RDP 9602: Consumption and Liquidity Constraints in Australia and East Asia: Does Financial Integration Matter? 3. Estimation And Results

Motivated by the discussion in Section 2, the estimating equation including the proxy for the shadow price on the constraint and the unrestricted error correction is:

The proxy is a current-dated variable, and is likely to be correlated with the consumption innovation, rendering the estimated coefficients biased and inconsistent. This is notably so when current income is used as the proxy since innovations in current income are likely to be correlated with unforecastable innovations in permanent income which, liquidity constraints aside, are related to innovations in consumption in standard models (Muellbauer 1983 and Blanchard and Fischer 1989). Accordingly, an instrumental variables procedure is used for the liquidity constraint proxies and the real interest rate. Lagged consumption and income are not instrumented since they are pre-dated and so orthogonal to the error term.

The instrumental variables are lagged two periods.^[4] In the case of income, the use of data for which the reporting interval exceeds the planning interval generates spurious correlation between the current consumption growth and the first lag of income growth, rendering first lags inadmissible as instruments (Deaton 1992, pp. 96–98). Using the second lag has the additional benefit of excluding possible problems with measurement error or transitory income. Nelson (1987) and Attanasio and Weber (1989) report, however, that the effect of temporal aggregation on the rejection of REPIH is minimal, and its importance can be judged by comparing the results for OLS and IV estimation.

To introduce the material, a basic model, popular in the literature, in which the growth of real per capita non-durables consumption is regressed only on a constant and the growth of real per capita income, was estimated using instrumental variables. The results are reported in columns (2) and (3) of Table 3, with the coefficient of determination in column (4). To form a view on the constancy of the regression coefficients, the equation was estimated for income growth with a multiplicative decade dummy (excluding Australia and the Philippines because of the short sample), shown in columns (5) to (7). Generally, the diagnostics were sound.^[5]

The results indicate a high degree of variation in the income dependence of non-durables consumption across countries and time. In Australia, non-durables expenditure growth has been independent of income growth. This is not a problem of finding appropriate instruments since the coefficient is insignificant even using OLS (and in OLS it is minuscule). Non-durables consumption growth has been tied to income growth in Hong Kong, Japan, Korea, Singapore, Taiwan and Thailand, but has fallen (by more than one standard error) or become insignificant in the 1980s and 1990s in Japan and Singapore. The dependency of non-durables expenditure on income has remained unchanged (that is, within one standard error) in Hong Kong, Taiwan and Thailand but has increased in Korea. In Indonesia, Malaysia and the Philippines, total consumption expenditure growth depends on income and in the case of Malaysia, this is increasingly so over time.

These results largely accord with priors about the degree of financial repression and capital account openness in these economies, as outlined in de Brouwer (1996a). For most of the period, Australian financial markets have been open and highly developed. Domestic financial repression was lower and access to international capital markets greater in Japan and Singapore in the 1980s than in the previous decade. Financial and exchange reform in Korea, Taiwan and Thailand is only recent. Oddly, the coefficient on income growth in Hong Kong is consistently high.

The results and interpretation above are only tentative since relevant variables, such as the real interest rate and demographic change, and other variables which enable more specific identification of the form of the constraint have been excluded. Accordingly, a grid search procedure was applied to identify whether the real interest rate, demographic variables and the proxies for liquidity constraints other than current income growth were significant in economies for which non-durables expenditure data are available. The estimation procedure was as follows. An equation including current income growth, the first lags of consumption and income, the real interest rate and one of the three demographic variables was estimated, and insignificant regressors eliminated. This procedure was applied to the three definitions of the three demographic variables. Current income growth was then replaced by the other proxies for liquidity constraints and the procedure re-applied. If these proxies were statistically significant, current income growth was returned to the regression to test their relative explanatory power. Not all proxies were available for all countries and estimation periods vary with data availability. Tables B.1 through to B.7 in Appendix B present the statistically significant results for Australia, Hong Kong, Japan, Korea, Singapore, Taiwan and Thailand.

The results vary substantially between countries. The coefficient on the real interest rate is positive and significant in Japan, Singapore and Thailand, indicating the importance of inter-temporal wealth effects in consumption and suggesting that the basic Keynesian model with consumption solely dependent on income is inadequate. Demographic variables are relevant in Japan, Taiwan and Thailand. In Japan, the change in the dependency ratio is significant and negative. In Thailand, the change in the proportion over 64 years is significant and negative. In Taiwan, however, the dependency ratio is significant and positive. A linear trend was included in these regressions to test whether demographic change is picking up a simple trend in the data, but it was not significant.

The significance or otherwise of the eight proxies for the shadow price listed in Table 2 are summarised in Table 4, with n indicating not significant, y indicating significant and n/a indicating that the series is not available for that country. In all countries, except Australia and Hong Kong, an error-correction between income and consumption was statistically significant. It was argued earlier that an error-correction may be due, among other things, to aggregation over households or to liquidity constraints. But the result that the error-correction is insignificant only for the two countries with among the most open and developed financial markets in the region suggests that the error-correction arises because of liquidity constraints rather than aggregation effects. Consistent with the results for the basic model outlined in Table 3, current income growth is a significant determinant of consumption growth for all countries (except Australia). In all cases, except real residential property prices in Japan, income growth added the most explanatory power of all the proxies for liquidity constraints, which is not surprising since it is a catch-all variable.

