Information about Asia and the Pacific Asia y el Pacífico
Chapter

III. Money Demand and Financial Liberalization

Author(s):
Robert Corker, and Wanda Tseng
Published Date:
March 1991
Share
  • ShareShare
Information about Asia and the Pacific Asia y el Pacífico
Show Summary Details

This section investigates the implications of financial liberalization for money demand in the Asian countries, focusing on the stability and predictability of broad and narrow monetary aggregates during the 1980s. The possible effects of financial liberalization on money demand are discussed, followed by the theoretical and empirical issues that arise in estimating money demand functions. Then the empirical results are presented and analyzed and conclusions are drawn together.

Effects of Financial Liberalization on Money Demand

The existence of a stable and predictable relationship between monetary aggregates, economic activity, prices, and interest rates is a crucial element in the formulation of monetary policy. Financial liberalization that improves the quality of economic signals, alters the institutional environment, and expands the array of financial opportunities creates potential for instability in money demand.

For the purpose of analyzing the possible effects on money demand, it is useful to group financial liberalization into three broad categories. First, interest rates that better reflect the economic return and riskiness of financial assets could prompt portfolio shifts. The nature of such onetime shifts in the demand for money would depend on the monetary aggregate in question and on which interest rates were liberalized: if, for example, interest rates on time deposits increased after liberalization, the demand for broad money might rise at any given level of income, but the demand for narrow money might decline.

In principle, the portfolio shifts could be predicted by money demand functions containing appropriate interest rate terms. But, in practice, interest rates might not have varied sufficiently prior to liberalization to allow for measurement of their effects on portfolio choices. As interest rates became more flexible, previously latent roles for interest rates in determining money demand could, therefore, become transparent.

Second, shifts to indirect monetary policy instruments might alter the observed relationship between monetary aggregates, economic activity, prices, and interest rates. In particular, the effective rationing of credit under direct credit controls could be alleviated following a shift to indirect monetary policy instruments making money more demand—as opposed to supply—determined. Because money demand relationships estimated prior to the policy change might not have properly identified money demand behavior, they could prove to be poor predictors of monetary developments after the policy change.

Third, measures to improve the functioning and depth of financial markets could prompt both portfolio shifts and alter the sensitivity of money demand to changes in incomes and interest rates. For example, measures to promote competition among financial institutions could lower transactions costs in financial markets and cause money demand to respond more rapidly to interest rate changes than before. Alternatively, changes in the regulation of financial institutions could prompt a reassessment of the relative risk of different financial assets, leading to both discrete portfolio adjustments and a change in the interest elasticity of money demand.

More generally, measures that promote financial market development could result in the availability of new, attractive assets leading to gradual portfolio shifts away from monetary assets. These shifts might occur independently of developments in income and interest rates. In addition, the returns on the new assets might become important determinants of money demand. Such assets could include foreign assets, if external capital flows were liberalized, as well as domestic assets such as stocks and bonds. However, to the extent that the policy of financial market development encouraged reintermediation, that is, attracted domestic savings away from unofficial “curb” markets, the direction of the effects on broad monetary aggregates would be less clear cut: as the size of curb markets diminished, the demand for all financial assets, including money, might increase.

In summary, financial liberalization could lead to onetime or more gradual shifts in the level of money holdings, as well as to changes in the measured income and interest elasticity of money demand. In some cases, financial liberalization may help to make more transparent the relationship between money and interest rates that may not have been detected previously.

As a practical challenge, it is important to distinguish between the various effects of financial liberalization because they have different implications for understanding the evolution of monetary aggregates. For instance, onetime changes in money holdings might be associated with temporary instability in money demand relationships during the period of liberalization. But changes in the income and interest rate elasticity of money demand, as well as new influences on money demand, might take longer to detect, rendering money demand relationships unstable over a longer period.

As discussed in Section II, virtually all Asian countries undertook financial liberalization during the 1980s, although the pace and scope of reforms varied substantially. For most of the countries, financial liberalization proceeded gradually and on many fronts: it may, therefore, be difficult to detect significant, discrete shifts in money demand. Only in a few countries, notably Indonesia, was liberalization concentrated in one or two major reform initiatives around which more clear-cut breaks in behavior might be expected.

Regardless of differences in financial liberalization, monetary developments during the 1980s in these countries shared some similar features. In particular, the ratio of broad money (defined here as currency plus demand and time deposits in the banking system) to income tended to increase, reflecting rapid growth in quasi-money balances: the share of quasi-money in broad money increased in all countries (Chart 2).8 By contrast, the ratio of narrow money (currency plus demand deposits in the banking system) to income, although generally more volatile than that of broad money to income, did not show a consistent trend across countries: in some countries it exhibited a downward trend, while in others it showed little trend or even tended to rise. However, it is not clear from Chart 2 whether the monetary trends of the 1980s differed significantly from those of the 1970s. And even if such breaks were apparent, it would be important to separate the effects on monetary developments resulting from financial liberalization from those arising from the changed economic environment of the 1980s. To do this, a more formal analytical framework, described below, is required.

