The strategies that remained all made money—so why did the original pool of capital still lose?
When looking at performance, I start with a question that’s often overlooked: What happened to the strategies that stopped being shown?
Here’s a completely fictional example. Ten strategies each start with 10,000 yuan on the same day, with no additional funds added. Five gain 20% and remain on display; the other five lose 30%, then shut down, with their remaining cash held until the end of the observation period.
Looking only at the five that remained, the average return is 20%. But if we include all ten from the start, the total capital falls from 100,000 yuan to 95,000 yuan—a 5% overall loss. None of the surviving performance curves was falsified; the displayed sample changed, and the conclusion reversed.
This is called survivorship bias. A preprint published in March 2026 on Indian small-cap stocks also compared backtest results using current constituents with those using historical constituents. Its specific findings can’t be directly applied to crypto, but they remind us to keep track of samples that exit.
When evaluating BTC, ETH, and SOL trading strategies, I care more about whether we can reconstruct the complete list of strategies available at the time, and their actual gains or losses when they stopped running. If exit records can’t be found, we can’t claim that the average of the strategies shown today represents the result of picking one at random back then.
Fix the starting point and sample first, then compare returns. The second image is a photo of trading research materials; neither the person nor the screen is proof of live trading by this account.
$BTC $ETH $SOL
Tap my profile picture to view live copy-trading performance
When looking at performance, I start with a question that’s often overlooked: What happened to the strategies that stopped being shown?
Here’s a completely fictional example. Ten strategies each start with 10,000 yuan on the same day, with no additional funds added. Five gain 20% and remain on display; the other five lose 30%, then shut down, with their remaining cash held until the end of the observation period.
Looking only at the five that remained, the average return is 20%. But if we include all ten from the start, the total capital falls from 100,000 yuan to 95,000 yuan—a 5% overall loss. None of the surviving performance curves was falsified; the displayed sample changed, and the conclusion reversed.
This is called survivorship bias. A preprint published in March 2026 on Indian small-cap stocks also compared backtest results using current constituents with those using historical constituents. Its specific findings can’t be directly applied to crypto, but they remind us to keep track of samples that exit.
When evaluating BTC, ETH, and SOL trading strategies, I care more about whether we can reconstruct the complete list of strategies available at the time, and their actual gains or losses when they stopped running. If exit records can’t be found, we can’t claim that the average of the strategies shown today represents the result of picking one at random back then.
Fix the starting point and sample first, then compare returns. The second image is a photo of trading research materials; neither the person nor the screen is proof of live trading by this account.
$BTC $ETH $SOL
Tap my profile picture to view live copy-trading performance