The AI's performance was tested by feeding it with tweets and asking it to generate stock tickers to buy and sell for day trading. The results showed a monthly return of 0.546% after adjusting the market factor, and by strengthening the signals by netting the buy and sell signals, the strategy based on the improved signals earned 3.717% per month. However, the study also revealed that the AI's output may be biased towards large caps and negative ones about small caps, reflecting the media it's trained on. The study concludes that while AI day-trading may sometimes work under narrow conditions, its stock picks may not necessarily be better than random.
Key takeaways:
- Sangheum Cho, of the World Bank’s Development Research Group, has been using ChatGPT to pick buy-sell portfolios for day trading with significant success.
- ChatGPT's stock recommendations are based on the prevailing mood from tweets from Bloomberg and the Wall Street Journal, with only 7% of the tweets mentioning specific shares.
- Despite some inconsistencies and potential biases, ChatGPT was able to generate a trading strategy that earned 0.546% per month after adjusting the market factor, and up to 3.717% per month with improved signals.
- However, the study also highlighted some limitations of using AI for day trading, including the inability to ask for reasoning behind picks, the need for significant sanitising of portfolios, and potential biases in the AI's training data.