🏷️#Backtesting
4 articles
What Happens When You Split a Random Technical Strategy Across 1, 3, 5, or 15 Accounts? Measuring the Effect of Diversification Alone With 5.7 Years of Real Data
Each account trades once a day, risking 2% per trade at RR 1:1. Holding this fixed, we ran the same signal split across 1 account, 3 accounts (by strategy), 5 accounts (by weekday), and 15 accounts (strategy × weekday) over 5.7 years of real data from 2021. The strategy itself loses — a 47.8% win rate with negative expectancy. Even so, splitting into 5 or more accounts kept every one of the 6 years in the black, while the 1-account version had a losing year. Of the 90 account-years across the individual accounts, 50 were losing years — yet bundled together, the losing years disappear. That's the effect of diversification. Finally, moving only the win rate with the same setup: 55% wins gives ¥17.34M a year, while 45% gives just ¥1.1M and losing years return. Diversification firms up the floor; whether you get a big upside comes down to edge.
Is a Challenge "Expected-Value Positive Just by Buying It"? I Built a Coin-Toss EA and Backtested It 16 Times [Analysis]
Placing SL/TP at spread x 40 makes the cost ratio independent of the currency pair. Verified with an exact solution plus 24 real EA runs, a single challenge's expected value comes out to +¥119,000 against a ¥108,800 fee (95% CI +¥18,000 to +¥381,000). Raising the number of daily attempts up to the daily loss limit shrinks the time to resolution from 139 days to 7 days without lowering the pass rate, and across multiple accounts, simply using a different random seed per account shifts the probability of finishing positive from 56% to 98%. The one remaining hole is a stop-loss getting jumped by a gap.
The Consistency Rule Specifically Targets Trend-Following: What 10,000 Runs on a Real EA Showed [Opinion]
We connected a real EA to Fintokei's server and compared over-trading, trend-following, and one-sided betting with risk sized equally. Without a consistency rule, trend-following wins outright (70.4%), but the more profit concentrates in big wins, the harder it gets cut down — the purest trend-following setup dropped to a 0.0% pass rate at a 20% threshold. This gives a structural explanation for the observation that "winners cluster at firms with loose consistency rules."
Does Averaging Down Beat the Consistency Rule? What a Real EA Backtest and a Rulebook Review Actually Show [Opinion]
Averaging down (nanpin/DCA) turned out to be genuinely immune to the consistency rule. But at a 13.2% pass rate, it's useless on its own. Adding a stop-loss raises that to 41.2% while keeping the immunity — but it still falls short of trend-following's 52.9%. Checking the actual rulebooks, what's banned isn't averaging down itself — it's doubling your lot size.