Journal Issue
Share
Article

People’s Republic of China: 2011 Spillover Report—Selected Issues

Author(s):
International Monetary Fund
Published Date:
July 2011
Share
  • ShareShare
Show Summary Details

VI. CHINA SPILLOVERS: ANALYSIS FROM A GLOBAL VAR1

This chapter analyzes the spillovers from China using a global vector autoregression that allows for complex interactions among a large number of economies. The impact of cyclical factors, such as worsening credit quality, is assessed, along with exchange rate appreciation and structural reforms for rebalancing demand. Deteriorating credit portfolios would adversely impact trading partners, particularly those in the Asian supply chain. On the other hand, the benefit to trading partners of Chinese structural reforms and real appreciation depends on their location in the supply chain.

1. Global VAR. A global vector autoregression (GVAR) provides a dynamic multi-country framework suitable for the analysis of interdependence and international transmission of shocks. In this note, a GVAR model comprising 30 countries—21 advanced and 9 emerging—is used, based on Chen et al. (2010, IMF Working Paper WP/10/124; see Appendix I). The model has real and financial variables: industrial production (a proxy for GDP), real effective exchange rate, real money market rate, real share prices, and a measure of potential financial stress (the asset-weighted average expected default frequency of banks and nonfinancial corporates). The latter captures the role of credit in the transmission of financial and real sector shocks. Monthly data are used (Jan 1996–Dec 2008).

2. Spillovers. The model is used to analyze the spillover effects of policies or shocks from China. Given China’s rapidly changing economic structure, an estimated model like the one used here cannot speak to the long-term impact of changing policies or shocks. Conjunctural concerns regarding the pace and quality of credit growth are assessed by examining the effects of deteriorating balance sheets. Balance sheet quality is captured through estimated default probabilities (EDFs); these have predictive power on the strength of credit growth one year ahead (the unconditional correlation between credit growth and one-year lagged EDF is around -0.5), even though credit is highly influenced by official actions in China. Structural reforms in support of rebalancing the economy, such as financial sector reforms that help to level the playing field in production, would be expected to increase Chinese productivity. Only the aggregate impacts of such reforms and of exchange rate appreciation are assessed (it is not feasible to report standard errors, given that weighted maximum impulse responses are reported).

3. Financial shocks. A deterioration in the quality of non-financial corporate as well as financial balance sheets in China would have substantial negative effects on the rest of the world, highlighting China’s increasingly important role, including in the global supply chain.

  • Corporates. An increase in Chinese corporate default probabilities by 1 pp could lower industrial output in other economies by as much as ⅓ percent among emerging Asian economies and ⅕ percent of GDP in commodity exporting economies. Declining equity prices would have a qualitatively similar impact (see figure below).

  • Banks. A similar deterioration in bank balance sheets would have a somewhat larger impact. An increase in Chinese banks’ default probability by 1 pp could lower industrial output in Japan and emerging Asia by about ⅓ percent. U.K. industrial output would also be lower by about ½ percent. Many other economies, such as in the Euro Area, could also experience some decline in their industrial production (see figure below).

  • Robustness. Note that the measure of financial stress is indicative of problems in balance sheets that would affect real activity, at least until policies are implemented to resolve the problems directly or to transfer them out of the banks’ balance sheets and, as some expect, onto the balance sheet of the public sector. A financial stress scenario performed in the context of a structural macroeconometric model of the G20 (Vitek, 2010, IMF Working Paper WP/10/152) yields broadly similar results.

Illustrative Impact on Industrial Output of a Deterioration in Chinese Corporate Balance Sheets

(1 p.p. increase in default probabilities)

Illustrative Impact on Industrial Output of a Deterioration in Chinese Bank Balance Sheets

(1 p.p. increase in default probabilities)

4. Rebalancing. Structural reforms would be expected to boost productivity in manufacturing, which accompanied by greater import substitution would generally have an important impact on the supply chain. Economies upstream would benefit from higher output in China at lower cost, while the demand for inputs from suppliers would decline. A rise in industrial output by 1 pp would raise output by about 0.05 percent in economies at the end of the supply chain, such as advanced economies (see figure below). The impact downstream (mainly, emerging Asia) is estimated to be negative, albeit small, owing to increased capacity in China.

5. Appreciation. RMB real effective appreciation would lower output in China but have positive spillovers for most economies, especially commodity (e.g., Latin America) and producers of intermediate goods (e.g., emerging Asia). A 10 percent real effective appreciation could raise output by close to 1 percent in many advanced economies (e.g., Japan, U.K.) but less so in the U.S. (see figure below). Such an appreciation would also lower output in China. Symmetrically, a real effective depreciation boosts growth in China, at the expense of many partners.

Illustrative Impact on Industrial Output of Higher Productivity in China’s Industrial Sector

Illustrative Impact on Industrial Output of a 10 Percent Rise in China’s REER

Appendix I. The GVAR Model

Structure of the GVAR Model

The structure of the GVAR model can be summarized as the follows. Consider N+1 economies, indexed by i = 0, 1, 2,…, N, and a vector xit of ki domestic variables for each economy. Stacking the vectors of country-specific variables,

a VAR in xt would contain too many parameters to be estimated if the time dimension T of the data is not much larger than the number of economy N. Instead of regressing xi,t on

without any restriction, GVAR links xi,t to a k*i × 1 vector x*i,t, where

The weight ωlij captures the spillover effect of variable l of foreign economy j on variable l of domestic economy i. Since ωlij measures the relative importance of economy j to economy i, the spillover effect of variable l is in proportion to the weight chosen to measure the relative importance. Therefore, each economy’s component of GVAR is given as a VARX*(pi, qi):

where dt–s is the observed common factor of q × 1 dimension and εit is iid across time. Country-specific vector xi,ts* reflects interdependence among economies and serves as a proxy for the unobserved common effects across economies. The country-specific foreign variables and common factors are treated as weakly exogenous (if confirmed by statistical tests), i.e., they are “long-run forcing” country-specific domestic variables. The term “long-run forcing” means that in the equations for foreign variables, the coefficients on the error-correction terms are set to zero. The dynamics of foreign variables are not influenced by deviations from the long-run equilibrium path, in contrast to the dynamics of domestic variables.

