What is mean reversion in crypto markets
A beginner-friendly intro to mean reversion — what it is, why it exists, and where to look for it in crypto.
If you've spent any time around quant trading, you've heard the phrase "mean reversion" thrown around. It sounds technical, but the core idea is older than computers and shows up everywhere in nature and markets: things that drift far from their average tend to drift back.
Crypto markets are usually associated with the opposite — wild trends, parabolic moves, capitulation bottoms. So why would anyone look for mean reversion here?
The answer: not every part of crypto trends. Inside specific relationships — pairs of correlated assets, ratios between similar tokens — mean reversion is alive and well. You just have to know where to look.
A simple example
Suppose you've been holding 1 BTC and 10 ETH for a year. Bitcoin moves up 50%; ETH moves up only 10%. The BTC/ETH ratio has stretched in BTC's favour.
Historically, when this ratio gets too extreme, it tends to come back. Maybe BTC stalls and ETH catches up. Maybe BTC pulls back and ETH holds. Either way, the ratio compresses.
If you swap some BTC into ETH at the stretched extreme, you don't need to predict the absolute price of either. You just need the ratio to compress — which it does, statistically, more often than not in liquid pairs that aren't in a major regime change.
That's mean reversion in a nutshell: betting on relationships returning to normal, not on absolute price moves.
Why does this work?
Two coins in the same sector — say two layer-1 alts, or two LSDs — tend to share fundamental drivers. Total crypto market sentiment, sector-specific narrative, dollar strength, regulatory news — all of these affect both coins. When one runs ahead because of a temporary liquidity event or news cycle, the underlying drivers don't change, so the gap usually closes.
This works less well when:
- The drivers actually diverge (one project has a major catalyst the other doesn't)
- One coin enters a sustained trend that the other doesn't share
- Liquidity collapses, fragmenting price discovery
A good ratio screener filters for pairs where mean reversion is working right now, not pairs where it might work in theory.
How to test for mean reversion
Two main statistical tests:
Hurst exponent (1951): measures whether a time series mean- reverts (H < 0.5), random-walks (H ≈ 0.5), or trends (H > 0.5).
Augmented Dickey-Fuller (1979): tests whether a series has a stable mean to revert to (low p-value = stationary).
Apply both to the log-ratio of two prices. If both pass, the pair is statistically mean-reverting over the tested window.
For the full math with Python code, see our Hurst, ADF and walk-forward backtest guide.
What this strategy is not
It's not a way to predict price. You're not betting BTC goes up or down — you're betting BTC and ETH return to their typical relationship.
It's not high-frequency. Typical hold time between swaps is 1–6 months on the pairs we screen.
It's not free money. We have a whole post on where the strategy fails. Strong directional trends, low-volume alts, delistings, and short-history tokenized assets all break it in different ways.
Where to start
If you want to dig deeper:
- The methodology page on PairScan walks through the full filter pipeline.
- The open-source utility at github.com/pairscan/ratio-mean-reversion lets you run the filters on your own price data.
- The BTC vs Apple tradable ratio post shows mean reversion applied to a cross-asset pair (crypto vs tokenized US equity).
Or just open pairscan.io — it runs the four- filter pipeline plus walk-forward backtest on 170+ pairs every 6 hours, including cross-asset combinations with tokenized US equities. Free tier shows the top 3 pairs daily.
The strategy isn't a secret. It's classical statistics applied to crypto and tokenized-equity pairs. Understanding it takes about the same effort as understanding compound interest. Profitably executing on it takes more — that's where filters and discipline matter.