Warwick McKibbin has a piece in today’s Fin (subscription required) endorsing the Castles-Henderson critique of the modelling behind IPCC projections of CO2 emissions. He refers back to a paper, with David Pearce and Alison Stegman, for the Lowy Institute (PDF file). I’ve read the paper and I think it’s broadly correct. On the other hand, I’ve previously argued that the Castles-Henderson critique is invalid (see also here). So what gives?
Castles and Henderson claim that the use of market exchange rates, rather than Purchasing Power Parity (PPP) exchange rates to compare national incomes renders the IPCC results totally unbelievable. Terms like “fantastic assumptions”, “astronomical” and “extraordinary” abound. Although there’s not a lot of detail, the general impression given is that the IPCC projections are out by a factor of at least three and probably more.
By contrast, I’ve argued that at the aggregate level, it shouldn’t matter whether you use exchange-rates or PPPs. All that matters is demand for energy, and effects arising from the choice of exchange rate will cancel out.
McKibbin uses a disaggregated computable general equilbrium model. In such a model you’d expect changing the basis of exchange rates to have some effect on the predicted outcomes but not much. This is indeed what happens. As the paper states (p6)
We find that by 2050 the projection of emissions from fossil fuels use based on the MER measures of GDP gaps is 22% higher than our base projection (using PPP) and by 2100, projected emissions are 40% higher than baseline emissions. About half of the higher emissions are generated from countries that are classed as developing in 2002 and about half from industrial economies. These numbers are almost 3 times those found in Manne and Richels (2003) who undertake a similar exercise. There are a number of reasons for these differences, which are open to debate.
In the context of a projection to 2050, a variation of 22 per cent is insignificant. It’s far smaller than the range of variation within the IPCC estimates. The same is true in spades for a 40 per cent error in a prediction for 2100.
Moreover, as the comparison with Manne and Richels shows, the effect found by McKibbin et al is highly fragile. Almost certainly, there are plausible variants of the model in which using PPPs produces a higher estimate of emissions than using market rates.
The correct interpretation of the McKibbin et al and Manne and Richels studies is that, within the range of precision that any estimate of emissions in 2050 can have, it doesn’t matter much whether you use PPP or market exchange rates. This is, of course, directly opposite to what is claimed by Castles and Henderson.
Having said all this, I agree with McKibbin that the IPCC models were not done in the way an economist would like, and didn’t add a lot to our understanding of the economic issues involved in climate change. But this is no justification for endorsing the clearly erroneous claims made by Castles and Henderson.
John Quiggin 
John,
I find your comments on SRES a little odd. If I told you a carbon tax could reduce emissions by 20% by 2050 you would argue that it shouldn’t be used because the margin of error in our forecasts by 2050 is so large that 20% doesn’t matter – or if I could find a medical treatment that reduced the number of people who die by 2050 by 20%, this should not be tried because the margin of error in our population forecasts of 2050 are much larger that 20%. You clearly believe that nothing should be done today to reduce emissions because given the uncertainty about the future, every policy will be less than the forecast error of 2050? This is the dangerous proposition usually advocated by the “do nothing” lobby which I am surprised you endorse.
Secondly you have totally missed the point about the role of PPP versus MER in conditioning growth projections IN A CONVERGENCE MODEL. It is not just a numeraire issue (which is also the arguement of the SRES authors who also don’t seem to understand the growth convergence literature which they nonetheless claim to use in some scenarios). The proposition is simple – if you assume a larger gap in incomes per capita and you assume that the gap closes at some rate, then you need to grow faster in real terms to close the gap. Faster growth means higher energy use which means higher emissions (taking other things as given). More details can be found in my paper which you claim to agree with.
I thought the Castles critique ran as follows. The assumption is that convergence will occur. If PPP adjusted incomes are used then less measured convergence needs to occur so the effects of income growth on the demand for energy in developing countries will be less. Hence there will be a lower growth in emissions.
Maybe some wrong assumptions here but, given the assumptions, nothing wrong vabout the conclusions.
Warwick and Harry,
Obviously, I need to restate my point, so I’ll do it in two ways
1. The main variable of interest is energy use. In any consistent model framework, if incomes converge, so (approximately) should energy use. So it doesn’t matter what relative prices you use to summarise aggregate income – PPP or market exchange rates.
2. Using market rather than PPP exchange rates implies that poor countries need to grow faster to catch up than rich ones. But to fit observed patterns of energy use, it must also imply a lower income-elasticity of energy demand. These two effects should cancel out.
