November 26, 2004

View from Silicon Valley- The trouble with forecasts.

© copyright View from Silicon Valley, 2004. All rights reserved.

Semiconductor sales people produce sales forecasts on a regular basis. We all understand accurate sales forecasts are essential tools for loading billion dollar wafer fabs.

I've sold for companies who wanted weekly, monthly and/or quarterly roll-ups for three, six or twelve months, often down the exact part number and selling price, per customer, per project and per month. I've worked for companies who thought it was perfectly reasonable to pull sales people out of the field to produce an "emergency" forecast-- on nearly a bi-weekly basis. One sales rep firm I worked for produced 18 sales forecasts per month (not counting "emergencies"...).

Sales quotas are based on forecasts and sales people get paid better when they make quota. (Duh!) On the other hand, I did spend five years at a major semi firm where my Area Sales Manager routinely waited until the 9th or 10th week of a 13-week quarter to set individual quotas. (He wanted uniformity in the sales force's earnings.) Despite this practice, this company continued to describe itself as a "meritocracy."

As an aside, the most ridiculous requirement I experienced was a weekly summary due every Friday by 5:00pm, even though several of us were usually still on airplanes at that hour. After the second or third time I inquired about an action item this sales manager failed to follow-up on from my report, he insisted that even though he didn't actually read the reports, it was still urgent they be submitted, on time, each week. (I imagine he also learned it was helpful to scream at sales people while getting his MBA.)

Sales which do not match forecast, in theory, should always be a problem. However, in the real world, you (almost) never get fired for exceeding, even wildly exceeding, forecast. Sales below forecast will not only cut your commission but also cause missed promotions and even firings. (Ask Carly Fiorina.) Which way do we slant forecasts once we understand these consequences?

Of course, competent corporate management understands this forecast dynamic and compensates by driving for the highest possible forecast. The realization higher forecasts foster higher stock prices is a happy side effect in this era of large stock option grants.

Sorry if this boring, or remedial, but much of the above came rushing back when I heard the latest sales forecast from the Semiconductor Industry Association (SIA). I know the forecast "game" well and, apparently, so does SIA.

For background, SIA history sales figures show ($B, CY04 through Sept'04):

As chairman of Silicon Valley Residents, Inc., my fictional company whose charter is to help Silicon Valley residents (or at least me) plan for the future, I am compelled to point out several problems with this forecast:

1) If you ever(!) turn in an annual forecast exactly matching the prior year down to four digits, expect to be laughed out of the room. (Assuming you're not demoted or fired on the spot.)

Anyone who actually generates forecasts knows it is only possible to match numbers to this degree when you start from the answer and work backwards. (On the other hand, this is sometimes exactly how you are told to do your forecast but that's another missive…)

Sorry SIA, go back to the  drawing board. Your 2005 result is simply not credible.

2) The press release accompanying SIA's forecast claims the US and Europe will hold share over the next three years. If so, this is excellent news for the Silicon Valley economy. Plans will be made. Monies will be spent. However, it would also mark a departure from a seven-year trend (CY04 YTD):

                    1998    1999    2000    2001     2002    2003    2004     Since'98
USA share:  32.7%  31.8%  31.3%  25.7%  22.2%  19.4%  18.4%      -44%
Euro share:  23.2%  21.3%  20.7%  21.7%  19.7%  19.4%  18.3%      -21%

SIA, can you explain how and why the erosion of US and European market share will be stopped?

2A) Digging further into SIA's current forecast details, you find:

                         2004     2005     2006     2007      Since'04 
USA share:        18.5%   17.6%   17.6%   17.1%     -7.6%
European share: 18.5%   18.5%   18.3%   18.2%     -1.6%

Losing seven share points in the US market is not "stable."  SIA, we need an accurate bottom line to effectively manage our finances.

3) SIA is a trade association, not a bastion of independent research.  SIA's 1999 forecast was:

               1999    2000    2001    2002
'99 Fcst     144     173      209     235
(Actual      150     204      139     140)

OK, SIA, your web site admits the semi business is cyclical. "Up" years are followed by "down" years. Busts follow booms. Even so, you failed to "see" the "usual" downturn in 2001 after two strong years in 1999 and 2000. Today, you see no "usual" downturn in 2005 sales after two strong years in 2003 and 2004.

Do we need to get someone in here who actually understands cycles?

* * *

Even if we accept SIA's numbers, important aspects are overlooked:

A) At risk of sounding like a broken record, despite tremendous semi sales growth since 2002, Santa Clara County jobs continued to decline (through Oct'04):

(With +14Ku net new government +healthcare jobs since 1998, private payrolls are actually down -13.1%.)

B) Current semi sales growth has led, as it always does, to increased fab capacity. Per Semiconductor International Capacity Statistics:

3Q04 fab capacity was up +3.6% and utilization dropped.

Foundry capacity up +10% in 3Q04 on top of +10% in 2Q04.

Total 300mm capacity up +18% in 3Q04 on top of +92% in 2Q04.

Four to five additional new 300mm wafer fabs are under construction and will begin production in 2005.(&)

C) According to Denali, current semiconductor memory profitability did not match or exceed prior sales peaks:

                      1999   2000   2001   2002   2003   2004   2005
Est.Profits($B)   2.2     7.0   -13.5    -6.2    -6.3      6.5     -??

Conclusion: Unlike forecasts submitting by real semi sales people, the SIA does not forecast units or prices. Also unlike real semi sales people, there are no repercussions to SIA if they forecast too high.  (On the other hand, heads might roll if they are insufficiently optimistic.)  

With massive incremental wafer supply coming on-stream, prices will inexorably decline in 2005. (Prices already started dropping 3Q04.)

Unit demand must grow precisely as fast as prices fall to make SIA's 2005 forecast.