The cointegrating relationship between stock prices and trading volume: Evidence regarding the predictability of security returns.

Abstract

This study develops and tests the hypothesis that stock prices and trading volume are influenced by the same set of fundamental forces. The implications of this hypothesis for modeling and forecasting stock returns are also explored. Part 1 identifies several effects likely to contribute to the observed positive correlation between stock prices and trading volume. Among these are the various constraints that prohibit certain classes of investors from short selling; the disposition effect, which is the tendency for investors to hold losing investments too long and sell winners too early; and the prevalence of positive feedback trading strategies in financial markets. Part 1 also presents a simple supply and demand example which demonstrates that both asset prices and trading volume are influenced by the information signals received by traders in efficient markets. Part 2 presents empirical tests of the hypothesis that stock prices and trading volume are determined by a set of common factors. The presence of a common stochastic trend (cointegration) is shown to be consistent with the above hypothesis. Stock prices and trading volume are found to be cointegrated, which is interpreted as evidence in support of the common factor hypothesis. The theoretically correct method for modeling cointegrated variables, known as an error-correction model (ECM), explains over 4% of the variability in monthly stock returns from 1962-1991. An index of the total dividends paid to the Standard and Poor's 500 is included as an instrument for the information set hypothesized to be a common factor in stock prices and trading volume. After demonstrating that stock prices, trading volume and the dividend index are part of a trivariate cointegrated system, the dividend index is included into the ECM of stock prices. The explanatory power of the ECM rises to 6.5% due to the inclusion of the dividend index. Part 3 develops a forecasting model of monthly stock returns based on the ECM presented in Part 2. Out-of-sample forecasts of monthly stock returns are generated from this model, as well as other forecasting models chosen from the literature on the predictability of security returns. Under a variety of conditions, both with and without transactions costs, trading rules generated by the forecasting model of Conrad and Kaul (1989) consistently outperform a buy and hold strategy. The ECM forecasting model performs no better than a macroeconomic or random walk model, and underperforms a buy and hold strategy in the presence of transactions costs. The overall finding from Part 3 is that the simple time series model of Conrad and Kaul generates forecasts which beat the market conclusively for the thirteen year period spanning January, 1978 to December, 1990.