Specification and Estimation Issues

Following is an outline of an empirical model that will be used to investigate the stability and predictability of money demand. In this regard, there are a number of interrelated issues concerning the specification and estimation of the model that have a direct bearing on how instability might be detected and analyzed.

Specification Issues

Theories of money demand attempt to explain one or more of the characteristics of money, stressing, in particular, the role of money in fulfilling transactions needs and providing a store of value.9 However, regardless of their theoretical background, a wide variety of models of money demand converge ultimately to a specification in which money depends on a scale variable (income or wealth), prices, and interest rates. This relationship can be written as

where, M is money; P the general price level; Y aggregate incomes; Ra the interest rate on alternative assets: and Ro the return on money.10 Variables other than interest rates are expressed in logarithms. A priori, the coefficients a1, a2 and a4 are expected to be positive, while the coefficient a3 is expected to be negative. For a narrow monetary aggregate, the “own” return would be close to zero and the coefficient a4 probably irrelevant; this would not be the case for a broader monetary aggregate that included a sizable interest-bearing component. In a purely portfolio model of asset demands, the coefficient a3 would be equal in size, but opposite in sign, to the coefficient a4 so that money demand only depended on relative asset returns. The coefficient a1 in most theoretical models would be expected to be close to one: indeed, this condition is usually imposed as, for example, in the earlier study of monetary policy in Asian countries (see Aghevli and others (1979)).

Whereas equation (1) is a fairly general theoretical specification of a money demand function, it assumes that money holdings adjust instantaneously to desired levels following a change in prices, incomes, or interest rates. In practice, this would not be the case for two main reasons. First, the adjustment of financial portfolios is not costless and individuals and corporations will lend to strike a balance between the cost of deviating from desired holdings and the cost of adjusting the level of their actual money holdings. Second, most theoretical models suggest that money demand depends on expected prices, incomes, and interest rates that need not always be related to actual levels of these variables. Allowing for some slow adjustment of expectations in response to new information would again introduce some sluggishness into the adjustment of actual money holdings to desired money holdings. In principle, it would be best to separate out the two sources of adjustment lags, but in practice it is difficult.11

Empirical attempts to capture the sluggish adjustment of money demand toward desired equilibrium holdings have often employed the assumption of partial adjustment in which a fixed proportion of the difference between desired and actual holdings diminishes each period.12 Recently, however, this approach has been criticized as overly restrictive because it assumes that adjustment costs and expectations can be captured in a very specific, simple fashion. In this analysis, an error correction dynamic specification is used. The error correction specification—to be discussed more fully below—can be thought of as a more general, intertemporal version of partial adjustment in which expectations are based on available information (see Nickel (1985)). The error correction money demand function can be written in the form:

where the symbol ∆ represents a first difference of a variable and the subscripts attached to the variables represent lagged values. M¯ stands for the fitted values from equation (1).

The distributed lags of first-difference terms are a mathematical representation of a general, unrestricted, short-run relationship between the variables. This generality acknowledges a degree of agnosticism about the true structure of adjustment costs and expectations mechanisms. The term with the coefficient b6 represents deviations from long-run equilibrium and gives the equation its error correcting properties: if money demand is below that consistent with desired long-run holdings, money demand increases. The static long-run equilibrium properties of the money demand function represented by equation (2) are identical to equation (1).13

Estimation issues

Recent econometric theory and practice have focused more attention on the time series properties of the data typically used in regression analysis. A large number of time series used in economic analysis are nonstationary (i.e., they have a persistent tendency to increase or decrease over time), which implies a number of restrictions on their use in regression analysis. In particular, regression of one nonstationary series on another can give rise to the so-called spurious regression problem and lead to incorrect statistical inferences.14 The nonstationary problem can be removed by first differencing the data, but in doing so, potentially interesting information about long-run equilibrium relationships between economic variables is lost.

Work by Granger and others has, however, found a way to advantageously exploit the spurious regression problem.15 The basic result of their analysis centers on the notion of cointegration: linear combinations of two or more nonstationary time series can, in some circumstances, turn out to be stationary.16 If this is the case, the linear combination can be legitimately included in a regression with other stationary variables (such as first-differenced nonstationary variables) with the twist that the combination variable—or cointegrated series—contains the useful long-run information that would otherwise have been lost through first differencing of all variables in the equation.

