Resumen de The predictive power of quarterly earnings per share based on time series and artificial intelligence model
The purpose of this study is to compare the forecasting ability among an Autoregressive Integrated Moving Average (ARIMA) model, Transfer Function (TF) model, Artificial Neural Network (ANN) model and Genetic Algorithm (GA) model. To evaluate forecasting accuracy, two dimensions are taken into consideration: (a) deviation between an actual quarterly Earning Per Share (EPS) value and forecasted quarterly EPS value, and (b) direction changes from quarter to quarter between an actual quarterly EPS value and forecasted quarterly EPS value. Both the quarterly basic EPS (BEPS) and diluted EPS (DEPS) data were applied in order to forecast the future quarterly basic EPS. Empirical results have shown that the TF model outperforms the ARIMA model. Therefore, the time lags setting of the TF model is adopted in the other two models: GA and ANN. The empirical results reveal that the GA model has the best forecasting accuracy under both BEPS and DEPS, while the ANN model has been shown to have the worst forecasting accuracy under both BEPS and DEPS. In addition, there is not enough evidence to support that the using of diluted EPS data would yield higher accuracy than that of using basic EPS data in the aspect of deviation.
