An article in Bloomberg this morning said that they asked ChatGPT to design a market-beating ETF, and it responded that it couldn't because stocks were too unpredictable. I disagree. Mungofitch has shown us numerous examples of mechanical investing systems that beat the market.

Some stocks are more predictable than others. Take JNJ for example, or BRK-A. If one plots the stock price of JNJ versus revenue per share or earnings per share over the last 65 years (log-log), one gets a straight line with high r^2. Further, if one plots revenues or earnings per share over that period versus year, one gets a smooth curve that can be extrapolated with reasonable certainty. Combining the two graphs gives an estimate of the stock price, say, five years in the future that while not always correct, is pretty close. Same thing for BRK-A if one uses stock price and book value per share.

I think if ChatGPT focused on stocks for which the stock price tracked revenues, earnings or book value closely, and for which revenues, earnings or book value changed smoothly over time, then it could predict future stock prices with reasonable accuracy. The stock price 5 or 10 years out would be more predictable than one year out. It's like predicting a major league baseball player's batting average. The prediction is more certain for consistent hitters, and more certain over the next 5 or 10 games than over the next one game.