How to Evaluate a Strategy With Drawdown, Turnover and Tail Risk

Track compounded net return, maximum drawdown and recovery time, gross turnover, and expected shortfall over a specified horizon. Add average and peak exposures, estimated capacity and net performance under higher costs. Report these alongside Sharpe and a benchmark over the same dates. Each metric describes a different way a strategy can become hard to hold or trade.

Three questions to ask of the P&L path

With net returns $r_t$ , calculate wealth $W_t=\prod_{u\le t}(1+r_u)$ and its running peak $H_t=\max(1,W_1,\ldots,W_t)$ . Maximum drawdown as a positive magnitude is

\operatorname{MDD}=\max_t \left(1-\frac{W_t}{H_t}\right).

Record the peak and trough dates and the first subsequent date at or above that peak. If the peak has yet to be regained, report the drawdown as ongoing.

Define gross turnover at a rebalance as the sum of absolute traded notionals divided by equity just before trading. When estimated from weights, compare the new target with actual weights after price drift. Some reports halve two-sided turnover, so state the convention before comparing values. With losses $L_t=-r_t$ and a sample whose worst 5% consists of exactly $k$ observations, estimate 95% expected shortfall as their average.

An equity path of 100, 120, 96, 108 and 126 has a maximum drawdown of $1-96/120=20\%$ . The earlier peak of 120 is recovered only when equity reaches 126. A path can finish with a strong total gain and still require enough capital and conviction to survive its interim loss.

At one rebalance, an account with $100,000 in equity sells $25,000 of one asset and buys $25,000 of another. Gross turnover is 50%. At 10 basis points of cost per dollar traded, the cost is $50, or 5 basis points of account equity. If the five worst losses in 100 days are 2%, 3%, 4%, 5% and 6%, empirical daily 95% expected shortfall is 4%. Five observations offer little precision, so add stress scenarios and show how the estimate changes when a severe day is removed.

Drawdown and expected shortfall look at different objects. Drawdown is path dependent. Losses that arrive back to back can deepen the drawdown even if their individual magnitudes are ordinary. One-day expected shortfall instead summarises the severity of the worst daily losses, irrespective of order. Report the horizon and the number of observations in the tail. With one year of daily data, a 99% tail contains only about two or three days of probability, so the estimate deserves a sensitivity check.

Three questions to ask of the P&L path

Compare a loss threshold with losses beyond it