Short answer
If trading costs are nonlinear, then the threshold should not be a static line applied to every market state and every position size. The signal has to clear the cost curve that actually exists at the moment the trade would happen.
Learn / Research process
Back to learn
Answer page / research process
Topic cluster / Signal calibration and thresholdsHow should you set a trading threshold when costs are nonlinear?
When costs rise with size, volatility, or thin liquidity, the threshold should clear a moving net-value hurdle rather than a fixed raw score cutoff.
What to remember
- The edge per trade shrinks as participation rises.
- Volatility spikes can make the same threshold too eager.
- Shared-capital portfolios may face different effective costs than isolated single-strategy tests.
Why one global cutoff fails
Spread, impact, and adverse selection often get worse in stressed tape or at larger size. A raw signal level that was profitable in calm conditions can become a losing trade when liquidity thins or the portfolio has to move more capital through the same opportunity.
- The edge per trade shrinks as participation rises.
- Volatility spikes can make the same threshold too eager.
- Shared-capital portfolios may face different effective costs than isolated single-strategy tests.
A better thresholding frame
Think in net value, not raw score. That usually means converting the score into an expected edge estimate, subtracting the state-dependent cost of trading, and only acting when the remaining margin is large enough to justify the operational risk.
How to test it honestly
Run the rule through quiet and stressed windows, vary the cost assumptions around the boundary, and inspect whether nearby threshold settings produce similar behavior. If the strategy only works when one exact cutoff meets one optimistic cost curve, the threshold is probably telling you more about the backtest than about the market.