How Much In-Sample Data to Use in Forecasting? Evidence from a Simple Stock Returns Model

10 Pages Posted: 14 May 2009

Date Written: May 14, 2009

Abstract

Using a simple and well-established model for predictive power this letter assess how much in-sample data is required to obtain good out-of-sample forecasts. Specifically using the present value predictive model for monthly stock returns we conduct a backward recursive exercise where the out-of-sample period and the end of the in-sample period are held constant but the start of the in-sample period is rolled backwards. Using RMSE measure for eight international markets results show that in-sample periods of between ten and fifteen years produce reasonable forecasts across markets and forecast horizons.

Keywords: Dividend Yield, Returns Predictability, Forecasting, Backward Recursion

JEL Classification: C22, G12

Suggested Citation

McMillan, David G., How Much In-Sample Data to Use in Forecasting? Evidence from a Simple Stock Returns Model (May 14, 2009). Available at SSRN: https://ssrn.com/abstract=1404625 or http://dx.doi.org/10.2139/ssrn.1404625

David G. McMillan (Contact Author)

University of Stirling ( email )

Stirling, Scotland FK9 4LA
United Kingdom