Papineau 2006 Experience curves in RE technologies

Papineau, M. (2006). "An economic perspective on experience curves and dynamic economies in renewable energy technologies," Energy Policy 34, 422-432.

• Unlike Klaassen et al (2005), this paper uses price of the equipment (turbine, module, etc.) rather than total investment costs per kW (though one regression did use price of wind power as the DV).
• Depending on the data available, uses either cumulated capacity (MW), cumulated electricity GWh), or cumulative output (total industry shipments) as independent variables.
• Estimates learning equations of the form logP = a + blogX for individual countries and for panels.
• Implies that a 1% increase in X (capacity, electricity, or output) will result in a b% decrease in the dependent variable - usually price.
• The progress ratio (PR=2^b) gives the percent of the previous price the new price will be after each doubling of the independent variable. The learning rate (LR) is 1-PR.
• Regressions of price against any of these dependent variables give statistically significant b parameters, and show learning ratios of between 10% and 20% for solar and 5-10% for wind.
• However, adding time trend (year) variable reduces significance and reduces learning ratio.
• Possible reasons for results: Use of price (rather than cost) as DV; price and cost changes affected by entry of new competitors and changes in industry concentration; price may change as result of gov't subsidies without change in production efficiency; since output determined by price AND price determined by output, the random disturbance term will be correlated with the regressor and estimated parameters will be biased and inconsistent.
• Learning equations with R&D (and R&D expenditures divided by annual sales, and RD/S*cumulated output) instead of capacity were estimated, and the only results shown included the time trend variable. Coefficients on R&D variables were very small, significant only in the case of PV in the US, and occasionally were positive (wrong sign). Doesn't say where R&D data came from, but seems to be IEA gov't expenditure on R&D on various RE technologies.
  
• Concludes by saying the results support introduction of "imperfect foresight and stochastic uncertainty of learning rates in energy system models."
• Also says increased funds should be allocated to R&D, though this conclusion can't be based in the results of her econometric results, since R&D funding (when time trend variable included) apparently had very small impact on price reductions. However, author was not explicit about the measurement of the independent or dependent variables in these estimations.