Portfolio Drawdown Beyond Correlation

Two strategies look attractive on their own. Their average correlation is low, and combining their backtests produces a smoother line. That is a useful starting point, but not a complete portfolio risk analysis. A portfolio can still suffer concentrated losses when positions overlap, liquidity disappears or several strategies react to the same event. Correlation describes one aspect of co-movement. Drawdown describes the path of the combined account.

For a trader, the important question is not simply whether two return series are different. It is whether their combination improves the result under a realistic capital, sizing and execution policy. To answer that, you need synchronized observations, explicit exposure and a chronological portfolio ledger. Adding the headline numbers from separate reports is not enough.

Build the portfolio on a common clock

Suppose strategy A trades early in the session and strategy B trades near the close. Their trade lists have different lengths and timestamps. Pairing the first trade of A with the first trade of B creates an arbitrary correlation that does not represent simultaneous market exposure. Align both strategies to a meaningful common time grid before measuring co-movement.

Daily marked-to-market returns can be useful for a daily capital view. Intraday account limits require finer checkpoints. Closed-trade equity can miss a large temporary loss before a trade recovers. If your portfolio must survive an intraday loss rule, evaluate equity at the relevant intraday frequency rather than assuming that end-of-day survival implies intraday survival.

Include legitimate zero returns when a strategy was flat on a valid observation date. Keep missing data distinct from zero exposure. If strategy B's feed was unavailable for an afternoon, entering zero profit falsely treats that period as a safe absence of risk. Data alignment is an economic modeling decision, not just a spreadsheet convenience.

A five-session example

The following fixed-dollar net profits are deliberately constructed. Both strategies use the same account currency. There are no deposits or withdrawals, and transaction costs are already included in each observation. The portfolio takes both strategies at the displayed sizes.

Session	Strategy A	Strategy B	Portfolio
1	$200	−$100	$100
2	−$100	$200	$100
3	−$100	−$100	−$200
4	$200	−$100	$100
5	−$400	−$300	−$700

Portfolio cumulative profit is $100, $200, $0, $100 and −$600. Its peak is $200 and its final trough is −$600, producing an $800 maximum closed-session drawdown. With $10,000 initial equity, the percentage drawdown from the $10,200 peak is 800/10,200, or approximately 7.84%. Dividing by initial capital gives an 8% capital loss measure, which is a different denominator.

Strategy A's maximum drawdown is $400. Strategy B's is $500. Adding those maxima gives $900, not the actual portfolio maximum of $800. Individual drawdown episodes do not necessarily start at the same time or share the same peak. Always compute drawdown from the combined equity path.

Correlation and covariance still matter

For two return series with weights wA and wB, portfolio variance is the sum of their weighted variances plus twice the product of the weights and covariance. Covariance can be written as correlation times the two standard deviations. This describes return dispersion under the selected sample and weighting assumptions. It does not specify the order of losses.

Portfolio variance =
  wA² × volatilityA² + wB² × volatilityB²
  + 2 × wA × wB × correlation × volatilityA × volatilityB

Assume each strategy has 1% daily volatility and each receives a 50% return weight. The table shows the implied portfolio volatility for three hypothetical correlations. These are algebraic scenarios, not estimates of any actual strategy pair.

Correlation	Portfolio daily volatility
−0.50	0.50%
0.00	0.71%
+0.90	0.97%

The diversification benefit can shrink substantially if co-movement rises during stress. A full-sample correlation averages calm and difficult periods together. Inspect relationships during negative portfolio days, high volatility, major announcements and shared position exposure. Those conditional samples will be smaller, so avoid presenting their estimates as precise constants.

Low correlation does not identify the risk mechanism

Two strategies may have low daily correlation while both carry the same overnight gap risk. Their ordinary gains and losses can offset, but a single market discontinuity can affect both. Conversely, positively correlated strategies may still complement one another if they trade at different horizons and are sized so that the combined exposure remains controlled.

Look at what each strategy actually holds. A breakout model and a mean-reversion model trading the same futures contract are not independent assets merely because their code differs. A strategy trading an index and another trading a highly related index can share a common equity-market exposure. Labels are not a substitute for measuring the positions.