More powerfully still, the existence of cointegrating relationships turns out to be synonymous with the existence of a long-run equilibrium relationship between the economic variables.17 The proof of this result also establishes a correspondence between cointegration and the error correction model, which has developed in tandem with the new results on the properties of time series: if a cointegrating relationship exists, then it can be legitimately used in an error correction model. In the context of the analysis here, if the linear combination of money, prices, income, and interest rates in equation (1) formed a cointegrating relationship, then it would also represent the long-run relationship between these variables. Furthermore, the residuals formed by subtracting actual money balances from the fitted values of the long-run relationship described by equation (1) would be stationary and would form a legitimate error correction term to include in equation (2).

A number of methods have been proposed to estimate error correction models which, while producing consistent results in the statistical sense, generate quantitatively different estimates for key parameters in finite samples.18 The method adopted here follows that of Engle and Granger (1987) and, after preliminary analysis to establish the order of integration of the data series, employs a two-stage estimation process. In the first stage, equation (1) is estimated directly by ordinary least squares (OLS) regression and without regard for the short-run dynamics of money demand. The residuals of this regression are then tested for stationarity to determine whether the regression equation might represent a cointegrating relationship between money, prices, income, and interest rates. Engle and Granger show that, if this is the case, the OLS estimates of the coefficients are consistent and converge rapidly to the true long-run behavioral parameters. If stationarity is rejected, the data do not support the existence of a stable long-run relationship between the variables.

In the second stage, an error correction model (equation (2)) is estimated for those monetary aggregates for which cointegrating relationships can be found. Again, Engle and Granger show that OLS provides consistent estimates of the parameters in this equation, even where all regressors are not strictly exogenous. Equation (2) is specified first as a general distributed lag of first differences in the dependent and independent variables plus the lagged residuals from equation (1)—the error correction term. This general specification is reduced to a more compact form by eliminating insignificant variables.19

Testing for Stability

While economic theory supports the notion that money demand is likely to be a stable function of income, prices, and interest rates, this stability is predicated on an unchanging institutional environment.20 Indeed, changes in the institutional environment are widely believed to be responsible for observed long-run cycles in the income velocity of money (see, for example. Bordo and Jonung (1987)). Thus, money velocity is typically found to trend downward—money growth exceeds income growth because of monetization of the economy—in countries that are at an early stage of their economic development and that have relatively immature financial markets. As the degree of sophistication of the economy increases, the velocity trend is typically reversed because financial market innovations permit agents to economize on their holdings of money.

The likelihood that velocity exhibits long-run cycles suggests that, over typical data sample sizes (say, a couple of decades), one may estimate a money demand function that is dominated by a medium-term phenomenon such as monetization This possibility places some practical limits on the notion of stability: it may not be possible to identify a money demand function that is stable for all time but it may be possible to estimate a relationship that can explain monetary developments during the sample period with reasonable predictability In a specific example, one might estimate the long run income elasticity of money demand (coefficient a2 in equation (1)) to have been fairly constant over the sample period, but, perhaps reflecting monetization, greater than one. The implication that the share of monetary assets in income would increase without limit in the long run is, however implausible. Nevertheless, reasonable stability of the income elasticity and other key parameters in the historical data period would provide some grounds for concluding that monetary developments might remain predictable in the future, at least over the duration of the policy horizon (say, a few years).

Financial liberalization is one of the major vehicles through which the institutional environment is changed and is thus likely to affect the stability of money demand relationships both in the long and short runs. Recalling the discussion in the section on the “Effects of Financial Liberalization on Money Demand” above, liberalization might cause a onetime level shift in money demand (instability in the intercept term. a0 of equation (1)); alter long-run elasticities (instability in the coefficients a1, a2, a3, and a4 of equation (1)); or lead to short-run unpredictability of money demand (instability in the “b” coefficient in equation (2)). In addition, new influences may become important after liberalization (one of the “a” coefficients in equation (1) may acquire, or lose, statistical significance) or the precision by which monetary developments may be predicted might change (the “a” and “b” coefficient estimates might be unchanged, but the equations’ residual variance might increase).

