As mentioned above, our sample includes firms from Hong Kong, Indonesia,
Japan, Korea, Malaysia, the Philippines, Singapore, Taiwan, and Thailand, and the
benchmark countries, Australia and New Zealand. We examine the exposure of firms in
these countries to fluctuations in their currencies against the U.S. dollar, the Japanese yen,
the British pound, and the euro (the deutschmark prior to 1999). The sample extends from
January 1990 through March 2002 and includes an average of eighty firms per country.
The stock return data are taken from DataStream. For each country, we examine
the returns of the largest firms included in the major market index. When possible, we use
the returns from the largest 100 firms. When returns from fewer than 100 firms are
available, we use all of the available firms in the index. The list of firms, and their
industry classification and each market value are given in a file available from the authors.
In Table 1, we report the correlations among weekly observations of the major
currencies against the currencies of each of the countries in our sample. 6 As the table
shows, the correlations are not uniformly high. In particular, the value of every country’s
currency against the dollar shows little correlation with its value against the other major
currencies. This reaffirms that it may be informative to include the exchange rates
separately, rather than using only a trade-weighted exchange rate.7
To estimate exchange rate exposure itself, we add fluctuations in all four currencies
6 7 Graphs of the exchange rates are given in Appendix Figure 1. We might have expected multicollinearity to be an important problem. In that case, one might as well look only at a weighted average of exchange rates, as was done in most of the past work mentioned in note 3. A high degree of multicollinearity makes it impossible to sort out the individual effects of the explanatory variables. Too much multicollinearity (while not biasing the coefficient estimates) causes the standard errors on the individual coefficient estimates to be high, thereby misleadingly leaving the individual coefficients statistically insignificant. One then finds only joint significance, without individual significance. Here, multicollinearity turns out not to be a serious problem; while some of the correlations reported in Table 1 are high, there is still enough independent currency variation that we find widespread significance of the individual currencies.