In a logically inconsistent model, for example one where incomes were modelled using market rates, but energy demand was modelled using PPP rates, the outcome would be that, when incomes converged, formerly poor countries would have much higher energy demand [per person] than rich ones. This doesn’t happen in the IPCC models, or in the McKibbin et al model.
In fact, the McKibbin et al model gets higher emissions in both sets of countries under MER. This is totally inconsistent with the Castles-Henderson critique.
Note: In the original version of the post, I wrote PPP in the para above for MER. This was an error.
John,
Actually no. Your first point is like arguing that the number of beaches per capita should converge everywhere so that people in Bolivia will have the same beaches per capita as people in Australia. Australians are endowed with beaches whereas Bolivians are not. We have coal endowments which other countries do not (as well longer distances to drive different heating/cooling requirements etc i.e. countries have dirrent consumption patterns as well as different endowments of carbon base energy). Coal is partly tradeable but transport costs makes it more efficient for Australia to embody this carbon in our exported goods hence we have larger emissions per capita than others. Our emissions are based on this endowment. In fact in the data there is not a fixed relationship between energy use and GDP nor between carbon emissions and energy use (the point of the first part of our paper). The SRES people have models that assume there is a fixed relationship(as you seem to with your back of the envelope). This is rejectable when you estimate substitution elasticities which we do in our modeling research but others don’t – they just assert.
On the second comment, you have misread our results. Our benchmark income gaps are measured in PPP. We impose the MER mistake in our counterfactual which shows emissions higher. Not sure where you got your last comment from but it wasn’t from the paper? We don’t say that the SRES is wrong nor that everything Castles and Henderson say is absolute truth – we argue that the Castle’s and Henderson point on emissions projections if convergence is used is correct in our model and we see why it might be rejected in other models but that is by assumption in those models.
Warwick,
It’s obviously right that you shouldn’t get exact convergence. But if the Castles-Henderson critique were right the overestimation of growth in currently less-developed countries should produce an overestimate of energy demand in those countries so that, when income convergence is achieved, those countries have systematically higher energy use than countries that are currently developed. This doesn’t happen in SRES or, as far as I can see, in your model (in the case when MER is used instead of PPP).
I apologise for mis-stating your results in my last para, saying PPP when I meant MER. I’ve corrected this and noted the change.
My point is that if the Castles-Henderson critique were valid, use of exchange rates would understate less-developed country income [implying higher growth to reach convergence] and overstate developed country income [implying lower growth]. So a switch from PPP to MER should produce estimates with higher energy use for less-developed countries, but the same or lower energy use for developed countries.
John,
There are two issues here. Once is the point you make which involved the initial carbon intensity of GDP which of course will change depending on whether you measure GDP in terms of PPP or MER (emissions are what are measured independently of this measurement issue related to GDP). The other involves the profiles of emissions over time as a result of this effect plus the growth differentials. I agree with your point that the initial carbon intensity changes (i.e. China really doesn’t look that energy inefficient if measured in terms of PPP). We adjust for this in the experiment. The result is that the net effect of the re normalization PLUS the different emission paths due to change in relative prices due to changes in growth rates etc is what we calculate. The net effect calculated allows for the normalization issues (which is why we build empirically base models rather than using back of envelope calculations with restrictive assumptions!). I believe 20% higher emission in MER terms versus PPP terms by 2050 is non trivial. This may not happen in other models. It is smaller in the Manne and Richels MERGE model but that model only has trade in oil and gas plus an aggregate good and no flows of capital so that higher growth in developing countries tends not to raise global incomes.
It’s obviously better to have a detailed empirical model rather than a back-of-the-envelope aggregate estimate. But it’s generally true that an economically consistent back-of-the-envelope estimate will match the corresponding detailed model to within 20 per cent or so.
My back-of-the-envelope aggregate estimate was that changing PPP for MER should make no difference. This matches your 2050 estimates, and those of Manne and Richels to within 20 per cent.
Castles and Henderson also presented a back-of-the-envelope arguments, suggesting that changing PPP for MER should make a huge (though not precisely specified) difference, to the point where the MER estimates were “astronomical”, “extraordinary” and so forth. Clearly, the detailed modelling results don’t show this, or anything like it.
I agree that detailed model estimates are better than back-of-the-envelope, but the most important thing is to get the economics right, and the assumptions logically consistent, in the first place. Castles and Henderson did not do this, and came up with nonsense as a result.