Can ChatGPT Generate Stock Tickers to Buy and Sell for Day Trading?

Abstract

This paper examines the generative feature of ChatGPT for empirical asset pricing. I show that ChatGPT can generate stock tickers that provide a profitable day trading strategy. Using input prompts as multiple Twitter posts, including both macro and firm-specific news by major news providers, I ask ChatGPT to generate lists of stock tickers to buy and sell. The trading strategy based on the buy and sell lists earns significant long-short returns in open-to-close intraday trading. By asking again about the reason for generating those stock tickers, keywords of ChatGPT’s answer suggest that tech stocks are important for generating the buy lists, whereas sector- or industry-level analysis is important for generating the sell lists. In particular, ChatGPT’s buy and sell lists consist of economically linked stocks through the supply chain, resulting in lower industry concentration than those of their matching groups. The performance is attributable to the stock selection within each industry, the short leg of the strategy, and stronger in the difficult-to-arbitrage stocks, implying that ChatGPT signals’ applicability of extracting mispricing signals in text data. As most of the Twitter data consists of non-firm-specific news, this finding sheds light on the literature by showing that ChatGPT can process a bulk of seemingly non-firm-specific news to generate firm-specific mispricing signals.

Keywords: ChatGPT, Empirical Asset Pricing

JEL Classification: G10, G11, G12, G14