Re: Attempt at a Schema

["J. Barkley Rosser, Jr." <rosserjb@xxxxxxx>]

Goncalo,
     Wow!  Is this your schema, or have you borrowed
it from someone else?
    Some picky points.  I think your characterization of
Arrow and Debreu is simply way off base.  I also don't
see any of the versions of Keynes being lumped in with
Lucas.  How do you get that one?  And, I think that one
should be careful about using the term "intuitionism," given
its special use in mathematical logic to refer to systems
that allow the existence of the "excluded middle."  Finally,
I am surprised that you did not bring up Bayes in the discussion
of subjectivism.  Clearly, he is the key figure.
      More generally, I think maybe you have overcomplicated
the discussion, interesting as your cases are.  A somewhat
simpler, and I think quite useful, categorization is due to
Tony Lawson in a 1988 piece in the JPKE, "Probability and
uncertainty in economic analysis," JPKE, 1988, vol. 11, no. 1,
pp. 38-65.  Especially on p. 48 he presents a neat little
diagram that provides four cases based on two distinctions.
The distinctions are subjective versus objective and
numerically measurable versus not numerically measurable.
He identifies Savage and Friedman as in the subjective but
numerable category and Keynes as in the subjective but
not numerable category (although I have already noted that
Keynes allowed for other cases as well as this one).  In
the objective numerable he puts Muth and Lucas and in the
objective non numerable he puts Knight.
Barkley Rosser
-----Original Message-----
From: Goncalo Fonseca <fonseca@xxxxxxxxxxxxxxxxxx>
To: Post Keynesian Thought <pkt@xxxxxxxxxxxxxxxx>
Date: Monday, December 18, 2000 11:54 AM
Subject: Attempt at a Schema


