Sims and Sargent were announced as winners of the Nobel Prize in economics on Monday. I've been a heavy user of vector autoregression models in my career including this recent paper, which discusses some of the issues that Sims discussed in his 1980 paper "Macroeconomics and Reality". I met Sims once at Princeton. I visited the department for a few days to work with David Bradford. I gave a presentation on applying time series econometrics to climate change, which he attended. That was a bit scary :)
BTW, this is also the 500th post on Stochastic Trend!
Well done! But the link to your latest VAR paper did not work for me. I'd also like to see the MS of your own paper on econometrics & CC at Princeton.
ReplyDeleteTim - I fixed the link - thanks for pointing it out. I'll see if I can dig out my slides from 10 years ago and send to you. There was no formal paper as such. But we published a couple of papers either side of my visit there (Climatic Change, 2000 and JGR, 2002). I remember that a big focus of my talk was about carbon absorption...
ReplyDeleteMany thanks. With not much text, your Princeton paper is difficult to follow, but looks good. Your then under review JGR paper with Kaufmann is much more accessible. But I do have a few problems with it.
ReplyDeleteFor example, I am not sure why you consider emissions of CO2 and SO2 as stochastic/random walks. They are rather deterministic, I would have thought.
That explains why there have been amazing reductions in SO2 and NOx emissions in USA since 1990 (see my comments on Muller, Mendelsohn, and Nordhaus AER 2011 below).
You are right that total flows of CO2 to and from the atmosphere are about equal - at about 200 GtCO2 - but that means the average residence time is much less than 100 years. In fact Houghton puts it at only 4 years (2004:31, also TAR 2001). That also means that the total flux up is about 7-8 orders of magnitude larger than the anthropogenic emissions (not merely 2 as RK and DS say).
Then I am very doubtful that the astro-chemistry of radiative forcing involves lagged variables. The mix on any given day is determinate and separate from the radiative forcing of the previous day. Temperature itself does exhibit some stochastic properties, but they are adequately dealt with by first-differencing, as is indeed implicit in the term climate change.
So I do not agree with K&S 2002 when you state "differencing the time series is inappropriate because it removes the long-run driving trends…." (para.16.). Radiative forcing is unique to time and place given the forcing factors and so is deterministic not stochastic.
In Fig 2 of K&S you have temperatures for the SH from 1860. Amazing, when there was only a handful of met stations there until 1910, with just one in all of SH Africa.
Finally, I would have liked to see atmospheric water vapour brought to account. A particular batch of evaporation may not be "long-lived" but there always is [H2O] and it is not a constant across the globe either spatially or temporally.
All the same a very impressive paper, should get you a Nobel some day!
Tim