In the below case study, 3 AI agents cooperatively studied around a month of historical Bitcoin prices and then collectively produced an action sequence (buy or sell) for the following week, resulting in odds consistently greater than 60%. For more details, please read more here:
www.xinvisionq.com/subpages/caseStudies.html.
Thats a huge result of 60% but can we really trust AI to compete with human in regards to predictive movement of the market? I mean yes trading is hard and has historical results and thats also our basis as trader, does it mean we can rely fully on them maybe if not now more than ever in the future.
In comparison to humans, AI agents don't "think" all they do is compute, but that isn't a bad thing necessarily because they don't have any emotions. We humans tend to overthink things too much, basically we're rational and irrational, and this doesn't play out to our advantage especially for trading which is just being able to make decisions constantly under incomplete information. No matter how hard we try, nobody can accurately predict the absolute price of crypto at any given time, so what's more important is finding the market's overall trend of the recent past, and somehow when the multiple AI agents cooperate and learn the historical trading data they are able to come up with a trading strategy to produce decent short horizon forecasts. The emphasis here is studying more recent historical data, around 30 data points to then predict the foreseeable near future of about 5-7 data points. Basically, the longer the forecast horizon the odds will merge closer to 50-50, which is exactly consistent with the well-known random walk of the market.