Christiano and Eichenbaum (1992), Christiano et al. (1994), Pagan and Robertson (1994), Sims and Zha (2003) and Leeper (1995) find that in a univariate framework, the monetary policy shocks cannot be successfully identified. Therefore the interpretation for the response of interest rate to monetary policy shock might be mistaken. As a result, a multivariate framework, such as a vector autoregression model, has been recommended to study such issues. Consequently, the threshold vector autoregression approach has been applied to many studies. Shen and Chiang (1999) use a threshold vector autoregression model in which data is subdivided into low and high inflation regimes. They hypothesize that interest rate responds inversely to increased monetary growth in the low inflation regime and positively to increased monetary growth in the high inflation regime. Charles (1999) incorporates the nonlinearity in a vector autoregression analysis including output, prices and money supply. He finds that monetary policy has asymmetric effects on the macroeconomic variables. Balke (2000) employs threshold vector autoregression analysis, with credit condition as a threshold variable for the U.S., and reports that output responds more to monetary policy in a credit-rationed regime. Mandler (2010) studies regime dependence in the effects of systematic and nonsystematic monetary policy in the U.S. with a threshold vector autoregression model. The empirical evidence shows a change in inflation dynamics is the most important source of the transition of the U.S. economy from the high inflation state into the low inflation state, while the change in the monetary policy reaction functions has only very little effect. Bunzel and Enders (2010) also use inflation threshold to estimate a nonlinear Taylor rule. According to the previous literature and the well-developed threshold models, exchange rate pass-through to China s disaggregated electronics imports may show asymmetries in the low and high inflation regimes. That is, the reactions of China s disaggregated electronics imports to nominal effective exchange rate may vary in different inflation regimes. However, based on the previous literature, first, most studies focus on exchange rate pass-through to imports in developed countries, rather than developing countries, yet not to China s disaggregated electronics imports. Second, although the standard recursive threshold vector autoregression analysis has been employed in some studies, these studies do not extend this methodology into threshold nonrecursive structural vector autoregression. Third, the paper further extends the asymmetric exchange rate pass-through to disaggregated electronics imports, that is, imported electronics capital goods, imported electronics intermediate goods and imported electronics consumption goods. The significant advantages of the threshold nonrecursive structural vector autoregression analysis are that it allows for both regime change and asymmetries. The inflation threshold allows for analyzing exchange rate pass-through asymmetries to electronics imports empirically. The inflation threshold model separates the low inflation regime and the high inflation regime with 9

11 different sets of model parameters. In this paper the inflation threshold groups the sample data during 2003 and 2012 into two regimes, that is, the low and high inflation regimes. When inflation exceeds a given threshold value, this time point belongs to a high inflation regime, whereas the time point belongs to a low inflation regime when inflation is lower than the given threshold value. As a result, by finding out the inflation threshold value, the whole time sample can be divided into the low and high inflation regimes. All these operations above allow the possibilities to study the asymmetries of exchange rate pass-through to disaggregated electronics imports in different inflation regimes. However, there is another question: how to analyze the exchange rate pass-through to China s electronics imports in each inflation regime? In order to solve the raised problem, the paper extends the recursive structural vector autoregression approach to the nonrecursive structural vector autoregression approach (see Kim and Roubini (2000) and Kim(2001)), because the nonrecursive vector autoregression can identify the coefficients of the contemporaneous relationships. Thus the impulse responses of China s disaggregated electronics imports to nominal effective exchange rate can be captured in the low and high inflation regimes. Therefore, the threshold nonrecursive structural vector autoregression analysis is a very important new method to study the asymmetric exchange rate pass-through to China s disaggregated electronics imports. Chapter 3 employs the threshold nonrecursive structural vector autoregression analysis to investigate whether there are asymmetric exchange rate pass-through effects on China s disaggregated electronics imports in the low and high inflation regimes. In particular, this study highlights the asymmetric exchange rate pass-through effects on China s disaggregated imported electronics capital goods, intermediate goods and consumption goods. The empirical evidence extends the literature in three ways: first, the reactions of disaggregated electronics imports to nominal effective exchange rate shocks are inflation-regime-dependent and asymmetric. Second, for imported electronics capital goods and intermediate goods, the impulse responses of these two categories to nominal effective exchange rate show asymmetries between the low inflation regime and the high inflation regime. In the low inflation regime, exchange rate pass-through to imported electronics capital goods and intermediate goods are smaller than that of a high inflation regime. This is because a decrease in production costs occurs domestically in the low inflation regime so that foreign electronics exporters have to decrease their selling prices in terms of RMB to maintain the market share in China. Thus, the prices of imported electronics capital goods and intermediate goods decline obviously in a low inflation regime. However, the imported electronics capital goods only decrease slightly in the high inflation regime. The reason may be that production costs may increase due to a high inflation regime in China. The production costs for the domestic electronics firms increase greatly, thus there are strong demands for the imported electronics capital goods and intermediate goods. And also China s electronics imports firms are lack of price negotiation power. As a result, a large number of Chinese electronics imports firms have to accept the prices determined 10

