← Quantery Blog

How to read drawdown in a backtest | Quantery Blog

October 23, 2026 · 8 min readbacktestingdrawdownrisk

In Quantery, maximum drawdown is the largest fall from a prior peak to a later low among portfolio values recorded at rebalance dates. It tells you how far the simulated strategy fell at those observations before recovering or reaching the end of the test. Read it as a description of the sampled path your rules produced under the backtest's assumptions. It isn't a forecast of the next loss.

The headline depth is only the start. You also need to find when the decline happened, how long the portfolio stayed below its old peak, what the benchmark did over the same stretch, and whether the damage came from the screen or the holding policy. A backtest can finish ahead while spending an unbearable amount of time underwater. The ending return won't tell you that.

What does maximum drawdown measure?

At each rebalance date in Quantery's portfolio series, keep the highest observed value reached so far. Compare the current observed value with that prior peak:

drawdown = 1 - (current portfolio value / prior peak)

The result is zero at a new high and positive below it, expressed as the magnitude of the loss. Maximum drawdown is the largest observation in the series. Its start is the observed peak. Its bottom is the observed trough. If the portfolio later exceeds that peak at a rebalance date, that observation closes the episode. A deeper fall and recovery between weekly or monthly rebalances won't appear in the reported value or curve, so any timing or duration read from that curve has the same sampling limit.

This makes drawdown path-dependent. Two hypothetical portfolios can begin and end at the same values while earning those results in very different orders. One may climb with shallow setbacks. The other may suffer a deep loss, recover late, and still print the same ending return. An ending return collapses the path into a destination. Drawdown puts the path back.

That path matters because the backtest assumes the rules remained in force. It doesn't model the researcher's willingness to keep following them after months of losses, nor does it model a decision to cut exposure halfway through. A strategy you would abandon at its low isn't represented by the recovered line shown afterward.

Quantery reports maximum drawdown beside its hypothetical portfolio result. Start there, then use the portfolio curve and benchmark to locate the episode. How to read the full backtest report covers the other pieces that belong around it: coverage, event results, the benchmark, and the portfolio policy.

Drawdown depth and duration answer different questions

Depth measures how far the portfolio fell from its previous peak. Duration measures how long it remained below that peak. A deep decline that reverses fast creates a different research problem from a shallower decline that drags across filing cycles and rebalance dates.

The long decline can be harder to follow. Each rebalance provides another chance to doubt the rule, change a threshold, or replace the thesis. That behavior isn't captured by the simulated portfolio. The line keeps obeying the saved rules without getting tired.

Duration also helps diagnose what happened. If the portfolio remains below its peak while new qualifiers enter, the problem may extend beyond one bad holding. If the decline begins with one crowded entry cohort and stops after those positions leave, timing or concentration deserves attention. If the strategy and benchmark remain underwater together, broad market exposure may explain much of the pain.

A finished test can leave the final drawdown open. In that case there is no recovery date inside the sample. Don't treat the sample boundary as a recovery. The test stopped while the strategy was still underwater. Extending the end date may eventually show a recovery, but that becomes a different run with more information.

Compare the strategy with its benchmark during the same decline

Raw drawdown answers, "How much did this simulated portfolio lose from its own peak?" It doesn't answer whether the screen failed relative to its opportunity set.

Open the benchmark over the same dates. If both lines fell together and the strategy fell less, the drawdown may be market exposure with some resilience. If the benchmark held up while the strategy sank, the screen or portfolio policy added damage. If both fell and the strategy recovered later, the recovery gap is part of the result even when their eventual returns look similar.

Benchmark choice still matters. A broad large-company index gives poor context for a thesis that lives in small, thinly traded companies. The mismatch can make a market-wide decline look like thesis-specific failure, or hide a style shock that hit the selected universe harder. Use the benchmark selection workflow before interpreting the gap.

Keep raw and relative drawdown separate. Raw drawdown approximates the loss the hypothetical account had to endure. Relative performance asks whether the rules did better than the chosen alternative during that loss. A strategy can pass one question and fail the other. You need both answers.

Find what caused the worst stretch

Maximum drawdown is an output, not a diagnosis. Work from the peak toward the trough and ask which moving part changed.

Start with entry cohorts. A cluster of companies entering on the same rebalance date can make a nominally diversified portfolio one shared bet. The names may sit in one industry, respond to one macro shock, or carry the same accounting weakness. Grouping the decline by entry date and sector shows whether several rows are really one event.

Then check concentration. An equal-weight portfolio begins with equal allocations, but holdings don't remain equal as prices move between rebalances. A severe loser can still dominate part of the fall, and several correlated losers can dominate it together. Testing dependence on a single outlier explains how to reconstruct contribution and rerun without the leading contributor. Apply the mirror version to the largest loss.

Now inspect the exit policy. A strict gate near a noisy threshold can force exits after a bad move and admit the company again after a small reversal. A long maximum hold can keep a broken qualifier after the feature that admitted it has faded. Exit on gate failure can do the opposite and lock in every brief threshold crossing. The drawdown may belong to the holding policy as much as the entry rule.

Finally, check data coverage around the episode. A shrinking effective universe can change what the screen was able to select. Missing price history can remove possible replacements. Point-in-time fundamentals can be sparse for an early part of the test. Coverage problems don't automatically explain a loss, but they can turn the drawdown into evidence from a narrower universe than the title of the thesis suggests.

Stress-test the drawdown without tuning it away

Once you've formed a diagnosis, change one assumption per rerun. Preserve the original report. Its unpleasant path is evidence, and editing until it disappears destroys that evidence.

Useful reruns include:

• Change the rebalance cadence while keeping the thesis and exit rule fixed. This tests whether the loss depends on the observation clock.
• Change the maximum holding period while keeping entries fixed. This tests whether stale positions extended the damage.
• Toggle exit on gate failure. This tests whether the decline came from holding through deterioration or churning around a boundary.
• Remove the dominant sector or entry cohort in a controlled reconstruction. This tests concentration, with the narrower universe stated up front.
• Split the calendar at a date chosen without reference to the trough. This tests whether the result repeats outside the episode without using the loss itself to design the sample.

Every rerun is hypothetical and excludes costs. A setting that reduces drawdown by trading more often may look safer in the report while creating more spread, slippage, and tax drag outside it. The backtest doesn't give that activity a free pass just because the curve gets smoother.

Don't optimize for the smallest drawdown on the page. That invites the same search problem as optimizing for the highest return. You can keep trying exit rules until one sidesteps the historical trough by accident. Predeclare the changes that have an economic reason, record failed variations, and use a holdout period before keeping the revised rule. The backtest research log is built for this discipline.

What can drawdown establish?

A drawdown can establish that a saved set of rules, applied to the available point-in-time data and a specified portfolio policy, produced a particular losing path in the tested sample. It can show whether the loss coincided with the benchmark, one sector, one cohort, or an exit decision. It can show whether modest changes to the policy leave the diagnosis intact.

It can't establish the worst loss those rules could face. The historical sample contains only the shocks that occurred while usable data existed. A future decline can be deeper, last longer, arrive faster, or combine exposures the test never encountered. The reported portfolio also leaves out trading frictions and taxes.

The practical use of drawdown is to improve the question you ask next. If one cohort caused the fall, test timing concentration. If the benchmark fell with it, test the claim in relative terms. If a boundary created churn, test the boundary. If every sensible rerun preserves the same painful stretch, keep the pain in the thesis record.

A backtest curve is easy to admire after it recovers. Read it from the peak toward the trough instead. That's where you learn whether the rules are sturdy enough to keep testing, and which rule you should change before the next run.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles