Don’t Touch the Yield

There’s a widespread idea in finance and economics land that there’s something wrong, dangerous, even unnatural about persistently low interest rates.

This idea takes its perhaps most reasonable form in arguments that the fundamental cause of the Great Financial Crisis was rates that were “far too low for far too long,” and that continued low interest rates, going forward, will only encourage speculation and new asset bubbles. Behind, or anyway alongside, these kinds of claims is a more fundamentally ideological view, that owners of financial assets are morally entitled to their accustomed returns, and woe betide the society or central banker that deprives them of the fruit of their non-labor. You hear this when certain well-known economists describe low rates as the “rape and plunder” of bondowners, or when Jim Grant says that the real victims of the recession are investors in money-market funds.

I want to look today at the “reaching for yield” version of this argument, which Brad Delong flagged as PRIORITY #1 RED FLAG OMEGA for the econosphere after it was endorsed by the Federal Reserve’s Jeremy Stein. [1] In DeLong’s summary:

Bankers want profits. … And a bank has costs above and beyond the returns on its portfolio. For each dollar of deposits it collects, a bank must spend 2.5 cents per year servicing those deposits. In normal times, when interest rates are well above 2.5 percent per year, banks have a normal, sensible attitude to risk and return. They will accept greater risk only if they come with returns higher enough to actually diminish the chances of reporting a loss. But when interest rates fall low enough that even the most sensible portfolio cannot reliably deliver a return on the portfolio high enough to cover the 2.5 cent per year cost of managing deposits, a bank will “reach for yield” and start writing correlated unhedged out-of-the-money puts so that it covers its 2.5 percent per year hurdle unless its little world blows up. Banks stop reducing their risk as falling returns mean that diversification and margin can no longer be counted on to manage them but instead embrace risks. 

It is Stein’s judgment that right now whatever benefits are being provided to employment and production by the Federal Reserve’s super-sub-normal interest rate policy and aggressive quantitative easing are outweighed by the risks being run by banks that are reaching for yield. 

Now on one level, this just seems like a non-sequitur. “Banks holding more risky assets” is, after all, just another way of saying “banks making more loans.” In fact, it’s hard to see how monetary policy is ever supposed to work if we rule out the possibility of shifting banks’ demand for risky private assets. [1] An Austrian, I suppose, might follow this logic to its conclusion and reject the idea of monetary policy in general; but presumably not an Obama appointee to the Fed.

But there’s an even more fundamental problem, not only with the argument here but with the broader idea — shared even by people who should know better — that low interest rates hurt bank profits. It’s natural to think that banks receive interest payments, so lower interest means less money for the bankers. But that is wrong.

Banks are the biggest borrowers as well as the biggest lenders in the economy, so what matters is not the absolute level of interest rates, but the spread — the difference between the rate at which banks borrow and the rate at which they lend. A bank covers its costs as reliably borrowing at 1 percent and lending at 4, as it does borrowing at 3 percent and lending at 6. So if we want to argue that monetary policy affects the profitability of bank lending, we have to argue that it has a differential effect on banks funding costs and lending rates.

For many people making the low-rates-are-bad-for-banks argument, this differential effect may come from a mental model in which the main bank liabilities are non-interest-bearing deposits. Look at the DeLong quote again — in the world it’s describing, banks pay a fixed rate on their liabilities. And at one point that is what the real world looked like too.

In 1960, non-interest-bearing deposits made up over two-thirds of total bank liabilities. In a system like that, it’s natural to see the effect of monetary policy as mainly on the asset side of bank balance sheets. But today’s bank balance sheets look very different: commercial banks now pay interest on around 80 percent of their liabilities. So it’s much less clear, a priori, why policy changes should affect banks interest income more than their funding costs. Since banks borrow short and lend long (that’s sort of what it means to be a bank), and since monetary policy has its strongest effects at shorter maturities, one might even expect the effect on spreads to go the other way.

And in fact, when we look at the data, that is what we see.

Average interest rate paid (red) and received (blue) by commercial banks. Source: FDIC

The black line with diamonds is the Federal Funds rate, set by monetary policy. The blue line is the average interest rate charged by commercial banks on all loans and leases; the solid red line is their average funding cost; and the dotted red line is the average interest rate on commercial banks’ interest-bearing liabilities. [3] As the figure shows, in the 1950s and ’60s changes in the federal funds rate didn’t move banks’ funding costs at all, while they did have some effect on loan rates; the reach-for-yield story might have made sense then. But in recent decades, as banks’ pool of cheap deposit funding has dried up, bank funding costs have become increasingly sensitive to the policy rate.

