The benefits of forecasting inflation with machine learning: New evidence
复制并扩展了Medeiros等人（2021）的研究，在美、加、英三国数据上比较多种机器学习方法预测通胀的效果，发现随机森林在疫情期间表现不佳，而随机波动模型和梯度提升方法更准确。
Summary Medeiros et al. (2021) (Journal of Business & Economic Statistics, 39:1, 98–119) find that random forest (RF) outperforms US inflation forecasting benchmarks. We replicate the main results in Medeiros et al. (2021) and (1) considerably expand the set of machine learning methods, (2) analyse the predictive ability of both the initial and extended sets of methods on Canadian and UK data, (3) add results on coverage rates and widths of prediction intervals and (4) extend the sample from January 2016 to October 2022. Our narrow replication confirms the main findings of the original paper. However, the wider replication results suggest that other methods are competitive with RF and often more accurate. In addition, RF produces disappointing results during the coronavirus pandemic and subsequent high inflation of 2020–2022, whereas a stochastic volatility model and some gradient boosting methods produce more accurate forecasts.