Why the Sharpe ratio misleads for high-frequency scalpers
Posted: Sat Oct 31, 2026 6:59 pm
The Sharpe ratio divides average return by volatility, annualized. It was designed for portfolios with returns measured over days or months. For scalping systems, with thousands of trades and short holding periods, annualization can produce enormous numbers that mean little.
Part of the problem is the assumption of independent, normally distributed returns. Scalp results have fat tails, clustering, and dependence on market conditions. The ratio understates the risk of rare large losses.
Also, since scalpers are flat most of the time, volatility of daily returns captures only the active periods. Different conventions for computing returns yield widely varying values.
Better measures for this context include the distribution of daily or weekly returns, maximum drawdown relative to average return, the percentage of profitable days, and the worst day as a share of the monthly profit.
Compare strategies using the same method, and look at the shape of the equity curve, not only a summary figure.
When a Sharpe above five appears in a scalping backtest, suspicion is warranted. Costs or data are likely wrong.
What measures complement or replace the Sharpe ratio when evaluating scalping systems?
Part of the problem is the assumption of independent, normally distributed returns. Scalp results have fat tails, clustering, and dependence on market conditions. The ratio understates the risk of rare large losses.
Also, since scalpers are flat most of the time, volatility of daily returns captures only the active periods. Different conventions for computing returns yield widely varying values.
Better measures for this context include the distribution of daily or weekly returns, maximum drawdown relative to average return, the percentage of profitable days, and the worst day as a share of the monthly profit.
Compare strategies using the same method, and look at the shape of the equity curve, not only a summary figure.
When a Sharpe above five appears in a scalping backtest, suspicion is warranted. Costs or data are likely wrong.
What measures complement or replace the Sharpe ratio when evaluating scalping systems?