Size and stock market integration: a study of Canadian firms

Abstract

In this article the restrictions imposed on excess returns by a dynamic optimization model are tested on stock market data from the Toronto Stock Exchange (TSE), from which ten size-portfolios have been formed. The model implies that all excess returns should move proportionately if assets are perfectly integrated. The restriction that all size portfolios are governed by one single latent variable is rejected over the sample period 1961–2002. It is established that this rejection is due to the presence of the smallest size portfolio, especially during the second half of the sample period. The uncertainties of the late 1980s and 1990s appear to require the presence of a second latent variable. No definite conclusions can be drawn regarding these sources of risk even if the return on the market portfolio and exchange rate fluctuations play an important role.

Acknowledgements

I am grateful to Benoît Carmichael and Alain Cohen for giving me access to their data from the CFMRC database and to Christine Hébert for her competent research assistance.

Notes

¹Doidge et al. (2006) find for example that exchange rate movements can have a significant impact on firm value, and that this impact differs according to the amount of international trade. Financial innovations are also found to impact firms differently according to size in Christoffersen et al. (2006).

²For simplicity, the βs are assumed constant. This is supported by Ferson and Harvey (1991) and Evans (1994) who found that the time variations of asset returns seem to come mostly from the variables in X_t and very little from the factor loadings themselves.

³In an alternative specification, the exchange rate variable was replaced by the return on the Dow Jones. The model's restrictions were rejected more strongly even if the variable was statistically significant. The relatively high correlation between the Canadian and the US returns, at 0.7, also seemed to cause some problems in identifying the factor loadings associated with each variable.