Quant Style Regimes In Football

[andy28]

I’ve been running a draw backing model that’s been performing well overall, but recently I hit a strange patch where it just dried up. I assumed it was normal variance and kept it running, but the pattern was too clean to ignore a clear losing spell followed by a clear return to form. That made me dig deeper to see if anything structural had changed.

I used to rely on rolling averages, but they’re too heavy in Excel, so I switched to an Exponential Moving Average (EMA) for league goals. That’s when everything clicked. During the losing run the league wide average goals per game had climbed to around 3.2, which is obviously not ideal when you’re backing draws. That elevated scoring environment lasted for about 50 matches before dropping back down. Once it fell to Below 2.6 the model immediately started hitting again, I’m 4/4 since the drop.

When I split the data by scoring environment, the behaviour made sense. In the high goal period, home wins surged and away wins dipped. When the average dropped, draws spiked again, home wins fell, and away wins rose. I mentioned this to GPT and it said something interesting, that what I’ve built is basically a quant‑style regime model similar to what’s used in financial markets. That sent me down a rabbit hole, and it turns out traders use regime detection all the time to understand when a strategy is in or out of season.

It made me think about the comments you often see on here people saying favourites are winning more than they should, or that home teams are dominating, or that draws have disappeared. Maybe these aren’t just random swings. Maybe they’re temporary structural regimes in how a league behaves.

And maybe the cause isn’t in the usual data I track. It could be things like referees being told to enforce something differently, VAR behaviour shifting, fixture congestion, weather patterns, tactical trends, or even psychological cycles within teams. All of these could affect the whole league without showing up directly in spreadsheets.

So the question I’m throwing out there is whether football actually these kinds of regimes, the same way financial markets do, or whether match results are too independent for that idea to hold. Because from what I’ve seen my model didn’t suddenly stop working the environment changed, and when the environment reverted the model came back to life.