RDP 9107: The Cost of Equity Capital in Australia: What can we Learn from International Equity Returns? 5. Estimates of the IAPM

For estimation, the MSCI World index in SDR terms is taken as the relevant market portfolio. A broader market index, including other types of assets, might be desirable: this could be the subject of subsequent work. All analysis is in pre-tax terms. This may be justified by factors like those discussed below in Section 7(b), or by the observation that equities are often held by institutions with relatively low tax rates. The estimation period is the period since the float of the Australian dollar, January 1984 to December 1990. Thus, it includes the stockmarket crash of October 1987. There may be arguments to suggest that the crash was a “Peso-problem” type occurrence and that these data will give undue attention to that episode. However, if memories of such episodes are long, this may not be inappropriate.

(a) Stockmarket Returns

The analysis uses the data described in Section 2, and the accumulation indices for the All Industrials and All Resources produced by the Australian Stock Exchange. Following standard practice, monthly stockmarket returns are calculated as the difference in the logged accumulation series. The following equation is estimated:

where R_w denotes returns in the world market denominated in SDRs, and R_i denotes returns in each of the 11 national markets denominated in their national currency.^[23]

The results are shown in Table 4.^[24] The estimates reveal that individual country returns are explained to a significant extent by world market returns. For some markets such as the US and Japan, this is hardly surprising since those markets constitute a significant proportion of the world market. But returns in smaller countries such as Australia and Sweden are also highly correlated with world market returns. The results also reveal that for all countries, the parameter estimates on the world market term (which will be referred to as the beta estimate) are close to unity.

However, beta estimates may be biased if there are omitted factors which happen to be correlated with the world market return. An obvious candidate for Australia is the influence of commodity prices. To take account of these possible biases, two other factors are included to explain the local currency stockmarket returns, consistent with stockmarkets being determined in an APT framework by a number of factors. These extra factors are variables for metals and oil prices, both measured in SDRs, and expressed as differences of logs.^[25] Metals prices are measured by the Economist index on the Tuesday nearest the end of the month. Oil prices are measured by the price of West Texas Intermediate oil on the New York Mercantile Exchange.

The results from including these extra factors are also included in Table 4. Metals and oil prices are significant explanators for a number of countries, with signs that tend to be consistent with resource endowments. For example, for Australia and Canada, both variables show positive signs; for the Netherlands (home of Royal Dutch Petroleum) oil prices carry a positive sign, while for Japan both variables show negative signs. But as expected, the world stockmarket variable remains the most important explanator. In the Australian market, the resources sector is estimated to have a beta that is above unity, though the difference is not statistically significant. One reason why the beta estimate for the resources sector rises following the inclusion of other factors could be that oil prices may have affected the world stockmarket. When oil prices rise, as in August and September 1990, the world market may fall, but the resources sector will be less affected and may even benefit. Thus the energy sector may appear to have low or negative covariance with the world market at such times, but this effect is removed by taking account of the other factors.

But, as discussed in Section 3, one determinant of the observed betas in each country should be the degree of leverage in that market. That is, returns in countries which have higher debt/equity ratios might be expected to show greater volatility. So, to draw inferences as to whether or not the underlying risk (i.e. the asset beta) of a particular national market is greater or less than in other countries, we should try to take out the effects of differing debt/equity ratios across countries.

One problem is that the debt/equity ratio used will ideally be a forward-looking one, since future financing decisions will affect the risk of future cash flows. However, there is no good indicator of such intentions. (One reason why firms might not announce future issues would be that it may increase the cost of raising funds in the current period.) Hence, we must be satisfied with using observed debt ratios as an approximation. Another problem is that debt/equity ratios have varied significantly in my estimation period. Poterba (1991, p. 28) shows that Japanese debt/equity ratios fell significantly through the second half of the 1980s, while US ratios rose. In addition, Australian debt/equity ratios also tended to rise somewhat over this period. As a result of these changes, the use of debt/equity ratios from any single year will be open to dispute, but the use of the middle year of the sample may be the least arbitrary choice. Debt/market value ratios for 7 countries in 1987 from Borio (1990, p. 11) and EPAC (1990, p. 18) are used. Corporate tax rates for foreign countries are obtained from Borio (1990, p. 20). Based on these, and equation 5 above, we can estimate the unlevered (or asset) betas that are implied for each country. After normalisation to unity, these are shown in Table 5 below, along with their transformed standard errors.^[26]

As can be seen, after taking out the effects of leverage, the implied unlevered (or asset) betas estimated for Australia are between 1.5 and 2 standard errors greater than unity. The reason for this is straightforward: if Australian equities have lower debt ratios than other markets, but show average volatility (as measured by levered betas), it follows that they would show greater than average volatility if they had greater use of debt. Hence there is some evidence that Australia is a risky country in the CAPM sense.

