Research On Constructing Statistical Arbitrage Strategy Based On Time - Varying Weighted Kernel Density Estimation

The main idea of Pairs Trading-a widely used statistical arbitr age strategy, is to find some best-related investment products at first, and among them find a pair of products which possess long-term eq uilibrium relationship (also in general called co integration relations hip).When unreasonable deviation appears in this pair of investmen t varieties,investors began to position in the corresponding securitie s and the stock market, buying relatively undervalued investment va rieties and selling relatively overvalued varieties. When the spread o f the paired species returns to the non-arbitrage interval, the positio n can be closed. The biggest difference between the strategy and th e traditional way is that it is concerned with the relative spread of the two varieties of investment, rather than the price trend of each product. Also if taking this strategy, you should establish a short o pening and long opening in the market at the same time and this h edge mechanism can effectively avoid the market system risk, so it is also a market neutral strategy, because the income has low corr elation with the overall trend of the market.In this paper, the author will try to set up Pairs Trading from the perspective of the cumulative percentile and this method of the kernel density estimation in nonparametric is mainly used here.Whe n setting up the strategy of pair trading, we usually tend to find out the two time series of closing price is in consistent or not fi rstly. If it is,the price trend of these two sequences is basically con sistent. If the market is perfect, then its corresponding cumulative percentile of the price in each moment should be the same.But act ually the market is not so perfect,or ineffective, which resulted in the difference of cumulative percentage of price in each moment. when there is a serious price deviation, the cumulative percentile deviation will also indirectly appears. The pair trading mode is fact ually established on basis of this idea. In this paper, the author is going to use the method of weighted kernel density estimation to e stimate the distribution function of the price at every moment. At 1 ast, an empirical analysis is carried on agricultural futures, chemical futures and metal futures in order to testify the feasibility and stab ility of the method according to the idea of pair trading.