Page created: 2023.04.07
Last updated: 2023.12.03
“3 traders”
Image extraction: DALL・E2
ーーーーーーーーーーーーーー
ChatGPT Specification History:
2023.05.22 – current
ChatGPT4 (+Browse with Bing)
2023.04.24 – 05.19
ChatGPT3.5 (via ChatSonic)
ーーーーーーーーーーーーーー
Addendum (2023.09.29):
The measurement was not possible from the 2nd week of July to the 5th week of September because the browsing function of ChatGPT4 was stopped, but it resumed in the 1st week of October after “Browse with Bing” was restored.
ーーーーーーーーーーーーーー
Addendum (2023.05.21):
Starting from the measurements conducted between May 22nd & 26th, 2023, ChatGPT4+Browsing is being used. Now that ChatGPT can access the latest information (as of May 21st, 2023, still in beta), there is no longer a need to rely on other services (although ChatSonic was user-friendly).
ーーーーーーーーーーーーーー
As of April 2023, ChatGPT is a hot topic worldwide, and I personally find it interesting, so I decided to try it out by having it provide weekly predictions for the movements of US stocks, exchange rates, and cryptocurrencies.
ChatGPT is defined as a generative AI that calculates and connects the next possible words and apparently read up to 5 trillion words, and it has rapidly spread to the general public worldwide over the past month. However, it seems that ChatGPT does not read so-called “latest information” such as yesterday’s or today’s economic information, so I wondered if it would be impossible to predict stock price movements.
Nonetheless, I learned that a generative AI called Writesonic reads the latest information (Google’s data) and can be used in conjunction with ChatGPT through an API, so I decided to give it a try. By the way, Writesonic’s chat is called ChatSonic.
Therefore, it should be called “ChatGPT & ChatSonic economic forcast” rather than “ChatGPT economic forcast,” but the name is too long, and since it is based on ChatGPT, I will call it “ChatGPT economic forcast” for now.
However, even if you ask something like “Tell me next week’s stock prices”, it doesn’t give a clear answer. Of course, they can’t predict it. It is impossible to perfectly foresee the future of stock prices and exchange rates, which are formed by incorporating all events currently happening in the world and those that will occur during the forecast period, even if you read 5 trillion words of past events and words. The future is always uncertain.
However, it might be possible to say something like “the possibility of this happening is high” based on assumptions such as “if there are no major events” or “based on past patterns.” I have a faint hope in this regard and will start taking statistics first.
Method of statistics
As a statistical method, judge whether to buy or sell based on the following 4 items:
SPX500
EUR/USD
JPY/USD
BTC/USD
assuming that it holds a position at the beginning of the week and settle at the end of the week.
As for the timing of questions to AI, I ask before the Australian market opens within the previous weekend (Saturday or Sunday).
Regarding BTC/USD, there is no opening or closing price, but the same timing as the other 3 items is used as the measurement period.
Judgment of the result is that “buy” is correct if it has risen even by 1 cent. “Sell” is correct if it has dropped even by 1 cent. It will be a draw if it is the same.
The chart to be used is the MT4 of Thinkforex that I use for foreign exchange trading.
The results will be published on the score page.
To be fair, the method of extracting answers from AI is quite forceful. It starts with “You are a hedge fund trader” and instructs, “You must have a position.” However, since this is just a pastime, please forgive any complaints, etc.
…If it turns out to be “not good at all” after six months or a year, or if other tools that attract more interest than ChatGPT appear, or if I simply get tired of it, I will naturally discontinue it.