Subject: Re: Support as insurance
From: "Stephen J. Turnbull" <>
Date: Tue, 30 Nov 1999 12:34:09 +0900 (JST)

In re: insurance economics.  Academic economics actually matters
directly to insurance; the effects of moral hazard and adverse
selection are sufficiently strong that not understanding them can put
you out of business.

>>>>> "Ben" == Ben Tilly <> writes:

    Ben> Judging from your response to Ben Laurie I think we can agree
    Ben> on the following statement:

Please don't; you'd both be at least partially wrong.  :-)

    Ben> "The potential for an insurance company to make a healthy
    Ben> profit is dependent upon the fact that they are in a position
    Ben> to diversify the risk of your disaster over many insurers and
    Ben> you are not.  The fact that insurance companies do
    Ben> consistently make healthy profits depends upon their using
    Ben> available information better than their competitors do."

First, terminology:  insured = the person subject to risk, insurer =
person selling insurance against risk.

Second, profitability of insurance companies does not depend entirely
on diversification.  Insurance companies are financial intermediaries.
They cannot depend on the law of large numbers entirely, but must
build up reserves against the relatively rare event that many
independent risks all occur at the same time.  In the meantime they
invest the reserve, in bonds or stocks.  So if the capital markets are
doing well, the insurance companies also do well.

Third, diversification is not a necessary consequence of many
insureds.  Y2K is not fully diversifiable by having many insureds, if
they all use the same program there is _no_ diversification.  Even if
they all use different programs, if they are all based on the same
BIOS or system library there may be no diversification.  Insurance
companies must understand the nature of the stochastic process
underlying the risks very well.

Fourth, using available information well is important, but not always
legal (eg, in some states life insurance companies are not allowed to
charge whites or women more for annuities, although auto insurance
companies are allowed to charge women less for liability insurance).
This means you must go out and _collect_ information actively.  Again, 
understanding how the insured risk correlates with observable
characteristics of the insureds is critical.

    Ben> What you illustrate very well is the role of equal access to
    Ben> information.  In a free market an insurance company that is
    Ben> allowed to discriminate on significant risk factors whether
    Ben> or not they are "fair" (eg pre-existing conditions, gender,
    Ben> race), will consistently win over one that does not.  Unless
    Ben> government steps in and regulates what factors may be
    Ben> considered, the result is a disconnect between social notions
    Ben> of "fairness" and economic considerations.

Even when the government steps in, in practice what happens is that
those factors are considered, albeit in attenuated fashion.  Not
allowed to do HIV tests?  Check (through thorough background checks)
whether the person is homosexual.  Not allowed to invade privacy that
way?  Demand an interview, and depend on the interviewers' judgement
about whether the person is gay.  Get your wrist slapped because an
interviewer blows the whistle?  Red-line (a term in banking and
insurance for discrimination conducted on the basis of residential
geography when direct discrimination is not allowed for some reason)
the Castro district (an area of San Francisco famous for its culture,
one part of which is that many gays live there).  And so on.  (I don't
know whether any of these specific practices actually occur, but
similar ones are widespread.)

The problem is that the more indirect the information, the less people
with low risks benefit from insurance (as probability of loss
increases, "deductibles" and premia increase), and eventually their
net benefit becomes negative.  So they are chased out of the insurance
market altogether, raising the risk, and therefore the prices, in the
risk pool which actually buys the insurance.  This phenomenon is
called "adverse selection," and is related to but quite different from
"moral hazard," which Russ has mentioned.  (There is also the
"unfairness" to those who possess the selection characteristic but not
the characteristic relevent to the risk.)

Economists recommend allowing the insurance company to collect the
directly relevant information to insurance pricing, and have an income
transfer policy to balance unfairness in endowments.  Of course, that
assumes no social stigma, such as the very strong one long associated
with homosexuality.  But this in general should be the case for
software.  (Although the "cookie" application suggested below gives me 
pause, given that 90% of the spam that gets through my filters now is
amateurs advertising porno web sites.  There must be a lot of them,
and I don't suppose you want your s/o knowing you frequent them....)

I don't know how you would apply this to the topic at hand (pricing
software warranties).  But I would start by collecting browser cookies 
on my web site.  Not to recommend this to FSBs, but this is one reason 
why big vendors do that kind of thing, I'm sure.

    Ben> If this kind of model proves popular then there is a
    Ben> potential for establishing a reinsurance market, probably
    Ben> linked with a certification service.  In this model a small
    Ben> contractor would get certified by the company as producing
    Ben> quality work, and would then be authorized to provide
    Ben> guarantees as above and reinsure the risk with the
    Ben> certification provider.  Questions about providing

Small quibble.  You don't reinsure with the certifier.  Think of bond
raters like Moody's.  You would be more likely to contract with a
company with a good rating.

Also, I don't think a company would get certified, or at least it
wouldn't stop there.  I doubt you could sell (re-)insurance at a
profit based on a company certification due to moral hazard.
Certifiers would also certify divisions, projects, and products.  So I
also see specialists functioning like UL: Y2037 Laboratories, I18N
Certification Inc, etc.

    Ben> certification for free software come up from time to time, as
    Ben> well as questions of how valuable it is.  Well this is one
    Ben> model for how you can do it and make the certification truly
    Ben> meaningful.

Russ Nelson wrote:

    >> That's one of the reasons why medical insurance (the real
    >> thing, not the ersatz stuff that employers provide) doesn't
    >> cover pre-existing conditions.  And if you seen an insurance
    >> plan that does, it's not insurance, it's socialized medicine.

It's not socialized medicine.  Socialized medicine is compulsory, paid 
for by taxes.  It's a (presumably) convenient bundle of insurance and
prepaid care.  Often with one subsidizing the other.

University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences       Tel/fax: +81 (298) 53-5091
What are those two straight lines for?  "Free software rules."