Of Blind Men, Elephants and the Economy/Talk Show I

[From Bill Powers (2011.05.28.1207 MDT)]

Adam Matic 2011.05.28 0050 gmt+1 --

AM: Perhaps only more free time could be a common goal that everyone wants
to have more than before. Not that every single human really wants
more free time, but that might be a useful simplification. If there is
a fixed number of needs an agent has, and in the beginning he can't
even correct all the errors, then "more free time" might be an
ultimate goal.
It could be achieved by reorganizing the amounts off goods to be
produced, bought or sold.
So, in the beginning, an agent produces what he needs, but when he
consumes it, there is not enough to fulfill the need. There is also an
error in "free time". Reorganization begins and some goods are
produced in excess and offered in the market, some are bought from the
market

COMMENT: BP: The moderator was trying to question the assumption of some special characteristic of all people to be incorporated into a model of the economy. Instead of assuming that everyone always wants more of anything he has or is getting, a better approach is to define a variable or a variable parameter of behavior, and construct the model so that if the variable doesn't have the correct value, the model will behave differently from the real population. If no value will work, the model is wrong. That way your model qualifies as a falsifiable hypothesis.

This is the reason for the "How many people ..." kind of question -- it was intended as a reminder that people are not all alike in their perceptions or their preferences. Some people may be "rational actors," while not everyone is. So if you want to use that premise, first you have to admit the possibility that it may be incorrect, and then you have to set up the model to catch the mistake if the premise turns out to be false. You have to see what difference it makes to the model's behavior if everyone is a rational actor, most people are, most people are not, or nobody is. If the model's behavior depends entirely on the first possibility, and the possibility is false, then you will find yourself making up good reasons why it's still true. There's a lot of that sort of model-rescuing excuse-making going around.

People do not generate behavior; they control. Their actions -- their behaviors -- are a result of both internal and external factors: reference signals and perceptual functions on the inside, and disturbances and feedback functions on the outside. What is constant or repeatable is not behavior, but the controlled results of behavior. Even the desired result can change when a system at a higher level experiences a disturbance. Any model that assumes a fixed behavior pattern or fixed preferences, therefore, is likely to predict poorly, and more poorly still if there is no provision for changing the model's parameters to fit the real control processes when they change.

Final word of, one hopes, wisdom: when we make models we don't have to guess how they will behave, because we can run the model and see. I think it's best simply to make the components of the model itself as realistic as possible and not assume that one can guess what will happen when they are assembled and set into motion. We don't make models to prove a theory is right; we make them first to find out how they will actually work, then to test whether they work right, and finally to see if the theory behind them is right. The more vulnerable we make the model to failure if it's wrong, the more confident in it we can be when we can't find anything wrong with it.

Best,

Bill P.

[From Adam Matic 2011.06.5 0015 gmt+1]

[Bill Powers (2011.05.28.1207 MDT)]

COMMENT: BP: The moderator was trying to question the assumption of some
special characteristic of all people to be incorporated into a model of the
economy. Instead of assuming that everyone always wants more of anything he
has or is getting, a better approach is to define a variable or a variable
parameter of behavior, and construct the model so that if the variable
doesn't have the correct value, the model will behave differently from the
real population. If no value will work, the model is wrong. That way your
model qualifies as a falsifiable hypothesis.

AM:
Could you give an example of a an appropriate variable or a variable
parameter of behavior? I'm not sure I understand that part.

This is the reason for the "How many people ..." kind of question -- it was
intended as a reminder that people are not all alike in their perceptions or
their preferences. Some people may be "rational actors," while not everyone
is. So if you want to use that premise, first you have to admit the
possibility that it may be incorrect, and then you have to set up the model
to catch the mistake if the premise turns out to be false. You have to see
what difference it makes to the model's behavior if everyone is a rational
actor, most people are, most people are not, or nobody is. If the model's
behavior depends entirely on the first possibility, and the possibility is
false, then you will find yourself making up good reasons why it's still
true. There's a lot of that sort of model-rescuing excuse-making going
around.

AM:
That looks like a good point. Do you think that "rational" behavior
would first need to be modeled independently from the model of
economy, and then later incorporated?

AM:

Final word of, one hopes, wisdom: when we make models we don't have to guess
how they will behave, because we can run the model and see. I think it's
best simply to make the components of the model itself as realistic as
possible and not assume that one can guess what will happen when they are
assembled and set into motion. We don't make models to prove a theory is
right; we make them first to find out how they will actually work, then to
test whether they work right, and finally to see if the theory behind them
is right. The more vulnerable we make the model to failure if it's wrong,
the more confident in it we can be when we can't find anything wrong with
it.

AM:
It took me some time to understand this. Thank you. Things are a bit
more clear now.

Best
Adam

[From Bill Powers (2011.06.04.1635 MDT)]

Adam Matic 2011.06.5 0015 gmt+1]

Could you give an example of a
an appropriate variable or a variable

parameter of behavior? I’m not sure I understand that part.

Do you think that
“rational” behavior would first need to be modeled
independently from the model of economy, and then later
incorporated?

The rational behavior would show up as your idea of how individuals make
decisions to buy and sell things. So they are the economy; the
rest is infrastructure. If you think people try to estimate the best
return on investment, put that in your model. If you think (as I do) that
they have needs and desires for goods and services, and opinions about
how much work they want to do, and how much money they want to have in
the bank, then model those control systems. You can allow for a range of
the needs and desires, and the importance given to them (loop gain). The
parameters generally have to do with the reference conditions and the
gain – that is, how much effort ta person will put out to counteract a
given amount of error. Also we need to adjust which goods are wanted and
in what quantities. The first question is not how a rational actor would
behave, it’s whether the rational actor model is right. You’d better
settle that question about any model before investing a lot of effort in
building and testing it.
In my last model, I proposed that managers responsible for pricing adjust
prices to control inventory. If inventory is increasing they lower
prices, and so on. Of course other managers might be controlling for
maximizing profit, which could lead to interesting conflicts with the
managers who adjust pricing, and with other managers who negotiate wages
and dividends and adjust such costs of production. I didn’t get into such
complications, but eventually one would have to deal with them. This
model behaved in ways that looked reasonable, but I never got beyond the
initial stages.
The point of a model is to get it working and see what it does. That
gives you something to start improving on. The model will do
something, and it’s up to you to see if what it does is
reasonable. You don’t have to guess right the first time. Every time you
run the model it is likely to do something unexpected or unexplained.
That gives you the information you need for the next revision.

Best,

Bill P.