Artificial Intelligence Models Analyze Financial Markets and Investor Behavior
Researchers are running a project that adapts Google's BERT architecture to examine financial portfolios and predict investor behavior and stock returns.
Researchers from Chicago Booth, Princeton, Harvard, and NYU have launched the Market Genome Project, which applies transformer-based artificial intelligence models to financial portfolios in order to understand market dynamics and asset prices.
Market Genome Project and Artificial Intelligence
Following the research paper on attention mechanisms published in 2017, artificial intelligence models were adapted for capital markets. Ralph S. J. Koijen from Chicago Booth, Motohiro Yogo from Princeton, Xavier Gabaix from Harvard, and Robert J. Richmond from NYU launched a joint study within this scope.
The Role of the Demand System in Asset Prices
Koijen and Yogo began developing the demand system approach in 2015. Examining institutional stock data, the researchers determined that demand shifts account for more than 80 percent of the differences in stock returns.
BERT Architecture Adapted for Portfolios
To analyze fund data, the researchers adapted Google's BERT transformer architecture for portfolios. By masking stocks in portfolios during training, the model gained the ability to predict missing assets.
Results Differing from Traditional Methods
The asset-based model developed outperformed traditional firm characteristics in explaining relative valuation and return movements. The model captured true demand patterns beyond standard industry classifications.