Font Size: a A A

Statistical Arbitrage Of High-frequency Data Based On State Space Model

Posted on:2015-07-17Degree:MasterType:Thesis
Country:ChinaCandidate:F Y ZhangFull Text:PDF
GTID:2309330431458399Subject:Applied Statistics
Abstract/Summary:PDF Full Text Request
Capital market is a market full of all kinds of risk, Once handled poorly, investor may lose the principal,what’s worse, the investor may suffer great losses. We launch the CSI300stock index futures and margin in2010, which changes trade pattern from one-way to two-way in China Stock market. However, the loss ratio of participants remains high. People have been looking for an investment strategy that has nothing to do with the market trend and gets stable profits.Statistical arbitrage of quantitative investment is a neutral strategy. It does not rely on market trend and bases the arbitrage on historical data, guided by the fundamental analysis.Statistical arbitrage increases a bit of risk, but it can get more chances than those of the traditional risk-free arbitrage.There are many statistical arbitrage strategy is very rich. Co-integration model strategy is often adopted in statistical arbitrage, which has some advantages and is operable in practice. People are not satisfied with it, however. So far, we have got more effective method, such as the wavelet analysis and neural network methods. Some start from the residuals and use GARCH model and O-U model to innovate treatment methods. And some start from the optimal threshold and look for trading rules to gain higher yields. This paper improves the coefficients inthe cointegration model, introduces time-varying parameter state space model, and adopts kalman filter algorithm to estimate state vector. We change fixed coefficients in the traditional cointegration model to time-varying coefficients, which can reduce the volatility of residual, make price sequence better fitted, in order to find an effective arbitrage opportunities.This paper selects for30-minute,15-minute and5-minute high frequency data to develop the state space model, using the most commonly used threshold rules to conduct arbitrage analysis. It shows that there are many arbitrage opportunities within a day in the actual market, making up for the inadequacy of only using daily closing price data. According to the results of statistical arbitrage in the samples,30-minute data sequence arbitrages12times,15-minute data15times and15-minute data30times. Price arbitrage cumulative yield of30-minute data sequence is6.7904%, that of15-minutes data sequence is2.3403%, and that of5-minute is only 0.7459%. Out of the sample, the numbers of arbitrage of different frequency data are the same, that is, unwinding4times, no extreme stop-loss and cumulative yield is over5.5%. In conclusion: the higher the frequency of the data, the more the number of arbitrage; but due to the extreme stop situation and transaction costs, arbitrage cumulative yield would lower instead. It indicates that it is not so suitable to choose high frequency data to conduct statistical arbitrage, such as five-minute and1-minute data. According to the empirical results in this paper, using30-minute data can get relatively stable high yield.In this paper, the empirical analysis demonstrates that using the high-frequency data to develop state space model can effectively conduct statistical arbitrage, the strategy model can be used to guide the investors to avoid market risk, and obtain the stable income. According to the research process and results in this paper, we make the following recommendations on the practice of statistical arbitrage:First,sequence should have a high correlation between asset price and trading and be more active; Second, asset prices sequence should not include many sudden extremes; Third, it is better to adopt the high frequency data rather than a day, such as30-minute data; Forth, besides the cointegration model, state space model also is an effective method to conduct statistical arbitrage.
Keywords/Search Tags:statistical arbitrage, time-varying parameter, state space model, kalmanfiltering, high frequency data
PDF Full Text Request
Related items