Short time series of share prices with financial results in day-ahead forecast - The warsaw stock exchange main market example

Forecasting of future time series values basing on past values is one of the domain of Artificial Intelligence (AI) interests. Classical approaches are based on different learning and deep learning methods where predictions and forecasting tasks can be improved by additional time series correlated with predicted one. In this paper the fusion of time series data with non time-series data (companies current financial reports) for one-day ahead forecast of share price change is proposed. The day-ahead forecast problem is formulated as two classification problems: share price increase/decrease two-class problem and share price increase/decrease/no change three-class one. It is shown that such data fusion can also lead to forecast accuracy improvement.