12 by the foreign exporters. Thus in the low inflation regime, exchange rate pass-through to imported electronics capital goods and intermediate goods are smaller than that of the high inflation regime. Third, the impulse responses of imported electronics consumption goods to nominal effective exchange rate show no asymmetry between the low and high inflation regimes. In the low and high inflation regimes, exchange rate pass-through to imported electronics consumption goods is complete. Compared with imported electronics capital goods and intermediate goods, the differentiation of imported electronics consumption goods widely exists. The imported electronics consumption goods cannot be easily substituted by the domestic electronics goods. Therefore, the imported electronics consumption goods are much less sensitive to nominal effective exchange rate changes. Take Apple electronics products as an example, even though RMB appreciates, the prices of these products will stay stable in terms of RMB because the domestic consumers still prefer to choose these imported electronics products rather than the domestic produced electronics goods. Thus complete exchange rate pass-through happens to imported electronics consumption goods in the low and high inflation regimes. The paper is organized as follows. Chapter 1 summarizes the causes of China s high inflation, utilizing the threshold non-recursive structural VAR modeling. Chapter 2 describes the role of inflation regime in the pricing-to-market to China s electronics exports by exponential smooth-transition autoregressive analysis. Chapter 3 estimates the asymmetric analysis on exchange rate pass-through to China s disaggregated electronics imports. Section 4 summarizes the results and concludes. 11

13 Chapter 1 What Causes China s High Inflation? A Threshold Structural Vector Autoregression Analysis China s recorded astonishing economic growth implies a necessity to understand its inflation. The present paper employs threshold nonrecursive structural vector autoregression analysis to explore the asymmetric effects of macro-variables on inflation in low and high inflation regimes. The empirical evidence demonstrates, first, that the reactions of inflation to various shocks are inflation-regime-dependent and asymmetric. Second, monetary policy influences China s high inflation and adjusting the domestic interest rate in China may be an effective way to control inflation in a high inflation regime, but not in a low inflation regime. In a high inflation regime, a high inflation rate may cause the macro-authorities to increase the domestic interest rate, in an attempt to stabilize high inflation. Third, contrary to expectations, the world oil price is not a strong cost-push factor in a low inflation regime. Oil price increases may increase inflation in a high inflation regime, but there is no such obvious effect in a low inflation regime. Finally, China s nominal effective exchange rate influences inflation in both low and high inflation regimes. A nominal effective exchange rate appreciation might be effective in controlling domestic inflation in both regimes. 1.1 Introduction China s inflation oscillated greatly between 2001 and 2012, coinciding with several rounds of inflationary and deflationary periods. China experienced a new round of high inflation due to fiscal stimulus and expansionary monetary policy following the global financial crisis in When the economy changes from an expansionary regime to a contractionary regime, the dynamic adjustment of inflation to monetary policy is likely to change. As Shen et al. (1999) demonstrate, the central bank will respond differently to changes in economic variables in high-inflationary and low-inflationary environment. Many regime-switching models have been developed to analyze these types of regime changes, but the present paper might be the first to employ the threshold nonrecursive structural vector autoregression (VAR) analysis, aiming at exploring the asymmetric effects of macro-variables on inflation between low and high inflation regimes in China. The present paper explores the following questions. First, will China s inflation respond asymmetrically to macroeconomic shocks in low and high inflation regimes? Second, does monetary policy influence China s high inflation? If so, to what extent does monetary policy affect the inflation in low and high inflation regimes? Third, is the world oil price a strong cost-push factor for China s high inflation in low and high inflation regimes? Finally, does China s nominal effective exchange rate influence inflation in low and high inflation regimes? 12

