From Pixels to Profits: Trading Arbitrage Portfolios Based on Image Representations

43 Pages Posted: 26 Oct 2023 Last revised: 11 Jan 2024

Date Written: December 16, 2023

Abstract

This paper explores a novel approach to statistical arbitrage by utilizing Convolutional Neural Networks (CNNs) to predict directional shifts in excess returns of arbitrage portfolios, which are constructed based on multifactor models. Using image representations of historical return co-movements to identify nonlinear predictive relationships, the study applies CNNs to extract relevant geometrical return patterns from the data. The empirical results illustrate that the proposed image-based arbitrage strategies yield significant excess returns, which are not explained by common risk factors. Further investigations into the sources of these excess returns - namely omitted factor momentum, leverage and margin constraints, and lottery demand - do not conclusively account for the observed profits.

Keywords: CNN, machine learning, images, finance, representation

Suggested Citation

Paluszkiewicz, Niklas, From Pixels to Profits: Trading Arbitrage Portfolios Based on Image Representations (December 16, 2023). Proceedings of the EUROFIDAI-ESSEC Paris December Finance Meeting 2023, Available at SSRN: https://ssrn.com/abstract=4612557 or http://dx.doi.org/10.2139/ssrn.4612557

Niklas Paluszkiewicz (Contact Author)

Ulm University ( email )

Albert-Einstein-Alee 11
Ulm, D-89081
Germany
+49 731 5023599 (Phone)

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
220
Abstract Views
475
Rank
262,046
PlumX Metrics