[From Bill Williams 21 January 2004 3:25 AM CST]
My previous CSGnet posting "He screwed the Pooch" slipped
out of my hands before I intended to post it. As a result
being asked to improve the formating of my email-- complaints
for which I am grateful. As a result a student found me a
much better email program, than the one I'd been previously
using.
But, on to substansive issues. Those of you who were at Boston
may have recall my attempting to make the point that econmics
is a serious field of inquiry, or at least issues in economic
theory ought to be taken seriously. I also attempted to make a
point about it being a good idea in economic theory to observe
some elementary arithemetic rules-- such as treating equal signs
as if they were important. That is if you use an equals sign,
then treat the quantities on the two sides of the sign as being
equal to each other. Rick's H. Economicus I argued didn't do
this-- that is treat equal signs as they are meant to be treated
in arthemetic operations. Rick, didn't then, and hasn't since
then, regard his violation of the rules usually used in adding
and subtraction etc, as a substansive comment.
When I see an error-- such as a violation of arithmetic-- in a
fundamental feature of a paper, I quickly loose interest in the
paper. Unless that is there is a possiblity of publishing a
correction in a journal. Since this wasn't a possiblity in regard
to Rick's H. Economicus scheme, I lost all interest. Rick, however,
pestered Bill Powers until Powers did what seemed to me to be a
through job of criticism of H. Economicus. Immeadiately after
encountering Power's criticism, Rick decided he "wasn't paricularly
good at economic modeling." This is a judgement to which I can
enthusiastically agree. Recently, however, Rick seems to have
forgotten this earlier and perceptive conclusion, and returned to
thinking his economic modeling skills are quite good. I can't agree.
Since Rick has taken it upon himself to make judgements about what
is and is not PCT, and HPCT correct, the claims that Rick makes
regarding his efforts to do economic modeling should be of concern
to all those interested in the future of efforts to applying
control theory in efforts to improve the quality of human experience.
Rick has ignored criticisms-- mine, and Bill Powers of the foundations
of economic modeling. Worse, Rick has after efforts have been made to
show him what is wrong, then complained of a lack of substansive
critical and constructive comment.
As, I said in Boston Economics is a serious business. It is also a
difficult business. I've made reference to the criticism that Powers
made of Rick's paper several times recently. Rick has misrepresented
what Powers had to say. Documents never "speak for themselves" but
I will leave it to CSGnet readers, the moment, to "judge for yourselves."
Bill Williams
[From Bill Powers (2003.02.07.1037 MST)]
Rick Marken (2003.02.06.2200)--
>OK. I imagine the diagram could be better.
Then make it better. It's not much help as it stands. It is hiding errors.
>I guess one thing I was trying to do was model at the aggregate level,
>the same level at which TCP was describing the behavior of the circular
>flow. The variables whose behavior I am trying to model are aggregate
>variables, like GNP. So the agents are really conceptual agents.
That's not an answer to my objection. If you want to see what the model
predicts about the growth of GNP, you don't make GNP a controlled variable
and give it an exponentially-rising reference level. If you do that, you'll
get out of the model precisely what you put into it, an
exponentially-rising GNP. That is exactly what TPC did: declared that
economic output depends on exponents, and then gave the impression that he
had proven that it rises exponentially. In other words, the way you're
going to prove that GNP rises exponentially is to make sure it rises
exponentially. What kind of modeling is that? It would be like proving that
the outfielder runs along a curved path by giving him reference levels for
direction of running in X and Y that correspond to the curved path you
observe. In your baseball model, you at least show how the directions of
running are functions of error signals in some _other_ control systems than
the direction-of-running systems. The fact that they are curved _emerges_
from the operation of the model; there's nothing in the model that
specifies the curvature directly, as your rising reference level for GNP
specifies how GNP is to behave.
What bugs me is that you know better than this. You have never, to my
knowledge, made this mistake before.
>Individuals compare income (PQ' at the individual level) to expenses
>(P'Q' at the individual level). My PQ' - P'Q' controller is a
>representation of the aggregate control over their balance sheets
>exerted by many individuals.
