Journal of Comparative Economics29, 80–94 (2001)doi:10.1006/jcec.2000.1705, available online at http://www.idealibrary.com on

Real Exchange Rate Misalignment in China:An Empirical Investigation1

Zhichao Zhang

St. Antony’s College, Oxford University, Oxford OX2 6JF, England

Received May 28, 1998; revised November 27, 2000

Zhang, Zhichao—Real Exchange Rate Misalignment in China: An Empirical Investigation

Based on the theory of equilibrium real exchange rate, this paper estimates the behavioralequilibrium exchange rate and the resulting misalignment in China. Evidence shows chronicovervaluation in China’s central planning period, but economic reforms have brought thereal exchange rate closer to equilibrium. The cumulative effect of exchange rate reform ledto a substantial real depreciation of the Chinese currency after 1981. Indications are thatChina now has a proactive exchange rate policy with the nominal exchange rate used as apolicy tool to attain real targets.J. Comp. Econ., March 2001,29(1), pp. 80–94. St. Antony’sCollege, Oxford University, Oxford OX2 6JF, England.C© 2001 Academic Press

Journal of Economic LiteratureClassification Numbers: C13, C32, F31, P21, P33.

1. INTRODUCTION

Throughout the central planning period from the mid- 1950’s to the late 1970’s,exchange rate policy in China, as in other command economies, was subordinateto trade policy, which in turn was a residual in centralized materials balancing.The exchange rate was mainly a translator, serving as a mere accounting devicelinking foreign trade and the domestic economy (Mah, 1972). Economically, itwas relevant only for nontrade transactions, such as foreign tourism and remittancefrom overseas’ Chinese. Changes in the exchange rate could not affect directly theoverall balance of trade because the volumes of imports and exports were fixedby the trade plan. Although, in theory, changes in the exchange rate may havebudgetary consequences that could potentially produce real effects (Ames, 1953),the trade-off between fiscal balance and price stability prevented such changesfrom occurring (Zhang, 1997).

1 The author thanks the Wai Seng Fu Foundation for financial support and Gillian Rathbone forassistance. Thanks also go to John Bonin and two anonymous referees, whose detailed comments weremost instructive.

800147-5967/01 $35.00Copyright C© 2001 by Academic PressAll rights of reproduction in any form reserved.

EXCHANGE RATE MISALIGNMENT IN CHINA 81

Under this institutional setup, Chinese exchange rate policy was beset withthree basic problems: an unrealistic exchange rate that did not reflect conditionsof relative scarcity of foreign exchange; a rigid exchange rate regime that rejectedtotally market forces in exchange rate decisions; and a malfunctioning mechanismso that the exchange rate failed to signal and guide economic agents to efficientresource allocation (Chen, 1992). Against this background in the late 1970’s, Chinalaunched a reform of its exchange rate policy. The objectives were to rationalizethe level of the exchange rate, to make full use of the exchange rate as an economiclever, and to establish a managed, uniform floating rate system while graduallyrendering the Chinese currency convertible.

The first reform step took place in 1979 when the government decided to adopt,in parallel to the official exchange rate, an Internal Rate for Trade Settlements,operative from January 1, 1981. This marked the first official recognition thatthe exchange rate was overvalued. Once this internal rate was introduced, furtherefforts were directed to reforming the official exchange rate itself. In particular, in1984, China frequently adjusted the official rate. The reform devaluations resultedin the official rate being at par with the internal rate by the end of 1984, effectivelyrendering the latter redundant. From January to October 1985, the nominal officialexchange rate continued to be devalued repeatedly, but in mini steps, reflecting thefact that the reform focus had switched to the search for an appropriate institutionalarrangement for exchange rate changes. Major devaluations followed in July 1986,December 1989, November 1990, and January 1994.

Although the International Monetary Fund has classified China as a managedfloating exchange rate regime since 1987, China only officially admitted to sucha regime in 1991. In 1994, the government announced that it was adopting amanaged floating rate regime based on a uniform rate, coupled with a move topartial convertibility on current account. This policy announcement was related tothe emergence of the foreign exchange swap market (Lu and Zhang, 2000).

The basis for the swap market was foreign exchange retention schemes that al-lowed Chinese exporters to retain a portion of their export proceeds over which theyhad autonomy in spending. A market for the retained foreign exchange emergedfirst in 1981; from 1985 the momentum of development quickened. In this market,surplus units swapped their entitlement to foreign exchange with deficit units at aprice determined, more or less, by demand and supply. This price was known asthe swap exchange rate. With the swap market, the effective exchange rate thatChinese exporters received was the weighted average of the official and the swapexchange rates, with the weights being determined by the retention ratio. Whileat times the official exchange rate was pegged, the swap exchange rate was freedgradually. Its fluctuations implied that the effective exchange rate in China wasflexible, rendering inaccurate the notion that China had a fixed rate regime duringthe period.

