汇率决定模型的样本外预测能力研究:基于分位数回归方法

Out of sample predictability of exchange rate model :based on quantile regression

汇率决定模型的有效性问题一直受到学界的质疑。通过改变汇率决定模型的参数估计方法,汇率决定模型的预测效果会显著改善。本文针对G-7国家从1982年至2012年的数据,采用分位数回归法来估计汇率决定模型的三种基本形式,并将汇率决定模型的样本外预测能力与汇率的自回归模型进行了比较。研究结果表明,汇率决定模型在样本外预测能力上优于汇率的自回归模型,汇率决定模型是有效的。在汇率的不同分位数上,各个影响因素对汇率的影响程度不一样,汇率过度贬值或过度升值时,要重新考虑四个因素的影响程度及其对应的汇率调整政策。

Validity of exchange rate determination theory is always doubted in the international economics. By changing the parameter estimation method of exchange rate determination model, its predictability will be improved significantly. This paper collects the data for G7 currencies from 1982 to 2012, and compares the out-of-sample predictability of three kinds of exchange rate determination models with an autoregression model by using quantile regression. The findings support the predictability of exchange rate structural models is super to an autoregression model, and these structural models are effective. The paper analyzes the determinants of exchange rates in G7 countries based on the quantile regression, and finds out that the impact of factors on exchange rate are very different under different quantiles of exchange rates. When exchange rates are over-depreciated or over-appreciated, we should reconsider the actual impact of these factors and the adjustment of exchange rate policies.