Latest Cowan LRA Model's Global Semi Sales Forecast Results For 2013 and 2014

Latest Cowan LRA Model's Global Semi Sales Forecast Results For 2013 and 2014

Loading
Loading Social Plug-ins...
Language: English
Save to myLibrary Download PDF
Go to Page # Page of 7

Description: Updated global semiconductor sales and sale growth expectations for 2013 and 2014 as derived by exercising the Cowan LRA forecast model are presented and discussed. These latest updated results are based upon October 2013's global semi sales as abstracted from the WSTS's October 2013 HBR, Historical Billings Report, posted on their website on Dec. 5th, 2013.

 
Author: Mike Cowan (Fellow) | Visits: 2658 | Page Views: 3158
Domain:  High Tech Category: Semiconductors Subcategory: Big Picture 
Upload Date:
Short URL: http://www.wesrch.com/electronics/docxEL1SE1J5APKCT
Loading
Loading...



px *        px *

* Default width and height in pixels. Change it to your required dimensions.

 
Contents:
2013, 2014 Global Semiconductor Sales and Yr-O-Yr Growth Expectations per the Cowan LRA Forecasting Model
(Derived From the WSTS's Oct. 2013 HBR, Historical Billings Report)
Date of Latest Monthly Forecast Update Write Up: Dec 06th, 2013; Author: Mike Cowan
For the past six consecutive months I have provided (hopefully) interested weSRCH readers my latest monthly forecast write-ups

summarizing 2013 and 2014’s global semiconductor sales and sales growth forecast numbers. Each monthly forecast update is derived by
exercising the Cowan LRA, Linear Regression Analysis, forecasting model operating on each successive month’s latest actual global
semi sales abstracted from the WSTS’s monthly HBR. This month’s updated forecast write-up is based on October 2013’s actual
sales of $26.116 billion which was reported on 12-05-13 via posting on the WSTS’s website (@ http://www.wsts.org/TeaserLeft/Historical-Billings-Report). It should be highlighted that this month’s October 2013 actual global semiconductor sales reported by
the WSTS represents an historical high for the semiconductor industry’s actual October monthly sales handily topping the
previous record high actual October sales of $24.890 billion which occurred last year – that is, for October 2012.
The table shown below provides a high-level overview of the model’s latest, updated forecast expectations for global semiconductor
sales projections for both the fourth quarter and second half of 2013 as well as 2014’s full year sales numbers broken out on a quarterly,
half-year, and full year basis, respectively. These latest 2013 and 214 global semiconductor sales forecasts are compared to their previous
years’ global semi sales thereby providing year-over-year sales growth expectations for each of the identified time periods. As exhibited
in the table, 2013’s latest full year semiconductor sales forecast came in at $304.21 billion which is up 0.25 percent from the previous
month’s forecast estimate of $303.45 billion. The resulting 2013 year-over-year sales growth expectation increased slightly to 4.3
percent, up 0.2 percentage points from last month’s sales growth forecast estimate of 4.1 percent.
COWAN LRA MODEL's 2013 & 2014 GLOBAL SEMI FORECAST EXPECTATIONS
QUARTERLY, 1/2 YEAR, AND FULL YEAR SALES / SALES GROWTH ACTUAL AND FORECAST NUMBERS
FOR 2012, 2013 AND 2014 (UPDATED => 12-06-13)
2 0 1 3 FRCSTED
2013 Yr-O-Yr
2 0 1 4 FRCSTED
2014 Yr-O-Yr
Time
2 0 1 2 ACTUAL
SALES ($B)
Sales Growth (%)
SALES ($B)
Sales Growth (%)
Period
SALES ($B)
$75.141
1st QTR
$69.833
$70.449
0.9%
6.7%
2nd QTR
$73.094
$74.645
2.1%
$77.574
3.9%
$152.715
1st HALF
$142.926
$145.094
1.5%
5.3%
$83.584
3rd QTR
$74.430
$80.552
8.2%
3.8%
$236.298
3 QTRs
$217.357
$225.646
3.8%
4.7%
$78.565
5.9%
$81.523
4th QTR
$74.205
3.8%
2nd HALF
$148.636
$159.117
7.1%
$165.107
3.8%

FULL YEAR

$291.562

$304.211

4.3%

$317.822

4.5%

Sources: Cowan LRA Model, WSTS (NOTE: 2012, 1Q13, 2Q13 And 3Q13 Are Actual Sales Numbers)

