Enhanced Modeling and Parameter Estimation for Long-Horizon Return Predictability in Pure Jump Point Processes

• Activity Number: 344 - Methods in Financial Econometrics
• Type: Contributed
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Meng-Chen Hsieh* and Clifford Hurvich
• Companies: Rider University and New York University
• Keywords: return predictability; long memory; Cox Process; pure jump process; tick-level data; fractional Gaussian process

Abstract

There has been significant research interest on return predictability over the past decades. Results from prior research suggest that returns are predictable over long-run horizons. Moreover, recent financial literature shows that volatility is associated with long-term return predictability. Hsieh and Hurvich (2019) presented a coherent asset price model based on a pure jump process for returns and variance. They proved conditions under which there exists long-run return predictability. In this work, we extended previous results with the following three key contributions: (1) we generate the model by allowing for a flexible time lag in the underlying fractional Gaussian process, (2) we introduce an estimation procedure for model parameters and, (3) we propose a Monte-Carlo simulation-based hypothesis test on return predictability. Our hypothesis test shows an improved power on detecting return predictability while maintaining the nominal size compared with the test based on a non-parametric framework.