"System change is now inevitable. Either because we do something about it, or because we will be hit by climate change. '...

"We need to develop economic models that are fit for purpose. The current economic frameworks, the ones that dominate our governments, these frameworks... the current economic frameworks, the neoclassical, the market frameworks, can deal with small changes. It can tell you the difference, if a sock company puts up the price of socks, what the demand for socks will be. It cannot tell you about the sorts of system level changes we are talking about here. We would not use an understanding of laminar flow in fluid dynamics to understand turbulent flow. So why is it we are using marginal economics, small incremental change economics, to understand system level changes?"

Saturday, October 24, 2015

IAMs Cannot Be Fixed

Also in re IAMs from David Roberts's Vox article, and rather a separate point:
There are tons of other interesting results buried in this monster paper — for instance, out of various parameters, uncertainty about future productivity growth has by far the largest implications for outcomes, "which suggests that uncertainty in GDP growth dominates the uncertainty in emissions"
Please note - GDP growth is a parameter of IAMs; it's not an output, it's an input. This alone makes them unfit for purpose, as I determined independently, but David wrote up before I got around to it.
Right now, the most frequently used climate IAMs treat economic growth as mostly or entirely exogenous. Here’s a very brief explanation of endogenous vs. exogenous variables in economic models:

Some economic variables are determined by our models, while others are usually assumed to be determined by factors outside of our models. We call the former endogenous variables and the latter exogenous variables.

For econometric applications, the crucial difference between an endogenous and an exogenous variable is that we must assume that exogenous variables are not systematically affected by changes in the other variables of the model, especially by changes in the endogenous variables.

In other words, an exogenous variable is something you plug into the model, not an outcome of the dynamics within the model.

So what does it mean for GDP growth to be treated as exogenous in economic models? It means that growth cannot be systematically affected by endogenous variables like, say, temperature — by definition, in these models, climate impacts cannot affect the rate of GDP growth. Climate impacts might subtract something from total economic output, but they do not alter the rate or trajectory of growth.
This doesn't mean that complacent economic models are better than disturbing economic models nor the other way around. It means that this approach is of no value whatsoever in assessing long term climate risk. At all. Period.

Now people have finally got around to realizing this and are plugging in a fveedback from climate to the growth parameter. If course, they aren't going to publish the results. It's not hard to imagine the rosy models turning collapsitarian with this minor tweak. That doesn't make them right, of course. Presumably an approach with such a fundamental flaw as to tread GDP as a parameter is pretty flawed in other ways.

It's funny how people who criticize climate scientists tend to be such empiricists, obsessing over every bump and wiggle in every chart and using each little excursion as an excuse to dismiss a fairly mature body of physical theory. But when it comes to economics, they are utterly happy with the economists utter indifference to data collection and their silly handwaving predictions.

If our circumstances weren't so dire it would be funny.


Jonathan Koomey said...

Yes, exogenous GDP projections assume that climate will have no effect on the economy, but we know that the "current trends continue path" will interfere with the orderly development of civilization by the end of this century (and probably sooner). And that means the modeling results are as good as gibberish (although there are some insights that can still be gleaned from them if you are careful and knowledgeable).

For more on what to do instead of trying to calculate an optimal economic path decades hence, see my open access article in ERL:

Koomey, Jonathan. 2013. "Moving Beyond Benefit-Cost Analysis of Climate Change." Environmental Research Letters. vol. 8, no. 041005. December 2. [http://iopscience.iop.org/1748-9326/8/4/041005/]

Also see

Koomey, Jonathan G. 2012. Cold Cash, Cool Climate: Science-Based Advice for Ecological Entrepreneurs. Burlingame, CA: Analytics Press. [http://www.analyticspress.com/cccc.html]

Steve Bloom said...

Michael, is your "Now people" paragraph referring to Gillingham et al. (public copy here, BTW)?

Let us note who isn't a co-author of that paper. I see that Tol has now attached himself to the Helix Climate project, maybe because they were taking anyone who wanted to volunteer. It will be interesting to see how that blows up in their faces.

Amusing to see Stoat defending economists, but then again he did buy Northern Rock stock after the initial fall.

