Printer Friendly

Is a household debt overhang holding back consumption?

ABSTRACT The recent plunge in U.S. home prices left many households that had borrowed voraciously during the credit boom highly leveraged, with very high levels of debt relative to the value of their assets. Analysts often assert that this "debt overhang" created a need for household deleveraging that, in turn, has been depressing consumer spending and impeding the economic recovery. This paper uses household-level data to examine this hypothesis. I find that highly leveraged homeowners had larger declines in spending between 2007 and 2009 than other homeowners, despite having smaller changes in net worth, suggesting that their leverage weighed on consumption above and beyond what would have been predicted by wealth effects alone. Results from regressions that control for wealth effects and other factors support the view that excessive leverage has contributed to the weakness in consumption. I also show that U.S. households, on the whole, have made limited progress in reducing leverage over the past few years. It may take many years for some households to reduce their leverage to precrisis norms. Thus, the effects of deleveraging may persist for some time to come.

**********

The bursting of the U.S. housing bubble inflicted enormous damage on household finances. Besides contributing to a significant decline in the net worth of homeowners, the plunge in home prices left many of those who had borrowed voraciously during the credit boom highly leveraged, meaning that they had very high levels of debt relative to the value of their assets. Analysts often assert that this "debt overhang" created a need for household deleveraging that, in turn, has been depressing consumer spending and impeding the economic recovery.

The past few years have indeed seen both a sizable decline in aggregate household debt and weak growth in aggregate consumer spending. However, the nature of the relationship between these two developments is not well understood. According to the simplest models used by economists, a household's consumption is determined by its income (actual and expected), wealth, preferences, and the return it earns on savings. In slightly more refined models, the uncertainty faced by a household plays a role, as does its ability to borrow. However, debt does not typically exert an independent influence on consumption in traditional models; rather, borrowing is presumed to vary with consumption, as the latter rises and falls in reaction to changes in its determinants.

The traditional framework points to many factors that may be contributing to the lackluster performance of consumer spending in recent years. Wealth losses, weak income growth, and limited availability of credit, as well as a more uncertain and pessimistic outlook for future income, would all be expected to have depressed spending. Within the traditional framework, the observed decline in debt over the past few years would be interpreted as the result of weak consumption growth rather than a driving force in and of itself.

This paper asks whether a need to reverse the run-up in leverage that arose from the credit boom and subsequent collapse in home prices is in fact contributing to the recent weakness in consumer spending. To test this deleveraging hypothesis, I look at whether the households with the greatest mortgage leverage several years ago have reduced their spending the most, all else equal. I use household-level data so that I can control for other factors that might have led highly leveraged households to have different patterns of consumption than their counterparts with less leverage.

High levels of debt and leverage might have had an independent influence on consumer spending for several reasons. First, some households may target a given level of leverage; the sharp rise in leverage that occurred with the slump in home prices may have induced these households to pare back their consumption in order to pay down debt. Second, financial institutions are typically less willing to lend to more highly leveraged households. As a result, the rise in leverage has impeded some households from borrowing more to finance consumption and has prevented others from raising their discretionary cash flow by refinancing into lower-rate mortgages.

To set the stage for my analysis of deleveraging, I begin by examining how some households ended up with so much leverage in the first place. I find that, as of 2007, homeowners in states that had experienced the most pronounced housing booms tended to have considerably more mortgage debt than homeowners in other states. However, even the most indebted of these households did not at that time appear excessively leveraged, because they had seen so much home price appreciation. In other words, they appeared to have fairly solid balance sheets under the assumption that home prices would remain flat or increase going forward. In the end, of course, these homeowners were left in a precarious situation when the rise in home prices proved to be a bubble: their mortgages often came to exceed the value of their homes, and they had limited, if any, ability to borrow more, refinance, or sell their homes in the face of a shock to income that made it difficult to make their (relatively high) mortgage payments.

I also find that after the housing bubble burst, highly leveraged households had larger declines in spending than their less leveraged counterparts despite having smaller changes in net worth, suggesting that their leverage weighed on their consumption above and beyond what would have been predicted by wealth effects alone. Results from regressions that control for wealth effects and other factors that might have influenced consumption are consistent with this view. Not surprisingly, highly leveraged mortgage borrowers also had more difficulty meeting their loan payment obligations in the wake of the home price bust, and nearly a fifth of them were no longer homeowners by 2011.

