Better strategies compound linearly. More strategies compound roughly with the square root. The math is decisive.

The mistake most quants spend years making

Most quantitative researchers begin their careers chasing a Better Strategy. The premise is intuitive: more skill, more research, more iteration, eventually produces a superior alpha source. The Sharpe ratio goes up. The drawdown comes down. The strategy, in your imagination, becomes The One.

The premise is also wrong in a specific and consequential way. The marginal return to improving a single strategy is, in almost every realistic setting, smaller than the marginal return to adding a second strategy with low correlation to the first. This is not a contrarian opinion. It is a direct consequence of how risk-adjusted returns combine across uncorrelated sources, and it has been understood at least since Markowitz wrote down portfolio theory in 1952. The persistence of the single-strategy mindset in 2026 is a sociological phenomenon, not a mathematical one.

This post lays out the case for multi-strategy systematic investing — not by reciting textbook diversification, but by walking through the specific decisions, trade-offs, and operational realities that the single-strategy mindset hides. The argument is in four parts: the mathematics of diversification, the statistical advantages of multi-strategy thinking, the risk-management benefits, and the operational costs.

The mathematics of diversification

Start with the simplest possible version of the argument. Two strategies, A and B. Both have annualized Sharpe of 1.0. Returns are uncorrelated. Each contributes equally to portfolio volatility (equal vol weighting).

The portfolio’s Sharpe is √2 ≈ 1.41 times the individual Sharpe. That is a 41% improvement, achieved without making either strategy better. The math: variance is the average of individual variances when uncorrelated; mean is the average of individual means; the ratio of mean to standard deviation scales by 1/√(1/N) = √N.

With N uncorrelated strategies of equal Sharpe S, the portfolio Sharpe is S·√N. The improvement is not bounded; it scales without limit as you add diversifying sources.

Reality is messier. Correlations are rarely zero. Define the average pairwise correlation of strategy returns as ρ. The portfolio Sharpe of N equal-Sharpe-S strategies at average correlation ρ is approximately:

S · √(N) / √(1 + (N − 1)·ρ)

For ρ = 0, you recover S·√N. For ρ = 1 (fully correlated), the portfolio Sharpe collapses back to S — adding more strategies doesn’t help. The interesting range is in between.

Worked numbers, S = 1.0:

— N = 2, ρ = 0.0: portfolio Sharpe = 1.41. — N = 2, ρ = 0.2: portfolio Sharpe = 1.29. — N = 2, ρ = 0.5: portfolio Sharpe = 1.15. — N = 5, ρ = 0.0: portfolio Sharpe = 2.24. — N = 5, ρ = 0.2: portfolio Sharpe = 1.69. — N = 5, ρ = 0.5: portfolio Sharpe = 1.29. — N = 10, ρ = 0.0: portfolio Sharpe = 3.16. — N = 10, ρ = 0.2: portfolio Sharpe = 1.91. — N = 10, ρ = 0.5: portfolio Sharpe = 1.33.

Two observations. First, even at realistic correlations (0.2 to 0.3, typical of systematic strategies that look “different”), the improvement is substantial. A 5-strategy portfolio at ρ = 0.2 has 1.69x the Sharpe of any single component. That is the equivalent of finding a strategy with 70% more skill — without finding it.

Second, the marginal benefit per added strategy diminishes. The jump from N = 2 to N = 5 at ρ = 0.2 adds 0.40 to portfolio Sharpe. The jump from N = 5 to N = 10 adds only 0.22. The first few uncorrelated strategies matter most. This has the implication that after 5-10 strategies, the value of adding more comes mainly from reducing the average pairwise correlation, not from sheer count.

Why “improve the strategy” feels like the answer

If the math is so clear, why does the single-strategy mindset persist?

Three reasons.

Reason one: the cognitive cost of inferiority. Adding a Sharpe-0.6 strategy to a Sharpe-1.0 strategy feels like a step down. The new strategy looks worse than the old. Allocating weight to it feels like a downgrade. In fact, at correlation 0.2 to the first strategy, a Sharpe-0.6 second strategy at vol-equal weight raises the portfolio Sharpe from 1.0 to ~1.15. It is mathematically additive. But it does not feel additive.

Reason two: the legibility of single-strategy performance. A single strategy has one equity curve, one Sharpe, one drawdown. It is summarizable in one number. A multi-strategy portfolio requires a dashboard — risk contributions, effective N, correlation stability, drift relative to design. The portfolio level is harder to talk about, harder to defend in a meeting, harder to celebrate.

Reason three: the operational compounding cost. A second strategy means more code, more monitoring, more failure modes, more sizing decisions, more rebalancing logic. The operational complexity grows non-linearly with N, even as the statistical benefit grows sub-linearly. There is a real threshold below which a single, well-understood strategy is operationally superior to several poorly understood ones.

This third reason is the legitimate one. The first two are bias. The framework I lay out in this series assumes you have the operational capacity to handle 5-10 strategies — which most serious quant operations do, even if they don’t realize it. The barrier is rarely capacity. The barrier is mindset.

Statistical advantages beyond the Sharpe boost

The Sharpe-boost math is the headline. Three more subtle statistical advantages compound the case for multi-strategy.

Reduced sample-size dependency. A single strategy with 24 months of data has a Sharpe estimate with standard error of about 0.7 on annual scale. Even if the true Sharpe is 1.0, the observed Sharpe could plausibly be anywhere from 0.3 to 1.7. Decisions based on this single number are decisions made under massive sampling uncertainty.

