Small confidences can be more off
When predicting small probabilities of events occurring, there is more inherent risk in those small probabilities. They can be way off since they are likely based on little data (events that have occurred rarely) and small shifts in the expected distribution can have large shifts in Probabilities have orders of magnitude.
↑ 2 References
“Since we cannot replay history thousands of times and count the outcomes, a statement that some event will occur with a probability of .01 or .001 or .0001 or .00001 is essentially a readout of the assessor’s subjective confidence. This includes mathematical analyses in which scientists plot the distribution of events in the past (like wars or cyberattacks) and show they fall into a power-law distribution, one with “fat” or “thick” tails, in which extreme events are highly improbable but not astronomically improbable.7 The math is of little help in calibrating the risk, because the scattershot data along the tail of the distribution generally misbehave, deviating from a smooth curve and making estimation impossible. All we know is that very bad things can happen.” (Steven Pinker, Enlightenment Now)