Re: Goodheart's Law

["John O'Donnell" <jackodonnell@xxxxxxxx>]

Jonathan D Halvorson wrote:
>
> "Goodheart's law" is surely an overstatement, but not quite in the way
> that John asserted (which is the same thing I've always heard in the past:
> all and only noncausal correlations collapse when used for control
> purposes.)

For now I will concede that it MAY be true that it is not
only non causal correlations collapse under use as controls
but I will still take the position that the non causal
condition is a true requirement. The basis of this is that
an asserted cause of a non reliable correlation may not be a
true cause. Primary among these untrue asserted causes is
the behavior of people. Such a "cause" is merely an
expectation and not an immutable cause.

> Instead, the correlation can collapse even if it does reflect an
> underlying causal relationship. In particular, you can expect it to
> collapse if the regularity is being exploited at the expense of those
> whose actions constitute the regularity. To put it very simplistically: A
> rational agent has an interest in avoiding such exploitation, and must
> alter her behavior so that she is no longer exploited in order to remain a
> rational optimiser.  Since it is ridiculous to assume perfect
> information, computational power, etc., whether or not an exploited
> regularity actually will change depends on the information agents receive
> and understand, on whether all things considered it costs more to change
> than to allow others to profit at one's expense, and so on.

As I read this, this is an assertion that just such a source
of correlation (i.e. -- the actions of an agent) is being
asserted as causal. Without further argument I concede such
a correlation is likely to fail, if not sooner then later.
To put it in my limited mathematical nomenclature (Recognize
also that the "d" presented here will mean either partial or
total derivative, as the case may require.):

Let inflation = Inf and X(n) be any influence on Inf, then:

d(Inf)/dt = Sum {[d(Inf)/dX(n)]*[dX(n)/dt]} for n = 1 to ??,
where one value of n represents the quantity of money and
another represents a behavior of people to allocate the use
of that money while other n's represent the many other
things that influence the value of money.

I say that no matter what people chose to do with their
money, the fact remains that because [d(Inf)/dM]*[dM/dt] is
always positive then no matter what the choices made by
people in allocating the use of their money, the supply of
money can always be changed to compensate for that action.

However, because the actions of people cannot be predicted
with any kind of certainty, any correlation that asserts
that there will be any particular amount of change in Inf
for a given change in M will certainly fail. Only the
direction of change that would have been, absent the
influence of people's behavior or any other non accounted
for effect, will be consistent and the amount of change in
the quantity of money, given that money is defined as the
fiat issue of a government, is unrestricted except that it
cannot go below zero.

Any correlation that asserts a relationship with any X(n)
that is based on people's behavior and Inf will also fail.

> Other correlations, on the other hand, may actually be strengthened when
> they are transparently used for control purposes. The effect on the
> market when the Fed changes interest rates is surely made more
> reliable because people expect it to have a certain effect.

This is another of the false indicators of control. The
control in the case of monetary policy is always the
availability of money (i.e. -- the issue of the CB) or the
most prevalent money substitute (i.e. -- intermediated
credit of commercial banks) and not the inter bank interest
rate "set" by the CB.

The interest rate is only one of several variables affected
by changes in the availability of money. ("Money" now used
to also include money substitutes.) The relevant measure
that is also affected by such changes in the quantity of
money is the value of money.

An example of the difficulty of using the wrong variable as
the target for monetary policy is the recent Japanese
experience. There the target inter bank rate was taken to
zero but the currency was was allowed to deflate. The
correct response would have been to continue creating money
even while the inter bank rate was at zero such that the
value of the currency would be maintained constant. This
would have, in turn, monetized the government's debt and
eliminated any concern about fiscal irresponsibility.

This last is one of the problems that becomes visible when
one stops treating the debt denominated in the currency of
an issuing government the same as debt issued by
corporations or governments that must be repaid with
something other than its own issue. While the accounting of
such debt held by the CB is called "debt held by the public"
(at least it is accounted as such in the US) it is in fact
debt held by the government and could just as easily simply
be taken off the accounts without any consequences.

> If I have one major beef with most economists, including many
> post-keynesians, it is that they equate causal relationships with highly
> stable causal regularities. When we are dealing with the
> causes and effects of actions (reasons-responsive agents), there is no
> justification for this assumption. It must be established by the evidence
> in each case, and even then the possibility for variation in the
> regularity must be respected.

Does this explain the difference to your satisfaction? Or,
do you have an example of an asserted causal relationship
that does not fall in this category of "behavioral"
consequences?

> On Fri, 29 Dec 2000, John O'Donnell wrote:
>
> > GGard97342@xxxxxx wrote:
> > >
> > > Is not the Phillips Curve a case study for Goodhart's Law?
> > >
> > > The story is that Phillips was taken to see Edward Heath, British Prime
> > > Minister  1970-74, and persuaded Heath to use his theory as a basis of
> > > policy. The result was the curve jumped several notches to the right on the
> > > inflation scale.
> > >
> > > Most people regard Goodhart's Law as a good joke. I believe it is far more
> > > than that, and feel it is a pity that the rest of Goodhart's work is not as
> > > soundly inspired.
> >
> > Goodhart's Law -- "Any observed statistical regularity will
> > tend to collapse once pressure is placed upon it for control
> > purposes" -- is an over statement of the problem. It isn't
> > that statistical regularity will collapse upon use; it is
> > only such correlations that lack causal relationships that
> > will collapse. The Phillips curve is a relationship that
> > lacks causality, dQ/dM = 0 is not.

--
			-- jbod

		Tax Privilege, Not People
___________________________________________________
Come visit and see a new economic perspective --
       http://www.geocities.com/CapitolHill/1067
           Comments/arguments welcome.
.