Study Of Trading Strategies Based On Stationary Process And Technical Analysis

Market efficiency hypothesis is a basic assumption of classical financial theory,many famous financial models are based on this assumption.However,during recent years,many researchers find the phenomena that this assumption does not hold under several circumstances.From the perspective of be behavioral finance,these phenomena are due to dealers' different psychology,or information asymmetry.This thesis tries to discover these phenomena under the framework of stationary process,and explore the strategies that can gain stationary revenue.The first part studies the methodology of pair trading,and provide two successful cases.Many previous researches investigate the pair trading strategy under the framework of cointegration theory,however,there is no persuasive theory that can provide the reason why there exist stable cointergration relationships among different securities' price.Thus the cointegration relationship in the training sample may not work for out-sample data.Based on this defect of cointegration arbitrage,we propose a new model to describe the process of paired securities' price.In this new model,we seperate the paired price into two parts:a filter based on moving average,and residuals.We pointed out that,under the assumption that securities' price logarithmic increment is(weakly)stationary,the residual process is also weakly stationary,furthermore,this new model requires no cointegration relationship.Based on this new model,we propose a new method for pair trading.We further seperate the revenue of arbitrage strategy into two parts:a part of risk,and a part of certainly income.Then we propose the theory and method to get the best estimation for the pairing ratio ?.Under the circumstances of high frequency arbitrage,dealers need to finish the deal with high speed,thus instead of last price,the trading strategy has to buy at the ask price and sell at the bid price.We proposed the modified trading strategy,which considers the bid-ask spread.We use the calendar spread arbitrage in CSI300Futures as an example,and examine the strategy's performance under 3 different data frequencies,and get positive return under each of the 3 circumstances.The return and return-risk ratio is highest when we use data with the highest frequency.Further more,we compare the performance of arbitrage back test results using the trading price of of bid-ask price with that using last price,the performance with trading price of last price is much better than the one using bid-ask price,which means the performance of high frequency arbitrage strategy is overestimated when we use last price as transaction price.At last,we show an example of inter-commodity arbitrage as another successful example.The second part studies the theoretical basis of trading strategy under the theory of stationary process.Under the hypothesis that log return is stationary,trading strategies based on stationary process have the properties that cumulative log return of unit time is stationary,and the average profit converges over time.These properties ensure the persistence and stability in profitability of trading strategy.Then we show the station-arity of several popular technical indicators,and propose the MACD strategy.We use CSI300futures to test the performance of MACD strategy,during the testing period of nearly 4 years.This strategy provide 26.86%annual return with Sharpe ratio of 2.17.We further propose the method to enhance a stationary strategy,and proof that the op-timal weight of every signal should be proportional to standardized prediction of return,then we provide the method to estimate the standardized prediction of return under the framework of stationary process.Then we use a classical KD strategy as the strategy to be enhanced.The original KD strategy provides the performance of annual return of 18.08%,and Sharpe ratio of only 1.2.After enhancement,the enhanced KD strategy have the annual return of 52.33%,Sharpe ratio ascends to 3.99.At last of this part,inspired by the mean-variance model,we discuss the construction of a portfolio of strategies,with the objective function being the tradeoff between expected return and variance,and the condition of all weight being non-negative.The result of empirical study shows that us-ing a portfolio of optimized-weighted strategies may help reduce the risk of strategy and increase the return-risk ratio while the return does not decrease.The last part studies the predictive power of 4 popular pairs of two-day bullish and bearish Japanese candlestick patterns in Chinese stock market.Based on Morris' study,we give the quantitative definitions of these four pairs of two-day candlestick patterns.When testing predictive power of candlestick patterns on short-term price movement,many researchers deem the occurences of candlestick patterns as independent events,however this method may lead to the following two problems:Firstly,there exists high correlation among the samples because of the overlap of holding period among different stocks;Secondly,this method can not give the appropriate result when the occurences is not evenly distributed along the time axis.Thus we study this question from the perspective of portfolio strategy,we propose two stock portfolio strategies,one measures the absolute return,another one measures relative return.Under the assumption that stocks' price follows the process which is logarithmic increment stationary,we prove that both the two strategies' daily return are stationary.Then we use Step-SPA test to test the profitability of these two strategies.To ensure the robustness of our result,we use two different sample sets,one is medium sized stocks,another is the set of large sized stocks.Test results show that the predictive power differs from pattern to pattern.3 of the 8 patterns show the strongest predictive power regardless of stock capital size.Compared with large sized stocks,candle stick patterns show better predictive power in the medium sized stock sample.The method used in this part provides a new framework for event study research.