14 The published literature shows that monetary policy is widely affected by asymmetry. Morgan (1993) finds that asymmetry exists in a number of widely accepted economic models. Cover (1992) argues that positive and negative money-supply shocks have asymmetric effects on output. Such asymmetric effects are also asserted by DeLong and Summers (1988), who compare the effectiveness of different demand management policies in the USA and other industrial nations before and after World War II. They find evidence of asymmetry in business cycles. There was considerable improvement in the US economy after World War II thanks to the more stable financial system and improved macroeconomic policies. Beaudry and Koop (1993) find that allowing for the impulse response of gross national product to be asymmetric, the effect of negative innovations on gross national product are observed to be much less persistent than the effects of positive innovations. Thoma (1994) shows that money income causalities across sample periods are highly correlated with the level of real activity. When real activity declines, money income causality becomes stronger, and when real activity increases, the relationship becomes weaker. Kandil (1995) shows that asymmetric effects are expected to differentiate the cyclical impact of positive and negative demand shocks on output growth. In regards to asymmetric effects of macro-variables on inflation, the previous published literature further reveals that the economic growth and inflation nexus might not necessarily be linear and can be unstable. Friedman (1968) suggests that the changes in inflation rate might induce regime switching and asymmetry. Threshold models provide an intuitive way to capture the nonlinearity. Tong and Lim (1980) introduce a univariate threshold autoregressive model to demonstrate that this model exhibits nonlinearity. Tong (1983) also employs the threshold model, with inflation chosen as the threshold to separate the whole sample into low and high inflation regimes. In this framework, the money supply is assumed to be an exogenous variable. It is also assumed that the public and the central bank can influence the money supply. The money supply movements are not only influenced by the monetary authorities, but also by feedback from other economic forces. However, a limitation of this threshold model is the utilization of the univariate framework. Tsay (1998) extends the univariate threshold principles to the multivariate threshold frame. The studies by Leeper and Gordon (1992), Gordon and Leeper (1994), Christiano and Eichenbaum (1992), Christiano et al. (1994), Pagan and Robertson (1994), Sims and Zha (2003) and Leeper (1995) find that in a univariate framework, monetary policy shocks cannot be successfully identified. Therefore, the interpretation for the response of interest rate to monetary policy shock might be mistaken under a univariate framework. As a result, a multivariate framework, such as a VAR model, is best suited to studying such issues. The threshold autoregression approach has been applied to many studies. Shen and Chiang (1999) use a threshold VAR model in which data is subdivided into low and high inflation regimes. They hypothesize that the interest rate responds inversely to increased monetary growth in a low inflation regime and positively to increased 13

15 monetary growth in a high inflation regime. Charles (1999) incorporates nonlinearity in a VAR analysis including output, prices and money supply. He finds that monetary policy has an asymmetric effect on the macroeconomic variables. Balke (2000) employs threshold VAR analysis, with credit condition as a threshold variable for the USA, and reports that output responds more to monetary policy in a credit-rationed regime. Mandler (2010) uses a threshold VAR model to study regime dependence and the effects of systematic and nonsystematic monetary policy in the USA. The empirical evidence shows that a change in inflation dynamics is the most important source of a transition of the US economy from a high inflation state to a low inflation state, while a change in monetary policy reaction functions has very little effect. Bunzel and Enders (2010) also use the inflation threshold to estimate a nonlinear Taylor rule. Hwang and Wu (2011) find that there is an asymmetric and nonlinear relationship between inflation and economic growth in China; Liu and Pang (2011) point out that asymmetric effects exist between monetary policy and inflation. The reaction of inflation to various macro-variables varies and may show nonlinearity in both low and high inflation regimes. Most studies focus on the asymmetric effects of monetary policy in developed countries, rather than developing countries, such as China. Although standard recursive threshold VAR analysis has been applied in some studies, those studies do not extend this methodology to threshold nonrecursive structural VAR. A significant advantage of threshold nonrecursive structural VAR analysis is that it allows for both regime change and nonlinearities when analyzing the causes of China s high inflation. The inflation threshold allows for modeling nonlinearities empirically. The inflation threshold model separates a low inflation regime and a high inflation regime using different sets of model parameters. The inflation threshold separates sample data from 2001 to 2011 into two regimes: low and high inflation regimes. When inflation exceeds a given threshold value, the time point belongs to a high inflation regime. The time point belongs to a low inflation regime when inflation is lower than the given threshold value. Introducing an inflation threshold value allows division of the time sample into low and high inflation regimes. The inflation threshold enables the study of asymmetric effects in different inflation regimes. However, there is another question: How do we analyze the causes of China s inflation in low and high inflation regimes? In order to solve this problem, the present paper employs a nonrecursive structural VAR model. This paper extends the recursive structural VAR approach to a nonrecursive structural VAR approach (see Kim and Roubini, 2000; Kim, 2001), because a nonrecursive VAR can identify the coefficients of the contemporaneous relationships. Thus, by employing nonrecursive structural VAR analysis, the impulse responses of inflation to macro-variables can be captured in low and high inflation regimes. Therefore, threshold nonrecursive structural VAR analysis is a very important new method for studying the causes of China s high inflation. The present paper uses threshold nonrecursive structural VAR analysis to investigate 14

