Correlation & Diversification in Trading

At a glance

Core principle: Count independent bets, not tickers
Crisis behaviour: Correlations rise exactly when diversification is needed
What to measure: Strategy return correlation, not instrument correlation
Practical limit: Most portfolios contain far fewer independent bets than positions

Key takeaways

Adding positions reduces risk only if those positions are driven by different factors. Five technology stocks is one bet held five times.
Correlations are unstable and rise during stress, so diversification provides the least protection at the moment it is most needed.
Measure correlation between strategy returns, not between the instruments traded: two strategies in different markets can be the same bet.
Genuine diversification comes from combining different return sources, such as trend following and liquidity provision, rather than more instruments.
A simple concentration audit, grouping exposures by factor, usually reveals that a portfolio of ten positions carries two or three real bets.

The diversification illusion

A portfolio holding ten different stocks appears diversified. If all ten are large technology companies, the portfolio has one exposure repeated ten times, with slightly reduced company-specific risk and unchanged sector and factor risk.

Portfolio volatility with N equal positions, each with
volatility s and average pairwise correlation p:

   Portfolio vol = s x sqrt( (1/N) + (1 - 1/N) x p )

10 positions, each 30% volatility:

   p = 0.0   ->  portfolio vol =  9.5%   (true diversification)
   p = 0.3   ->  portfolio vol = 17.9%
   p = 0.6   ->  portfolio vol = 23.9%
   p = 0.9   ->  portfolio vol = 28.7%   (almost no benefit)

The number of positions barely matters once correlation is high.
The effective number of independent bets is roughly 1 / p.

How correlation changes the risk of a portfolio.

The final line is the practical takeaway. At an average correlation of 0.6, ten positions provide the diversification of roughly one and a half independent bets. Adding an eleventh correlated position accomplishes almost nothing.

Correlation rises when you need it most

Historical correlations are computed over calm periods that dominate the sample. During crises, participants sell whatever is liquid regardless of fundamentals, and previously independent assets move together. This is a structural feature of forced deleveraging, not an anomaly.

Equity sector correlations converge toward one during broad selloffs, so sector diversification within equities provides much less protection than the historical figures suggest.
The equity-bond relationship is not constant. It has been negative in some decades and positive in others, particularly when inflation rather than growth drives markets.
Strategy correlations converge too. Carry, premium selling, and short volatility strategies all depend on calm conditions and fail simultaneously.
Liquidity is the common factor. Assets that share no fundamental driver can move together purely because the same participants are forced to sell both.
Plan with stressed correlations. Compute the correlation of your positions using only the worst 5 percent of market days, which is a far more useful input than the full-sample figure.

Correlating strategies, not instruments

The more useful analysis is at the strategy level. Two strategies can trade entirely different markets and still be one bet if they rely on the same underlying condition.

Apparently different	Actually the same bet	Shared dependence
Long equities and short volatility	One risk-on position	Calm, rising markets
Currency carry and credit spread trades	One carry position	Absence of stress
Equity momentum and growth stock selection	One factor exposure	Continuation of recent leadership
Trend following in stocks and in commodities	Partially independent	Trend persistence, but different drivers
Mean reversion in indices and gap fading	One reversion bet	Short-term overreaction
Long bonds and long gold	Partially related	Falling real yields

To measure this properly, record daily or weekly returns for each strategy separately and compute the correlation between those return series. If two strategies show a correlation above roughly 0.6, they should share a single risk budget rather than each receiving a full allocation.

Building genuine diversification

1
Diversify across return sources first
Trend following, mean reversion, carry, and liquidity provision fail under different conditions. This is the most valuable axis and the one most often ignored.
2
Then across asset classes
Equities, rates, currencies, and commodities respond to different drivers. Futures make this accessible in a single account.
3
Then across timeframes
A multi-month trend strategy and a multi-day reversion strategy have low correlation even in the same instrument.
4
Only then across instruments
Adding more tickers within the same strategy and asset class is the weakest form of diversification, though it does reduce idiosyncratic risk.
5
Set explicit exposure limits
Positions per sector, per factor, and per direction. These caps do the work that intuition fails at.
6
Audit regularly
Quarterly, group every position by its dominant driver and count the groups. That count, not the position count, is your diversification.

Frequently asked questions

How many positions do I need to be diversified?

It depends entirely on correlation. With genuinely uncorrelated positions, ten to fifteen captures most of the available benefit. With highly correlated positions, no number is sufficient, because the portfolio volatility approaches the individual position volatility regardless of count. Count independent drivers rather than positions.

Why did my diversified portfolio fall together?

Because correlations rise during stress. Forced selling affects everything liquid simultaneously, so assets that normally behave independently move together in exactly the conditions where you wanted the diversification. This is why stressed-period correlation, rather than full-sample correlation, is the relevant input for portfolio construction.

Are bonds still a good diversifier for equities?

The relationship depends on what is driving markets. When growth expectations dominate, bonds have often risen as equities fell. When inflation dominates, both have fallen together, as occurred in 2022. Treating the negative correlation as permanent is a common and costly assumption.

How do I measure correlation between my strategies?

Record each strategy’s daily or weekly return series separately, then compute the correlation matrix. Do this on returns rather than on positions, because that captures the actual joint behaviour. Anything above roughly 0.6 should be treated as one strategy for risk-budgeting purposes.

Is diversification always beneficial?

Diversifying across genuinely different return sources is the closest thing to a free improvement in trading. Diversifying by adding more instruments within the same strategy adds complexity and cost without much benefit. The distinction between those two activities is where most portfolio construction goes wrong.

Test this idea before you trade it

Describe the rules in plain language and AlgoTrader AI turns them into a structured strategy blueprint with a configurable historical backtest, cost assumptions, and exportable code.

Build a backtest

Keep reading

Referenced by

Educational use only. This guide explains how a strategy works. It is not investment advice, not a recommendation, and no result described here is a forecast. Test any approach on historical and out-of-sample data, size positions conservatively, and never risk money you cannot afford to lose.