Australia is an outlier in that neither current income growth nor any variable outlined in Tables 1 or 2 could systematically explain the growth in non-durable consumption. This supports the proposition that non-durables consumption in Australia is unconstrained and smoothed. The sample period is relatively short (1975–1994), and when it is extended back to 1970 using the total private expenditure deflator in place of the non-durables deflator, the coefficients and equations are only significant when decade income growth dummies are included. In this case, income growth is significant in the 1970s but not in the 1980/90s (see Table B.5), consistent with the results in Table 3. The irrelevance of current income does not follow for broader definitions of consumption, although the interpretation is muddier when services and durables are included since issues of data smoothing and stock adjustment arise. When consumption is defined as total private consumption expenditure, income growth, the error-correction, the proportion of the young working age population, and inflation are significant explanators. The coefficient on the share of young workers is negative, consistent with the view that the accumulation of reputation and collateral reduce constraints on smoothing. Rising inflation also depresses real consumption, presumably (in an inter-temporal model) because it increases uncertainty in real income^[6] and it suppresses the real interest rate when the nominal interest rate does not adjust instantaneously and completely, thereby generating a negative intertemporal substitution effect. The result that the basic Hall model is rejected for total expenditure is not surprising given the lumpiness of durables expenditure and the expectation that liquidity constraints operate most strongly with this type of expenditure. The insignificance of financial type proxies is consistent with the developed and open financial markets of Australia. Similar to Blundell-Wignall, Browne and Tarditi (1995), but unlike Debelle and Preston (1995) who use a long span of quarterly data, there is no evidence of a secular decline in the liquidity constraint on total expenditure.

The evidence of liquidity constraints in East Asia is stronger, although the outcome varies substantially by country. Consider Hong Kong and Singapore, two of the most financially developed economies in East Asia. In the case of Hong Kong, current income growth and the nominal loan rate are the only significant proxies, but in both cases the explanatory power is low. Hong Kong's financial markets are in general large and well-developed (although its domestic banking market was cartelised and domestic deposit and loan rates segmented somewhat from the domestic money market during this period (de Brouwer 1995)). There is no secular decline in the dependency on the proxies. In Singapore, the evidence of a liquidity constraint over the whole period is more robust, with the error-correction and current income growth both significant. Substituting for other proxies of liquidity constraints, financial deepening and the nominal interest rate are both significant and signed as expected (the former positive, the latter negative).^[7] The significance of financial deepening implies that the constraint is declining over time as the depth of markets expands. Domestic markets were substantially liberalised in 1975 and capital controls removed in 1978, so one would expect a break in the regression between the 1970s and the 1980s/90s. Indeed, the coefficient on both income growth and financial depth falls over decades. The results suggest that liquidity constraints have eased in Singapore over time, due to financial deregulation.

Japan provides a different perspective on constraints and financial deregulation. Like Singapore, it initiated capital account reform in the middle of the sample period (1980) but it deregulated its bond, money and non-traded financial markets only slowly over the 1980s and 1990s, with the liberalisation of bank deposit rates, for example, extending from 1985 to 1994 (de Brouwer 1995, 1996b) and the Bank of Japan only stopping window guidance in 1991. The real interest rate, the change in the dependency ratio, the error-correction and current income growth are significant – a classic example of constrained intertemporal optimisation with demographic change. When current income growth is excluded, financial depth and real residential land prices are also significant proxies of the liquidity constraint.^[8]

At first glance, the significance of financial depth suggests that in Japan, like Singapore, deregulation has enabled households to expand their consumption. There are, however, two pieces of contrary evidence. Firstly, when sub-sample decade dummies are included, the liquidity constraint, measured as income growth or real land prices, is significant in both periods and does not decline in the 1980s/90s relative to the 1970s. Secondly, real residential property prices are strongly significant and explain consumption growth ‘better’ than the money/GDP ratio, which is consistent with the claim that money in this case only proxies asset prices. As shown in Appendix A, money/GDP and real residential property prices follow the same trend and in the early 1970s and the late 1980s both rose and fell sharply. In the early 1970s, when domestic financial markets were closed, controlled and narrow, the monetary expansion fed directly into asset price inflation. In the second half of the 1980s, when domestic financial markets were being deregulated, expansionary monetary policy also fed directly into asset price inflation. In both episodes, policy was expansionary and asset prices rose, but only in the latter period were markets (being) deregulated, suggesting that deregulation has not yet had a separate identifiable effect on non-durable consumption. The tentative evidence in Table 3 of declining constraints looks incorrect: it is too early to identify an effect on consumption. Given that the capital account was liberalised at the start of the 1980s, however, the results are consistent with the interpretation that international financial openness is not sufficient for consumption smoothing when domestic markets remain undeveloped or controlled.

