Maximization (and Giffen effect)

[Martin Taylor 2007.11.11.23.01]

[From Bill Powers (2007.11.08.0656 MDT)]

Rick Marken (2007.11.07.2230) --

I think it would be appropriate to acknowledge the late Bill Williams' priority in analyzing the Giffen effect as a control process. In fact when he first approached me about PCT, it was to seek my help in constructing a working computer model of the Giffen Paradox, which we renamed the Giffen Effect once the model had explained it.

Following up this comment, there was a good discussion relating to a control theory of value and the Giffen effect on the ECACS forum in 2003-4 (there were quite a few other good discussions of PCT there, too). Here's what Bill Williams said in one message <http://ecacs.net/discus/messages/192/243.html>:

---------------Quote from Bill Williams------------
Martin,

I am perhaps beginning in myself to see what has happened in the economics profession. After thinking about a problem that goes unsolved for a sufficient period of time a reluctance begins to develop to put any more effort into solving the problem. I am I think only part way to this sort of orrientation, but I think I can perceive in myself the sort of "Oh no! Not that question again!" sort of feeling.

Still your comments have provoked me to give some thought to the price-index issue once again. And, perhaps my thinking is has some useful content even if it doesn't quite solve the problem.

Suppose we take a person who has adjusted their expenditure to a static situation in which they distribute their income between the consumption of commodities at various prices. Suppose we use a control theory analysis to do this employing the technique I developed for a two commodity case. Suppose in this static situation we varied the consumer's income at some level of income the consumer could purchase a quantity of all the commodites that would satisfy the reference level for each commodity. At this point adding any more income to the consumer's budget would not increase their purchases. (I am assuming that the consumer is not involved in a contest of competitive display with a neighbor of the sort that Veblen considered in his notion of conspicious consumption.) Then we could experiment by increasing the prices of various goods in random combinations. Most of these random combinations of price changes would result in price quantity behaviors of the normal downward sloping demand curve type. However eventually it would be expected that Giffen good type behavior would be found. Finding price points at which the Giffen good (People should call it a Giffen commoddity but no on does.) becomes active because it identitifies a point at which a possible positive feedback effect might be generated. That is, when the price of the commodity-- the Giffen good is increased the consumer buys more of the product and this might result in a further price increase of the product-- a positive feedback loop.

So now we have a consumer for whom we have identified a combination of income and prices for which the consumer is experiencing no error at one end of the scale and various points were the consumer is experiencing a Giffen type effect as a result of a price increase. Now using a sufficiently sophisticated technique we could infer a control system that would simulate the consumer's behavior. At the point at which the consumer's budget allowed them to purchase a quantity of comodities that matched their reference level running the control system would generate no error. At any lower level budget errors would begin to be experienced, and these errors could be computed.

Supposing all this, which makes my head spin, we would have something like or close to what we pretend that we have now with the headonistic prices indexes. Plus, we would have identified points at which the consumer's behavior would in combination with some circumstances regarding supply generate unstable behavior in the market.

Could in princple something like this be done? Well, I could see that there would be a lot, a great lot of difficulties involved in actually carrying out such an experiment, inferring a control system and then calculating the consumer's error. Or, would one want some measure such as the consumer's error times the gain of the system? I'm not sure. Error times the gain might be a better measure of the consumer's conflict as a result of not having a sufficient budget to purchase the reference level of all the commodities.

So, taking the standpoint of a theorist with any resonsiblity for actually carrying out such a system, this is how one might per impossible construct a control theory price index.

If we removed the restriction against the interaction between consumer's errors as a result of conspicious consumption we might see that what ever the level of prices (however determined) and the consumers budget the error index for the consumer wouldn't change-- because it was being determined as a result of the competition for status by way of conspicious consumption. From an economists standpoint ( an economist adopting a control theory perspective ) if the main source of error for the consumer was the competitive display of conspicious consumption, then it might be reasonable to reduce the consumer's budget. The consumer would then consume less be feel no worse off because everyone else was consuming less.

In terms of policy this control theory type index is more interesting from a policy standpoint than is the hedonistic index. All the hedonistic index tells one is that consumers are better off if they have more income. And, I suppose if it actually worked it would be a way of judging the effects of various policies-- but it doesn't really do this. A control theory price index would tell policy maker about where the giffen points were located ( good to know if one is interested in a stable market ) and also indicate when conspicious consuption was dominating the situation- which would be good to know if one was setting taxes on goods that did not contribute to the genuine well being of a consumer.

