Intelligent techniques for economic time-series

Abstract

Prediction of a non-stationary time-series is hard as the frequency components and their amplitudes in the series vary randomly over time. This paper proposes a two-step approach for prediction of such non- stationary time-series. The first step employs Taylor series to approximately predict the next time-point value in the series from its current and last two preceding sample values. The Taylor series-based prediction, however, presumes that the time-series is locally stationary. The second step includes reinforcement learning inspired error correction in the Taylor series-based prediction results using an adaptive probabilistic learning matrix. The said matrix is used to encode the scale factors (error coefficients) in the learning equation from an estimate of errors incurred in Taylor series-based prediction. The learning is performed by a reward/penalty type adaptation of conditional probabilities involved in the probabilistic learning matrix. On convergence of the matrix, the saved probabilities are used to determine the appropriate scale factor needed to scale the estimated error in Taylor series-based prediction with an aim to rectify the prediction result by adding the scaled error to the predicted time-series value. Experiments undertaken confirm that the proposed technique outperforms the state-of-the-art prediction models by a significant margin of prediction accuracy.