16 whether macroeconomic variables have asymmetric effects on China s inflation in low and high inflation regimes. In particular, this study pays attention to the asymmetric monetary effects in low and high inflation regimes, and attempts to capture the different impacts on China s inflation in low and high inflation regimes. The empirical evidence extends the literature in four ways. First, the reactions of inflation to various shocks are inflation-regime-dependent and asymmetric. Second, adjusting the domestic interest rate in China may be an effective way to control inflation in a high inflation regime, but not in a low inflation regime. The results are consistent with findings of the previous study (See Van Der, 2004 ) that when inflation breaches its target level, adjustment in interest rates is required to move in the same direction as the expected inflation. Third, oil prices might not be as important as generally considered in pushing China s high inflation. Oil price increases may increase inflation in a high inflation regime, whereas there is no such obvious effect in a low inflation regime. The most intuitive explanation is that in a high inflation regime oil price shocks are indicative of the increased scarcity of energy. In a high inflation regime, the speed of adjustment of inflationary expectations is rapid. Thus, the oil price increase induces firms to invest in more energy-efficient capital because of the inflationary expectation in the future. Finally, a nominal effective exchange rate appreciation might be effective in controlling domestic inflation in both regimes. This is because a stronger RMB may very well curtail Chinese domestic inflation. Chinese consumers will find cheaper foreign goods in their markets and, in return, this will put downward pressure on similar domestic goods. Chinese importers of foreign raw materials and intermediate components will enjoy lower relative prices. Furthermore,prices on imported petroleum and petroleum products will also be reduced. The present paper is organized as follows. Section II summarizes the threshold non-recursive structural VAR modeling method and presents the details of the identification scheme for inflation shocks. Section III describes the data and presents a preliminary analysis. Section IV estimates the inflation threshold. Section V discusses the empirical results using threshold nonrecursive structural VAR analysis. Section VI illustrates the robustness of empirical results. Section VII summarizes the results and concludes. 1.2 Analytical Framework The present paper uses the Tsay (1998) method to estimate the threshold value and threshold variables. The effects of inflationary expectations might take some time to manifest themselves. Therefore, the threshold variable may be composed of lagged values of the inflation rate or its moving average. Following Tsay (1998), Shen and Chiang (1999) and Balke (2000), a threshold VAR can be interpreted as follows: 15

17 Z t = + A Zt + B ( L) Zt 1 + ( σ + A Zt + B ( L) Zt 1) I( Ct d, τ ) σ + ε, (1) t where Z t = ( ΔOILt, GAPt, ΔPRt, ΔIFt, ΔNEERt ) is a vector containing oil price, output gap, inflation, interest rate and nominal effective exchange rate. 0, if Ct d < τ I ( Ct d, τ ) = {, (2) 1, if C > τ t d where C t d represents the threshold variable that determines the dynamics of Z t in different regimes, τ represents the inflation threshold value, and I represents an indicator function. I equals 0 when C t d is less than τ and 1 otherwise. Therefore, Equation (1) can also be described as: Z t σ + A Z t + B ( L) Z t 1 = { σ + σ + ( A + A ) Z t + ε. t 1 + [ B ( L) + B 2 ( L)] Z t 1 + ε. t if I = 0 if I = 1. (3) If I = 0, the relevant coefficients are σ 1, A 1 and B 1 (L), and the relevant coefficients are σ 1 + σ 2, A 1 + A 2 and B 1 (L) + B 2 (L) when I = 1. B 1 (L) and B 2 (L) represent polynomial matrices. ε t represents a vector of serially unrelated structural innovations. A 1 and A 2 represent the structural contemporaneous relationship in different regimes, and Equation (1) is thereby generalized to a multivariate framework. Using this multivariate framework avoids the limitation of the single equation approach, and it is able to capture the dynamic interactions of oil price, output, interest rate, inflation and nominal effective exchange rate. Not only inflation but also oil price, output, interest rate and nominal effective exchange rate shocks may have an impact on inflationary conditions and, furthermore, affect economic activity. In order to capture these interactions, the present paper has employed nonrecursive structural VAR analysis. The analysis allows for both regime change and nonlinearities in low and high inflation regimes, depending on the value of inflation threshold. Thus, this section generalizes the threshold model proposed above so as to incorporate the inflation regime into the dynamic system. For each inflation regime, nonrecursive structural VAR analysis is introduced to describe the analytical framework. The nonrecursive structural VAR model is based on the following considerations: first, oil price is included in the structural VAR model, following McCarthy (2000) and Hahn (2003), and supply shocks are captured by the world oil price index. This is because if the monetary contraction is carried out when inflationary pressure is being faced, price deflation and economic recession are not only due to the tight monetary policy but also due to the original 16

