Font Size: a A A

A Study On On-line Portfolio Strategies Based On Market Anomalies And Forecasting Stock Prices

Posted on:2017-02-11Degree:MasterType:Thesis
Country:ChinaCandidate:Z J PengFull Text:PDF
GTID:2279330503485599Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
The core problem of investment portfolio selection is how to allocate financial assets to achieve an optimal portfolio under uncertain circumstance. Because financial market is an extremely complex system, the investors have to face an endlessly changing situation when they are engaging in investment practices. In order to achieve the goal of maximizing return, the investors need to adjust their strategies according to the changing situation without any future information. Thus, investment portfolio selection is nothing but a dynamic, on-line problem. The modern portfolio selection theory based on mean-variance model has been extensively studied by many researchers and lots of valuable results have been obtained in the static case; however, people have seldom dealt with it in the dynamic case because the solution to the model is too difficult. In recent years, on-line algorithm has been applied to portfolio selection which is based on universal portfolio. This thesis tries to explore the portfolio selection problem systematically and thoroughly by means of the character of market’s anomaly and on-line learning algorithm. The main contributions of this thesis are listed in the following three aspects:1. We propose to take use of the robust L1-median to construct a new on-line portfolio strategy which is a universal portfolio strategy. Based on historic information, this thesis would use the robust L1-median to forecast the price momentum and give a new on-line portfolio strategy with maximizing prospective return. Not only has simple computation, this new strategy but also uses as much historic information as it can. Furthermore, this thesis proves that the new strategy has universal portfolio property. Finally, taking experiment on real market data, the results show that this new strategy can obtain better performance.2. We propose two new switching portfolio strategies which reflect different momentum effect or reversion effect. Focused on designing the switching regime among several basics strategies and combing the switching regime with market anomaly, this thesis sets different switching probability which reflects different momentum effect or reversion effect, and gives tow new switching portfolio strategy. Furthermore, this thesis gives and proves the return threshold of the momentum switching strategy. Finally, taking experiment on real market data, the results show that these two new strategies can obtain better performance.3. We construct a passive aggressive strategy via multi-period asymmetry mean reversion. Most of the existing on-line reversion strategies employ the assumption that mean reversion is multi-period symmetry or single-period asymmetry, while the real mean reversion usually is multi-period asymmetry. According to this problem, this thesis investigates the on-line strategy with both multi-period and asymmetric mean reversion. This thesis gives a new on-line portfolio strategy by means of setting piecewise loss function and combing the loss function with passive aggressive algorithm. This new strategy has linear time complexity. Although this new strategy has not a good property like universal portfolio property, the results of experiments taken on four real markets show that the new strategy has better performance and can bear higher transaction cost rate than the existing strategies.In general, these theses investigate the on-line portfolio selection strategy based on the character of market’s anomaly and the methods of forecasting the price. This study not only enriches the modern portfolio theory, especially on-line portfolio theory, but also has important practical significance for on-line investors.
Keywords/Search Tags:on-line portfolio selection, market anomaly, price forecasting, piecewise loss function, passive aggressive algorithm
PDF Full Text Request
Related items