The swap market enabled a considerable proportion of foreign exchange, inits later days about 80% of China’s trade transaction, to be allocated through the

82 ZHICHAO ZHANG

market. This undermined China’s rigid system of exchange controls and resulted ina relatively free flow of foreign exchange regulated by supply and demand forces.Moreover, the swap market facilitated the reform of the official exchange rate byproviding exporters with a more depreciated secondary rate and information aboutthe realistic level of the exchange rate. As a result, it improved the links betweendomestic and foreign prices and the effective exchange rate reflected domesticmarket conditions at the margin.

This process reflected one critical aspect of China’s reform, namely the rational-ization of the exchange rate to provide proper price signals. This reform assumedthat the exchange rate was not an equilibrium price, so that changes were neededto correct exchange rate misalignment. Of considerable interest are the extent towhich the Chinese exchange rate was misaligned in the prereform period and thequestion of whether the changes brought it toward equilibrium. Despite repeatedclaims and indeed widespread agreement that exchange rates in centrally plannedeconomies were overvalued (Desai and Bhagwati, 1979), empirical research onthe issue has been scant. To a large extent, this is due to the fact that estimatingovervaluation is challenging. This exercise requires determining the equilibriumreal exchange rate in the first place and then measuring the degree of deviationof the actual real exchange rate from this equilibrium rate. Only in recent yearshave methods for estimating the equilibrium real exchange rate been advancedadequately by the development of relevant statistical tools, particularly unit-rooteconometrics. This paper is an application of this methodology to China.

The present study aims to assess empirically the misalignment of China’s realexchange rate from the prereform era to the recent reform period. Section 2 surveysalternative approaches to measuring misalignment of the real exchange rate. InSection 3, a basic analytical model of the equilibrium real exchange rate followingMontiel (1999b) is presented. Econometric formulation and cointegration analysisare carried out in Section 4. The variable space in which the relevant economicfundamentals are included is based on the Montiel model. The Johansen approach isemployed to determine the number of cointegration vectors. A unique cointegratingrelationship is identified based on economic theory and the data. Using thesecointegration coefficients, Section 5 derives the equilibrium real exchange rate ina behavioral sense for China and determines the degree of misalignment during theperiod under investigation. The effect of the reform on misalignment is estimatedin this section. Section 6 contains conclusions.

2. APPROACHES TO MEASURING REAL EXCHANGERATE MISALIGNMENT

The real exchange rate can be defined as the relative price of tradable to non-tradable goods, or the Salter ratio (Corden, 1994),

RER= PT/PN, (1)

EXCHANGE RATE MISALIGNMENT IN CHINA 83

where RER is the real exchange rate andPT and PN are the domestic prices oftradable and nontradable goods. In Hinkle and Montiel (1999), this ratio is calledthe two-good internal real exchange rate for tradables and nontradables. Note thata decline in the real exchange rate denotes appreciation, while an increase in theRER indicates depreciation.

The equilibrium real exchange rate is defined traditionally as the rate that isconsistent with the simultaneous achievement of internal and external balance(Williamson, 1985). Then, misalignment is the difference between the actualand the equilibrium real exchange rate. For developing countries, three basic ap-proaches to RER misalignment are found in the empirical literature: a measurebased on purchasing power parity (PPP) (Balassa, 1990; Cottani et al., 1990;Chinn, 1998), a measure using the black market premium (Quirk et al., 1987), anda model-based approach (Edwards, 1988, 1994; Dollar, 1992; Elbadawi, 1994;Baffes et al., 1999).

The first approach uses deviations of the actual RER from some base year inwhich the RER is believed to be in equilibrium to calculate misalignment. Inade-quate consideration is given to changes in the equilibrium RER caused by funda-mentals because this approach assumes an unchanged equilibrium real exchangerate throughout the period. In the second approach, the black market exchange rateis used to measure misalignment. However, no empirical evidence establishes arobust correlation between the two. The informational content of the parallel rateis limited, since its deviation from the official rate could be positive or negativein response to a shock at various times along the adjustment path (Montiel andOstry, 1994). The third approach uses a formal model for determining the equi-librium RER. Its principal advantage is the capability of incorporating changesin the equilibrium real exchange rate. One model, the macroeconomic balanceapproach popularized by Williamson (1985), identifies internal and external equi-libria and calculates the underlying exchange rate either by applying a comparativestatic calculation or by running simulations using large macroeconomic models(Williamson, 1994). Inevitably, this approach faces complications resulting from aconsiderable amount of parameter estimation, time lags in comparative static mod-els, or the size of model specifications in simulation models. Clark and MacDonald(1998) provide an excellent review of this method with numerous references.

