Quantitative Trading Strategy Based On Multivariate HMM Model

The domestic quantitative investment field has experienced more than ten years of development,the industry has begun to take shape,and quantitative strategies have gradually formed a system.Investors are paying more and more attention to quantitative investment strategies,especially in the context of the large asset management industry's evolution to net worth and technology.There are mainly two categories: multi-factor models and statistical arbitrage models in current various quantitative products.With the gradual improvement of linear models based on traditional quantitative algorithms,whether asset allocation or new factor mining,the linear system has been very high perfect,meanwhile its bottleneck has gradually emerged.It is difficult to mine new factors,especially for non-linear problems.With the development of machine learning in the fields of algorithm,computing power,and big data,it greatly reduces the difficulty and cost of machine learning.The advantage of machine learning is to find the non-linear relationship between objects.However,the current application of machine learning in the field of quantitative investment is not well.The first reason is that it is hard to accept for investors because of the black-box characteristic of machine learning.The second reason is that it is difficult to accurately describe the dynamic market in data processing because of complexity of financial data and over-fitting of algorithms.It results in poor performance outside the samples.This article adopts a multivariate hidden Markov model,takes labelled technical analysis as the observation sequence input of the model,and uses the future earnings as the hidden state to construct a quantitative trading strategy based on multivariate hidden Markov model.Hidden Markov model is a generative model of machine learning,which has relatively better white-box characteristics.At the same time,this article adopts supervised learning method to further improve the interpretability of the model.This article takes the five important indexes of the A-share market,Shanghai Stock Index,Shenzhen Component Index,CSI 300,SSE 50,and CSI 500 as the research objects,and selects the 2005 to 2012 market data as the model training data,the 2013 to 2020 market data as back testing data.The strategy in this article can obtain an average 117% extra return relative to the benchmark for a long time on the test objects and above 15% annualized return.Compared with the existing applications of hidden Markov models in the financial field,the main innovations of this article are as follows:(1)By introducing multi-dimensional technical index quantification technology,creatively applying multivariate hidden Markov models to financial market trend prediction.(2)Adopting multivariate hidden Markov supervised learning method to exploratively improve the interpretability of learning model in financial market applications.This article is divided into five parts: the first part is introduction,which introduces the research background and research cause;the second part is theoretical basis,which introduces related financial investment theory,quantitative investing theory,technical analysis theory and the algorithms of standard and multivariate hidden Markov model;the third part is model construction,including the selection and labeling process of technical indicators,the definition of hidden state of supervised learning,the process of multivariate hidden Markov model,and how to construct a quantitative trading model based on the learning results;the fourth part is empirical analysis,including empirical results and explanation of the results;the fifth part is the conclusion,including summarization of the research results,shortcomings and the next plan.