Network-Based Financial Forecasting: A Statistical and Economic Analysis

Abstract

One of the main challenges facing researchers and industry professionals for decades is the successful prediction of asset returns. This paper enriches this endeavor by an in-depth analysis of topological metrics of correlation networks applied to financial forecasting. While academic research often focuses on statistical performance metrics, industry professionals are more interested in the economic value-added of competing forecasting approaches. Since statistical significance does not automatically imply economic significance, this article devotes attention to both types of performance metrics. We show that the benchmark mean model is indeed difficult to beat when it comes to statistical performance metrics. However, considering economic metrics, network-based predictors generate a clear value-added, which also applies to the multi risky asset allocation dimension.