RDP 9209: Financial Liberalisation and Consumption Behaviour 4. Estimates of λ by Decade and the Liquidity Constrants Interpretation

The following steps were taken to overcome some of the concerns that arose in applying the J and P approach. First, utility was assumed to be represented as in equation (6), so that non-liquidity constrained consumers behave according to equation (7):

where

This avoids the non-stationarity problems discussed in the previous section. Second, it was assumed that current consumption of liquidity-constrained consumers is a constant fraction of current income – so the expected rate of growth of consumption equals that of disposable income. Even if liquidity constrained, it seems unreasonable to assume that Keynesian consumers always spend exactly all of their income. This implies:

Equations (13) and (14) are weighted according to the proportion of the population λ that are liquidity constrained:

where e_4t = λ_e3t + (1−λ)e₂t

Third, quarterly time series of all variables are employed. Fourth, common consumption definitions are utilised to facilitate international comparisons. Equation (15) was estimated for total consumption for eight OECD countries.^[3] It was also estimated for total consumption less the purchases of durables for that subset of countries for which such data are available.^[4]

Instrumental variables estimation was used to examine whether the degree of excess consumption sensitivity (which is measured by the λ parameter in equation (15)) is linked to imperfect capital markets.^[5] This linkage is examined mainly by exploiting the knowledge that financial liberalisation has progressed over time, particularly in the 1970s and 1980s. If the liberalisation process seen in financial markets has caused liquidity constraints to be progressively relaxed, then estimating equation (15) for successive time periods should tend to indicate a generally reduced size and significance of the λ parameter. Three time periods are chosen, the 1960s, the 1970s and the 1980s (see Table 1 for precise dates). These, of course, may not be the economically most relevant time periods, given the gradual and diffuse nature of deregulation which makes it difficult to identify clear structural shifts. Nevertheless, with this approach less emphasis is placed on the cross-country ranking of λ than in the J and P paper. Formal tests can be applied to differences in λ values for individual countries between decades.

(a) Nested Model Estimates for Consumption

Table 1(p27) presents the results from estimating equation (15), in its change in logarithms form, for eight large OECD countries. Overall, the results suggest the excess sensitivity parameter λ varies across countries, as does its pattern over time. The United States, Japan, Canada and Australia show evidence of declining liquidity constraints. For the United States economy, there is no evidence that liquidity constraints were lessened in the 1970s compared to the 1960s. However, the λ parameter falls from a significant value of 0.47 in the 1970s to a statistically insignificant value of 0.25 in the 1980s, which is consistent with reduced liquidity constraints. For Japan, the magnitudes of the estimated λ parameter suggest that liquidity constraints in the 1980s are less severe than they were in either the 1960s or 1970s. The size and pattern of estimated coefficients for Canada and Australia are quite similar. The λ estimate for Australia is significant in all three decades, but it declines in value during the 1970s and 1980s. The pattern for Canada conforms closely to expectations, since Canada was one of the first countries in the sample to deregulate its financial markets.

The results for Germany, France and Italy are similar in that they show no evidence of declining liquidity constraints. For Germany, the λ parameter is significant at the 5 per cent level for all three subsample periods, and increases in value from 0.37 in the 1960s to 0.67 in the 1970s and to 0.98 in the 1980s. Given that German households are known to have a strong preference for saving, increasing precautionary saving behaviour in the more uncertain environments of the 1970s and 1980s, together with the presence of liquidity constraints, might explain our results. For France, a significant excess sensitivity parameter (λ=0.40) is estimated when the extended period of the 1960s and 1970s together is used (see final column of Table 1). The robust errors estimate of the 0.31 coefficient for France in the 1980s suggests little easing of liquidity constraints in this period. Similarly, the Italian results show little change in the estimated values of λ (within the range of 0.43 to 0.47) between decades.

The United Kingdom results appear to have little in common with the two groups of countries discussed so far. The PIH appears to be accepted by the data in the 1960s and 1970s, but the size and significance of the λ parameter both increase during the 1980s. Possible reasons for this are discussed in Section 6.

Campbell and Mankiw (1989) observe that an estimate of the excess sensitivity parameter close to zero, even if statistically significant, supports the PIH, while a large value of this coefficient suggests rejection. According to this criterion, the pattern of the λ parameters across decades generally suggests easing liquidity constraints in four of the countries considered, but not in Germany, France, Italy and the United Kingdom. However, it is possible to test more rigorously whether declines in the λ parameter in successive time periods are significant in a statistical sense. The results, based on the unit normal distribution, are given in Table 2. The null hypothesis is that the λ value in the most recent period is smaller than in the preceding period. These results are a strong qualification to any interpretation based on the Campbell and Mankiw criterion. Only in a few instances (indicated by asterisks) does this test accept the null hypothesis of falling liquidity constraints.

(b) Asymmetries in the Nested Model

Implicit in the above tests is the proposition that liquidity constraints encountered will have the same effect regardless of the direction of movement in disposable income. This assumption may be unrealistic, since reductions in income are likely to be more constraining than increases – at least for large changes in income. If the consumer is rationed in credit markets and current income falls substantially then consumption must fall. If income increases under the same credit market conditions, consumption may or may not increase. This asymmetry may be important for individuals with little or no non-human wealth, for whom necessary expenditure is a very large percentage of disposable income. Therefore, the presence of this type of asymmetry can be taken as further evidence of liquidity constraints. Moreover, as these constraints unwind over time with financial liberalisation, so should the magnitude of the asymmetry.

