Strategies · Library · Stocks
§3.10Mean reversion
Mean-reversion – weighted regression
A variant of §3.9 Mean-reversion – single cluster
A wider form of the group mean-reversion trade. Regress stock returns on risk factors (industry groups and other characteristics), weighting calmer stocks more, and bet against the leftover. This makes positions neutral to every factor, not only one group.
Open in the builderAs a long-only proxy; see how ARCANE uses it.
- Asset class
- Stocks
- Held for
- Days
- Data
- ARCANE has part of the data
- Symphony
- Builds as a long-only proxy
The rule
- Signal
- The part of each stock's return left over after a weighted regression on industry and other risk-factor exposures.
- Entry and exit
- Rebalance often (daily to weekly); bets shrink as leftovers fade.
- Sizing
- Dollar position proportional to the negative of the weighted leftover; weights per stock often the inverse of its variance; book nets to zero.
How it fails
- Factor model choice drives results
- Short leg needed
- High turnover
- Estimation noise with many stocks
Data it needs
- daily prices
- industry classification
- risk-factor exposures
Prices exist; industry tables and factor exposures for stocks are not in ARCANE's data.
How ARCANE uses it
A Call that a stock lags its sector once market and sector moves are removed could use this framing, with the factor list supplied by hand.
Builds as a long-only proxy. Hold the laggard of two sector ETFs over a window; the regression, many-stock universe and short leg are lost.
Filed under the mean reversion card
Buy last week’s biggest losers and sell last week’s biggest winners, betting that short, sharp moves partly undo themselves.
A paper plan from this card holds 1 to 10 days, stops at 1× the expected move and uses at most 50% of your risk budget. The card
Source: Kakushadze and Serur (2018), 151 Trading Strategies, §3.10, p. 50 of the SSRN version. Described in our words. Mean reversion is ARCANE’s filing, not the book’s.