Cointegration and Long-Horizon Forecasting
研究协整变量的预测，发现长期预测中忽略协整关系不会降低标准多变量预测精度，简单单变量Box-Jenkins方法同样准确，并指出标准精度指标忽视协整关系维护的缺陷，提出替代方案。
We consider the forecasting of cointegrated variables, and we show that at long horizons nothing is lost by ignoring cointegration when forecasts are evaluated using standard multivariate forecast accuracy measures. In fact, simple univariate Box–Jenkins forecasts are just as accurate. Our results highlight a potentially important deficiency of standard forecast accuracy measures—they fail to value the maintenance of cointegrating relationships among variables—and we suggest alternatives that explicitly do so.