The VARX* can be estimated economy by economy using the ordinary least squares (OLS) method or rank-reduced approach if the cross-dependence of the idiosyncratic shock is sufficiently small, that is:

all i≠j, l and s.

From equation (3), it can be seen that

Where zit =(xitxit*) and Wi is an appropriately defined weighting scheme. Thus, stacking (4) across i, the endogenous variables can be solved for in a global system:

thus

Where p = max{pi, qi}, r = max{ri}, and

Equation (8) is a VAR for the complete set of domestic variables for all economies.

The advantage of the GVAR model is that it makes the estimation of (8) feasible by accounting for interdependence among economies and then estimating the partial system on a economy-by-economy basis, which implies allowing for modeling a large number of economies. The impulse response is computed based on (8).

The vector for domestic variables is given by:

where edfbit denotes the logarithm of asset-weighted average expected default frequency (EDF) of banks and edfntt for (nonfinancial) corporates rit is the real money market rate, yit is the logarithm of industrial production, pits the logarithm of real share price index, and qit is the logarithm of the real effective exchange rate.

The vector for foreign variables for each economy except the United States is given by:

We do not construct foreign effective exchange rates to minimize the number of parameters to be estimated, since information about foreign economies’ currency is captured in the (trade-weighted) real effective exchange rate qit.

The foreign variable for the United States is constructed as:

Given the large influence of the U.S. financial variables on global markets, the U.S. foreign financial variables are less likely to be weakly exogenous for the U.S. domestic variables. That is the main reason we do not include the U.S. foreign financial variables in the equations for the United States.

The spot oil price is included as a common factor dt–s to remove the common component in the reduced form residuals. Another candidate for inclusion as a common factor could be the index of global stock price volatility VIX, to ensure that the EDF shocks are purely idiosyncratic. However, because the VIX is driven by volatility in U.S. share prices, it is not weakly exogenous to the U.S. variables. Adding it separately will not augment the information content of the model.

Equations (3) and (4) show that the spillover effect of a foreign variable on a domestic variable is proportional to the weight ωlij, which measures the relative importance of economy i to economy j in the transmission. Since the transmission channels for financial variables are likely to be different from the transmission channels for the variables measuring real activity, we use financial weights to construct foreign financial variables—EDFs, real money market rate, share price index and real effective exchange rate—and trade weights for industrial production.

Impulse Responses

Given the short sample period, the study focuses on short-run dynamics. The model is estimated in first differences as the macroeconomic and financial data are found to be integrated of order 1. Identifying the complete set of shocks in equation (8) and computing the impulse response functions in a GVAR model is not straightforward. It requires imposing an enormous amount of identification restrictions due to the large number of economies covered in the study. Therefore, we identify shocks following the approach in Dees, Di Mauro, Pesaran and Smith (2007) and Binder, Chen and Zhang (2009).

After identifying the EDF shocks, we compute impulse responses of the other variables in the global solution in equation (8) based on correlations between the reduced form shock of each variable and the identified structural shock of the EDF. Such an identification scheme means that zero correlation between the structural EDF shocks and other domestic variables in each economy need not be imposed and the transmission of the shock is determined without any additional restrictions.

Data Description

VariableDescriptionSourceNotes
edfbAsset weighted one year ahead expected default probability of (broadly defined) financial firmsMoody’s KMVData for China from March 1996 to April 1997 are not available, and are interpolated in a linear manner.
edfnAsset weighted one year ahead expected default probability of non-(broadly defined) financials firmsMoody’s KMVMissing data for October 1996 is interpolated.
yLogarithm of industrial production indexGDS for Australia and New Zealand; CEIC for Brazil, China, Hong Kong SAR, Indonesia, Malaysia, Philippines, Singapore and South Africa; IFS for all other economies.Data for China is the value added of industry, which to our knowledge the closest available measure of the industrial production. The series is spliced with the implied value from the year on year growth value from 1995 January onwards.

All data from CEIC and for India are available in seasonally unadjusted form and adjusted using Census X12 in EViews.
rMoney market rate deflated by consumer price index (CPI)Money market rates are from IFS and CEIC.

Consumer price indices for Australia and New Zealand are from GDS, while the rest economies are from IFS. The 7 day weighted average CHIBOR is used for China.
Data for Sweden from December 2004 onwards are not available in the IFS, and the policy-related interest rate from the GDS is taken instead. Missing data for September 1992 is interpolated.
psLogarithm of share price index deflated by CPIIFS
qLogarithm of real effective exchange rateData for Hong Kong SAR, Indonesia, Mexico and Turkey are from CEIC, while the rest are from IFS.
poLogarithm of world spot petroleum priceIFS

Prepared by Papa N’ Diaye (APD) and Nathan Porter (SPR).

Other Resources Citing This Publication