I wrote a review (in two parts:
http://www.bayesianinvestor.com/blog/index.php/2008/04/15/predictocracy-part-1/
and
http://www.bayesianinvestor.com/blog/index.php/2008/04/15/predictocracy-part-2/)
of Michael Abramowicz's book Predictocracy that is is somewhat more oriented
toward laymen than this mailing list is. This message discusses additional
aspects of the book that are mainly of interest to designers of prediction
markets.
I suggest reading that review before reading this message.
Abramowicz is worried about the fact that prediction market (PM) subsidies
such as the market scoring rule do not subsidize by a predictable amount
(i.e. they may end up losing less money than planned). It's unclear why he
considers lower than expected payments a problem. It might imply inadequate
incentive to price discovery, but the connection looks weak enough that
I prefer to use other measures to assess the adequacy of the incentive
(although I don't have any great measure in mind). He suggests making the
subsidy more predictable by paying an independent firm to provide liquidity
(i.e. a predictably small bid-ask spread). That approach provides a good
incentive for that market maker to figure out the optimal price, but that
market maker has a clear incentive to minimize the rewards available to
other traders. So it seems to be effective when we need one researcher to
figure out the right price (i.e. when PMs provide little advantage over
the best alternatives) and less effective at rewarding the aggregation of
information from many traders (i.e. when PMs are most likely to be valuable).
He points out a potentially fixable problem with the standard approach
toward subsidies - unless the market maker that is providing the subsidy
starts with an unusually good estimate of the right price, much of the
subsidy goes to the initial traders who correct the most obvious mispricing,
when we might prefer to concentrate most of the subsidy on the traders who
do the harder job of correcting the last small mispricings. A well thought
out schedule of starting with a small subsidy and increasing it as a
function of time and/or trading volume should improve the effectiveness
of the subsidy. Another solution that he suggests is the use of two PMs
(the first one used to predict the result of the second) with different
market scoring rules (I have little intuition about how well this would
work).
He has an interesting idea of limiting the amount by which a trader can
move the market based on the trader's reputation. One advantage of this
is to reduce the risk of the governemt treating the PM as gambling (by
only allowing traders who have demonstrated skill to make large trades).
It will also help reduce manipulation and public concerns about manipulation
(I expect public concerns will be more important than actual manipulation).
He has good suggestions about using market prices (mainly prices of
insurance) to improve government approaches to regulation. If banks needed
to get a small amount of private insurance in order to get FDIC coverage,
the prices of that insurance would be an effective substitute for many of
the detailed rules that are currently used to limit the moral hazard of
FDIC insurance.
On page 218, he provides some weak reasons to believe that PMs will be
more efficient than stock markets, mainly based on the lack of any systematic
uptrend in PM prices. Since his reasoning applies as well to many commodity
futures prices, that effect ought to be measurable based on available data
(has anyone done this?). He neglects effects which might cause PMs to be
less efficient, such as low volume and longshot bias for unlikely PM
contracts.

--
------------------------------------------------------------------------------
Peter McCluskey | When someone is honestly 55% right, that's very good
www.bayesianinvestor.com| Whoever says he's 100% right is a fanatic