Systematic Portfolios: Diversification That Works
Ten strategies are only diversified if they do not all lose in the same week. Historical correlation will not tell you that.
Diversified on paper
A portfolio can hold strategies across equities, rates, currencies and commodities, on short and long horizons, and still be one bet. If all of them depend on the same background condition, such as steadily trending markets or plentiful liquidity, they will fail together when it changes.
Backtests often hide this, because across a long calm period the strategies show low correlation. When the condition reverses, they move as one, at exactly the moment diversification was supposed to help. This article is informational and is not investment advice.
Why correlation misleads
A correlation coefficient summarizes how two return series moved over a chosen window and says nothing about why. Two strategies can be near zero for years and then converge sharply, which leaves a number that was accurate for the period and useless as a forecast.
Exhibit 1: Three ways to test whether strategies are independent, and what each can and cannot show.
| Test | What it shows | Limitation |
|---|---|---|
| Historical correlation over the full sample | Average co-movement | Dominated by calm periods; may exclude the regime that matters |
| Correlation in stressed periods only | Behavior when it counts | Few observations; depends on which crises are in the data |
| Review of signal logic | Whether strategies rely on the same condition | Requires access to how each signal is built |
The third test is the most informative and the least used. It asks a plain question of each strategy: what has to be true in the market for this to work? If several strategies give the same answer, they are one position with several names.
How sizing undoes diversification
Even a well-chosen set of strategies can be concentrated by how capital is assigned, and a common practice is to give more to whatever has done well lately.
Recent winners tend to be those best suited to current conditions, and they are also likely to be the ones most alike. Sizing them up loads the portfolio onto the shared dependency it was meant to spread.
List strategies by share of capital and again by share of risk. If the top of both lists is the same few names, and they rely on the same condition, the portfolio is less diversified than its count suggests.
Sizing by risk contribution addresses this directly, as shown in risk budgeting across a multi-strategy portfolio.
Questions for a provider or an internal team
- Which market condition does each strategy depend on?
- How did the strategies behave relative to each other in the worst periods?
- Were those periods part of the data used to design them?
- How is capital assigned, and when was that last reviewed?
- Has anyone outside the design team examined the signal logic?
Testing independence is one layer, and building systems on different logic from the start is another, which we cover in architectural diversification across market regimes.
Key takeaways
- For each strategy, write one sentence on the market condition it needs in order to make money.
- Group strategies whose sentences are the same, whatever asset class they trade.
- Recalculate correlations using only the worst historical periods for the portfolio.
- Compare each strategy’s share of capital with its share of portfolio risk.
Examine how diversified a system really is
Synergy Nexus Quant can explain how its systems are tested for independence and how position sizing is tied to risk contribution.
Backtesting: The Easiest Number to Fake
Any strategy looks good if it is tuned against the same history often enough. What matters is when the rules were fixed.
Algorithmic Trading Compliance Amid Shifting Markets
The rules on algorithmic trading controls differ between Canada, the United States and Europe. What supervisors want to see demonstrated is much the same.
Risk Budgeting Across a Multi-Strategy Systematic Portfolio
Giving every strategy the same capital feels balanced. In risk terms it rarely is.