Sure, there ain't free money in quant strategies, just premia for poorly understood or bona fide unknown risk factors. Here we seem to have a kurtosis premium, which is highly conditional to make things worse. As the name suggests, it tends to be negative with the variance component of vola declining... but the complex conditionality destroys prediction rather effectively, even if you conveniently happen to care only about the first two moments (with st dev being significantly lower overnight, one has to admit).