Looking at the most recent cycle, the decline in the Fed Funds rate from around 5 percent in 2006-2007 to the zero of today has been associated with a 2.5 point fall in bank funding costs but only a 1.5 point fall in bank lending rates — in other words, a one point increase in spreads. The same relationship, though weaker, is present in the previous two cycles, but not before. More generally, the correlation of changes in the federal funds rate and changes in bank spreads is 0.49 for 1955-1980, but negative 0.38 for the years 1991-2001. So Stein’s argument fails at the first step. If low bank margins are the problem, then “super-sub-normal interest rate policy” is the solution.

Let’s walk through this again. The thing that banks care about is the difference between what it costs them to borrow, and what they can charge to lend. Wider spreads mean lending is more profitable, narrower spreads mean it’s less so. And if banks need a minimum return on their lending — to cover fixed costs, or to pay executives expected bonuses or whatever — then when spreads get too narrow, banks may be tempted to take underprice risk. That’s “reaching for yield.” So turning to the figure, the spread is the space between the solid red line and the solid blue one. As we can see, in the 1950s and ’60s, when banks funded themselves mostly with deposits, the red line — their borrowing costs — doesn’t move at all with the federal funds rate. So for instance the sharp tightening at the end of the 1960s raises average bank lending rates by several points, but doesn’t move bank borrowing rates at all. So in that period, a high federal funds rate means wide bank spreads, and a low federal funds rate means narrower spreads. In that context the “reaching for yield” story has a certain logic (which is not to say it would be true, or important.) But since the 1980s, the red line — bank funding costs — has become much more responsive to the federal funds rate, so this relationship between monetary policy and bank spreads no longer exists. If anything, as I said, the correlation runs in the opposite direction.

Short version: When banks are funded by non-interest bearing deposits, low interest rates can hurt their profits, which makes them have a sad face. But when banks pay interest on almost all their liabilities, as today, low rates make them have a happy face. [4] In which case there’s no reason for them to reach for yield.

Now, it is true that the Fed has also intervened directly in the long end, where one might expect the impact on bank lending rates to be stronger. This is specifically the focus of a speech by Stein last October, where he explicitly said that if the policy rate were currently 3 percent he would have no objection to lowering it, but that he was more worried about unconventional policy to directly target long rates. [5] He offers a number of reasons why a fall in long rates due an expectation of lower short rates in the future would be expansionary, but a fall in long rates due to a lower term premium might not be. Frankly I find all these explanations ad-hoc and hand-wavey. But the key point for present purposes is that unconventional policy does not involve the central bank setting some kind of regulatory ceiling on long rates; rather, it involves lowering long rates via voluntary transactions with lenders. The way the Fed lowers rates on long bonds is by raising their price; the way it raises their price is by buying them. It is true, simply as a matter of logic, that the only way that QE can lower the market rate on a loan from, say, 4 percent to 3.9 percent, is by buying up enough loans (or rather, assets that are substitutes for loans) that the marginal lender now values a 3.9 percent loan the same as the marginal lender valued a 4 percent loan before. If a lender who previously would have considered a loan at 4 percent just worth making, does not now consider a loan at 3.9 percent worth making, then the interest rate on loans will not fall. Despite what John Taylor imagines, the Fed does not reduce interest rates by imposing a ceiling by fiat. So the statement, “if the Fed lowers long rates, bank won’t want to lend” is incoherent: the only way the Fed can lower long rates is by making banks want to lend more.

Stein’s argument is, to be honest, a bit puzzling. If it were true that banks respond to lower rates not by reducing lending or accepting lower profit margins, but by redoubling their efforts to fraudulently inflate returns, that would seem to be an argument for radically reforming the bank industry, or at least sending a bunch of bankers to jail. Stein, weirdly, wants it to be an argument for keeping rates perpetually high. But we don’t even need to have that conversation. Because what matters to banks is not the absolute level of rates, but the spread between their borrowing rate and their lending rate. And in the current institutional setting, expansionary policy implies higher spreads. Nobody needs to be reaching for yield.

