Should We Give the Forecasters a Break?
Laura Peters, Senior Editor -- Semiconductor International, 2/1/2001
 |
The release claimed that only Advanced Forecasting accurately predicted fourth-quarter 2000 peak growth for the industry. Others predicted peak dates ranging from 1Q01 to 4Q02, including Gartner Group's Dataquest, VLSI Research, IC Insights, Infrastructure and The Information Network. Representatives of Advanced Forecasting, who claim 90% accuracy in predicting turning points over the past 14 years, were "prevented from participating in the press conference at ISS 2001," SEMI's Industry Strategy Symposium, while several optimistic firms were included.
As a former employee of a market research firm, I can identify with the forecaster's plight. At the same time, I respect the work that Moshe Handelsman, Advanced Forecasting's president, has done. He uses a statistical, economic behavioral model to predict significant turning points - a unique approach. Advanced Forecasting's comments are legitimate when the company notes that its competitors did not accurately predict the start of the boom in 1992, the recession of 1996 and the 1998 recovery. But it's not as if the forecasters are not trying.
As identified by Bill McClean in his refreshing look at industry cycles (see Semiconductor International, January 2001), market downturns typically occur due to global recession, overcapacity and/or excess inventory. We know the industry cannot always get a handle on global economic conditions, but it can better control the overcapacity and inventory components of the equation by better managing the supply chain.
Some forecasters base their projections on demand. However, even if you could predict demand on a device-by-device basis (a task made more difficult by the proliferation of device types), other factors, such as increasing manufacturing efficiencies coupled with rapidly shrinking die size, increasing yields, shrinking product lifetimes, etc., lead to wide ranges of output and demand. This forecast is unacceptable too.
So what can we learn from the past? Despite the industry's 17% average growth rate, it has never grown at a rate between 10 and 20% in any given year - it is always greater than 20% (closer to 40% in peak years), or in the single digits, positive or negative. Cycles of growth are shortening because chip manufacturers correct for demand more quickly these days. Also, companies are improving their supply chain management and embracing the Internet as a vehicle for made-to-order rather than made-to-forecast methods. But this switch is not going to happen overnight.
I'm sure forecasters would like to factor in the chances of a global recession, and the (consequential) effect of the stock market. Then they can factor in the effect of the 300 mm conversion, including yields. And then add the real build rate of foundries and IC manufacturers, not company projections. Then consider the momentum factor - where everyone gets on the latest bandwagon and significantly influences its direction.
We should give the forecasters a break. They scrutinize articles from all segments of the industry, talk with top executives, track capital spending, examine computer/Internet/networking demand, demand for games, digital appliances and Palm Pilots. However, the market is a moving target. Unless the analysts gain access to key information outside their reach - inventory numbers, real capacity numbers, yields, die sizes and die size changes, all in real time - the accuracy of forecasting is not likely to change significantly.
Possibly Advanced Forecasting has the best methodology in our business, but the others are worthwhile too. Maybe we just need to change the way we interpret their forecasts.
