Quite a lot really. Unless, of course, you’re looking at the wrong models in the wrong way. As Robert S. Pindyck does. I do have some sympathy for the paper, but its badly written, somewhat confused, and the author has failed to emphasise some key distinctions.
To begin with where I agree, I’m fairly happy with his assertion that “certain inputs (e.g. the discount rate) are arbitrary, but have huge effects on the [social cost of carbon] estimates”. I’m only “fairly” happy, because to say that the discount rate is “arbitrary” is stupid (which is probably a hint that this thing hasn’t been peer-reviewed; its only a working paper); what he means is, that people of good faith can nonetheless disagree about what the correct value should be. And it is certainly true that how much you think future climate damage should be costed now depends rather heavily on the discount rate. However, this is just the bleedin’ obvious, so he gets no points for that. He also whinges about the uselessness of the models, but fails to realise that for exploring the impacts of different discount rates, or different damage functions, they are quite illuminating.
His paper is about “integrated assessment models” (IAMs) – the things used to turn parametrised versions of climate change into damage estimates. I know little about them, so I’m not going to say much about that bit. Of course his title doesn’t specify that. And while his language in the abstract when he says the models’ descriptions of the impact of climate change are completely ad hoc, with no theoretical or empirical foundation is probably defensible, you have to read it fairly carefully to realise he is talking about the “damage functions” bit there – he isn’t talking about the physical-climate-bit at all in the abstract. Later on he is more careless:
There are two types of inputs that lend themselves to arbitrary choices. The ﬁrst is the social welfare (utility) function and related parameters needed to value and compare current and future gains and losses from abatement. The second is the set of functional forms and related parameters that determine the response of temperature to changing CO2e concentrations and (especially) the economic impact of rising temperatures.
The arbitrariness of “(especially) the economic impact of rising temperatures” may or may not be correct – like I say, that’s not my thing; though I know enough that damage isn’t just temperature, it ought to be precip too, at least. But he’s quite wrong to say that “the set of functional forms and related parameters that determine the response of temperature to changing CO2e concentrations” is arbitrary. That’s just ignorance on his part.
He does talk about the physical-science-bit, wrapping the discussion around climate sensitivity, which is what it all boils down to at this level. But I don’t think he knows what he is talking about; and instead he just talks up the uncertainties. What he should have written for this section (prefixed with “I know nowt about CS…” is “Compared to the uncertainties elsewhere, the uncertainty in CS is small and we may as well assume 3 oC. Or 2 oC. Or somewhere like that”. Instead he repeats the Roe/Baker heresy even down to the Gaussian bit which I pointed out was wrong ages ago.
mt will be happy that he does point out that the IAM models intrinsically don’t deal with “catastrophic outcomes” well, since they’re built to be smooth, i.e. they’re built not to have any catastrophic outcomes. So he gets a point for that.
What to Do?
Well, don’t read his paper or the “What to Do?” since it doesn’t really go anywhere. Instead, you need to think like Karl Popper (I think; though I could easily have the wrong guy; lets call him “KP” for now) who was talking about building political systems. And laying in to the likes of Plato, who designed their systems around philosopher-kings, who by their very nature were wise and good (even if they were instructed to lie to the populace; but that’s another matter). KP tried to say No; we should not design our systems on the basis that our rulers will be Good; on the contrary, we should design them to survive rulers who are Bad.
Similarly, the answer to “oh noes, our IAM models suffer from unknown damage functions and uncertain discount rates” is not “fiddle with the models” or “select simpler more subjective models”; its to design models that work even if you don’t know such things. Like – aha, you guessed it, I know – a carbon tax that we can ramp up or down slowly as needed.