There may be a number of reasons why asset betas might be higher in Australia than overseas. The obvious one is if there is more market or cyclical risk in the Australian economy and Australian stockmarket than in other countries. The Australian market has, for example, more resource stocks than most other countries. These are relatively risky, as the asset beta estimates in Table 5 indicate. But the Australian industrials sector also appears to have a relatively high asset beta. This may reflect the fact that Australia has fewer stocks in some low-beta sectors such as consumer goods and services, and utilities, which tend to be government-owned in this country but are often publicly listed in other countries. In addition, industrial stocks in Australia are probably affected somewhat when the resource sector suffers.

A further reason could be the particular arrangements in some other countries (notably Japan and Germany) by which banks have equity holdings in firms whose debt they also hold. As debt-holders, they may make concessions at times when the firm is in trouble. Thus, there is a case for arguing that some of what is measured as “debt” in these countries, is more like equity. Thus, measured debt/equity ratios may be overstated a little, and unlevered betas (especially in Germany) may not be quite as low as my figuring suggests. On the other hand, these financial arrangements may reduce the risks of bankruptcy, and may make equity safer. As a result, for those equity-holders who are not also debt-holders, Japanese and German stocks may still be relatively low-beta investments.

What are the implications if asset betas are higher in Australia than overseas? According to the CAPM, it is the asset beta that is the primary input into the required rate of return on an asset. Indeed, if we assumed a certain value for a world equity premium, we could estimate the effect on the cost of equity in Australia. Estimates for particular countries often put the equity premium at something over 6 per cent. However, there are many who are surprised by the magnitude of this historical risk premium, hence the growing literature beginning with Mehra and Prescott (1985) trying, but failing, to explain it using theoretical models of asset pricing. Assuming for illustrative purposes a required world premium of 4 per cent, we can simply multiply by the estimated asset betas for each country to get an estimate of the equity premium that might be observed in each country if all countries had average debt/equity ratios. This rough figuring would suggest that the equity premium might be about 5.2 per cent in Australia or 1.2 per cent above the average world equity premium. Within this total, the industrial and resource sectors would be estimated to have equity premia of around 5.0 per cent and 5.7 per cent, respectively. This figuring should be considered as indicative only, but it does not seem implausible.

(b) Exchange Rate Risk

Since investment in a particular national equity market will always be denominated in the currency of that nation, the riskiness of each market to a foreign investor will also depend on the risk of that national currency. And if the relevant measure of risk for any asset is the correlation of the asset's returns with the world market return, we must examine whether exchange rate returns in any country are correlated with the world stockmarket return.

Table 6 contains estimates from regressions explaining exchange rate changes by the world market return and changes in a number of commodity prices.^[27] The data for exchange rate changes are measured as SDRs per unit of domestic currency so that increases correspond to appreciations. These other factors are consistent with exchange rates being determined by an APT model, and were selected with particular reference to Australia. Again I include metals prices and oil prices, as well as an index of rural prices.^[28]

As might be expected, exchange rate movements are far less well explained than local stockmarkets. For Australia, oil, metals and rural prices are all significant explanators of the exchange rate, consistent both with natural resource endowments, and some other empirical work.^[29] For other countries, rural prices are estimated to have larger effects than seems reasonable, casting some doubt on the estimates. The regressions also suggest that world stockmarket returns have only weak explanatory power for exchange rates. However, they suggest that the exchange rates of Australia and Canada are positively correlated with the world stockmarket return, while the exchange rates of some European countries (roughly speaking, the Deutschemark bloc) appear to be negatively correlated with this measure of the world stockmarket.

While world stockmarket movements are not generally included as regressors in exchange rate equations in Australia,^[30] there seems to be no reason why they should not be. In particular, world stock returns are measured very precisely and contain significant information about expected future outcomes in the world economy. This approach may not be too much at odds with the usual practice of explaining exchange rates in terms of largely domestic factors (e.g. domestic interest rates), since it may well be that most domestic factors (especially in a small economy such as Australia) have international causes. In addition, to the extent that some purely domestic factors do impact on the exchange rate, they may be diversifiable for the typical world investor, and if so, are of little concern. As for the criticism that theoretical exchange rate models provide no role for variables such as stockmarket returns, it should be remembered that such models have not proved especially robust from an empirical point of view. Exchange rate markets often seem to be driven by sentiment: it may be that a variable that measures the performance of world equity markets can capture some of these factors.

