Information about Asia and the Pacific Asia y el Pacífico
Chapter

Appendix II Money, Activity, and Prices in China: An Analysis of Causality

Author(s):
David Burton, Wanda Tseng, Kalpana Kochhar, Hoe Khor, and Dubravko Mihaljek
Published Date:
September 1994
Share
  • ShareShare
Information about Asia and the Pacific Asia y el Pacífico
Show Summary Details

The principal purpose of this analysis is to study the joint movements of prices, activity, and monetary and credit variables in China over the past decade. The aim is twofold. First, the analysis attempts to shed light on the causal relationships underlying the recurrent macroeconomic cycles in China. In particular, the hypothesis that stop-go cycles were due in large part to autonomous increases in aggregate demand, especially for investment, is examined. Second, in light of the current efforts to reform the monetary policy framework and move away from direct credit controls, the study examines the transmission mechanism of monetary policy, that is, the channels through which monetary policy changes influence the economy. The first part of this appendix contains a brief discussion of the vector autoregression (VAR) methodology used and the ways in which the results can be interpreted. The second part presents the main results of the estimation of VARs for China.

Methodology

One way to conduct an analysis of the joint movements of major macroeconomic variables is to construct a full-fledged structural model spelling out the theoretical relationships between them. This approach also requires assumptions to be made about the exogeneity of certain variables vis-à-vis others in the model and the exclusion of some variables from some of the equations of the system. An alternative, more flexible way is to specify and estimate a VAR in which no a priori assumptions need to be made about either the structural relationships underlying the variables in the model or the exogeneity of particular variables. This methodology allows the researcher to manipulate reduced-form systems so as to characterize joint movements among a wide range of potentially endogenous variables. In general, the VAR is set up as follows:

where X, Y, and Z are the main variables of interest in the model and B(L) is a 3 x 3 matrix of polynomials in the lag operator, with

BOLOXt = BOXt, BtL1Xt = B1Xt–1, and so on.

This study will exploit the “agnostic” approach of the VAR technique to examine empirically some prior notions about the causality pattern of certain macroeconomic variables that have played a crucial role in the cycles in China. Specification of a VAR generally involves selecting the variables represented by X, Y, and Z, as well as their lag lengths. Clearly, the availability of data constrains both these choices. For this analysis, the set of variables includes broad money, narrow money, currency in circulation, domestic credit, net domestic assets, fixed investment, retail sales, imports (all in nominal terms), industrial production (real), and inflation. The models have been estimated using quarterly, seasonally adjusted data over the nine years from 1985 to 1993.100° The variables were all tested for unit roots and found to be nonstationary in levels and stationary in first differences. All variables are therefore expressed in log differences. Based on the Schwarz-Bayesian Information Criterion, a constant and two lags of each variable have been used in the VAR.101

One interesting way of summarizing the results of a VAR estimation is to examine the Granger causality between the variables in the model (Engle and Granger (1987)). A variable x is said to “Granger cause” another variable y if lagged values of x are significant in the equation for y. The estimation procedure produces a set of F-test statistics that test the null hypothesis that the coefficients of all lags of a particular variable in each equation are zero. The figures shown in the tables below are the probabilities of accepting the null hypothesis. For example, a p-value of less than 0.1 on variable x in the equation for y would imply that lagged values of x have significant incremental predictive power in the evolution of y.

Another way of summarizing the results of a VAR is to decompose the variance of the forecast error of a variable of interest into the shares attributable to innovations in each of the other variables in the system. To do this, it is necessary to first transform the autoregressive representation of the system into a moving average representation, as follows:

In general, the covariance matrix Σ of the residuals, ∊, is not diagonal, that is, off-diagonal elements of the matrix representing covariances between any two variables in the system are nonzero. The residuals of this moving average representation thus need to be orthogonalized, as orthogonalization eliminates the covariances among the shocks to each explanatory variable, so that the covariance matrix of the residuals is diagonal. This process allows the relative impact of each variable to be identified separately. In this study, the Choleski factorization procedure is used, in which the covariance matrix Σ is factorized into two matrices S and S’. The matrix S is “lower triangular,” that is, its diagonal elements are positive, the elements below the diagonal are nonzero, and the elements above the diagonal are zero; it can be shown that such a factorization is unique, that is, there exists only one matrix S that is lower triangular so that Σ = SS’.

One important issue in the orthogonalization procedure is the choice of ordering of the variables, as the ordering can affect their measured impact. This is particularly true when the covariances among the residuals are significant. In general, variables that are expected to have little or no predictive power for other variables should be put last. The issue of ordering the variables is discussed further below.

Results

Granger Causality Tests

The first VAR system (VAR1) consists of broad money, domestic credit, fixed investment, industrial production, and inflation.102Table 14 contains a summary of the results. Each column represents an equation of the model.