[1] The DeLong post doesn’t give a link, but I think he’s responding to this February 7 speech.
[2] As Daniel Davies puts it in comments to the DeLong post:

If the Federal Reserve sets out on a policy of lowering interest rates in order to encourage banks to make loans to the real economy, it is a bit weird for someone’s main critique of the policy to be that it is encouraging banks to make loans. If Jeremy Stein worked for McDonalds, he would be warning that their latest ad campaign carried a risk that it might increase sales of delicious hamburgers.

[3] Specifically, these are commercial banks’ total interest payments from loans and leases divided by the total stock of loans and leases, and total interest payments divided by total liabilities and interest-bearing liabilities respectively.

[4] Why yes, I have been hanging around with a toddler lately. 

[5] Interesting historical aside: Keynes’ conclusion in the 1930s that central bank intereventions could not restore full employment and that fiscal policy was therefore necessary, was not — pace the postwar Keynesian mainstream — based on any skepticism about the responsiveness of economic activity to interest rates in principle. It was, rather, based on his long-standing doubts about the reliability of the link from short rates to long rates, plus a new conviction that central banks would be politically unable or unwilling to target long rates directly.

Planned Service Changes

[Edit, 4-30-14: I put this post up a week ago and then took it down after a few hours because, seriously, there is no way I am going on hiatus. But apparently it’s bad form to put a post up and then delete it, so in the interests of historical integrity I’m putting it back.]

This blog has never had a high volume of posts, but it’s going to drop to zero for the next few months.

As some of you know, I’m in the final stages of my PhD at UMass-Amherst (the Gondor of the austerity wars). I’ve been working on this thing for quite a few years, and could happily work on it quite a few more — except, damn it, I went and got a job. Starting next fall, I’ll be an assistant professor in the economics department at Roosevelt University in Chicago.

It’s a good job. I like the department a lot: it’s unapologetically heterodox and serves mostly working-class students; I like my new colleagues and I don’t mind moving back to Chicago, where I lived for most of the ’90s. Of course my mother thinks I should be at Harvard, and I do harbor fantasies of teaching at the PhD level. But that’s not going to happen, and short of that, Roosevelt is about ideal for me. So I’m happy.

But! I do have to get the dissertation done and defended before then. So, rewarding as this blog is — and it really is rewarding; I think I have the best readers in the econosphere — I need to shut it down. Next post you see from me, will be after the thesis is submitted. slow the pace of posting, from its already low levels.

Honestly, you probably won’t even notice the difference.

Aggregate Demand and Modern Macro

Start with a point on language.

People often talk about aggregate demand as if it were a quantity. But this is not exactly right. There’s no number or set of numbers in the national accounts labeled “aggregate demand”. Rather, aggregate demand is a way of interpreting the numbers in the national accounts. (Admittedly, it’s the way of interpreting them that guided their creation in the first place). It’s a statement about a relationship between economic quantities. Specifically, it’s a statement that we should think about current income and current expenditure as mutually determining each other.

This way of thinking is logically incompatible with the way macroeconomics is taught in (almost) all graduate programs today, which is in terms of optimization under an intertemporal budget constraint. I’ll avoid semi-pejorative terms like mainstream and neoclassical, and instead follow Robert Gordon and call this approach modern macro.

In the Keynesian income-expenditure vision — which today survives only, as Leijonhufvud put it, “in the Hades of undergraduate instruction” — we think of economic actors as making decisions about current spending in terms of current receipts. If I earn $X, I will spend $Y; if I earn one dollar more, I’ll spend so many additional cents. We can add detail by breaking these income and expenditure flows in various ways — income from dividends vs. incomes from wages, income to someone at the top decile vs someone at the bottom, income to urban households vs income to rural ones; and expenditures on services, durable goods, taxes, etc., which generate income in their turn. This is the way macro forecasting models used by business and government were traditionally constructed, and may still be for all I know.

Again, these are relationships; they tell us that for any given level of aggregate money income, there is a corresponding level of aggregate expenditure. The level of income that is actually realized, is the one for which desired expenditure just equals income. And if someone for whatever reason adjusts their desired level of expenditure at that income, the realized level of income will change in the same direction, by a greater or lesser extent. (This is the multiplier.)

I should stress that while this way of thinking may imply or suggest concrete predictions, these are not themselves empirical claims, but logical relationships.

