dominant assurance contracts for funding public goods, especially free software

[Kragen Sitaker]

Open-source software, and increasingly information in general, is a
public good.  Information generally is beneficial and nonrival, and
it's becoming increasingly difficult to exclude non-payers from its
benefits.  Open-source software in particular explicitly makes no
pretense of excluding anyone and makes a virtue of necessity.

Public goods
------------

Generally public goods tend to be underprovided; that is, the body
politic spends little enough effort producing them that a little more
effort would have greater social benefits than social costs --- but
the benefits are available to everyone, while the costs are borne by
the one person making the incremental extra bit of effort.

We have four general ways to provide public goods.

First, individuals or firms can provide them unselfishly; for example,
Richard Stallman spent many years writing free software to give away
to everyone, so that today we could be free of proprietary software
licenses.  He would be a happier man today had he spent those years
not working with computers at all, but everyone else is better off
today because of this enormous and ongoing sacrifice.

Second, individuals or firms can provide them selfishly as long as
their own benefit from the public good thus produced exceeds their
cost.  For example, if it's worth five minutes for me to have a
command to change my Emacs colors to red on black, and I can build
such a command in that time, I will produce it.  The fact that
hundreds of thousands of other people may save three or four minutes
because I did this is of no consequence.

This strategy justifies larger amounts of investment for larger firms.
For example, if a million US dollars spent on Linux development by IBM
produces 30 million US dollars of excess value among users of
computers in general, and IBM's Global Services unit has a 10% market
share and can extract half this excess value, IBM has made back 1.5
million dollars on its million-dollar investment.

This is an important reason behind most free/libre software
development today, and it explains some observed phenomena, such as
large IT companies contributing disproportionately to the open-source
effort and the "scratch an itch" effect.

Third, consortia (or cartels) of individuals or firms can work
together to pay for the production of public goods.  In a way, this is
a variant of the second strategy, but the consortium might be
shorter-lived and more flexible than a traditional firm.  These
consortia suffer from prisoners' dilemma problems: all the
participants benefit from the provision of the public good, whether or
not they pay their dues each year.  As with other prisoner's dilemma
problems, these consortia require a lot of effort and social
interaction to sustain, work better when they're long-lived, and work
better with a relatively small number of members.

Fourth, governments can legally require the payment of taxes to fund
public goods such as hospitals, policing, schools, public health, and
PyPy.  The existence of governments in general is justified by
prisoner's-dilemma problems, and funding public goods by many
individual contributions is just such a problem, as explained in the
previous paragraph.

One common structure for a consortium that minimizes the risk to
contributors is the "assurance contract," in which if some
participants fail to pay their dues (or, in some cases, if
sufficiently many participants fail to pay their dues) then everyone's
dues are refunded.  This dramatically reduces the magnitude of the
prisoner's-dilemma problem: if you do not pay your dues, you may still
be able to "free ride" on the dues of others, but if you do pay your
dues, you at least know that they will not be wasted.

Dominant assurance contracts
----------------------------

In 1996, Alex Tabarrok invented a modification of the assurance
contract he calls the "dominant assurance contract," which may
represent a fifth path to public goods that obsoletes the second,
third, and fourth, in most cases.  That is, it may supersede taxation,
traditional cartels and consortia, and provision of public goods for
private benefit.

In the form he described in his paper, a dominant assurance contract
is a contract offered by an escrow agent to members of the public; it
is of the form, "In exchange for my payment of $S, if K or more people
donate $S to this same cause, the escrow agent will create public good
X (at cost $C), and keep all the donated money.  If fewer than K
people donate $S, the escrow agent will return $S + $F to each of the
donors and will not create the public good."

The difference between this and the standard assurance contract is
that the escrow agent is risking $F * (K - 1) of their own money on
the proposition that many people will donate; if enough people donate,
the escrow agent gets to keep the excess over $C.  They are permitted
to set K larger than C/S, but presumably if they set it too much
higher, contributors will set up another escrow agent who's less
greedy.

In the standard assurance contract, if you believe that other people
will not commit money, you have no incentive to commit money.  In
Tabarrok's invention, you do have this incentive.  His mathematical
analysis shows that in such a situation, even with limited information
of one another, rational agents have a substantial probability of
funding every public good that is expected to be more valuable than
its expected cost of production.

I haven't finished understanding Tabarrok's analysis of the
limited-information case.

I suspect that the analysis extends to a more general case, in which
each contributor chooses the amount of their own contribution $S, the
escrow agent performs the project if the total contributions are over
some basic amount, and the extra refund is a specified percentage of
the contribution rather than a specified dollar amount; but Tabarrok
does not mention this in his paper.