NOTE => ALL ITALICIZED NUMBERs ARE FORECAST ESTIMATEs

As highlighted in previous monthly write-ups, the Cowan LRA Model was extended in order to predict the sales forecast numbers for all
four quarters of 2014. Consequently, the above table also provides updated sales forecast estimates for next year's sales and
corresponding sales growth forecasts. As shown the updated resulting full year 2014 sales prediction came in at $317.82 billion. Relative
to the model's just updated full year 2013 sales forecast of $304.21 billion, 2014’s newly updated sales growth forecast estimate becomes
4.5 percent which is a downward drop of 0.2 percentage points relative to last month’s published 2014 sales growth forecast of 4.7%.
The second table (below) captures the latest updated forecast "prognostications" for seventeen other market watchers’ sales growth
expectations for 2013 and 2014 – ranked from high to low based on year-over-year sales growth forecast expectations for 2013.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

VARIOUS FORECASTERS "THINKING" (FORECAST WISE) - Ranked from HIGH to LOW
UPDATED: Dec. 6th, 2013
2013 Yr-O-Yr
2012 SALES
2013 SALES
Sales Growth
FORECASTER
(ACTUAL - $B) (FORECAST - $B) (FORECAST - %)
Objective Analysis (Jim Handy)
------"Approach 10%"
VLSI Research (ICs Frcst)
$245.8
$264.6
7.7%
BofA Merrill Lynch Global Research
$291.562
$313.756
7.6%
$313.041
DataBeans
$292.190
7.1%
IDC
------6.9%
Future Horizons (Malcolm Penn)
------6.9%
$308
Semi Intelligence (Bill Jewell)
$292
5 - 6%
Semico Research
------5 - 6%
Gartner
$299.912
$315.390
5.2%
$317.9
IHS - iSuppli
$302.9
5.0%
IC Insights
------4.8%
WSTS / SIA
$291.562
$304.309
4.4%
COWAN LRA FRCST MODEL
$291.562
$304.211
4.3%
TSMC (Morris Chang, CEO)
------4.0%
Value Line Publishing (A. House)
------3.0%
$298
ABI Research
---3.0%
The Information Network
------1.5%
Carnegie Group (Bruce Diesen)
-------1.0%
Sources: Various Market Researchers As Identified

NOTE => ALL ITALICIZED NUMBERS ARE FORECASTS

MOST RECENT
FORECAST
DATE/MONTH
8-05-13
10-25-13
6-28-13
Jan 2013
7-29-13
7-23-13
11-20-13
9-19-13
10-10-13
12-03-13
10-07-13
12-03-13
12-06-13
10-17-13
10-04-13
9-12-13
Aug 2012
May 2013

