The role of data limitations, seasonality and frequency in asset pricing models

We demonstrate that the estimates of the Capital Asset Pricing Model (CAPM) parameters significantly differ across samples, which are based on different days of the week (representing different seasons). Our evidence suggests that the “noise” in the data is not an issue. We also show that parameter differences for mixed samples, which contain information on different seasons, are too small to distort statistical analysis. This is so because parameter estimates converge to some values (but not necessarily to the true values) for high frequencies or as the sample size becomes large. Our evidence suggests a possibility that the CAPM may hold empirically if seasonality in the parameters of the population model are taken into consideration.