2004 · 45 citations · 12 references
ComputingBusiness EconomicsFinanceForecastingMachine LearningTime Series AnalysisFinancial ForecastingForecasting MethodsNeural NetworksShort-term ForecastingTime Series ForecastingTime Series PredictionCausalityFinancial Time Series ForecastingModeling NonlinearityNeural ForecastingNeural Time Series ForecastingNonlinear Forecasting ModelsNonlinearityStock Market Returns
Recent studies show that there is a significant bidirectional nonlinear causality between stock return and trading volume. This research reinforces the results presented previously and we further investigate whether trading volume can significantly improve the forecasting performance of neural networks, or whether neural networks can adequately model such nonlinearity. Neural networks are trained with the data of stock returns and trading volumes from standard and poor 500 composite index (S&P 500) and Dow Jones Industry index (DJI). The results are used to compare with those networks developed without trading volumes. Daily data is applied to train neural networks in order to test whether trading volumes can help in short-term forecasting. Directional symmetry (DS) and mean absolute percentage error (MAPE) are both employed to test the result of robustness. Empirical results indicate that trading volume has little effect on the performance of direction forecasting. Sometimes it may lead to over-fitting. For forecasting accuracy, trading volume leads to irregular improvements.
12
A. Ronald Gallant, Peter E. Rossi, George Tauchen · Review of Financial Studies · 1992 · 1.4K citations · Full text