2014 SALES
(FORECAST - $B)

2014 Yr-O-Yr
Sales Growth
(FORECAST - %)

$343.962

9.6%

$354

15%

$316.636
$317.822

4.1%
4.5%

As previously mentioned the latest Cowan LRA Model's semiconductor sales growth expectation for 2013, derived from the WSTS's just
released October 2013 HBR actual sales numbers, jumped slightly to 4.3 percent compared to the model’s prior month’s published 2013
sales growth forecast prediction of 4.1 percent (see last month’s forecast write-up at: http://electronics.wesrch.com/paper-details/docxEL1SE1J5AUSIV-2013-2014-global-semi-sales-and-yr-o-yr-sales-growth-expectations-per-cowan-lra-forecast-model). Thus, this month’s
Cowan LRA Model’s year-over-year sales growth forecast ranks 13th in the list (high to low) among the eighteen semiconductor market
researchers that the author regularly follows as detailed in the table above. Moreover, the mean 2013 sales growth expectation for the
overall population is 5.1 percent with a relatively wide range of plus 10 percent to minus 1.0 percent. Note that the WSTS just released
its Autumn 2013 semiconductor market sales forecast numbers for 2013 and 2014 which are reflected in the table.
The third table (shown below) displays the model’s 2013 month-by-month sales and sales growth forecasts revealing their historical
“evolution” covering the past sixteen months (since September 2012) as well as for 2014 over the last eight months starting in May of
this year. One should keep in mind that the month-by-month global semiconductor sales and associated sales growth forecast estimates
are derived from a purely mathematical model that features linear regression analysis techniques operating on the past 29 + years of
historical, actual sales numbers from 1984 through 2013 YTD, Year To Date, that are gathered, assembled and published each month by
the WSTS (see the write-up appended below for an overview summary of the Cowan LRA Model).
BY-MONTH EVOLUTION OF COWAN LRA MODEL's GLOBAL SEMI SALES FORECAST EXPECTATIONS FOR 2013 / 2014
MONTH Of Sales Data =====>
JUL 2012
AUG 2012
SEPT 2012
OCT 2012
NOV 2012
DEC 2012
DATE Of Forecast Update
9-10-12
10-10-12
11-07-12
12-10-12
1-09-13
2-07-13
$321.096
$316.237
$307.044
$311.629
$309.244
$302.022
2013 Sales Forecast ($B)
-1.5%
-2.9%
1.5%
-0.8%
-2.3%
SEQUENTIAL PERCENT CHANGE (% )
2012=>13 Sales Growth Frcst Est.
7.7%
7.4%
5.2%
6.4%
5.5%
3.6%
MONTH Of Sales Data =====>
JAN 2013
FEB 2013
MAR 2013
APR 2013
MAY 2013
JUN 2013
DATE Of Forecast Update
3-07-13
4-08-13
5-07-13
6-09-13
7-05-13
8-06-13
$310.667
$303.695
$296.310
$300.317
$303.197
$300.685
2013 Sales Forecast ($B)
0.46%
-2.2%
-2.4%
1.4%
1.0%
-0.8%
SEQUENTIAL PERCENT CHANGE (% )
6.6%
4.2%
1.6%
3.0%
4.0%
3.1%
2012=>13 Sales Growth Frcst Est.
YTD 2013 Actual Sales Grwth TREND (%)
7.7%
1.9%
0.9%
0.4%
1.4%
1.5%
$312.397
$317.854
$320.603
$317.403
2014 Sales Forecast ($B)
SEQUENTIAL PERCENT CHANGE (% )
1.7%
0.9%
-1.0%
----2013=>14 Sales Growth Frcst Est.
5.4%
5.8%
5.7%
5.6%
JUL 2013
AUG 2013
SEPT 2013
DEC 2013
OCT 2013 NOV 2013
MONTH Of Sales Data =====>
DATE Of Forecast Update
9-09-13
10-07-13
11-07-13
1-0x-14
2-0x-14
12-06-13
$303.944
$304.145
$303.453
2013 Sales Forecast ($B)
$304.211
1.1%
0.07%
-0.23%
0.25%
SEQUENTIAL PERCENT CHANGE (% )
4.2%
4.3%
4.1%
2012=>13 Sales Growth Frcst Est.
4.3%
YTD 2013 Actual Sales Grwth TREND (%)

2014 Sales Forecast ($B)
SEQUENTIAL PERCENT CHANGE (% )

2013=>14 Sales Growth Frcst Est.

2.4%
$320.449
1.0%
5.4%

3.4%
$319.709
-0.23%
5.1%

4.0%

3.9%

$317.826
-0.6%
4.7%

$317.822
-0.0013%
4.5%

SOURCE: Monthly Runs Of Cowan LRA Forecasting Model, From Sept. 2012 through December 2013 (Date Of Forecast)

Also in the above table, I would like to draw the reader’s attention to the YTD 2013 Actual Sales Grwth TREND (%) line which tracks
the month-by-month evolution of the actual cumulative year-over-year sales growth percentage. Details of the by-month YTD global
semiconductor sales for both 2012 (full year) and 2013 (through Nov. 2013 Est.) are shown in the 4th table (below) along with the
corresponding by-month YTD 2012 to 2013 sales growth percentages. As shown this month’s Oct. 2013 actual YTD sales growth
percentage fell slightly to 3.9 percent with a 0.1 percentage point decrease from last month’s result (at 4.0 percent) following a turning
point (point of inflection) six months ago associated with April 2013’s low percentage of a 0.4 percent YTD sales growth. Also included
in the table are the latest updated 2013 sales and sales growth expectations derived from October’s sales for comparison.
YTD SALES
(TIME FRAME)

2012 YTD 2013 YTD 2012=>13 YTD
SALES ($B) SALES ($B)SALES GRWTH

JAN

$21.482

$23.128

7.7%

FEB YTD

$43.426

$44.233

1.9%

MAR YTD

$69.833

$70.449

0.9%

APR YTD

$93.664

$93.997

0.4%

MAY YTD

$116.617

$118.308

1.4%

JUN YTD
JUL YTD

$142.926
$166.536

$145.097
$170.600

1.5%
2.4%

AUG YTD

$189.575

$195.928

3.4%

SEPT YTD

$217.357

$226.015

4.0%

OCT YTD

$242.247

$251.762

3.9%

NOV YTD - EST.

