Reproduce “Integrating LSTM Networks with Neural Levy Processes for Financial Forecasting” on OpenResearch

About this paper

Mohammed Alruqimi, Luca Di Persio

Researchers from the University of Verona developed a hybrid forecasting framework combining Grey Wolf Optimizer-tuned LSTM networks with stochastic financial models. The LSTM-L´evy model, calibrated using an Artificial Neural Network, consistently achieved superior prediction accuracy with R² values up to 0.998 across three real-world financial datasets.

Key insights

• Hybrid models that strategically combine deep learning (LSTMs) with robust stochastic financial processes (like L´evy-Merton) can achieve superior predictive performance by leveraging the strengths of both paradigms.
• Artificial Neural Networks (ANNs) are a highly effective and computationally efficient method for calibrating complex stochastic financial model parameters, outperforming metaheuristic algorithms and specialized SDE solvers.
• The L´evy-Merton Jump-Diffusion model, with its capacity to account for sudden price jumps and heavy tails, provides a more accurate representation of real-world asset price dynamics when integrated with deep learning.

Summary by alphaXiv.