[pzhon]

[ QUOTE ]

[ QUOTE ]

I think he comes up with really bad conclusions using this model. His results are still much more conservative than from other models. I'm not sure why. I think he made a serious miscalculation.

[/ QUOTE ]

The general conclusion is something I've seen alluded to in other posts; that even a superior player can't afford to pass up high EV+ situations.

His example uses a player that has an ROI of 100% and he concludes that D (representing the probability that you would win an all-in confrontation) would only need to be .58 in the smaller tourneys and .56 in the larger tourneys (given typical payout schedules).

Are you saying D is higher using other models?

[/ QUOTE ]
No, it is much lower in every other model I've seen.

Even with this one, I don't see how he goes from the prize distribution to a value of D other than the probability of doubling up. In his model, if the tournament pays less than half of the players, you have to double up to get any prize. Either there is a critical part of his model which he didn't include, or else he miscalculated.

His model of chip accumulation is missing a few things, including time. If I double up on the first hand, this is very different to me than doubling up at the end of the 4th level. It means I can still apply my skill advantage to accumulate more chips (on average) for 4 extra levels. This means doubling up on the first hand is better than doubling up at the typical time, and I should accept a lower probability of doubling up in order to do so quickly.

For example, suppose I start with 1000 chips, and expect to accumulate 100 chips on average by doubling up 55% of the time at the end of level 4, and busting out 45% of the time. If I have 2000 chips at the start of the tournament, how many should I expect to have at the end of level 4? I think the answer should be between 2100 and 2200. If 2000 chips at time 0 is worth 2200 at time 4, then D~50%. If it is worth 2100 at time 1, then D~52.5%. Both of these are lower than my 55% probability of doubling up.

That points out how implausible it is to require a value like 56%, higher than the probability of doubling up normally in Guerrera's examples. It also shows that this model is not so great, and isn't worth studying that deeply.