N/A

$276.942

4.1%

NOV YTD - ACT.

$266.109

TBD

TBD

DEC YTD

$291.562

2013 FRCST

-----

$304.211

4.3%

Source: WSTS's Oct. 2013 HBR (DEC 2013)

Finally, the model is forecasting that next month’s (November 2013) actual sales estimate will come it at $25.180 billion. This
corresponds to a predicted November 3MMA forecast sales of $26.895 billion which is normally reported, each month, by the SIA as
part of their monthly Global Sales Report press release posted on its website at the beginning of the month (see the following URL =
http://www.semiconductors.org/news/global_sales_report_2013/). Also note (see table above) that the Nov. YTD estimated sales
forecast for 2013 is projected to come in at $276.94 billion which would yield, if realized, a corresponding 2013 November YTD sales
growth forecast estimate of 4.1 percent, which is 0.2 percentage points lower than the model’s latest predicted full year 2013 sales
growth expectation of 4.3 percent.

The Cowan LRA Model’s Semiconductor Forecast – A “Succinct” Overview of Its Forecasting Methodology
Developed by Mike Cowan (Independent semiconductor industry analyst and market watcher)

A unique and practical global semiconductor sales forecasting model created by the author facilitates the determination of
future global sales of the semiconductor industry. The Cowan LRA (Linear Regression Analysis) Model, which forecasts
global semiconductor sales, is a strictly mathematically based model that features statistical analysis of the past 29 years
of historical, monthly actual global semiconductor sales numbers that are collected and published by the World
Semiconductor Trade Statistics (WSTS) organization. The model is a dynamic, mathematically pure view of near-term
worldwide semiconductor sales looking forward over the next five to eight quarters.
Since the model is purely mathematically-based, it is devoid of any economic assumptions or emotional biases. It
exploits linear regression analysis operating on the "appropriately transformed" actual monthly semiconductor sales
numbers that are gathered and reported by the WSTS. This mathematical transformation of the actual monthly sales
numbers thus "renders" the global semiconductor sales data highly linear and, therefore, very amenable to linear
regression statistical analysis techniques. The numerical transformation of the past 29 years of monthly actual sales
numbers -- from 1984 through 2012 -- that is invoked is not a complicated mathematical expression but very straight
forward and "makes sense physically," thereby yielding extremely high linear regression correlation coefficients of greater
than 0.97. In exercising this forecasting model each month, a total of nine distinct sets of linear regression parameters (of
the format: y = mx + b) are employed to calculate the resulting global semiconductor sales forecast estimates for each
one of the next eight quarters associated with the model's forecast horizon as well as producing next month’s sales
forecast expectation.
It should be emphasized that each month's actual global sales number released by the WSTS (as posted on its website)
is a "lagging indicator" since it is released a full month after the fact, that is, it is backward looking. The Cowan LRA Model,
however, "turns" this lagging monthly sales number into a "leading indicator" by virtue of its near-term forecasting
capability looking out over each of the next eight quarters. This is the "beauty" of the model and, therefore, makes it
dynamic in the sense that it is run each month using the most recent actual global semiconductor sales numbers
published by WSTS via their monthly Historical Billings Report. Thus, it can rigorously "track" the near-term sales forecast
outlook of the global semiconductor industry on an almost real-time basis. Consequently, the model's monthly sales
forecast expectations do not "sit still" but evolve each month since conditions can change rapidly and unexpectedly in the
semiconductor industry, and market forecasters are hard pressed to keep up with these monthly changes. How can
industry management be sure that a forecast issued two, three or more months ago is still relevant to what's happening in
today's dynamic semiconductor market? With the Cowan LRA Model’s forecast results being issued on a monthly basis
immediately following the WSTS’s posting of its monthly HBR on its website, this can, therefore, help eliminate this very
real world dilemma.
ABOUT THE AUTHOR

