Cutting Through the Noise: A Deep Learning-Enhanced Particle Filter for High-Frequency Market Forecasting

Christos Konstantinou*, Gbenga Ibikunle, Khaladdin Rzayev, Yi Cao

*Corresponding author for this work

Research output: Working paper

Abstract

We introduce a deep learning-enhanced particle filter designed to reduce market microstructure noise in high-frequency stock prices. Unlike traditional approaches, our method does not rely on conventional assumptions about price dynamics. By tailoring noise filtration strategies to specific characteristics of financial instruments, we demonstrate significant variations in prediction performance across market environments: while ML significantly improves forecasting accuracy for stocks with higher market microstructure noise, it adversely affects predictive capabilities in low-noise settings by extracting signal rather than noise. These contrasting outcomes, statistically significant across noise quartiles, highlight the importance of selective application of noise reduction techniques in financial forecasting.
Original languageEnglish
Publication statusUnpublished - 2024

Fingerprint

Dive into the research topics of 'Cutting Through the Noise: A Deep Learning-Enhanced Particle Filter for High-Frequency Market Forecasting'. Together they form a unique fingerprint.

Cite this