A 5-strategy portfolio of comparable strategies, each with 24 months of data, has a portfolio Sharpe estimate whose standard error is smaller — roughly by √5. The portfolio number is more robust, even though each component’s number is identically noisy.

Selection-bias dilution. Each strategy in a portfolio is, individually, a survivor of some selection process. Its observed Sharpe is inflated relative to its true Sharpe. The inflation is approximately √(2·ln N_trials) for the maximum of N_trials.

When you combine many selection-tainted strategies into a portfolio, the inflation does not disappear, but its impact at the portfolio level is reduced. The portfolio Sharpe is a weighted average of inflated individual Sharpes, and the weighted average inherits less of the inflation than the maximum did. This is not a substitute for proper deflation (DSR and similar), but it is a real second-order benefit.

Regime robustness. Different strategies tend to perform well in different regimes. Trend strategies do well in trending markets, mean-reversion in choppy ones, carry in low-vol periods, defensive strategies in stress. A portfolio of strategies across regimes has lower regime-conditional drawdowns than any single strategy. The portfolio survives more market environments because its components fail in different ones.

Risk-management benefits

Beyond Sharpe and statistical advantages, multi-strategy investing offers risk-management benefits that single-strategy operators simply cannot access.

Lower drawdown for the same return. The math of diversification doesn’t just improve Sharpe; it also reduces drawdown. Empirically, a 5-strategy portfolio at modest correlation has a max drawdown roughly 50-60% of any individual strategy’s max drawdown, at the same expected return. The improvement is even more dramatic for tail measures like Conditional Value at Risk.

Ability to take less leverage. A higher-Sharpe portfolio can achieve the same return target with less leverage. This is operationally important. Less leverage means more margin buffer, lower margin call risk, more capacity to survive transient stress, and easier scalability if you grow capital.

Diversification of operational risk. Single-strategy operators have, by definition, single-point-of-failure operational exposure. A bug in the signal code, a data feed issue, a broker problem — any of these can take down the entire book. A multi-strategy portfolio with independent infrastructure for each strategy spreads operational risk. Not perfectly, but meaningfully.

Negotiation leverage with capital providers. Allocators evaluate single-strategy operators differently from multi-strategy ones. A multi-strategy book with a coherent framework is treated as an investment grade asset; a single-strategy book is treated as a more concentrated bet. The implications for capital cost and fee structure are real.

The operational costs (and why they’re worth it)

Building a multi-strategy book is harder. The operational costs are real and should not be downplayed.

You need monitoring infrastructure for each strategy independently — daily P&L attribution, weekly drift checks, monthly stress tests. You need a rebalancing process — when to re-allocate, by how much, on what trigger. You need a strategy-lifecycle process — how new strategies enter, how underperformers exit, how the transition is managed. You need documentation discipline — what was decided when, by whom, with what rationale, to be reviewed quarterly.

In my experience these operational costs are roughly linear in the number of strategies, perhaps slightly super-linear. They are not negligible. But they are also surmountable, and the leverage they create is significant.

The single-strategy operator can spend 80% of their time on research and 20% on operations. The multi-strategy operator typically spends 50-60% on research and 40-50% on operations. The mix is uncomfortable for researchers who joined the field for the research. It is also the correct mix for someone managing real capital.

A practical framework

The remainder of this series, and the two blog posts that follow, lay out the practical framework. The next post compares the major allocation methods (equal weight, inverse vol, risk parity, mean-variance) under realistic constraints. The post after that walks through the end-to-end implementation of a multi-strategy portfolio from research through monitoring.

The framework is not exotic. The components have been studied for decades. What I argue is that the practical decision points — when to add a strategy, when to remove one, how to size, how to monitor, what to override — are where the value gets created or destroyed. The math is well known; the discipline is what separates portfolios that survive from portfolios that don’t.

Closing argument

A better single strategy improves your returns linearly. More strategies, properly combined, improve your returns by roughly √N. Over a decade, the compounding difference between linear and √N improvement is enormous.

The mindset shift is small and uncomfortable. You stop optimizing the strategy. You start optimizing the portfolio. You evaluate decisions at the portfolio level, even when they feel like downgrades at the strategy level. You build operational infrastructure that lets you run more strategies reliably. You accept that the marginal hour is better spent diversifying than perfecting.

In the practical experience of running multi-strategy books, this is the single largest source of compound improvement available to a quantitative operator. It is also the most consistently overlooked.

If you build systematic strategies one at a time and you’ve never seriously asked the multi-strategy question, you are leaving compound returns on the table. The math doesn’t care about your preferences. It just sums.

References

• Markowitz, H. (1952). Portfolio Selection. Journal of Finance 7(1), 77–91.
• DeMiguel, V., Garlappi, L., Uppal, R. (2009). Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy?. Review of Financial Studies 22(5), 1915–1953.
• Black, F., Litterman, R. (1992). Global Portfolio Optimization. Financial Analysts Journal 48(5), 28–43.
• López de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.
• Bailey, D.H., López de Prado, M. (2014). The Deflated Sharpe Ratio. Journal of Portfolio Management 40(5).
• Ledoit, O., Wolf, M. (2004). A Well-Conditioned Estimator for Large-Dimensional Covariance Matrices. Journal of Multivariate Analysis 88(2), 365–411.