The two-stage estimation procedure used here provides a systematic framework for investigating the stability of money demand functions. At the first stage, the tests for cointegralion between money, income, and interest rates provide information about the existence of stability in long-run money demand relationships, A rejection of cointegration, for example, would indicate that money, income, and interest rates have not been related in a particularly stable, predictable fashion over the course of the data sample, perhaps because of significant changes in the institutional environment. The tests do, however, require a large number of observations to produce convincing results: where degrees of freedom are limited, the tests are, at best, indicators of medium-term stability. Furthermore, the specific tests (described in Table 1) have limitations in certain statistical circumstances. In particular, the tests would have a bias toward incorrectly rejecting cointegration if the residuals in equation (1) were stationary but had a strong autoregressive component, suggesting that a degree of judgment is required when test statistic values are close to critical values.21

Table 1.Asian Countries: Long-Run Relationships Between Money, Income, and Interest Rates
Income ElasticityInterst Rate Semielasticity1Tests for Cointegration2Cointegration Status
CRDWDFADF
Indonesia
Real narrow money1.16−0.660.563.883.72Yes
(1974:2–1989:4; n=63)(28.1)(3.0)
Real broad money31.58−2.051.105.253.47Yes
(1974:2–1989:4; n=63)(42.9)(12.6)
Korea
Real narrow money0.79−0.840.352.753.41No
(1970:1–1989:4; n=80)(35.6)(3.6)
Real broad money1.00−0.780.613.742.59Possibly
(1970:1–1989:4; n=80)(…)(14.0)
Malaysia
Real narrow money1.110.623.744.25Yes
(1970:1–1989:4; n=80)(74.0)(…)
Real broad money1.63−1.650.483.293.38Possibly
(1970:1–1989:4; n=80)(66.9)(5.1)
Myanmar4
Real narrow money1.270.674.833.39Yes
(1970:1–1989:3; n=79)(37.9)(…)
Real broad money1.430.272.692.06No
(1970:1–1989:3; n=79)(26.9)(…)
Nepal
Real narrow money1.750.323.024.45Possibly
(1970:1–1989:4; n=80)(48.2)(…)
Real broad money2.600.202.003.39No
(1970:1–1989:4; n=80)(46.5)(…)
Philippines
Real narrow money50.67−1.160.553.072.94Possibly
(1973:1–1989:4; n=68)(9.0)(4.4)
Real broad money1.470.784.053.00Yes
(1973:1–1989:4; n=68)(25.0)(…)
Singapore
Real narrow money60.86−1.170.454.743.17Yes
(1975:1–1989:4; n=60)(55.4)(5.65)
Real broad money1.37−2.130.462.373.29No
(1975:1–1989:4; n=60)(81.0)(10.3)
Sri Lanka
Real narrow money0.92−1.600.220.550.45No
(1978:1–1989:4; n=48)(12.8)(5.6)
Real broad money1.220.460.402.584.49Possibly
(1978:1–1989:4; n=48)(29.6)(2.0)
Thailand
Real narrow money0.85−1.530.322.182.50No
(1977:1–1989:4; n=52)(40.8)(6.1)
Real broad money1.72−2.460.331.951.59No
(1977:1–1989:4; n=52)(32.5)(6.1)
Note: The numbers in parentheses under the coefficient estimates are t-scatisties.Source: fund staff estimate.

Deposit rates for Furrow money and opportunity cost variables for brood money. See Appendix II for detailed definitions and sources.

The tests for cointegration are described in Engle and Granger (1987) and Engle and Yoo (1987). CRDW (cointegrating regression Dtjrbin-Wacson) and DF (Dickey-Fuller) tests assume the error structure of the long-run relationship is first-order auforegressive For the ADF (augmented Dickey-Fuller) test, the error structure is assumed to be fourth order Significance levels depend on the number of observations (n) and the number of variables in the cointegrating relationship (including the dependent variable). Approximate significance levels are:

(a) Two variablesCRDWDFADF
Sample size1%5%10%1%5%10%1%5%10%
n=501.000.780.694.323.673.284.123.292.90
n=1000.510.390.324.073.373.033.733.172.91
(b) Three variablesCRDWDFADF
Sample size1%5%10%1%5%10%1%5%10%
n=504.844.113.734.453.753.36
n=1004.453.933.594.223.623.32

The Indonesian broad money relationship includes a positive intercept shift dummy for the period 1988:4 onward.

The narrow money relationship for Myanmar includes a negative Intercept shift dummy for the period 1967:3 onward. For broad money, there is a positive intercept shift dummy for 1985:4 onward and a negative intercept shift dummy 1957:3 onward.

The Philippine narrow money relationship includes a negative shift dummy for 1983:4 onward.

The narrow money relationship for Singapore includes a term equal to the one-period-ahead 12-month change in the Singapore dollar against the U.S. dollar.

Note: The numbers in parentheses under the coefficient estimates are t-scatisties.Source: fund staff estimate.

Deposit rates for Furrow money and opportunity cost variables for brood money. See Appendix II for detailed definitions and sources.