Mike Cowan, the developer of the Cowan LRA Model, is a 48-year semiconductor industry veteran. He enjoyed a 36.5-year career
at IBM's Microelectronics Division in East Fishkill, N.Y., where he was involved in many facets of semiconductor development
and manufacturing engineering, including both technical and management responsibilities. Over his last ten years at IBM, as a
senior technical staff member, he was involved in strategy development and competitive analysis activities focused on the
semiconductor industry, and developed a wide range of top-down and bottom-up models to predict the dynamics of the
semiconductor industry. After retiring from IBM in Jan. 2002 he became an independent semiconductor industry watcher /
analyst, upgraded and perfected his sales forecasting model and then launched his monthly global semiconductor sales
forecast activities in August 2002. Cowan earned both BS and MS degrees in Physics from Wayne State University in Detroit,
Michigan, and an MS in Electrical Engineering from Syracuse University in New York. For more information and / or questions
regarding either the model and/or the monthly forecast results, the reader is encouraged to contact Mike directly at
mikedcowan(at)verizon(dot)net.

The following three articles were posted by the editors, Jeremy Young and Rob Lineback, of The Semiconductor
Reporter website back in August and September of 2002 in order to introduce the Cowan LRA Model to their
readership as part of the kick-off of a monthly global S/C sales forecasting feature on their website
(semireporter.com). These articles provide additional insight regarding the modeling technique and forecast
results based upon their views of the value and benefits of the Cowan LRA Model as a forecasting methodology
and approach relative to the model’s monthly output.

Introducing the Cowan LRA Model semiconductor revenue forecast (1st Article)
The Semiconductor Reporter, August 5, 2002, 8 a.m. EDT

Conditions change rapidly and unexpectedly in the semiconductor industry, and market forecasts are hard put to keep up
with the changes. How can industry management be sure that a forecast issued two, three, or more months ago is still in
line with what's happening in the market?
To address the issue of monitoring forecast accuracy given the latest data, The Semiconductor Reporter presents a new
type of semiconductor market forecast, the Cowan LRA Model Forecast, developed by IBM Microelectronics veteran Mike
Cowan. Based on linear regression analysis (LRA) of historical worldwide semiconductor revenue data published by the
SIA and WSTS, the model is used to generate an updated forecast each month immediately following the monthly release
of the SIA/WSTS monthly revenue number.
About Mike Cowan (side bar)
Mike Cowan has a 36-year history at IBM's Microelectronics Division in East Fishkill, N.Y., where he was involved in many facets of semiconductor development and
manufacturing engineering, including both technical and management responsibilities.
Over his last ten years at IBM, as a senior technical staff member, he has been involved in strategy development and competitive analysis focused on the
semiconductor industry, and has developed a number of top-down and bottom-up models to predict the dynamics of the semiconductor industry.
Cowan earned both BS and MS degrees in Physics at Wayne State University in Michigan, and an MS in Electrical Engineering at Syracuse University in New York.

The proprietary Cowan LRA Model generates forecast estimates for worldwide semiconductor revenues -- an estimate of
what the following month's number will be, as well as estimates for the next four quarters and the current calendar year.
The estimates are derived using a dynamic model based on historical actual monthly semiconductor revenues over the
past 18 years. "The model incorporates both the cyclic and the seasonal nature of the actual historical data and relies on
statistical analysis of the historical data employing linear regression analysis of certain combinations of the actual data,"
Cowan said.
All market researchers revise their forecasts from time to time, either on a schedule or on an as-needed basis when they
see actuals drifting too far off the predicted course. Tellingly, they seldom refer to their previous forecast numbers, which
are now "wrong," when they announce their latest forecast.
Cowan, on the other hand, will publish his new forecast every month in a running tally matrix that shows the evolution of
revenue forecasts for each quarter over time as each month's new actual-revenue data is fed into the model, refining its
"view" of the future. These dynamic estimates are continuously changing, typically by a small amount each month. The
method reflects the reality that as a future quarter gets closer, revenues for that quarter can be more accurately predicted
-- based on current and historical data.
Based on worldwide semiconductor revenue data released Friday for the month of June 2002 ($13.276 billion), the Cowan
LRA Model is currently forecasting estimated 2002 total worldwide semiconductor revenues of $137.05 billion, a 1.4%
decrease from the 2001 total. The revenue forecast figures are all up slightly from the forecast generated by the model
based on May numbers.
Turning points in the market can be pinpointed in Cowan's running tally, as the "arrow" shifts from up to down -- when the
monthly adjustment to the revenue forecast stops pushing the number up and starts pushing it down (and vice versa).
Looking at the model's predictions as monthly actuals are fed into it sequentially from 2001 and going into 2002, for

