IMPROVING EARNINGS PREDICTIONS WITH NEURAL NETWORK MODELS

In this paper we develop a generalized deep neural network model to predict quarterly earnings. Using a diverse range of predictors consisting of fundamental, technical and sentiment data the resulting model outperforms existing timeseries models such as the Fama-French 2006 regression model and comes close in prediction accuracy to sales analysts’ estimates. This is achieved by handling some known issues in time series models such as seasonality and non-linearity of the earnings while improving predictions with additional explanatory variables that reflect the expectations of the market. Thus, we add to the existing literature a comprehensive and innovative neural network model that provides solutions to known challenges in forecasting and closes the gap between statistical models and sales analysts.