18 negative supply shocks. To identify the monetary policy effect, the world oil price index is introduced as a proxy for supply shocks. Second, to identify the demand-side effect, the paper follows Hahn (2003) and constructs the output gap by applying the Hodrick Prescott (HP) filter to the industrial production index. Third, the interest rate is used to allow for the effects of monetary policy on inflation. The nonrecursive structural VAR analysis can identify monetary policy shocks; thus, it can reveal whether or not domestic inflation is caused by mismanaged monetary policy. Fourth, the nominal effective exchange rate is used in nonrecursive structural VAR analysis. Compared with the bilateral exchange rate vis-à-vis the US dollar, the effective exchange rate is the right choice to identify the monetary policy effect for a country with diversified trading partners, such as China. Fifth, the consumer price index (CPI) is included in the nonrecursive structural VAR model to measure the domestic price inflation. The nonrecursive structural VAR model can be interpreted as follows: G ( L) Z t = e t, (4) where G(L) is a matrix polynomial in the lag operator L, Z t is an n 1 data vector, and e t is an n 1 structural disturbance vector. e t is serially uncorrelated and var(e t ) = Λ. Λ represents a diagonal matrix where diagonal elements are the variances of structural disturbance; hence, structural disturbances are assumed to be mutually uncorrelated. Then, the reduced-form equation can be estimated as: Z = B( L) + ε, (5) Z t t where B(L) is a matrix polynomial (without the constant term) with lag operator L. Bernanke (1986) and Sims (1986) suggest that the nonrecursive structural VAR allows the imposition of restrictions to contemporaneous structural parameters. G 0 is a non-singular coefficient matrix of L 0 in G(L), which represents the contemporaneous coefficient matrix in the structural form, and G 0 (L) is the coefficient matrix in G(L) without contemporaneous coefficient, G 0. This relationship can be described as: 0 G ( L) = G + G ( ). (6) 0 L Therefore, the parameters in the reduced-form equation and those in the structural form equation can be described as: 17

19 B ( L) = G G. (7) 1 0 Λ 1 0 Only through sample estimates of Σ can the maximum likelihood estimates of Λ and G 0 be obtained. The right-hand side of Equation (7) has n 2 unknown parameters to be estimated. Because Σ contains n (n + 1)/2 known parameters, here at least n (n 1)/2 restrictions are necessary. Therefore, at least n (n 1)/2 restrictions imposed on G 0 are needed to achieve identification. In the VAR modeling with Cholesky decomposition, G 0 is supposed to be triangular. Whereas in the structural VAR approach G 0 can be recursive or nonrecursive structure as long as it has enough restrictions for achieving identification. In this paper, the data vector is {OIL, GAP, IF, PR, NEER}, where OIL is the 3-spot oil price index (Dubai, the UK and Texas), GAP is the output gap in China, IF is domestic inflation in China, PR is the interest rate in China and NEER is the nominal effective exchange rate in China. For the restrictions on the contemporaneous structural parameters, G 0, we follow Sims and Zha (2006) and modify their model. The present paper proposes a structural VAR approach with contemporaneous restrictions. It allows modeling nonrecursive restrictions across different equations. To indentify the structural shocks, a Sims Bernanke decomposition of the nonrecursive matrix and a variance covariance matrix of the reduced-form VAR residual (ε t ) are used to generate the structural disturbance (e t ). Hence, the contemporaneous relationships between the structural disturbances and the reduced-form VAR residuals can be demonstrated as follows: e e e e e OIL GAP IF PR NEER = ε OIL = g21ε = g31ε = g42ε OIL OIL GAP = g51ε OIL + ε GAP + g32ε GAP + g43ε IF + g52ε + ε + ε GAP IF PR + ε + g34ε NEER PR + g35ε NEER, where e OIL, e GAP, e IF, e PR and e NEER are the structural disturbances, which represent oil price shocks, output gap shocks, inflation shocks, interest rate shocks and nominal effective exchange rate shocks, and ε OIL, ε GAP, ε IF, ε PR and ε NEER are the reduced-form residuals (i.e. unexpected movements of OIL, GAP, IF, PR and NEER, respectively). Hence, contemporaneous relationships can be modeled as: 18

