报告题目：
Combining Forecasts from Multiple Experts for Multiple Variables
报 告 人：
Prof.Long Zhao, National University of Singapore Business School
报告时间：2024年12月26日（周四）10:30~12:00
报告摘要：
We address the challenge of combining point forecasts from multiple experts across multiple variables to draw inferences about an unknown variable. A significant amount of prior research has focused on aggregating forecasts from multiple experts about a single variable into a consensus forecast (we refer to this as the separate inference). However, in practice, each expert frequently provides a forecast for each variable across multiple variables. Because the forecast errors can be correlated across the variables, the decision-maker may additionally benefit from pooling forecasts for the other variables (we refer to this as the pooled inference). In our model, we assume that the forecast error consists of three sources of randomness: an error due to the variable-specific factor, an error due to the expert-specific factor, and an idiosyncratic noise. When the covariance structure of the forecast errors is known, we find that the pooled inference is structurally equivalent to the separate inference but with a smaller idiosyncratic noise, thereby allowing the decision-maker to weigh the forecasts from heterogeneous experts more judiciously. Despite this benefit, relative to the separate inference, the value of additional information from pooling is not always strictly positive: It is equal to zero when experts are exchangeable (so there is no need to weigh forecasts differently) and also when there is no idiosyncratic noise (so there is no further room to reduce the idiosyncratic noise). Empirically, we show that pooled inference can outperform benchmark aggregation methods that are based on separate inference, even when it involves estimating a large covariance matrix. Drawing from the empirical results, we further provide decision-makers with practical guidelines on when to choose pooled inference.
报告人简介：
Long Zhao is an assistant professor from NUS business school. His research focuses on data-driven decision making with applications in forecast aggregation, portfolio optimization, and terrestrial water cycle. His work has been published on Operations Research and Proceedings of National Academy.
欢迎有兴趣的师生积极参加！