In this presentation we discuss the role of data mining in time-series predictions, including those of market-trend data and stock quotes. These are characterized by stationary and non-stationary time series that may depend on non-quantifiable non-numeric measures, as well as recent information that may be significant in predicting near-term trends. Such predictions should, therefore, consist of predictions of non-stationary and stationary time series, as well as the abstraction and integration of non-numeric information in predictions. In this talk, we survey various techniques for trend predictions and data mining of market-trend data. We then propose the use of intelligent agents for the abstraction of non-numeric information, the decomposition of non-stationary time series into multiple stationary time series, and the prediction of trends using artificial neural networks. Finally, we illustrate our techniques for predicting stock-market data.