Correlation Between Strategies
Running several methods at once diversifies only to the extent they fail at different times. Most methods within one style fail together.
MadStockAlerts Research · Updated August 28, 2026
What to take away
- Strategies within one style are usually highly correlated with each other.
- The relevant question is whether their drawdowns coincide.
- Correlation of returns understates it if the tails coincide.
- Genuine diversification across strategies requires different mechanisms.
- More strategies also means more parameters, and more scope for overfitting.
MAD Academy Training Video · 0:46
Two Strategies, One Bet
Strategies that look different can depend on the same market condition, which means they fail in the same week.
This lesson is part of a Stock Alerts + Tools plan.
Where the correlation comes from
Two trend-following methods with different parameters are the same idea implemented twice. They enter at similar times, hold similar positions and lose money in the same conditions, which are ranges.
| Combination | Typical correlation of results |
|---|---|
| Two trend methods with different lookbacks | Very high |
| Trend and mean reversion | Low, and sometimes negative |
| Two mean-reversion methods | High |
| Equity long-short and equity long only | High, unless genuinely hedged |
| Methods across different asset classes | Lower, and it rises in a crisis |
The second row is the only genuinely diversifying combination in the list, and it is the one that requires holding two methods that will each look wrong while the other is working.
Drawdowns rather than returns
A correlation coefficient computed on monthly returns can understate the relationship that matters. What matters is whether the bad periods coincide, and tail dependence is not captured by a linear correlation.
Scroll the chart sideways to see all of it.
- Method A
- Method B
What genuine diversification requires
- Different mechanisms: one profiting from continuation and one from reversion is a real difference.
- Different horizons, since a method holding for months fails in different conditions from one holding for hours.
- Different inputs, since two price-based methods share the input that drives both.
- Different asset classes, with the caveat that they converge in a crisis.
The cost of adding strategies
Each additional method brings its own parameters, its own fitting risk and its own monitoring burden. Two methods with a genuine difference in mechanism diversify more than five variations on one, and the five carry five times the scope for having been fitted.
The sample size problem also multiplies. Evaluating one method requires a long record; evaluating five requires five long records, and the capital and attention available do not multiply with them.
Allocating between methods
Once several methods are running, capital has to be divided between them, and the division is subject to the same estimation problems as any other allocation.
| Approach | What it assumes |
|---|---|
| Equal weight | Nothing. No estimates to be wrong about |
| Weight by past performance | That the recent ranking persists, which recency bias makes appealing |
| Weight by inverse volatility | That volatility is more estimable than return, which it is |
| Optimised weights | That expected returns can be estimated, which they cannot |
The second row is the one that occurs by default rather than by decision. Capital drifts toward whatever has worked recently, which is the same drift the rebalancing article describes and has the same consequence.
The first and third rows are the conclusions the efficient frontier article reaches for asset allocation, and they hold here for the same reason: an approach that requires fewer estimates is frequently better than a better approach that requires more.