OT margin of victory is not as meaningful as regulation.

Interestingly enough, Ken Pomeroy did a study on this very thing, and found out that overtime metrics actually shows slightly better correlation with predictive results to the final 5 minutes of regulation in a tied ballgame. While the study is locked behind a paywall, I'll summarize what it showed:

__Garbage time is real:__

@10min left: A 23+pt margin renders the final 10min meaningless from a predictive standpoint (i.e. score effects will basically be completely random those final 10minutes and are not predictive of future play)

**@5min left:** A 15+pt margin renders the final 5min meaningless from a predictive standpoint (In fact a 22+pt margin starts showing negative predictive correlation, i.e. the winning team steps off the gas and allows meaningless buckets to the worse team that will not be repeated in a rematch)

**@2min left:** A 13+pt margin renders the final 2min meaningless from a predictive standpoint (In fact a 20+pt margin starts showing negative predictive correlation, i.e. the winning team steps off the gas and allows meaningless buckets to the worse team that will not be repeated in a rematch)

__Score effects in close games shows decent correlation to predictive success:__

@10min left: When margin is between 0-6pts, score effects in the final 10min show a positive correlation of 0.21-0.15 to future success

**@5min left:** When margin is between 0-6pts, score effects in the final 5min show a positive correlation of 0.15-0.13 to future success

**@2min left:** When margin is between 0-7pts, score effects in the final 2min show a positive correlation of about 0.13 to future success

**In OT:** Score effects in the final 5min of OT show a positive correlation of 0.16 to future success.

TLDR: Basically, the numbers (from thousands of games) suggests that overtime is a fairly important indicator and predictor of future performance against that same opponent. In fact it's an even better indicator (albeit slightly) than even the final 5min or 2min of regulation in a tight game. It also shows that 20+pt leads with 10min to go, 15+pt leads with 5min to go, and 13+pt leads with 2min to go can be considered blowouts and predictive algorithms should consider stopping data collection at that point as a predictive indicator.