Abstract

As a consequence of market globalization, it is now harder than ever before accurately to attribute portfolio risk to country and sector positions. The increase in industrial concentration with the growth of global stocks has led to a biased sector composition in many smaller markets. This makes it very difficult to separate the impact of domestic market movements from global industry effects. This is particularly so when a few stocks can account for a large proportion of the market capitalization of a single country or region. At the portfolio level this can lead to overestimating the contribution of country factors to portfolio risk.

Here we present a new solution to this problem. Roll (1992) was the first to examine industry composition bias in local market indices; though Heston and Rouwenhorst (1994, 1995) were later to question some of his conclusions. Using the approach of Grinold et al. (1989), he estimated a set of industry-neutral country indices by regressing observed returns on a set of dummy-indicator country and sector variables. However, this approach makes some strong assumptions about the relative sensitivity of the asset returns to these factors. We therefore propose an iterative algorithm. The estimated indices are used then to estimate the asset return sensitivities which, in turn, are used to re-estimate the factor indices. By imposing a set of identifying restrictions on these regressions, we ensure, first, the algorithm converges and, second, the solution has an obvious interpretation.

We estimate our model on...