The most similar study of deleveraging to date is a paper by Atif Mian, Kamalesh Rao, and Amir Sufi (2011), which compares spending patterns across U.S. counties with different average degrees of household leverage. That paper finds that retail sales dropped much more sharply in counties with higher leverage. One challenge in interpreting these results is that the counties with the most leverage also tended to be those with the largest home price declines, such that a powerful wealth effect, in addition to any deleveraging effect, should have been depressing consumption in these areas. My paper goes beyond the Mian, Rao, and Sufi analysis to show that high leverage appears to be associated with weak consumption growth even after accounting for wealth effects.

The limitations of my data source make it difficult to quantify the precise effects of deleveraging on the macroeconomy. However, using data that extend through 2011, I show that U.S. households, on the whole, have made very limited progress in reducing leverage over the past few years. Important financial strains persist, as evidenced by the fact that there was essentially no reduction between 2009 and 2011 in the share of homeowners reporting that they were somewhat or very likely to have problems making their mortgage payments over the coming year.

I. Background

In this section I set the stage for the analysis to come. I discuss the macroeconomic backdrop in order to provide context for why high household leverage is a key policy issue, and I provide more detail about the possible channels through which leverage might be influencing household spending. I also explain how my work fits in with previous studies of household indebtedness.

I.A. The Household Debt Crisis and the Macroeconomy

The lackluster economic recovery during the past two and a half years has spurred discussion about whether the United States will experience a "lost decade" of stagnant economic growth as Japan did following the bursting of its own property price bubble in the early 1990s. Carmen Reinhart and Kenneth Rogoff (2009) present evidence that the weakness in the U.S. economy is likely to persist for a very long time. Examining a large number of severe financial crises in developed and emerging economies over several centuries, Reinhart and Rogoff document that the economic slumps that follow tend to be deep and protracted. They note that it is "beyond contention that the [recent] U.S. financial crisis [was] severe by any metric" (p. 467), the implication being that the U.S. economy is likely to share a similar fate.

Household debt plays a key role in the narrative supporting this view. Outstanding consumer loans and, especially, residential mortgage loans rose significantly during the credit boom in the early and middle part of the last decade (figure 1). The subsequent sharp increase in the number of households having problems making mortgage payments, which began before the economy fell into recession and joblessness rose, suggests that many households took on more debt during the boom than they could sustain over the long run. Although rapidly rising home prices meant that mortgage leverage--as captured by the aggregate mortgage loan-to-value ratio--barely budged for much of the credit boom, it rose sharply after home prices turned down in mid-2006 (figure 2). Many analysts think that this "debt overhang" and the ensuing process of deleveraging have held back consumption and the broader recovery over the past few years and will remain a headwind against economic growth for some time to come.

It is important to explore the veracity of this narrative. As policymakers gauge whether additional fiscal and monetary stimulus might be justified, they need to understand how the still-elevated level of aggregate household leverage bears on the underlying strength of the economy. Moreover, a better understanding of the implications of high leverage might shed light on the benefits of specific policy interventions. For example, Joseph Gagnon (2011) and others have argued for improvements in programs that allow "underwater" borrowers (those with mortgages exceeding the value of their homes) to refinance, so that more households can benefit from the low mortgage rates that have resulted from accommodative monetary policy. Other analysts have advocated reducing mortgage principal in order to revive the economy (see, for example, Goodman 2011).

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

I.B. What Is the Relationship between Household Debt and Spending?

In the simplest models used by economists, households can borrow as much as they wish, and a household's spending at any given time is based on its expected lifetime resources, interest rates, and tastes. This level of spending, together with the household's current income, determines its current saving or dissaving (borrowing). If incomes are expected to rise over time until retirement, as they typically do, households in this constraint-free world will tend to take on debt, on net, when young, move into positive net worth as they age, and then run down their net worth in retirement. Of course, evidence suggests that in the real world many households cannot borrow as much as they wish. These liquidity-constrained households may at some points in their lives have to accept levels of consumption that are low relative to their lifetime resources.

