PardeeCenterIFs
commented
1 month ago Today, during the modeling Pod, we discussed this function, and after work from Barry we all decided to update this function to flatten the slop even more, going from a coefficient of -0.1 to -0.05. This is because once countries reach the level of 20, at about $315k per capita, then the alphas were being kept constant at 20% or (0.2), which produced a jump in Economic Growth, shown in SSP5 in the late years of a 2200 horizon. This limit of 20% has also been relaxed to 5%.
The function in question is:
GDP/Capita (PPP) Versus Cobb-Douglas Alpha (GTAP 5)
In the context of new Constant Dollar data being used, namely 2017 Constant Dollars for GDP, many of our functions have been updated, but couldn't find data to update this one, so I asked Barry if we should use Forecast instead, or how was this originally created and this is his answer:
Barry: "Although the name of the function says GTAP5, I don’t remember ever estimating it statistically. In fact, my memory in contrast is that I looked to economic literature to judge how the share of value added going to capital (about 50% for low income countries) decreases with development relative to the rising share going to labor. That literature could have given rise (I think probably did) to the green line below for the function. I see also the function that includes “Estimated” in the title that has a much more rapid decrease in the capital share and therefore rise in the labor share. Perhaps GTAP data does have the data to estimate the shares indicated and either Mohammod and I could have fit a regression line to it. But the fall in capital share is so fast that I can see why we (probably I) would have decided that the more subjective/literature based formulation was preferable.
Another note: both functions, especially the blue line, are somewhat dangerous because they have no saturation toward a level that protects the capital share. I think a curve that saturates at something like 30% at perhaps $300 to 500 per capita would actually be a safer function for analysis. But the model code might be bounding the green line at the 35% level in transient mode."