Table 14.Summary Results of VAR1
Dependent Variables
Broad moneyDomestic creditFixed investmentIndustrial productionInflation
R20.590.450.360.610.58
P-valueof Q (15)10.810.270.960.130.88
P-value of F-tests of exclusion restrictions2
Explanatory variables
Broad money0.06*0.180.200.04*0.07*
Domestic credit0.280.320.570.610.07*
Fixed investment0.960.520.450.09*0.86
Industrial production0.260.890.250.110.05*
Inflation0.00*0.02*0.950.600.00*

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero. An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero. An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

Based on the results of these F-tests, a “causal mapping” of VAR1 can be derived (Chart 11).

Chart 11.Causal Mapping of VAR1

At the outset, it should be noted that these dynamics are very complex and, in some cases, difficult to interpret. This being said, the mapping shown in Chart 11 seems to suggest the following dynamics for the Chinese economy.

First, as no other variable in the system appears to have any predictive power in the equation for fixed investment, it could be said that fixed investment is “causally prior,” in the Granger sense, to the other variables in the system. This is, prima facie, in line with the casual empirical observation that fixed investment tends to lead the economic cycles. As noted in Section IV, the investment system has been based predominantly on the central planning system of project approvals rather than on a market mechanism. Moreover, with the decentralization of decision making, overall control over the volume and composition of investment has weakened, as local governments increasingly ignore central directives in the quest for rapid development of their local economies.

Second, the effects of fixed investment on the economy appear to work through its impact on industrial production. As was observed in the most recent cycle, the strong growth in fixed investment during 1991–92 led to rapid growth in industrial output and, in turn, to the overheating of the economy and rising inflation.

Third, with the exception of the inflation equation, changes in domestic credit have little explanatory power in the other equations. The limited role of domestic credit in the dynamics is somewhat unexpected, given the continued significant role of the credit plan in the economy.103 However, there are several reasons to attribute this limited role to the major structural changes that have taken place in the economy during the past decade. First, the domestic credit variable is based largely on the coverage of the credit plan, which is primarily relevant to the SOEs. However, the most dynamic part of the industrial sector in the past few years has been the nonstate sector, which now accounts for more than one half of industrial output and which has had to meet most of its credit needs outside the credit plan. Second, the past five years have witnessed very rapid increases in foreign direct investment, equity issues, and external borrowing, including through the issuance of bonds, all of which have become increasingly important avenues of financing. Another avenue is the diversification of domestic capital markets through the expansion of various forms of government and enterprise stock and bond issues.104 Third, it is possible that much of this investment was financed by funds channeled through NBFIs. In China, many of these NBFIs are subsidiaries of the specialized banks, and, in many instances, investment funds were channeled from the specialized banks through their subsidiary NBFIs. In terms of the monetary accounts, such financing would be captured in the net domestic assets of the banking system.

For this reason, the VAR has been re-estimated using net domestic assets instead of domestic credit (VAR2). The results are summarized in Table 15 and imply the causal mapping depicted in Chart 12.

Table 15.Summary Results of VAR2
Dependent Variables
Broad moneyNet domestic assetsFixed investmentIndustrial productionInflation
R20.590.520.400.630.54
P-value of Q(15)10.630.670.990.210.64
P-value of F-tests of exclusion restrictions2
Explanatory variables
Broad money0.640.480.480.10*0.39
Net domestic assets0.270.940.260.290.23
Fixed investment0.830.990.640.10*0.83
Industrial production0.10*0.07*0.300.05*0.18
Inflation0.00*0.00*0.990.820.01*

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero, An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero, An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

Chart 12.Causal Mapping of VAR2

Note that the dynamics of the system are largely the same as in VAR1. Fixed investment remains causally prior to the other variables in the system. Its impact on the rest of the economy appears to work through industrial production. Broad money and net domestic assets are “caused” by changes in inflation and industrial production. However, changes in the broader credit variable—net domestic assets—are also not significant in explaining any of the equations in the system.

In view of the possibility that the high degree of correlation between net domestic assets and broad money may have resulted in their joint effects being picked up by only one of the variables, the VAR was re-estimated by dropping broad money from the system and including only net domestic assets (VAR3). The results are different from the previous two cases (Table 16); the causal mapping is shown in Chart 13.

Table 16.Summary Results of VAR3
Dependent Variables
Net domestic

assets
Fixed

investment
Industrial

production
Inflation
R20.490.360.560.50
P-valueof Q(15)10.900.960.640.51
P-value of T-tests of exclusion restrictions2
Explanatory variables
Net domestic assets0.07*0.10*0.06*0.10*
Fixed investment0.950.420.340.92
Industrial production0.09*0.380.08*0.10*
Inflation0.00*0.880.430.00*

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero, An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

The reported p-values are the probabilities of accepting the null hypothesis that there is no serial correlation in the error term.