The intertemporal optimization approach followed in modern macro is based on a different set of logical relationships. In this framework, agents know their endowments and tastes (and everyone else’s, though usually in these models agents are assumed to be identical) and the available production technology in all future periods. So they know all possible mixes of consumption and leisure available to them over the entire future and the utility each provides. Based on this knowledge they pick, for all periods simultaneously (“on the 8th day of creation” — that’s Leijonhufvud again) the optimal path of labor, output and consumption.

I realize that to non-economists this looks very strange. I want to stress, I’m not giving a dismissive or hostile summary. To anyone who’s done economics graduate work in the last 15 or 20 years (a few heterodox enclaves excepted) constructing models like this is just what “doing macroeconomics” means.

(For a concrete example, a first-year grad textbook offers as one of its first exercises in thinking like an economist the question of why countries often run current account deficits in wartime. The answer is entirely in terms of why countries would choose to allocate a greater share of consumption to periods when there is war, and how interest rates adjust to make this happen. The possibility that war leads to higher incomes and expenditure, some of which inevitably falls on imported goods — the natural answer in the income-expenditure framework — is not even mentioned. Incidentally, as this example suggests, thinking in terms of intertemporal allocation is not always necessarily wrong.)

In these models, there is no special relationship between income and expenditure flows just because they happen to take place at the same time or in any particular order. The choice between jam today and jam tomorrow is no different from the choice between blackberry and lingonberry jam, and the checks you get from your current job and from the job you’ll hold ten years from now are no more different than the checks from two different jobs that you hold right now are. Over one year or 50, the problem is simply the best allocation of your total income over your possible consumption baskets — subject, of course, to various constraints which may make the optimal allocation unachievable.

My point here is not that modern models are unrealistic. I am perfectly happy to stipulate that the realism of assumptions doesn’t matter. Models are tools for logical analysis, not toy train sets — they don’t have to look like real economies to be useful.

(Although I do have to point out that modern macroeconomics models are often defended precisely on the grounds of microfoundations — i.e. more realistic assumptions. But it is simply not true that modern models are more “microfounded” than income-expenditure ones in any normal English sense. Microfoundations does not mean, as you might imagine, that a model has an explicit story about the individual agents in the economy and how they make choices — the old Keynesian models do at least as well as the modern ones by that standard. Rather, microfoundations means that the agents’ choices consist of optimizing some quantity under true — i.e. model consistent — expectations.)

But again, I come not to bury or dispraise modern models. My point is just that they are logically incompatible with the concept of aggregate demand. It’s not that modern macroeconomists believe that aggregate demand is unimportant, it’s that within their framework those words don’t mean anything. Carefully written macro papers don’t even footnote it as a minor factor that can be ignored. Even something anodyne like “demand might also play a role” would come across like the guy in that comic who asks the engineers if they’ve “considered logarithms” to help with cooling.

The atomic units of one vision are flows — that is, money per time period — between economic units. The atomic units of the other are prices and quantities of different goods. Any particular empirical question can be addressed within either vision. But they generate very different intuitions, and ideas of what questions are most important. 

Still, it is true that the same concrete phenomena can be described in either language. The IS curve is the obvious example. In the Hades of the undergraduate classroom we get the old Keynesian story of changes in interest rate changing desired aggregate expenditure at each given income. While in the sunlit Arcadia of graduate classes, the same relationship between interest rates and current expenditure is derived explicitly from intertemporal optimization.

So what’s the problem, you say. If either language can be used to describe the same phenomena, why not use the same language as the vast majority of other economists?

This is a serious question, and those of us who want to do macro without DSGE models need a real answer for it. My answer is that default assumptions matter. Yes, with the right tweaks the two models can be brought to a middle ground, with roughly the same mix of effects from the current state and expected future states of the world. But even if you can get agreement on certain concrete predictions, you won’t agree on what parts of them depend on the hard core of your theory and what on more or less ad hoc auxiliary assumptions. So Occam’s Razor will cut in opposite directions — a change that simplifies the story from one perspective, is adding complexity from the other.

For example, from the income-expenditure perspective, saying that future interest rates will have a similar effect on current activity as current interest rates do, is a strong additional assumption. Whereas from the modern perspective, it’s saying that they don’t have similar effects that is the additional assumption that needs to be explained. Or again, taking an example with concrete applications to teaching, the most natural way to think about interest rates and exchange rates in the income-expenditure vision is in terms of how the the flow of foreign investment responds to interest rates differentials. Whereas in the modern perspective — which is now infiltrating even the underworld — the most natural way is in terms of rational agents’ optimal asset mix, taking into account the true expected values of future exchange rates and interest rates.