example, the turning point was observed in the February 2002 estimates for upcoming quarters: they started getting
bigger when February's actuals were factored in.
Even though the model is a new one, its "predictions" for previous quarters and years can be seen by stepping it through
the historical data and comparing its forecasts to actual outcomes. Mr. Cowan is willing to share historical "track record"
tables with interested readers, and is interested in hearing feedback and fielding questions. In addition, those who wish to
dig deeper into the picture developed by the Cowan LRA Model may be interested in studying Mr. Cowan's Running Tally
Matrix spreadsheet. Please email him at mikedco@attglobal.net.

What's the point of a dynamic forecasting model? Who needs a forecast that won't sit still? (2nd Article)
By Jeremy Young; The Semiconductor Reporter, August 5, 2002, 8 a.m. EDT

There are numerous market analysts applying a range of methodologies to the task of predicting what the semiconductor
market is going to do in the near- and long-term future. Why do we need another one? In particular, why do we need a
dynamic forecast that changes every month?
After all, most corporations go through one major forecasting exercise each year, and then make adjustments once or
twice during the year if conditions change. Who could possibly take on the task of rejiggering their company's plans -every month? Ouch!
We all hope that will not become necessary. However, in reality the market itself is in a constant state of flux, and
forecasts that are made only once every six months, or even every quarter, cannot hope to stay in close step with the
market's moves.
Here's the point: even if we don't tear up the budget and the production plan every month and do it over, it pays to check
in frequently to make sure that the market reality has not made all our plans obsolete. We may have a forecast that we
hang our hat on, and it may be the forecast of record for many months in a row. But if things keep changing, we need to
keep checking to see whether market pressures are pushing the numbers outside the boundaries we set for it. We need to
know at the earliest possible point when it does become time to tear up the plan and make a new one.
That's why we got excited when we started looking at Mike Cowan's LRA Model forecast. Here is an impressive
forecasting methodology based on linear regression analysis that swallows another data point each month when the SIA
releases the monthly revenue figures, and instantly puts out a revised forecast. Here is a forecast approach that accepts
the reality that things constantly change: the forecast is dynamic, and as we watch it over time, we expect not only to
glean important information from its latest forecast, but from the changes that are occurring in the numbers it puts out
each month.
How can anything but a dynamic forecast remain true to a dynamic marketplace?
Keep calibrated
Readers who are deeply involved in analyzing market trends will want to examine the details of Mr. Cowan's LRA Model
forecast estimates, and analyze its dynamic behavior. They may also want to have a look at his monthly Running Tally
Matrix, which lays out in detail how the new data, including monthly revenue figures and revisions to previous months'
revenue numbers, produce changes in the forecast numbers. (Please email your request for the Running Tally Matrix to us
at editor@semireporter.com.)
Readers who don't have the time to get into it that deeply may simply want to check the forecast once a month to see if
the model has spotted a significant market shift or a turning point in the market's progress -- to see if it's just "steady as
she goes," or whether something new is showing up in the numbers that bear closer attention.
Market forecasters may want to keep an eye on the monthly forecast changes to get an early handle on whether they
need to look at their numbers again. In a sense, the Cowan LRA Model Forecast is a way to keep other forecasts
calibrated.
VLSI Research, for one, has recently said that it will update its forecasts every month -- in recognition, we believe, of the
same market reality that makes the Cowan LRA Model useful. Market researchers are not actually clairvoyant. And as
new data comes out, they need to continuously check and adjust.
We hope you will find The Cowan LRA Model Forecast a useful early warning system for your forecasting and planning
process. With luck, you won't have to change course more often -- but it could make you quicker on your feet when the
need does arise.

'Back-to-the-future' forecasts aim to clear up crystal ball (3rd Article)
By J. Robert Lineback, The Semiconductor Reporter; Sept. 4, 2002, 10 a.m. EDT

After seeing mixed results from chip industry forecasts in the past 10 years, Mike Cowan decided to take a crack at
finding new statistical approaches to predicting semiconductor sales based on widely available data and as few
assumptions as possible.

