An Empirical Application Of Statistical Arbitrage In The Stock Market

Statistical arbitrage is a trading strategy. It aims to construct such a portfolio that contains a pair of stocks which are co-integrated. When the spread is too high, we can sell the higher-priced security and buy the lower-priced security with the idea that the mispricing will correct itself in the future. And when the spread is too low, we can deal in the opposite direction. Statistical arbitrage is market neutral, which means no matter how severe fluctuation the market is the portfolio can hedge the system risk in the market.So far many scholars have make study in statistical arbitrage. These research focus on pairs selection and trading design. This paper will make an empirical test about the effect of common models in our market based on the existing research. These models include Normal distribution, GARCH model, and so on. The result is that GARCH model is not a suitable model as it produces too many trading signals, while Normal distribution is a good model. Take the stock of Minsheng Bank and Merchants Bank for example, the portfolio can obtain every17percent annualized return during2007and2011.At the same time, many papers focus on one-one pairs, this paper tries to deal with one-many pairs by using principal component analysis. The result is that the two methods have respective advantages and disadvantages. Both of them produce many trading opportunity. The former need a high correlation between the pairs, and the latter is more suitable for stocks that have small float market capitalization and fluctuate severely. Finally this paper applies catastrophe theory into statistical arbitrage. What catastrophe theory means is that there exists a change point, and the statistical character of time series before and after the point is different. This paper use likelihood estimate to find the change point of index and stocks. By using change point, the return becomes higher and the risk becomes lower.