Recency bias phenomenon

[Valuist]

This has become a buzzword, or buzz phrase in handicapping circles, mostly in the analytical crowd. They didn't invent the term; it is an actual psychological phenomenon. But does it really exist in handicapping?
Doesn't a so-called recency bias discount the factor of current form? Its based on two assumptions; first is that people overemphasize what they saw last. That may well be true. But the second is that the public sees EVERY game. They assume the public knew what Eastern Washington did three days ago, or what Texas AM did in their last two games. The public can't stay on top with over 300 teams. Can you imagine betting the Kentucky Derby based strictly on horses 2 year old form? What about college teams who add impact players at semester break? What about the added confidence of improved play? It sounds good in theory, but I question it.

 

[railbird]

the analytical crowd are the least analytical people i know, they use about 20 buzz words that make them feel sharp, but they are far from it.

 

[cheapseats]

What qualities or ability does the public have, enmass. Seems like a different way of chasing and defining steam.

 

[ComptrBob]

There is also anti-recency bias that uses stale data to compute handicapping ratings.

KenPom is one of the worst, e.g. in 2018, took him 20 games to flush Wisconsin, 15-18, from his Top 20 eventually to finish as 70th.

 

[Valuist]

Look at the job world, the dating world. What is often stressed? First impressions. People are inherently lazy and stubborn. IMO a first impression bias trumps recency bias.

 

[John Kelly]

Look at the job world, the dating world. What is often stressed? First impressions. People are inherently lazy and stubborn. IMO a first impression bias trumps recency bias.

First-impression bias TRUMPS recency bias.

Let's keep politics out of the discussion.

 

[John Kelly]

You make good points, Valuist.

Recency is critical, especially when a good team goes bad or a bad team suddenly improves.

You want to find CHANGES that influence performance.

The more subtle the change, the better for the shrewd handicapper looking to outsmart the competition.

Sometimes you can see the light come on for a team or horse.

Other times, you can spot when a bulb is about to go dead or a favored horse is going the wrong way.

 

[WildBill]

I think this is evolving constantly. Yes recency bias is very real, you see it almost every week in the NFL betting markets. However, there is some merit to it, NFL teams evolve very fast and just a small change like an injured player or a change in plays/schemes can make an impact so hard to create some sort of model that reflects true balanced out value.

Personally, I think the real opportunities are to embrace the mix of randomness and luck that impact sports and assume those balance out over time. I used to try to model two sports and come up with "true" pricing that could beat the books, but I gave up long time ago. Someone else out there surely can build a better model and it was too hard to predict if the modeled team was going to show up that day or not. So you can say my methods today are a mix of recognizing that recent performance outside of trend or model is quite normal, but be prepared to take advantage when it does start to regress to some sort of normal which can be either a drop off or an improvement. This is where you can attack both recency bias driven betting lines and even those fairly sharp bettors who are just using model numbers.

 

[Foresthill]

You make good points, Valuist.

Recency is critical, especially when a good team goes bad or a bad team suddenly improves.

You want to find CHANGES that influence performance.

The more subtle the change, the better for the shrewd handicapper looking to outsmart the competition.

Sometimes you can see the light come on for a team or horse.

Other times, you can spot when a bulb is about to go dead or a favored horse is going the wrong way.

If I may, let me simplify this: Life is dynamic. Thus, sports is dynamic. Also, randomness plays a role.

The two extremes in term of skill vs randomness: Chess -- all skill, and the lottery -- all random (or "luck", if you prefer). Everything else is in between.

 

[cheapseats]

Forest, you go so far as to say it's a zero sum game; life that is.

 

[BenoitPaire2]

I almost got sucked into it too.. luckily was bailed out.
On Monday, I foolishly bet sac state -11. Thought they would come out hot after some bad performances.... they then play the worst game of the year by far. Lose wire to wire against Idaho, one of the worst teams in the country

I watched the game and there offense was putrid as always.. but the defense was surprisingly bad. They looked to me like they quit on Katz.

I decide to take Weber State pk for one of my biggest bets of the year based on the performance Monday and thinking that they quit.. Tonight they played their asses off.. best offense I’ve seen them run in a while. I luckily got bailed out by Harding scoring 44— but it almost burnt me

 

[Valuist]

From the recent ESPN podcast, Preston Johnson, who is a modeler, said a number of professional handicappers who model said they have been increasing the weight of recency and he likely would tweak his model as well to account for recency.

 

[kane]

From the recent ESPN podcast, Preston Johnson, who is a modeler, said a number of professional handicappers who model said they have been increasing the weight of recency and he likely would tweak his model as well to account for recency.

I saw that, and good for him to realize and admit that he can still learn, that he doesn't feel the need to stay stuck in his ways, and that when presented with another way of doing things, he's willing to adapt and change. I've said it before, but imo he's the only one worth listening to on that show, and unless I'm mistaken, he's also the only professional bettor on the show as well

 

[ComptrBob]

From the recent ESPN podcast, Preston Johnson, who is a modeler, said a number of professional handicappers who model said they have been increasing the weight of recency and he likely would tweak his model as well to account for recency.

This was in a discussion of the Raptors 15 win streak.

Its pretty clear from his comments that Preston Johnson has a very simplistic model that probably doesn't work well. He also said something very telling, "I'm not someone that is able to program very well, at a level, so at least I could go back and test some win streaks and evaluate them … ". A successful modeler has to have considerable database skill and the ability to filter and program queries that result in betable edges. By his own admission, sounds like he is deficient in this area. He did point out that streaks also depend on your opponent' strength. Schwimer chimed in that his group has backtested 2,3,4 game win streaks (probably just generically) and the streaking team tends to under perform going forward. But again, not all streaks are equal.

I've been incorporating "diminished stale data" algorithms in my models for 30 years (since the early 90's) which essentially is the pretty much the same as "increased weight on recency". All data must be weighted and discounted differently. For example, some college hoops data from the previous year can get stale immediately, whereas some March Madness data taken from 20 years ago can be very useful.

What both cappers missed was what was pointed out here, just using the Sagarin ratings of Top 16 wins (actually Top 15 since they don't play themselves), the Raptors are now 15-14, a record of 40-15 which means their record against the "Bottom 14" is 25-1. The 15 game win streak came against 14 "Bottom 17" teams, plus the 9th rated 76ers (14th rated Pacers (2), 15th rated Nets, and 16th rated Spurs).