The tests for cointegration are described in Engle and Granger (1987) and Engle and Yoo (1987). CRDW (cointegrating regression Dtjrbin-Wacson) and DF (Dickey-Fuller) tests assume the error structure of the long-run relationship is first-order auforegressive For the ADF (augmented Dickey-Fuller) test, the error structure is assumed to be fourth order Significance levels depend on the number of observations (n) and the number of variables in the cointegrating relationship (including the dependent variable). Approximate significance levels are:

(a) Two variablesCRDWDFADF
Sample size1%5%10%1%5%10%1%5%10%
n=501.000.780.694.323.673.284.123.292.90
n=1000.510.390.324.073.373.033.733.172.91
(b) Three variablesCRDWDFADF
Sample size1%5%10%1%5%10%1%5%10%
n=504.844.113.734.453.753.36
n=1004.453.933.594.223.623.32

The Indonesian broad money relationship includes a positive intercept shift dummy for the period 1988:4 onward.

The narrow money relationship for Myanmar includes a negative Intercept shift dummy for the period 1967:3 onward. For broad money, there is a positive intercept shift dummy for 1985:4 onward and a negative intercept shift dummy 1957:3 onward.

The Philippine narrow money relationship includes a negative shift dummy for 1983:4 onward.

The narrow money relationship for Singapore includes a term equal to the one-period-ahead 12-month change in the Singapore dollar against the U.S. dollar.

At the second stage, conventional statistical tests can be carried out to gauge whether the coefficients of the full error correction model are stable across different subperiods of the data sample. It may be the case, for example, that financial liberalization has not affected the long-run relationship between money, income, and interest rates, but has rendered somewhat unpredictable short-run deviations from long-run equilibrium.

Of course, any instability uncovered by the tests at either the first or second stage cannot necessarily be attributed to financial liberalization. It could also be possible that the money demand function was misspecified or that behavioral properties cannot be deduced using available data. Alternatively, instability could reflect inappropriate treatment of expectations: for example, money demand may react quite differently to anticipated, as opposed to unanticipated, events. For these reasons, additional judgmental procedures may be required to trace the source of any instability.

Results

This section presents general results concerning the nature and stability of money demand in the Asian countries: individual country results are reported in more detail in Appendix III. The results suggest that, in most of the countries (Thailand was the only exception), a stable long-run demand function exists for at least one of the monetary aggregates.22 In about half of the countries, narrow money appears to be most reliably related to income and interest rates, while broad money is favored in the other countries. Short-run money demand functions appear to be less stable, although it is difficult to pin any instability to specific episodes of financial liberalization.

Estimation Results

The estimated long-run income elasticities of narrow money demand vary substantially among the Asian countries: for about half of the countries, the elasticity exceeds unity, while for the remainder the elasticity is less than unity (Table 1).23 Measured income elasticities in excess of unity in Indonesia, Malaysia, Myanmar, and Nepal may indicate monetization, while elasticities less than unity in Korea, the Philippines, Singapore, Sri Lanka, and Thailand may reflect agents’ economizing on holding cash balances. While Singapore might be expected to be in the second group because of its relatively developed financial system, and, conversely, Nepal and Myanmar might be expected to be in the first group, it is not clear that the estimated elasticities for the remaining countries are related to their respective stages of financial development.

Interest rates (and deposit rates in particular) generally have significant negative coefficients in the narrow money equations. Furthermore, regression analysis favored the inclusion of interest rates rather than price inflation as the opportunity cost variables of narrow money demand.24 This result held regardless of whether actual, lagged, or leading values of inflation were used to proxy expected price movements. The exceptions are Malaysia, Myanmar, and Nepal where no opportunity cost effects could be found, possibly (in the latter two cases) because of a lack of variability in interest rates in the sample period. For Singapore, expected exchange rate changes were found to play a significant role in determining demand for narrow money in addition to interest rates.

The long-run income elasticities of broad money demand exceed unity in all but one case, mirroring the trend decline in the velocity of broad money in Asian countries during the 1970s and 1980s (Table 1). The exception is Korea for which the estimated income elasticity equals one: a declining gap between interest rates in curb markets and the return on broad money explains the observed downward trend in velocity for this country. The large income elasticities for broad money demand in most Asian countries do not necessarily reflect monetization, which is a phenomenon usually associated with narrower monetary aggregates. Instead, the large elasticities may be capturing growth in wealth in excess of that of income, that, in turn, reflects high savings rates in many Asian countries.25 Rapid growth of savings in the form of time deposits would be consistent with the trends in the composition of money described in the section on “Effects of Financial Liberalization on Money Demand” above.