As a policy measure we might want to institute what I might call economic integration. Rather than allowing the wealthy to live in gated communities in which the forces of conspicious consumption run without limit, we might compell the wealthy to live next to the impoverished. I can even forsee a slogan--"Unequall neighborhoods are inheriently discriminatory" -- which (no question) they are.

Somehow I don't see the National Science Foundation being eager, at least under the current administration, to fund, supposing it was possible, the development of a control theory counter part to orthodoxy's method of computing a price level.

Martin, you talk about computing a price index in terms of statistical mechanics. However, when I think about the implication of a control theory index what comes to my mind is political dynamics.

Bill Williams

-----------------end of Bill Williams message-------

Incidentally, this quite serious conversation, along with others, was happening at a time when Bill was playing silly-buggers on CSGnet.

If you want to see more about Bill's last thoughts relating to the Giffen effect and other economic notions, go to <http://www.ecacs.net> and do a search for "Giffen effect". It's mentioned in 8 different threads, though some of them are branches off earlier ones. (For the newer CSGnet members, ECACS was designed as a forum rather than a mailing list for discussion PCT in a way that enabled the discussions to be kept active, so that a response six months or several years later could still be relevant.

I'm not sure whether you have to be registered on ECACS to do a search. You do, if you want to post, although nobody has for the last few years (which I think is a pity, as many of the CSGnet discussions would benefit from an ability to revisit them when someone has a new idea (or is a new participant and sees an old issue from a new direction).

Martin

[From Bill Powers (2007.11.12.0004 MDT)]

Martin Taylor 2007.11.11.23.01 –

Welcome again, Martin.

It’s good that Bill Williams had a place where he could develop his ideas
more freely without running into my criticisms. Somehow when he and I
discussed these things, he was unable to distance himself enough from the
subject matter (or me) to tolerate even suggestions as to how his
approach to modeling might be made to work a little better. He actually
had a good intuitive feel for complex systems – I see in the discussion
you posted some strong suggestions that he was seeing the same
multidimensional Giffen Effect that I posted about a day or so ago. But
as this post shows, it was hard for him to make the transition to an
orderly and workable model – that’s why he came to me in the first
place, for help in turning his basic understanding of the Giffen Effect
into a runnable computer model that would demonstrate the effect. But
that created its own difficulties, because I found out how to do it
rather quickly using PCT, in one weekend with him – but I was not an
accredited economist.

I’m convinced that the “silly buggers” stuff (as you call it)
he introduced to CSGnet was a consequence of his untreated diabetes,
which ended up killing him. We were friends for over 25 years, and it is
very sad that it ended as it did. Mary and I both liked him a lot, and
when he visited us in Durango there was no trouble at all that I could
see. A year or two later things started to go downhill.

Thanks for the nice diagram of the relationship of the reorganizing
system to the hierarchy. It shows what I intended very clearly. Dag
Forssell also provided a clearer picture than my initial diagram; it
appears in the paperback edition of B:CP on page 191 in place of my
original diagram.

I halfway expected you to show up at the Manchester meeting; your former
colleage Phillip Ferrel (Farrell?) did, and we got along very well. You
might have enjoyed it, I think.

As to the Kalman Filters, I think something similar really does happen at
some level in human brains, but the Modern Control Theory people want to
substitute it for the negative feedback model. I don’t think they
seriously propose it as a model of how living systems work, but even as a
method of designing artificial systems it doesn’t excite me much – very
complex, slow, and computation-intensive, and really more like
pre-feedback-theory approaches than the elegance of negative feedback
hierarchies. But who knows? Perhaps there is something there we need to
consider. As you say, it does look like a form of planning, but does it
really propose anything that’s not already implicit in the PCT model? I’m
probably too prejudiced to say.

Best,

Bill P.

[Martin Taylor 2007.11.12.10.53]

[From Bill Powers (2007.11.12.0004 MDT)]

Martin Taylor 2007.11.11.23.01 --

Welcome again, Martin.

Thanks. I was at a meeting and workshop in LA for a week.