But the relevant point for this paper is that the estimates above provide some weak evidence that the Australian exchange rate is correlated with world stockmarket movements. That is, the Australian dollar may be something of a “fair weather” currency: this may not be much of a surprise to many. And since foreign investors can only invest in the Australian stockmarket by incurring Australian exchange rate risk,^[31] this apparent “exchange rate beta” may require an additional risk premium before foreign investors will hold Australian equities.^[32] Similarly, Australian investors may find it more favourable to invest overseas if exchange rate risk provides some insurance at times when stockmarkets fall. Again, this finding is preliminary, but it seems relatively plausible.

Footnotes

Strictly speaking, the CAPM and the IAPM are based on excess returns, i.e. returns above the risk-free rate. In other work I have estimated the results in Table 4 and Table 6 using excess returns (over a weighted average short-term rate), and obtained almost unchanged results. In this period, the estimated standard deviation of monthly world stockmarket returns is 52 times greater than that of the monthly risk-free rate, so it is hardly surprising that the stockmarket component dominates other movements. [23]

For brevity, diagnostic statistics for these equations have not been provided. Note, however, that Durbin-Watson coefficients for the estimates tend to be quite close to 2, so I have not looked further at any dynamic adjustment process. On this point, Solnik (1988, p. 42) notes: “Some investigators have attempted to find leads or lags between markets. However, no evidence of a systematic delayed reaction of one national market to another has ever been found. The existence of such simple market inefficiencies is, indeed, unlikely, since it would be so easy to exploit them to make an abnormal profit.” [24]

In preliminary regressions, a rural commodity price variable was also included, as with the exchange rate results, but it was not significant for any country. [25]

The unlevered beta estimates used in these calculations are from the multi-factor rather than the one-factor regressions. The implications of this choice are discussed below in Section 7(a). The debt/market value ratios for the industrial and resource sectors are my own estimates, and are based on Statex data for these sectors for 1987, adjusted in line with the EPAC number for the Australian market as a whole. [26]

Note that I do not attempt to divide price movements into “expected” and “unexpected” components. Since all variables are financial prices, measured on an end-period basis, any “expected” component would represent a profit opportunity. [27]

The index of rural prices is based on Australian export weights and end-month SDR prices for wheat, beef and sugar. Wool was excluded since the reserve price scheme (which was set in Australian dollars) could induce spurious correlation with the exchange rate. [28]

For example, Macfarlane and Tease (1989) also find evidence that the Australian dollar is affected by commodity prices. They find that the response to commodity prices is greatest when the Australian dollar is measured against the Deutschemark. In effect, they are estimating the effect on two currencies: these results suggest that the DM responds negatively to commodity prices, which explains their finding. [29]

An exception is Cosset (1984) who finds no stable role for such effects in data for the period March 1973 to February 1980. One reservation about Cosset's work is that all exchange rates are measured against the US dollar. However, if the US dollar behaves perversely, this method would imply that all other countries' exchange rates do as well. Using SDR rates (or some other average) seems more sensible. [30]

It is, of course, possible to use forward markets to perfectly hedge foreign currency cashflows that are known with certainty. However, equity returns are uncertain (and fairly volatile) and therefore cannot be perfectly hedged. In any case, because hedging involves persuading someone else to bear risk, even a risk that can be perfectly hedged will attract a risk premium if it is not diversifiable. [31]

The evidence of correlation between the Australian dollar and the world stockmarket implies that debt denominated in Australian dollars could carry a risk premium as well. Smith and Gruen (1989) find evidence that Australian risk-free assets yield higher returns over recent years than foreign risk-free assets, but are unable to explain the difference in a consumption-CAPM (CCAPM) framework. If the equity premium on the world market was 4 per cent, my estimates of a beta of 0.13 could imply a risk premium from this factor of around half a percentage point. Thus, the IAPM might be a better explanator of exchange rates than a CCAPM model, just as Mankiw and Shapiro (1986) have shown that the CAPM outperforms the CCAPM in the US equity market. [32]