Mean reversion strategy in Python

What is mean reversion?

Mean reversion is the idea that prices oscillate around a fair value and that unusually large deviations tend to correct. Instead of chasing trends, a mean-reversion trader fades them: buying when an asset is stretched below its average and selling when it is stretched above.

It works best in range-bound, liquid markets and breaks down in strong trends, where 'cheap' keeps getting cheaper. The art is measuring 'stretched' robustly and cutting losers before a reversion that never comes.

The logic

We measure deviation with a rolling z-score: how many standard deviations the current price sits from its 50-bar mean. A z-score of -2 means price is two standard deviations below average, a statistically notable dip.

The rule is symmetric: go long when the z-score drops below -2, go short when it rises above +2, and stay flat in between. A stop-loss caps the damage when a 'cheap' market keeps falling.

mean reversion in Python

Here is the full strategy in the Manifold-BT expression DSL. It imports manifoldbt, builds the signal, configures realistic execution, and runs the backtest against Rust.

import manifoldbt as mbt
from manifoldbt.indicators import close
from manifoldbt.helpers import time_range, Slippage, Interval

# Z-score of price vs its 50-bar moving average
z = close.zscore(50)

# Long 2σ below the mean, short 2σ above, flat in between
strategy = (
    mbt.Strategy.create("mean_reversion")
    .signal("z", z)
    .size(mbt.when(z < -2.0, 1.0, mbt.when(z > 2.0, -1.0, 0.0)))
    .stop_loss(pct=4.0)
)

start, end = time_range("2021-01-01", "2026-01-01")
config = mbt.BacktestConfig(
    universe={"binance": ["ETHUSDT"]},
    time_range_start=start,
    time_range_end=end,
    bar_interval=Interval.hours(1),
    initial_capital=10_000,
    fees=mbt.FeeConfig.binance_perps(),
    slippage=Slippage.fixed_bps(2),
    warmup_bars=50,
)
store = mbt.ingest(provider="binance", symbol="ETHUSDT", symbol_id=1,
                   interval="1h", start="2021-01-01T00:00:00Z",
                   end="2026-01-01T00:00:00Z")
result = mbt.run(strategy, config, store)
print(result.summary())
mbt.plot.tearsheet(result, show=True)

Backtest it with Manifold-BT

The config runs on hourly ETH bars with Binance perpetual fees and two basis points of slippage, so the result reflects real trading friction, not a frictionless ideal. Because mean reversion trades often, those costs matter a lot.

Tune the lookback window and z-score thresholds with a parameter sweep before trusting any single setting. A rule that only works at exactly z = -2 and a 50-bar window is overfit; one that works across a plateau of nearby settings is more likely to be real.

Frequently asked questions