Ventana software support forum

At 01:00 PM 4/12/98 -0400, you wrote:
>I have a question on choosing a modeling technique. I need to build a m=
odel
>describing effects of marketing effort on the market-- in a small sector=
. ...
> My model needs to look like a marketing flight simulator.
...
> The question is whether on not SD is suitable, and if suitable, is it t=
he
best >modeling technique available?

System dynamics seems like a good choice for your problem, for two reason=
s:
- Tools for building flight-simulator interfaces are more highly evolved
for SD software than for any econometric or time-series package I know of
(except maybe Excel)
- Just about anything worth doing in econometrics can be done better with=
SD

I have looked at econometrics and time series
>modeling books, and have noticed some things that concern me. In
econometrics,
>sales as a function of advertising exhibit =93ratchet effects.=94 The b=
ase
case of
>a ratchet effect is that sales go up once people become aware of a produ=
ct
>(through advertising), but a decline or halt in advertising won=92t caus=
e sales
>to return to their pre-advertising levels. I do not know how to create =
an
>effect like this in SD modeling.

This is really a matter of model, not method. SD models of marketing
problems often include many effects that can cause ratchets, as well as
their inverse (sales increase at first but decrease in the long run).
Forrester describes some advertising effects like this in Industrial
Dynamics, Chapter 16 (have we invented anything new since 1961?). In
general its easier to include feedback effects in an SD model than with
conventional econometric tools.

Some examples from SD models Ive seen:
- advertising "primes the pump" for word-of-mouth, which then sustains
sales even if advertising stops
- advertising gets people to try a product; they then become habituated t=
o
it and continue to use it when advertising stops
- a promotion gets people to buy a lot of a product, but then sales fall
because people have full larders (or get sick of the product, or wait for
it to wear out, or wait for the next promotion, or ... )

Many of these structures are already clearly articulated in the marketing
literature, and its easy to replicate them in an SD model. Good
econometric work is founded on these structures, but Ive also seen a lot
of bad work. The bad stuff tends to just estimate sales as a function of
marketing (and a few other variables like price) with a few lags thrown i=
n
to capture larder or ratchet effects. This doesnt give you a structural
understanding of whats going on, yields a model that is not robust to
extreme conditions, and when combined with the typical limited scope and
short time horizon of available time series data, is likely to be serious=
ly
biased.

In fact, I fear that the complexity of a
>system like this makes an SD model ineffective. Managers will not be ab=
le to
>get their minds around the problem (actually, defining the problem is a
sticky
>situation in my case) or model, so they would not be able to get as much
out of
>it as a conceptual model.
> What are your thoughts? Where should I go to learn about choosing mode=
ling
>techniques for different systems?

Im not sure what you mean by a conceptual model here. But if managers ar=
e
having trouble getting their heads around the conceptual issues, a
regression model definitely will not help them. An SD model could add a l=
ot
of value just by taking an inventory of mental models, especially making
the stock-flow structures clear. Then you can move to simulation, using
estimation and other sources of data where appropriate.

- Tom


****************************************************
Thomas Fiddaman, Ph.D.
Ventana Systems http://www.vensim.com
34025 Mann Road Tel (360) 793-0903
Sultan, WA 98294 Fax (360) 793-2911
Tom@Vensim.com http://home1.gte.net/tomfid/
****************************************************