评价贝叶斯向量自回归模型在区域经济预测中的表现

Evaluation on Performances of the Bayesian Vector Autoregression Model in Regional Economic Forecasting

贝叶斯向量自回归(BVAR)利用先验的统计信息能够克服时间序列数据较短的困扰,理论上在我国区域经济预测中应该具有良好的效果。绝大多数区域预测模型文献缺乏"真正"意义上的样本外预测误差评价研究,但我们早期对民族八省区主要经济指标2010—2015年的预测为本文详细评价BVAR模型实际预测误差提供了绝佳的机会。以民族地区为例,本文的分析表明,BVAR模型的预测误差非常小,预测能力令人非常满意。同时本文也分析并指出进一步提高BVAR模型预测精度的努力方向。

Using prior information, the Bayesian Vector Autoregression （BAVR） model can overcome the problem of insufficient time series observation, and in theory, have better performances in regional economic forecasting. In literature on regional economic forecas- ting, most papers failed to evaluate alternative model＇s real out-of-sample forecasting errors. But our previous research, forecasting main economic variables of the Minority Areas from 2010 to 2015 with BVAR, provides a rare opportunity to do such evaluation study. Based on evidences from the Minority Areas, this paper demonstrates that the BVAR model has superior forecasting performance. Also, this paper points out the ways of how to improve regional BVAR model＇s forecasting precisions.