基于隐式马尔可夫模型市场的择时策略研究

发布时间：2018-03-03 04:29

本文选题：量化　切入点：择时　出处：《浙江工商大学》2017年硕士论文　论文类型：学位论文

【摘要】：本文将隐式马尔可夫模型应用于中国股市得到了不错的效果,在模型的回测中针对各个问题分别进行处理。在回测的设计方面将模型设计出一种不断更新的模式,即以m天的数据作为模型的训练数据每隔n天进行数据更新,这里的m与n作为交易方案中重要的参数以最大或模型样本内夏普比率为目标进行网格搜索优化。在交易的形成方面根据各个隐藏状态在模型训练样本内的收益情况进行打分,选取得分最高的隐藏状态进行交易,在交易的实施方面以每日的开盘价进行交易,收盘价进行结算。在回测标的的选择方面分别以沪深300、中证500和上证50三个标的分别进行回测,从回测结果上看,模型在样本内均能取得不错的效果,从三个市场的回测净值曲线上不难看出模型均能很好的应对2009年的牛熊转换,但是模型在2015年下半年均出现了巨大的回撤,这说明模型在较为复杂的市场情况下不能很好的做出反应,当除去2015年的快速下跌导致的回撤后模型的回测效果明显得到改善。为了验证模型的稳定性,分别针对不同的更新速度以及模型训练期进行回测,回测结果上看出模型存在广阔的参数高原,因此可以认为模型在参数优化的方面过拟合的可能性较小,除此之外针对不同的模型输入特殊特征分别进行回测,从结果上看出模型在不同的输入特征下均能取得不错的效果,这也说明了模型的稳定性,最终认定基于隐式马尔可夫模型的择时策略是个较为有效的策略。
[Abstract]:In this paper, the implicit Markov model is applied to the Chinese stock market with good results, and the model is dealt with separately in the reverse-test of the model. In the design of the model, a new model is designed. That is, the training data using m-day data as the model is updated every n days. The m and n are used as important parameters in the transaction scheme to optimize the mesh search with the maximum or sharp ratio in the model sample. In the formation of the transaction, the transaction is graded according to the earnings of each hidden state in the model training sample. The stealthy state with the highest score is chosen to trade, and the daily opening price is used to carry out the transaction. The closing price of the stock market is settled. In the selection of the target for the back test, the three marks, namely Shanghai and Shenzhen 300, China Stock Exchange 500 and Shanghai Stock Exchange 50, are respectively selected. From the result of the review, the model can all achieve good results in the sample. It is not difficult to see from the net return curve of the three markets that the models can well cope with the transition in 2009, but the models all showed a huge retreat in the second half of 2015. This shows that the model does not react well in the more complex market conditions, and the model's backtracking effect is significantly improved after the withdrawal caused by the rapid fall of 2015, in order to verify the stability of the model. According to the different renewal speed and the training period of the model, the results show that there is a wide parameter plateau in the model, so it can be considered that the model is less likely to be over-fitted in parameter optimization. In addition, according to the special characteristics of different model input, the results show that the model can achieve good results under different input characteristics, which also shows the stability of the model. Finally, the time selection strategy based on implicit Markov model is considered to be a more effective strategy.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【参考文献】