Best time frame for HTR modeler

[Bernie]

The HTR modeler defaults to the last 100 races, but this number can be changed. Has anyone researched what time frame gives the best results? It seems to me that too many races would dilute the information on what is working best right now, while too little data could give erroneous conclusions due to chance happenings

[Glen]

Bernie, not sure of the answer but I does seem that the results do change with the
begining of the meet vs middle/end of the meet
spring vs fall
and so on.

I prefer to look at all races (995) and then short term (last 15 races or so) to see if there are any pace biases.

[tbrown]

I like to look at the Records option and see each line for the last 20 races-this gives me a good sense of what is winning right now. I only wish the velocity numbers were in that screen.
I like to see the overal last few hundred, too, to see what long tern trends are still working.

[Glen]

Last 40 Dirt Sprints at LAD PL1-Ev1 and Ep1 and F horses
10-22-01 LAD-Win Records: 040 PL: 1 Avg Par: 093 Avg Fld: 9.6
------------------------------------------------------------------------------
Factor Rnk1 Top2 Top3 Top4 Top5 Top6 LONGSHT ROI-1 ROI-2 ROI-3
----------------------------------------------------------------------------
MLO 25% 47% 65% 82% 88% 90% 00/07 $0.60 $0.74 $0.89
HTR 28% 50% 63% 70% 88% 98% 02/07 $0.73 $1.04 $0.94
(K) 30% 52% 68% 73% 88% 98% 01/07 $0.78 $0.98 $0.95
P-Scan 25% 43% 65% 77% 82% 90% 01/07 $0.88 $0.46 $1.10
POW 35% 47% 65% 73% 90% 98% 01/07 $1.18 $0.33 $1.46
CLA 20% 40% 52% 65% 75% 85% 00/07 {$0.54 $0.60 $0.63
PER 30% 45% 57% 68% 73% 82% 02/07 $0.95 $0.96 $0.63
PROB% 30% 52% 65% 77% 88% 93% 01/07 $0.79 $0.88 $0.66
EFF 22% 40% 50% 63% 73% 82% 03/07 $1.01 $1.10 $0.39
JKY 20% 43% 57% 63% 70% 77% 02/07 $0.61 $1.15 $0.85
TRN 13% 30% 47% 60% 73% 75% 02/07 $0.31 $1.20 $0.59
ACL 15% 32% 50% 60% 77% 85% 02/07 $1.04 $0.94 $1.59

¦ VEL 17% 40% 50% 60% 73% 80% 01/07 $0.58 $1.58 $0.43
¦ Fr1-Ev 30% 50% 65% 77% 88% 90% 03/07 $2.43 $0.66 $0.79
¦ Fr2 08% 22% 43% 50% 60% 70% 01/07 $0.31 $1.36 $1.41
¦ Fr3 03% 17% 28% 45% 57% 63% 01/07 $0.08 $0.73 $1.91
¦ A/P 20% 40% 52% 63% 70% 82% 03/07 $0.66 $1.49 $0.58
¦ E/P 32% 52% 63% 75% 85% 93% 03/07 $1.75 $1.39 $0.44
¦ S/P 15% 28% 40% 60% 68% 70% 01/07 $0.44 $1.04 $0.73
¦ F/X 20% 38% 47% 73% 85% 88% 02/07 $0.56 $1.44 $0.49
¦ L/P-Lv 08% 10% 22% 43% 57% 63% 00/07 $0.33 $0.06 $2.01
¦ (*) 25% 02/07 $1.24
¦
¦ ----------------------------------------------------------------------------
¦ RS Starters Won-Pla-Sho Win% Impact $ROI ITM% %Race Winners
¦ ----------------------------------------------------------------------------
¦ F 0008 004 001 001 50 4.79 4.38 75 10
¦ E 0083 013 011 008 16 1.50 0.73 39 33
¦ P 0112 015 013 013 13 1.28 0.68 37 38
¦ S 0146 008 012 015 05 0.52 0.68 24 20
¦ R 0034 000 003 003 00 0.00 0.00 18 00
¦
¦ ----------------------------------------------------------------------------
¦ p-Scan differential Win% AccumWin% Pla% Sho% Longshots
¦ ----------------------------------------------------------------------------
¦ 0.00 to -0.99 45 45 40 20 01
¦ -1.00 to -1.99 18 63 15 13 01
¦ -2.00 to -2.99 13 75 15 20 01
¦ -3.00 to -3.99 10 85 08 13 03
¦ -4.00 to -4.99 08 93 00 08 00
¦ -5.00 to -9.99 08 100 23 28 01

[Bernie]

Glen,

Would you rather look at the 40 race results you show above or the 15 race results mentioned in your earlier posting? What if the 15 race results looked good in certain areas but the 40 race ones didn't - or vice versa?

