“The institutional investor 11”
Image extraction: DALL・E2
The “ChatGPT Economic Forecast” was an initiative that involved having ChatGPT, a leading generative AI, make market predictions and then measuring the financial outcomes. Despite some periods being immeasurable, the project was conveniently conducted for approximately one year in total.
It spanned from the week of April 10-14, 2023, to the week of April 1-5, 2024.
The experiment repeated a question starting with “You are a hedge fund trader…” every weekend.
The direct motivation was a desire to play with the novelty of generative AI. Additionally, there was an interest in assessing whether investment information from major, reputedly reliable websites available online was indeed valuable.
Defining ChatGPT as a “machine that collects information and generates a summarized text” (at least that was the only use in the ChatGPT economic sentiment), the act of asking ChatGPT to predict weekly market fluctuations essentially boiled down to the question:
Can investments based solely on information from major websites be profitable?
The conclusion was “not really profitable.”
In summary,
US stocks: Slightly negative
Forex: Break-even
Crypto: Positive
The results weren’t bad but not great either.
For example, U.S. stocks are almost a tie for the negative, but in any case, it is losing out on index investments. Currencies have not been profitable either.
The positive outcome for crypto was largely due to riding the bullish market that began in February 2024. The performance from the previous year to the beginning of this year was significantly negative, to the point of being unbearable.
For Bitcoin, a win-loss record of 16-20 suggests a commendable effort, but the extreme volatility in prices means that buying at the beginning of the week and selling at the end is not advisable.
From the perspective of a trader like me, who considers not setting a stop-loss in swing trading unthinkable, this amounts to excessive risk-taking and is akin to gambling.
ChatGPT’s Prediction Tendency
This seems to reflect the tendency of major sites when forecasting the market. The results suggest that while ChatGPT can be somewhat accurate in trending markets, it fails in range-bound markets.
In essence, the predictions were bland and non-committal.
However, it’s worth noting that, by nature of the experiment, ChatGPT was almost always forced to take a position weekly. Even in situations where the initial response was indecisive between buying or selling, further probing often forced a decision.
Thus, the lack of success in range-bound markets might be somewhat forced but, given the experimental rules, this conclusion (albeit aggressively) is accepted.
This tendency to perform well in trending markets but not in range-bound ones is common among most traders.
Conclusion
Excluding lucky trades in BTCUSD, the conclusion is that ChatGPT’s forecasts are not useful for investing. Or rather, for such results, it’s unnecessary to consult a chatbot.
Nonetheless, this was an expected outcome, and a year’s worth of data was desired for a satisfactory conclusion, hence the continuation for a year.
In any case, the outcome perfectly suits this blog, dubbed “A Personal Trader’s Failure Notebook.”
Investing isn’t so easy that following machine or major website forecasts can yield profits – a conclusion that’s probably obvious without needing to be told, but it was proven over a year.
Looking Forward
The results from April 2023 through a year, influenced by the bullish market environment for US stocks, indeed arouse curiosity about what would have happened in a correction phase.
Similarly, there’s slight curiosity about how the performance would continue during the crypto market’s upward trend.
However, there seems no need to continue this method or anything similar, unless a new kind of “play” idea arises for collecting data through a different approach.
For instance, there’s been mention that ChatGPT (or similar generative AI) might be well-suited for selecting stock picks.
(Reference: https://www.finder.com/uk/ai-investing)
Yet, that alone doesn’t inspire a desire to gather such data, as it seems it would only replicate this experiment’s findings.
Even so, since it’s all in good fun, perhaps there’s no need to take it too seriously and just enjoy it as it is.
In any case, it’s great to live in an era where high-quality toys like ChatGPT are easily accessible for enjoyment.
