Sensitivity analysis
How do you test whether a conclusion survives when one input behind it is changed?
In plain words
Sensitivity analysis re-runs a conclusion while moving one input at a time and holding the others fixed, to see how far the output shifts. If the output barely moves, the conclusion is robust; if a small change in the input flips it, the conclusion rests on that input.
Why it matters
Most risk judgements depend on assumptions chosen by whoever built the model: a reference price, a volatility estimate, a threshold. Sensitivity analysis shows which of those choices the verdict actually depends on, and therefore where the judgement is fragile rather than well supported.
In a real week: ETH, week of 19 April 2021
For ETH in the week of 19 April 2021, a rule that calls a week strong when its return exceeds the historical median reads a 3.79% close, the 58th percentile of 166 weeks, as strong. Change one input: measure the same week from its low of 2055.00 on Tuesday 20 April at 07:00 UTC to its high of 2644.47 on Thursday 22 April at 16:00 UTC, and the week shows a 28.68% range, far larger than the closing figure.
A common mistake
Mistaking sensitivity for probability. It shows how much a conclusion depends on an input, not how likely that input is to change, and moving several inputs at once hides which one drove the result.