Or, what got me thinking about this in the first place. I’ve been reading a lot of empirical work on credit constraints and business investment in the Great Recession — I might do a post on it in the next week or so, though an academic style literature review seems a bit dull even for this blog — and three things have become clear.

First, the commitment to intertemporal optimization means that New Keynesians really need financial frictions. In a world where current output is an important factor in investment, where investment spending is linked to profit income, and where expectations are an independently adjusting variable, it’s no problem to have a slowdown in investment triggered by fall in demand in some other sector, by a fall in the profit share, or by beliefs about the future becoming more pessimistic. But in the modern consensus, the optimal capital stock is determined by the fundamental parameters of the model and known to all agents, so you need a more or less permanent fall in the return on investment, due presumably to some negative technological shock or bad government policy. Liberal economists hate this stuff, but in an important sense it’s just a logical application of the models they all teach. If in all your graduate classes you talk about investment and growth in terms of the technologically-determined marginal product of capital, you can hardly blame people when, faced with a slowdown in investment and growth, they figure that’s the first place to look. The alternative is some constraint that prevents firms from moving toward their desired capital stock, which really has to be a financial friction of some kind. For the older Keynesian perspective credit constraints are one possible reason among others for a non-supply-side determined fall in investment; for the modern perspective they’re the only game in town.

Second, the persistence of slumps is a problem for them in a way that it’s not in the income-expenditure approach. Like the previous point, this follows from the fundamental fact that in the modern approach, while there can be constraints that prevent desired expenditure from being achieved, there’s never causation from actual expenditure to desired expenditure. Businesses know, based on fundamentals, their optimal capital stock, and choose an investment path that gets them there while minimizing adjustment costs. Similarly, households know their lifetime income and utility-maximizing consumption path. Credit constraints may hold down investment or consumption in one period, but once they’re relaxed, desired expenditure will be as high or higher than before. So you need persistent constraints to explain persistently depressed spending. Whereas in the income-expenditure model there is no puzzle. Depressed investment in one period directly reduces investment demand in the next period, both by reducing capacity utilization and by reducing the flow of profit income. If your core vision of the economy is a market, optimizing the allocation of scare resources, then if that optimal allocation isn’t being achieved there must be some ongoing obstacle to trade. Whereas if you think of the economy in terms of income and expenditure flows, it seems perfectly natural that an interruption to some flow will will disrupt the pattern, and once the obstacle is removed the pattern will return to its only form only slowly if at all.

And third: Only conservative economists acknowledge this theoretical divide. You can find John Cochrane writing very clearly about alternative perspectives in macro. But saltwater economists — and the best ones are often the worst in this respect — are scrupulously atheoretical. I suspect this is because they know that if they wanted to describe their material in a more general way, they’d have to use the language of intertemporal optimization, and they are smart enough to know what a tar baby that is. So they become pure empiricists.

In Leon Fink’s wonderful history of the New York health care workers’ union 1199, Upheaval in the Quiet Zone, he talks about how the union’s early leaders and activists were disproportionately drawn from Communist Party members and sympathizers, and other leftists. Like other communist-led unions, 1199 was kicked out of the CIO in the 1940s, but unlike most of the others, it didn’t fade into obscurity. Originally a drugstore-employees union, it led the new wave of organizing of health care and public employees in the 1960s. Fink attributes a large part of its unusual commitment to organizing non-white and female workers, in an explicitly civil-rights framework, and its unusual lack of corruption and venality, to the continued solidarity of the generation of the 1930s. Their shared political commitments were a powerful source of coordination and discipline. But, says Fink, it was impossible for them to pass these commitments on to the next generation. Yes, in 1199, unlike most other unions, individual leftists were not purged; but there was still no organized left, either within the union or in connection to a broader movement. So there was no way for the first generation to reproduce themselves, and as they retired 1199 became exposed to the same pressures that produced conservative, self-serving leadership in so many other unions.

I feel there’s something similar going on in economics. There are plenty of people at mainstream departments with a basically Keynesian vision of the economy. But they write and, especially, teach in a language that is basically alien to that vision. They’re not reproducing the capacity for their own thought. They’re running a kind of intellectual extractive industry, mining older traditions for insights but doing nothing to maintain them.

