Testing international asset pricing models with mutual fund data

The paper is a first attempt to test international asset pricing models with mutual funds as assets by incorporating time-varying betas and risk premia in a straightforward way. Funds that invest internationally should be priced according to international asset pricing models. A diverse set of Swiss and U.S. mutual funds with an international investment policy is employed to test some of these models and explore the major sources of risk and their variation over time.;Following the reasoning of Adler/Dumas (1983) I include currency positions along with the world wealth portfolio in the list of potentially priced factors. I argue, that in light of empirical evidence on the predictability of currency returns, currency positions can only be important factors, if their premia are allowed to vary over time, since average currency returns are very close to zero. By modeling the time-variation in fund betas the approach in this paper accommodates movements in conditional moments and risk premia. A Bayesian Variable Selection technique is used to explore parsimonious versions of the general model.;I find that both currency factors and time-variation in a few betas are statistically and economically important. Currency betas are particularly important to price bond funds. However, pricing errors are large and standard multivariate tests of a zero intercept vector reject both unconditional and conditional models. Only within the set of U.S. mutual funds am I unable to reject the conditional models in a version that incorporates the Italian Lira and the corresponding currency forward premium.;Sensitivity checks on both managed and passive assets indicate that the rejection of conditional models that allow for time-varying betas is partially due to estimation problems for a small subgroup of funds and equity markets which are very volatile and not well-diversified. Large beta estimates for time-varying factors lead to large pricing errors for these assets.;The shrinkage property of the Bayesian variable selection approach tends to reduce extreme beta estimates and thereby eliminates many cases of high average pricing errors. However, this technique also eliminates most of the conditional currency betas.;While conditional beta models improve time series explanations of international asset returns, they don't seem to compensate for the poor ability of the World market and currency factors to explain cross-sectional differences in international asset returns.