Even households with positive net worth often choose to hold some debt. This behavior arises in part because of the convenience of using credit cards, but a more important consideration is the need to borrow to purchase a home: when a household wants to own a home, and its desired housing services can only be provided by a property whose value exceeds the household's wealth, borrowing is the only option. In this case, the household not only has a motivation for borrowing but also can use the home as collateral, to create an ability to borrow that would not otherwise exist.

Two factors appear to have been especially important to the rapid growth of U.S. household debt in the early and mid-2000s. First, financial innovation continued to broaden households' access to credit and lower the cost of credit for households that already had access; this process (which had been under way for decades) in some ways accelerated over the period, amid lagging financial regulation (see Dynan 2009). For example, the increasingly popular "nontraditional" mortgages allowed households with lower or more variable incomes, less wealth, and weaker credit records to finance the purchase of a home. Second, and perhaps relatedly, rapid home price appreciation fueled growth in mortgage debt and other household debt. Most notably, higher home prices increased desired spending through a wealth effect, and some of that higher spending was financed by borrowing. (1) Moreover, higher home prices increased the value of collateral against which constrained households could borrow to finance their desired spending.

A household's leverage is often measured by the ratio of its debt to its assets, or, in work focusing on housing, the ratio of a homeowner's mortgage balance to the value of the underlying home. The use of such measures probably reflects in part lenders' emphasis on these ratios when setting the interest rate on loans or determining their willingness to lend in the first place. In any case, such measures are closely related to the traditional leverage ratio of assets to net worth used in the corporate finance literature. If D represents debt, A represents assets, and NW represents net worth, then

(1) D/A = 1 - 1/A/NW.

Economic theory suggests that household spending and balance sheets should have changed in several ways in response to the approximately one-third drop in national-average home prices since their peak in mid-2006. The direct effect of this decline was a drop in household assets and net worth with no change in debt, leaving households more leveraged than before. This loss in wealth should have led households to spend less and therefore to save more out of their current income; over time, that higher saving should push net worth back up again. Further, there are two reasons to expect that this rebuilding of wealth following the decline in home prices should have led to lower debt. First, lower household spending--on consumer goods and services as well as homes--would be generally associated with a reduced desire to borrow to finance such spending. Second, homeowners had a reduced capacity to borrow because the decline in home prices means that they had less collateral against which to do so.

The point is that traditional wealth effects alone should have led to an endogenous reduction in debt. Debt also probably fell endogenously for other reasons. In particular, weak income growth as well as uncertainty and pessimism about future income prospects likely damped consumer spending and, in turn, depressed the need to borrow to finance that spending.

However, some part of the decline in household debt may have been driven by the high levels of debt, leverage, and debt service themselves. Households that experienced a sharp increase in leverage when home prices declined might simply have been uncomfortable with being so leveraged. Other households may have felt the need to reduce debt because their debt-service obligations increased to unsustainable levels after low, time-limited "teaser" interest rates obtained at the time of mortgage origination expired. When home prices are stable or rising, borrowers with teaser rates can typically refinance into new low-interest-rate loans before the interest rate on the original mortgage increases, but borrowers that fell into negative equity with the recent home price decline would not have been able to do so. Further, households' discomfort with their recent level of leverage and debt-service obligations may have been exacerbated by the heightened probability of job loss; such a dynamic has been formalized recently in a model presented by Christopher Carroll, Jiri Slacalek, and Martin Sommer (2012).

Lenders' behavior--in conjunction with high levels of leverage--may also have contributed to the decline in household debt. Households with high leverage (and high debt-service obligations) generally have more difficulty obtaining loans from financial institutions and have had particular difficulty in recent years because of the sharp tightening of loan standards in the wake of the financial crisis. Although credit conditions have been gradually thawing over the past couple of years, the supply of credit remains considerably more restricted than normal, particularly for mortgages (see Bernanke 2012).

These considerations might help explain both the low levels of new borrowing and (in some cases) the high rate of loan default in recent years. (2) Further, some of these considerations--if they have weighed heavily enough on households--may have provided an additional motivation to reduce spending and raise saving beyond that related to changes in wealth and the other traditional determinants of consumption. Yet the empirical aggregate consumption functions used by many policy-oriented economists traditionally do not include debt or leverage as an explanatory variable, instead capturing balance sheet considerations solely by including aggregate net worth (see Dynan 2012). It is important to explore, then, whether high debt and leverage might be having an independent influence on consumption, in order to assess whether the traditional approach might be leading analysts seriously astray.

