Linear Forecasting

Following the post I made about the downturn in Semico’s Inflection Point Indicator I received a forecast presentation from Mike Cowan. If you don’t know, Mike produces a linear regression analysis (LRA) of data released by the WSTS (World Semiconductor Trade Statistics). Over the last six, seven years he’s been quoted in many of the industry trade rags. Typically when I write something about a forecast he makes sure I know that he’s out there by sending his take.

"Hi Carl - I saw your posting entitled "Semis: A Bad Year Ahead?" (dated 6-23-11) at the following URL = http://blogs.forbes.com/carljohnson/2011/06/23/semis-a-bad-year-ahead/.

In your posting I noted you invoked the boldness of Semico in "Predicting all the way through the first few months of next year ......"

Therefore, I thought that I would "dazzle" you with my boldness per the following "quantitative" predictions of global semi sales as derived from the application of the Cowan LRA forecasting model. See the write below and the attached PowerPoint file for my "boldness"!!!

All righty then Mike….. Have a go! The more the merrier!

Before we get to the forecast some color is in order. When one uses "total semiconductor sales" they include categories like optoelectronics, sensors and discretes. I mention this because some of the industry forecasters exclude these and focus on the Integrated Circuit (IC) portion of the WSTS report. They do this because it has an impact on those who want to model capital spending as a percent of sales. Many believe it is more accurate to use IC sales only when modeling the cap-spending ratio.

Okay, LRA stuff… I really don’t have a problem with the methodology – it’s purely mathematical and it adjusts on the fly. You put the latest numbers in each month and out comes a projection for the next month and the next year. Setting up some ranges for the next monthly release allows you to build a scenarios for the year. Pretty straightforward stuff.

Here's a table detailing current projections (click for full size):



Mike provided his thoughts:

"Therefore, as the above scenario analysis table lays out, depending on the actual WSTS (to-be-released) May 2011 global semi sales number, the forecasted year 2011 sales estimate as determined by the model could vary between a low of $317.947 billion and a high of $325.778 billion. The corresponding 2010 to 2011 sales growth forecast estimate would then vary between 6.6% and 9.2%, respectively. "

"Note - last month's previously published Cowan LRA Model's sales growth forecast estimate for 2011, which was based upon April 2011's 'actual' sales (of $23.519 billion), came in at (plus) 8.0 percent based upon the model's 2011 sales forecast estimate of $322.328 billion as shown in the table immediately below."

Here's that table (click for full size):

You can find many of Mike's past monthly forecast reports on the WeSRCH website (URL = http://electronics.wesrch.com/?ts=1184274345&category=Semiconductors). Type "Cowan LRA" in the search box and they will pop up.

I suppose if you want to win the annual semiconductor forecasting trophy (is there one?) or do some quantitative (only) business planning this could be a useful model. Seems to me that if the projections were more focused – like on a specific category, say, NAND Flash, they might be even more useful.

To each his own…..