I spent ~85 USD of GPU time testing an AI price model on 150 coins. It found real signal. It didn't prove any extra money.
The model: Kronos, an AI trained on candle data. I fine-tuned it and tested it on months it had never seen.
What it got right: in both test periods it ranked tomorrow's winners above the losers better than chance (t-stats 3.3 and 5.2).
What it didn't do was beat a simple baseline. I built the same strategy from plain momentum + volatility, then added the AI's forecasts. After costs, in mid/small caps: +26.2 vs +28.7 bp/day. The difference, +2.5, has a range from −14 to +18. That's noise.
So "the AI predicts the market" can be true and still not pay. Always ask: better than what, after fees?
Have you ever seen an AI trading tool compared to a simple baseline? 👇
Not financial advice. Research test, no live money.
#Aİ #TradingBots #algoTrading #tradingfees $BTC
The model: Kronos, an AI trained on candle data. I fine-tuned it and tested it on months it had never seen.
What it got right: in both test periods it ranked tomorrow's winners above the losers better than chance (t-stats 3.3 and 5.2).
What it didn't do was beat a simple baseline. I built the same strategy from plain momentum + volatility, then added the AI's forecasts. After costs, in mid/small caps: +26.2 vs +28.7 bp/day. The difference, +2.5, has a range from −14 to +18. That's noise.
So "the AI predicts the market" can be true and still not pay. Always ask: better than what, after fees?
Have you ever seen an AI trading tool compared to a simple baseline? 👇
Not financial advice. Research test, no live money.
#Aİ #TradingBots #algoTrading #tradingfees $BTC