I.C. Previous Literature

Relatively little attention was given to household debt issues before the recent crisis. Much of the literature instead focused on whether credit constraints explained the excess sensitivity of aggregate consumption to aggregate income (see, for example, Ludvigson 1999). At the household level, Kathleen Johnson and Geng Li (2007) found that the consumption of households with low liquid assets and high debt-service burdens was more sensitive to changes in income than the consumption of households with low liquid assets alone. There was also some interest in the role of appreciating homes as collateral for borrowing-constrained households, particularly as home prices began to rise rapidly early in the 2000s (see, for example, Iacoviello 2005 and Disney, Bridges, and Gathergood 2010). But some research from previous decades did give heed to the possible role of household debt in economic downturns. For example, Frederic Mishkin (1977) argued that fears of excessive debt-service burdens induced a deleveraging that contributed to the severity of the 1973-75 recession.

Much more research has focused on household debt, particularly mortgages, since the financial crisis. This newer literature includes papers that look at the early rise in defaults among subprime borrowers (Mayer, Pence, and Sherlund 2009), the interplay between the borrower's choice to default and the lender's choice to modify the terms of the mortgage (Foote and others 2009), strategic defaults by underwater borrowers (Bhutta, Dokko, and Shan 2010), and the relationship between defaults and securitization (Keys and others 2010). The research has yielded a number of interesting and important findings.

Nearly all of this more recent work, however, has used mortgage records or credit bureau data. Those data sources have shed light on important issues regarding the crisis, but they have their shortcomings. Most notably, the background information about the debt holders is typically limited to what one would find on a loan application. Researchers have partly mitigated this problem by merging this information with additional data such as average income by zip code, but the potential for such merges is limited, and the information is still not household-specific. An important strength of the household survey data set used in my analysis is that it provides rich background information about the borrowers that I study. (3)

In addition, most of the past work has been backward-looking, aimed at exploring the causes of credit distress. There has been fairly little work that ties credit distress and, especially, deleveraging to economic activity. (4) Two notable exceptions are papers by Mian and Sufi (2011) and Mian and others (2011), which look at employment and spending patterns in U.S. counties with different degrees of leverage on household balance sheets. One limitation of these papers is that the counties with the most leverage also tend to be the counties with the largest home price declines, such that the degree to which these authors' finding of soft recent economic activity reflects a special deleveraging effect as opposed to traditional wealth effects is unclear.

II. Data Sources

Although macroeconomic data have been the basis for much casual analysis of deleveraging, such data have limited value for understanding the true linkages between household debt and consumption. My analysis will rely primarily on the Panel Study of Income Dynamics (PSID), the longest-running representative longitudinal survey of U.S. households. As I show, the information in the PSID is broadly consistent with macroeconomic developments in recent years. The data set also provides clear evidence of the central role that home prices played in explaining why households accumulated so much debt during the boom.

II.A. The Need to Use Micro Data

U.S. statistical agencies publish timely estimates of aggregate household debt and related variables at a quarterly frequency. However, these data have limited value for understanding the causes and consequences of the household debt crisis. In particular, the aggregate measures may not adequately capture important debt-related pressures in subgroups of the population. For example, the 2002-06 increase in aggregate household debt shown in figure 1, although more concentrated over time, was no larger in magnitude than the rise over the preceding two decades, which did not have particularly pernicious consequences. Indeed, the earlier rise in debt likely benefited households by allowing them to better smooth their consumption over the business cycle and over the life cycle. A key difference between the rise in debt in the early to mid-2000s and that in earlier decades was that the latter was fairly spread out across the population (Dynan 2009), whereas the former saw concentrations of households taking on very large amounts of debt (see, for example, Mayer and others 2009). Because the aggregate data essentially masked this trend, policy analysts who took the traditional approach of focusing on aggregate measures greatly underestimated the amount of risk building up in the financial system before the crisis.