An alternative model, the Behavioral Equilibrium Exchange Rate (BEER), dif-fers fundamentally from other methods in the notion of equilibrium it adopts. In thisapproach, the relevant notion of equilibrium is not derived from macroeconomicbalance; rather it is determined by an appropriate set of explanatory variables. Theactual real exchange rate is said to be in equilibrium in a behavioral sense whenits movements reflect changes in the economic fundamentals that are found to berelated to the actual real exchange rate in a well-defined statistical manner (Clarkand MacDonald, 1998). Instead of calculating the level of the real exchange ratethat is consistent with internal and external balances, the BEER model is esti-mated using the fundamental determinants of the actual real exchange rate. The

84 ZHICHAO ZHANG

systematic relationship between the actual real exchange rate and its fundamentaldeterminants is estimated directly. Based on this relationship, the long-term equi-librium path for the real exchange rate, which is taken to be the equilibrium realexchange rate, is derived. This method is well suited to developing countries inwhich large and complex models are often not feasible because of data limitations.

Estimation of the BEER model is dependent on theoretical guidance for thechoice of an appropriate set of economic fundamentals. Following Edwards (1988,1994), there is now a large body of literature on identifying such a set of variables.In a recent study, Montiel (1999b) develops a model that synthesizes previousmodels of the equilibrium real exchange rate. According to Montiel, the long-runequilibrium real exchange rate is determined by the steady-state values of the pre-determined variables and permanent (sustainable) values of both policy variablesand exogenous variables. The set of variables that may act as the long-run de-terminants includes the four following components. First, domestic supply-sidefactors and particularly the Balassa–Samuelson effect arising from faster produc-tivity growth in the traded-goods sector than in the nontraded goods sector areconsidered. Second, fiscal policy, such as permanent changes in the compositionof government spending between traded and nontraded goods, is relevant. Third,changes in the international economic environment, including changes in an econ-omy’s external terms of trade, the flows of external transfer, the world inflationrate, and the level of the world real interest rates, are important. Fourth, a liber-alization of commercial policy, for example, a reduction in export subsidies, mayaffect the long-run real exchange rate.

Once the fundamentals are identified according to theory, the BEER approachestimates directly the systematic relationship between the actual real exchangerate and these long-run determinants. In this respect, the BEER model benefitsfrom the recent development of the cointegration technique that permits the equi-librium relationship between the real exchange rate and its determinants to beexamined directly. In essence, cointegration analysis tests for the existence of anequilibrium relationship among economic variables that involve a systematic co-movement among them in the long run. If cointegrated, one would expect the realexchange rate and its fundamental determinants to be related to each other in asystematic way. That is, the real exchange rate has a mean reversion property inthe long run, where the mean can be viewed as the equilibrium rate. Therefore,the cointegrating equation can capture a steady-state relationship between actualvalues of the real exchange rate and economic fundamentals. The estimated coin-tegrating equation can be used to approximate the equilibrium real exchange rateand to gauge misalignment. Based on cointegration analysis, the BEER model ispromising (Montiel, 1999a); we take this approach in estimating misalignment inChina.

Recently Chou and Shih (1998) have estimated the equilibrium exchange rateof the Chinese currency using the PPP approach and the shadow price of foreignexchange (SPFE) model. The current study differs from Chou and Shih (1998) in

EXCHANGE RATE MISALIGNMENT IN CHINA 85

that it follows the BEER approach, which is based on the standard theory of theequilibrium real exchange rate. The period of examination covered by this studyis also longer, as it spans the mid 1950’s to the recent past. Hence, we examine theevolution of China’s exchange rate policy from a new perspective.

3. MODEL ESTIMATION AND COINTEGRATION ANALYSIS

Following Baffes et al. (1999), Montiel’s theory can be expressed in a generalform,

ln e∗t = β ′Fpt , (2)

wheree∗ is the equilibrium real exchange rate,Fpt is a vector of permanent values

for economic fundamentals that are identified by theory, andβ is a vector ofcoefficients for the long-run fundamentals. To estimateβ, an empirical model,which is to be consistent with Eq. (2) but relates to observable variables since theequilibrium real exchange rate is not observable, is specified. This relationship canbe captured in a cointegrating equation of the form

lnet = β ′Fpt + ωt, (3)

whereωt is a stationary random variable with zero mean, indicating that the ac-tual real exchange rate and the economic fundamentals are cointegrated. If thiscointegrating equation holds, the real exchange rate will have a mean reversionproperty in the long run. The mean of this cointegrating relationship can be iden-tified econometrically as the equilibrium rate since it is the level toward which theactual real exchange rate gravitates over time. Our task is to determine whether thiscointegrating relationship exists by testing for the stationarity of a linear combina-tion of the variables of interest. Then, the cointegration parameters can be used asestimates of the parameter vectorβ in Eq. (2) and the equilibrium real exchangerate can be derived.

In order to carry out the cointegration analysis, we formulate the variable spaceas

(RER, FINVEST, GCON, EXPORTG, OPEN). (4)

In (4), RER is the real exchange rate of the Chinese currency, Ren-min-bi (RMB).The others are the determinants of the equilibrium RER, as suggested by Montiel’stheory. The variable FINVEST is investment represented by the index of gross fixedcapital formation, at fixed prices with 1952= 100, which determines the domesticsupply capacity and can also be viewed as a proxy for technological progress.The next variable, GCON, is the index of government consumption, at currentprices with 1950= 100, which captures the effect of fiscal policy. EXPORTG isthe growth rate of China’s exports, approximated by the one-lag difference of the

86 ZHICHAO ZHANG

log of exports. In the literature, the variable terms of trade is often used to representchanges in the international economic environment. In China, however, no consis-tent data are available for this variable, so we use instead the variable EXPORTGto reflect this effect. Finally, the variable OPEN is the degree of openness measuredas the ratio of the sum of imports plus exports to GDP in domestic currency andis included to capture the effect of commercial policy.

