3.5 Robustness
4.7.4 Generalized impulse response analysis
Following Koop et al. [1996], the Monte Carlo integration procedure used to com- pute the non-linear generalized impulse response functions for the VAR-SV model discussed in Section 4.2.4 is outlined as follows:
1. Take a Gibbs draw of the states; βt, h•,t and at, using the algorithm in Ap- pendix 4.7.2
2. Identify the time varying structural impact matrix: At, using the algorithm in Appendix 4.7.3
3. Using the Gibbs draws of the current states along with the impact matrix, simu- late future paths of the elements in the covariance matrix: h•tandat, 20 periods into the future.
4. Using the impact matrix found in Step 2, recover the structural innovations using the identity: ut = Atet, where ei,t ∼ N(0, 1) and i ∈ {1, 2, 3, 4}. The benchmark model is then given by: E[yt+k|It], whereIt denotes the informa- tion set up to time t.
5. Using the structural innovations found in Step 4, compute the contemporane- ous impact of a unit shock using the identity: ˜ut = Ate˜t. For instance if we shocky1,t then ˜et = (e1,t+1,e2,t,e3,t,e4,t)where ei,t is thei-th entry of et. The model with the shock is then given by: E[yt+k|e˜t,It], where It denotes the same information set in Step 4.
6. Derive the generalized impulse response functions by taking the difference be- tween the shocked model in Step 5 and the benchmark model in Step 4. We repeat this procedure every 100th draw from the stationary distribution obtained in the MCMC Algorithm described in Appendix 4.7.2. This yields a set of 2000 responses of the four endogenous variables over the accepted rotations at each point in time. The representative impulse responses at each point in time are then taken to be the median of this distribution.
Chapter5
Time-varying Macroeconomic
Effects of Energy Price Shocks: A
New Measure for China
5.1
Introduction
Despite the important of China as a major economic player and energy consumer in the world, there has seen a small literature on the relationship between energy price shocks and China’s macroeconomic variables (Tang et al. [2010] Du et al. [2010], Cunado et al. [2015], Wei and Guo [2016], Herwartz and Plödt [2016] and Cross and Nguyen [2017a]). Consistent with existing literature in the US and other devel- oped economies, the standard approach to modeling energy price shocks has been to examine the effects of an exogenous, unanticipated rise in the price of imported crude oil prices (see, e.g., Hamilton [1983], Kilian [2008a, 2009, 2014], Peersman and Van Robays [2009], Peersman and Van Robays [2009], Lippi and Nobili [2012] or Baumeister and Peersman [2013b] among others). While an examination of the rela- tionship between oil price shocks and Chinese macroeconomic variables is of interest in it’s own right, a deeper investigation into the structure of China’s quarterly energy expenditure shares reveals that the use of oil prices as a proxy for modeling more general energy price shocks paints an incomplete picture. After establishing this fact, the objective of this paper is to propose a new index of quarterly energy prices, which accurately reflects both the structure of China’s total energy expenditure shares on primary commodities, along with intertemporal fluctuations in international energy prices. Once established, the index is then used alongside three key macroeconomic variables; inflation, real GDP growth and a short term interest rate, to investigate the effects of energy price shocks on China’s macroeconomy over the past two decades. To provide evidence in support of the claim that oil prices are not a good proxy for
80Time-varying Macroeconomic Effects of Energy Price Shocks: A New Measure for China
more general energy price dynamics, Figure 5.1 compares the expenditure shares on primary energy commodities: coal, crude oil and natural gas, within China and the US, over the period 1993Q1-2016Q2.1
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 Percent 0 20 40 60 80 100 Oil Coal Natural Gas (a) The US 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 Percent 0 20 40 60 80 100 Oil Coal Natural Gas (b) China
Figure 5.1: Primary energy consumption share by fuel type (1993-2016)
Note: Data is retrieved from EIA database website (World Energy Balances), measured in energy units: thousand tonnes of oil equivalent (ktoe). Shares in 2015 and 2016 are assumed equal to the previous year 2014.
