Power, Not Fear: A Collusion-Based Account of Betting Market Inefficiency

Abstract

We examine racetrack betting market inefficiency. We argue that the overround – an established measure of inefficiency – may be seen as a reflection of bookmakers’ collusive returns maximising behaviour, rather than simply as their response to an adverse selection problem. We test, and find empirical support for, several proposed market effects of bookmaker (market maker) collusion. Besides number of race participants, overround varies significantly according to (1) whether or not the race is the last at the race meeting, (2) racetrack location, (3) day of the week (weekday vs. weekend), (4) type of race (handicap vs. non-handicap), and (5) type of race event (Flat/National Hunt). The study provides fresh insight into the origins of betting market inefficiency, where the link between market structure and context and bookmaker behaviour is a key feature. Regarding the broader implications, the study alerts us to the potential for financial market inefficiencies such as the bid-ask spread to be explained, in part, as a consequence of market makers’ opportunistic behaviour wherever and whenever it is possible for intermediaries to exploit their understanding of investor behaviour.

Key Words:

• Market Inefficiency
• Overround
• Tacit Collusion
• Betting

JEL classifications:

• D03
• D43
• G14

Notes

1. The study’s findings have wider appeal, given the accepted similarities between markets for state contingent claims (betting markets) and other financial markets (see, e.g., Levitt, 2004; Snyder, 1978). Shin (1992, 1993) explains the similarities between the bookmakers’ OR and the bid-ask spread in other financial markets.

2. Bookmaking is explained further in Section 3.

3. On very rare occasions, sum of implied probabilities may be <1 (Coleman, 2007), in which case we would have an “underround”.

4. While we use a horse-race example to illustrate the OR and betting market efficiency, the points made apply to any event, sporting or otherwise, for which there is more than one possible outcome, and for which bookmaker odds are available.

5. We say “in principle” because the ability of the OR to signify an overall transfer of monies from punter to bookmaker depends on the “making of a perfect book” (see Dowie, 1976). A perfect book exists, to the extent that the bookmaker is able to lay each possible event outcome to lose the same amount of money. In practice, market complexities often prevent a perfect book from being realised. Some bookmakers may also adopt deliberately long positions with one or more race entrants potentially to enhance profits while increasing risk exposure.

6. Any returns to the bookmaker must contribute to covering the costs of bookmaking. These costs can include, for instance, administrative costs, travelling expenses, licensing fees, salaries and wages, and bad debts. For large bookmaking businesses such as Ladbrokes and William Hill, the advent of person-to-person Internet betting exchanges (e.g., Betfair.com) has and is threatening bookmaker margins, as demonstrated in an apparent reduction in OR per runner. While acknowledging these points, our study focuses on examining race-based factors which may influence the magnitude of the OR, independent of the effects created by external pressures.

7. The total number of bookmakers permitted to operate at any given race meeting is determined by the number of list positions or “pitches” available. This number is set by the British Horseracing Authority in conjunction with racecourse owners. A key determinant is the physical size and layout of the “betting ring” at the particular racetrack. Within the total permitted, the actual number of bookmakers operating at a given race meeting will depend on, for instance, the quality and nature of the race meetings (some bookmakers prefer Flat to National Hunt racing), the number and geographical proximity of alternative race meetings on that day, the location of the particular racetrack, and the expected number of race goers. For the individual bookmaker, a significant determinant is whether or not s/he has or can obtain a licence to operate at a given racetrack.

8. In the context of racecourse attendance, we categorise Fridays as a weekend rather than a weekday. Compared to Monday to Thursday, Friday racecourse attendances tend to include considerably more “works outings” and similar organised events involving recreational race goers.

9. A list of the racetracks comprising each category is available on request.

10. The reasons for excluding Sundays from data collection are threefold: (1) the generally lower frequency or incidence of Sunday race meetings compared to other days of the week; (2) the selectivity of Sunday racing (not all racetracks engage in Sunday racing); and (3) given the still relative “newness” of Sunday racing, the greater uncertainty regarding the type of clientele frequenting race meetings on a Sunday.

11. Bank holiday Mondays were excluded from data collection, given the likelihood of significant numbers of recreational bettors patronising race meetings on these days.

12. The 994-race sample comprised 539 Flat races and 455 National Hunt races. There are 463 handicap races and 531 non-handicap races.

13. Basically, the closer both values are to one, the lower the collinearity among the predictor variables (Blaikie, 2003).

14. An alternative would be to use interaction terms to test our five second-order propositions. In each case, a separate variable representing the product of the interaction between the predictor variable and number of runners would be added to a regression model comprising the five main effect variables (e.g. last race effect, race type, etc.). Thus, there would be five additional variables (interaction terms), the significance of which would indicate the sensitivity of the predictor variables to field size. Unfortunately, however, the use of binary variables (1, 0) to measure the predicted market effects renders the use of interaction terms problematic. A particular problem is that the interaction between, say, Flat-NH effect (Flat = 1, NH = 2) and number of runners (range 3–18) creates a product term for which a score of, for example, 10, could represent either a Flat race comprising ten runners (1 × 10) or a NH race comprising five runners (2 × 5). The mathematical problems (and problems of interpretation) are even greater for binary variables of 1, 0. For these reasons, it is inappropriate and potentially misleading to employ interaction terms to test second-order propositions.

15. This argument draws on the point that horses competing in handicap races tend to have accumulated more public form than horses competing in non-handicap races (which can sometimes involve race entrants, all of which are having their first public outing. Examples are maiden two-year-old races taking place at the beginning of the Flat season). The potential for insider trading is thus considered to be greater in the latter than the former.