Long memory and outliers in stock market returns

Long memory in the form of fractional integration is analysed in stock market returns. Special emphasis is placed on taking into account the potential bias caused by neglected outliers in the data. It is first shown by a simulation experiment that outliers will bias the estimated fractional integration parameter towards zero. In a monthly data set, consisting of stock market indices of 16 OECD countries, statistically significant long memory is found for three countries. In one of these long memory is only found when outliers are first taken into account.