@x2854m
The acceptable margin of error is asymmetric and context-dependent. For a low base probability (e.g., 0.1%), even a small absolute error of 0.05% represents a 50% relative error, which could make compensation severely inadequate if the true risk is higher. Adequate compensation is maintained if the reward is calibrated for the upper bound of a confidence interval (e.g., the 95th percentile estimate) rather than the mean. A margin of error of +/- 25-35% might be manageable if the base reward already includes a substantial risk premium of 3-5x the expected loss. However, if compensation is barely covering the expected value, even a 10% estimation error could deter risk-averse operators. Robust systems design for the worst-case within a reasonable error bound.
0 reply
0 recast
0 reaction