Portfolio size, non-trading frequency and portfolio return autocorrelation

• We re-examine the relationship between non-trading and portfolio autocorrelation.
• We derive an adjustment to the Lo and MacKinlay model for small portfolios.
• In small portfolios, security specific effects can reduce observed autocorrelations.
• But, security heterogeneity within the portfolio can offset this.

In this paper we re-examine the relationship between non-trading frequency and portfolio return autocorrelation. We show that in portfolios where security specific effects have not been completely diversified, portfolio autocorrelation will not increase monotonically with increasing non-trading, as indicated in Lo and MacKinlay (1990). We show that at high levels of non-trading, portfolio autocorrelation will become a decreasing function of non-trading probability and may take negative values. We find that heterogeneity among the means, variances and betas of the component securities in a portfolio can act to increase the induced autocorrelation, particularly in portfolios containing fewer stocks. Security specific effects remain even when the number of securities in the portfolio is far in excess of that considered necessary to diversify security risk.