Categories
DSP

The Best Mother Wavelet for Stock Prediction Using Adaplet Method

It has been determined the best mother wavelet for stock prediction of SONY 2006 and BNI 2012 based on the Adaplet Method (The Adaptive Filter which is use wavelet as initial coefficients). The Mother Wavelets, which is used, are Coiflet 1-5, Daubechies 1-5, and Symlet 1-5. The prediction analysis includes overshoot, autocorrelation error and data pattern conformity, three days prediction, and segmentation. According to the overshoot analysis, it shown that for all the data, the overshoot at the beginning of data increased as its wavelet level increased. While using Daubechies 1 and Symlet 1 produced smallest overshoot among other wavelets (112.2%). The autocorrelation error of data pattern prediction indicates conformity with the original data. As its wavelet level increased, the autocorrelation error pattern also ramped (near zero). Coiflet 5 and Daubechies 1 produced the smallest mean square error (MSE), which is equal to 0.0147; meanwhile, Coiflet 1 shows the best result with an average error of 0.001 in next three days prediction analysis. On the other hands, Symlet 3 shows the best MSE of 1.213. Symlet offers the best result, according to best wavelet sequence assessment of each method.

(click here for download the paper)

Categories
DSP

Prediction Results Analysis of Sony 2006 and Toshiba 2006 Share Data using Adaplet Method (Wavelet based Adaptive Filter)

The pattern prediction results from the Sony 2006 and Toshiba share data have been analyzed using Adaplet Method, a wavelet based adaptive filter (adaptive filter which its initial coefficients are wavelet coefficients), which is based on Coiflet, Symlet and Daubechies Wavelets. According to these wavelets, it has shown that, for all data, as its wavelet level increased the anomaly patterns at the beginning of data also increase. For Coiflet wavelet, each result is different but the error correlations are going to smooth (approaching zero). For Symlet Wavelet, the result is not quite different and the error correlations also not so good. For Daubechies Wavelet, the result is not quite different between wavelet levels, and the error correlations is not so good (always oscillates). So the wavelet chosen affects the prediction pattern results and the error correlations, which for these data, the Coiflet Wavelet offer good prediction pattern result and good error correlation.

(download PDF here)