On this page
Formula and intent
SQN = √N × (mean trade result ÷ standard deviation of trade results), where N is the number of trades. The fraction is the per-trade Sharpe ratio; the square root of N converts it into a t-statistic-like quantity. Two systems with the same per-trade Sharpe differ in SQN if one produced four times as many trades: its SQN is twice as large.
The intent is to reward consistency across many trades rather than a high average on a few. Trade results are usually expressed in R so that position size and instrument do not distort the standard deviation.
Relationship to other measures
- SQN is Sharpe per trade times √N. A high Sharpe on ten trades and a modest Sharpe on a thousand can give the same SQN.
- Unlike expectancy, it is dimensionless once results are in R.
- Unlike the profit factor, it uses every trade's deviation from the mean, so a few outliers raise the standard deviation and lower the score.
- It is not corrected for the number of configurations tried; on the best of a sweep it is as optimistic as any other statistic.
In Senzoukria
The Replay performance panel reports SQN in its System quality section with the hint '√N × mean / standard deviation (Van Tharp). Accounts for the number of trades.' The panel also writes a short verdict from the trade count and the SQN, with wording for no edge, weak, average, solid and suspicious; the suspicious case is triggered when the figure is too good for the sample size and a handful of trades is carrying the result. When the SQN cannot be computed the verdict says to judge on the curve, not on a missing number.
In the backtest panel, SQN is one of the walk-forward objectives, described as 'quality, accounts for the number of trades', alongside expectancy per trade and Sharpe per trade.
Common mistakes
- Comparing SQN across samples with very different trade counts without noting that √N is doing the work.
- Computing it in dollars across instruments. Position size and tick value inflate the standard deviation.
- Reading the desktop verdict labels as certifications. They describe this sample; the solid case is worded as the most a backtest can ever prove.
- Optimising a sweep on SQN and then reporting the same SQN as evidence. Use the deflated Sharpe for that.
Related
In the same section
- Objective function
- Sample size
- Sortino ratio
- t-statistic
- Stacked imbalances
- SPX vs SPY options
- Static drawdown
- Spot-perp basis
This page in other languages
Frequently asked questions
- Why does SQN increase with the number of trades?
- Because the formula multiplies the per-trade Sharpe by the square root of the trade count. The design assumption is that the same average and dispersion observed over more trades is stronger evidence of a real edge. That assumption breaks when the trades are not independent, for example many trades inside one session that share the same market regime.
- What does the suspicious verdict in the Replay panel mean?
- It appears when the SQN is high relative to the sample size, a pattern that usually means a few large trades dominate the result. The panel suggests checking the trade distribution below the summary. It is a prompt to inspect, not a judgement that the strategy is wrong.