[FIGURE 3 OMITTED]

Looking beyond the aggregate data is likely to be equally important in assessing the nature and implications of the deleveraging that has occurred since the bursting of the credit bubble. For example, the recent behavior of the aggregate personal saving rate might suggest that deleveraging is not an important force holding back consumer spending. The saving rate has, in fact, risen in recent years, from just below 1 1/2 percent at its low point in 2005 to an average of 4 1/2 percent over the past year (figure 3). But given the conventional wisdom that the marginal propensity to consume out of wealth is between 3 and 6 cents on the dollar (see, for example, Davis and Palumbo 2001), the decline in household net worth alone would predict an increase in the saving rate on the order of 4 to 8 percentage points--much higher than the realized increase. Absent other factors, deleveraging, if important, should have raised the saving rate yet further. Accordingly, skeptics argue that the observed rise in the saving rate is too small for deleveraging to have been an important force.

The problem with such arguments is that many factors are currently affecting consumption and saving, some of which--including low interest rates and consumption smoothing in the face of transitory disruptions to income--are probably boosting consumption and reducing saving. (5) Aggregate data likely offer too little variation to identify any independent effect that deleveraging might have had on consumer spending. In addition, because aggregate data do not provide a good picture of the financial situation of the most-indebted households, they also cannot tell us how much more deleveraging is to come. Given the importance of these issues for the prospects of the U.S. economy, it is essential to study deleveraging with household-level data.

II.B. The Panel Study of Income Dynamics

This paper uses household survey data from the PSID to examine the household-level underpinnings of the run-up in borrowing in the 2000s and the subsequent deleveraging. This survey collects a rich set of background information from its participants that can shed light both on what led some households into such precarious financial positions and on what, if anything, they have done subsequently to reduce debt and rebuild their net worth. The resulting findings are thus a complement to the existing body of micro data-based research on the mortgage crisis, nearly all of which is based on administrative financial data records where the background information is largely restricted to what is on a loan application.

Launched in 1968, the PSID is a panel survey of households conducted by the Institute for Social Research at the University of Michigan. Participating households were at first surveyed every year, but beginning with the 1997 wave, the frequency was changed to every other year. The most recently released full wave contains information from about 8,000 interviews conducted in 2009. In February 2012 a very limited set of preliminary data from the 2011 wave of the PSID was released, including the information needed to construct net worth as well as the results from a special module on foreclosures and mortgage payment problems. The survey's documentation cautions that these data are subject to revision, but given the importance of using timely information for the questions at hand, I make use of them in the analysis below.

The PSID contains fairly extensive information about mortgages on primary residences as well as loans used to finance motor vehicle purchases. Balances on other common types of household debt--such as credit cards, student loans, medical and legal bills, and loans from relatives--were reported as a group until 2011, when the questionnaire was changed to collect more detail. (6) The PSID also provides some information about assets and about net equity in businesses, vehicles, and second homes, so that one can create a limited measure of households' net worth. (7) Data on expenditures on food and a few other items are available for most waves of the survey; questions about many more categories of spending were added between 1999 and 2005, such that a broad (although still not complete) measure of consumption can be constructed for the 2005, 2007, and 2009 waves. The special module on foreclosures and mortgage payment problems first appeared in the 2009 survey.

I calculate the mortgage debt of each household as the sum of the balances of any first and second mortgages on its primary residence. I calculate consumer debt as the sum of outstanding balances on up to three vehicle loans (backed out from information on the original balance of the loans and the payment history) plus the reported sum of balances on other types of consumer loans. Debt-service obligations are derived from information on required loan payments, except for the "other loan" category, where, because much of this debt is presumably credit card debt, I follow the Federal Reserve's convention of assuming that the required monthly payment is 2.5 percent of the balance. (8) I calculate the net worth of each household as the sum of the values of its primary residence, its private annuities and individual retirement accounts, any other stocks or bonds or mutual funds, its bank accounts, and its net equity in businesses, vehicles, and second homes, minus the household's mortgage debt and its non-vehicle-related consumer debt.

