Online poker cheating

One of the projects I’ve been engaged in lately is trying to catch players who are cheating in online poker.

(This is unrelated to the Pokernomics analysis I have also been doing — although at a glacial pace. For those of you who sent me hand histories we will be getting you ring game analyses within the next couple weeks.)

It turned out to be harder than I thought it would to catch online poker cheats. I have found lots of behavior that looks awfully suspicious in the data, but it doesn’t seem like the players do much better when they are behaving suspiciously than when they are not. So far, I can’t figure out whether what looks suspicious to me is not cheating, or whether cheating of the sort I’m looking at doesn’t provide a player much of an edge.

Recently, some highly regarded young players got caught playing under multiple identities in the same tournament. (See here and here, although if you aren’t a poker insider the posts are a little confusing). Thanks to Scott Cunningham for pointing me to these two links.

Because of these two projects on poker, I figured I better play a little myself to understand the game better. I was surprised how much fun it was. I was a big loser initially, even in low stakes games. Now I’m still a loser, but not as much, and at much higher stakes. I’ve even had the honor (??) of losing to the guys who just got caught cheating.

But the best news is that my wife Jeannette quickly picked up the game and now does not consider her day complete if she can’t slip in a few sit-n-go no limit hold ’em tournaments after the kids go to sleep. I married well.


cc

http://www.fulltiltpoker.com/multiMedia.php podcast (sorry)

Pokernomics

[...] One rogue economist tackles attempts poker. You’d think that poker would be on most Econ 101 syllabi. [...]

Derek

This is a problem I have thought about a lot as well. Part of the difficulty comes from the various ways in which poker cheating can happen. Some cheats might display themselves through their play (e.g., squeezing a victim by re-raising each other until the victim folds his hand fearing the nuts, then one cheater making a suspicious fold on a later street to avoid having to reveal the deception at showdown) vs. completely invisible cheating (e.g., sharing of hole card information, which in a game like Omaha, creates a real edge).

My feeling is that it will be hard to create a diagnostic test that identifies cheaters, because some cheaters will not behave suspiciously (e.g. type 2 cheaters), while many monkeys who play online will appear to be type 1 cheaters when in fact they are simply idiots. Thus the dual problem of being both overinclusive (false positive) and underinclusive (false negative).

I think a more interesting approach would simply be to create a dB measuring the statistical likelihood that a given combination of players is seated at the same table together. The more frequently the players are seated at the same table, the more likely it is that they are cheaters. This would be particularly true in lower limit games where the sheer number of tables would likely mean that even full-time online pros are rarely seated together with great frequency. However as you climbed the limits, and the player pool shrank, this method would work much less well--think, Chau Giang, Doyle Brunson, Jen Harmon, etc. in the big Bellagio game. They play together at the same table with near 100% correlation, but obviously, they arent cheating.

Anyhow, setting aside these situations, where the pool of players/available tables is large enough, it should be possible to create a distribution of "likelihood" that Player X has of being at the same table with Player Y. Look principally for the players whose correlation is several standard deviations off the norm, and you have a pretty good set of "possible cheaters."

The next step would be to figure out if you can slice this subgroup further.

Read more...

scott cunningham

The way in which these cheaters were caught highlights the challenges of catching them. For instance, Party Poker only caught these guys because others reported them. The first guy, JJProdigy, was in a big multi-table tournament (MTT) playing with two accounts. When his highly ranked "JJProdigy" account got knocked out, his second unknown account one, and he seemed to want the notoriety of winning a large field MTT, so he told some people it was actually him. One of them told someone who told someone who told Party, who only then took action against him. A similar thing happened against ZeeJustin. That they were kids might explain why they couldn't keep it quiet.

But the ability to catch cheaters doing this is very hard, as Party can only go by whether the same computer is on the same tournament at the same time. And it's not too difficult to get around that, if you don't mind investing in multiple IP addresses, computers, etc. Many professional players are quite cynical about online poker, saying that the cheating is so pervasive that it's no longer profitable.