[Glen]

Bernie, good question and not sure of the answer. My generic answer would be what ever looks best. I had mentioned last week before the BBS contest that Ev1 was killer at LAD over the short term. When I made that statement it was over maybe a 20 race sample. Historically speaking(if I am correct), Ev1 and longshots have not been common place at LAD. But Carl was able to get a nice $19 winner Ev1 last Sat. Guess it is holding up.

Honestly, I would wager Ev1 will suck from here on out at LAD. So sell short on Ev1... Bet the ground crew gets back to doing what they do and get the rack back to normal. Just my guess since these trends generally do not last very long.

BTW Bernie, you made some nice selections last weekend. Especially on the turf sprinter. How would you classify yourself as a Handicapper...ie what method? Ev? Class? Trip? Modeling? All of the above?

[Glen]

One more thing...Check out the records on the last 40 dirt sprints at LAD. 34 of the 40 have an early ev/lv balance. Ie horse has lower Ev ranking than Lv ranking.
Example of Early Ev/Lv Balance: Ev3/Lv9
Example of Late Ev/LV Balance: Ev5/Lv1

Pretty amazing to me the early speed bias that only 6 of the last 40 winners were horse who ran late.

Win records LAD 040 races Paceline mode 1
Date R# D/S Par Horse Qp RS Ev Lv VEL CLA (K) HTR PScan_r $Pay MLr
------------------------------------------------------------------------------
22Oct 08 6.5D 093 Win 6 *F 1 8 5 4 5 4 -1.9 4 18 4
22Oct 07 7.0D 098 Win 7 *F 1 9 3 3 1 1 0.0 1 08 2
22Oct 06 6.5D 093 Win 3 P 1 9 5 7 6 5 -3.3 7 19 8
22Oct 05 6.5D 097 Win 3 P 3 9 9 1 2 5 0.0 1 09 2
22Oct 03 5.5D 091 Win 3 S x x x 9 5 5 -4.3 7 18 7
22Oct 02 6.0D 093 Win 2 P 4 4 2 6 6 5 -3.1 5 21 4
21Oct 09 6.0D 090 Win 3 S 4 4 1 2 1 1 -0.6 3 04 2
21Oct 08 7.0D 098 Win 3 P 5 2 1 7 4 2 -1.6 3 05 3
21Oct 07 6.0D 091 Win 6 P 1 4 1 1 1 1 0.0 1 05 1
21Oct 06 6.0D 092 Win 5 P 3 1 1 1 1 1 0.0 2 06 2
21Oct 04 5.5D 094 Win 6 E 3 6 5 8 5 5 -3.5 6 22 6
21Oct 03 6.0D 090 Win 2 S 4 3 4 9 9 9 -7.6 9 84 9
21Oct 01 6.0D 094 Win 7 E 1 5 4 6 3 3 -2.8 7 12 3
20Oct 09 6.0D 098 Win 7 *F 1 6 1 7 2 2 0.0 1 20 4
20Oct 07 6.0D 087 Win 5 P 1 5 2 2 1 1 -6.5 3 08 2
20Oct 05 7.0D 099 Win 2 E 3 5 2 5 1 1 -0.3 3 08 1
20Oct 04 7.0D 094 Win 4 P 3 7 6 1 1 2 0.0 1 04 1
20Oct 03 6.0D 087 Win 1 S 3 1 3 4 4 4 -2.5 6 14 4
20Oct 02 6.0D 098 Win 2 S 5 1 3 3 2 2 -2.1 2 06 1
19Oct 09 6.5D 090 Win 4 *E 2 4 1 2 1 1 -1.2 2 03 1
19Oct 07 6.0D 090 Win 4 P 2 4 2 1 1 1 -0.4 2 06 1
19Oct 05 7.0D 096 Win 3 E 2 7 2 3 2 3 0.0 1 07 2
19Oct 03 6.0D 093 Win 1 S 5 3 2 1 1 1 0.0 1 04 2
18Oct 09 5.5D 093 Win 6 E 4 4 5 2 2 3 -5.2 3 09 3
18Oct 07 6.5D 094 Win 6 *P 1 3 2 3 3 2 -4.1 2 05 2
18Oct 05 6.0D 087 Win 4 P 2 7 7 2 2 6 -0.7 2 08 2
18Oct 03 6.0D 089 Win 4 P x x x 6 5 5 -0.9 3 19 4
18Oct 02 6.0D 089 Win 4 *P 2 4 1 2 1 1 -0.9 2 03 1
18Oct 01 5.5D 094 Win 7 E 2 4 4 5 3 3 -2.0 3 16 3
15Oct 06 6.0D 091 Win 2 S 1 3 2 6 6 6 -3.1 6 58 5
15Oct 05 6.0D 098 Win 7 *E 1 5 2 2 2 2 -0.6 3 09 3
15Oct 03 6.0D 091 Win 6 *E x x x 1 1 1 0.0 1 03 1
15Oct 01 6.0D 093 Win 2 S 6 3 6 4 5 6 -1.9 5 10 3
12Oct 04 6.0D 101 Win 6 *E 2 5 3 1 2 2 0.0 1 06 1
08Oct 08 6.0D 097 Win 6 E 4 7 8 5 3 6 -1.1 3 10 3
08Oct 07 7.0D 098 Win 3 P 5 7 5 5 6 3 -2.3 4 31 7
08Oct 05 6.0D 097 Win 7 *F 1 9 8 3 3 2 -1.1 4 24 4
08Oct 04 6.0D 093 Win 5 E 7 7 9 4 3 4 -4.7 4 09 4
08Oct 03 6.0D 091 Win 6 E 1 5 4 4 5 5 -2.1 4 08 5 _
08Oct 02 6.5D 088 Win 1 P 2 9 6 2 2 2 0.0 1 04 1 _