Korea, Taiwan and Thailand are examples of economies with regulated or repressed domestic and financial systems and controls on the capital account. They are an interesting contrast to Australia, Hong Kong and Singapore. In the case of Korea, the error-correction and current income growth are the only significant explanatory variables.^[9] The coefficients are stable over sub-periods, indicating no change in liquidity constraints. While these results are consistent with the casual observation that financial markets in Korea are among the least developed and liberalised in this sample set, Korea is also known as a country which has regularly used controls on non-durable consumption imports (particularly from Japan) to control demand and the current account (Hasan and Rao 1979, p. 271 and Kim 1991, p. 47). The insignificance of financial variables suggests that it may be trade controls which give rise to the constraint.

The results for Taiwan and Thailand, on the other hand, show stronger evidence that financial development affects consumption. For Taiwan, the change in the dependency ratio, the error-correction, current income growth and the margin between loan and deposit rates are jointly significant with the expected signs (except for the change in the dependency ratio). When current income growth is replaced with other proxies for the liquidity constraint, financial depth, the deposit rate and the loan rate enter significantly with the expected sign. For Thailand, aging, real interest rates, the error-correction and current income growth are significant and signed as expected. When current income growth is replaced in the equation for Thailand, financial depth, deposit and loan rates, and the loan-deposit margin are significant with the expected sign. The significance of both the interest rate margin and financial depth point to financial repression or lack of development as causes of the liquidity constraint.^[10] There is little evidence of falling liquidity constraints in either country: the coefficients on decade dummies for current income growth and other proxies for the constraint are not significantly different over time. This is not surprising since substantive reform in both countries is only relatively recent. While the capital account had been partially liberalised by the late 1980s in Taiwan, systematic controls on inflow still remain (Lee 1990, pp. 160–161), covered interest parity did not hold from 1991 to 1994 (de Brouwer 1996a) and additional foreign exchange controls have been implemented occasionally (for example, mid 1992). Substantive liberalisation of the Thai capital account only took place in May 1990 and April 1991 (de Brouwer 1996a). Moreover, domestic interest rate liberalisation was only implemented in 1989 in Taiwan and in 1992 in Thailand (de Brouwer 1995).

There are two final comments to be made about the modelling procedure and results. Firstly, while the proxies for the shadow price of the liquidity constraint are not derived from first principles, when they are statistically significant they are signed as expected, and this lends support to the model. For example, financial deepening (and real interest rates) are always positively correlated with consumption, and widening interest margins and nominal interest rates are always negatively correlated with consumption.

Secondly, the results are probably not the outcome of data mining by which a grid search over a series of proxy variables has been used to select only those relationships which are significant (Lovell 1983). In the first place, all the proxy variables are relevant a priori to identifying the impact of the liquidity constraint on consumption, so degrees of freedom have not been ‘wasted’. More to the point, statistically significant proxies tend to be bunched together for countries which have less developed, free and open financial markets rather than spread uniformly across all countries as would be the case if the results were purely random over a set of countries. It seems less plausible in this case, therefore, that the Type 1 error of rejecting the null hypothesis that the coefficient is different from zero when it is true is occurring to the detriment of correct inference. Whatever the case, the process is transparent since all the variables tested are cited and the standard error of the significant variables is presented in Appendix B. The marginal significance levels of the proxy variables is usually less than the standard 5 per cent level, often substantially so.

Footnotes

The instruments used vary with the country examined but are generally natural logs of per capita real income, per capita real consumption, real exports, real US GDP and the unemployment rate, in levels and in first differences. Lagged consumption and income are not instrumented since this does not affect the results. [4]

The equations were estimated using the instrumental variables option in Microfit 386, and the standard diagnostics are Sargan's test of misspecification, a Lagrange multiplier test for first-order serial correlation, Ramsey reset functional form test, Bera-Jarque normality test and the Ramsey reset heteroscedasticity test. The error term in the Japan equation was serially correlated and corrected using the Newey-West procedure with a Parzen window of 2 lags. [5]

The model does not directly address the issue of uncertainty and precautionary saving but Blanchard and Fischer (1989, pp. 288–291) present a model where consumption is decreasing in real income uncertainty. [6]

Real residential property prices were not significant but real residential property price inflation entered with a positive and significant coefficient in OLS regressions. A suitable set of instruments could not be found for IV estimation. [7]

The Nikkei 225 stock price index was also significant but the equation was only marginally significant. [8]

The coefficient on the curb loan rate is significant and negative but only in OLS and not IV estimation. The results across countries are rarely sensitive to the estimation procedure, which suggests that in this case the problem is finding suitable instruments for the curb loan rate. [9]

It was argued above that the money/GDP ratio reflected wealth effects for Japan rather than financial deepening, and so it is necessary to answer the question here of why this may not also be the case for Taiwan and Thailand. Data for residential property prices for Taiwan and Thailand are not available and so the question cannot be answered conclusively but there is strong indirect evidence against it. Firstly, the interest margin is also significant for these two countries, which is corroborating evidence of the effect of financial repression. Secondly, significance of the money/GDP ratio does not mean that asset prices are also significant and vice versa. In the case of Australia, real residential property prices are barely significant at the 10 per cent level but the money/GDP ratio is not. [10]