Read more...

scott cunningham

Derek - that type of test would only work when a player controls the table he sits at, such as ring games. In a MTT, a player can't control where he sits. I'd be interesting in knowing, though, how much of a change in EV occurs when a person increases the number of accounts he's operating in that MTT. What's the likelihood that a person playing 3 accounts will end up at the same table at some point in a 200-person touranment? What's the change in EV with each additional account?

cc

There will never be a great professional economist poker player. Any economist who can master the game would have dropped out of their Ph.D. program way before candidacy...or would have been weeded out in jr. high as they made their fortune and avoided grad school poverty...

tim in tampa

Chris Ferguson is one of the only individuals who really comes from an academic background suited (no pun intended) to poker analysis who has been a major champion. I always believed that those of us who work with this sort of thing ought to be able to dominate poker -- but it's not really working out that way. In fact, I've found I play better on pure instinct than when I use statistics and game theory. That having been said, David Sklansky made quite a career both as a player and writer using almost a purely statistical approach.

I'm not finished working through this, especially considering my dissertation-in-progress is ON poker, but so far, the results have been far from my expectations.

Derek

Scott:

Cheating in MTTs is harder to detect, Im sure, than ring games, but I dont play them, so I could care less. That said, your question about the actual +EV edge from multiple buyins in MTTs is a good one. Let's begin by saying that there is some structural edge for sure, b/c it reduces the volatility of results for the player by creating multiple chances. As you know, you can't win a tourney w/out winning some coin flips (e.g., AK vs. QQ), so simply having the ability to have multiple entries helps b/c the arbitrariness of the coinflips gets spread over a large sample, thus you get a "second chance" (in a different account) when other players would be eliminated. Second chance = lower volatility of results.

As for the EV edge itself, since ab initio everybody's buyin is equal, the skilled players will be able to multiply their natural edge, while the average or weaker players will magnify their disadvantage. In other words, if you're dead money coming in, doubling your buyin will just increase the amount of dead money at stake. OTOH, if you have a +EV skill edge b/c you're Phil Hellmuth, you multiply your edge by doubling the amount of money at risk. This is what Mason Malmuth would basically call a non-self weighting strategy, which is key to all gambles with +EV outcomes.

Beyond the ability to increase your bet when you have an edge, there is the +EV component from cheating tactics themselves. In order to exploit this edge, you need to be seated at the same table as one of your alter identities. This fact alone is impossible to manipulate in the big MTTs. But lets say luck gets you seated at the same table as your alter egos. What is your edge here?

Im not a tourney player, but I would expect that there is some edge. Hypothetically, lets say Identity 1 and Identity 2 got seated at the same table during the random draw of the MTT. You'd have 20% of the chips in play at that table, vs. every other player who would have only 10% individually. Question: does having a chip advantage create a greater +EV edge that is disproportionate to the money required to buy those chips?

I hypothesize yes. Again, Im not a tourney player, but you would have a few moves clearly available only to you. For example: (1) chip dumping to become the big stack so you can attack small stacks and threaten them with elimination; (2) additional hole card information; (3) re-bluff stealing.

By and large, I dont think the strict "cheating" aspects of the multi-identity cheaters is that big. I really think they get an advantage from the structural edges I first mentioned (i.e., do-agains on coin flips, and non-selfweighting bets thru multiple entries).

The fact that ZeeJustin was able to make the final table in an MTT with multiple accounts was quite impressive, and really, is more of a testament to his skill as a player. For guys like him, who are strong tourney players, I think the edge is quite real. (Although if I were a good tourney player like him, I'd cheat not in MTTs but in the large buy-in SNGs. Having two entries on a $500+$50 would be quite advantageous.)