So are they trying to make PQ' = P'Q'? That denies leakage right there; if
there is actually leakage of money from the system, it is impossible for
PQ' to equal P'Q', if I interpret those symbols correctly. Your model
contains no source of money to replace the leakage disturbance. so the
books clearly cannot balance. You claim in your paper that the manager
makes PQ' = P'Q' by changing the price, but changing the price does not
supply the needed money. Your system will run out of money -- remember that
leakage happens continuously, not just once. A continuous inflow of money
is required to compensate for it. If you have set up the equations so this
is not necessary, your model is incorrect.
>
>The evidence I gave that the model is working properly (given what I
>wanted the model to do) is presented in Table 1, p. 169. This table
>shows that the model does what it was designed to do: produce behavior
>(of Q/Q' and inflation, at least) that matches the behavior of these
>variables, according to TCP's model, as a function of leakage.
But it does that only because you specifically made it do that. And there
has to be an error in your model, because Table 1 shows the rate of growth
increasing at a constant 13% a year, while in fact the amount of money
available for spending is decreasing at the assumed leakage rate. There is
no way that the rate of spending can be increasing according to the
indicated inflation rate while the rate of growth is remaining constant in
terms of goods purchased. The amount of goods may be increasing year by
year, but inflation is increasing faster, so there is a net loss of output
in constant dollars -- exactly as the Circular Flow model predicts.
This is what I mean by not making sure your model works correctly before
drawing conclusions from it.
> I'm sure I could have done a better job of
>validating the model. But the model did do what I built it to do.
So why didn't you just make GNP increase at 13% per year, so you could say
the model is correct, since GNP increases at 13% per year? Essentially
that's what you did by making the reference level increase 13% per year.
> I
>agree that the discussion might have been too sketchy. But I don't know
>what detailed evidence (other than what I presented in Table 1) you
>would want to prove to yourself that this is a "proper" model.
Look at Econ004.pas. It's not complete and it may even be mistaken, but it
is a proper model. No behavior is simply assumed to happen and then made to
happen that way. The books actually balance, both for goods and for money.
If you put leakage into it, the total amount of money in the system will
decline and if no new money is supplied, the system will go broke. In your
model, all the money can be removed and the system will go right on working.
>I don't think your conclusion that "of course leakage had no effect" is
>correct. The increasing reference for GNP corresponds (somewhat) to
>TCP's intrinsic growth rate; I think it represents the growth in demand
>for GNP that results simply from population growth.
None of that is in your model, and your model can't really work for that
reason. Leakage has no effect in your model because you have not provided
any way for it to have an effect. Leakage can't step in and rewrite your
equations to give itself an effect.
> I constructed the
>model to mimic the behavior of TCP's model. And my model succeeds at
>doing that.
Except for the effect of leakage. Imitating the behavior is trivial. I can
construct a model for stopping a car by saying its velocity decreases
exponentially to zero. That duplicates the behavior of the real car. But
the whole point is how that result is achieved; the essence of the model is
in the how, not the what. Your model contains no how, or no valid how.
> I thought that the effect of leakage on growth rate would
>"fall out" of the model. But it didn't, and that's how I discovered that
>TCP had put the effect of leakage on growth rate in as an assumption.
But you did the same thing: you assumed a constant growth rate by making
the reference level grow at a constant rate. Of course there was no way
leakage could have an effect on growth rate when you specifically set up
the model so _nothing_ could materially affect growth rate.
> In
>fact, _change_ in leakage does affect growth rate.
How could it? You show the growth rate in Fig. 1 as constant at 13% per
year. If you changed the leakage, it would still be 13% per year. If you
think that a change in leakage affects growth rate, you've made another
mistake. It might have a transient effect, but there can be no other kind
of effect. And it is not a transient effect we're talking about.
>I really have very little experience with Delphi. And I really don't
>have any time to work on that model; I'm juggling 3 projects at work and
>3 other PCT related projects at home right now. And I don't think I'm
>particularly good at economic modeling, anyway. So I think I'll just let
>you guys proceed.
That's beginning to sound like the wisest course. But if your home projects
with modeling are to fare any better than the economic model, you should
consider very carefully what we're discussing here. Your model works by
magic, and that is not how you have normally made models in the past. If
you don't figure out exactly where you went wrong, you could make the same
mistake again.
Best,
Bill P.