Abstract: The trading signal index system of cointegration strategy based on the fixed threshold is easy to trigger the trading too early, which leads to the cointegration strategy closing the stop loss at the peak of spread fluctuation, which is not suitable for the volatility aggregation time series. Based on the state space model, the time-varying coefficient cointegration statistical arbitrage strategy is constructed, the time series data of five state-owned banks’ A-share closing price are used for empirical analysis, GARCH model is used to extract the volatility agglomeration information of spread series, and wavelet de-noising technology is used to optimize the time-varying trading signal index system. The Sharpe ratio of wavelet de-noising GARCH time-varying threshold cointegration strategy is as high as 2.284 in the turbulent period from January to June of 2018, which is 18.4% higher than that of non de-noising time-varying threshold cointegration strategy and 109.5% higher than that of fixed threshold cointegration strategy in the same period; In the stable period from July to December of 2018, the sharpe ratio of the three strategies was lower than -2, and the sharpe ratio after noise reduction was slightly lower than that of the non noise reduction strategy. The empirical results show that: in the period of market turbulence, the time-varying trading threshold based on GARCH model can effectively capture the volatility agglomeration information, improve the accuracy of the model position when the jump spread occurs, and the wavelet de-noising optimization can reduce the possibility of triggering trading due to the wrong noise signal, and significantly improve the bank strategy Sharpe ratio. In the stable period, wavelet de-noising time-varying threshold cointegration strategy and non de noising, fixed threshold cointegration strategy are not suitable for the stable period.

Key words: co-integration statistical arbitrage； trading signal index； wavelet denoising； GARCH model； sharp ratio