Re: Semiconductor Industry analysis

From: John Conover <>
Subject: Re: Semiconductor Industry analysis
Date: Sun, 4 Dec 94 11:54 PST

Dan hauer writes:
 > Most attemps at statistical analysis of the semiconductor industry for
 > the purpose of predicting the future have proven to be inaccurate both
 > in amount and in timing. And when you look for accuracy in both amount
 > and timeing I don't brlieve it has ever occurred.

Yea, but this is *NOT* a statistical method. In point of fact, it
confirms your observation that statistical methods will not work in
the analysis of SC markets.

 > There are many dynamics that affect the S.C. industry. A couple of the
 > key forces are:... that the SC industry is no longer a block but
 > a group of many sub industries that feed off of applications in other
 > industries...Rapid sales increases and decreases are fueled more by
 > perception than by a true balance of supply and demand...manufacturing
 > capacity for the "right products at the right time" have a tremendous
 > carry over effect to other related products.

Yes, and that is how the fractal theories of markets works. It is a
methodology that addresses the issues of complexity of systems,
without attempting to address the underlying causality. It is
obviously best to address the causality issues (through some form of
reduction and/or statistical technique) where the causality can be
formalized, but this is seldom the case in complex (eg., capitalist)
markets. In point of fact, it can be shown that a complex market place
can *NEVER* be analyzed on a causality basis.

 > The related product carry over has caused big problems repeatedly in
 > both the forecasting and manageing of companies in the industry. Memory
 > over or undersupply greatly affects all companies selling to the computer
 > markets {which is about 60% to 70% of us} Availability of Windows 95 will
 > also affect the timing and magnatude of many companies sales...Problems
 > with the application or customer acceptance of KILLER new products also
 > has a large affect. A recent example is the forecasted surge in home TV
 > top boxes for all kinds of uses. Well to date the volumes have been very
 > disappointing, due to many and varous reasons, but, oversupply
 > and additional available capacity has had a domino effect on other markets.

That's right. Actually, the technical name for "fractal" processes is
"innovative process." It presumes that "innovative processes" are the
"engine" that makes the market function a non-linear function
(otherwise, it would be a straight line.) Although it is impossible to
predict who will do the innovation in a market segment, some one
will. Additionally, although it can not predict when the innovation
will take place, innovations will appear in "clumps," as opposed to a
random basis. (A "bear" market is really an innovation of sorts,
although probably not by design.) So this can be used as a predictive
mechanism. This concept was originated by the "programmed traders,"
and has been very successful for the last quarter century. Programmed
trading is not a statistical method-it is a technique that exploits
the dynamics of the situation. It is very possible (and is usually the
case, for that matter) that more money is made when the market goes
down, than when it goes up. (Case in point: the 1987 crash, where the
programmed traders picked up 2% ownership of corporate America in 3
hours.) The statement "buy low, sell high" wouldn't work otherwise-and
is simply a statement that essentially says "exploit the dynamics of
the market." Fractal analysis is the formal methodology to do this.
In essence, it states that although the market has unpredictablilty,
it is also not random (there is a difference.)

For example, a related non-linear dynamic system is the weather. If
you have a year with lower rainfall, chances are significantly higher
that next year will also have lower rainfall. Although the causality
is not understood (and never can be,) and although the weather can
never be predicted from year to year, the fact that dry spells and wet
spells "clump" together can be exploited to enhance market share in
umbrellas. Note that this is not a statistical method-a statistical
method would say there is a 50% chance of next year being wet/dry
because that is the way it has been for all of recorded
time. Statistical methods assume a random distribution, and we know
that the weather is not random, since the weather this year *DOES*
influence what the weather will be next year.

 > Therefore, it is apparent to me that staticical analysis is an important
 > data point in the process of forecasting, but that many other factors add
 > up to have a dynamic force that is far greater than statisical momentum
 > or predictors. Over the years, I have gained a reputation for having a
 > good ability to forecast sales, markets, products, trends, etc. and frankly
 > I have used a lot of statistics to analysis the the past but little stats.
 > to project the future..........................................DAN HAUER..

That is exactly what fractal analysis says to do. (If you look at the
formal fractal methodology, it really does not tell you anything you
do not already know-statistics don't work, and can be misleading, be
nimble and be quick, be observant and play the dynamics. Unfortunately,
this is exactly counter to MBA/MBO methodology-which is a problem in
corporate America.)


John Conover,,

Copyright © 1994 John Conover, All Rights Reserved.
Last modified: Fri Mar 26 18:58:11 PST 1999 $Id: 941216132443.7940.html,v 1.0 2001/11/17 23:05:50 conover Exp $
Valid HTML 4.0!