The study I linked earlier found the following predictive validities for life outcomes using the same sample group described in your link. The correlation is more than 0, but generally quite weak. I’d expect the army test used in your link had greater correlations due to sample bias. An army vocational test is going to be more predictive of life outcomes among a biased sample that is mostly persons applying to the army than among the general population in a similar way that MCAT score would be more predictive of life outcomes among a biased sample that is mostly persons applying to med school than among the general population.
Earnings at Age 35 – 0.04
Incarceration at Age 35 – <0.02
Welfare at Age 35 – <0.02
Depression --<0.02
I’m not familiar with the algorithms used for predicting the fall in oil price or market crash, but I expect that they used a variety of sources of information to improve their prediction instead of just looking at a single stat. It’s a similar idea with colleges, applying for a job, or just about any other group that has other useful information about the sample group available to them. College use a variety of measures to predict success including high school grades, high school curriculum, LORs, essays, achievements outside of the classroom, etc instead of just looking at a single test score. So the question isn’t whether a particular test scores has a more than 0 correlation with life outcome, it’s more what that test score adds to the prediction beyond the other criteria used in evaluation of applicants.