Design Of Online Portfolio Strategies With Historical Information

As one of the important branches of modern portfolio theory,the online portfolio theory studies how investors can adjust strategies with appropriate weights in a complex market environment to achieve the purpose of maximizing long-term returns.The efficient market hypothesis holds that investors' behavior is rational and that investors cannot obtain excess returns through historical information,e.g.price,turnover rate,volume,weight distribution,etc.However,with the development of behavioral finance theory,the efficient market hypothesis is constantly being questioned and challenged,and the irrational behavior of investors is gradually known by people.In addition,considering more realistic scenarios,scholars began to focus on online portfolios under the influence of different friction factors.Therefore,this thesis will mainly explore online portfolios that consider investors' irrational behavior and friction factors,e.g.,transaction cost and price impact,by mining and using different historical information.Combining the research results of behavioral finance theory and market microstructure theory,this thesis will use online learning technology to build the strategy model,and finally obtain the effective online portfolio strategy.The main contributions of this thesis are listed in the following four aspects:(1)Using historical price information,different strategies are proposed from the two perspectives of pattern matching update and passive active learning:?For historical price information,we use the nearest neighbor algorithm and correlation coefficient to define a similar set of price information.Based on the similarity set,we obtain an expert strategy related to parameter settings by the pattern matching update algorithm.Then,we integrate this expert strategy using the average integration method and propose a new online portfolio strategy.Besides,the universal property of this strategy is given and proved then,which means this strategy can obtain the logarithmic return of the optimal steady readjustment strategy that is approaching the benchmark.Meanwhile,this pattern matching strategy will also be applied to part(4).Based on this strategy,we will design an online strategy that considers price impact and transaction cost.? For historical price information,we use correlation coefficient and define a similar set of price information.The relative prices and correlation coefficients in the similarity set are used as predictions of future price trends and corresponding probability weights,and weighted averages are used to obtain future relative price estimates.Using the passive aggressive update algorithm,we propose an online strategy which is different from pattern matching.Finally,the two strategies are tested on 12 datasets from different financial domestic and foreign markets.The empirical results show that these two online strategies using historical price information can achieve higher returns,and the performance of the returns does not depend significantly on the choice of parameters.(2)Using price volume and text information,on the basis of measuring price reversal with price correlation,we introduce market anomalies and investor irrational behavior,utilize heuristic algorithm and then propose two different mean reversion strategies:?According to the characteristics of disposal effect,the anchor price deviation degree is used to construct the anchor price deviation as a feature of price reversal.Combined with the correlation characteristics of price reversal,a weight transfer equation is constructed to obtain the mean reversion strategy considering the disposal effect.This strategy is tested on the 6 datasets from domestic market in part(1).The empirical results show that this strategy can achieve higher returns,and the performance of the returns does not depend significantly on the choice of parameters.?By crawling investor posting data from the Choice Stocks Fortune,we construct investor sentiment indicators that take into account emotional lag and persistence.Combining investor irrational behavior characteristics,we construct two different new weight transfer equations based on two weight transfer equations that only consider the price correlation and the disposal effect.Utilizing heuristic algorithm,we obtain two mean regression strategy considering investor sentiment.Then,we use different domestic market datasets to conduct empirical tests on the two strategies.The results show that these two strategies have better performance.(3)In the case of sufficient market liquidity,we consider the design of online portfolio strategies under the influence of transaction costs.Combining with online active learning technology and using the idea of reducing the weight adjustment frequency,we propose an online active portfolio selection framework with considering transaction costs.Under this framework,the problem of online portfolio selection with transaction costs is decomposed into two specific questions,'when to activate weight update' and 'how to update weight'.We use the first-order and second-order information of portfolio weights to construct the activation probability,and then generate activation signals to control and select the timing of activating weight update.Next,we use the passive aggressive update algorithm,and add the Li regular constraint to propose two specific online active strategies with transaction costs.Finally,we test the two strategies on the same 12 datasets in part(1).The empirical results show that compared with existing online strategies,the two strategies proposed in this part have better performance under the influence of transaction costs.(4)Under the illiquid market environment,we estimate the impact cost with historical trading information,and study online portfolio strategy affected by both price impact and transaction cost.In view of the shortcomings of existing online portfolio models that consider transaction cost,we propose an online portfolio model that includes a portfolio rebalancing period.From the perspective of optimal execution of portfolio rebalancing,the impact cost of price impact is estimated.Combining the impact cost and transaction cost to calculate the transaction cost factor,we use the pattern matching strategy proposed in part(1)and obtain a new online strategy.Finally,we use demostic dataset to test the proposed strategy.The empirical results show that:?For all stocks in the dataset,whether their permanent or temporary impact coefficient may have an order of magnitude difference in their results,and the larger tradable shares,the smaller impact coefficient.?this strategy performs better than the baseline and online active strategy proposed in part(3)under the influence of transaction cost and price impact.Speccialy,this strategy can bear certain initial investment scale and transaction rate.