Thanks for the pointer to your paper, Jonathan, which I will read. The abstract raises a conceptual objection from me, which is that I don't think it's reasonable for economic modeling to shoot for a warming target, rather it needs to be an emissions one, with GCMs informing us, however poorly, as to the linkage between the two.

Michael, when you referred in the previous post to where the fat tail might be found, was that about unknown/unquantified feedbacks? If so note this Beeb piece, in which Vladimir Romanovsky pulls the plug on the Schuur et al. projection published only six months ago. I was expecting this to happen given the notional trend of such things going back ten years or so (Bayes is my friend), but had thought it would take until next year. OTOH it's probably not fair to count Romanovsky's remarks against Schuur et al. until they're turned into a paper, so maybe the year is still good.

Jonathan Koomey said...


Thanks for your comments. I do understand the complexities of converting a warming limit to an emissions pathway, and agree that it isn't simple. My response is not to abandon the warming limit but instead to adopt an evolutionary frame, in which we use the warming limit to develop emissions pathways, and then track the emissions and warming in each year needed to meet that warming limit. Then we use cost effectiveness analysis to determine where to start reducing emissions. If we hit the emissions target but warming appears to be going faster than the model runs indicated, we'd need to ramp up emissions reductions.

The warming limit implies the needed emissions pathway, however imprecisely, and instead of accepting these as gospel, we need to adapt as measurements dictate.

Happy to chat more about this and hear your thoughts.


Michael Tobis said...

Jonathan, you are always welcome to do a guest post here and I can probably get you a guest post on ATTP which gets more traffic these days.

Jonathan Koomey said...


Many thanks for this offer, and for your thoughtful commentary. Let me think about what I can contribute that's new and different, and then we can discuss.


William Connolley said...

> GDP growth is a parameter of IAMs; it's not an output, it's an input. This alone makes them unfit for purpose

Compare: "CO2 growth is a parameter of AOGCMS..."

William Connolley said...

> It means that this approach is of no value whatsoever in assessing long term climate risk. At all. Period.

Don't understand that. Why not?

> Amusing to see Stoat defending economists, but then again he did buy Northern Rock stock after the initial fall.

You're making things up. I have no NR shares.

Michael Tobis said...

"Compare: "CO2 growth is a parameter of AOGCMS..."

Not a terrible assignment, but I'd hope you would know the answer.

An AOGCM is a simulation, with climate-like behaviors of a system emerging from forst principles. An IAM is a back of the envelope generalization. The word "model" has entirely different meanings.

An AOGCM is tested against enormous amounts of data, both in statistical and nowadays in prognostic modes. An IAM leaves room to argue about paramaeters but no methodology for testing or constraining them.

As if that weren't enough, the key question to which an IAM is put is the prognostication of how climate impacts affect growth, and yet a mysterious quantity called "underlying growth" is generally put in as a number at the beginning. The analogy to AOGCMs fails because while CO2 concentration trajectories are ordinarily specified, it is not some simple function of CO2 concentration that is the intended output. IAMs simply beg the question to which they are put.

I know of numerous applications of AOGCMs. I know of only one for climate IAMs, which is to establish a social cost of carbon. They are not suited for purpose, because the presumption of underlying growth persisting through the period simply begs the question as to how big a problem the carbon presents.

Steve Bloom said...

Tch, made me go dig it out:

"he did buy Northern Rock stock"

"You're making things up. I have no NR shares"

"In order to prove my financial acuity, when Northern Rock fell to about 200p and the govt guaranteed its deposits, I bought £250 worth, believing it would bounce back. Its now down to 90p, and the news now is that its to be nationalised. My shares are now worth so little I hardly care, but this looks to be a disaster in the making, or rather in the continuing."

Note my use of past tense. Presumably the nationalization subsequently erased them, but my statement remains technically and substantively correct.

But I do agree with William that GCMs and such could be (and should be) subject to vetting similar to the IAMs (although Michael is right that they'll look a lot better).

"If we hit the emissions target but warming appears to be going faster than the model runs indicated, we'd need to ramp up emissions reductions."

