RDP 9209: Financial Liberalisation and Consumption Behaviour 5. Cross Country Correlation of Residuals and Pooled Results

Adrian Blundell-Wignall Frank Browne Stefano Cavaglia and Alison Tarditi

September 1992

A worrying aspect of the Table 1 results is the general inability to reject the null hypothesis that the declines in λ are significant according to the unit normal test. This could be due to the use of an inefficient estimation procedure. An econometric issue not addressed in most of the previous literature concerns the possible importance of cross correlations between the error terms in the test equations for the countries being studied. Thus, for example, income and consumption shocks in one country could be translated through standard international linkages to income and consumption shocks in others. Alternatively, common shocks (e.g. oil price changes) could be important. It is possible to take account of this problem by pooling the data for a number of countries and using a Seemingly Unrelated Regression Estimation (SURE) procedure to estimate individual country parameters. This should enable more efficient estimates of the λ parameter.

More efficient estimates might also be obtained if the λ parameter can be estimated jointly for a number of countries where financial deregulation is thought to have been broadly similar, provided such a restriction is accepted by the data. This requires the countries considered suitable for pooling to be chosen on a priori grounds, i.e. on the basis of what is known about the deregulatory policies in each.

(a) Country Groupings Based on Information about Deregulation

Financial regulations generally fall into two broad categories:

  1. “rate/quantity” regulations on bank deposits and loans, including ceilings on bank deposit rates and quantitative measures that have similar effects (credit ceilings, capital controls, etc.); and
  2. “powers” regulations governing the extensiveness of activities of individual financial institutions and their competitiveness.

In general terms, there are considerable differences in emphasis between countries in the extent to which “rate/quantity” regulations have been removed and/or “powers” regulations still apply. Developments are summarised in Table 5. The United States, the United Kingdom, Canada and Australia moved relatively early and with some rapidity in removing rate/quantity and powers regulations. While some powers regulations still apply, their financial system may be described as highly competitive. Japan too has made important steps in the 1980s, removing capital controls at the beginning of the decade, and gradually introducing market alternatives to regulated bank deposits throughout the decade. Developments proceeded more cautiously in France and Italy, with capital controls being removed only gradually throughout the 1980s and rate/quantity and powers regulations still applying fairly extensively over the full sample period used here. While Germany was one of the first countries to remove rate/quantity regulations in the 1960s and 1970s, it has been relatively slow to implement “powers” deregulation. As a result, competition between German banks has remained muted, and short-term financial instruments paying market returns have not been readily available as alternatives to bank deposits throughout the 1970s and 1980s.

Table 5: Financial Liberalisation in the 1970s and 1980s
  Rate/Quantity Deregulation of Intermediaries Powers Deregulation Competition Between Intermediaries Foreign Exchange Deregulation
 
  (Rapid Liberalisation in the 1980s)
United States Mainly in the late 1970s and early 1980s. From the mid-1970s important. Always deregulated in 1970s and 1980s.
Japan Carried out gradually through the 1980s. Gradual introduction of new instruments, mainly in 1980s. For all the 1980s (not 1970s).
United Kingdom Controls widely used until 1980. Being gradually carried out mainly from the mid 1980s. Cartel-like behaviour evident. Removed controls in 1979.
Canada Always deregulated in 1970s and 1980s. Always deregulated in 1970s and 1980s. Always deregulated in 1970s and 1980s.
Australia Controls widely used until early 1980s. Regulations eliminated and foreign bank competition introduced in mid 1980s. Removed controls from 1983.
  (Countries Slow to Liberalise)
Germany Always deregulated in 1970s and 1980s. Strongly controlled and little deregulation in 1970s or 1980s. Cartel-like behaviour evident. Always deregulated in 1970s and 1980s.
France Controls widely used in 1970s and 1980s. Being gradually carried out mainly from the mid 1980s. Cartel-like behaviour evident. Controls widely used and only in late 1980s phasing out begins.
Italy Credit ceilings used until 1983. Ready availability of short Treasury Bills since 1975, but strong regulation of intermediaries. Cartel-like behaviour evident. Highly regulated in 1970s and most of the 1980s – some recent easing.

Source: OECD.

