Over the past few weeks, the Ivory Tower wing of the “kick the can consensus” has been busy sharing their upbeat interpretations of the CBO’s latest debt projections. Paul Krugman wasted no time declaring that the projections were “distinctly not alarming” in an article that ended with the assertion that “kicking the can down the road … is the responsible thing to do.” And Krugman’s not alone. Anyone scanning op-eds and blogs has seen a story that the CBO gave an all clear for more fiscal stimulus, or at least delayed restraint. But Brad Delong’s approach was the most detailed of those that I came across.
He published an article on Seeking Alpha last week with essentially three components:
- A graph showing the changes in the CBO’s long-term projections for potential GDP and attributing those changes to “diminished risk taking, slack investment and unemployed workers losing and not acquiring skills.” Implying these are the obvious reasons, he asks: “What else could it be?”
- A conclusion that any fiscal tightening would make our long-run deficit and debt situation worse because the CBO’s potential GDP projections imply an eta of 0.22. If you just went, “Huh???” then hang on, I’ll explain in a minute.
- A long excerpt from chapter 24 of John Maynard Keynes’s classic text, The General Theory of Employment, Interest, and Money.
In my reply below, I’ll suggest that Delong’s interpretation of the CBO’s projections is incorrect. I’ll also argue that he’s using a model that’s fundamentally flawed. And then I’ll conclude with some general observations about economic research and modeling.
Interpreting the CBO’s projections: Keynes vs. Keynes
Here’s the graph that Delong includes in his article:
He focuses on the two colored lines, which show the CBO’s 2007 and 2013 projections for potential GDP. He attributes the gap between the lines to the real economic factors that I listed in (1) above, and this attribution is a critical step in his thesis. It leads him to conclude that fiscal tightening would worsen the long-term debt picture. He further implies that there’s no other plausible explanation for the decline in the CBO’s projections.
I disagree. A more believable explanation is that the CBO realized it was wrong about potential growth in 2007, just as the rest of the world realized it was wrong. All of the optimistic themes of the day, such as the Great Moderation and the unlimited supply of global liquidity, were proven false. And everyone who makes forecasts ratcheted them down when they realized their errors, in hindsight. This includes forecasters at the CBO and virtually everywhere else. As Tyler Cowen likes to say, “we thought we were richer than we were.”
The key is that we’re talking about estimates and predictions, not real events. Delong’s interpretation would be perfectly valid if the projection paths were/are real events. But the 2007 path never actually happened and the 2013 path certainly hasn’t happened, either. Even without the forecasts, potential GDP is a hypothetical idea about what could have happened and can’t be observed as, say, inflation can be observed. When the CBO looks backward to past years, its figures are estimates of this hypothetical idea. When it looks forward to future years, the figures are predictions of estimates of a hypothetical idea. How real is that?
In any case, the estimates and predictions tend to anchor on the current environment as these things always do. And the difference from one to the next is explained mostly by new starting points and adjustments for past estimation and forecast errors, not real-life changes in the economy’s true growth potential.
Because there’s a Keynes passage for all occasions, I’ll suggest that Keynes understood this explanation as well as anyone. He famously stressed our fallibility in predicting the future in chapter 12 of The General Theory. Because long-term expectations are so difficult “as to be scarcely practicable,” he suggested that:
In practice, we have tacitly agreed, as a rule, to fall back on what is, in truth, a convention. The essence of this convention – though it does not, of course, work out so simply – lies in assuming that the existing state of affairs will continue indefinitely, except in so far as we have specific reasons to expect a change.
By choosing a passage from The General Theory’s final chapter to close his article, I’ll suggest that Delong missed by 12 chapters in finding an answer to his rhetorical question: “What else could it be?”
