FW: REPLY PCT and an economic emergency (SD7270)

Fred_Nickols1 · December 8, 2008, 2:57pm

[From Fred Nickols (2008.12.08.0756 MST)]

Here's another from the SD list. This one points to what might be an informative model. See the very end of the post I'm forwarding.

···

--
Regards,

Fred Nickols
Managing Partner
Distance Consulting, LLC
nickols@att.net
www.nickols.us

"Assistance at A Distance"

-------------- Forwarded Message: --------------
From: "SDMAIL Tom Fiddaman" <tom@ventanasystems.com>
To: System Dynamics Mailing List <sdmail@lists.systemdynamics.org>
Cc: System Dynamics Mailing List <sdmail@lists.systemdynamics.org>
Subject: REPLY PCT and an economic emergency (SD7270)
Date: Mon, 08 Dec 2008 12:21:21 +0000

Posted by Tom Fiddaman <tom@ventanasystems.com>

John Voyer correctly notes that positive feedbacks in the economy are not news.
The multiplier-accelerator and other theories have been around for a long time.
Where Powers really goes astray is with the notion that we can build a model,
reach consensus, and solve the problem technocratically. The system is too
complex for that. If you read the blogs of academic economists, you'll find
dozens of theories about the financial crisis, and little formal basis to
distinguish among them. By "formal" I really mean "model," and models have not
been visible in the discussion, at least to me.

The nugget of wisdom in Powers' message is that positive feedback can work both
ways, and that efforts to maximize a system's growth rate can have destabilizing
side effects. Like the electric power system, the economy works best when it's
on, and is difficult to restart if inadvertently switched off. Optimizing for
robustness might be a better option. Ironically, there's little evidence that
efforts to increase the growth rate actually do anything - per capita GDP growth
has been nearly constant in advanced economies since the industrial revolution.

To some extent, it might be possible to remove positive feedbacks, as Powers
proposes. One might, for example, reconsider the implications of money creation
through lending. I doubt that it's possible or desirable to eliminate all
positive loops though - after all, if you eliminate capital accumulation, growth
stops. The focus of most macroeconomic policy though is creation of
countercyclic
negative feedbacks. It's not clear that those policies are well understood (see
Forrester thesis below), or that they work in extreme conditions (as we're
seeing
now).

However, I think it's wrong to blame the rules and place faith in the invisible
hand. Mark-to-market, for example, might have contributed to the problem by
creating one of Powers' positive loops (falling prices force asset liquidation
to
meet capital requirements, driving prices further down). However, in a world
where undistorted signals were sufficient for stability, investors would have
seen that coming and priced assets accordingly, and the crisis would never have
occurred. The worst of the subprime lending was in the private sector; it wasn't
the GSEs who were driving ARMs, low-documentation loans, and other junk.

The real problem is not Fed policy or FNMA, it's that people got excited about
rising asset prices and forgot to think about fundamental value and risk. As a
result they took on too much leverage and too many non-transparent instruments.
There were bubbles long before there were regulators.

The real challenge here, I think, is that a lot of the perverse positive loops
are in people's heads. Education helps, but people forget, so it's an ongoing
process. Forgetting is amplified by evolutionary effects. It's hard to
distinguish between a robust strategy and risky speculation. Given the
difficulty
with attribution, it's quite likely that selection pressure will favor managers
who generate short term returns, and thus good times will gradually drive out
good sense. To some extent this is the problem of evil, and we will never fix
it,
but we do need to make some headway on such problems, otherwise the financial
crisis will be merely the first of a number of global catastrophes.

If you want a good example of SD work in this space, take a look at Nathan
Forrester's thesis.
http://dspace.mit.edu/handle/1721.1/15739
At least take a look at the introduction (pgs 7-18) and conclusions (215-217).
You can download the model from my site here
http://metasd.com/models/index.html#Business

Tom
Posted by Tom Fiddaman <tom@ventanasystems.com>
posting date Sun, 07 Dec 2008 10:21:53 -0700
_______________________________________________

Bill_Powers1 · December 8, 2008, 5:45pm

[From Bill Powers (2008.12.08.0924 MST)]

Fred Nickols (2008.12.08.0756 MST)]

I don't know if my replies to your posts are going to the originators, so I'm copying this to Tom Fiddaman.

-------------- Forwarded Message: --------------

From: "SDMAIL Tom Fiddaman" <tom@ventanasystems.com>
To: System Dynamics Mailing List <sdmail@lists.systemdynamics.org>
Cc: System Dynamics Mailing List <sdmail@lists.systemdynamics.org>
Subject: REPLY PCT and an economic emergency (SD7270)
Date: Mon, 08 Dec 2008 12:21:21 +0000
>
> Posted by Tom Fiddaman <tom@ventanasystems.com>
>
> John Voyer correctly notes that positive feedbacks in the economy are not news.
> The multiplier-accelerator and other theories have been around for a long time.
> Where Powers really goes astray is with the notion that we can build a model,
> reach consensus, and solve the problem technocratically.

