Survivorship bias, the silent killer.
It’s the single most common reason a crypto backtest lies to you — and it’s nearly invisible. Here’s what survivorship bias is, why crypto suffers from it worse than any market, and what we do about it.
Only asking the survivors
The classic image: in WWII, engineers studied returning bombers and proposed armouring the spots that were most often hit. A statistician pointed out the mistake — the planes hit elsewhere never came back. They were studying only the survivors, and drawing exactly the wrong conclusion.
Crypto backtests make the same error constantly. If you test a factor on “the coins trading today,” you’ve quietly deleted every coin that delisted, got abandoned, or went to zero. Your sample is a curated list of winners — and almost anything looks profitable on a list of winners.
An enormous graveyard
Equities have survivorship bias too, but crypto’s is extreme. Thousands of tokens have died — failed projects, exploited protocols, exit scams, exchange delistings. The set of coins still trading is a heavily filtered minority of survivors, which means a factor measured on it can look strong for a reason that has nothing to do with predictive power.
On a universe of survivors, “buy the risky ones” always looks smart — because the risky ones that blew up aren’t in your data.
Bring the dead back into the test
The fix is to put the failures back. We reconstruct a graveyard of delisted and collapsed coins from exchange archives and re-run validation with them included, handling their deaths honestly (the loss is realised, returns are capped, illiquid names are floored). Then we ask: does the factor survive when the dead are in the room?
- Funding crowding survived — its tradeable spread was essentially unchanged with dozens of dead peers added (attenuated, but intact and still positive in both regimes).
- Low volatility did not — its “edge” was largely a survivorship mirage, and its apparent strength came from shorting dying microcaps you could never actually trade. (The full story.)
Clean the data before adding factors
Until a result survives a survivorship correction, we treat it as directional, not proof — and we say so. Cleaning the universe comes before stacking factors, every time. It’s unglamorous, it kills good-looking ideas, and it’s exactly why the one signal we kept is worth trusting.
Common questions
What is survivorship bias in simple terms?
It is what happens when you study only the things that lasted. If you only ask survivors whether something was safe, you miss everyone it killed. In backtests, testing only on coins that still trade today deletes every coin that failed — so the average looks far better than reality.
Why is survivorship bias worse in crypto?
Because crypto has an enormous graveyard. Thousands of tokens have been delisted, abandoned, or collapsed to zero. A universe of "coins trading today" is a heavily filtered set of winners — so any factor measured on it can look strong for the wrong reason.
How does Ioxer handle it?
We re-run validation with delisted and collapsed coins added back from exchange archives, and we treat survivor-only results as directional, not proof. Funding crowding survived that correction (attenuated but intact); low volatility did not — its edge was largely a survivorship mirage.
Ioxer is research, not investment advice. IOX is a crowding read — not a price prediction, not a buy/sell signal.