Serial correlation is the tendency of a strategy's results to depend on the results before them. Positive serial correlation means winning trades cluster and losing trades cluster; negative means they alternate. A strategy with independent results has neither, and its streaks are exactly what its win rate implies.
Why it arises
Most strategies have it, and in the positive direction, because they depend on the market's state. A trend follower wins repeatedly while a trend runs and loses repeatedly while the market ranges. A dip buyer wins through a rising market and loses through a falling one. The regime, not the trade, decides the run, and trades taken in the same regime are not independent draws.
What it does to drawdowns
Monte Carlo reshuffling, described on its own page, assumes independent trades. When results are positively serially correlated, the real losing runs are longer than any reshuffle produces, and the real drawdowns are deeper than the DD95 estimate. The estimate is still useful, and it is still too small, by an amount that depends on the strength of the correlation.
Measuring it
Compute the correlation between each trade's result and the previous one, and the one before that, over the history. Values near zero mean independent; values of 0.1 to 0.3 are common in regime-dependent strategies and worth noting. A runs test, which compares the number of streaks to the number expected, gives the same information. Either is a few minutes in a spreadsheet, and a verified record that publishes them is rare.
A strategy has a 45 per cent win rate and 400 trades. If its results were independent, the longest expected losing run is about nine. Its history shows a run of fifteen, and its lag-one correlation is 0.22. The Monte Carlo DD95 at one per cent risk is 19 per cent; the historical maximum was 24, and a block reshuffle that keeps runs of trades together rather than individual trades gives a DD95 of 27. The strategy is fine. Its risk setting should be chosen on 27, not 19.
Streaky results are not a defect; they are the signature of a strategy that depends on market state, which is nearly all of them. The consequence is that drawdown estimates built on reshuffling individual trades understate the risk, and the risk setting should allow for it. Where a vendor publishes a block-bootstrap drawdown figure, one that reshuffles runs rather than trades, prefer it. Where they publish only a plain Monte Carlo, add a margin.