基于隐马尔科夫模型、马尔科夫状态转换模型的商品期货价格预测研究与实证分析

发布时间：2018-04-03 12:37

本文选题：商品期货　切入点：价格预测　出处：《山东大学》2017年硕士论文

【摘要】：投资者参与期货交易进行套期保值、投机或者套利。如果能够对期货价格波动做出及时准确的预测,可以有效降低交易风险,乃至获取超额收益。本文主要基于隐马尔科夫模型(Hidden Markov Model,HMM)相关的理论,研究商品期货价格预测、期货市场状态预测的相关问题。本文采用国内期货市场上交易相对比较活跃的螺纹钢期货品种作为实证研究对象,首先在具有高斯输出的隐马尔科夫模型(Gaussian Hidden Markov Model,GHMM)下,采用处理后的收盘价序列作为单个特征因子进行实证分析。随后放宽价格序列不相关的假设,采用马尔科夫状态转换自回归模型(Markov Switching Autoregressive Model,MS-AR)进一步做实证检验,最后采用具有高斯混合输出的隐马尔科夫模型(Gaussian Mixture-Hidden Markov Model,GM-HMM)进行数据模式匹配,并进行价格预测。本文首先介绍了 HMM模型的发展历程,然后介绍了与模型相关的理论基础,最后利用螺纹钢连续(RB888)数据进行实证分析。实证过程主要包括模型参数估计、解码和预测三个步骤。在GHMM和GM-HMM模型下,参数估计使用的是Baum-Welch算法,通过设定初始参数值并将训练样本数据传入模型,经过迭代估计得到最优参数估计结果。预测过程是利用已经学习得到的模型,对未来一段时间内期货价格、市场状态进行预测。在价格预测中,本文使用GHMM模型预测了未来价格的概率分布,使用GM-HMM模型基于"单日预测"和"加权预测"的方式,预测了未来价格的走势。状态预测即预测市场所处的状态以及市场状态是否会发生转变("牛转熊"或者"熊转牛")。在MS-AR模型下,模型的参数估计主要通过Hamilton滤波过程来实现。本文实证结果表明,GHMM,MS-AR和GM-HMM模型对于价格预测,都具有一定的精确度。本文研究内容和方法,对于投资者看清市场,及时获得市场走势预测,并进行相应交易具有一定的指导意义。
[Abstract]:Investors participate in futures trading for hedging, speculation or arbitrage.If the futures price fluctuation can be predicted timely and accurately, the transaction risk can be effectively reduced, and even the excess return can be obtained.Based on the theory of Hidden Markov Model (hmm), this paper studies the related problems of commodity futures price forecasting and futures market state forecasting.In this paper, the relatively active rebar futures in the domestic futures market are used as the empirical research object. Firstly, under the output of Gao Si's Hidden Markov Model (Gaussian Hidden Markov Model GHMMM),Using the processed closing price sequence as a single characteristic factor to carry on the empirical analysis.Then the assumption of irrelevance of price sequence was relaxed, and Markov Switching Autoregressive model MS-ARA was used to further test the data pattern matching. Finally, Gaussian Mixture-Hidden Markov Model GM-HMMM with Gao Si mixed output was used for data pattern matching.And carries on the price forecast.This paper first introduces the development of HMM model, and then introduces the theoretical basis related to the model. Finally, the empirical analysis is carried out by using the continuous steel bar RB888 data.The empirical process consists of three steps: model parameter estimation, decoding and prediction.In the GHMM and GM-HMM models, the parameter estimation is based on Baum-Welch algorithm. By setting the initial parameter value and passing the training sample data into the model, the optimal parameter estimation results are obtained by iterative estimation.The forecasting process is to predict the futures price and market state in the future by using the model that has been learned.In price forecasting, GHMM model is used to predict the probability distribution of future price, and GM-HMM model is used to predict the trend of future price based on "one day forecast" and "weighted forecast".State forecast is to predict the state of the market and whether the market state will change ("Bull to Bear" or "Bear to cattle").Under the MS-AR model, the parameter estimation of the model is mainly realized by the Hamilton filtering process.The empirical results show that both GHMM MS-AR and GM-HMM model have some accuracy for price prediction.The research contents and methods of this paper have certain guiding significance for investors to see the market, get market trend forecast in time, and carry on the corresponding transaction.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F724.5;F764.2

【相似文献】