Skip to main content

Parameter Sensitivity (Box Plots)

What it is​

Parameter Sensitivity analysis uses box plots to show how the Sharpe ratio (or other metric) is distributed for each value of each optimized parameter. While heatmaps show combinations of two parameters together, sensitivity box plots isolate each parameter independently — answering: "How much does changing this one parameter affect performance, regardless of what the other parameter is set to?"

This is critical for understanding which parameters your strategy is most sensitive to. A parameter with tight box plots across all values has minimal impact — you can set it to almost anything. A parameter with dramatically different box plots at different values is a key driver of performance and needs careful selection.

How to access it​

Navigate to the Sensitivity tab in the optimization analytics popup. Available on Plus plans and above.

The optimization analytics popup is accessed via the table icon in the View Panel after your optimization job completes. See The Strategy Panel & View System for full details.

What you see​

For each optimized parameter, a series of box plots is displayed:

  • X-axis — Each value tested for that parameter (e.g., EMA period: 5, 10, 15, ..., 65).
  • Y-axis — Sharpe ratio (or selected metric).
  • Each box plot shows the distribution of Sharpe across all combinations that used that specific parameter value. The box shows the interquartile range (25th to 75th percentile), the median line, and whiskers extending to the min/max values.

If you optimized two parameters with 13 values each, the EMA sensitivity chart has 13 box plots — one per EMA value — each summarizing the 13 combinations that used that EMA value (across all DEMA values).

How to interpret it​

High sensitivity (parameter matters a lot):

  • Box plots at different values are at significantly different heights. For example, EMA period 10 has a median Sharpe of −0.3 while EMA period 35 has a median Sharpe of 1.2.
  • This parameter drives performance. Selecting the right value matters, and you should pay attention to which values produce the best medians.

Low sensitivity (parameter doesn't matter much):

  • Box plots at all values are roughly the same height and width. Changing this parameter doesn't meaningfully change outcomes.
  • You have more freedom with this parameter. Choose based on other considerations (faster execution, fewer trades, etc.) rather than optimizing for performance.

What to look for in each box:

  • Tall boxes (wide interquartile range) — Performance varies a lot depending on what the other parameter is set to. This value works well with some combinations but poorly with others.
  • Short boxes (narrow interquartile range) — Performance is consistent regardless of the other parameter. This is a more reliable value.
  • Median line position — Higher median is generally preferred, but a high median with a very tall box (high variance) may be less reliable than a slightly lower median with a short box.

Example​

Sensitivity analysis for a DEMA × EMA crossover on BTC shows:

EMA parameter sensitivity:

  • EMA values 5–15 have low medians (below 0) and tall boxes — short EMA periods are unreliable.
  • EMA values 25–45 have high medians (0.8–1.2) and shorter boxes — this is the sweet spot.
  • EMA values 55–65 have moderate medians (0.3–0.5) — functional but not optimal.

DEMA parameter sensitivity:

  • Box plots are more uniform across all DEMA values — medians range from 0.4 to 0.8.
  • Conclusion: the strategy is more sensitive to EMA period than DEMA period. The EMA choice matters most; the DEMA choice is secondary.

The trader focuses on getting the EMA value right (25–45 range) and treats the DEMA value as a fine-tuning parameter.