An important tool in the study of photovoltaics is the Learning Curve (AKA: Experience Curve). See Gregory Nemet's excellent work (1,2) for a primer on the subject. I've built an excel based learning curve simulator that can be downloaded here.
If you want to guesstimate the future cost of photoelectrics, a learning curve analysis is a good start. If you want to translate the future costs of photoelectrics into the future costs of photoelectricity then use the output of the learning curve calculator as an input into the levelized electricity cost calculator. The result is what's known in professional circles as a scientific wildass guess (SWAG) - on the street it's known as mathematical masturbation.
Current Production Costs
The current best of breed PV companies can produce crystalline photoelectric panels for $1.30/Watt. These costs are broken up into two parts: silicon costs (i.e. polysilicon cost) and non-silicon costs (i.e. wafer, cell and module processing costs).* Silicon and Non-silicon costs are around 70 and 60 cents/watt respectively (March 2010). These costs can be further broken down into...
Crystalline PV panels are made from polysilicon. Polysilicon is the same material used to make computer chips. In fact, scrap polysilicon from the computer industry has been the primary feedstock supply to the PV industry until relatively recently.
The expansion of the PV industry throughout the 1990s led many observers to note that the supply of scrap poly was going to run short of demand in the early part of the millennium - this is exactly what happened. The same observers noted that the supply crunch was going to lead to higher silicon prices but most everyone underestimated how high the prices would go. Why did they guess low? Short answer: the success of Germany's Feed-in tariff.
Anyways, the silicon crunch led to extremely high silicon prices and a commiserate boom in new polysilicon refining capacity. As a result the current polysilicon environment is one of glut and falling prices. The guessing game we currently face is figuring out how far and how fast polysilicon prices are going to come down. Further down the rabbit hole we go...
There are two types of refineries used to make polysilicon. Siemens process based refineries and Fluidized Bed Reactor (FBR) type refineries. The historical cost of production for Siemens refineries is around $25 to $30/kg while FBR plants have production costs a smidge under $20/kg. This begs the question, why isn't everyone building FBR plants? My gut feeling is that expanding Siemens based silicon production over the last few years was the quickest means to an end. FBR, while superior from a costs standpoint, was new and therefore riskier from a deployment standpoint.
So... My guess for polysilicon. Contract poly prices are going to come down to $30/kg in the next few years and approach $20/kg within 10 years. Fluidized bed reactors should start to dominate new refining plant construction in the medium term.
Wafer Processing Costs
Best of breed (Renesola, LDK) wafer processing costs are currently a little over 30 cents/Watt. These costs are expected to come down to 25 cents/Watt in the near term based on increased scale and improved manufacturing.
*The exact definitions of silicon and non-silicon costs vary by manufacturer. For example, Yingli and Trina define the terms as I have above. Suntech on the other hand includes wafer processing costs in their definition of silicon costs.