I recently posted my own reaction to the CBO’s debt projections, including a chart with their last six years of growth forecasts. And when you look at the projected growth rates you can see how the lines in Delong’s chart would ratchet downward based on simplistic forecasting methods. But the bigger issue, in my opinion, is that the CBO habitually understates debt for many reasons, including the assumption that the economy returns to full employment and sticks there, which of course never happens. There’s an historical pattern that we’ve seen over and over where governments aren’t transparent about their debt problems, which leads policymakers and economists to claim there’s no rush to fix them, and they kick the can until it’s too late. I’m currently writing a series of articles demonstrating this process at work in the U.S. today (using word games to make my points – if you’re not interested in my thesis but like word games you may still be entertained.)
The wonderful world of 0.22 etas
Now let’s turn to Delong’s assertion that the CBO’s forecasts imply an eta of 0.22. This depends critically on his list of reasons for the gap between the 2007 and 2013 projections. Replace those reasons with my explanation, and the CBO is saying absolutely nothing about eta. But what does eta mean? In the notation that Delong uses in papers on fiscal policy, eta describes the long-term effects of a shortfall between actual economic output and potential output. (For anyone familiar with the idea but not Delong’s notation, it’s the hysteresis effect.) If eta were truly equal to 0.22, this means that a fiscal stimulus that lifts economic output by $100 in the short-term will also increase output by $22 in every year beyond the short-term as far as you can see into the future. We’re talking some pretty powerful stimulus.
Let’s take an example. The Treasury determines that the economy’s fate rests on me, alone, and sends me a check for $100 without any offsetting taxes or spending reductions. I spend the $100 and the economy expands in the short-term by that same $100 (using a multiplier of 1 for simplicity). And in one hundred years time, I’m resting in peace (I hope), but the economy is still kicking out an extra $22 of output thanks to this check that I received today and this wonderful thing called eta. I’ll say it again: one hundred years from now, an extra $22 is being spent somewhere just because of the $100 that I spent today. Delong suggests an even greater effect, though, based on his belief that the multiplier is greater than 1. Using a multiplier of 1.5, as he did in one application of his model, year 2113 spending rises by $33 due to the $100 check from today.
Maybe it’s just me, but I can’t imagine how fiscal stimulus could possibly be that strong and permanent, or even close. And again, the CBO’s projections say no such thing. But let’s just assume I’m wrong. If eta really were 0.22, then we need to turn to the next step in Delong’s argument. This is the model that he and Larry Summers introduced in a paper last year, which I’ll call it the DS model. What should we know about the DS model? As with any model, you start by looking for the key assumptions. I’ll discuss just a few of these, because you don’t need to go any further than that. And if abstract models bore you, then don’t bother with this part at all. Just skip the next five paragraphs and pick up below. I’ll explain why at the end.
My optional critique of the DS model
First, the DS model assumes that the government never acts to reverse an increase in debt, nor does it try to stop an upwards trend in debt. In the latter assumption, the model doesn’t even allow for the possibility that debt could be trending upwards. It considers only a temporary jump based on a single instance of fiscal stimulus. It’s hard to know where to begin critiquing these assumptions. I’ll set aside the fact that our debt is, well, trending upwards, and point out that fiscal stimulus is often followed by restraint. It may be hard to remember that this actually happens, but in the 1980s alone it happened about a half dozen times. And it’s relevant here because the main cost of fiscal stimulus is the fact that you can’t add to the debt too many times without eventually taking steps to subtract from the debt. Otherwise, you’ll have to accept some combination of a handful of unpleasant consequences, all of which the DS model is ignorant.
Nobel Laureate Thomas Sargent put it this way: “There’s a fundamental truth that everyone has to understand: what the government spends, the public will pay for sooner or later, whether in taxes or inflation or having their debt defaulted on.” It seems irresponsible to base fiscal policy advice on a model that purports to quantify the costs of stimulus and yet ignores Sargent’s fundamental truth.