For several countries (Indonesia, Korea, Malaysia, the Philippines, and Thailand) broad money demand was found to be negatively related to an opportunity cost variable defined as the return on alternative assets minus the average interest rate paid on broad money holdings. A steady decline in the opportunity cost of holding broad money in most of these countries during the 1970s and 1980s contributed to—but by no means can explain all of—the rising share of money in incomes. The declining opportunity cost, in turn, partly reflects some of the changes in the structure of broad money in this period: as the share of interest-bearing quasi-money in broad money increased, the average return on broad money also increased for a given level of deposit rates. Indeed, the interest-bearing proportion of broad money is now so large for many of these countries that the opportunity cost variable is approximately an interest rate differential term (i.e., deposit rates minus the return on nonmonetary assets) implying that broad money is affected by relative asset returns as opposed to the general level of interest rates.

A negative interest rate effect (as opposed to a relative asset return effect) was found for Singapore, while for Sri Lanka, only the return on money appeared to matter, entering the long-run money demand function with a positive coefficient. No interest rate effects could be found for Myanmar or Nepal.

Stability of Money Demand

The formal tests for coinlegration suggest that at least one monetary aggregate in most of the countries investigated appears to have been related in a stable fashion to developments in income and interest rates over the long run. Long-run stability in money demand is also suggested more informally by the similarity between the long-run income elasticities estimated here and those for six of the countries reported in Aghevli and others (1979) (see tabulation below). The latter estimates, which were based on a much earlier sample period stretching back, in some cases, to the 1950s, confirm a remarkable constancy in long-run income elasticities over two very different historical periods.

Estimated Long-Run Income Elasticities of Money1
CountryNarrow MoneyBroad Money
Indonesia1.16 (1.63)1.58 (1.85)
Korea0.79 (…)1.00 (…)
Malaysia1.11 (1.23)1.63 (1.65)
Myanmar1.27 (…)1.43 (…)
Nepal1.79 (…)2.62 (…)
Philippines0.67 (0.85)1.47 (1.54)
Singapore0.86 (1.34)21.37 (1.33)
Sri Lanka0.92 (1.08)1.22 (1.48)
Thailand0.85 (0.68)1.72 (1.49)

Values in parentheses are taken from Table 3 of Aghevli and others (1979).

In Aghevli and others (1979), the estimated income elasticity is not statistically significant.

Values in parentheses are taken from Table 3 of Aghevli and others (1979).

In Aghevli and others (1979), the estimated income elasticity is not statistically significant.

In detail, the estimated long-run narrow money demand functions for Indonesia, Malaysia, Myanmar, and Singapore are likely to be cointegrating relationships, and, therefore, likely to be stable relationships. The same is possibly true—although the formal test results are less convincing—for Nepal and the Philippines. The results for Myanmar and the Philippines require some qualification, however. In the case of Myanmar, an intercept shift dummy variable was needed to take into account the “demonetization” of September 1987 (see Section II). In the case of the Philippines, an intercept shift dummy variable was required to explain developments in narrow money after the fourth quarter of 1983.

It is not clear what the dummy variable in the Philippines’ case is capturing.26 Certainly, the early 1980s saw the introduction of significant financial liberalization in this country, and some shift in money demand behavior might have been expected around this time. More generally, however, the Philippines experienced a period of financial turmoil after 1983, including high inflation rates, and it is also likely that the dummy variable reflects a failure to fully account for pessimistic expectations that may have precipitated a flight from narrow money in this period.

While the formal statistical tests are satisfied, the stability of the long-run relationship for narrow money in Indonesia was questionable and a break in behavior after the June 1983 financial liberalization package is possible. In particular, estimates based on the subperiod 1974 to the second quarter of 1983 fail to identify the significant interest rate effect estimated for the full sample. In addition, the income elasticity is a little higher in this subperiod, although there is no evidence of a onetime shift in the level of narrow money holdings in 1983. These results provide tentative evidence that the June 1983 measures, which included interest rate liberalization, increased the transparency of the relationship between narrow money and interest rates.

The estimated relationships for Korea. Sri Lanka, and Thailand do not appear to represent stable, long-run narrow money demand functions, Attempts to identify the source of instability, and, in particular, whether financial liberalization played a significant role, proved inconclusive, reflecting, in part, the gradual nature of liberalization in these countries. For Korea, the sharp rise in the velocity of narrow money in the early 1980s is not well explained by the estimated equation. This rise in velocity may have been related to the availability of new, alternative financial assets, but, equally, the rise may be related to the high rate of inflation in the early 1980s.27 For Sri Lanka, developments in financial variables do not explain the abrupt reversal of the upward trend in narrow money velocity after 1984: instead, the measured income elasticity increases sharply in the second half of the 1980s. This reversal may be related to political turmoil in this country which may have encouraged the public to hold a larger proportion of their savings as liquid assets. For Thailand, the main source of instability is the measured interest rate elasticity. Interest rates gain significance as the data sample progresses (in the period 1978–85 they are not significant at all) possibly reflecting gradual liberalization during the 1980s. It should be noted, however, that since interest rate data were only available from the late 1970s onward for Sri Lanka and Thailand, the tests for long-run stability of money demand are not very powerful.

