​Some thoughts on Buffet Gambler, models, and crashing (tl;dr, too).

[pokerjoe]

 

 

I don’t know BG’s methodology, but let’s assume he had an organized approach or a model. Let’s also assume it had been successful for some decent number of bets before this year (I’ve never audited his record, but a few respected posters vouch for him, and RAS did as well).

 

His approach, whatever it was, collapsed this MLB season, and not by a little bit. Why might it have done that?

 

It could be that he’d only been lucky in the previous years, and this summer was just a case of the cards breaking even. No depth of record changes the fact that, for anyone with a history of winning, they may only have been lucky. And don’t give me stats analysis of the rear-facing low chance that his past record was just luck. You wouldn’t defend a lottery winner as skillful because the chance of winning that on luck alone is so low. Survivor’s bias is an especially strong phenomenon among forum touts.

 

It could also be that he really had an edge in past years, but lost it for emotional reasons. If he’d been handicapping (that’s what I mean by “organized approachâ€), as opposed to modeling or rating, then he may have been knocked off his game.

As a poker player, I’m well aware that edge isn’t static, even in the same game against the same players. You can’t say “That guy had an edge and therefore always will.â€

People have “A†games, “B†games, “C†games. Maybe BG is a handicapper who got rattled, tilted, whatever, and played catch-up, and got beat worse, and fell apart.

 

It could also be that he modeled and the market caught up to him.

I have an idea that any model will eventually collapse because of its past success.

I’ve been through it myself a number of times. A full MLB season in which I crushed, followed by a year in which I lost back all the previous summer’s winnings by June. Two years of >60% football betting, <40% the third year. I had several good years of CBB (with my own Pomeroy-ish model), that went flat after Pomeroy went public. And then there was this past NBA season where my model, with over 2k bets the previous three seasons, at 58%, crashed so bad I stopped betting by Christmas (I’ll try to write another post about this later; it would be too long to include here, but let’s just say I could have been the Buffet Gambler of NBA betting).

 

I mentioned this idea to a poker friend with a successful model and he vehemently counter-argued. Of course. When my models were winning, the last thing I’d want to hear was that I was doomed. I understand. And he might be right, too, because he’s modeling in a smaller, more obscure market, but also because he might just have an awesome model.

 

But here’s the idea (“idea,†guys. I’m not saying this is true): any stats or combination of stats that correlate to winning or line movements will eventually be sussed out and neutralized by the market, leaving you with a break even proposition if you’re lucky, a wrong-side betting monster if you aren’t.

 

If you have a model and bet enough, with enough success, to affect the line, those line moves will be broken down, and every possible correlating stat relationship will be found (I personally analyze the hell out of interesting line moves). But even if you don’t create CLV, but “merely†win ATS, because ATS success is also number crunched, those correlations will be found, too (though not so quickly).

Even the most chocolate combinations of public stats are not so brilliant that no one else will find them.

In theory, if your model loses edge, it will simply make fewer and fewer bets. If it’s been seeing a game as 3 when it was lined 5, it will now see the game as 3 but only find it at 4, as the market catches up. No loss, just no bets, no profits.

 

But lines vary from projected numbers for a reason. For example, if you used a really simple math model, like Sagarin (which I believe is nothing more than scores run through the Excel add-in, Solver), you’ll create a power rating. Obviously it’s primitive (I wouldn’t say it’s a “model,†because I think of a model as a rating based on a summation of the team’s individual players’ abilities), but it is math-based.

 

But Sagarin-type ratings don’t work because where it varies from the line, it varies with purpose. It should, long run, hit 50% only. This issue—that lines vary from any model or rating with a purpose—is the crux of the problem. As the market (other modelers and handicappers) catch-up to you (start to include the variables you’ve been especially smartly handling, in somewhat the way you do), if they get ahead of you, the lines will still vary from your projected numbers, but differently than they did before. Only your weakest picks will remain. Only your errors. You’ll end up betting only where the market has discounted your model.

 

I’ve expressed that poorly. Let me illustrate.

 

Suppose you went to a universe with a sport where people hadn’t figured out that there was such a thing as home field advantage. You’d have a huge market advantage, even with only adding HFA to Sagarin-style ratings.

 

Then suppose that the alternate universe market did figure out (with no public pronouncement, obv) that HFA existed. Thereafter, you’d no longer have an advantage, clearly. In this example you would know it, because you’d be going from almost only betting home teams to a fair mix of home and away. But generally modeling edge is distributed over many variables complexly arranged, so you probably wouldn’t know that the market had caught you.