The challenge was transforming the most reliable data available--historic sales numbers compiled by the World
Semiconductor Trade Statistics (WSTS) organization--into some kind of reliable and useful leading indicator for market
trends.
"I went back to the basics, and just looked at historical data--the month's actual sales data as it is reported by WSTS and
the SIA," said Cowan, who retired from IBM Corp.'s Microelectronics Division in January after 36 years at the company.
"After looking at what I call the fine structure of that data, I discovered that with the right combinations of the data, it
could be applied to linear regression analysis and used as the basis of forecasting."
The result is the Cowan LRA Model, which is being used each month on The Semiconductor Reporter to update a
dynamic forecast for chip revenues based on historical sales data released by the Semiconductor Industry Association. The
new forecast from Cowan's model is being issued today (see the results), based on the July sales data released on Tuesday
by the SIA.
Cowan works with the single-month sales data, and his proprietary model incorporates both the cyclic and the seasonal
nature of the actual historical data collected over the past 18 years. Linear regression analysis (LRA) is applied to certain
combinations of the actual sales data from the 18-year period (see Aug. 5 story).
"Obviously one would like to have the sales data sooner," he said, referring to the time lag involved in collecting it from
around the world and then the SIA releasing the information in the U.S. market. "It is certainly a lagging indicator, and it
has been a source of frustration for a good many years."
But using all of the sales data from the past 18 years, Cowan said he has worked out the right combinations of data and
calculations to turn the historical numbers into a look forward.
According to Cowan, the beauty of this approach is that it is dynamic and provides "momentum vectors" month-to-month,
quarterly, and yearly, to "give you a flavor of the direction that the industry is going."
"It is a forecast estimate. No forecast is going to be accurate 100% of the time, all of the time," added Cowan, but he also
believes the LRA model has proven that it is not "happenstance statistics" because it works well when stepped through the
historical data from 2001--which "was certainly a pivotal year in demonstrating that things had changed," he said.
"It was 2001 when everyone [in the industry] got very frustrated. When you looked at the prognostications and laid them
out in a significant way across a wide range of market researchers, you just saw that everyone was way off the mark,"
Cowan said.
A variety of factors have caused fundamental changes in the chip industry in recent years, including a shift in emphasis
from stand-alone PCs to communications and an increase in low-cost systems for consumer markets. Sporadic growth
cycles and the unprecedented lack of visibility in 2002 have caused some market research firms to revisit and rework their
forecasting techniques, with a few moving to monthly updates of annual outlooks (see Sept. 3 story).
Cowan agrees that semiconductor forecasting has fallen into a certain degree of disarray following the worst chip industry
downturn in history last year and the weak recovery period of 2002. Part of the problem, he suggested, can be traced back
to top-down and bottom-up forecasting models that start with estimates of system shipments or wafer fab capacity and
then apply assumptions against those models.
"The supply-demand approach was okay [in the 1990s], and you could get ballpark numbers, but it left a little bit to be
desired," Cowan said. "But then came the big frustration in 2001."
That led Cowan back to widely available and most often historical data, which he figured could say something about the
future if only tickled with the right amount of statistical analysis.
Like most researchers, Cowan protects the specifics of his linear regression analysis and models, but he said his approach
"leverages existing data so that underlying assumptions are not way out in left field"--meaning off the mark like most
were in 2001.
Cowan said he also has other models besides the one that works off monthly WSTS/SIA sales data. Another of his
forecasting models uses widely available quarterly results from U.S.-based semiconductor companies and a revenue
consensus for leading chip makers from Wall Street chip analysts.

The IBM retiree believes consensus forecasting is currently the best approach for industry projections. The WSTS group
holds a worldwide consensus forecast summit twice a year for its fall and spring projections, but the biannual frequency is
not enough to keep up with the changing market conditions, Cowan noted.
So he has found that building a "consensus outlook" from the projections of a half dozen or more chip analysts at
brokerage firms works well as a consensus element in the forecast built off of quarterly results from U.S. companies.
"This is a new technique that also features existing data in terms of providing a forecasting methodology," Cowan
explained. "It cleans off the old crystal ball."

Subscribe
x