Application of XCSR Model for Dynamic Portfolio Selection

Abstract

ABSTRACT Security selection is the most time-consuming problem in investment process. Today, investing environment is more complex than before and investors can’t see through the information frame, the reality behind which they are unable to scrutinize. Static model cannot provide appropriate solutions when current phenomena are completely different from that of the training period. Learning classifier system (LCS) is possessing with dynamically learning mechanism to evolve internal rules in response to changes in the external environment. This study employs real number version classifier system (XCSR) to investigate constructing dynamic stock portfolio. We examine the constituents of the TSEC Taiwan 50 Index and take those stocks daily transaction data to generate a number of technical indicators as input factors for XCSR model. The empirical results revealed that this study's XCSR model yields return on investment is significantly better than the Buy and Hold model. This research also indicates that classifier system can effectively monitor market fluctuations and the proposed model can help investors obtain relatively optimal returns. Keywords: Extended Classifier System, Dynamic Portfolio, Framing Effect