Although RER is defined as the relative price of tradable to nontradable goods,proxies have been used widely because of data limitations on prices of tradable andnontradable goods. A common proxy is constructed using the nominal exchangerate (E) multiplied by a fraction consisting of the foreign wholesale price index(WPI∗) in the numerator and the domestic consumer price index (CPI) in the de-nominator, i.e., RER= E(WPI∗/CPI). In the present study, the nominal exchangerate of RMB against the U.S. dollar is used forE. However, as mentioned in thefirst section, the effective exchange rate that Chinese exporters received after 1981was the weighted average of the official and the secondary exchange rates. To re-flect this fact, the nominal exchange rate that we use to compute the real exchangerate is a weighted average of the two rates with the swap rate having a weight of0.44, which was the average of the foreign exchange retention ratio from 1986 to1993. From 1981 to 1985 when swap transactions were insignificant, the InternalRate for Trade Settlements was used as the secondary rate. Its weight is 0.8 inthe calculation because, during this period, this rate was applicable to trade andrelated activities, which in turn accounted for 80% of China’s foreign exchangeincome. For WPI∗, the U.S. wholesale price index for finished industrial goodsis used. However, for China’s domestic price level, the retail price index is used,since no consistent CPI data are available during the sample period.2

In the econometric formulation of the system, an impulse dummy is constructedto account for the impact of unusual events because of the nonnormal residuals ofthe equation determining the RER. The problem is with the outlier observations ofthe RER in 1961 to 1962 and 1981. To address this problem, the dummy is set at−1 for 1961 and 1962 to account for the impact of the great famine that caused theChinese RER to appreciate through a sharp rise in domestic inflation and at+1 for1981 because of the sudden jump in the RER (depreciation) due to the introductionof the secondary exchange rate. The dummy is zero for all other years. While the

2 Most data sets are sourced from the International Monetary Fund’s International Finance Statistics(the CD-ROM version, 1999 May). Exceptions are the following. The data on the Chinese officialexchange rate against the dollar from 1953 to 1955 are taken from the “Manual of Exchange Rates”published by China’s State Administration of Foreign Exchange Controls (1986). All exchange ratesare average values for the period. The Chinese GDP before 1980 is taken from the Department ofNational Economic Accounting of the State Statistical Bureau (1997). Unlike national gross outputunder the material production system, this GDP series is a newly published data set based on theexpenditure approach and at current prices. The index of gross fixed capital formation is also from thissource, with the exception that figures for 1996 and 1997 are the author’s estimation. Data on China’sgovernment consumption, gross capital formation, imports, and exports from 1952 to 1977 are takenfrom various issues of the Statistical Yearbook of China; all series are at current prices.

EXCHANGE RATE MISALIGNMENT IN CHINA 87

TABLE 1Model Evaluation Diagnostics

Diagnostic tests (multivariate)

Vector AR 1-2 F(50, 85)= 0.98131 [0.5210]

Vector normality Chiˆ2(10)= 10.979 [0.3591]

Vector Xiˆ2 F(300, 7)= 0.05324 [1.0000]

Note.Figures in brackets areP values. F(50,80) means that the test has anF-distributionwith 50 degrees of freedom in the numeratorand 80 degrees of freedom in the denominator.Chiˆ2(10) refers to theχ2 test with 10 degreesof freedom.

impulse dummy is entered unrestrictedly, a constant term is restricted to lying inthe cointegration space.

The empirical investigation starts with a general system for the sample periodfrom 1952 to 1997. Since the data are annual, given the number of variables and thesample size, a lag length of 2 was selected. However, if the lag length is increasedto 4, there is no essential change in empirical results, so the two-lag system issufficient. Table 1 presents the diagnostic test results for this model. In the tableand all other estimation reports, variables in lowercase letters are in log form.

Test results for individual equations indicate no serious problems, so these are notpresented. For the system as a whole, the diagnostic tests involve tests for the fol-lowing hypotheses. First, there is no serial correlation, labeled Vector AR in Table 1,which tests for second-order residual autoregression. Second, the system residualsare distributed normally, Vector normality. Third, there is no heteroscedasticity,Vector Xiˆ2. None of the system test results rejects the null hypothesis, so thisdynamic system with one lag is acceptable.

Next, cointegration is investigated using the Johansen approach and the degreeof integration of the system is determined in the process. The Johansen methodis an essential tool for estimating cointegration in a multivariate system and testsfor the number of cointegrating relationships in a VAR system. Table 2 reports thecointegration statistics using the Johansen technique.

