KIT - Chair of Financial Economics and Risk Management Teaching - Research and Thesis - Supervised learning Projects - Deep Learning Statistical Arbitrage (Deep Learning)

Traditionally, statistical arbitrage involves trading strategies that exploit pricing inefficiencies between similar assets. These strategies rely on statistical models to identify temporary mispricings.
With advancements in machine learning, particularly deep learning, there's been a shift toward using data-driven methods for arbitrage. Deep learning models like transformers and convolutional networks are used to enhance predictive accuracy and improve trading decisions by identifying patterns in asset returns.
The key idea behind the paper is to exploit temporal price differences between similar assets by creating arbitrage portfolios (:= trades relative to mimicking asset portfolios.). These portfolios are formed as residuals from conditional latent asset pricing models, capturing the unexplained variation in asset returns after accounting for systematic factors.