NHL Model Notes
I overhauled the NHL model entirely during the offseason between the 2017-18 and 2018-19 seasons. I suspect that 2014-15 was an auspicious year for the former model, because the model in 2018-19 barely resembles the original model, and I've never had a model perform as consistently as NHL 2018-19. False confidence from 2014-15 induced two seasons of mediocre results and a terrible 2017-18; which was the impetus for a full rebuild. NHL 2017-18 was a necessary evil; it was modeling rock-bottom. I had to finally over-correct and used everybody else's fried numbers to the point where I knew they were highly speculative and unreliable.
One thing I can tell you after wrestling with NHL models built using principles and advanced metrics from various "experts" and advanced analytics site operators; the NHL is the most widely-misunderstood major North American sport, from a quantitative perspective. If you want to succeed modeling the NHL, you have to take the base data, without the little spins and adjustments that everybody adds to the numbers, and advance it yourself. There are no tried-and-true advanced NHL metrics or projection systems available to the public, unlike what Fangraphs provides in the MLB realm.
After finally operating a dialed-in NHL model for a full season, I can now appreciate just how low-variance the NHL is. Going forward, I expect the NHL to be our most consistent, and possibly most profitable sport.