The Limper - NFL - 2018 - Post Mortem

Hitting a pathetic 43.5% ATS, this was the worst season the model has managed since 2009, the 2[SUP]nd[/SUP] year I began posting NFL projections. Re-running the season, using the data collected for 2018 (minus the scores, of course), the model produces a respectable 61.7% ATS rate, so where it went wrong, week-to-week, is a mystery; one I won’t solve until the off-season when I’ll go under the hood and take a look. That said, a few things stand out, like some teams defensive secondary play, which varied wildly – one week to the next. Matchups and injuries don’t explain it, home/away also fails, so what’s the reason? Also, QB changes for teams like the Ravens and Jags threw the model under the bus, as the tendencies of their entire offenses changed. Offensive play-calling for some teams (like the Vikings and Titans) got worse as the season progressed, while other teams (the Giants and Colts) remarkably improved. Officiating, which went from a flag every play, for half the season, to letting the players play, also killed any hope of reliable past performance projections. Thing is, none of these developments sufficiently explains the model’s poor performance at projecting reliable MOVs. It could still project straight-up winners and losers at 62.6% SU, but accurate margins of victory – the most important criteria – remained out of reach; and I won’t know why until I put the Limper in dry dock and can see what’s what below the water-line.

FWIW - Dogs killed it this season, going 53.6% ATS, while managing only 33.4% SU. The Bears, at 12-4 ATS, were the hands-down best at beating the Vegas lines, while the Falcons (who – too – late - finally rid themselves of the burden of Steve Sarkisian), were at the bottom with a lame 5-11 ATS.

Anyway – that was then. The playoffs are a whole new season, and – fingers crossed – I’m hoping the model comes through in 2019.

GLTA

keep up the work

a consistent 43% hitter is just as useful as a 57% model