I had this conversation with a friend at a top department the other day:

  what do you think? is this kind of critique valid/useful?
11:17 AM him: its totally true
11:18 AM and you wouldn’t know what was getting baked into the cake unless you were trained in the literature
  I only started understanding the New Keynsian models a little while ago
  and just had the lazy “they are too complicated” criticism
11:19 AM now I understand that they are stupidly too complicated (as Noah’s post points out)
11:20 AM me: so what is one supposed to do?
  if this is the state of macro
 him: i dunno. I think participating in this literature is a fucking horror show
 me: but you don’t like heterodox people either, so….?
11:21 AM him: maybe become a historian
  or figure out some simple variant of the DSGEmodels that you can make your point and publish empirical stuff

This is where so many smart people I know end up. You have to use mainstream models — you can’t move the profession or help shape policy (or get a good job) otherwise. But on many questions, using those models means, at best, contorting your argument into a forced and unnatural framework, with arbitrary-seeming assumptions doing a lot of the work; at worst it means wading head-deep into an intellectual swamp. So you do some mix of what my friend suggests here: find a version of the modern framework that is loose enough to cram your ideas into without too much buckling; or give up on telling a coherent story about the world and become a pure empiricist. (Or give up on economics.) But either way, your insights about the world have to come from somewhere else. And that’s the problem, because insight isn’t cheap. The line I hear so often — let’s master mainstream methods so we can better promote our ideas — assumes you’ve already got all your ideas, so the only work left is publicity.

If we want to take questions of aggregate demand and everything that goes with it — booms, crises, slumps — seriously, then we need a theoretical framework in which those questions arise naturally.

[*] Keynes’ original term was “effective demand.” The two are interchangeable today. But it’s interesting to read the original passages in the GT. While they are confusingly written, there’s no question that Keynes’ meant “effective” in the sense of “being in effect.” That is, of many possible levels of demand possible in an economy, which do we actually see? This is different from the way the term is usually understood, as “having effect,” that is, backed with money. Demand backed with money is, of course, simply demand.  

UPDATE: The Cochrane post linked above is really good, very worth reading. It gives more of the specific flavor of these models than I do. He writes: In Old Keynesian models,

consumption depends on today’s income through the “marginal propensity to consume” mpc. 

Modern new-Keynesian models are utterly different from this traditional view. Lots of people, especially in policy, commentary, and blogging circles, like to wave their hands over the equations of new Keynesian models and claim they provide formal cover for traditional old-Keynesian intuition, with all the optimization, budget constraints, and market clearing conditions that the old-Keynesian analysis never really got right taken care of. A quick look at our equations and the underlying logic shows that this is absolutely not the case.  

Consider how lowering interest rates is supposed to help. In the old Keynesian model, investment I = I(r) responds to lower interest rates, output and income Y = C + I + G, so rising investment raises income, which raises consumption in (4), which raises income some more, and so on. By contrast, the simple new-Keynesian model needs no investment, and interest rates simply rearrange consumption demand over time. 

Similarly, consider how raising government spending is supposed to help. In the old Keynesian model,  raising G in Y = C + I + G raises Y, which raises consumption C by (4), which raises Y some more, and so on. In the new-Keynesian model, the big multiplier comes because raising government spending raises inflation, which lowers interest rates, and once again brings consumption forward in time.

Note, for example, that in a standard New Keynesian model, expected future interest rates enter into current consumption exactly as the present interest rate does. This will obviously shape people’s intuitions about things like the effectiveness of forward guidance by the Fed.

UPDATE 2: As usual, this blog is just an updated, but otherwise much inferior, version of What Leijonhufvud Said. From his 2006 essay The Uses of the Past:

We should expect to find an ahistorical attitude among a group of scientists busily soling puzzles within an agreed-upon paradigm… Preoccupation with the past is then a diversion or a luxury. When things are going well it is full steam ahead! …. As long as “normal” progress continues to be made in the established directions, there is no need to reexamine the past… 

Things begin to look different if and when the workable vein runs out or, to change the metaphor, when the road that took you to the “frontier of the field” ends in a swamp or in a blind alley. A lot of them do. Our fads run out and we get stuck. Reactions to finding yourself in a cul-de-sac differ. Tenured professors might be content to accommodate themselves to it, spend their time tidying up the place, putting in modern conveniences… Braver souls will want out and see a tremendous leap of the creative imagination as the only way out — a prescription, however, that will leave ordinary mortals just climbing the walls. Another way to go is to backtrack. Back there, in the past, there were forks in the road and it is possible, even probable, that some roads were more promising than the one that looked most promising at the time…

This is exactly the spirit in which I’m trying to rehabilitate postwar income-expenditure Keynesianism. The whole essay is very worth reading, if you’re interest at all in the history of economic thought.