The estimated long-run broad money functions for Indonesia and the Philippines appear to be co-integrated relationships, and there is also weak support for cointegration in the estimated relationships for Korea, Malaysia, and Sri Lanka. For Indonesia, however, an intercept shift dummy is required for the period following the substantial lowering of reserve requirement ratios in the October 1988 financial reform package, which precipitated a surge in bank intermediation. Interestingly, no intercept shift was detected after the major financial reform package of June 1983, and income and interest rate elasticities are almost identical in the subperiod 1974–83 to those estimated for the full sample.

The estimated relationships for Myanmar, Nepal, Singapore, and Thailand do not appear to represent stable long-run broad money demand functions. In the case of Nepal, for example, the long-run income elasticity was substantially different in the 1970s (over three) compared with the 1980s (just under two). The income elasticity for Thailand also declines over the course of the sample period, reflecting a slowdown in the trend decline in broad money velocity that cannot be explained by developments in interest rates. By contrast, the instability in the case of Singapore shows up as a rising interest rate elasticity in recent years. More generally, the instability of money demand in Singapore may reflect an inability to capture the high degree of currency substitution between Singapore dollars and the Asian Dollar Market.

The existence of stable error correction models of money demand was limited to just four countries: Indonesia and Malaysia (narrow and broad money equations) and Korea and Sri Lanka (broad money equations only). The equations for these four countries had satisfactory statistical properties and passed formal tests for parameter constancy (Tables 2 and 3).28 In Indonesia, the estimated coefficients in the error correction equations for broad and narrow money did not appear to have been significantly affected by either the 1983 or 1988 financial reform packages.29 In Korea, however, there was evidence of residual autocorrelation in estimates based only on data from the 1980s, possibly indicating an omitted variable in more recent subperiods when financial liberalization intensified.

Table 2.Asian Countries: Error Correction Equations—Narrow Money1
Dependent VariableLagged Dependent VariableDeposit Rates Lagged OnceReal IncomePricesError Correction TermConstantR¯2SE (Percent)DWAUTO (5)2FORE (16)3
Stable, well-specified equations
Indonesia
Nominal money0.70−0.45−0.100.020.572.02.0011.3823.56
(1974:4–1989:4)(7.8)(2.5)(2.7)(3.2)
Malaysia
Real money0.240.15−0.200.010.102.51.985.3116.08
(1970:3–1989:4)(2.1)(1.4)(3.2)(3.0)
Poorly specified or unstable equations
Myanmar
Nominal money0.480.52−0.440.0140.674.02.2212.94
(1970:1–1989:3)(7.0)(4.8)(5.7)(1.1)
Nepal
Nominal money0.680.27−0.110.010.431.81.5120.4418.96
(1970:3–1989:4)(7.6)(2.1)(3.3)(1.1)
Philippines
Real money0.57−0.340.20−0.130.010.412.41.979.8634.12
(1973:3–1989:4)(5.6)(2.0)(2.1)(2.7)(1.7)
Source: Fund staff estimates.

Specification as equation (2) in the main text. Variables other than the error correction terms are all first differenced. Variables other than interest rates are in logarithms. Numbers in parentheses are t-statisties.

Test for (up to) fifth-order autocorrelation; see Godfrey (1978). Distributed as χ2 with 5 degrees of freedom. A value greater than the critical value 11.07 implies possible autocorrelation at the 5 percent significance level.

Test for parameter stability in the last 16 periods of the data; see Hendry (1980). Distributed as χ2 with 16 degrees of freedom. A value greater than the critical value of 26.30 implies parameter instability at the 5 percent significance level.

Equal to –0.22 in 1987:4.

Source: Fund staff estimates.

Specification as equation (2) in the main text. Variables other than the error correction terms are all first differenced. Variables other than interest rates are in logarithms. Numbers in parentheses are t-statisties.

Test for (up to) fifth-order autocorrelation; see Godfrey (1978). Distributed as χ2 with 5 degrees of freedom. A value greater than the critical value 11.07 implies possible autocorrelation at the 5 percent significance level.

Test for parameter stability in the last 16 periods of the data; see Hendry (1980). Distributed as χ2 with 16 degrees of freedom. A value greater than the critical value of 26.30 implies parameter instability at the 5 percent significance level.

Equal to –0.22 in 1987:4.