On this basis our sample of countries can be divided into two groups.[8] In the first group, the United States, Japan, the United Kingdom, Canada and Australia are classified as countries that have implemented substantial liberalisation policies. The second group consists of the continental European countries, Germany, France and Italy, which have been much slower to deregulate. Having decided on this separation of countries, the test equations can be estimated for each group of countries as a system using a SURE technique and the instrumental variables for income already described above.

(b) Empirical Results

The results of this estimation procedure for group 1 are displayed in Table 6. No cross-equation parameter constraints are imposed in the top panel, and the results differ from those in Table 1 only to the extent that they take into account possible contemporaneous cross-correlation of residuals. Panel 2 of the table displays the results that emerge when the excess sensitivity parameter is constrained to be the same across countries in each subperiod. The validity of this constraint is tested using a likelihood ratio test. Finally, the joint constraint of equal slopes and intercepts is imposed across countries for each subperiod and again the validity of this constraint is tested using the chi-squared test based on likelihood ratios. This result is reported in panel 3 of the table. The United Kingdom is excluded because the data did not accept the restriction that its λ value was the same as the other countries in the group.

Table 6: Pooled Results: Group 1
(United States, Japan, Canada and Australia)
    1960s 1970s 1980s 1960s/1970s
United States µ′
 		 0.003
(0.001)		 0.002
(0.001)		 0.005
(0.001)		 0.004
(0.001)
  λ
 		 0.42**
(0.16)		 0.43**
(0.11)		 0.01
(0.15)		 0.25*
(0.12)
Japan µ′
 		 0.010
(0.003)		 0.008
(0.002)		 0.005
(0.001)		 0.009
(0.002)
  λ
 		 0.42**
(0.16)		 0.28**
(0.07)		 0.14
(0.08)		 0.30**
(0.07)
Canada µ′
 		 0.003
(0.002)		 0.004
(0.002)		 0.004
(0.002)		 0.005
(0.002)
  λ
 		 0.46**
(0.16)		 0.25
(0.16)		 0.16
(0.12)		 0.14
(0.16)
Australia µ′
 		 0.004
(0.001)		 0.003
(0.001)		 0.002
(0.001)		 0.004
(0.001)
  λ
 		 0.35**
(0.07)		 0.19
(0.14)		 0.15
(0.11)		 0.15
(0.12)
Log likelihood   525.2 536.4 483.7 1012.6
United States   0.004
(0.001)		 0.003
(0.001)		 0.004
(0.001)		 0.004
(0.001)
Japan µ′
 		 0.010
(0.002)		 0.008
(0.002)		 0.005
(0.001)		 0.010
(0.002)
Canada   0.003
(0.002)		 0.004
(0.002)		 0.004
(0.002)		 0.004
(0.001)
Australia   0.004
(0.001)		 0.003
(0.004)		 0.002
(0.001)		 0.004
(0.001)
  λ
 		 0.38**
(0.06)		 0.29**
(0.05)		 0.14**
(0.05)		 0.24**
(0.05)
Log likelihood   525.0 535.4 483.6 1011.8
United States µ′
 		 0.004 0.003 0.004 0.004
Japan   (0.001) (0.001) (0.001) (0.001)
Canada λ
 		 0.47** 0.33** 0.14** 0.34**
Australia   (0.06) (0.05) (0.05) (0.05)
Log likelihood   517.7 532.6 481.8 1002.5

Note: A SURE estimation technique and the same instrumental variables as for Table 1 results are used here. Standard errors are shown in parentheses.

For the other countries in group 1, the excess sensitivity parameters have changed somewhat in value relative to the Table 1 estimates and, as expected, the corresponding sample standard errors have fallen in all cases. The results shown in the top panel show declining λ values in the 1980s compared to either the 1970s or the 1960s for all countries. The revised results for the unit normal tests are shown in Table 7. In contrast to the earlier results shown in Table 2, the decline in λ in the 1980s compared to either the 1970s or to the 1960s is significant for both the United States and Japan. In the case of both Canada and Australia, the value of λ is significantly lower in the 1980s compared to the 1960s.