The two statistics of particular interest are the maximum eigenvalue statistic,λmax, and the trace statistic,λtrace, in Table 2.3 Both test statistics reject the null ofno cointegration in favor of at least one cointegrating vector in the system. Further,according to the trace statistic, the cointegration rank could be more than 1, as thenull hypothesisρ ≤ 1 can be rejected. Although the test may suggest otherwisewhen the trace statistic is adjusted for degrees of freedom, the outcome is very close

3 The Max statistic tests that there areρ cointegration vectors against the alternative thatρ+ 1 exists,i.e., h0: rank= ρ, h1: rank= ρ + 1, while the Trace statistic tests the null hypothesis that there are atmostρ cointegration vectors against more thanρ cointegration vectors, i.e.,h0: rank= ρ, h1: rank> ρ.

88 ZHICHAO ZHANG

TABLE 2Cointegration Statistics

Ho: rank= ρ λmax 95% Ssa λtrace 95% Ssa

ρ= 0 43.84∗∗ 34.4 33.88 108.3∗∗ 76.1 83.72∗ρ ≤ 1 27.1 28.1 20.94 64.5∗∗ 53.1 49.84ρ ≤ 2 17.78 22.0 13.74 37.4∗ 34.9 28.9ρ ≤ 3 13.87 15.7 10.72 19.62 20.0 15.16ρ ≤ 4 5.751 9.2 4.444 5.751 9.2 4.444

Note.The first column is the null hypothesis to test the number of cointegratingrelations. The next column is the maximum eigenvalue statistic,λmax, that testswhetherρ = 0 againstρ = 1, orρ = 1 againstρ= 2, etc. The column with theheading 95% gives the 95% critical values taken from Osterwald-Lenum (1992).∗∗ denotes rejection of the null hypothesis at the 1% significance level;∗ denotesrejection at the 5% significance level. The column “Ssa” gives the small-sample-adjusted values ofλmax or λtracefrom Reimers (1992).λtraceis the trace statisticthat tests the null thatρ = r (r = 1, 2, . . . ,n − 1) against the alternative thatρ > r .

to not rejecting the null. Furthermore, there is some evidence that the cointegratingrank could even be more than 2, because the trace statistic also rejects the null ofρ ≤ 2, as shown in the fourth row. However, this evidence is weak, as it rejects thenull only at the 5% significance level and the small-sample-adjusted value ofλtrace

is smaller than the critical value. Based on these tests, we take the cointegrationrank to be 2. Since the cointegration rank is not full, the system cannot be I (0).On the other hand, no value of the modules of the eigenvalues of the companionmatrix of the dynamics (not shown) is 1 or greater than 1, suggesting that novariable is I (2). This leads to the conclusion that the variables in the system are I(1) processes.4

To determine uniquely the cointegration vectors, restrictions were imposed onthe identified cointegrating relationships based on economic arguments and testresults. Given that there are two cointegrating vectors, restrictions are placed onα, which is the loading factor or the adjustment speed, andβ, which is the vectorof the cointegration coefficients. The log likelihood ratio test, labeled LR-test inTable 3, reveals that the restrictions cannot be rejected. Table 3 presents the testresults and the cointegration estimates.

Of the two cointegrating vectors, the first row of theβ vectors in Table 3 isrecognizable as the real exchange rate relation, the parameters of which are ourprincipal interests. Next we consider the weak exogeneity of the variables. If the

4 I thank one of my referees for advising me to test unit roots for the possibility of a structural break.Pretests have been conducted on individual variables to test for unit roots with structural breaks usingthe recursive approach suggested by Banerjee et al. (1992) and the critical values therein. However,the variables turn out to be I (1) processes. Currently, the property and critical values for testing forunit roots with structural breaks in a system context are not available.

EXCHANGE RATE MISALIGNMENT IN CHINA 89

TABLE 3Restricted Cointegration Analysis

β ′rer finvest gcon exportg open Constant1.0000 −0.37400 0.32254 3.3892 −0.87818 −2.6683−1.5301 0.00000 0.00000 −4.5529 1.0000 3.7490

Standard errors ofβ ′rer finvest gcon exportg open Constant0.00000 0.11469 0.13534 0.57300 0.13932 0.711620.14851 0.00000 0.00000 0.98952 0.00000 0.16066

α

rer finvest gcon exportg open−0.15232 0.00000 −0.36768 0.00000 0.000000.00000 0.00000 −0.24438 0.00000 −0.10463

Standard errors ofαrer finvest gcon exportg open0.037125 0.00000 0.063579 0.00000 0.000000.00000 0.00000 0.044994 0.00000 0.032400

LR-test, rank= 2: Chiˆ2(6)= 6.9642[0.3242]

Note.β ′s are coefficients of the cointegration vectors andα′s are coefficients of the adjustmentspeed or the loading factors. For the log likelihood ratio test, i.e., LR-test, Chiˆ2(6) refers to theχ2 test with 6 degrees of freedom and the figure in brackets is theP value.

variables are weakly exogenous to the system, there will be no loss of informationfrom not modeling the short-run behavior of these variables, and so estimationof the multivariate model can be conditioned on such variables. Weak exogeneitycan be determined by examining whether every estimate in the entire row ofα

estimate is zero; see Hendry (1995) and Harris (1995) for the testing procedurein such a context. In our case, where the cointegration rank is 2, if one of thetwo rows of theα estimates is zero and therefore does not need to be modeled interms of short-run dynamics, the cointegration analysis can be simplified by usingthe single-equation Engle–Granger procedure. However, as Table 3 shows, this isnot the case for the real exchange rate in China. Hence, we retain the Johansentechnique, which is a multivariate system estimator.