It's good that Bill Williams had a place where he could develop his ideas more freely without running into my criticisms. ...I see in the discussion you posted some strong suggestions that he was seeing the same multidimensional Giffen Effect that I posted about a day or so ago. But as this post shows, it was hard for him to make the transition to an orderly and workable model -- that's why he came to me in the first place, for help in turning his basic understanding of the Giffen Effect into a runnable computer model that would demonstrate the effect. But that created its own difficulties, because I found out how to do it rather quickly using PCT, in one weekend with him -- but I was not an accredited economist.

I'm convinced that the "silly buggers" stuff (as you call it) he introduced to CSGnet was a consequence of his untreated diabetes, which ended up killing him.

I can't speak to that, but he told me that he enjoyed applying "The Test" by introducing disturbances in his CSGnet postings that might produce predictable following messages from other participants. I suggested he might also enjoy the messages that would follow serius postings, but with no effect.

Thanks for the nice diagram of the relationship of the reorganizing system to the hierarchy. It shows what I intended very clearly.

I'm glad to hear that. It makes so much sense to me -- and in light of the "12th level" thread as a whole, it suggests how a hierarchic system can build from the ground up.

I halfway expected you to show up at the Manchester meeting; your former colleage Phillip Ferrel (Farrell?) did, and we got along very well. You might have enjoyed it, I think.

Farrell. It's not surprising he understood you quickly, since he was a co-author on the Layered Protocol papers in the PCT special issue of IJHCS. He is a Ph.D. control engineer.

As to the Kalman Filters, I think something similar really does happen at some level in human brains, but the Modern Control Theory people want to substitute it for the negative feedback model.

If that is so, it would seem to be so obviously inefficient that one would wonder how the idea could survive. But is it so? You mentioned in your long message the possibility that an imagination-mode model might enable continued effective action if there were gaps in the real-world data stream. In the past we have talked about perceptual "filling in" based on experience, the "Artificial Cerebellum" on the output side, and an "imagination short circuit" connection. All of these could be considered as components of a model. The only real extension I see is in the complexity of the "imagination short circuit", and even that may not really be an extension.

As you say, it does look like a form of planning, but does it really propose anything that's not already implicit in the PCT model? I'm probably too prejudiced to say.

To me, "the PCT model" is a phrase with many possible interpretations. In the past, I've used terms like "basic PCT", "classic PCT", "accepted PCT", "speculative PCT", etc. There's the PCT that seems to be absolutely required as the fundamental basis (and to me, definition) of life. That "basic PCT" is defined by the laws of non-equilibrium thermodynamics. It is the only version of PCT that I think could not be contradicted by future discoveries.

Beyond basic PCT, there are PCT models based on observation, analysis, or plausible speculation. These models gain acceptance for various reasons, and one strong reason is that Bill Powers tends to think they might be correct (and that Bill Powers has proposed them, which usually implies he thinks they might be correct). Why this reason works is that Bill Powers often sees clearly just what is implied by basic PCT. But to accept that as sufficient reason is to appeal to authority, not science.

So we ask "What is really implicit in 'the PCT model'?" If "the PCT model" is "classic PCT", meaning the strict HPCT hierarchy, with imagination loops, then I would say that the Kalman Filter as planning model is implicit. It's not been made explicit. If "the PCT model" is basic PCT, analysis and simulation might discover whether it is implicit (as well as whether the hierarchy itself is implicit).

Which is a long-winded way of saying that I do agree that the Kalman filter as a way of planning and of compensating for data gaps is probably implicit in HPCT if the imagination loops are accepted as a part of the hierarchy.

Martin

[From Rick Marken (2007.11.12.1130)]

Martin Taylor (2007.11.12.10.53) --

I can't speak to that, but he told me that he enjoyed applying "The
Test" by introducing disturbances in his CSGnet postings that might
produce predictable following messages from other participants. I
suggested he might also enjoy the messages that would follow serius
postings, but with no effect.

What is un-serious about posting in order to test for controlled
variables. It seems like it was his attitude toward the variables
people controlled that were un-serious. And he had to know what
variables people were controlling for if he know what the response to
his disturbances would be. I think, for example, that Bill knew that I
was controlling for a control theory view of economics and when my
replies to his questions on CSG confirmed that fact he found it
amusing because he thought that approach to economics was so
ridiculously wrong, something he made clear after my talk at the 2000
meeting in Boston.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2007.11.12.15.00]

[From Rick Marken (2007.11.12.1130)]

Martin Taylor (2007.11.12.10.53) --

I can't speak to that, but he told me that he enjoyed applying "The
Test" by introducing disturbances in his CSGnet postings that might
produce predictable following messages from other participants. I
suggested he might also enjoy the messages that would follow serius
postings, but with no effect.

