Visualizing Earnings to Predict Post-Earnings Announcement Drift: A Deep Learning Approach

Jon A. Garfinkel; Paul Hribar; Lawrence Hsiao

doi:10.2139/ssrn.5040374

Back

Visualizing Earnings to Predict Post-Earnings Announcement Drift: A Deep Learning Approach

Working paper

Open access

Visualizing Earnings to Predict Post-Earnings Announcement Drift: A Deep Learning Approach

Jon A. Garfinkel, Paul Hribar and Lawrence Hsiao

SSRN

12/02/2024

DOI: 10.2139/ssrn.5040374

Files and links (1)

url

https://doi.org/10.2139/ssrn.5040374View

Open Access

Abstract

We examine the potential of a deep learning model to use visualized earnings data to predict post-earnings announcement drift. After transforming quarterly earnings time series into bar chart images, we employ a convolutional neural network (CNN) to detect patterns and features within these visualizations that correlate with post-announcement drift. Out-of-sample tests reveal that the CNN-identified features significantly predict post-announcement returns, outperforming traditional drift predictors. This predictive capability remains consistent over time, is not accounted for by existing risk controls or known return anomalies and is robust across various model configurations. Our findings highlight the promise of applying AI to visualized financial data as a novel approach to predicting earnings changes and equity returns.

Details

Title: Subtitle: Visualizing Earnings to Predict Post-Earnings Announcement Drift: A Deep Learning Approach
Creators: Jon A. Garfinkel - University of Iowa, Tippie College of Business
Paul Hribar - University of Iowa, Accounting
Lawrence Hsiao - National Taiwan University
Resource Type: Working paper
DOI: 10.2139/ssrn.5040374
Publisher: SSRN
Number of pages: 69 pages
Language: English
Date posted: 12/02/2024
Academic Unit: Accounting; Finance
Record Identifier: 9984770889302771

Metrics

28 Record Views