We leverage the capabilities of GPT-3 to generate historical business descriptions for over 63,000 global firms . Utilizing these descriptions and advanced embedding models from OpenAI, we construct time-varying business networks that represent business links across the globe. We showcase the performance of these networks by studying the lead–lag effect for global stocks and predicting target firms in M&A deals. We demonstrate how masking firm-specific details can mitigate look-ahead bias concerns that may arise from the use of embedding models with a recent knowledge cutoff, and how to differentiate between competitor, supplier, and customer links by fine-tuning an open-source language model .

Motivated by McLean and Pontiff (2016), we study the pre- and post-publication return predictability of 241 cross-sectional anomalies in 39 stock markets. We find, based on more than two million anomaly country-months, that the United States is the only country with a reliable post-publication decline in long-short returns. Collectively, our meta-analysis of return predictors suggests that barriers to arbitrage trading can create segmented markets and that anomalies tend to represent mispricing instead of data mining.