What is un-serious about posting in order to test for controlled
variables. It seems like it was his attitude toward the variables
people controlled that were un-serious. And he had to know what
variables people were controlling for if he know what the response to
his disturbances would be. I think, for example, that Bill knew that I
was controlling for a control theory view of economics and when my
replies to his questions on CSG confirmed that fact he found it
amusing because he thought that approach to economics was so
ridiculously wrong, something he made clear after my talk at the 2000
meeting in Boston.

Bill W isn't here to answer for himself, so all I can do is to suggest you look at the (sometimes voluminous) discussions on the ECACS forum. Unfortunately, ECACS had the same problem wth Marc Abrams that CSGnet had -- starting a discussion sensibly and reasonably and slowly but exponentially diverging into angry ranting. He eventually got banned, but his messages are still on the Forum.

Without knowing how you and he interacted in 2000, may I suggest that there might be more than one way of looking at economics through a PCT lens, and that just possibly some knowledge of conventional economic theory might be useful, even when you don't believe it (as Bill apparently did not)?

Martin

[From Rick Marken (2007.11.12.1250)]

Martin Taylor (2007.11.12.15.00) --

Bill W isn't here to answer for himself, so all I can do is to
suggest you look at the (sometimes voluminous) discussions on the
ECACS forum.

I've read them. The stuff on economics just didn't ring my bell. I
thought Bill's dad's book was the best economics book I've ever read,
even with the many errors in it. The basic organizing idea made a
whole lot more sense to me than anything I've read in economics texts.

Without knowing how you and he interacted in 2000

After my talk he went on about how non-economists (like me) shouldn't
venture any ideas about economics. It was pretty weird, but we got
along fine in face to face meetings. I guess he was able to hold his
hatred for me down when we met in person.

may I suggest that
there might be more than one way of looking at economics through a
PCT lens

Sure. But the only other PCT view of economics I've seen is Bill
Powers'.I like his stuff, too.

and that just possibly some knowledge of conventional
economic theory might be useful, even when you don't believe it (as
Bill apparently did not)?

I think knowledge of some conventional economic data can be useful but
conventional economic theory -- to the extent that it does not see
people as input controllers -- can only be confusing.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2007.11.12.17.33]

[From Rick Marken (2007.11.12.1250)]

> Martin Taylor (2007.11.12.15.00) --

> and that just possibly some knowledge of conventional

economic theory might be useful, even when you don't believe it (as
Bill apparently did not)?

I think knowledge of some conventional economic data can be useful but
conventional economic theory -- to the extent that it does not see
people as input controllers -- can only be confusing.

Confusing doesn't necessarily mean wrong, though. I think "wrong" is th word you want. If so, I might well agree.

The same issue arose at the LA workshop, where the theme was the visualisation of networks. Underlying network visualisation is a whole area called "social network Analysis", part of which considers what happens when something is done to alter the structure of a network. This is treated as a mathematical question: take out a key node, and you change the possibilities for message passing from one part of the network to another. That's fine, perhaps, if the nodes are inanimate, such as computer network servers, but it doesn't wash if the nodes are people with references to get messages from one part of the network to another. They will act to reclose the gap. I pointed this out to one or two people (I called it "network homeostasis" to sound academic), which caused a kind of hesitant "aha" reaction. It does matter that humans try to achieve goals!

Martin

As to the Kalman Filters, I
think something similar really does happen at some level in human brains,
but the Modern Control Theory people want to substitute it for the
negative feedback model.

If that is so, it would seem to be so obviously inefficient that one
would wonder how the idea could survive. But is it
so?
[From Bill Powers (2007.11.12.1545 MDT)]

Martin Taylor 2007.11.12.10.53 –

If you look up some threads on MCT on the web you’ll see that the old
“PID Control” model (i.e., negative feedback using
Proportional, Integral, and Derivative terms in the output function) is
rather sneered at as old-fashioned. Of course that’s a very limited
design, too, but it’s used as a code-word for old-fashioned control
theory in general, meaning pure negative feedback control. You may recall
that Oded Maler was arguing that model-based control based on sensing
outside temperatures and computing heat losses was the best design for a
home thermostat. He was convinced that model-based computations could
control physical devices instantaneously (it really seemed that he
thought that). I think that was Oded – could it have been someone else?
It seems that there are a lot of people in the Netherlands who take this
stance, people with various ties to cybernetics and Ashby’s Principles of
Cybernetics which is very MCT. I don’t feel like going back and looking
up all that stuff again – it was somewhat depressing.