4. MISALIGNMENT AND EFFECT OF THE REFORM

Based on Table 3, the implied behavioral equilibrium real exchange rate can bepresented as in Eq. (5) with standard errors in parentheses:

BEER

= 2.6683+ 0.374finvest− 0.32254gcon− 3.3892exportg+ 0.87818open

(0.71162) (0.11469) (0.13534) (0.573) (0.075521).

(5)

90 ZHICHAO ZHANG

In Eq. (5), BEER is the behavioral equilibrium exchange rate in log form. Thepositive sign of the variable finvest suggests that a rise in investment is likely todepreciate the real exchange rate, as this tends to shift spending toward tradedgoods, a result consistent with Baffes et al. (1999). On the other hand, gcon hasa negative sign, implying that an increase in government spending causes the realexchange rate to appreciate. This is likely in China because government consump-tion has a low import content. The result for export growth, the variable exportg,indicates a negative elasticity so that an increase in the growth rate of exports ap-preciates the real exchange rate, which is consistent with other empirical research(Elbadawi and Soto, 1997). Under China’s institutional setup, an acceleration ofexports may increase import capacity, which in turn leads to a domestic expan-sion and hence a real appreciation. The openness variable displays a positive sign,implying that trade restrictions will cause the real exchange rate to appreciate.This confirms the finding in the literature that economic closedness is associatedwith overvaluation. Using the derived equilibrium real exchange rate in Eq. (5),movements of the actual and the equilibrium real exchange rate in China from1954 to 1997 are graphed in Fig. 1.

Figure 1 displays an interesting pattern of misalignment in China. The actualreal exchange rate was overvalued during much of the prereform period from 1957

FIG. 1. Misalignment of actual real exchange rate from equilbrium.

EXCHANGE RATE MISALIGNMENT IN CHINA 91

to 1977. Only briefly, between 1971 and 1973, was the exchange rate undervaluedand this resulted from the world’s general adoption of floating exchange ratesand China adjusting its nominal exchange rate correspondingly. However, after1978, the pattern changed as economic reform was launched in China. The actualRER now fluctuated closely around the BEER. Periods of overvaluation occurredbriefly in 1982 to 1983, 1985, and 1996; however, these were short-lived and themagnitude was generally smaller than in the prereform period. In fact, the Chinesecurrency was undervalued in 12 to 20 years from 1978 to 1997 and in 4 of the otheryears the real exchange rate was very close to equilibrium. Regarding exchangerate reform, Fig. 1 displays a pronounced jump or devaluation in the actual realexchange rate in 1981 when the secondary exchange rate was introduced. After thattime, the currency moved closer in line with the equilibrium rate in a period thatincluded several nominal devaluations and the development of the swap exchangemarket as a reform experiment.

To examine the extent to which exchange rate reform reduced misalignment inChina, we consider a further exercise. A formal test for the effect of the reform isattempted using the structural time series model of Harvey (1989). A model forthe real exchange rate is formulated in which a reform variable is included as theexplanatory variable. This reform variable, labeled reform 81 in Table 4, is a stepdummy taking the value of unity from 1981 onward and zero otherwise. A smoothtrend model, which specifies a fixed level and a stochastic slope, is estimated.5 Theestimation procedure converged very strongly, indicating that the specification ofthe model is good. Table 4 reports the estimated coefficients and postinterventiondiagnostics.

Table 4 indicates that the coefficient on the reform variable is 0.557; the com-pound rate of growth over the sample period is therefore exp(0.557)− 1, whichequals 75%. In other words, the cumulative effect of the reform is a 75% rise ordepreciation in the real exchange rate in China. This indicates that the exchangerate reform has brought the Chinese real exchange rate down substantively from itsovervalued level. To evaluate the model, we subject it to the postintervention tests.The number of periods chosen for the postintervention predictive test is 14, so thatwe can look at predictions from 1982 onward after the exchange rate reform in1981. The Chow test shows that it is not significant at the 1% level. Next, to examinethe predictive power for the 1990’s, 7 within-sample periods are chosen. The Chowtest statistic is again not significant. Hence, the model is well behaved and accept-able, so that our conclusion about the effect of the exchange rate reform is validated.

5. SUMMARY

Based on the economic theory of equilibrium real exchange rate, this paperestimates the long-run equilibrium path for the real exchange rate in China using the

5 Additionally, the model includes a slope intervention for 1962, when China was hit by famine, andan irregular dummy for 1993, when there was an impulse outlier in the variable rer.

