【摘要】 对已有的马尔科夫转移动态因子模型提出了一个新的参数估计方法——两步最大期望(Expectation-maximization, EM)法，通过对马尔科夫转移动态因子模型进行重新参数化，将其转换为混合动态因子模型，并将因子与状态均视为潜变量，利用EM算法实现对重新参数化后的参数以及因子得分的估计。将因子得分视为已知数据、状态视为潜变量，针对每个因子序列建立马尔科夫转移自回归模型，利用EM算法对依状态变化的截距项和自回归系数进行估计，并对状态与拐点进行识别。通过数值模拟验证该方法的有效性，并将该模型与估计方法用于我国沪深股市股票数据分析中，对股市行业周期进行度量和识别。更多还原

【Abstract】 A novel parameter estimation method, the two-step expectation-maximization(EM) approach, is presented for existing Markov switching dynamic factor models. The Markov switching dynamic factor models are re-parameterized to transform them into a mixed dynamic factor model, with both factors and states treated as latent variables, the EM algorithm is employed to estimate the parameters and factor scores in the re-parameterized model. With factor scores taken as known data and states as latent variables, Markov switching auto-regressive models are established for each factor sequence. The EM algorithm is utilized to estimate the intercept terms and auto-regressive coefficients that vary with states, and to identify states and turning points. The effectiveness of this method is verified through numerical simulations. The model and estimation methods are applied to analyze a stock market data set from the Shanghai and Shenzheng stock exchanges in China, aiming to measure and identify industry cycles in the stock market.更多还原