Abstract:

Markowitz's mean-variance portfolio model pioneered modern portfolio theory. However， due to the financial assets being non-normal and time-varying distributed， the efficiency of the mean-variance portfolios is difficult to achieve， which makes investors face serious welfare losses. High dimensional dynamic higher-order portfolio can effectively solve the existing drawbacks of the classical mean-variance portfolios； however， its application also meets several difficulties.A time-varying higher-order co-moment estimate， labeled as VC-MF-TVSNP， is proposed by combining a varying-coefficient multi-factor model and a time-varying semi-nonparametric （TVSNP） model. The model specification， estimation， and selection approaches are given in this paper. The multi-factor model can efficiently reduce the “curse of dimensionality” problem in the time-varying higher-order co-moments estimation， and the semi-parametric structure can efficiently solve the “model misspecification” problem. Then a high-dimensional dynamic high-order moment investment analysis is given based on the component stocks of the Chinese CSI 300 index.The empirical studies show that the VC-MF-TVSNP model can effectively capture the time-varying structure of higher-order co-moments of asset returns， and it is more suitable for the latent structure of asset returns. High-dimensional dynamic portfolio based on the VC-MF-TVSNP model can generate higher and more stable economic value， which is further confirmed by robust analysis.To a large extent， the VC-MF-TVSNP model solves the “curse of dimensionality” and the “model misspecification” problem efficiently， which can provide a more precise estimation of high dimensional time-varying higher-order co-moment estimation rather than the existing approaches， and give investors a better reference for asset allocation.

Key words: multi-factor model, semi-nonparametric distribution, time-varying higher-order co-moments modeling, dynamic portfolio