Binomial Autoregressive Moving Average Models With an Application to U.S. Recessions
提出二项自回归移动平均模型，用于分析二元时间序列，可避免长滞后马尔可夫模型的维度灾难，并通过美国经济衰退数据验证其优于传统马尔可夫模型。
Binary Autoregressive Moving Average (BARMA) models provide a modeling technology for binary time series analogous to the classic Gaussian ARMA models used for continuous data. BARMA models mitigate the curse of dimensionality found in long lag Markov models and allow for non-Markovian persistence. The autopersistence function (APF) and autopersistence graph (APG) provide analogs to the autocorrelation function and correlogram. Parameters of the BARMA model may be estimated by either maximum likelihood or MCMC methods. Application of the BARMA model to U.S. recession data suggests that a BARMA(2, 2) model is superior to traditional Markov models.