It is tempting, when developing models to analyse risk and return in global equity portfolios, simply to modify an essentially local model. Our aim is to develop an explicitly global model, giving preference wherever possible to global over local effects. We also attempt to minimise prior assumptions on likely stock exposures. We reject the standard single market, single industry, single exchange rate classification in favour of a more general approach, which seems more consistent with the reality of large multinationals, operating in a number of countries and industries.

Our model gives preference to global factors in the decomposition of returns. We have used it to demonstrate that global influences account for a major part of the returns to stocks and markets. The covariance structure of global portfolios is complex and we have demonstrated that the conventional method of market, exchange rate and industry classification is an oversimplification. The majority of companies have statistically significant betas against more than one market, and many have multiple industry and exchange rate exposures.

We also find some evidence of regional effects among our local market residuals. We find strong evidence of global industry effects, and whilst their impact varies from country to country they constitute an important source of risk (and return).

First we use a standard model of global equity risk and return simply to allow a contrast with the alternative model we developed. This standard model is characterised by its initial focus on the local relationship between stock return and market return, and its subsequent use of global factors to explain part of the residual covariance.

We have built a model that seeks to give pre-eminence to global influences. In addition, we try to impose restricting assumptions only where these are consistent with the likely fundamental characteristics of global stocks. Thus global factors are not restricted to operating only as part of the local market residual. We do not impose single market, exchange rate and industry exposures, nor are betas restricted to unity.

The standard model is still a local model, extended to the global context. The only non-zero market beta is against the local market. This is determined before other effects are estimated. Thus global factors operate only on the residuals. Even there, when global industry betas are restricted to a single industry, binary zero returns will tend depend on which effect is measured first. Since we are looking to build an explicitly global model we give priority global factors the best chance of being influential.

The residual terms estimated represent the pure" influence of the world market, global industries and local markets. The order of decomposition clearly loads as much explanatory power (ie influence) into the world and global industry effects as possible. In effect it says that the return to a stock is due first to currency return and second simply to an exposure to global equity in general. Any residual return is due first to global industry returns. Anything left must be due to local effects.

We include the world market because the correlation structure of global equity returns points strongly to the existence of an important factor in returns, which we deem to be a world market effect. The inclusion of this factor is therefore consistent with observed patterns of equity returns. Its inclusion also significantly reduces the correlation between the global industry factors, which makes interpreting industry factor betas easier.

The model is estimated on 13,000 stocks from 23 markets, although we focus on the largest 10-15 markets since these dominate in capitalisation terms. We use eight global industry definitions: energy, utilities, transport, consumer goods, capital goods, basic industries, resources and financials. Monthly data from January 1987-Decembe r 1993 were used.

This model allows for multiple exposures to markets, exchange rates and global industries. We can test for the number of statistically significant factor betas on a stock-by-stock basis. What we want to know is whether the conventional use of single market, exchange rate and industry exposure is consistent with our empirical analysis. Table 1 shows the percentage of stocks with a given number of statistically significant factor betas. Thus, for instance, out of the 400 securities analysed in Canada 10% have statistically significant exposure to any market residual term. However 66% have exposure to two or more market residuals. In fact, across the larger markets shown the majority of stocks have two or more market exposures. Thus most stocks are influenced not only by their local market, but by other "foreign" market residuals. This is not surprising given the widespread occurrence of cross-border operations. From this we conclude that single market classification of stocks is not consistent with the empirical evidence in this explicitly global framework.

Similar conclusions can be drawn about single exchange rate and industry classification. The occurrence of multiple exposures is lower than that found for markets, but for a larger number of stocks single currency and industry classification is not consistent with the evidence from this model. Overall we conclude that multiple exposure to markets, exchange rates and global industries is highly pervasive, that this is consistent with the likely fundamental characteristics of stocks and that it is likely to be concentrated among larger securities.

A question of great interest to investors in global markets is the relative importance of different sources of risk and return. The block separable nature of our model allows the clear separation of contributions to variance from markets, industries, exchange rates and specific effects. That is, since each of the main factor groups are orthogonal with respect to each other their impact on stock or portfolio returns is quite separable.

Table 2 shows the percentage contribution to market risk and return from the main factor blocks, and allows the magnitude of these effects to be compared with the local market residual. The global factors account for a significant part of equity market returns. The combined influence of the world market factor, exchange rates and global industries explains between a low of 52.7% of returns (Hong Kong) and a high of 96.4% of returns (Japan). In between the UK is very dependent on global effects, with these accounting for 81.5% of returns, while the figure for the US is 68%. Note that since we are working from a US dollar base there is only a small amount of currency influence, and that this accounts for the relatively low figure. The influence of the world market and of global industries is in fact relatively high.

We are dealing with capitalisation-weighted indices, so it is likely that the degree of dependence on global effects will be a positive function of market capitalisation. This presents no problem since from a global perspective big countries simply constitute a bigger slice of the world than small ones. Nevertheless it is of interest to examine the relative dependence of markets on global influence in the context of their relative capitalisation - is a market particularly global given its size - and we explore this question below.

The relative power of the three factor blocks varies between countries. For instance the UK, Netherlands and Australia are particularly exposed to industry effects; Japan, Australia and Canada are particularly exposed to exchange rate movements.

Table 3 shows diagnostic data on the standard deviations of both the unadjusted local market and the local market residual. It also shows the correlation between the two series in each country. In effect the correlation shows how globally integrated a market is - a low number indicating that much of the return to the localmarket is attributable to global influences.

All of the market residual returns have lower standard deviations than those of the original local series. This is the point of this type of model: some part of local returns is attributed to factors other than the local market. The final column shows capitalisation on a ranked basis.

Plotting the correlation between local market and local market residual returns against the log of capitalisation shows, as expected, a relationship between capitalisation and the extent of global dependence. Equally, however, there is wide variation in the extent of global dependence.

Edward Fishwick is director of research at Quantec in London"