中文摘要：

配对交易是统计套利中最主要的交易策略，但随着市场有效性的逐渐提高，该策略的获利机会正变得越来越有限，传统的固定参数交易模型已难以保证配对交易一直获得最大利润，交易模型的参数不仅需要优化，而且还需要动态地、自动地调整优化值，因此有必要研究开发具有人工智能属性的参数动态优化交易模型，这对于提升交易模型的盈利能力和执行效率具有重要意义。自适应配对交易模型是对传统的协整配对交易策略进行改进，推出一种基于强化学习模式的新型统计套利交易模型；将Sarsa强化学习算法和8-greedy策略与新模型相结合，把模型参数的确定方法由传统的主观经验法和固定参数法改进为自适应模式的动态参数优化法；编制的计算机程序仿真实现了基于新模型的套利交易全过程，涵盖模型参数的动态优化、套利交易的模拟操作以及交易绩效的测量评估；以中国债市交易量最大的5种债券为样本，构建4组配对组合，采用Johansen协整检验法、T检验和Robust稳健性检验等方法对交易模型和测试结果进行实证分析。研究结果表明，新模型的运行效果全面优于传统模型。新模型显著提升了交易系统的获利能力，收益率和索提诺比率大幅提高；同时降低了投资风险，最大回撤出现明显下降；还提高了套利交易的执行效率，交易次数明显减少，套利成本下降；具有持续学习的能力，能促进累计收益率不断上升并最后收敛于最大值。研究结果还表明，协整配对交易在中国债券市场同样具有有效性，能够获得显著正收益。将强化学习思想与协整配对交易策略相结合，设计开发出一种新型配对交易模型，实现了模型参数的自适应动态调整。这种改进型交易模型有助于应对传统配对交易策略获利能力的下降，进一步提高配对交易策略的效?

英文摘要：

Pairs trading is one of the major statistical arbitrage trading strategies. However, its profit opportunity has become scarcer due to the improvement of the market efficiency. The traditional fixed parameter trading models are no longer sufficient for eternal profit maximization. The parameters of the trading models need not only to be optimized but also to be done so dynamical- ly in an automatic manner. Therefore, it is necessary to develop a trading model of which parameters are dynamically optimized with artificial intelligence, as it may be of significance in improving the profitability and efficiency of trading models. A new type of statistical arbitrage trading model is proposed based on the reinforcement learning mode, improving the tradi- tional cointegration trading strategy; Applying the Sarsa algorithm and 8-greedy strategy to the new model, the key parameters in the new trading model can self-adapt to reach the optimal values, instead of judging from professional experience or insisting on determined parameters just like the traditional strategy; A computer simulation is designed to run through the complete process of the new trading model including model parameters self-adapting adjustment, securities transaction, and trading performance eval- uation. The trading simulation and empirical tests such as Johansen cointegration test, t-test, and Robustness test are conducted on four bend pairs that are composed of the top five bonds with the largest trading volumes in the mainland markets. The results show that the new model outperforms the traditional one in all aspects. It significantly enhances the profitability of the trading system while reducing the drawdown risks ; It improves the efficiency of arbitrage trading as it reduces the number of transactions and thus transaction costs ; It possesses ability to learn continuously so that it increases the accumulated return step by step and eventually converges to the highest level. The results also reveal that the cointegration trading strategy is eff