Tabarrok also doesn't extend his analysis to a multi-round case; if
contributors expect that the knowledge they reveal by committing
donations in a failed first round could be used to price-discriminate
against them in a second round, they might withhold their donations if
the price $S is nearly as high as the expected value they will receive
from the public good.

Consequences for Free Software
------------------------------

It should be possible for well-known projects and people to fund
future work in this way.  Contributors might offer money (which may or
may not need to be held in escrow) or in-kind contributions, which are
often more valuable on software projects.  In some cases, individual
features or bugfixes might have separate funds; in other cases,
contributors might simply provide a general budget for some period of
time to a group like the Apache Software Foundation, which would
allocate it as they currently do --- but from a much larger base of
contributors.

Giving free-software users an incentive-compatible way to fund small
open-source projects could dramatically expand the
free-software-writing community and reduce its current growing
centralization in large firms like IBM.

Consequences for Movies
-----------------------

At present, the movie industry is quite concerned about the new ease
with which people can copy movies.  Obviously for any individual
movie, this is a pure benefit: more people can enjoy the movie with
less work than was previously required.  However, it also converts the
movies into a public good.  They've always been nonrival; now they're
increasingly nonexcludable.

Much of the money spent to make a movie is spent by a studio during
the filming and editing of the movie, and only a small fraction of the
royalties from playing the movie in theaters, selling it to video
stores, etc., ever returns to the actors, editors, and other workers
who created the movie.  The studio invests the expected box-office
value of the movie, minus its expected profit, into the making of the
movie.

In effect, the studio is acting as a risk-arbitraging proxy for the
moviegoers.  The studio assumes the risk that the movie will turn out
worse than expected, and accepts the benefit if it turns out to be
better than expected; the moviegoers pay the actual value they expect
to receive from watching the movie (having heard others' reviews, and
being limited to a fixed set of ticket prices); and the studio carries
a large negative "float" in the mean time, from the time filming
starts to the time the box-office receipts start to roll in.

The moviegoers could just as well use a dominant assurance contract to
fund the movie production up front, assuming the risk themselves, and
paying a lower "ticket price" as a result.  Presumably they would fund
a substantial number of movies that turned out not to be good enough
to bother watching, so they would probably end up spending about the
same amount of money per movie they actually watched.  (Minus, of
course, the profits of the movie studios.)

Moviegoers who wanted an intermediary to continue to aggregate risk on
their behalf could still use "studios", funded through dominant
assurance contracts; "producers" would manage a fund of donations by
evaluating screenplays, directors, actors, etc., and fund several
films each year from their fund.  Producers who made worse choices
than people would have made on their own would find themselves out of
a job.

Tabarrok's paper claims to cover the case of increasing-returns public
goods with a point of sharply decreasing returns (at which spending
should stop); I think dominant assurance contracts will work for many
kinds of public goods, but moviemaking is an excellent fit to his model.

However, copyright places the risk on the artist, while dominant
assurance contracts place the risk on the artist's fans.  So dominant
assurance contracts may be more appealing to established artists than
to unknown or avant-garde artists.

Consequences for Government
---------------------------

Government money goes to two places: public goods and corruption.
Taxation is an extremely effective way to fund public goods, but
because both taxpayers and government officials can improve their lot
by cheating, corruption invariably accompanies it.

Some government functions cease to work properly if they are beholden
to those who benefit from them; police and courts, for example.  These
might not be candidates for funding from dominant assurance contracts.
Other government functions are necessary for the enforcement of the
dominant assurance contract itself; for example, national defense and,
again, courts, might not be fundable in this fashion.

However, especially in socialist places like France and San Francisco,
much public money is merely spent on things expected to benefit the
public more than their cost, but whose loss would not result in social
collapse.  For example, public transit, emergency care, street lights,
unemployment benefits, public schools, and basic research.

Especially local government projects could be funded this way.

Probably initial construction could be funded in this way better than
ongoing maintenance could.  Imagine that each year, the citizens
decided whether to put money into paying teachers that year; if the
contract fails one year, it will be much harder to restart the school
the next year, since all the teachers will have gone off to find new
jobs in the mean time.

Perhaps this can be solved with some kind of variant of the idea.

Objections to Dominant Assurance Contracts
------------------------------------------

I've talked to several people about this idea, and they've raised many
interesting objections.

Some point out that the models used in the proofs in Tabarrok's paper
assume that people are motivated solely by self-interest, a
demonstrably incorrect assumption, and so the proofs may not hold in
the real world.  For example, revenge and altruism are other common
motivations.

