A threshold is a tunable cutoff that turns a trading signal into action. When signal > threshold, the strategy sends a BUY order; when signal < -threshold, it sends a SELL order. Signals between those bounds produce no trade, so the setting controls how strong a predicted price move must be before risking an order. section 4.1.1
Figure 4.2 A proprietary trading strategy. The strategy computes a signal from public market data. If signal > threshold, the strategy sends a BUY order to the exchange. Similarly, if signal < -threshold, the strategy sends a SELL order; threshold is a parameter.
The order regions make the threshold’s role visible: raising it narrows the signals that trigger trades, while a negative threshold creates overlapping BUY and SELL conditions. section 4.1.1
Treat threshold as a continuous, ordered parameter whose value should be chosen against a business metric such as markout profit. Very low settings can cause frequent losing trades; very high settings can leave the strategy trading too rarely. Negative settings are invalid because the buy and sell regions overlap. section 4.1.1 section 4.2.1
A response-surface workflow tunes a threshold with fewer aggregate settings than dense testing:
- Choose at least three plausible values using domain knowledge; three points allow a parabola to estimate a maximum between measured bounds. section 4.1.1 section 4.2.1
- Randomize threshold selection at trading decision times and collect repeated profit measurements for each setting, then compare aggregate means with standard errors. section 4.1.2
- Fit a surrogate model of markout_profit(threshold), optimize its estimated profit over candidate values, and treat the resulting setting as an estimate rather than a direct production measurement. section 4.1.4
- Run a final production measurement at the estimated optimum; accept it when estimated profit falls within the aggregate measurement’s approximate 95% confidence interval. section 4.1.5
Interpolated trading threshold
Measure markout profit at thresholds 0.5, 1.0, and 1.5 with 15,000 individual observations per value. Fit a parabola through the aggregate results, search the fitted curve, and obtain an estimated optimum such as threshold_opt = 1.09. Before deploying it, measure profit again at 1.09 and compare the estimate with the observed mean plus or minus two standard errors. section 4.1.2 section 4.1.4 section 4.1.5
Threshold can remain one coordinate of a larger configuration. With order size as a second continuous parameter, measure a design spanning both parameters; interpolation and optimization then operate on markout profit as a function of threshold and order size. Response surface methodology is presented as best suited to three or fewer parameters. section 4.2 section 4.2.1
