◢The Composed Market priced on request Open the partner account
paid
Affiliate disclosure. The partner link in the masthead and in the band beside the copy on this page is a sponsored link to a partner operator, and this site may be paid if you open an account through it, at no extra cost to you. It carries rel="sponsored noopener" and opens in a new tab. A desk about the price of a market a customer asks for should not leave its own funding unsaid: one link funds the site, no operator and no market is named, rated or recommended anywhere on it, and the site carries no affiliate ranking of any kind.
The Composed Market / The refusals
A price exists, and the customer still does not get it

Why a request is priced and then refused

A refusal is not one behaviour. A request can fail because the combination cannot be modelled at all, because it can be modelled and the price is worse than the operator will offer, because a leg rests on a fact nobody will confirm, or because the legs are so tightly linked that no joint price is meaningful.

Desk spec
requests
500
accepted
218
refused on price
96
never priced
131
the composed priceThe number returned for a set of outcomes the customer named. It is not the product of the legs: on the samples 5.60 multiplies and 4.80 is offered, because the legs happen together more often than independence implies.
the requestA customer asking an operator to invent a market. 500 arrived in the sample month, 218 produced a price, 131 could not be priced at all, and the median answer took 3 hours 40 minutes.
the void legA leg that never happens, repaired by re-pricing the combination from its remaining legs. Dividing the price by the void leg gives 4.50 where the real three-leg price is 4.80 - a 6.7% error.
Direct answer

A refusal has four causes and they are not interchangeable. Of 500 requests in sample B, 131 could not be modelled at all and 96 were priced and refused - the 96 dividing into 38 with no model for the combination, 27 below the minimum accepted price, 19 resting on an unverifiable statistic and 12 whose legs were correlated beyond the limit.

Refused is not the same as impossible

The 131 requests that were never priced were never going to be: no model covered the combination, or no market existed for one of the legs. The 96 that were refused are a different thing entirely. A price was produced, it was a real price for a real risk, and the operator declined to offer it. That is a commercial decision about the shape of the book, and it is the only one of the four reasons that could have gone the other way.

What the samples show

Sample I - the 96 requests that were priced and refused, by reason
ReasonRequestsShare of the 96What the reader could have done
No model for the combination3839.6%Nothing: the market did not exist
Price below the minimum accepted2728.1%Ask for fewer legs or a shorter stake
A statistic the operator cannot verify1919.8%Restate the leg on a settled statistic
Correlation beyond the accepted limit1212.5%Nothing: the legs are too close to one outcome
Every priced-then-refused request96100%84 of the 96 could have been re-asked
sample I - what the refusals say about the product requests in the month = 500 never priced (no market existed) = 131 -> 26.2% priced and refused = 96 -> 19.2% of those refused, the reader could have re-asked = 38 + 27 + 19 = 84 of those refused, nothing could have helped = 12 -> 12.5% requests that never reached a customer as a price = 131 + 96 + 37 = 264 of 500 = 52.8% and the largest single reason on the whole queue is not a price at all: it is a market the operator cannot model.
sample I - the correlation limit, in one figure a combination whose legs are nearly the same outcome: leg A, the team to win = 62.50% leg B, the team to score first = 40.00% if the two happen together 55.00% of the time, joint chance measured = 55.00% -> composed price = 1 / 0.55 = 1.82 joint chance if independent = 0.625 x 0.40 = 0.2500 -> price = 4.00 the two prices differ by 2.18 = 120% of the composed price where the gap between the two answers grows faster than the model's confidence, the combination is refused rather than quoted at a figure nobody would accept.

Why the reasons are worth separating

A reader who is told only that a request was refused cannot tell whether to ask again with smaller legs, to reword a leg onto a statistic the operator settles, or to stop asking. The four reasons have four different answers: the first and the fourth are dead ends, the second is a question about size, and the third is a question about how the leg was written. None of the forty terms pages in sample F distinguishes them, and 7 state only that a request may be refused.

Five things to do with a refusal
  • Ask which of the four reasons applied. The answer decides whether asking again is worth anything.
  • If the reason is price, ask for the same combination at a smaller stake rather than for a shorter combination.
  • If the reason is an unverifiable statistic, restate the leg on a statistic the operator settles - and note that the price will change with the wording.
  • If the reason is correlation, stop: a set of legs that close to one outcome is not a market, whatever it is called.
  • Keep the refusal. A refusal that is not recorded cannot be distinguished later from a request that was never answered.

Read next