Can AI Replace Stock Analysts? Evidence from Deep Learning Financial Statements
Study shows AI deep-learning models can outperform human analysts in predicting 12-month-ahead stock prices.
What it examines
This paper investigates whether AI, specifically neural-network-based deep learning, can outperform human equity research analysts in predicting 12-month-ahead stock prices using financial statement data, historical stock prices, and interest rates.
What it concludes
AI has the potential to outperform human analysts in stock price prediction, especially for firms with high investment and uncertainty. Future research could explore more complex models and additional data types to further enhance prediction accuracy.
Evidence objects
AI has the potential to outperform human analysts in stock price prediction, especially for firms with high investment and uncertainty. Future research could explore more complex models and additional data types to further enhance prediction accuracy.
key_findings bullet 1 · key_findings · validation V0
Raw abstract and provenance
1 Can AI Replace Stock Analysts? Evidence from Deep Learning Financial Statements G. Nathan Dong * ` March 31, 2024 ABSTRACT Equity research analysts have long been regarded as experts in predicting the performance of stocks. With the rapid advancemen ts in artificial intelligence (A I), the question arises as to whether neural-network based deep-learning te chnology can surpass human analysts in this task. In this research, we apply a model-free d eep-learning technique to equity research by training a simple neural-network (without pre-specified financial models or pre-defined accounting ratios) to predict 12-month-ahead st ock price. Our AI model that runs on an i.expensive computer equipment is capable of processing vast amounts of financial statement raw data over a 14-year period in a fraction of the time and cost it would take a human analyst ($1,450 in equipment, 15 hours in time, and $2.5 in electricity, to be exact).? It outperforms human
Source row: 351 · abstract type: unknown