HOW I TESTED IT
Momentum inverted in 2026. The obvious fix: watch recent performance, and flip to the opposite rule when momentum stops paying.
I tested that over six different lookback windows. The best one returned +1.783% per rebalance against a +1.737% baseline - a difference indistinguishable from noise. Every other window was worse. None repaired 2026.
The reason is simple: by the time trailing performance tells you the regime flipped, you have already taken the losses that told you.
THE NUMBERS
Baseline: +1.737%/reb
Best adaptive rule: +1.783%/reb
Difference: noise
Best fix for 2026: still -0.66%
Adaptive rules feel like the answer to regime change. They are usually just a slower version of the same bet, with the losses already paid for the signal.
Part of a public experiment: 5,063 USDT to 1,000,000 USDT, every result published - including the ones that make me look bad.
Run on $SOL $BTC
#Binance #Futures #Backtesting #TradingBot #CryptoData
Momentum inverted in 2026. The obvious fix: watch recent performance, and flip to the opposite rule when momentum stops paying.
I tested that over six different lookback windows. The best one returned +1.783% per rebalance against a +1.737% baseline - a difference indistinguishable from noise. Every other window was worse. None repaired 2026.
The reason is simple: by the time trailing performance tells you the regime flipped, you have already taken the losses that told you.
THE NUMBERS
Baseline: +1.737%/reb
Best adaptive rule: +1.783%/reb
Difference: noise
Best fix for 2026: still -0.66%
Adaptive rules feel like the answer to regime change. They are usually just a slower version of the same bet, with the losses already paid for the signal.
Part of a public experiment: 5,063 USDT to 1,000,000 USDT, every result published - including the ones that make me look bad.
Run on $SOL $BTC
#Binance #Futures #Backtesting #TradingBot #CryptoData
