Econometrics Research Problem

[dollerAday]

Hey guys I have to do an econometric research report on a specific topic of my choosing. I want to do it involving poker, something along the lines of proving it is predominantly a game of skill over the long run. But I have been thinking about it and I’m not exactly sure what variables I could have to prove that it is skill based other than money earned. On the other hand I was thinking I could have variables like aggression, vpip, pfr and prove that there is an optimal strategy involved in poker. Another problem I see as a potential problem is finding the data for this, I have over a million hands logged in on PT but I’m not sure how I’d be able to export this data to get a regression model. If I’m able to export this data I was thinking that I would take everyone out over a certain amount of hands, like a couple thousand and use them in my sample.

[BlueBear]

You need tons on mined hands to do this research. Check the software forum.

[ QUOTE ]
On the other hand I was thinking I could have variables like aggression, vpip, pfr and prove that there is an optimal strategy involved in poker.

[/ QUOTE ]

This research you are proposing is solely statistically based and will prove nothing about optimal play from the game-theory perspective. The best you can do it to show multiple correlation between the different parameters.

[ QUOTE ]
I have over a million hands logged in on PT but I’m not sure how I’d be able to export this data to get a regression model

[/ QUOTE ]

This research report becomes a programming project in order to extract the data efficiently from a large database. You can use SQL queries to extract data faster I believe.

[ QUOTE ]
If I’m able to export this data I was thinking that I would take everyone out over a certain amount of hands, like a couple thousand and use them in my sample.

[/ QUOTE ]

Many stats like win rates and showdown stats take many ten thousands (or even hundred thousand) of hands to converge to a realistic estimate.

[dollerAday]

Do you think there is any other way at approaching the data other then the way I have suggested?

[BlueBear]

The way you're approaching it is basically fine (as long as it is undergraduate report, not a phd thesis). Data-mine like crazy, purchase hand histories, build large samples, load it in Pokertracker/Excel, analyze correlations - eg. PFR/VPIP against win rate.

If you approach this the right way, the results will be very interesting.

[dollerAday]

Okay I realize that it takes tens if not hundreds of thousands of hands to find a true win rate but is finding one necessary? The other parameters such as vpip/pfr can be found in much much less hands and if I get enough samples of these wouldn’t they average out to around the true win rate? I just don’t think it will be that easy for me to find large sample sizes of players with extremely high vpip and such because they simply will bust before they can play an adequate amount of hands.

[hitch1978]

That statistic in itself can be of significant use, can it not? If exposed against the correct backdrop.

Just 2 small points, although I am far from an expert.

1) I would think that you should have a copy of TOP by your side while you are putting it together.Sklansky bucks is why results converge to what we deserve them to be.

2) Please post a copy here when you are done.

[JMAnon]

[ QUOTE ]
Okay I realize that it takes tens if not hundreds of thousands of hands to find a true win rate but is finding one necessary? The other parameters such as vpip/pfr can be found in much much less hands and if I get enough samples of these wouldn’t they average out to around the true win rate? I just don’t think it will be that easy for me to find large sample sizes of players with extremely high vpip and such because they simply will bust before they can play an adequate amount of hands.

[/ QUOTE ]

It is difficult to answer your question without knowing what your assignment is. Generally, econometric research involves looking for statistically significant correlations in data. If your sample size is large enough to generate a statistically significant correllation, it doesn't usually matter how big your 95% or 99% confidence intervals are. The main point is to show the relationship, not its precise magnitude (unless you are analyzing the data with a particular practical purpose in mind).

If you run a regression with $ won / hand as the independent variable and, for example, VPIP as the dependent variable, Stata (or SPSS) will tell you whether there is a statistically significant correlation between the two within your existing sample. You can also include your other variables (aggression, went-to-showdown, etc.) in a multivariate regression and see which are significant, if you have time to extract the data. You probably won't need more hands than you have to show a statistically significant correlation between VPIP and $won / hand.

With respect to getting rid of outliers, you might throw out the top and bottom 5% of players in terms of $won / hand, but it probably won't affect your results much unless you have an unusual data set.

It is true that your confidence intervals will shrink as you increase your sample size, but you probably don't need particularly small confidence intervals for your assignment.

edit to add: an ordinary least squares regression should be fine as there isn't really a reason to think that VPIP and $ won / hand are endogenous.