Financial market integration tests: an investigation using US equity markets

Abstract

Abstract The literature on international financial market integration contains conflicting results, in part because there exists no economic benchmark of integration to which the statistical results can be compared. We provide such a benchmark. We subject the NYSE, AMEX and NASDAQ exchanges to a series of integration tests that use a generally accepted multifactor asset pricing (MAP) model. If we accept the premise that these three exchanges are as well integrated as any equity markets are likely to be, then our results provide a benchmark by which to adjust the significance levels used in similar international financial market integration tests. We estimate both a constant and a time-varying coefficient version of the MAP model. Both versions reject the hypothesis that the three US exchanges are integrated. Our interpretation is that these integration tests are too powerful when confronted with the practical realities of integrated markets, and the usual confidence intervals need to be substantially enlarged. The time-varying coefficient model rejects the integration hypothesis three to four times more frequently than it should. Our results can also be interpreted as a specification test of MAP models. The risk premia estimates of the fixed-coefficient model seem to be well behaved, and they do not reveal any obvious misspecification of the model. In contrast, possibly because of the low number of observations in the cross-sectional regressions, the time series of the risk premia exhibit various instabilities and undesirable properties. Finally, we find some evidence that the NASDAQ may not be as well integrated with the NYSE and the AMEX as the latter two are with each other.