Second, the DS model assumes an unlimited supply of investors who are essentially price-takers – they’re indifferent to the size of the government’s deficit and debt. In other words, bond yields never adjust to the amount of supply that’s being pushed onto the market. We all know there’s a long history of economic models being built on this basis, on the premise that credit and financial markets don’t really matter. And we also know this is one of the reasons that conventional theory is repeatedly proven useless by real events.
With these assumptions, the DS model only considers one drawback to deficit spending – the added debt service cost, although assuming that interest rates never rise. Imagine where we’d be today if all of our past policymakers had looked at deficits in this way. Start with the Clinton administration, when we last took action to manage our finances responsibly, and then work backwards. Bush 41 defied key supporters and prioritized fiscal restraint over personal popularity. Truman and Eisenhower whittled down the mountain of World War II debt. Even Johnson and Reagan acted to reign in the deficits that had threatened to rise out of control during their administrations. And so on. Not one of these presidents based their decisions on a mathematical calculation of an increase in debt service costs, while assuming that the interest rate never rises. They were concerned about much bigger issues. That fiscal profligacy isn’t sustainable. That debt can’t rise forever. That there aren’t an unlimited number of investors willing to buy it. That the bond market makes sure that your quadrillionth dollar of issuance doesn’t carry the exact same yield as your trillionth or billionth. Remember Clinton’s famous question: “You mean to tell me that the success of the program and my reelection hinges on the Federal Reserve and a bunch of fucking bond traders?” And Eisenhower’s belief that “continued deficit spending is immoral”?
Thankfully, our past leaders didn’t base their decisions on the DS Model. If they had, it’s doubtful they would have found the motivation to reverse past increases in debt. Or even slow the growth in debt. If they had used the DS Model, we’d be Argentina.
Is it science or sales?
Now I’ll explain why I’m happy for you to ignore my critique of the DS model. Imagine that I enjoyed building models about make-believe economies and wrote a paper about a different model that I think works better than the DS model. And imagine that I had the prestige in the academic world to attract attention to my paper. Would this matter? Of course not. Not one important policy decision is based on the academic debate about eta or anything else in the DS model. That’s not how the world works. The Congressmen and women who make our laws don’t know eta from beta from tau. All they know is that they have famous economists supporting their policy position. And if the DS model were somehow discredited, there’d be another to take its place.
Nariman Behravesh, in his book Spin-free Economics, summarizes the core problem as follows:
[Economists] often hide their value judgments behind complex models and research, with the structure and assumptions of the study often leading to the conclusions that the research favors. More upfront honesty about values would probably lead to more productive discussion and less confusion on the part of the public.
But I’ll go further and say that models always lead to the conclusions that the research favors. Economic models, after all, are essentially collections of assumptions. Their output is just another window on the author’s beliefs. Here’s another excerpt, this one from a pseudonymous author (me):
For any economic issue, an economist can be found to back virtually any viewpoint. But the general public is ill-equipped to participate in the debate, because economists communicate in their own impenetrable language. Like the screen that separates the Wizard of Oz from his visitors, their lack of transparency is a barrier that plays to their advantage. It enhances their status as “experts” and thwarts public scrutiny of their work. And it also tempts economists to produce dishonest research when it suits their purposes, as many have acknowledged.
In the particular case of Delong’s Seeking Alpha post last week, there’s an obvious red flag. This is the link between his conclusions and an assumption for eta, which is one of those parameters that can be anything you want it to be. Apart from the usual problems with inferring the future from the past, its past can’t be measured in the first place. Which makes it highly exposed to the researcher’s incentives and biases.
But there are other red flags as well. In my opinion, whenever you see research that attributes a policy recommendation to a model, you should assume you’re being sold. And like any sales process, the pitch can be deceiving. Is there any real, objective science behind the conclusion? Or, is the model just as subjective as the person who built it? Whenever I dig deeper, the answer is always the latter. And if you skipped my critique above, bravo, that tells me you’ve stopped at the red flags and put away your shovel.
For more on Brad DeLong, see: