A FDR Based Pairs Trading Strategy

The capital markets are rapidly expanding and evolving, and the regulation rules of national and international capital markets are constantly adjusted and removed. However, it remains difficult to understand and grasp of the trend of security market. Hedging will certainly become the most important part of investment field. Taking advantage of its simple nature, pairs trading, an important kind of hedging, is rapidly evolving in both developing and developed countries. Prior to 2010, the immature market condition that stocks cannot be selling short in China stock market hampered the application and development of pairs trading. However, since March 2010, margin trading has been launched, and finally investors can short stocks. Besides, a variety of hedging strategies that need to short stocks can finally be applied. As a simple hedging strategy, pairs trading will have a broader application prospect. What's more, the research of pairs trading, in some way, may be able to serve for the research of more complicated hedging strategies, and may reveal the secrets of excess return of hedging strategies.At the same time, with the rapid development of computing technology, the requirements for data-processing in more and more industries are getting higher. Multiple testing has become a new method to solve the problem concerning large-scale statistical inference. How to control the overall systematic error rate has become the primary problem, especially when we need to perform multiple tests of significance simultaneously. Since FDR method was first proposed by Benjamini to control multiple hypothesis testing error rate, FDR method has become the focus of research on multiple hypothesis testing. Practically, in pairs trading, we need to select pairs that have significant correlation from thousands of stocks, and this is just a typical problem of multiple hypothesis testing. Thus, we believe that we can apply FDR method in the stock selection model of pairs trading, and the excellent property that FDR method can control the overall error rate must improve the performance of pairs trading.In the aspect of trading model, this paper establishes a more complex and sophisticated pairs trading model. There are three kind of trading signals:open signal, close signal, and stop signal. Different from what were stated in previous literature, these signals are determined not by experience, but by data-driving, so that these signals are optimal trading signals. Besides, a moving window model, along with a window width, is also used to update everyday trading signals, which makes them dynamic signals. In addition, this paper also studies the influence of different trading signals on the average return, the average transaction frequency, the average stop frequency of pairs trading. It turns out that the linear variation of trading signals will result in the nonlinear variation of the three variables above. Unlike some previous papers that ignore transaction cost, this paper, based on current margin trading system and rules, quantifies the transaction cost of long and short positions and embeds the cost in the trading model in a way that the cost has a direct effect on the returns of pairs trading strategy. Also, different from previous cash neutral position, a new way that is based on cointegration regression model is used. This new delta neutral way to take a position not only is theoretically better, for it makes the allocation of asset match its corresponding risk, but also can achieve higher average return.After the establishment of the trading model of pairs trading, this paper proposes a new method to screen pairs-FDR method. Unlike the previous pairs trading strategies that screen pairs through the correlation coefficients of price serials, or through the minimization of the sum of squares of the difference between prices of pairs, a FDR-based pairs trading strategy uses the false discover rate method to find the pairs that still exhibit abnormal correlation, after applying a factor model to control the effect of systematic risks. Our empirical work shows that both the average return and the amount of pairs that are selected by FDR method are much higher than the return and amount of pairs selected by correlation method. What's more, FDR-pairs have higher Sharpe ratio, lower standard deviation, and a distribution of negative bias and leptokurtosis. Also, FDR-pairs have higher liquidity and longer trading period than pairs selected by correlation method, no matter how the positions are opened or whether the pairs have generated profit, have experienced loss or have been stopped. Generally speaking, both the liquidity of pairs in the training period and the liquidity of pairs in the trading period have a very significant positive impact on the return of pairs. In addition, the liquidity of pairs in the training period has a better interpretation of the return of pairs selected by correlation method. However, the liquidity of pairs in the trading period has a stronger explanatory power of the return of FDR-pairs.At last, this paper tries to establish the relation between pairs trading and market efficiency. The main content that is studied by many scholars who concern market efficiency is usually the deviation between the market value and the real value of stocks. It is generally believed that the stock liquidity will affect the market efficiency:the stronger the liquidity of a stock, the more frequently it is traded, the more attention paid to the movement of the price of the stock, therefore, once the deviation between the market value and true value of the stock is big enough to cover the transaction cost, the reversal transactions of investors will let the market value revert to the true value, or fluctuate around the true value. In short, the stronger the liquidity of the stock, the less likely the stock is mispriced, the less likely the relative price spread will have large fluctuations; on the contrary, the weaker the liquidity of the stock, the more likely the stock is mispriced, the more likely the relative price spread will have large fluctuations. Thus, in pairs forming period, the weaker the liquidity, the more likely the stock will be mispriced, and the more likely there will be trading opportunities, and this stock is more likely to be chosen by the stock selection model; in pairs trading period, the stronger the liquidity of the stock, even when it is mispriced, its market value will quickly revert to its true value. For pairs, it means that the relative price spread will quickly revert to its mean. Because the source of the revenue of pairs trading is its mean-reverting property, we can surely expect that higher revenue of pairs trading comes from these pairs consisted of stocks with higher liquidity.And the empirical study of this paper verifies these expectations. Through our empirical study, turnover rate, a liquidity indicator of market efficiency, has a significant positive impact on the return of pairs trading. And the liquidity can affect the efficiency of market in a way that the higher the liquidity of stocks, the more frequently stock are traded, and the less likely that stock are mispriced, thus the higher of market efficiency. Besides, we believe the excess return of pairs trading is due to the misprice of two paired stocks. Our other empirical work also indicates that the liquidity of unpaired stocks is higher than the liquidity of paired stocks, which is consistent with our expectation. No matter how the positions are opened, the liquidity of pairs have changed significantly during the training and trading period and have a huge impact on the return of pairs, especially the liquidity of pairs in the trading period.