Borrowing ≠ Debt

There’s a common shorthand that makes “debt” and “borrowing” interchangeable. The question of why an economic unit had rising debt over some period, is treated as equivalent to the question of why it was borrowing more over that period, or why its expenditure was higher relative to its income. This is a natural way of talking, but it isn’t really correct.

The point of Arjun’s and my paper on debt dynamics was to show that for household debt, borrowing and changes in debt don’t line up well at all. While some periods of rising household leverage — like the housing bubble of the 2000s — were also periods of high household borrowing, only a small part of longer-term changes in household debt can be explained this way. This is because interest, income growth and inflation rates also affect debt-income ratios, and movements in these other variables often swamp any change in household borrowing.
As far as I know, we were the first people to make this argument in a systematic way for household debt. For government debt, it’s a bit better known — but only a bit. People like Willem Buiter or Jamie Galbraith do point out that the fall in US debt after World War II had much more to do with growth and inflation than with large primary surpluses. You can find the argument more fully developed for the US in papers by Hall and Sargent  or Aizenman and Marion, and for a large sample of countries by Abbas et al., which I’ve discussed here before. But while many of the people making it are hardly marginal, the point that government borrowing and government debt are not equivalent, or even always closely linked, hasn’t really made it into the larger conversation. It’s still common to find even very smart people saying things like this:

We didn’t have anything you could call a deficit problem until 1980. We then saw rising debt under Reagan-Bush; falling debt under Clinton; rising under Bush II; and a sharp rise in the aftermath of the financial crisis. This is not a bipartisan problem of runaway deficits! 

Note how the terms “deficits” and “rising debt” are used interchangeably; and though the text mostly says deficits, the chart next to this passage shows the ratio of debt to GDP.
What we have here is a kind of morality tale where responsible policy — keeping government spending in line with revenues — is rewarded with falling debt; while irresponsible policy — deficits! — gets its just desserts in the form of rising debt ratios. It’s a seductive story, in part because it does have an element of truth. But it’s mostly false, and misleading. More precisely, it’s about one quarter true and three quarters false.
Here’s the same graph of federal debt since World War II, showing the annual change in debt ratio (red bars) and the primary deficit (black bars), both measured as a fraction of GDP. (The primary deficit is the difference between spending other than interest payments and revenue; it’s the standard measure of the difference between current expenditure and current revenue.) So what do we see?
It is true that the federal government mostly ran primary surpluses from the end of the war until 1980, and more generally, that periods of surpluses were mostly periods of rising debt, and conversely. So it might seem that using “deficits” and “rising debt” interchangeably, while not strictly correct, doesn’t distort the picture in any major way. But it does! Look more carefully at the 1970s and 1980s — the black bars look very similar, don’t they? In fact, deficits under Reagan were hardy larger than under Ford and Carter —  a cumulative 6.2 percent of GDP over 1982-1986, compared with 5.6 percent of GDP over 1975-1978. Yet the debt-GDP ratio rose by just a single point (from 24 to 25) in the first episode, but by 8 points (from 32 to 40) in the second. Why did debt increase in the 1980s but not in the 1970s? Because in the 1980s the interest rate on federal debt was well above the economy’s growth rate, while in the 1970s, it was well below it. In that precise sense, if debt is a problem it very much is a bipartisan one; Volcker was the appointee of both Carter and Reagan.
Here’s the same data by decades, and for the pre- and post-1980 periods and some politically salient subperiods.  The third column shows the part of debt changes not explained by the primary balance. This corresponds to what Arjun and I call “Fisher dynamics” — the contribution of growth, inflation and interest rates to changes in leverage. [*] The units are percent of GDP.
Totals by Decade
Primary Deficit Change in Debt Residual Debt Change
1950s -8.6 -29.6 -20.9
1960s -7.3 -17.7 -10.4
1970s 2.8 -1.7 -4.6
1980s 3.3 16.0 12.7
1990s -15.9 -7.3 8.6
2000s 23.7 27.9 4.2
Annual averages
Primary Deficit Change in Debt Residual Debt Change
1947-1980 -0.7 -2.0 -1.2
1981-2011 0.1 1.3 1.2
   1981-1992 0.3 1.8 1.5
   1993-2000 -2.7 -1.6 1.1
   2001-2008 -0.1 0.8 0.9
   2009-2011 7.3 8.9 1.6

