| This thesis examines financial market integration in North-America from January 1984 to December 2003, using two basic CAPM and APT test models. We introduce a methodology valid in finite samples for the CAPM model. A pivotal statistic is introduced to correct for the so-called dimensionality curse which affects the critical points of the LR test statistic under the null hypothesis. When using this methodology, the null hypothesis of integration is strongly rejected for all sub-periods. Our results differ from those obtained in previous studies such as Mittoo (1992) using an asymptotic methodology. Next, an APT model with pre specified factors is used in order to test the null hypothesis of integration. The factors used are the Fama and French factors. In the latter two-pass test context, we introduce a split sample methodology in order to correct for the pre-estimation of BETAS. Moreover, we consider (and form) Fama and French factors for Canada for the 1984--2003 period. With this methodology, we again strongly reject the hypothesis of integration except for two sub-periods. Fama and French factors appear to have a different effect on the Canadian and American stock returns. |
