Sunday, November 30, 2014

Are US Households Done Deleveraging?

This Tuesday, I'll be  at Joseph Stiglitz's event at Columbia University on finance and inequality, presenting my work with Arjun Jayadev on household debt. You can find the latest version of our paper here.

In preparation, I've been updating the numbers and the results are interesting. As folks at the Fed have noted, the post-2007 period of household deleveraging seems to have reached its end. Here's what the household debt picture looks like, in the accounting framework that Arjun and I prefer.

The units are percent of adjusted household income. (We can ignore the adjustments here.) The heavy black line shows the year-over-year change in household debt-income ratios. The bars then disaggregate that change into new borrowing by households -- the primary deficit -- and the respective contributions of interest payments, inflation, income growth, and defaults. A negative bar indicates a factor that reduces leverage; in most years, this includes both (real) income and inflation, since by raising the denominator they reduce the debt-income ratio. A positive bar indicates a factor that increases leverage; this includes interest payments (which are always positive), and the primary deficit in years in which households are on net receiving funds from credit markets.

Here's what we are seeing:

In 2006 and 2007, debt-income ratios rose by about 3 percent each year; this is well below the six-point annual increases earlier in the 2000s, but still substantial. In 2008, the first year of the recession, the household debt-income ratio rises by another 3 points, despite the fact that households are now paying down debt, with repayments exceeding new borrowing by nearly 8 percent of household income. This is an astonishing rate of net repayment, the greatest since at least 1931. But despite this desperate effort to deleveraging, household debt-income ratios actually rose in 2008, thanks to the sharp fall in income and to near-zero inflation -- in most years, the rise in prices automatically erodes the debt-income ratio. The combination of negative net borrowing and a rising debt burden is eerily reminiscent of the early Depression -- it's a clear sign of how, absent Big Government, the US at the start of the last recession was on track for a reprise of the Depression.

Interest payments make a stable positive contribution to the debt-incoem ratio throughout this period. Debt-service payments do fall somewhat, from around 7 percent of household income in 2006 to around 5 percent in 2013. But compared with other variables important to debt dynamics, debt-service payments are quite stable in the short-term. (Over longer periods, changes in effective interest rates are a ] bigger deal.) It's worth noting in particular that the dramatic reduction in the federal funds rate in 2007-2008 had a negligible effect on the average interest rate paid by households.

In 2009-2012, the household debt-income ratio does fall, by around 5 points per year. But note that household surpluses (i.e. negative deficits) are no larger in these years than in 2008; the difference is that we see resumed positive growth of inflation and, a bit later, real incomes, raising the denominator of the debt-income ratio. This is what failed to happen in the 1930s. Equally important, there is a sharp rise in the share of debt written off by default, exceeding 3 percent in each year, compared with a writeoff rate below one percent in all pre-recession years. Note that the checked bar and the white bar are of similar magnitudes: In other words, repayment and default contributed about equally to the reduction of household debt. If deleveraging was an important requirement for renewed economic growth then it's a good thing that it's still possible to discharge our debts through bankruptcy. Otherwise, there would have been essentially no reduction in debt-income ratios between 2007 and 2012. [*]

This much is in the paper. But in 2013 the story changes a bit. The household debt-income ratio rises again, for the first time since 2008. And the household balance movers into deficit, for the first time since 2007 -- for the first time in six years, households are receiving more funds from the credit markets than they are paying back to them. These events are linked. While the central point of our paper is that changes in leverage cannot be reduced to changes in borrowing, for the US households in 2013, it is in fact increased borrowing that drove the rise in debt-income ratios. Inflation and income growth were basically constant between 2012 and 2013. The 5-point acceleration in the growth of the household debt-income ratio is explained by a 4.5 point rise in new borrowing by households (plus a 1.5 point fall in defaults, offset by a 1-point acceleration in real income growth).

So what do we make of this? Well, first, boringly perhaps but importantly, it's important to acknowledge that sometimes the familiar story is the correct story. If households owe more today than a year ago, it's because they borrowed more over the past year. It's profoundly misleading to suppose this is always the case. But in this case it is the case. Secondly, I think this vindicates the conclusion of our paper, that sustained deleveraging is impossible in the absence of substantially higher inflation, higher defaults, or lower interest rates. These are not likely to be seen without deliberate, imaginative policy to increase inflation, directly reduce the interest rates facing households, and/or write off much more of household debt than will happen through the existing bankruptcy process. Otherwise, in today's low-inflation environment, as soon as the acute crisis period ends leverage is likely to resume its rise. Which seems to be what we are seeing.

[*] More precisely: By our calculations, defaults reduced the aggregate household debt-income ratio by 20 points over 2008-2012, out of a total reduction of 21.5 points.


  1. Hey JW. This is interesting. You mention the "existing bankruptcy process" --- did you look in particular at the effects of the 2005 bankruptcy "reform" on default rates and lending? I guess you're looking at long run dynamics so that's not really as important as the monetary policy shocks of the 80s?

