HOW I TESTED IT
Rank 45 perpetuals by 5-day return. Long the top 5, short the bottom 5, hold 15 days, rebalance. Equal weight, dollar-neutral.
The out-of-sample control is the one that had destroyed everything before it: fix the settings on old data, then run them on data the search never touched. In-sample +1.711% per rebalance, out-of-sample +1.801%. The mean held.
It also survives costs far worse than real: still +0.48% per rebalance even at a 1.50% round trip, six times what Binance actually charges.
THE NUMBERS
Mean per rebalance: +1.737%
In-sample: +1.711%
Out-of-sample: +1.801%
Win rate: 55.6%
Being long and short at the same time means the market's direction cancels out. What is left is whether the winners keep beating the losers - which is a real question, not a bet on the market going up.
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 $BNB $BTC
#Binance #Futures #Backtesting #TradingBot #CryptoData
Rank 45 perpetuals by 5-day return. Long the top 5, short the bottom 5, hold 15 days, rebalance. Equal weight, dollar-neutral.
The out-of-sample control is the one that had destroyed everything before it: fix the settings on old data, then run them on data the search never touched. In-sample +1.711% per rebalance, out-of-sample +1.801%. The mean held.
It also survives costs far worse than real: still +0.48% per rebalance even at a 1.50% round trip, six times what Binance actually charges.
THE NUMBERS
Mean per rebalance: +1.737%
In-sample: +1.711%
Out-of-sample: +1.801%
Win rate: 55.6%
Being long and short at the same time means the market's direction cancels out. What is left is whether the winners keep beating the losers - which is a real question, not a bet on the market going up.
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 $BNB $BTC
#Binance #Futures #Backtesting #TradingBot #CryptoData
