Forecasting the Standard & Poor's 500 stock index futures price: interest rates, dividend yields, and cointegration

Abstract

Daily Standard & Poor's 500 stock index cash and futures prices are studies in a cointegration framework using Johansen's maximum likelihood procedure. To account for the time varying relationship(basis) between the two markets, a theoretical forward price series is constructed using interest rate and dividend yield data. Out-of-sample forecasts from error correction models are compared to those from vector autoregressions (VAR) fit to levels and VARs fit to first differences. This comparison suggests that modeling these price series as a VAR in levels and with the inclusion of data on interest rates and dividend yields, rather than any of the other models considered, has the lowest forecast error rate. This has the implications of affecting the way that forecasting and hedging models are constructed.