The Sanders Economic Analysis Flap

Not often is a single economics paper propelled to the front of the news cycle, but this is what happened to Gerry Friedman’s analysis of Sandernomics.  The bottom line as I see it is that, while Friedman was doing his best (as he saw it) to promote his candidate, he ended up harming him.  The ultimate effect will not be too great, assuming most voters forget what happened more than a week ago, and the Sanders campaign pivots a bit to limit the damage.  Nevertheless, I think there are lessons to be learned, especially for economists and similar researchers.

I suspect that what we witnessed was a sort of Rosie Ruiz moment.  Rosie Ruiz, you’ll remember, was a woman who entered the New York and Boston Marathons in 1980, leaving the routes early to hop on a subway, and then rejoining each race just before the finish line.  No one might ever have noticed, except she placed first among women in Boston by a wide margin, and then the scrutiny began.  Eyewitness reports confirmed that she had not actually completed either race, and the result is that she was stripped of her records.

Now, I’m not saying Friedman was cheating or violating any other academic rules, but he clearly cut several corners, not anticipating how the spotlight would come down on him once his work was disseminated.  My guess is that he thought the most important thing was to get his results to the public before the next round of primaries, pulled lots of late-nighters and didn’t stop to think about what would happen when he would suddenly become a focus of national attention.

So what were some of those corners?

1. He never presented his model.  The appendix to his report jumps immediately to parameter estimates, but there is no list of all parameters nor a formal model displaying how they relate to one another.  I take it that the implicit model calculates GDP growth from spending projections subject to a multiplier, and that this translates into labor demand with productivity as a residual.  The microeconomic results are determined by macro outcomes plus additional sector-specific factors.  There does not appear to be a simultaneous relationship between macro and micro (especially labor market) outcomes, which is a cause for concern.  Moreover, there is no discussion of the history of his model: who else has used it and what its track record has been.  I can imagine that these are the sorts of things one would put aside if one were in a race to publish, but Friedman should have expected that academic economists would savage him for this.

2. His estimate of the national income multiplier is suspect.  He gives a verbal description of how he calculated the multiplier, but it’s not entirely clear: “I assume the multiplier is two in first-quarter of 2009 and then falls by 20 times the reduction in the output gap.”  Can you write that equation?  But what makes it really suspect is that the multiplier remains close to 1 (.87) even at the end of the analysis horizon in 2026 after a decade of unprecedented economic growth.  It leaves the impression that Sanders can simply expand national income almost indefinitely by perpetual increases in spending.  It is the multiplier assumption, in conjunction with the structure of his model (whatever it is), that gives Friedman shockingly large predictions for income, employment and productivity gains.

3. There is no sensitivity analysis.

4. He apparently ended his work as soon as he generated his results.  This is an instructive mistake, in my opinion.  When I teach statistics and research methods, I always emphasize that, while models produce results, results test models.  Look at your output.  If it appears to be unreasonable you better be able, based on your model structure, to tell a plausible story that justifies it.  Otherwise you have evidence of a flaw in the model itself.  The highly unlikely predictions for GDP, employment and productivity should have sent Friedman back to the drawing board to see if one or more of his assumptions might be questionable.  Of course, this describes an iterative process that would make it difficult to meet what he may have believed were his political deadlines.

5. He didn’t circulate his results to skeptics.  The people you feel closest to are the most likely candidates to be reviewers, but the feedback you actually need is from those whose bias is to reject you and your work.  Friedman knew he was going to release his report to be devoured by a pack of wolves, so having an honest wolf or two provide a pre-release assessment would have been a good idea.  His list of acknowledgments does not suggest he did this.  Granted, it is not always possible to find a critic willing to play this role, but I’m pretty sure that there are Clinton partisans among economists in Gerry’s wider circle that he could have drawn on.

I realize I’m being hard on Gerry in this post mortem.  I know him a tiny bit (just a few moments here and there), and he seems like a very nice guy.  But I think he lost track of the larger picture and probably now regrets some of his choices.  The rest of us can learn from his experience.

Incidentally, for the record, I don’t for a moment believe I could have done better myself.  First, I’m not a macro guy, so I wouldn’t take on the job in the first place.  Second, the careful, comprehensive, cross-checking approach I’ve described is beyond the capacity of any one person within a short timespan.  If the Sanders people needed an economic analysis asap, before South Carolina and Nevada, they should have assembled a team to carry it out.

And as for the Sanders campaign, they should thank Gerry for his efforts and then say that, given its results, his analysis represents the upper limit of what Sanders can accomplish, and they welcome more analyses providing a range of predictions.  As most commentators have pointed out, the attractiveness of the Sanders program is not based on its performance in economic models, which are unreliable predictors in any event.  It is enough to show that the program is internally consistent and not vulnerable to large downside risks.  More finely tuned analysis can wait for the aftermath of the election, when it will be time for drafting policies in detail.

UPDATE: I’ve been taken to task for assuming that Friedman is a Sanders supporter, which I inferred from his work with the Sanders campaign on health care reform finance, as well as the flashy headline numbers from his latest study.  If he is actually for Clinton I can’t understand why he seemed to be in such a hurry to get his work out.

I’ve also been directed to Josh Mason’s defense of full-bore stimulus to get the economy back to full employment and potential output.  Overall, I agree with Josh, but I still think the extremely rapid rates of growth, year after year, for a full decade are problematic.  Our current macro situation calls for expansionary policy, but there is no precedent in a developed economy for this sort of velocity over this duration.  After such a massive downturn followed by many years of no rebound, it’s not plausible that the US can return to the pre-2007 trend in less than a decade.

To repeat (just so it’s clear): on the policy point, Mason is entirely correct.  The right thing to do is to apply large amounts of fiscal and monetary stimulus as long as incomes and employment are depressed or until these measures show signs of being counterproductive for some other reason.