Forecasting the effect of stock repurchase via an artificial neural network

A simple static cascade-forward back-propagation artificial neural network (ANN) is utilized to forecast the effect of stock repurchase on the closing price of firm´s common stock. The input factors are composed of today´s closing price, index of the stock market and the amount of tomorrow-intended repurchase. A rule-based data clustering is used to group the repurchase days by selecting two records that are under the same conditions as the day before the next repurchase. A few initial predictions are less accurate than the classic accounting equation because the combinations of training set are very limited. However, after several repurchase days have passed, the ANN-based prediction usually introduces less error based on increasing amount of possible training sets. Therefore, this technique can be very useful if the repurchase is spanned in a quite long period.