Is this typical for conformal prediction? The prediction interval first narrows, then suddenly explodes toward the end.

I think the conformal predictions are based on the estimated (calculated) residuals for each fc timestamp. Please someone correct me here, if i am wrong. But this means that it is normal that the intervall might change over time based on the estimated residuals calculation during training. One could imagine that the interval gets bigger over time, but it could also fluctuate. But your plot could also mean that the interval was calculated on small samples and therefore shows such a weird pattern. maybe increase the respective params that the interval is estimated based on more samples. you might end up with some smoother output.

No it is not

@Valeriy do you have an explaination as to why this could be?