Simple Adaptive Filter as a Part of Information System for Market Data Analysis

Application of the simple least mean squares (LMS) adaptive filter of to the Warsaw Exchange Market (GPW) has been analyzed using stocks belonging to WIG20 group as examples. LMS filter has been used as a binary classifier, that is, to forecast the sign of changes in the (normalized) stock values. Two kinds of data has been used, namely, the differenced and double-differenced normalized close values of stocks. It has been shown that while the predictive power of LMS filter is virtually zero for the differenced series, it rises significantly in the case of double-differenced series for all analyzed stocks. We attribute this to the better stationarity properties of the double-differenced time series.