RDP 2001-06: The Effect of Macroeconomic Conditions on Banks' Risk and Profitability 4. The Profitability of Australian Banks 1960–1999

A weakness in the preceding analysis is that data are available for less than one full cycle in banks' credit quality. To analyse the effect of macroeconomic variation on banks over the longer term, profitability data, specifically, the after-tax return on assets earned by banks is considered.

Whilst banks' return on assets is influenced by their credit risk, the relationship between the two is not straightforward. Movements in the return on assets will reflect not just credit risk, but the full range of risks, including banks' exposures to movements in interest rates and exchange rates, liquidity risk and operational risks. Moreover, banks' return on assets reflects not just risk-taking, but also other factors such as the mix of on- and off-balance sheet business, operating efficiency, the level of competition within the banking market and regulatory constraints. The relationship between risk and return also depends upon whether banks price for risk, and the lags between taking on risk and the crystallisation of risk into realised losses. To the extent that banks earn higher returns by taking on riskier business, this will boost the return on assets. However, if a bank experiences losses beyond what it had provisioned for, such losses will reduce profitability. Over the 1990s, the return on assets and the impaired assets ratio exhibited strong negative contemporaneous correlation.

4.1 A Simple Comparison of Firm-specific and System-wide Variation

Figure 5 shows movements in the average level and distribution of asset returns since 1960.^[8] It can be seen that the credit problems of the early 1990s resulted in the largest losses in forty years. Up until the mid 1970s, there was little movement in banks' asset returns and the distribution of returns was quite narrow. Smaller banks were, however, more profitable than the larger institutions (the unweighted mean being around 0.3 percentage points higher than the asset-weighed mean until 1975).

From the second half of the 1970s, financial liberalisation allowed banks to increase their return on assets. The dispersion across banks also widened considerably. Between 1960 and 1975 the interquartile range averaged 0.3 percentage points; since then it has averaged 0.6 percentage points.^[9] The difference in returns across banks, however, became less influenced by bank size. On average since 1980 there has been little difference between the unweighted and asset-weighted means.

Decomposing the panel variance using analysis of variance techniques outlined in Section 3.1 indicates that while interbank variation in profitability exceeds through-time variation, most of the variation lies in the residual (Table 4). In contrast to the impaired assets data discussed above, when the asset-share weights are applied the share of interbank variation in total panel variation increases substantially. Although several of the smaller banks made losses in 1990, which caused the fall in the unweighted mean, in terms of the overall variability of the panel, this is outweighed by the large losses made by large banks in 1990, 1991 and 1992.^[10] The share of variability through time also falls when the asset weights are applied, indicating that the smaller banks' profitability tends to be more variable than the larger banks'.

4.2 Banks' Return on Assets and the Macroeconomy

Demirgüc-Kunt and Huizinga (1999), in their cross-country comparison of bank profitability, find that higher real interest rates and, to a lesser extent, higher growth in real per capita GDP are associated with stronger bank profitability. While their finding of a positive relationship between per capita output and bank profitability is consistent with other studies that focus on banking risk, the positive relation between profitability and real interest rates runs counter to the findings of the risk-based studies. The results can be rationalised, however, by the fact that in periods of high interest rates banks are often able to earn higher interest rate spreads as well as running higher risks (Reserve Bank of Australia 1999).

Figures 6 and 7 present the weighted-average return on assets and macroeconomic indicators of financial system stability. The longer run of data highlights the increase in the share of interest payments in corporate income during the 1980s, which was a product of a substantial increase in the corporate sector's gearing, and (in the second half of the 1980s) high real interest rates.^[11]

The slowdown in real GDP growth in 1990 coincided with the banks' losses in that year. The recession of 1983, however, was accompanied by only a slight reduction in banks' profits. Similarly, while the share of construction in GDP reached historically high levels in 1989–1990, peaks around the same level in 1971 and 1982 occurred in periods when bank profits were stable or rising.

The dramatic peaks in residential property price inflation in 1974 and 1989 preceded periods of contraction in real credit. It was only the later episode, however, that had any marked impact on bank profitability. The muted response in 1974 reflects the impact of the close regulation of banks during the 1970s.

To more closely quantify the effect of macroeconomic variation on banking profitability, the panel regression analysis presented in Section 3.3 is repeated taking the return on assets as the dependent variable (Table 5).^[12]

Like the impaired assets data, the annual profitability data provide an unbalanced panel. Of the 21 banks included in the panel regression, only 6 were in operation throughout the whole period. The banks included in the panel (and the years for which annual reports were available) are listed in Appendix A.