Here again, we see that while the growth of debt looks very different between the 1970s and 1980s, the behavior of deficits does not. Despite Reagan’s tax cuts and military buildup, the overall relationship between government revenues and expenditures was essentially the same in the two decades. Practically all of the acceleration in debt growth in the 1980s compared with the 1970s is due to higher interest rates and lower inflation.

Over the longer run, it is true that there is a shift from primary surpluses before 1980 to primary deficits afterward. (This is different from our finding for households, where borrowing actually fell after 1980.) But the change in fiscal balances is less than 25 percent the change in debt growth. In other words, the shift toward deficit spending, while real, only accounts for a quarter of the change in the trajectory of the federal debt. This is why I said above that the morality-tale version of the rising debt story is a quarter right and three quarters wrong.

By the way, this is strikingly consistent with the results of the big IMF study on the evolution of government debt ratios around the world. Looking at 60 episodes of large increases in debt-GDP ratios over the 20th century, they find that only about a third of the average increase is accounted for by primary deficits. [2] For episodes of falling debt, the role of primary surpluses is somewhat larger, especially in Europe, but if we focus on the postwar decades specifically then, again, primary surpluses accounted for only a about a third of the average fall. So while the link between government debt and deficits has been a bit weaker in the US than elsewhere, it’s quite weak in general.

So. Why should we care?

Most obviously, you should care if you’re worried about government debt. Now maybe you shouldn’t worry. But if you do think debt is a problem, then you are looking in the wrong place if you think holding down government borrowing is the solution. What matters is holding down i – (g + π) — that is, keeping interest rates low relative to growth and inflation. And while higher growth may not be within reach of policy, higher inflation and lower interest rates certainly are.

Even if you insist on worrying not just about government debt but about government borrowing, it’s important to note that the cumulative deficits of 2009-2011, at 22 percent of GDP, were exactly equal to the cumulative surpluses over the Clinton years, and only slightly smaller than the cumulative primary surpluses over the whole period 1947-1979. So if for whatever reason you want to keep borrowing down, policies to avoid deep recessions are more important than policies to control spending and raise revenue.

More broadly, I keep harping on this because I think the assumption that the path of government debt is the result of government borrowing choices, is symptomatic of a larger failure to think clearly about this stuff. Most practically, the idea that the long-run “sustainability” of the  debt requires efforts to control government borrowing — an idea which goes unquestioned even at the far liberal-Keynesian end of the policy spectrum —  is a serious fetter on proposals for more stimulus in the short run, and is a convenient justification for all sorts of appalling ideas. And in general, I just reject the whole idea of responsibility. It’s ideology in the strict sense — treating the conditions of existence of the dominant class as if they were natural law. Keynes was right to see this tendency to view of all of life through a financial lens — to see saving and accumulating as the highest goals in life, to think we should forego real goods to improve our financial position — as “one of those semicriminal, semi-pathological propensities which one hands over with a shudder to the specialists in mental disease.”

On a methodological level, I see reframing the question of the evolution of debt in terms of the independent contributions of primary deficits, growth, inflation and interest rates as part of a larger effort to think about the economy in historical, dynamic terms, rather than in terms of equilibrium. But we’ll save that thought for another time.

The important point is that, historically, changes in government borrowing have not been the main factor in the evolution of debt-GDP ratios. Acknowledging that fact should be the price of admission to any serious discussion of fiscal policy.

[1] Strictly speaking, debt ratios can change for reasons other than either the primary balance or Fisher dynamics, such as defaults or the effects of exchange rate movements on foreign-currency-denominated debt. But none of these apply to the postwar US.

[2] The picture is a bit different from the US, since adverse exchange-rate movements are quite important in many of these episodes. But it remains true that high deficits are the main factor in only a minority of large increases in debt-GDP ratios.