Churchill Downs plans to undertake a study of every US race run from 2004 to 2006 to try to determine the factors that influence handle. VP Tom Jenkins is leading the study and expects the complete the study in six to eight months. (Louisville Courier-Journal Article)

I have a similar study along these lines that is coming out in the American Journal of Economics and Sociology this month entitled “What Do Bettors Want? Determinants of Pari-Mutuel Betting Preference” (with three co-authors). Our study looks at 2,957 races at 12 racetracks from the Fall of 2002. The 12 tracks were Arlington, Aqueduct, Belmont, Churchill, Calder, Fair Grounds, Hollywood, Keeneland, Louisiana Downs, Laurel, Santa Anita, and Turfway.

The following are some of the results from the paper, none of which are surprising. 

Regression Results
	Win		Place		Show		Exacta		Trifecta
	Q1	Q2	Q1	Q2	Q1	Q2	Q1	Q2	Q1	Q2
Purse Elasticity	0.483		0.458		0.458		0.417		0.476
Stakes		24%		24%		35%		16%		15%
Starter Allowance		*		*		*		*		*
High Claiming		*		*		*		*		9%
Mid Claiming		-26%		-24%		-22%		-25%		-26%
Low Claiming		-44%		-40%		-37%		-41%		-48%
Optimal Field Size	10.9	11.0	11.0	11.1	11.9	12.1	12.2	13.0	10.4	10.4
Take Elasticity	-2.42	-3.14	-2.52	-3.21	-2.70	-3.48	-2.98	-3.64	-0.53	-0.72
Competition Index	-0.8%	-0.8%	-1.3%	-1.3%	*	*	-0.7%	-0.6%	2.2%	3.1%
Big Carryover	25%	33%	27%	35%	28%	35%	25%	34%	32%	43%
Maiden	*	-11%	*	-10%	*	-10%	-4%	-14%	*	-11%
Juvenile	*	4%	*	*	*	*	*	*	*	*
Turf Race	5%	15%	9%	19%	10%	20%	-5%	*	*	9%
Female	-3%	*	-2%	*	*	*	-3%	*	*	*
State Bred	-10%	*	-8%	*	-5%	*	-6%	*	*	14%
Off Track	-12%	-16%	-13%	-16%	-14%	-18%	-6%	-8%	-10%	-14%
Route	-5%	*	-5%	*	-3%	*	-3%	*	*	*
Other Races	-3%	-3%	-3%	-3%	-4%	-4%	-4%	-4%	-4%	-4%
Quinella Elasticity							-0.006	-0.004
Superfecta Elasticity									-0.012	-0.016
R squared	0.740	0.683	0.744	0.692	0.748	0.707	0.698	0.646	0.573	0.469
Observations	2957		2957		2957		2957		2957
*not significant

From the paper
      Two sets of SUR regressions are estimated to account for different measures of the quality of horses entered in a race. Regression Q1 uses the race's purse size as a measure of quality to determine the purse elasticity (the effect of a 1% increase in purse size on betting volume). Regression Q2 uses the five major race classifications and assigns dummy variables. Increased race quality does have a positive impact on wager dollars across racing pools. The estimated purse elasticity is inelastic for all wagers (between 0.41-0.48). When measured by race classification, the race track hierarchy, from low level claimers up to stakes races, is strongly related to predicted betting volume. Bettors prefer higher quality races. With allowance races as the control group, stakes races attracted an estimated 15-35% increase in betting volume, with the greatest effect on the show pool and the slightest effect on exactas and trifectas. Starter allowance and high level claiming races do not have a significantly different effect from allowance races. Mid-level (22-26% less) and low-level claimers (37-48% less) are less attractive to bettors.  The fact that stakes races are much more popular among bettors than all other races is not necessarily evidence of increased presence of informed bettors only, inasmuch as stakes races are heavily advertised and attract interest from all bettors. We would argue, however, that those who are not betting in claiming races are more likely to be the informed bettors.
       Races restricted to fillies and mares are found to slightly reduce volume for win, place, and exacta wagers in the purse size regressions only. Maiden races are shown to reduce betting volume for all wagers in the race classification regressions, and for exactas in the purse size regressions.  This evidence supports the idea that less information leads to less betting.
      For each wager, the number of betting interests is found to have a positive but diminishing effect on betting volume. The point at which increasing the number of competitors (optimal field size) in the race would actually begin to decrease betting volume is found to range from 10 on trifectas, to 11 on win and place wagers and 12 on show wagers and exactas. The estimates are not significantly different from one another, but the results indicate that the optimal field size from a betting perspective is between 10-12 horses. Since for each pool there is evidence of an optimal number of betting interests, there is support for the hypothesis that after some point, informed bettors reduce betting volume in response to increased noise.  
       A divergence in the betting volume by wager is found in estimates of the competitiveness of a particular race. A race is considered more competitive when the probabilities of outcomes are similar in a race.  The competition index is small in a more competitive race, therefore, an increase in the competition index indicates a less competitive race and a negative coefficient establishes that a more competitive race increases betting volume. A less competitive race is found to attract less money in the win, place, and exacta pools and more money into the trifecta pool. It is of interest that the highest probability wager, the show wager, and the lowest probability wager, the trifecta, both seem to attract more money (relative to other wagers) for less competitive fields. Perhaps this is because the informed bettors will play these races with more confidence when there are a few “sure things” that will finish in the money. If a race is contentious, informed bettors will stick to win, place or exacta wagers. Some bettors may focus on trying to make a big score, which in a wide open race can be done by betting to win or exactas, but for a less competitive race requires trifectas (unless favorites run off the board). 
       As for the estimated impact of track take, the elasticity of demand for a wager across tracks is estimated to be price elastic for straight wagers and exactas and price inelastic for trifectas. The other variable reflecting rate of return, a large pick-six carryover, has a significant impact on all betting pools, increasing betting volume by a predicted 25-43%. Building up large progressive carryovers helps build handle on all wagers.  Of course, high rates of return should attract both informed and uninformed bettors.  
       Contrary to our initial predictions, turf races attract additional volume for straight wagers (5-15% win, 9-19% place, 10-20% show) but there are no significant differences in trifectas and possibly a negative effect for exactas (-5% for regression Q1 while regression Q2 was insignificant). This is consistent with the findings of Ray (2002b, 2002c) who speculates that these races are “unique and often high quality.”  It may be the case that, like stakes races, advertising/high interest in turf races may attract all bettors.
       Suboptimal track conditions reduce the amount wagered across all pools. The distance of the race reduces volume for win, place, and exacta wagers in regression Q1.  Again, these reductions in volume support the hypothesis that bettors bet less when information is noisy. 
       As for competing races or similar wagers, quinellas and exacta are substitutes, as are trifectas and superfectas, though the effect is small in each case. Each additional competing race run in the same hour reduces betting volume 3-5% for each wager.
       The days of the week and the order of races did impact wagering. Saturdays are the most popular (and the control group) followed by Fridays and Wednesdays. Wagering increases throughout the day.