>Mat asked:
>
>>
>> What I am really trying to get at is a distinction between conceptions of
>> uncertainty based on something like "human nature" versus conceptions of
>> uncertainty based on the structure of our social reality.
>>
>
>Mat,
>
>Perhaps you could express yourself a little better if you had a menu to
>choose from.   So here is an attempt at a very basic breakdown of the
>different
>views.  I divide it into three general categories: "Ontological" theories,
>"Informational" theories and "Subjective" theories.  (I suppose
>"epistemological" relates to the last two types, but I believe the further
>distinction is useful).
>
>I welcome everybody to correct, add or deduct from this schema.
>
>___________________________
>
>I - ONTOLOGICAL
>
>Basic premise: randomness is really in the object/action of interest (e.g.
>tossing a die, flipping a coin).  von Neumann-Morgenstern expected utility
>theory (and most other non-linear, non-Archimedean, alternative expected
>utility theories) are dependent on assumption on these.
>
>(1) Classical:
>
>Probabilities are derived by applying the Principle of Cogent Reason (i.e
>physical symmetry implies equal probability) and the Principle of
>Indifference/Insufficient Reason (i.e. assign equal probabilities if you
>have no (physical) reason to think one outcome is more likely).
>
>Who: Laplace, Bernoulli.
>
>Pros: it is a nice clean definition, limits the discussion of probability
>assignments to random situations where these "physical" features are clear.
>Not overambitious.
>
>Cons: Not ambitious enough.  Can't say anything about "probabilities" of
>things when the alternatives are not clear, "adding up" problems emerge,
>etc.
>
>(2) Relative Frequentist:
>
>The probability of a particular event in a particular trial is the relative
>frequency of occurrence of that event in an infinite sequence of "similar"
>trials.
>
>Who: Richard von Mises, Reichenbach
>
>Pros: gives a very "empirical" definition of probability.  No metaphysical
>crap about probabilities being "out there" somewhere.
>
>Cons: actually can't ever be tested -- so still a bit metaphysical. Not
>very helpful either, e.g. how do you ascertain the relevant probabilities
in
>a "unique" situation?  Doesn't clarify what "similar" trials means.
>
>(3) Propensity
>
>(I never got this one down, but it goes something like this): probability
>represents the disposition or tendency of Nature to yield a particular
event
>on a single trial.  Propensties are assumed to objectively exist, even if
>only in a
>metaphysical realm, but are NOT identified with Classical probabilities and
>are NOT necessarily associated with long-run frequency..
>
>Who: Peirce, Popper, Suppes
>
>Pro: keeps "objective" probabilites and expands their applicability to
>practically any situation.  Not physically limited like (1), not
>frequency-dependent like (2).
>
>Cons: Huh?  Way too much metaphysics!  What's a "propensity" and how do you
>assess/measure this animal?
>
>II - INFORMATIONAL
>
>Basic premise: there is no randomness inherent in Nature. If knowledge of
>all the conditions relevant to a particular "random" situation were known,
>then there would be certainty (cf. Knight, 1921: p.219).  In the absence of
>all relevant information, probabilities are assigned as a measure of the
>lack of
>knowledge.
>
>(1) Logical Relationist
>
>There is an "objective" (albeit not necessarily always measurable) relation
>between knowledge and the probabilities that are deduced from it. Note:
>knowledge here is treated as disembodied and not personal.  Whoever has
this
>information, necessarily makes this probability assignment.
>
>Who: Keynes (most of the time), Harsanyi, Lucas
>
>Pros: tight "objective" relation between knowledge and probability makes
>informational theory precise.  Probabilities adjusted as information
becomes
>available.  If everyone has the same information, then there is also
>commonality of agreement among people as to what the probability
assignments
>are ("rational expectations").  Enables one to distinguish between "risk"
>(enough info to make probability assignments) and "uncertainty" (relevant
>info to make probability assignments not available)..
>
>Cons:  What does "disembodied" knowledge mean, in fact? Where does
>"relevant" knowledge begin and where does it end?  Also, it relies on
>"common priors" to pull itself through.
>
>(2) Heterogeneous Relationism
>
>There is a correspondence (not a function) between information and
>probabilities, so people can derive different probabilities from the SAME
>information. Differences in expectations can be maintained for a
>long time as long as data is consistent with beliefs. ("rational beliefs").
>
>Who: Keynes (sometimes), Kurz
>
>Pros: makes a hell of a lot of  sense.  Personal beliefs should affect how
>information is processed.  Even common priors and common information don't
>necessarily imply same expectations.
>
>Cons: probability assignments still move too deterministically with
>"disembodied" knowledge, i.e. still no room for "whim".
>
>III - SUBJECTIVE
>
>Basic premise: again, the world is not inherently random. However,
>probability assignements are not derived from "disembodied" knowledge, but
>from "embodied" knowledge, i.e. probability assignments are derived from
>personal beliefs about outcomes.
>
>(1) Revealed Belief
>
>Probabilities of events are subjectively assigned.   Can numerically deduce
>these assignments by observing a person's behavior and choices in random
>situations.
>
>Who: Ramsey, de Finetti, Savage
>
>Pro: can derive "probabilities" numerically as mathematical expressions of
>belief.  Personal belief (not disembodied knowledge) is always the guide to
>our actions.
>
>Cons: relies to heavily on the assumption that ONLY beliefs guides one's
>choices.  Too "behaviorist" in its implications.   Choices may reveal
>"probability" assignments that the person had no idea he made.  Relies on
>state-independent utility.
>
>(2) Intuitionist
>
>Way-out-subjective; probabilities are assigned by beliefs, but not
reducible
>to mathematical expression because other things (e.g. prejudices,
>state-preference, wishful thinking, whim, etc.) affect people's choices.
>People
>form probabilities on "intuitive" assessments about a random situation.
So,
>cannot "deduce" subjective probabilities cleanly from behavior, but must
>live with multiple numerical measures (if any are possible).
>
>Who: Good, B.O. Koopman
>
>Pros: Anti-behaviorist and romantic.
>
>Cons: Too romantic, too loose.   Doesn't even sound like probability
>anymore.
>
>(3) State Preference
>
>An extreme form of intuitionist theory which does not even bother to try to
>make
>probabilities derivable.  Preferences are formed over "states of the
world",
>only one of which emerges in any time period.  Beliefs, expectations,
>prejudices, wishful thinking, etc. are all subsumed in "state-preference".
>
>Who: Arrow, Debreu
>
>Pros: No probabilities! Beliefs/expectations do not need to be made
precise.
>
>Cons: Not everything is reducible to a finite set of "states of the world".
>People might not have knowledge of all possible states of the world.  Also,
>a "state of the world" has to be defined in a very expansive list of
>things -- do people really have preferences so well defined?
>
>_______________________
>
>And that's all I can think of off the bat.  So, Mat, which of these is
>closest to the kind of thing you're trying to get at?
>
>Goncalo
>
>
>
>