Consider net and gross exposure separately. A long position and an offsetting short position may reduce net market exposure while still generating execution costs, margin requirements and operational complexity. Whether an account nets the positions, and how the broker or platform treats them, affects the executable portfolio.

Capital allocation is part of the strategy

Combining two dollar-profit series assumes that both positions could have been funded at their original sizes. If each standalone backtest used the entire account's margin capacity, adding them together can create an infeasible portfolio. Define shared capital, maximum contracts, margin assumptions and the rule for competing orders before calculating performance.

There are several legitimate allocation policies. Fixed contracts keep the arithmetic straightforward but let risk change with volatility. Fixed fractions of capital create compounding and require a common equity update rule. Volatility targeting changes exposure as estimated risk changes. None is universally superior, and they should not be mixed without making the combined policy explicit.

For example, if A and B each request two contracts but the portfolio limit is three, an allocation rule must decide what happens. Priority by timestamp, proportional scaling and rejecting the later order produce different trades. A report that assumes all four contracts were filled is not evaluating the stated three-contract portfolio.

Stress the joint system

Begin with scenarios that reflect plausible common stresses: higher transaction costs, reduced liquidity, a shared gap, delayed execution and simultaneous losing sessions. Apply the stress consistently to all affected positions. Increasing costs for only one strategy because its result is less attractive would turn the exercise into selective storytelling.

Historical synchronized block resampling can preserve some observed cross-strategy dependence. Sample the complete vector of strategy returns for each selected session or block. Independently resampling each strategy breaks their shared timing and can overstate diversification. This is especially dangerous when portfolio failure depends on several losses arriving together.

Historical resampling still cannot create a missing crisis mechanism. A separate common-shock scenario can be useful, but clearly label its assumptions. For example, impose an additional adverse price move on all open positions at a specified checkpoint, then recompute equity and limits. Do not describe an invented stress magnitude as an estimated probability unless you have a defensible model for that probability.

Measure contributions during the actual drawdown

Once the portfolio's worst peak-to-trough interval is identified, calculate each strategy's profit over that same interval. These contributions explain which systems drove the realized portfolio episode. They differ from each strategy's own maximum drawdown, which may occur elsewhere.

In the five-session example, the portfolio peaks after session two. From that peak through session five, A contributes −$300 and B contributes −$500, totaling the $800 decline. This is a coherent attribution because the time boundaries are shared. Calling A's standalone $400 maximum drawdown its contribution to this portfolio episode would be incorrect.

Also inspect recovery time and liquidity needs. Two portfolios with equal maximum drawdown can differ in how long capital remains below its peak. A short sharp loss and a year of slow erosion have different practical consequences even if the headline percentage is identical. Report the duration in an explicitly defined calendar or trading-session unit.

Do not optimize away the only bad episode

Portfolio weights can be overfit just like entry parameters. If you try hundreds of combinations and keep the one that perfectly offsets a single historical crisis, the apparent diversification may not repeat. Record the weight search and evaluate the selected allocation on separate data.

A practical alternative is to start with a small number of economically motivated allocations and test sensitivity around them. If moving from 50/50 to 55/45 destroys the result, investigate why. A robust decision should not depend on a precisely tuned weight unless a convincing mechanism supports that precision.

Keep code versions and input sets distinct. Strategy A using input set B is not the same constituent as strategy A using input set C. If one constituent changes, create a new portfolio specification and retain the previous result. Reusing old returns under a changed label can make a portfolio appear diversified when its actual sources are mismatched.

A trader's review sequence

1. Reconcile each constituent's net returns and timestamps.
2. Align them on the calendar required by the account's risk rules.
3. Define shared capital, sizing, margin and order conflicts.
4. Calculate the combined equity before calculating its drawdown.
5. Inspect ordinary and stressed co-movement.
6. Attribute losses over the portfolio's actual worst interval.
7. Test the chosen allocation on data not used to select it.

The goal is not the lowest possible historical correlation. It is a portfolio whose behavior you can explain and whose capital requirements you can support. A simple combination with transparent exposure and stable results can be more useful than a mathematically elaborate allocation that relies on fragile estimates.

When reviewing a portfolio report, ask one final question: could this account actually have held these positions, at these sizes, at these times? If the answer is unclear, improve the ledger and allocation rules before interpreting the smoothness of the equity curve. The most important diversification benefit is the one that survives execution.