My answer is that this institution is merely a technique for
eliminating a conflict between self-interest and altruism in a
particular situation.  The game-theoretic models only show us which
direction self-interest points in a particular situation.  I believe
that funding of public goods that can be motivated by either altruism
or self-interest will work better than funding motivated only by
altruism.

I might be wrong; because the escrow agent, not the public, stands to
gain if the contributors are too generous, people might be
disillusioned if they contribute from altruism, and then they will
want revenge.  I think it's possible to avoid this situation, though.

A second objection is that real-world projects involve some risk, even
if funding is obtained.  For example, a bridge or a movie might go far
over budget, or even remain incomplete without further funding; a
software project might turn out to be less useful than anticipated; or
the escrow agent might just take the money and run.  This is a
particular problem in the kind of increasing-returns public good
Tabarrok wrote his paper about, since an unfinished bridge is no more
useful than no bridge at all.

In Tabarrok's model, all of these risks are lumped together as a
discount on individuals' expected value from the public good.  If you
think there's a 50% chance the bridge won't be finished, your expected
value from the bridge will be 50% lower, and so you'll only be willing
to risk half as much money.  Many public goods would still be fundable
under this constraint, since their returns are orders of magnitude
larger than their cost.

I think that the escrow agent has a self-interest incentive to
minimize these risks as much as is cost-effective, especially if they
can do so in ways other escrow agents can't.  In moviemaking,
completion bonds are sometimes used to decrease this kind of risk.

A third objection is that Tabarrok's model does not talk about
iterative games.  It induces preference revelation in the single-round
version he analyzes, and in a multi-round game, that might matter.
For example, the escrow agent might try to price-discriminate in a
second round, and the prospect of price-discrimination might
discourage honest preference revelation in the first round.

As far as I know, nobody has done the analysis to see if this is a
problem, nor analyzed possible fixes.

Some people have expressed doubt that such an institution could be
established in the real world because of their cynicism about
governments' motives.  This misses the point that this institution
needs no special support from the government --- merely enforcement of
financial contracts, which many governments already provide, and which
is sometimes provided by nongovernmental means as well.

For example, a group of grad students can use this institution to fund
a photocopier in the grad-student lounge at, say, UCLA; a group of
friends can use it to fund a hot tub at one of their houses (if they
all visit frequently); a neighborhood can use it to fund
neighborhood-watch efforts, litter pickups, or improvements to a
neighborhood park; PTAs can use it to fund school supplies for their
kids; users of a piece of software can use it to fund bug-fixes or
feature enhancements; employees of companies can use it to fund
ombudsmen or group events; and political activism groups can use it to
fund actions such as protests or lawsuits.

We can expect that, if this institution provides some public goods
better than taxation in the real world, it will gradually supplant the
use of taxation for those goods, and the responsibilities citizens
delegate to governmental institutions will gradually diminish.

One person asked about the distinction between this institution and
completion bonds, as used for moviemaking.  A completion bond is a
kind of insurance policy against going over budget.  It has very
little in common with a dominant assurance contract.

Someone also asked whether it was possible to model a dominant
assurance contract as a normal assurance contract with a separate
prediction market, like the Iowa Electronic Markets, in which people
traded idea futures on the likelihood of the completion of the
funding.  I don't know how to model it in those terms, although it
might be possible.

Finally, Tabarrok suggests that the market for escrow agents should be
highly competitive because there are low barriers to entry --- all you
have to do is write a three-line contract and hold some money,
assuming that the possible contributors first hold some kind of
competition to select which escrow agent they want to use.  I think
that's a big assumption, and that escrow agents are likely to wield
substantial market power by virtue of network effects, and
consequently extract substantial profits from this business.

A well-known escrow agent will be able to attract many more
contributors, and so will be able to require much less money from
each, which is likely to be a large incentive to use the well-known
agent.

Credits
-------

Most of the ideas in this essay, other than Alex Tabarrok's original
paper, came from conversations with with (in approximate chronological
order) Gordon Mohr, David Weekly, Meredith Patterson, Jon Atkins,
David Braginsky, Chris Hibbert, Rohit Khare, Ilya Shpitser, Vadim
Kogan, Alex Russell, R. Michael Harman, Jade Hoffman, Ka-Ping Yee,
Norm Hardy, Mark S. Miller, Alan Karp, Riana Pfefferkorn, Benjamin
Rahn, Chad Woodford, Seth David Schoen, Sean Wieland, Matthew
O'Connor, Faisal Jawdat, Russ Nelson, Mike Linksvayer, Melinda Owens,
Tim Sturge, Ryan Oprea, Bernd Ankenbrand, Sunah Caroline Cherwin, Simona
Nass, and several others.