 

And might you have a trap dis-advantage? If obviously the line should be X, but it’s Y, so you bet, you would actually be fading whichever part of the market now led the field. The market wouldn’t vary from you essentially randomly (as it would vary from a simple, pointless, long-discounted rating like Sagarin, or from the typical square guesser), it would vary such that you were being seduced into fading the sharpest of sharps.

 

Probably everyone who models handles some things well, some things average, some things poorly (by “things†I suppose I mean variables).

 

Back to this alternate universe. Suppose the market consists of two sharps (me and you) and a handful of squares. Hypothesize further that there are only three variables: injuries, HFA and general ability. Maybe you handle injuries better than me, maybe I handle HFA better than you, and maybe the team’s general abilities are obvious even to the squares, so that we gain no edge there.

If you learn how to handle HFA as well as I do, I would then end up only betting where I’d wrongly accounted for injuries. I wouldn’t just break even, I’d now be (unknowingly) fading your expertise (where a square would be only fading the juice).

 

Meaning, in the constant race for modeling advantage, it may always be only a matter of time before the market doesn’t just catch up to you, but kicks your ass as it passes you by.

 

The reason I quit betting NBA so quickly this year wasn’t only because I was getting tortured, it’s also because my bet percentage (% of listed games on which I wagered) dropped from 21% to 14%, totals and sides both. That was the big tip-off. (I should add that I dropped my model-based betting, which was the last model I had, and probably the last I’ll ever have, lol; I continued with my handicapped bets, which were fine, though they couldn’t get me even for the year).

 

I interpreted that reduction in bet frequency as meaning there’d been a big change in the NBA market this year, and I was on the wrong end of it. A very important (and probably the most profitable) third of my bets were gone, leaving me sucking wind. I described it vaguely to friends as there being a market change for a reason I couldn’t put my finger on (although I think I’ve since figured out the source of this major change in last year’s NBA market, and it’s some interesting shit genuinely worthy of major news coverage).

 

I wonder if BG never felt that the market had changed on him, or did feel that way but released picks anyway because he had no financial reason not to try to Yosh his way back to profits. I don’t know. I’m just shooting ideas because this is actually a huge issue for those of us who cap and model.

 

That leaves us here: models, to stay profitable, must be constantly changed. But if they are constantly changed (generously known as “tweakingâ€), then their past record of success must be thrown out, leaving you back to where you started: without basis for assuming edge.

[The Seer]

I had great years from 2006 to 2009 but I did not figure out why I was losing in 2010 until I found out I had cancer. Cancer out, got better.

[rito]

BG's posted MLB plays at peeps in 2012 crushed the closing line. His posted plays at Betting Talk in 2013 were ~0ev according to close (they won some 35 units in 150 plays though) he mentioned changes in the offshore market made him less concerned with that (prior to the season). And yes there were some major changes (pinny credit going away, limits going from 30 to 10k).

 

His line value this year has been worse(slightly -ev vs pinny close), over way more plays than his last few posted seasons combined. So what happened? I don't know. He mentioned his past posting always being just a handful of his plays while his tout service has given out everything. I also do know that his plays are all modeled, so something clearly went wrong there. There was a point early in the season when i was still betting them that some random factors that are all luck were all just running absurdly bad (like think 20 blown saves for vs 0 against), but once you get to -65 units in 600 plays (into 10c lines) you kind of have to assume you've lost your edge.

 

The thing about Modeling is that it's just as much art as it is science. Sports change over time, the gambling markets change over night (due to legal issues). I've made a number of models over the years that properly back tested (most people do not properly back test) show a profit, that going forward don't win. I don't know anyone that does this though that isn't constantly looking to improve their models.

 

I'm curious what you think happened in NBA, this clown on BT keeps saying all the syndicates got killed on nba totals last year -- but 2.5+ moves from pinny open to close ran pretty well.

[rito]

Go fuck yourself Raiders you low life scumbag.

[cyberbabble]

I'll be optimistic and go with pokerjoe option 2, psychological reasons. It could be an example of a first time starter being affected by the pressure. He not only has a forum reputation to maintain, but many of the people buying picks are friends. The usual tout doesn't care about his picks, he is only interested in signing up more people. I'll accept that BG is an honorable guy, trying to do well, and it must be difficult to fail his friends in this situation. He is trying too hard, second guessing his models, chasing by releasing too many picks, or some variation of these. He hits a small random losing streak and spirals down into serious losses instead of accepting a small loss and moving on.

 

This is like the standard situation of a poker player moving up in stakes, taking a small hit, tightening up too much, and making the losses worse. The poker player needs to drop back and do what works for a while. Then try to move up again.

 

I've never followed his picks. i wish him well. Maybe he can drop back to what worked for him in the past and do better in the future.