The reported p-values are the probabilities of accepting the null hypothesis that all lags of the explanatory variables are zero, An asterisk denotes rejection of the null hypothesis with 90 percent confidence.

Chart 13.Causal Mapping of VAR3

Fixed investment is no longer “causally prior” to the rest of the variables in the system; lagged values of net domestic assets are significant in explaining the evolution of fixed investment. Changes in net domestic assets are also significant in the equations for industrial production and inflation, although there is evidence of a feedback relationship between industrial production and inflation, on the one hand, and net domestic assets, on the other. This suggests that the broader concept of net domestic assets is a more meaningful credit variable than the narrower concept of domestic credit.

As for the issue of how monetary and credit impulses are transmitted to the rest of the economy, the results from the three VARs suggest that, as broad money and net domestic assets are significant in explaining the evolution of both industrial production and inflation, they could be seen as leading indicators for activity and prices.105 The finding that changes in net domestic assets help to predict industrial production suggests that the credit constraint hypothesis extended by Calvo and Coricelli (1993) may be relevant in the case of China. However, to subject the Calvo-Coricelli hypothesis to a more rigorous test, it would be more appropriate to limit the coverage of industrial output to the SOEs and specify the domestic credit variable in real terms.

Variance Decompositions

Table 17 shows the variance decompositions of the forecast error for inflation two years ahead. As the causality mappings charted above suggest relatively complex patterns of predictive power, several different orderings were examined. The sequence followed in the first ordering is broad money, domestic credit, fixed investment, industrial production, and inflation, while the sequence followed in the second is fixed investment, industrial production, inflation, broad money, and domestic credit. The variance decompositions suggest that, although inflation is explained at the end of a two-year horizon largely by its own innovations, it is also explained to a significant extent by innovations in industrial production and fixed investment. Innovations in money and domestic credit appear to have a smaller impact on inflation. In the third and fourth orderings, net domestic assets is substituted for the narrower domestic credit variable. In these variance decompositions, the broader credit variable is assigned a higher proportion of the variance of the forecast error for inflation. These findings would suggest that, first, the inflation process in China has been partly a result of structural changes (such as price adjustments and price controls), and, second, that activity variables, as well as the broader monetary and credit aggregates, are also important in the inflation process.

Table 17.Inflation: Forecast Error Variance Decomposition
VariablesPercentage of Inflation Forecast Error

Variance Explained by1
Ordering 1
Broad money5.0
Domestic credit2.0
Fixed investment13.0
Industrial production21.0
Inflation59.0
Ordering 2
Fixed investment12.0
Industrial production22.0
Inflation61.0
Broad money3.0
Domestic credit1.0
Ordering 3
Broad money8.0
Net domestic assets11. 0
Fixed investment11.0
Industrial production13.0
Inflation57.0
Ordering 4
Fixed investment8.0
Industrial production16.0
Inflation62.0
Broad money5.0
Net domestic assets10.0

Percentages refer to the decomposition of the forecast error two years ahead.

Percentages refer to the decomposition of the forecast error two years ahead.

References

    CalvoGuillermo A. and FabrizioCoricelliOutput Collapse in Eastern Europe: The Role of CreditStaff PapersInternational Monetary Fund Vol. 40 (March1993) pp. 3252.

    • Search Google Scholar
    • Export Citation

    EngleRobert F. andC.W.J.GrangerCointegration and Error Correction: Representation, Estimation, and TestingEconometrica Vol. 55 (March1987) pp. 25176.

    • Search Google Scholar
    • Export Citation
100Although longer time series are available for some of the variables, the length of the sample used here is constrained by that of the fixed investment series.
101The Akaike Information Criterion suggests the inclusion of a third lag of each variable. However, in the interest of conserving degrees of freedom, given the limited availability of data, the lag length chosen by minimizing the Schwarz-Bayesian criterion has been used.
102Changes in merchandise import value were also included in the system but did not alter the implied relationships and so were dropped from the model.
103One reason for the finding that domestic credit has little predictive power in the system could be the high degree of correlation between broad money and domestic credit. In this case, the inclusion of both variables could result in their joint effect being picked up by one, with little incremental predictive power being assigned to the other. For this reason, the VAR was re-estimated using either broad money or domestic credit but not both at the same time. However, the results were not noticeably different—domestic credit still had little explanatory power in the equations for the other variables in the system.
104To the extent that these alternative portfolio options result in a reduction in household and enterprise holdings of bank deposits and, therefore, in a reduction in broad money, it should be possible to observe a more direct link between fixed investment and broad money; however, as noted above, that link is not supported by the data.
105Domestic credit, in contrast, appears to have little explanatory power in most of the other equations in the system, with the exception of the inflation equation.

    Other Resources Citing This Publication