92 ZHICHAO ZHANG

TABLE 4Estimation of Reform Effects in a Structural Series Model

Estimated coefficients of explanatory variables

Variable Coefficient R.m.s.e. t valuereform81 0.557473 0.0656875 8.4867 [0.0000]∗∗Slp1962 0.0963044 0.0367823 2.6182 [0.0122]∗Irr1993 0.124328 0.0448524 2.7719 [0.0083]∗∗

Predictive tests (14 periods inside sample) for rer

Chow F(13, 28) test is 2.72112 [0.0129]∗Cusumt(28) test is −0.7228962 [1.5243]

Predictive tests (7 periods inside sample) for rer

Chow F(6, 35) test is 1.78873 [0.1300]Customt(35) test is −1.14247 [1.7390]

Note.R.m.s.e refers to root mean square error. The variable Slpl 1962 isa dummy variable for the effect of the great famine in Chine in 1962. It isconstructed to capture the structural break in the slope and is modeled by astaircase intervention that takes the values, 1, 2, 3, etc., starting in the periodafter the break. The variable Irr 1993 is an impulse intervention dummyvariable to capture the irregular disturbance in 1993. ChowF(·) refers tothe Chow test inF form for predictive testing testing and Cusumt(·) is a ttest for the cumulatice sum of the standardized prediction errors. The latteris designed to assess whether the mean of a processyt changes over time.Values in brackets are the two-sided probability values.

BEER approach. In this approach, the systematic relationship between the actualreal exchange rate and economic fundamentals is used as the basic equilibriumconcept. In a behavioral sense, the real exchange rate is in equilibrium whenits movements reflect the economic fundamentals to which it is related in a well-defined statistic manner. Cointegration analysis indicates that domestic investment,export growth, and the trade regime are long-run determinants of the equilibriumreal exchange rate in China.

Misalignment in China is estimated by using the unique cointegrating vector toderive the behavioral equilibrium real exchange rate. Chronic overvaluation oc-curred in China during the central planning period, which provides empirical evi-dence for claims in the literature about currency overvaluation in centrally plannedeconomies. However, economic reforms have changed the pattern of China’s realexchange rate and brought it closer to equilibrium. In the reform years, the rever-sion of the real exchange rate to equilibrium is faster. Overvaluation is short-livedand its magnitude is smaller. The structural time series analysis shows that thecumulative effect of exchange rate reform has led to a substantial real depreciationof the Chinese currency since 1981 when the reform was introduced. Furthermore,

EXCHANGE RATE MISALIGNMENT IN CHINA 93

undervaluation occurred in 12 of 20 years from 1978 to 1997. These results indi-cate that China now has a proactive exchange rate policy that employs the nominalexchange rate as a policy tool, varied either frequently or occasionally, to attaineither targets in the real sector or a real exchange rate target. Given China’s grow-ing economic significance, this exchange rate policy and its consequent tendencyto undervaluation of the real exchange rate invite further research.

REFERENCES

Ames, Edward, “The Exchange Rate in Soviet-Type Economies.”Rev. Econ. Statist.36, 4:337–342,Nov. 1953.

Baffes, John, Elbadawi, Ibrahim A., and O’Connell, Stephen A., “Single-Equation Estimation of theEquilibrium Real Exchange Rate.” In Lawrence E. Hinkle and Peter Montiel, J., Eds.,ExchangeRate Misalignment: Concepts and Measurement for Developing Countries.pp. 405–464. NewYork: The World Bank, 1999.

Balassa, Bela, “Incentive Policies and Export Performance in Sub-Saharan Africa.”World Devel.18,3:383–391, Mar. 1990.

Banerjee, Anindya, Lumsdaine, Robin L., and Stock, James H., “Recursive and Sequential Tests ofthe Unit-Root and Trend-Break Hypotheses: Theory and International Evidence.”J. Bus. Econ.Statist.10,3:271–287, July 1992.

Chen, Biaoru,Renmingbi Hui Lu Yan Jiu (Studies of the RMB Exchange Rate).Shanghai: HuadongShida Chu Ban She (East China Normal Univ. Press), 1992.

Chinn, Menzie D., “Before the Fall: Were East Asian Currencies Overvalued?” National Bureau ofEconomic Research Working Paper No. 6491, Cambridge, MA: NBER, 1998.

Chou, Win Lin, and Shih, Yi Chung E., “The Equilibrium Exchange Rate of the Chinese Renminbi.”J. Comp. Econ.26,1:165–174, Mar. 1998.

Clark, Peter B., and MacDonald, Ronald, “Exchange Rates and Economic Fundamentals: A Method-ological Comparison of BEERs and FEERs.” International Monetary Fund Working PaperWP/98/67. Washington, DC: IMF, 1998.

Corden, Warner Max,Economic Policy, Exchange Rates, and the International System.Oxford: OxfordUniv. Press, 1994.

