Sunday, November 20, 2022

Inflating the Cost of Carbon

Comprehensive evidence implies a higher social cost of CO2  is a recent article in Nature which claims to calculate how much worse off humans will be for each additional ton of CO2 released.  The costs are summed over a period of almost three hundred years, from now to 2300. As best I can estimate from Extended Data Figure 2, about two thirds of their social cost of carbon is incurred after 2100. 

This raises serious problems. To begin with, CO2 output as a function of GNP depends on the technology for producing power. An order of magnitude reduction in the cost of either nuclear power or storage would almost entirely eliminate the use of fossil fuels, as would the development of cheap fusion power, either of which could happen in the next fifty years. That makes any estimate of CO2 output over the next three centuries a guess about unknowable technological change.

Almost all of the article’s estimated cost of carbon is from either increased mortality or reduced agricultural output. Mortality from increased temperature depends on medical technology, home insulation and cooling technology, and probably other technologies. Agricultural yields depend on agricultural technologies. We have no way of predicting those effects.

How does the article deal with technological change? As best I could tell, it ignores it. It is predicting the effect of temperature changes on mortality over the next three centuries on the assumption that they will be dealt with using the medical technology of today, and similarly for other relevant technologies. It is predicting the effect of climate change on agriculture with the same assumption.

Even with fixed technologies, individuals can, given time, alter their behavior to optimize against a changed environment — dike against sea level rise, change what crops they grow and where, how well they insulate their houses. The article explicitly includes adaptation in their calculations of costs due to sea level rise and concludes that the result is only about 1% of their total but, as far as I could see, it says nothing about the effect of adaptation of human behavior to higher temperatures on mortality. The agriculture piece mentions adaptation, but it was difficult to tell what forms of adaptation it was considering or how the effects were calculated. It says nothing about effects of progress in agricultural technologies to deal with the changed environment.

One of the things that affects mortality due to climate is income. Poor people have to go out in hot or cold weather more than rich people, are less able to afford well insulated houses or in other ways protect against the effects of climate. The rate of growth of per capita GNP assumed in the article's modeling, shown on Figure 1, implies that average incomes will have roughly tripled by the end of this century, by 2300 increased about eleven fold. As best I could tell, the calculation of mortality due to climate change entirely ignored that.

There is a second possible problem with the mortality calculations, although I am less certain of it since it is not entirely clear how they were done. The paper refers to the effect on mortality of unit increases in ambient temperature. That might mean increasing temperature throughout the year by one degree, it might mean increasing the average temperature over the year by one degree with whatever pattern of increase over the year they found in their data.

Hotter summers increase mortality, milder winters decrease it. Temperature increase due to greenhouse warming is greater in cold times and places than in warm, a pattern that can easily be seen in the IPCC projections of how much warming will occur where and its tables and graphs of change in minimum and maximum temperatures with global warming.

The article on which the Nature article bases its cost of mortality finds an average ratio of reduction in mortality to increase in mortality of .36. The IPCC report shows the ratio of the increase in minimum temperatures to the increase in maximum temperatures varying by regions, as high as three in some. If the article is assuming a uniform increase, correcting that error would sharply reduce the net mortality, in some regions to zero.

All of the errors I have described result in increasing the social cost of carbon. The article spends two paragraphs on things that might make the SCC higher than its estimate, says nothing I could find about things that might make it lower. For the effect of the new calculations on EPS policy, see The US Environmental Protection Agency Introduces a New Social Cost of Carbon for Public Comment.


FraserOrr said...

In all fairness David, the assumption about technology being the same is a fair one. After all, agricultural, energy and medical technologies today are pretty much indistinguishable from those ins 1722. After all, the story of George Washington driving in his EV down to his industrially farmed fields to check on his flash freeze processing plant is almost legendary. And thank God they gave him antibiotics for his throat infection rather than barbarically bleeding him to death.

Chris Oldman said...

I've been looking into this, too, but from the PoV of the EPA's regulatory aspect. I'm still trying to wrap my head around it totally, but I'm curious what your thoughts are regarding the discount rate. In the article, it points out that a large part of the increase in the total cost is just because they're using a lower discount rate.

If I understand correctly, the discount rate is basically there to represent the time preference, i.e. what rate would "society" demand in order to be indifferent between having a dollar today vs. having the accumulated amount in the future.

They have decided that 2% is more reasonable because we've been in a low-inflation, low-interest rate environment recently. I'm not sure that follows. It seems to me, it's the future, world rate of return that should be the guide, but am very curious what thoughts you might have.

David Friedman said...

I find the 2% discount rate the least objectionable feature of the article. I think that's about what long term riskless real interest rates have historically been, although it's not a subject I am expert in.