Forecasting With Stable Seasonal Pattern Models With an Application to Hawaiian Tourism Data
提出稳定季节模式模型的多种变体，分别处理计数和连续数据，并利用条件独立性分离季节与趋势进行预测。在夏威夷旅游数据上，一种变体在长期预测中优于季节性ARIMA和传递函数模型。
We propose several variations of the stable seasonal pattern (SSP) model first introduced by Marshall and Oliver and study their prediction procedures. Depending on the type of data (count data or continuous variable), we propose different treatments. Previously SSP models have been applied to trendless data or adapted to trending data in ad hoc ways. In the models considered here, conditional independence allows the seasonal pattern and trend to be modeled separately, whereas prediction uses both efficiently. In an out-of-sample forecasting experiment conducted on Hawaiian tourism data, one of the proposed variations demonstrates its long-term forecasting potential relative to seasonal Box–Jenkins autoregressive integrated moving average and transfer function models.