To test this proposition, the current values of the change in disposable income are transformed as follows:

The hats indicate instruments obtained on y^d using the exogenous variables already noted (see footnote (5)). The test equation (15) is now rewritten as follows:

Tests for asymmetric behaviour based on equation (16) are carried out with a simple t-test. The null hypothesis is assumed to be λ₂ > λ₁ and the alternative that λ₂ ≤ λ₁. The results are presented in Table 3.

The null hypothesis of significant asymmetries is accepted at the 5 per cent level for the United States, Germany and Canada for the 1960s; Japan, Italy and Canada for the 1970s; and France, Italy and Australia in the 1980s. Furthermore, when asymmetry was accepted, the magnitude of the response of consumption to a fall in income was, in most cases, several multiples of the effect of an equivalent increase in income. In no case was the effect on consumption of an increase in income significantly greater than that for a fall. These results lend further support to the liquidity constraints interpretation of the excess sensitivity parameter. They also further corroborate the evidence already reported of easing liquidity constraints over successive decades. This conclusion is inferred on the basis of the magnitude of the average differences in the λ₁ and λ₂ estimates in successive subperiods (2.27, 0.29 and 0.22 for the 1960s, 1970s and 1980s, respectively^[6] (see the bottom panel of Table 3)).

(c) Real and Nominal Interest Rate Effects

An attempt to add more realism to the nested consumption model in equation (15) is made by relaxing the assumption that the real interest rate is constant. This allows for changing intertemporal substitution in consumption. The exclusion of the real interest rate from the test equation could conceivably be leaving real disposable income growth to pick up this changing intertemporal substitution effect, i.e. if real income growth and real interest rates are correlated over time. Nominal interest rate changes, in the absence of real interest rate changes, should of course have no effect on consumption unless households are prevented from consuming from their permanent income by imperfect capital markets. It has been argued that such an imperfection in personal credit markets is reflected in banks' practice of using virtually constant repayment-to-current income ceilings as criteria for loan qualification.^[7] An increase in the nominal rate of interest, for a fixed real rate, may cause this ceiling to be breached and the potential borrower to be denied a loan. Thus if the liquidity constraints theory is valid, consumption can be constrained by variations in both current disposable income and in nominal borrowing rates of interest. Unless nominal interest rate effects are controlled for, variations in the imperfectly measured real rates could capture liquidity constraint effects coming from changes in nominal rates.

To capture both intertemporal substitutions and liquidity constraints phenomena arising from interest rate changes, equation (15) is accordingly amended as follows:

An increase in last period's real borrowing rate of interest r_t−1 reduces that period's consumption relative to that of the current period (α>0). An increase in the nominal interest rate i_t in this period for a fixed real rate will tighten household liquidity constraints and dampen consumption expenditure (γ<0) if capital markets are imperfect.

The cross-country results from estimating equation (17) are given in Table 4. They indicate support for significant changing intertemporal substitution effects for the United States (for the 1960s/1970s subperiod and also for the 1980s) and for the United Kingdom (for the 1970s and the 1960s/1970s subperiods). There is also some weaker evidence for Japan favouring changing intertemporal substitution effects. Note, however, that for Italy in the 1960s this effect was significantly negative. These results are of some interest in the light of the failure of many recent empirical studies to uncover a significant positive intertemporal elasticity of substitution (see, for example, Mankiw, Rotemberg and Summers (1985), Hall (1978), Campbell and Mankiw (1989) and Bayoumi and Koujianou (1989)). Nominal interest rate effects arising from liquidity constraints are significant for Japan (1960s/1970s), for Germany (1980s), for France (1970s), and for Italy and Canada (1960s). However, for the United States, nominal interest rate increases appear to have promoted consumption in the 1980s.

There is, therefore, some evidence of both intertemporal substitution effects and liquidity constraint effects arising from real and nominal interest rate changes, respectively. However, in no instance does the presence of these effects require any substantive amendment of the conclusions already arrived at with respect to the changing pattern of liquidity constraints across countries or over time. Hence liquidity constrained consumers cannot be explained away by allowing households to redistribute their consumption over time in response to changes in the intertemporal relative price.

Footnotes

The countries in question are the United States, Japan, Germany, France, Italy, the United Kingdom, Canada, and Australia. Some of the results reported in Sections 4 and 5 were just reported in an early working paper version by Blundell-Wignall, Browne and Cavaglia (1991). [3]

These countries are the United States, Japan, France, Italy, the United Kingdom, and Canada. The results for total consumption less purchases of durables do not yield overall inferences which are sufficiently different from those for total consumption to warrant reporting, perhaps because of the emphasis on the time variation of λ. These results are available, however, on request. It is also worth recalling that use of total consumption is less likely to create the illusion of excess sensitivity due to the stock adjustment of durables (Bernanke 1985). [4]

The instruments actually employed for the lag level of disposable income are three lags of this variable itself, as well as three lags on the lag level of personal consumption, the unemployment rate, and total exports, all in per capita terms, as well as contemporaneous population and a time trend. [5]

These apparently systematic patterns may be biased by the very large λ₂ estimate for Italy for the 1960s. If, in fact, there were very few observations on Δy^- it might be operating as a dummy variable picking up some other influence. [6]

Wilcox (1989) refers to a recent American Bankers Association textbook on consumer lending which suggests that a borrower's capacity to repay a loan can be measured by the payment-to-income ratio. Wilcox concludes that, in practice, this means the current payment-to-current income ratio. [7]