Interesting analysis: thanks!
Let's remember that in the one modelling game where people pour unlimited resources and intelligence and for which there is now nearly 200 years of extremely rich data--the stock market--it is still not a reliable predictor, certainly not in the short term. And in macro economics generally, the Fed's predictors of long term economic trends are constantly shifting and remain at best an approximate game.
Compare all that to predicting basketball, with relatively very little data, but infinite variables. For example, in what you cite above about records against ranked teams, you don't factor in won-lost on home, away, and neutral courts. And yet, what exactly is a neutral court? But the sampling available is suspect because we don't know how to weigh games early in the season vs. late in the season, or against teams that are improving or regressing, etc.
Modelling sports outcomes over a single season is like modelling the weather: we know with absolute certainty that somewhere in New England it's going to rain today...most likely....probably....possibly....but where and when it's going to rain exactly, well, everyone better bring an umbrella from CT to Maine.