[eberetta1]

Maybe next year the system will be back to working just fine. Can't jump ship with something that has paid the bills in the past.

[Inkwell77]

This forum is bringing it with the quality posts recently.

 

Great stuff.

 

I don't know how many picks Buffet Gambler posted the past few seasons, but it was more than likely just good variance leading to his positive units.

The only guys I ever seen profit continually are guys who specialize in one sport, guys who rarely make plays, or sharps who are part of a "team" (ala RightAngleSports). Yisman may be the only example on the internet of an individual guy who continually wins in various sports. .

 

Look at all the touts currently on the internet and pretty much every single guy losses when there number of plays gets into the 1000's

 

John Kelly is a shining example of a guy who has won over multiple years, but he is still only up like 10 or 15 units altogether, and I think his total number of plays is still less than 1,000

[rito]

Winning players are going to generally be betting plays until the value is gone, so there's really no way for them to post plays.

[Propswatcher]

Agreed quality post. Actually had some discussion with some friends about the BG issue. We came to the conclusion of the same factors PokerJoe mentioned.

 

We also came to the conclusion that perceived edge has to the craziest thing about this racket. A handicapper with at perceived edge, but truly -EV will go broke faster than anyone. (a la Vegas loves amateur card counters)

 

Also enjoyed PokerJoes mention on survivor-ship bias. (the analogy I think of: take 100 dogs and have them each pick between two bowls of food with team names. After 100 picks or so, there will be some dogs hitting 65% and some hitting 35%.) That doesn't mean my lab is the next Billy Walters. I've known some touts that started like this. (Tracked the picks and took the best records to start selling picks.) Taleb's book Fooled by randomness goes into this, very applicable to sports betting. Winning does not necessarily = future winning/talent.

 

So in my opinion we are down to this. We can all learn a good lesson from BG and constantly rethink these questions as we make each wager or use our models. What constitutes an edge? how do you know when you have an edge in your wager? how do you know when you have lost your edge?

 

Few edge examples.

 

#1 CLV

#2 information you know that the market doesn't factor

#3 Better cost/price than the market. (FH's Packers at Golden Nugget +7 -120 fits into this.)

 

 

 

 

 

 

 

[Seanie Mac]

This is a great read

[pokerjoe]

I wish I knew more about BG's methodology. I did once read through some of his write-ups and--sorry to seem like I'm kicking him after the fact--they read like gibberish to me.

 

To distinguish handicapping from modeling, I define handicapping as rating teams and players by judgment. If you aren't putting a number on it, you aren't handicapping (in the same way that a golf or horse racing handicap entails a number or strokes or pounds), but if you're trusting your hopefully good judgement on how much to drop Ohio State's rating with Miller out, for example, then you're handicapping (making hand made ratings) but if you're using a program to weight such variables, you're modeling. And if you're using a program to rate entire teams, without breaking them down by players, you're rating.

 

Anyway, I think BG must have been handicapping, and because that entails judgment, and judgment can be thrown off (one of the advantages of modeling is the absence of emotionalism in rating decisions, if not betting decisions), then I'd go with my option #2 above: he just went on tilt.

 

And though I have some sympathy for him, I'd say he somewhat lost respect, not for running bad, which we're all just waiting to do, but for the Yoshing. That was unethical.

[trytrytry]

what are his season numbers. I have seen them earlier and been told they are really poor. what are the numbers like right now?

[taxman]

-66.27 units for baseball season

[taxman]

Without his unconditional love for the Tigers and Red Sox, and to a lesser degree Greinke, and Dickey, he would be much closer to break-even....

[GameBred]

The thing about Modeling is that it's just as much art as it is science. Sports change over time' date=' the gambling markets change over night (due to legal issues). I've made a number of models over the years that properly back tested (most people do not properly back test) show a profit, that going forward don't win. I don't know anyone that does this though that isn't constantly looking to improve their models.[/quote']

 

Can you please share your definition of "properly back tested" and the common mistakes people use when back testing as the subject interests me greatly.

 

Thank you.

 

[bookbraker]

I wonder how many customers BG had/has?

 

 

[WVU]

Without his unconditional love for the Tigers and Red Sox' date=' and to a lesser degree Greinke, and Dickey, he would be much closer to break-even....[/quote']

 

I wonder what he thought of the leaked Verlander pic?

[taxman]

Variance. Ass variance....

[akillies]

back when BG handicapped he crushed, for years. He posted detailed analysis on the rx everyday during mlb, until he got too much "feedback"

 

[joeybagadonuts]

This is a great read

Wow blueprint is back. How you been , pal

[bookbraker]

I wonder how many customers BG had/has?

 

I would guess about 50.