Right, Jonathan, but the climate response lag to increased GHGs means that to be safe we need to project feedbacks into the future even though we may have insufficient data/physics with which to do so. I realize this amounts to educated guessing, but that's the hand we've dealt ourselves.

andthentheresphysics said...

I also found that GDP growth is an input rather surprising and it's hard to see how it's somehow equivalent to CO2 growth in an AOGCM (well, unless IAM results are presented as "if GDP growth is...."). As I understand it, we have much more direct control over our future CO2 emissions, than we do over our future GDP growth.

Also, as most here probably know, a simple way to determine how long it will take for something to double - given a certain growth rate - is to divide the annual percentage growth rate into 70. So, GDP growth of 2% means that the world economy will double in 35 years. Hence, if we assume that GDP growth is at least 2% pa, we'll be richer in 35 years - than today - unless climate change can do damage that is equivalent to the world economy today (at least, I think that's right). This sounds implausible, so any assumption that GDP growth can be sustained would seem to essentially guarantee that we'll continue to get richer, for the foreseeable future anyway. On the other hand, if climate change were to influence GDP growth, then we could be considerable poorer than we would be in its absence. It would seem that to be able to compare different possible pathways is more relevant than simply trying to determine the impact of CC under the assumption of some kind constant underlying GDP growth.

William Connolley said...

My apologies; you're correct; I did say that. In my defense that was 2008; I don't recall trivia back that far.

Michael Tobis said...

My sympathies; I have Fuller to remember such stuff for me.

Michael Tobis said...

To be fair, the parameter is "underlying GDP rate"; the purpose of the IAM is, to the extent I can tell, to determine to what extent the "underlying" rate is "modified" by the climate. Essentially it is a perturbation model which linearizes the effect.

As people in real sciences know, you can't rely on a perturbation model's results until you prove that the perturbation is small.

For all the criticisms of climate models missing feedbacks that aren't missing, it seems that IAMs lack this first order feedback in the key property of interest, and I have seen no effort made extrinsic to the model to prove the small-perturbation approximation. Further, I doubt that it can be proved - that's back on my turf after all, and I can think of lots of ways that climate disruption could disrupt the growth economy.

Then there's two other key problems. 1) That a growth economy is indefinitely sustainable in a meaningful way and 2) that the purpose of policy is to optimize for growth. But even resisting the constant temptation to challenge these core assumptions, the IAM approach is undefended and probably indefensible.

There are plenty of other reasons to suspect that IAMs are so crude compared to the complexity of the underlying system (would you put much trust a GCM that runs in Excel and takes a single time step to 2100 AD?) that they are utterly uninformative, but there's no reason to dig into the weeds.

Unless my understanding is fundamentally wrong, these things are an embarrassment to the academy. They are evidence on the side of the Ridleys of the world that want the universities to be trade schools on the utilitarian grounds that they don't seem able to achieve much in the way of research.

Jonathan Koomey said...


Your understanding is correct, based on what I know. These and other issues have been laid out in some of the foundational critiques of the IAMs, but that community is very insular and hasn't taken these critiques to heart. Here are a few key references:

Ackerman, Frank , Stephen J. DeCanio, Richard B. Howarth, and Kristen Sheeran. 2009. "Limitations of Integrated Assessment Models of Climate Change." Climatic Change. vol. 95, no. 3-4. August. pp. 297-315. [http://link.springer.com/article/10.1007%2Fs10584-009-9570-x]

DeCanio, Stephen J. 2003. Economic Models of Climate Change: A Critique. Basingstoke, UK: Palgrave-Macmillan. [http://amzn.to/1wvkvDu]

Rosen, Richard A., and Edeltraud Guenther. 2015. "The economics of mitigating climate change: What can we know?" Technological Forecasting and Social Change. vol. 91, no. 0. 2//. pp. 93-106. [http://www.sciencedirect.com/science/article/pii/S0040162514000468]

Steve Bloom said...

Permafrost feedback not missing from the models? I'm apeaking here of GCMs; I imagine some ESMs and such may have it, although how realistically? Can you point to a paper or to where AR5 includes it? AIUI Schuur et al. (2015) was the first modelable estimate. Do you have other information?

To mention another, let's not forget all these fires going on.

I remembered it because it was *amusing* trivia, William. Michael should be so lucky for that to be the whole of what Fuller does.