Two points are worth emphasizing. In the first instance, while oil expenditure is clearly the dominant share of total energy expenditure in the US, the data reveals that coal, not oil, is the major source of energy expenditure in China. More precisely, the total expenditure on oil in the US contributed around 70 percent of the primary commodity energy expenditure share, with coal and natural gas contributing the re- maining 30 percent. In contrast, oil expenditure in China’s economy accounts for just 34 percent of the total energy expenditure share, with coal contributing to 62 percent and natural gas the remaining 4 percent. In addition to this fundamental difference in average expenditure shares, we also highlight the fact that while the primary en- ergy expenditure shares in the US are relatively stable, China’s expenditure shares have significantly changed over time. More precisely, the total energy expenditure on oil in 1993 was just 24 percent of total energy consumption, compared to 35 percent in 2016. Similarly, the total energy expenditure on natural gas has grown from just 2 percent in 1993 to 8 percent in 2016.
Combined together, these simple empirical facts suggest that the common use of oil prices as a proxy for more general energy price dynamics in studies on the US economy, does not extrapolate to the case of China. Instead, since it comprises the
1As later discussed in Section 2, this sample period corresponds to the longest set of available data
for China’s macroeconomic variables. We also highlight the findings of a recent report from the US Energy Information Administration (EIA), which stated that the combined expenditure on these three primary energy commodities accounted for 91 percent of total China’s energy consumption in 2012, compared to 81 percent in the US.
§5.1 Introduction 81
highest average expenditure share, a superior proxy is provided by coal. That being said, the second observation suggests that a superior proxy to using coal prices can be obtained by an energy price index that accurately reflects both the the structure of China’s energy expenditure shares along with fluctuations in international energy prices. For this reason, the first objective of this paper is therefore to develop such an index. Once established, we then investigate the effects of energy price shocks on China’s macroeconomy over the past two decades.
To this end, our empirical analysis employs a sufficiently rich set of time varying BVAR models: a traditional constant parameter VAR, a time varying parameter VAR, a constant parameter VAR with stochastic volatility and a fully flexible time varying VAR with stochastic volatility. The motivation for this set of econometric models stems from the recent work of Cross and Nguyen [2017a], who show that when modeling the relationship between China’s economic growth and global oil price shocks, a time-varying parameter BVAR with stochastic volatility provides superior in-sample fit as compared to its time-invariant counterparts.2 To elicit a distinction
between any sources of time variation within the endogenous relationship between energy prices and China’s macroeconomy and any time varying volatility in any of the innovations, we conduct a formal Bayesian model comparison exercise through which the relevant features of the data are identified. After selecting the best reduced form model, the associated structural VAR (SVAR) is identified through the use of a new set of agnostic sign restrictions, which are motivated by structural differences between China and the US.
Our analysis yields three intriguing results. First, from a modeling perspective, the time varying parameter VAR with stochastic volatility is shown to provide the best in-sample fit of the data. Next, positive energy price shocks have consistently gener- ated economic stagflation over the past two decades. Interestingly, while the inflation responses are consistent over the inflation period, the real GDP responses have de- creased by around half over the sample period. Next, the interest rate responses are found to depend on the choice of monetary policy instrument. More precisely, the response of the PBOC bank rate to an energy price shock was found to be increasing over the sample period, suggesting that the PBOC is becoming more focused on in- flation stabilization as compared to facilitating output growth. On the contrary, the
2Since it is closely related to our current paper, we highlight two major distinctions between Cross
and Nguyen [2017a] and the present study. First, the development and use of the energy price index allows us to model more comprehensive energy price dynamics than those attributed to oil prices. Next, the present study seeks to investigate the relationship between energy price shocks and China’s macroeconomy in general, not just real GDP growth.
82Time-varying Macroeconomic Effects of Energy Price Shocks: A New Measure for China
results under the required reserve ratio were found to be consistently positive over the sample period. These results are shown to be robust to both official national data and those developed by Chang et al. [2015], thus strengthening the conclusion that energy price shocks have of significant time varying effects on China’s macroecon- omy.
The remainder of the chapter is organized as follow. First, Section 5.2 presents the new energy price index and data sources. Next, Section 5.3 outlines the econometric methodology including the various model specifications, model comparison tech- niques, identification strategy and computation of the non-linear impulse response functions. Sections 5.4 and 5.5 then present the results along with various robustness checks. Finally, Section 5.6 concludes.