20 e e e e e OIL GAP IF PR NEER 1.0 g21 = g g g32 g42 g g g ε 0.0 ε g35 ε 0.0 ε 1.0 ε OIL GAP IF PR NEER. (8) Note that this is an overidentified system in which 11 elements of G 0 have been restricted to zero. 2 The present paper establishes the model to include an international/foreign variable, and the world oil price index. The first equation assumes that the oil price is contemporaneously exogenous to other variables. The second equation represents the real sector, and it is influenced by the oil price. The third equation represents inflation, and inflation is allowed to reflect contemporaneous information on all variables in the system. The fourth equation is the interest rate, and the monetary authority is assumed to adjust the interest rate after observing the performances of the real sector and domestic inflation. The last equation represents the nominal effective exchange rate. Sato and Ito (2008) use a VAR model to discuss the relationship between inflation and the exchange rate. They suggest including the nominal effective exchange rate in a structural VAR to analyze the causes of domestic inflation. The present paper assumes that the nominal effective exchange rate is influenced by the oil price and the industrial production index. 1.3 Data and Preliminary Analysis We use monthly data that cover from January 2001 to April The variables in this paper include the 3-spot oil price index (Dubai, the UK and Texas), China s industrial production index, China s consumer price index, the interest rate in China and China s nominal effective exchange rate, where a nominal effective exchange rate increase represents RMB appreciation. The dataset is large enough to contain two major low inflation periods after economic contraction: one in 2001 and another between the global financial crises. It also includes two major high inflation periods: one between 2007 and 2008, and another that starts in the latter half of The data sources for the variables are the IMF International Financial Statistics CD-ROM and the CEIC Asia Database. To ensure the stationarity of the variables, year-on-year growth rates are taken for oil price index, consumer price index, interest rate and nominal effective exchange rate, expressed as OIL, IF, PR and NEER. According to Hahn (2003), this paper also applies the HP filter to the industrial production index, and uses its cyclical component, expressed as outgap (GAP). 2 Because n = 5, exact identification entails n(n 1)/2 = 10 restrictions, where n represents the number of endogenous variables. 19

21 The augmented Dickey Fuller test and the Phillips Perron test are used to test the stationary of the time series properties for the five variables. The unit root tests are reported in Table 1. Table 1. Unit Root Test ADF test PP test Variables Level YOY Level YOY OIL 4.229*** 7.047*** *** GAP 4.732*** *** IF *** *** PR *** *** NEER *** *** Source: CEIC Asia Database. Notes: OIL represents the 3-spot oil price index (Dubai, the UK and Texas); GAP represents the cyclical component of the industrial production index; IF represents the consumer price index; PR represents the interest rate; NEER represents the nominal effective exchange rate. The null hypothesis of the augmented Dickey Fuller (ADF) and the Phillips Perron (PP) tests is that the variable is nonstationary. The sample period is from January 2002 to April For the level variables in the table, constant and time trend are included. For the year-on-year growth rate (YOY) variables in the table, only constant is included. ***, denotes significance at the 1-percent level. According to Table 1, both the augmented Dickey Fuller test and the Phillips Perron test indicate that most of the variables are non-stationary in level but all the variables are stationary in the year-on-year growth rate series or cyclical component-t series. 1.4 Estimation of the Inflation Threshold To capture the asymmetric and nonlinear impacts in different inflation regimes, first it is necessary to estimate the inflation threshold. The inflation threshold allows for modeling nonlinearities empirically. Therefore, in this section, the Akaike information criterion (AIC) will be applied to determine the inflation threshold for China, aiming at separating the low inflation regime and the high inflation regime using different sets of model parameters. Based on the inflation threshold value, the sample data during 2001 and 2011 can be grouped into low and high inflation regimes. When inflation exceeds a given threshold value, the time point is grouped with a high inflation regime. The time point is grouped with a low inflation regime when inflation is lower than the given threshold value. 20