Michael Tobis said...

Permafrost, forest feedbacks etc. are not part of the Charney system, and is easily folded back into the forcing. That is, carbon cycle coupling does not belong in a climate model properly construed.

I am not an enthusiast of coupled GCM/carbon cycle models (ESMs). I think they take brute force computing far too far. In my opinion doing so is make-work for the supercomputing industry, not real science, at the very least pending dramatically improved understanding of the carbon cycle.

Problems which can be separated intellectually should not be coupled computationally just to maximize the sheer cussedness of the resulting code. There's a place for the scientific mind in science. Hamming's law (search on "purpose of computing") applies.

Michael Tobis said...

"We have an economy where we steal the future, sell it in the present, and call it G.D.P." -Paul Hawken (via Alex Steffen)

andthentheresphysics said...

So, if IAMs really are simply linear perturbation calculations, doesn't that immediately imply that their underlying assumption has to be that the perturbations are small?

Michael Tobis said...

ATTP - yep, that's it in a nutshell.

Most economists are really very naive about mathematics.

Michael Tobis said...

"While both temperature sensitivity and the damage function coefficient are very important for the welfare effects of unmitigated climate change, there is much more evidence on values for the former (Ackerman et al 2010:1662). Temperature sensitivity has been the subject of a large amount of empirical research, with the International Panel on Climate Change’s (IPCC’s) (2007b) review of estimates finding a mostly likely value of 3 degrees Celsius, and a two-thirds probability of lying between 2 and 4.5 degrees. Because of uncertainty about feedbacks in the climate system, all of the 18 probability density functions reviewed have a positive skew or ‘fat right tail’ (Dietz and Asheim 2010:10). In contrast, the relationship between temperature changes and damages for large increases in global average temperature is arguably an area of radical uncertainty, depending on the response of biophysical systems to warming and second-round socio-economic responses to climate change impacts, all of which are extremely uncertain. In the Stern Review, PAGE draws an exponent for the function from a triangular distribution with a minimum of 1, mode of 1.3 and a maximum of 3, which provides small chances of damages increasing very rapidly with changes in temperature (Dietz et al 2007a:314). DICE uses a quadratic damage function, but the justification for quadratic damages is scant.21 While this functional form may be satisfactory for small temperature increases, its implications for large temperature changes are unconvincing. In particular, Ackerman et al and Weitzman independently show that, with quadratic damages, only around half of global output is lost when global average temperatures reach 19 degrees Celsius above pre-industrial levels (Ackerman et al 2010:1660, Weitzman 2010:14). Weitzman’s (2010:8) discussion of the potential for a range of catastrophic impacts associated with much lower temperature increases provides a good illustration of why this result, and the damage function that produced it, seem extremely conservative. "


I believe that this "50% loss" under 19 C warming is 50% loss superimposed on the underlying growth rate of some 3% compounded annually, so it's still a growth economy, just not as big as it would otherwise have been. But I'm not sure of that.

I now have to read that link as well as


Steve Bloom said...

So Michael, 'splain to me: How, if things are so easily folded back, was it possible for Schuur et al. to write very recently:

"Field observations reveal that abrupt thaw processes are common
in northern landscapes, but our review shows that mechanisms that speed
thawing of frozen ground and release of permafrost carbon are entirely
absent from the large-scale models used to predict the rate of climate change."

So the GCMs aren't at all up to snuff.

"Climate warming as a result of human activities causes northern regions
to emit additional greenhouse gases to the atmosphere, representing a
feedback that will probably make climate change happen faster than is
currently projected by Earth System models."

Neither are the ESMs.

"The Earth system models analysed for the IPCC AR51 did not include permafrost
carbon emissions, and there is a need for the next assessment to make
substantive progress analysing this climate feedback."

There was absolutely no reason for the AR5 to ignore this (albeit that less exact figures than those of Schuur et al. would have had to have been used), other than that it makes it obvious that 2C is impossible. Playing politics? Maybe we'll see something *after* Paris.

Michael Tobis said...


Inaction and climate stabilization uncertainties lead to severe economic risks

Martha P. Butler · et al


critiques how IAMs are used.