Read more...

scott cunningham

Derek - Great thoughts. To your last paragraph, it's interesting you say you would've cheated at more large buy-in SNGs over MTTs, as ZeeJustin had the reputation of being an excellent MTT and large buy-in SNG player. He played 10 SNGs simultaneously routinely (not uncommon, and part of how young players gain so much experience in a relatively short period of time), and would exploit such gains routlinely by having multiple accounts at the same large buy-in SNG. (Just rehearsing this again makes me shudder and never want to play online).

Derek

Scott:

I said I'd focus on SNGs instead of MTTs if I were a cheater, simply because I think you're likely to get more consistent, +EV results this way. (Maybe I say this because I can actually win SNGs, but I've only cashed once in an MTT.) Really, if I were cheating, I'd focus on the cash games, however. Multiple accounts playing pot limit Omaha would make you unbeatable.

cc

I was in London at the Gutshot Poker Club and got a glimpse of the real possibilities of maximizing revenue in cheating, the online crew. Everyone who has commented here, working together either virtually or in the same room, supervised by a poker zen master. Hitting ring games or MTT's. One can assume you could replicate the MIT/Bringing Down the House level of execution.

PSD

Fascinating topic and definitely one worthy of my first post!

Dr. Levitt, you seem on the right track when you say you found suspicious behaviour, but then for some reason you blow it off since they get poor results.

Tracking cheaters doesn't necessarily imply tracking winners. Perhaps cheating even leads to a false sense of confidence, which makes cheaters' performance (as a whole) worse than non-cheaters.

Regardless, the goal must be to uncover consistently suspicious behaviour.

As for what consitutes 'suspicious behaviour,' I guess that's where you need to set out a few hypotheses before data-mining.

One way to make the hypotheses would be to get in the mind of a cheater and think of situations where you might cheat. This is vague though, and probably requires guidance from known cheaters.

The only other thing I can think of is avg. time to play a hand. For these guys to play multiple games at once, i'm sure that over the course of 100+ hands their average playing time is at least a few deviations from the mean. I doubt they record how long it takes to play a hand though.

Read more...

aultimer

cc - you've heard of Jeff Yass?

scott cunningham

Derek - sorry, that was poorly worded. I knew you were saying that *if* you were a cheater, etc. I was just saying that your prediction is interesting in light of the fact that his expertise were SNGs.-sc

cc

ault, I've heard of Yass, but interesting stuff with his fervent passion for poker. Just to bring a bit more clarity from the online poker community. Some players open multiple accounts for rakeback and bonus deals. There are sign-up bonuses upwards of 100% on some sites to attract new customers. If you open an account, it is what we would call free money. Rakeback deals are ways that affiliates of poker sites are able to collect part of the rake back from the hosting site, normally 25-35% of the rake contributed by a player. Regarding the number of tables being played by some of the professional players: 6-8 tables is fairly common, with 12 being the most I've heard of. For SNG's, on suggestion made by a poker blogger is preventing players from choosing tables, having a queue system (similar to a card room).

keith_k

You may want to consult with Jonah Gelbach, a professor in the University of Maryland Economics Department. Although I don't think he has any publications, he just started teaching an undergraduate course titled "Economics of the Gambling Industry". http://glue.umd.edu/~gelbach/457/.

MLG

Scott - To be clear. ZeeJustin claims to have never entered two accounts in the same Sit n Go. Rather he used additional accounts to remain annonymous because people recognized his "ZeeJustin."

penxv

If you want to figure out how cheaters cheat... your best bet is to start cheating.

And then make the adjustments that any rational cheater would make.

Panzerkaetzchen

Interesting... my boyfriend is working on poker-tracking software that will enable players to access statistics about their opponents' play; in theory, it will give his customers an edge at the online poker tables.
If you would be interested in trying out a beta-version of his software, I'll let him know.

Gravity

I know some excellent tests for diagnosing cheating at poker. Here is one:

If player A (in early position) is dealt AK and player B (in late position) is dealt AA and player A does not call or raise the blinds, this is a strong indication that players A and B are sharing their cards.

It will be hard for people who do not know poker strategy to diagnose cheating unless the cheaters are really obvious.