Table 7: Unit Normal Tests of the Hypothesis of Declining λ Values in Later Relative to Earlier Periods — Pooled Estimates
  Inline Equation Inline Equation Inline Equation Inline Equation
United States 0.05 2.25** 1.25 1.87**
Japan 0.80 1.31* 1.51** 1.56**
Canada 0.93 0.45 −0.10 1.50*
Australia 1.02 0.22 0.00 1.53*
Joint λ 1.15 2.12** 1.28* 3.07**

Note: The test statistics presented in the table are calculated as follows:

where λE and λL are the estimated coefficients for the relevant earlier and later periods and σE and σL are the corresponding variance estimates. Z is approximately normally distributed with zero mean and unit variance for moderately large samples (a condition fulfilled here with 40 observations for most subperiods). The critical values for the normal distribution at the 5 per cent and 10 per cent levels are 1.65 and 1.29 respectively. Z values in excess of these lead to acceptance of the null hypothesis of declining λ's. One asterisk indicates that the null cannot be rejected at the 10 per cent level, and two that it cannot be rejected at the 5 per cent level. The absolute values of Z are presented in the table.

While the constraint that λ be identical across countries cannot be rejected for any subperiod, the additional constraint that the drift parameter also be the same across countries is rejected for two of the four subperiods. Applying the unit normal tests to the jointly estimated λ values in panel 2 of Table 6, significant liquidity constraint relaxation for this group of countries cannot be rejected for the 1980s compared to the 1960s (at the 5 per cent level) nor for the 1980s relative to the 1970s (at the 10 per cent level). However, no significant reduction in liquidity constraints is indicated for the 1970s compared to the 1960s, despite a substantial fall in the magnitude for the group excess sensitivity parameter. Abstracting from issues of statistical significance, and focusing on the magnitude of the common λ estimates for Group 1 (panel 2), the results say that the number of households which experienced liquidity constraints fell from 38 per cent in the 1960s to 29 per cent in the 1970s and to 14 per cent in the 1980s.

For the second group of countries (Germany, France and Italy), shown in Table 8, the pooled individual country results are broadly similar to those in Table 1. However, the likelihood ratio test rejected the imposition of a common λ across countries for the 1980s. None of the remaining constraints can be rejected at the 5, or even 10 per cent level of significance. Therefore, applying the normal tests to examine the significance of changes in the cross-country constrained λ for the 1980s compared to the other subsamples is clearly invalid. The 1970s common λ value is fractionally higher though not significantly different from that of the 1960s.

Table 8: Pooled Results: Group 2
(Germany, France and Italy)
    1960s 1970s 1980s 1960s/1970s
Germany µ′
 		 0.008
(0.002)		 0.003
(0.002)		 −0.003
(0.002)		 0.005
(0.002)
  λ
 		 0.31**
(0.12)		 0.65**
(0.18)		 1.11**
(0.19)		 0.50**
(0.14)
France µ′
 		 0.005
(0.003)		 0.006
(0.002)		 0.003
(0.001)		 0.004
(0.002)
  λ
 		 0.50
(0.26)**		 0.12
(0.19)		 0.25
(0.18)		 0.50
(0.20)**
Italy µ′
 		 0.006
(0.002)		 0.006
(0.002)		 0.005
(0.001)		 0.006
(0.001)
  λ
 		 0.43**
(0.15)		 0.41**
(0.16)		 0.42**
(0.10)		 0.55**
(0.12)
Log likelihood   257.40 409.40 401.20 651.80
Germany   0.007
(0.002)		 0.005
(0.002)		   0.004
(0.001)
France µ′
 		 0.006
(0.003)		 0.004
(0.002)		   0.004
(0.001)
Italy   0.007
(0.002)		 0.007
(0.001)		   0.006
(0.001)
  λ
 		 0.38**
(0.09)		 0.41**
(0.11)		   0.52**
(0.08)
Log likelihood   257.20 407.30   651.80
Germany France µ′
 		 0.007
(0.001)		 0.006
(0.001)		   0.005
(0.001)
Italy λ
 		 0.38**
(0.09)		 0.40**
(0.11)		   0.53**
(0.08)
Log likelihood   257.10 405.70   650.60

Note: A SURE estimation technique and the same instrumental variables as for Table 1 results are employed here. Standard errors are shown in parentheses.

Footnote

See Blundell-Wignall, Browne and Manasse (1990) and further references therein. [8]