Table 3.Asian Countries: Error Correction Equations—Broad Money1
Dependent VariableLagged Dependent VariableOpportunity CostReal IncomeReal Income Lagged OnceError Correction TermConstantR¯2SE (Percent)DWAUTO (5)2FORE (16)3
Stable, well-specified equations
Indonesia
Real money0.550.13−0.13−0.210.010.362.42.035.5321.23
(1974:3–1989:4)(5.2)(2.4)(…)(3.9)(3.4)
Korea
Real money0.48−0.100.010.242.41.747.435.12
(1970:2–1989:4)(5.0)(2.0)(4.2)
Malaysia
Real money0.23−0.39−0.120.020.162.71.995.3720.23
(1970:3–1989:4)(2.1)(1.8)(3.0)(4.8)
Sri Lanka
Nominal money0.600.234−0.150.020.481.71.743.488.75
(1978:2–1989:4)(6.2)(1.6)(2.4)(3.2)
Poorly specified or unstable equations
Philippines
Real money0.44−0.550.19−0.1150.010.492.72.215.7026.16
(1973:3–1989:4)(4.7)(3.6)(1.9)(2.2)(1.8)
Source: Fund staff estimates.

Specification as equation (2) in the main text. Variables other than the error correction terms are all first differenced. Variables other than interest rates are in logarithms. Numbers in parentheses are t-statistics.

Test for (up to) fifth-order autocorrelation; see Godfrey (1978). Distributed as χ2 with 5 degrees of freedom. A value greater than the critical value 11.07 implies possible autocorrelation at the 5 percent significance level.

Test for parameter stability in the last 16 periods of the data; see Hendry (1980). Distributed as χ2 with 16 degrees of freedom. A value greater than the critical value of 26.30 implies parameter instability at the 5 percent significance level.

Deposit rates.

Equal to –0.46 from 1983:4 onward.

Source: Fund staff estimates.

Specification as equation (2) in the main text. Variables other than the error correction terms are all first differenced. Variables other than interest rates are in logarithms. Numbers in parentheses are t-statistics.

Test for (up to) fifth-order autocorrelation; see Godfrey (1978). Distributed as χ2 with 5 degrees of freedom. A value greater than the critical value 11.07 implies possible autocorrelation at the 5 percent significance level.

Test for parameter stability in the last 16 periods of the data; see Hendry (1980). Distributed as χ2 with 16 degrees of freedom. A value greater than the critical value of 26.30 implies parameter instability at the 5 percent significance level.

Deposit rates.

Equal to –0.46 from 1983:4 onward.

The failure to find satisfactory error correction models for many of these countries implies that the short-run predictability of monetary developments is poor. While this result is consistent with the expected effects of financial liberalization on developments in the monetary aggregates, the result may also reflect, in part, the poor quality of some of the quarterly data used in the estimation.

Conclusions

Analysis of the stability of money demand in the Asian countries during the 1980s produces mixed results. For most countries, at least one monetary aggregate appears to have been related in a stable long-run fashion to income and interest rates. However, regardless of long-run stability, short-run deviations from long-run equilibrium were, on the whole, rather unpredictable. In general, it is difficult to link any instability to specific financial reform measures, partly because reforms have been introduced gradually and on many fronts in most countries.

Instability of the estimated money demand functions manifested itself in different ways, depending on the monetary aggregate and country in question. In a few cases (broad money in Indonesia in 1988, narrow money in the Philippines in 1983) onetime shifts in the money demand appear to have taken place. In other cases, instability is reflected in changing income and interest rate elasticities. For example, income elasticities have declined in recent years for broad money in Nepal and Thailand and risen for narrow money in Sri Lanka, while in Singapore the interest elasticity of broad money has risen significantly. Finally, in some cases (narrow money in Indonesia and Thailand), instability is reflected in the appearance of previously unimportant interest rate variables in the money demand function.

Interest rates are becoming increasingly important determinants of both broad and narrow money demand in the Asian countries. To some extent, this reflects a greater flexibility of interest rates during the 1980s which has permitted interest rate effects to be measured; the previous study of monetary policy in these countries had to make do with inflation rates as the only measure of the opportunity cost of holding money.30 In the case of broad money, the analysis here also suggests that simple interest rate variables are not the most appropriate terms in the money demand function. In particular, as the proportion of interest-bearing quasi-money in broad money has increased, broad money demand now depends in many cases on the relative return between money and other financial assets.

In conclusion, monetary aggregates remain useful policy variables, but this usefulness remains subject to qualification. Financial liberalization has clouded the predictability of monetary developments, at least in the short run. The nature of broad aggregates also appears to have changed significantly with implications for their controllability and relationship to ultimate policy targets.

    Other Resources Citing This Publication