A Novel Time-Scale Feature Based Hybrid Portfolio Selection Model for Index Fund

Index fund is one of popular form in portfolio management that aims at matching the performance of the specified benchmark index. Since investors are a diverse group who operate on very different time scales, a novel time-scale feature based hybrid model is proposed in this paper for portfolio selection of index fund. First, maximum overlap discrete wavelet transform (MODWT) is used as a preprocessing to decompose time-scale features. With Particle Swarm Optimization (PSO) optimizing the weight of each scale, our approach can effectively and automatically extract important time scale features and eliminate the noisy features. Then, applying a fast two-level clustering algorithm, homogeneous groups of securities are formed based on weighted time scale features. Last, representative stocks of each group are selected for tracking portfolio construction. The computational results on 8 indexes demonstrate the effectiveness of the proposed model.