Peter - glad you like the icon.
Relative to your question the model is strictly mathematical exploiting linear regression analysis (LRA) that "operates on" the past 26 years (1984 to present) of historical, actual monthly sales numbers (note - "appropriately transformed") as gathered and published by the WSTS.
As such the statistical analysis does not factor in any anticipated boundary conditions that the industry might experience in the near term but relies entirely on the historical experience of the industry as reflected in these actual monthly sales numbers over time.
The following paragraph re-enforces the above words and is abstracted from a PowerPoint presentation I have previously sent you that provides a more detailed overview of the model:
"Since the model is a purely mathematical forecasting approach, it is therefore devoid of any economic assumptions or emotional biases; that is, the model’s statistical analysis approach “captures the historical experience” of the industry as “reflected” in its actual monthly sales over the past 26 + years in order to portend the future sales direction of the global semiconductor industry covering each of the next five quarters."
PS - if your readers would like a copy of the complete, above mentioned, presentation they can send me an e-mail (mikedcowan(at)verizon.net) and I would be happy to dispatch them a copy for their perusal.
Thanks Mike....good icon.
Does your LRA model take into account that the second half could be supply-side constrained....we may have a partially front-loaded 2010 which would be different to the historical norm?
According to the latest run of the Cowan LRA (Linear Regression Analysis) forecasting model for predicting global S/C sales, the 2010 sales forecast estimate continues to improve.
The updated sales forecast, based on the recently released (actual) May 2010 sales of $24.007 billion, came in at $304.7 billion which corresponds to a projected 2010 sales growth of 34.6 percent.
These two updated results compare to last month's 2010 sales and sales growth forecast results of $301.9 billion and 33.4 percent, respectively.
Additionally, the model estimates that next month's global S/C sales, namely June's sales forecast estimate, will be $28.3 billion which, if realized, would represent an historical record high (5-week) monthly sales with the previous high sales of $26.3 billion set back in September, 2008.
Note that the previous historical high sales for June was $25.5 billion which also occured in 2008. Also note that subsequently the monthly sales during 4Q08 plunged quite dramatically as the worldwide economic problems "kicked off" and continued over the next two quarters, at least relative to global S/C sales historical experience.
Mike Cowan, Developer of the Cowan LRA forecasting model
This seems to suggest that the recovery is up and coming. Global chip sales is always a barometer to gauge the economic global activity. I think this is a good news. Again, China is showing that it is the future. Europe could have been hurt by the weaker euro. But generally, the world is up and running.
Great news for the ailing economy but before we start anticipating the next raise should we be asking what's driving these numbers. Is this a seasonal factor coupled with depletion of inventories last year or some genuine demand growth spurred on by iPads, Android phones and memory devices?
Surely we don't need another inventory glut which crashed the dram prices like in '08.
Great news indeed. It would be interesting to see a breakdown by sector to determine what product areas are actually seeing the growth. Especially since one of the the largest growth areas was in the Americas region, which has been a region of continuing economic instability in a number of areas. Surprising and very encouraging, but with a cautious optimism for the reasons Sanjib has stated in his prior post.
Blog Doing Math in FPGAs Tom Burke 24 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...