You mentioned
in your long message the possibility that an imagination-mode model might
enable continued effective action if there were gaps in the real-world
data stream.

That was what Oded was arguing. One has to admit that this is a possible
mode of control that can tolerate loss of input, and it surely exists,
but it’s incapable of handling any unexpected changes in the environment.
As a primary means of control it would be ineffective
(disastrous).

In the past
we have talked about perceptual “filling in” based on
experience, the “Artificial Cerebellum” on the output side, and
an “imagination short circuit” connection. All of these could
be considered as components of a model. The only real extension I see is
in the complexity of the “imagination short circuit”, and even
that may not really be an extension.

I think we have to accept that people do use model-based control. But it
works only under very limited conditions and can never achieve the
precision of negative feedback control which can prevent all the little
bumps and jogs and drifts that normally joggle the controlled variables
around unless opposed. Those little disturbances are cumulative through
time and are too numerous and trivial to be predicted – and have too
much cumulative effect to be ignored. Sometimes you have to use
model-based control, but you shouldn’t expect a very high quality or
bandwidth of control from it.

To me, “the
PCT model” is a phrase with many possible interpretations. In the
past, I’ve used terms like “basic PCT”, “classic
PCT”, “accepted PCT”, “speculative PCT”, etc.
There’s the PCT that seems to be absolutely required as the fundamental
basis (and to me, definition) of life. That “basic PCT” is
defined by the laws of non-equilibrium thermodynamics. It is the only
version of PCT that I think could not be contradicted by future
discoveries.

Well, your basic PCT is too abstract for me, but I don’t begrudge it to
you. I agree that much of “the PCT model” is indeterminate; it
will only slowly become less so.

Beyond basic PCT, there are PCT
models based on observation, analysis, or plausible speculation. These
models gain acceptance for various reasons, and one strong reason is that
Bill Powers tends to think they might be correct (and that Bill Powers
has proposed them, which usually implies he thinks they might be
correct). Why this reason works is that Bill Powers often sees clearly
just what is implied by basic PCT. But to accept that as sufficient
reason is to appeal to authority, not science.

Hear, hear. It’s a bit awkward to be the originator of a system of
thought, because people tend to give too much weight to my ideas, so I
really have to bear down on the disclaimers and constraints and modifiers
to offer an ordinary idea for ordinary consideration. I don’t always
remember to do that, and sometimes resent the fact that others don’t even
seem to bother with it, while they expect it of me all the time. Oh, the
burdens of being an incomparable genius.

Which is a
long-winded way of saying that I do agree that the Kalman filter as a way
of planning and of compensating for data gaps is probably implicit in
HPCT if the imagination loops are accepted as a part of the
hierarchy.

Something like that, I agree. But the Kalman Filter concept simply
dismisses control of random disturbances, its proponents evidently not
realizing that a negative feedback system can counteract the effects of
random distubances up to the bandwidth of the controller. In MCT a random
disturbance with a zero mean is simply dropped from the equations (except
for computing the uncertainty of an observation). That’s an admission
that MCT just doesn’t work very well when you don’t know everything about
the environment.

Best,

Bill P.

[Martin Taylor 2007.11.12.22.52]

[From Bill Powers (2007.11.12.1545 MDT)]

Martin Taylor 2007.11.12.10.53 --

As to the Kalman Filters, I think something similar really does happen at some level in human brains, but the Modern Control Theory people want to substitute it for the negative feedback model.

If that is so, it would seem to be so obviously inefficient that one would wonder how the idea could survive. But is it so?

If you look up some threads on MCT on the web you'll see that the old "PID Control" model (i.e., negative feedback using Proportional, Integral, and Derivative terms in the output function) is rather sneered at as old-fashioned.

OK. I'm not going to follow this up, and I'll take your word for it.

As for the rest of your long posting, all I will say is that I agree pretty well with all of it. And thanks for writing it.

Martin