  2. From 2000 on, we have a precise measure of the share of household debt discharged each year, thanks to the NY Fed's Consumer Credit Panel. That fraction rises from 0.9 percent in 2000 to 1.3 percent in 2003-2005, falls back to 1 percent after the bankruptcy reform and then rises to 3.7 percent in 2010, before falling gradually back to 1.7 percent in 2013. (The numbers you see in the figure are slightly larger because they are given as shares of household income.) Of course it is possible that defaults would have been even greater under the old law. But yes, I think it is safe to say that in the longer view changes in bankruptcy law are not an important factor in the rise of household debt. After all, despite the new law, the share of debt discharged in 2009-2012 was historically unprecedented. The data prior to 2000 isn't as good, but it looks like the annual chargeoff rate for household debt rarely if ever passed 0.5 percent, even during recessions.

  3. JW
    There's lots of great stuff in this paper, but I'm struggling with this:
    "In particular, the rise in household leverage since the early 1980s is entirely attributable to higher interest rates, lower inflation, and lower income growth, in that order; household borrowing played no role."

    My understanding of your data was that borrowing (defined as year over year change in debt, plus defaults) increased by 1.49 points in 1984-1993 relative to the previous period. Higher interest payments meant that households didn't get to buy larger houses or consume more (primary expenditure), but that's not the same as saying we can dismiss borrower/lender behavior as an explanation for rising debt/gdp.

  4. QE- You're right, my language was probably not as clear as it should be. I'm still figuring out the best way of stating this argument.

    It is true that household borrowing, inclusive of interest payments, was higher over 1984-1993 than in the previous period. Now if we define new borrowing as borrowing exclusive of interest payments, then there was no increase. But that invites the question, why is this a sensible definition?

    I would suggest a couple of answers. First, there is no evidence that the increase in debt starting in the early 1980s was the result of a CHNAGE in household behavior. Yes, of course households might have chosen to reduce current expenditure sufficiently to offset the increase in debt service payments and thus held debt-incoem ratios constant. So in that sense, it is true that the rise in leverage fro the 1980s does reflect household behavior. But there is no evidence that household expenditure was any LESS responsive to changes in debt service burdens in the 1980s than previously. On the contrary, household expenditure did decline substantially in the 1980s -- just not enough to offset the higher interest costs.

    Here's a simplified but basically realistic example. Imagine a world where aggregate income was constant. Now suppose that homeowners in the aggregate devote 30% of their income to housing, which is just enough to replace depreciated housing and service existing mortgage debt. So aggregate housing assets and liabilities for homeowners remain constant. Now suppose that as a result of developments in the financial sector, interest rates on mortgage debt rise, so that the total cost of servicing existing debt and maintaining existing housing exceeds 30% of household income. If households continue spending 30%, then there will be some combination of a decline in housing assets in the household sector , and a rise in household debt. But I think it would be misleading to describe these results as simply "increased household borrowing" and look for a change in household decision-making to explain them.

    Here's another way of thinking about it. The aggregate credit flows that we see combine both prices and quantities. It's as if we only saw the total amount people were spending on gas, and couldn't observe the quantity of gas purchased directly. If you saw an increase in aggregate spending on gas in this situation, you would not want to immediately ask "why did the supply of gas increase"? It would be quite possible to see an increase in aggregate expenditure as a result of a downward shift in supply, if demand were less than unit elastic. This is especially likely in the credit case, since (a) demand for interest payments on existing debt is, modulo defaults, perfectly inelastic; and (b) demand for new borrowing is not driven by consumption loans (this is an important background point), but by financing for assets -- houses, cars, and increasingly college degrees -- which are tightly bound up with the normal household lifecycle and with income generating activities.

  5. Thanks for the response.
    The example you gave does not seem realistic to me given that most household debt is amortized. When the interest on an adjustable rate mortgage goes up, a household doesn't have the choice to keep spending 30% on housing, they can either make the higher payment or become delinquent. Credit card interest could be allowed to build up, but that’s a minor component of household debt. You’re right to point out the burden of interest payments from existing debt, but its effect is probably to reduce demand expenditure (and indirectly, pull down household income) rather than increase debt levels. My sense is that primary balance and interest is a rather arbitrary decomposition for household borrowing that confuses more than it illuminates. Do we really want to look at primary surplus and conclude the mid-to late 80s, a period that included the S&L Crisis, was a period of restrained lending/borrowing behavior, but for the pernicious influence of high and falling effective interest rates?

    The historical comparisons of interest rates vs nominal growth rates that you make in your “Fisher Dynamics” paper can be made just as well using growth, inflation, default, and borrowing. The debt/GDP ratio grew faster in 1984-1993 period than the 1964-1983 period because borrowing was higher (attributable to financial deregulation and regional housing bubbles), because nominal income was lower (attributable to a shift in policy emphasis from full employment to fighting inflation and supporting financial markets) and because the stock of debt was higher to begin with. It’s also worth noting that the earlier period includes the interest rate shocks of 1979-1982 and two earlier recessions where debt/gdp fell because of a combination of high inflation and sharp drops in borrowing--muddles the narrative a bit.

    Sorry to quibble about this issue, your points in the paper about aggregate demand, consumption, defaults and the importance of the denominator of the Debt/GDP ratio are spot on.