The Merit of High-Frequency Data in Portfolio Allocation

Abstract:     
This paper addresses the open debate about the effectiveness and practical relevance of high-frequency (HF) data in portfolio allocation. Our results demonstrate that when used with proper econometric models, HF data offers gains over daily data and more importantly these gains are maintained over longer horizons than previous studies have shown. We propose a Multi-Scale Spectral Components model for forecasting high-dimensional covariance matrices based on realized measures employing HF data. Extensive performance evaluation confirms that the proposed approach dominates prevailing methods and validates the intuition that HF data used properly can translate into better portfolio allocation decisions.

Number of Pages in PDF File: 43

Keywords: spectral decomposition, mixing frequencies, factor model, blocked realized kernel, covariance prediction, portfolio optimization

JEL Classification: G11, G17, C58, C14, C38

working papers series