Cottani, Joaquin A., Cavallo, Domingo F., and Khan, M. Shahbaz, “Real Exchange Rate Behavior andEconomic Performance in LDCs.”Econ. Develop. Cult. Change39,1:61–76, Oct. 1990.

Department of National Economic Accounting, the State Statistical Bureau of China,Zhong Guo GuoNei Sheng Chan Zhong Zhi He Suan Zi Liao 1952–1995 (The Gross Domestic Product of China1952–1995).Sheng Yang: Dongbei Caijing Daxue Chu Ban She (North-East Univ. of Financeand Economics Press), 1997.

Desai, Padma, and Bhagwati, Jagdish N., “Three Alternative Concepts of Foreign Exchange Difficultiesin Centrally Planned Economies.”Oxford Econ. Pap.31,3:358–368, Nov. 1979.

Dollar, David, “Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidencefrom 95 Ldcs, 1976–85.”Econ. Develop. Cult. Change40,3:523–544, Apr. 1992.

Edwards, Sebastian,Exchange Rate Misalignment in Developing Countries.Washington, DC: TheWorld Bank, 1988.

Edwards, Sebastian, “Real and Monetary Determinants of Real Exchange Rate Behavior: Theory andEvidence from Developing Countries.” In John Williamson, Ed.,Estimating Equilibrium ExchangeRates.Washington, DC: Institute for International Economics, 1994.

Elbadawi, Ibrahim A., “Estimating Long-Run Equilibrium Real Exchange Rate.” In John Williamson,Ed., Estimating Equilibrium Exchange Rates, Washington, DC: Institute for International Eco-nomics, 1994.

94 ZHICHAO ZHANG

Elbadawi, Ibrahim A., and Soto, Raimundo, “Capital Flows and Long-Term Equilibrium Real ExchangeRates in Chile.”Revista Anal. Econ.12,1:35–62, June 1997.

Harris, Richard,Using Cointegration Analysis in Econometric Modelling.London: Prentice Hall andHarvester Wheatsheaf, 1995.

Harvey, Andrew C.,Forecasting, Structural Time Series Models and the Kalman Filter.Cambridge:Cambridge Univ. Press, 1989.

Hendry, David F.,Dynamic Econometrics.Oxford: Oxford Univ. Press, 1995.Hinkle, Lawrence E., and Montiel, Peter J., Ed.,Exchange Rate Misalignment: Concepts and Mea-

surement for Developing Countries.A World Bank Publication. Oxford: Oxford Univ. Press,1999.

Lu, Maozhu, and Zhang, Zhichao, “Parallel Exchange Market as a Transition Mechanism for ForeignExchange Reform: China’s Experiment.”Appl. Finan. Econ.10,2:123–135, Apr. 2000.

Mah, Feng-hwa,The Foreign Trade of Mainland China.Edinburgh: Edinburgh Univ. Press, 1972.Montiel, Peter J., “The Long-Run Equilibrium Real Exchange Rate: Conceptual Issues and Empirical

Research.” In Lawrence Hinkle and Peter J. Montiel, Eds.,Exchange Rate Misalignment: Conceptsand Measurement for Developing Countries. pp. 219–263, A World Bank Research Publication,Oxford: Oxford Univ. Press, 1999a.

Montiel, Peter J., “Determinants of the Long-Run Equilibrium Real Exchange Rate: An AnalyticalModel.” In Lawrence Hinkle and Peter J. Montiel, Eds.,Exchange Rate Misalignment: Conceptsand Measurement for Developing Countries, pp. 264–290, A World Bank Research Publication,Oxford: Oxford Univ. Press, 1999b.

Montiel, Peter J., and Ostry, Jonathan D., “The Parallel Market Premium: Is It a Reliable Indicatorof Real Exchange Rate Misalignment in Developing Countries?”Int. Monet. Fund Staff Pap.41,1:55–75, Mar. 1994.

Osterwald-Lenum, Michael, “A Note with Quantiles of the Asymptotic Distribution of the MaximumLikelihood Cointegration Rank Test Statistics.”Oxford Bull. Econ. Statist.54, 3:461–472, Aug.1992.

Quirk, Peter, Christensen, Benedicte Vibe, Huh, Kyung-Mo, and Sasaaki, Toshihiko,Floating ExchangeRates in Developing Countries: Experience with Auction and InterBank Market.InternationalMonetary Fund Occasional Paper 53, Washington, DC: IMF, 1987.

Reimers, Hans-Eggerts, “Comparisons of Tests for Multivariate Cointegration.”Statistical Papers33,4:335–359, Dec. 1992.

Williamson, John,The Exchange Rate System.2nd ed., Washington, DC: Institute for InternationalEconomics, 1985.

Williamson, John, Ed.,Estimating Equilibrium Exchange Rates.Washington, DC: Institute for Inter-national Economics, 1994.

Zhang, Zhichao, “Exchange Rate Reform in China 1978–1994,” D.Phil. thesis, Oxford: Oxford Univ.,1997.