Economists' mistakes on climate change

Some climate scientists have been getting mad at economists lately (see also here). As Karl Smith notes, most economic analyses of climate change tend to downplay the threat caused by climate change:

I wrote before that part of the problem for climate hawks is that even expected damages from climate change are not that large. Not zero by any means but not as high as they would like...Getting the result that the US would fall off a cliff due to climate change as projected over the next century is hard to produce.

Climate scientists, in contrast, are increasingly alarmed by the data. Who is right?

Well, I would not call myself an expert on the subject by any stretch, but I did take a class that introduced me to some well-known economic analyses of climate change. I have to say, I was not hugely impressed. The papers I read seemed to have two big limitations:

1. Backward-looking, local analysis

2. Limited models with few or no interaction terms

The first of these is basically what the climate scientists are complaining about. Empirical economists use the past to predict the future, but climate change of the type we are facing is unprecedented. All economic data comes from a time when climate change wasn't very severe. 

The problem with this is that very extreme bad events are not always simple extrapolations of mild bad events. Suppose we want to make a model of the cost of five-kilometer-long asteroid impacting the Earth. So we look in the past, and we find the average costs caused by small meteorites, and we multiply these by the ratio of the volume of these small meteorites to the really big one. If a 0.5-meter-long meteorite causes an average of $100,000 in damages, then a 5-kilometer-long meteorite will cause $12.5 trillion in damages. Not fun, but not too bad.

But this would obviously be silly talk! A small meteorite doesn't cause a very small amount of things like global crop failures, tsunamis, mega-earthquakes, supervolcano eruptions, and ice ages. A large meteorite causes large amounts of these things. In other words, there is a high degree of nonlinearity at work that just cannot be estimated from past data. Economists don't look at the tail risks of climate change - the low-probability doomsday scenarios - because we just don't have the tools to do so.

But we rarely admit this.

The second big problem with economic analyses of climate change is that the models used are very limited in scope. For example, consider this NBER working paper by respected climate-change economists Deschenes and Greenstone. It analyzes the cost of increased heatstroke deaths from higher temperatures. Or consider this paper by the same authors, published in the American Economic Review (the most prestigious economics journal) in 2007. It estimates the losses in agricultural output due to climate change. 

These are careful, serious analyses, and the authors are up-front and honest about what they are and are not trying to model. But they are nevertheless limited by the fact that each model studies only one isolated effect of climate change. Interactions between effects are just ignored. For example, suppose agricultural losses make us less able to afford the air conditioning that would prevent us dying from heatstroke. Or suppose the danger of death from heatstroke makes it too dangerous to go out and farm! 

Even worse, there are a whole plethora of possible effects of climate change - shifting coastlines, species destruction, increased migration, higher rates of disease, etc. There are possible interactions between any pair of these. There are even possible interactions between any three of these!  A realistic model of climate change costs - one you could hold up in front of a Congressional committee and say "Hey, here's what we're looking at!" - would have to take most of these into effect. But economists aren't yet into making that sort of model. (Note: if you know of exceptions, please let me know!)

As another example of this, take Karl Smith's defense of climate change economics, in which he analyzes the impact of shifting coastlines on the United States. He concludes that the costs of resettling Americans will be small. But what about lost farmland? What about disrupted supply chains and networks of trade? What about the effects of the inundation of all the coastal cities worldwide that buy the U.S.'s export products? What about the uncertainty created by not knowing where to rebuild, because we don't know how much more the seas will rise? And so on.

How OK is all of this? Ryan Avent defends climate change economics, saying that bad analysis is worse than no analysis at all:

I think it's possible—indeed, likely—that current models are failing to adequately capture the impact of climate change on the global economy. But this is how science works. You approximate reality poorly, then you learn from that and do a better job, and then you do a better job still.

But what exactly are we learning from these limited, flawed analyses? Does making a backward-looking, linearized model of one "isolated" effect of climate change help us make a better model in the future? Well, yes, if by "in the future" you mean "after climate change has already run its course, and we can see how wrong those limited models were." Because that is how science works - you test your models against the data. Lowball estimates of climate change costs won't get a rigorous test until it's too late.

Ryan continues:

Environmentalists and economists are fundamentally on the same side, supporting the use of data and the scientific method to reach reasonable, peer-reviewed conclusions and appropriate policy recommendations.

That may be true (at least for those of us who are not political shills). But it seems to me that economists could improve their backward-looking models by incorporating forward-looking climate science. We could loudly reiterate how little they (or any of us) know about the doomsday "tail risks" of climate change. And we could try to make more comprehensive models.