Again, a trend term is included. In each model the trend effect is small but significantly positive. Here the trend is taken to proxy the impact of deregulation. A trend term, rather than a distinct structural break, is included as the process of deregulation was a gradual one. Restrictions on bank interest rates and lending policies were progressively eased between 1973 and 1986, allowing banks to expand into new, more profitable, areas of business, and allowing banks greater control over their interest margins (Battellino and McMillan 1989).

Consistent with the negative correlation between impaired assets and profits during the 1990s, the coefficients on all variables have the opposite sign to that found in the models of impaired assets. This would suggest that banks have not fully priced for risk by increasing margins as risk increases. Rather than increased risk resulting in higher bank profits, higher risk has reduced bank profitability.

Consistent with the results from the first model of impaired assets shown in Section 3.2, the share of interest in corporate income and real credit growth display a strong relationship with bank profitability. Each percentage point increase in the share of interest in corporate income is estimated to reduce the return on assets by around 0.01 percentage points.

Similarly, the return on assets is predicted to fall by 0.03 percentage points for each percentage-point acceleration in real credit growth. Accelerated real credit growth is found to reduce bank profits with a lag of one year. Rajan (1994) and Calomiris et al (1997) predict that such a response should reflect a long-run, rather than short-run, effect of rapid credit growth on banks' risk. That the observed response is relatively quick largely reflects the rapid acceleration in credit growth during the late 1980s preceding the loan loss problems of 1990–1992. The impulse response functions shown in Figure 8 demonstrate that the strong relation between real credit growth and bank profits combined with the high variability of real credit growth (particularly during the mid 1970s) has seen real credit growth have a large effect on banks' return on assets.

The second model shows that real interest rates exert some influence beyond their effects on corporate gearing. In line with Diamond's prediction that higher real interest rates increase the likelihood of borrower defaults, increases in real interest rates reduce bank profitability: each percentage point increase in real interest rates lowers profitability by around 0.02 percentage points (this relationship, however, is not strongly significant). The negative relation found between bank profits and real interest rates runs counter to Demirgüc-Kunt and Huizinga's findings, suggesting that in Australia's case the effect of high real interest rates in impairing banks' credit quality has outweighed banks' capacity to earn higher margins. In contrast to the results presented in Section 3.2, real GDP growth did not exert a significant influence on banks' profitability beyond its effect on the share of interest in corporate income.

The third model shown in Table 5 combines gearing and credit growth measures with property price inflation. The importance of property as collateral underlying secured bank lending is borne out by the positive relation between residential property price inflation and bank profitability: a one percentage point increase in property price inflation is found to lead to an increase in banks' return on assets of 0.03 percentage points. This effect, however, is seen to take several years. Although the share of construction in GDP was found to have a strong effect on impaired assets during the 1990s, over the longer period it did not have a significant influence over the return on assets.

While all three models leave most of the variation in the return on assets unexplained, extending the first model does not greatly improve explanatory power. In addition, extending the first model does not greatly change the estimated coefficients on corporate gearing and credit growth.

Footnotes

Profit figures are adjusted to exclude government assistance provided to the State Bank of Victoria in 1990 and State Bank of South Australia in 1991. [8]

Excluding the four years 1989–1992, the interquartile range has averaged 0.5 percentage points since 1976. [9]

When the State Bank of Victoria failed in 1990, and when the State Bank of South Australia failed in 1992, each was the 5th largest in the industry at the time. In 1992, Westpac, then the largest bank, and ANZ, ranked third, both reported large losses. [10]

Since a robust measure of commercial property price inflation over the 1960s is not available, residential property price inflation is considered in its place (during the 1990s the two price measures broadly moved together). In addition, data on the share of interest payments in the household sector's income are not available prior to 1973 so the household sector is not included in the analysis that follows. [11]

As in the analysis of impaired assets presented in Table 3, the choice of fixed effects is supported by the apparent absence of heteroscedasticity across banks. The Breusch-Pagan test accepts (at the one per cent significance level) the hypothesis that the variance of each model's residuals does not differ across banks. Unlike the previous models, however, the Breusch-Pagan test accepts the hypothesis that the variance of each model's residuals differs across time periods. The return-on-assets models, therefore, are estimated using estimated generalised least squares to adjust for this form of heteroscedasticity (Judge et al 1988). [12]