主讲人：

梁润（北京大学汇丰商学院高级研究员）

主持老师：

（北大经院）王熙

参与老师：

（北大经院）王一鸣、刘蕴霆

（北大国发院）沈艳、黄卓、孙振庭、张俊妮

（北大新结构经济学研究院）胡博

时间：

2021年5月7日（周五）

10：00-11：30

地点：

经济学院107会议室

主讲人简介：

梁润，现任北京大学汇丰商学院高级研究员，曾担任IMF驻华代表处兼职经济学家，金融机构高级研究员，上海财经大学助理研究员。他的主要研究方向为应用经济学，研究领域为宏观经济预测，以及老龄化与人力资本投资。他硕士期间曾在国际顶级期刊发表过多篇物理学文章，博士转为经济学研究。2012年博士毕业后，他在International Journal of Economic Theory，经济研究、金融研究等国内外知名期刊上发表过多篇文章，并多次获得“远见杯”宏观经济预测前三名。

摘要：

We construct nowcasts and forecasts of China’s PPI inflation using a large panel of data series with different frequencies (monthly, ten-day, weekly, daily). Mixed frequency data are incorporated in the dynamic factor model in two approaches: one is to convert high frequency data to low frequency, which is monthly in our example, and apply the expectation maximization (EM) algorithm in estimation; the other is to treat low frequency data as high frequency (daily) data with missing observations in a specific pattern and apply Banbura and Modugno (2014) to estimate the dynamic factor model with missing observations. We compare the forecast accuracy of these two approaches with some other alternative forecasting models, such as random walk, univariate autoregressive model and dynamic factor model with only monthly data. Our empirical results show that the first approach outperforms other models in most cases and horizons. Models utilizing high frequency data generally perform better than those do not. As high frequency information flows in, the forecasting accuracy improves substantially.