It is non stationary

Machine learning at central banks

If the machine learning mimics the economy then it tries to come to a conclusion, find the tipping points.  That makes it non stationary.

Machine learning is just that, trying to  come to a conclusion.  The sandbox is like that,  finite boundary problem orbiting solutions that are error bound though not  quite complete.  At any given time we have a set of prices that covers variations in real yields, the typicals.  As long as quants hold, the market is parsing out the typical set, then precision i not supported and we reprice, set the pricing error change in the other direction.