Can this be explained???

[njcurveball]

I am looking at the following query. I have about 2 1/2 months of data and I do not download most of the minor tracks.

*** PLEASE NOTE, all of my queries use a $2 ROI for break even. $2.10 = 5% profit, etc.

MY DATABASE
ML > 5 52,109 STARTERS, $1.41 roi

ML > 5 AND NKGAP > -2 (this gives me the close calls)
938 starters, $2.06 ROI!!! WOW!!!

take out the Unknowns
ML > 5 AND NKGAP > -2 AND NUNK = 0
870 starters, $2.10 ROI

ok now what?

After many queries, this worked, but can someone explain why???

RCLASS starts tot$ ROI
---------- ---------- ---------- ----------
0 53 111.4 2.10
1 230 363 1.57
2 180 338.3 1.87
3 161 275.5 1.71
4 99 321.8 3.25
5 63 155 2.46
6 33 120 3.63
7 25 109.4 4.37
8 14 39.2 2.8
9 12 0 0

So with the following query

ML > 5 AND NKGAP > -2 AND NUNK = 0 AND RCLASS > 3
246 starters, $3.03 ROI (over 50% profit!)

so again I ask, WHY does Brad Lidge even give Pujouls a pitch near the plate? OK since I cant explain that, WHY does a BAD class ranking lead to profits?

any ideas?
Jim

[MikeDee]

Jim, in MHO what you are doing is best described as back fitting. You are starting with a small sample and then reducing even further by selecting filters that improve either the ROI or win%.

Perhaps someone with a larger Db will run your test and report back. You can also play it forward. Wait a week and then run your test on a new week of data that is not part of your sample. Most of these things with high ROI's have a way of evening out over time.

[fred4now]

Mike is right, sample is just too small. If you want to see if this is backfitting look at it by intervals to see if just a couple of high prices is leading you astray. When I am researching stuff like this I usually look at the results by the month and year.

Here is the results for your queries run on 2005 data
ROI of 1.00 is break even. (It is a lot easier if we are on the same page with ROI)

nMLO>5 ROI=.72
nMLO>5, nKGAP>-2 ROI=.93
nMLO>5, nKGAP>-2, nUNK=0 ROI=.93
nMLO>5, nKGAP>-2, nUNK=0, rCLASS>3 ROI=.93

I could run this over 5 years of data, but I think the results would be about the same.
If you have any others you would like run, just ask.

[km]

It does seem to be a backfit at first glance, but i like NJ's premise here.

I hunted for it in the latest newsletter with those negatives (lost by 20+ and worst rank speed figure) to find an obvious factor that would deter public betting and yet with some factors of HTR would still hold its own with win% enough to produce profits.

No doubt that a poor Class rating will be apparent in the pp's for most other bettors. But a strong K rating may get some overlays in that group.

I think you are fishing in the right pond NJ, just as Jerry did with the Chaos races: look for situations that cause conflict and confusion with the majority of other bettors.

There is one price play factor in HTR2 that seems to defy almost all of the blatent negatives that handicappers perceive: HTR = 1

K110 and HF are nearly 100% resilent too, but they get only lower priced horses, yet when negatives are applied, they can produce positive ROI without hurting the win rate too much.

This area of handicapping - often called "contrarian" (embraced by Mark Cramer) seeks out horses that have a clear surface negative(s) while still maintaining strong underlying probability to win the race and thus positive returns.

[njcurveball]

Thanks for the reply Fred. I guess this query was bleeding money in the beginning of the year while the other one was winning.

They must have flip flopped to wind up with exactly the same ROI. Although a loss of 7% is a pretty encouraging start. :-)

Thanks for your help!

Jim
:)