Wednesday, April 23, 2008

Simulations for Thursday, April 24th

Teams Playing Pitching Matchup Sim Favorite Simulator Odds Vegas Odds Log5 Winner
DBacks vs Dodgers E.Gonzalez vs C.Billingsley Dodgers 56.58% 56.52% NA DBacks
Giants vs Padres T.Lincecum vs C.Young Padres 56.00% 57.45% NA Giants
Cubs vs Rockies J.Marquis vs A.Cook Rockies 55.20% 55.56% NA Rockies
Phillies vs Brewers J.Moyer vs J.Suppan Brewers 52.98% 54.55% NA Phillies
Vegas Baseball Odds
Simulator Setup File Download
Method Daily Points Total Points
Xeifrank's Simulator 189.64 3607.03
LV Hilton Odds 187.04 3665.32
LV Hilton leads by 1.62%
Note: Log5 method won't be used until approximately May 15th.

7 comments:

Anonymous said...

Hey, I love the sims, but I'm not sure why you're doing points the way you are. If I understand correctly, you simply give points to each simulator equal to the percent chance they had predicted for the winning side. But then a simulator who just picked a 100% chance for the favorite would always have higher expected points than a more accurate simulator. Maybe I'm missing something?

Best,
Matt

Xeifrank said...

Matt, if one of the methods put the odds of a team winning at 100% and they didn't win then they'd get 0 points for that game. If you put 100% on all the favorites, then your score would be the summation of all of the actual win percentages of the favorites. In yesterday's games, all the favorites (or you could argue all but one) lost. That method would've got zero points.

Anyways, all I am trying to measure is which is more accurate; Vegas Odds, My Simulator or Log5 and I'd be happy to post any other method that was based on some kind of science/reasoning. One thing that hampers measuring how accurate my simulator is vs Vegas Odds, is that I base my data input off ZIPS Projections. So if they were WAY off, then I would do poorly compared to Vegas. ZIPS is not kind to the Diamondbacks, and they got off to a hot start, and my simulator suffered because of that. I have since updated the DBacks with one of the other popular projection methods. One thing I could try, but I just don't have the time for it, is to break my simulator up by projection methods. Compare Vegas odds and Log5 to my simulator using projection method ZIPS, Chone, Marcel, Pecota, Miner etc...
Since they are all pretty similar for the most part, I just went with the one that I am most familiar with.

Thanks for the thoughtful comment. If you have any ideas to a better method for making comparisons, please don't hesitate to comment on it. I am always looking for improvements to my methodologies.

vr, Xeifrank

Anonymous said...

Right, when the favorite all lose you'll get zero, but look at the big picture.

Say Team A has a 55% chance of winning, and your simulator has it exactly right, giving Team A that 55% chance. Meanwhile, I give Team A a 100% chance of winning. My expectation is that I will get 55 points for this game (100*.55 + 0*.45). Your expectation is that you will get 55*.55 + 45*.45 = 50.5 points.

I've been thinking about other ways to measure this, and unfortunately I haven't come up with anything great. One thing you could do is take the average of the Vegas projection and the sim projection and then have the two simulators "bet" on the outcome. For example, your sim says Team A is 70% to win and Vegas says 80%. You make a bet with a money line of 75%, where you take the underdog and Vegas has the favorite. If favorite wins, Vegas gets 1 unit. If underdog wins, you get 3 units (75/(100-75)). The problem is, it's somewhat hard to standardize units across games this way--the risk and reward will be different for every game, which will make some games more important to the standings than others. I guess you could just have your sim "risk" one unit every time, no matter which side he's taking. I don't know, just sort of thinking out loud...

Xeifrank said...

Matt, thanks for the further comment. I am not calculating points the same way you did in your example. If Team A has a 55% chance of winning and my simulator has it pegged at 55%. Picking the favorite with a 100% confidence factors gives you 100 points, while my simulator would get 55 points, not 50.5 points. A system that put 100% on each "Vegas favorite" would be another possible benchmark for me to compare with. Last night the "100 method" would've correctly picked one out of four games for a total of 100 points while the simulator got 189.64 and Vegas got 187.04. The "100 method" is a feast or famine method, but one that should be easily beat by an intelligent system over the long term. I can back track and list that methods results as long as it keeps up. Thanks.
vr, Xeifrank

Xeifrank said...

Ok, the 100-fave method is now being tracked. Looking back at all the games I've simulated the new method is slightly ahead of the simulator, but still well behind Vegas. In games that Vegas has listed as 50/50 (-105/-105) I will give this method 50 points no matter what.
vr, Xeifrank

Anonymous said...

It's not about one result, it's about the long-term.

In the 55% example, let's say they played the game 100 times. 55 times the favorite would win, and I would get 100 points each for a total of 5500 points. The other 45 times I would get nothing.

Your simulator, meanwhile, would get 55 points 55 times, and 45 points 45 times, for a total of 5050 points.

Do you see what I'm saying?

Xeifrank said...

Ok, I see your math. I will have to ponder this for a little while. :) vr, Xeifrank