Technocratically? That sounds really bad. Let's not try to solve it that way.

I didn't expect to find that positive feedback is new to SD modelers, or even to economists, although some of them still use the term to mean "appreciative words," at least when off duty. I haven't, however, seen much recognition of the fact that positive feedback works in both directions, not just in the direction of increasing a variable's magnitude. It's a model of unstable equilibrium, like a needle balanced on its point. As soon as the balance is disturbed in any direction, the needle will fall all the way in that same direction.

There's also the matter of positive feedback which is frequency-dependent, so that feedback that is nominally negative for static values or very slow changes in a variable becomes positive when changes occur at or above a certain frequency, or when the changes are delayed enough. The system will spontaneously oscillate at the lowest frequency where the feedback is positive and the gain around the loop is greater than 1. This is a special kind of runaway condition in that the runaway is in the form of ever-increasing oscillations, not simply a continuous increase or decrease. This would seem closer to what we observe as boom-and-bust cycles.

The system is too
> complex for that. If you read the blogs of academic economists, you'll find
> dozens of theories about the financial crisis, and little formal basis to
> distinguish among them. By "formal" I really mean "model," and models have not
> been visible in the discussion, at least to me.

I don't accept the idea that the system is too complex to model. How can you know that until you really try to do it? I'm not going to worry about what's possible or impossible until I try and fail (which has, of course, been known to happen). Models don't have to be as complex as the systems they represent -- if they were, there wouldn't be much point to modeling.

We may have a semantic problem here with the term "model". I use the term to mean, interchangeably, "simulation," which in turn can be interpreted to mean "numerically solving the set of simultaneous nonlinear differential equations describing the system." A simulation works by "acting out" the equations, whereas the analytical approach is limited to cases where the equations can actually be solved for the variables of interest using the rules of mathematics.

To do a simulation, you break the system down into modules with inputs and outputs, and either from first principles or from experimental data (or by hypothesis to be tested) write the equations for each module that describe how each output variable depends through time on the set of input variables for that module. Then you describe the connectivity in the system: how each input variable depends on output variables from other modules. Variables that do not depend on other system variables are the independent variables of the whole model, which can be set or varied arbitrarily to represent the experimental situation.

With the analytical approach you then try to solve this system of equations, which for system of any complexity is usually impossible even when many of the observed relationships are idealized. In a simulation, however, you simply connect the modules as indicated, turn on the stimulation, and let it run while you record the behavior of all the system variables.

With regard to a simulation of the economy, the detail you put into the model -- the size of the modules -- depends on the detail you want to get out of it. If you want to know how each individual in the economy behaves, you have to have at least one module for each individual, and the same number of simultaneous equations in the same number of variables. If you want to know only how sectors in the economy behave, you represent aggregates of individuals in terms of average properties and average variable values. This, it seems to me, is pretty standard practice in the modeling community, inside or outside economics, as well as in other places where modeling is a normal tool such as engineering, physics, and chemistry. With the capabilities of even ordinary desktop computers at our disposal, there's no need to balk at a little complexity. Our home computers can do a billion calculations while you brush a fly away.

What PCT brings to the modeling of the economy is a new set of principles of human behavior, which imply a very different model of Economic Man from the one assumed in all economic theories I have seen (not many). I think the usual model of the person used in economic theories is flat wrong, and I think I can prove it by, among other things, a collection of demonstrations in which people behave in ways that the usual economic model can't come anywhere near predicting correctly. Since human desires and human behavior drive the economy, a new model of the individual will necessitate changes in any model of the economy that takes human properties into account. We don't need to spend much time on models that don't.

I also think that by introducing this new model of human beings explicitly and making it part of a simulation, and by modeling the infrastructure of economics as faithfully as we can (the basic rules, such as paying for what you acquire, and acquiring what you pay for, and other such basic taken-for-granted laws of American economics), we can construct a test-bed simulation into which a person could plug any assumptions, any policy recommendations, any "facts" one wishes and see what the consequences will be. That's the whole point of my proposals concerning economic models.

Such a model doesn't have to be taken on faith. The way I think modeling has to be used is first to build the model so it can predict what is already known to have happened, of course on the basis of the information available and events prior to the time of the prediction. Then, with the model validated, we run the simulation using all the data available to the present time, and let the model show us what will happen next. That's how we do it in PCT, and how it's done in most of the hard sciences. I haven't seen this done in economics (but I haven't yet seen what has been done in System Dynamics).

I'm certainly not the best person to construct such an economic model. But at the moment, I'm a local expert on PCT and human control systems, and that part I do know quite a lot about. I'm not likely to make any serious mistakes about control systems, but when I try to model economic systems I will probably, and have probably, made a lot of mistakes. Either I have to fix those mistakes slowly and laboriously, or someone who knows more about economic facts and also learns PCT will have to do it, probably quicker.

···

==========================================================================================

I can't read Nathan Forrester's thesis since I'm not at MIT, but I'm grateful for the link to the models, though that means I have to do some work that I hadn't planned on. I'll report back when there's any progress.

Best,

Bill P(owers)