[delayjf]

Bernie,
Good question, I'd agree with the other respondants with this caveat. One featured I'd like to see installed on the modeler would be the ability to model by date/day of the week. When I use to make my own speed figures (Beyer style) I often found that the track was almost always faster on the weekends than the weekdays. I assume this to be because of track maintenance performed on the dark days. But if the speed of a track can change, so can any prevailing bias (early, late, %early etc) A classic example of this occured (in my opinion)this summer at DeLmar. The day before the Pacific Classic everthing was won with wide sweeping runs, lots of long-shots won is this manner. The day of the Pacific Classic was run, speed on the rail was dominating. Which is one reason for Skimming's huge win. Other friends of mine have notice this type of track bias shift at Hollywood Park on Fridays/Saturdays. To me it would be interesting to relate track maintenance to track bias.
I would also be interested in modeling by class as well. Is speed more effective at certain classes than at others? This type of analysis would have to be conducted at a meets end to get enough races to give the data more validity.
None of the above would probably help a handicapper short term, but depending on what you find, you may be in a postion to zig, while the public is zagging. If you could predict that, you'd be way a head of the crowd.
Good Luck
delayjf

[Bernie]

Delayjf,

Great ideas. Have you thought about sorting your racefiles into 2 groups - weekdays and weekends? Regarding class, have you noticed big differences between alw/stk and claiming races that the modeler lets you look at?

Glen,

My handicapping is still evolving but tends to be very data driven. Would you care to share more about how you integrate modeling into your other handicapping practices?

[delayjf]

Bernie,
I'm not that far yet, However, a weekend / weekday sort may or may not reveal as much as a track maintenance / day of the week sort. While you can separate claiming from the other races, I'm not sure lumping all claiming races would give you as definative an answer as say a 10,000 clm sprint model vs a 50,000 clm sprint model, again the problem is the small sample size which is why you would need to do this at the end of a meet. But the knowledge learned could set you up at the next meet and allow you to narrow your focus a bit. Perhaps Ken (Massa) has already done this or something simular. Ken are your there?
delayjf

[Glen]

delayjf,

If you get into access, you can test data by day of the week and also by Class ranges or Pars. Also, the reason why KM put the days of the week into the HTR2001 Past Performace was for that very reason you speak of. Better horses/more competitive races are run on the weekend vs weekday.

Bernie,

Below is a link to an old thread that mentions some modeling stuff that I look for.
http://www.homebased2.com/forums/showthread.php?threadid=400