Nonlinear Model Improves Stock Return Out of Sample Forecasting (Case Study: United State Stock Market)

Improving out-of-sample forecasting is one of the main issues in financial research. Previous studies have achieved this objective by increasing the number of input variables or changing the kind of input variables. Changing the forecasting model is another possible approach to improve out-of-sample forecasting. Most researches have focused on linear models, while few have studied nonlinear models. In the present study, we have reduced the number of variables and at the same time applied a nonlinear forecasting model. Oil prices have been used as predictors to predict return by application of a new artificial neural network nonlinear model named Deep Learning and its comparison with OLS and ANN methods. Results indicate that the applied non-linear model has higher accuracy compared to historical average model, OLS and ANN. It also indicates that out-of-sample prediction improvement does not always depend on high input variables numbers. On the other hand when using a smaller number of input variables, it is possible to improve this forecasting capability by changing the model and applying nonlinear models.