For my income measure I use total family income before taxes, as information about after-tax income is not available. I set nonhousing consumption equal to the sum of spending on vehicles, vehicle-related items, gasoline, transportation, furniture, clothing, tuition, other school-related items, and food (both at home and away from home). One complication is that the time period over which the expenditures are supposed to be reported varies by category, from "an average week" to the previous month to the previous year. In constructing my nonhousing consumption measure, I adjust spending for all of the components to be on an annual basis and add them together, essentially ignoring the time mismatch. (9)

I augment the PSID data with state-level information in order to better capture the macroeconomic conditions faced by each household. Specifically, I add state unemployment rates as well as indexes of state home prices produced by CoreLogic, a private firm that collects and distributes consumer, financial, and property information.

One long-standing issue confronting researchers using household-level wealth data is the treatment of outliers. The distribution of wealth in the United States is highly skewed, with a long right tail. As a result, extreme values will tend to have an undue influence on means and on results from ordinary least squares regressions. Measurement error in wealth data and small sample sizes further reduce the usefulness of such analyses. For these reasons I focus mainly on medians in the analysis of summary statistics below and apply a transformation that downweights the influence of outliers in the regression analysis.

The calculations presented throughout the paper are based on weighting the PSID observations. I use the longitudinal weights provided by the PSID for the core sample of households combined with the additional sample of immigrant families.

II.C Summary Statistics on Household Balance Sheets and Consumption

Table 1 reports some summary statistics for the households in the sample. All responding households from each wave are included, although for some variables, such as consumption and net worth, some observations are missing because the household did not report full data for the variable or its components. As discussed above, I show medians instead of means so as to avoid the influence of unduly large readings, particularly for balance sheet variables. Accordingly, the results describe the experience of the typical household rather than the average household. Although the means, in principle, might be more telling about aggregate conditions, they are unlikely to be representative given the relatively small numbers of wealthy households in the PSID sample. (10)

The movements in the variables between waves of the PSID are broadly consistent with other information on economic developments in recent years. Median pretax family income rose between the 2005 and 2007 waves, corresponding to an increase in income between calendar years 2004 and 2006, and rose again (modestly) between the 2007 and 2009 waves, corresponding to calendar 2006 and 2008; aggregate personal income, as published in the National Income and Product Accounts (NIPA, not shown in the table), also rose over both time spans. (11) Median nonhousing consumption rose between 2005 and 2007 but reversed that gain between 2007 and 2009, falling 8 percent; aggregate NIPA consumption (not shown) also rose between 2005 and 2007 and then edged down between 2007 and 2009. The larger decline seen in PSID consumption may reflect the fact that the median household did worse than the average household during the recession; it may also reflect the fact that my measure excludes less discretionary items such as payments for utilities and health care. Median net worth rose considerably between 2005 and 2007 but plunged over the next 2 years; aggregate household net worth, as published in the Federal Reserve's Flow of Funds Accounts (not shown), showed the same general pattern, but with less pronounced changes.

The homeownership rate in the PSID sample fell between 2005 and 2007, and again between 2007 and 2009. The median reported home value rose 18 percent between 2005 and 2007 and then reversed two-thirds of that gain between 2007 and 2009. Although this pattern is consistent with households recognizing a boom and bust in home prices, the timing differs from that seen in direct data on home prices. The CoreLogic national index of home prices peaked in April 2006 and by mid-2007 was just 2 percent above its reading 2 years earlier. Between mid-2007 and mid-2009, the CoreLogic index declined 24 percent. At face value this comparison suggests that households adjust their views of the values of their homes with a lag, or perhaps are too optimistic in general. However, a more complete analysis is needed before one can draw strong conclusions. (12)

The credit cycle seen in aggregate data appears to some extent in the PSID data as well. The fractions of households in this sample holding mortgage debt and vehicle debt fell between 2007 and 2009. Median vehicle debt for households holding such debt also declined between 2007 and 2009, but median mortgage debt for households with mortgage debt increased over that period. The latter pattern may reflect new homeowners, who tend to purchase smaller homes, being shut out of the market--indeed, Neil Bhutta (2012) concludes that first-time homebuying has been very weak, especially for households with less-than-excellent credit scores. It may also reflect the possibility that the credit crunch was felt mostly in the tails of the household indebtedness distribution, at least at the beginning.

The table also shows some clear limitations of the data. In particular, the fraction of households holding mortgages, at just over 40 percent, is considerably below the roughly 50 percent figure that shows up in the Federal Reserve's Survey of Consumer Finances (SCF), which is widely considered to have the best available data on U.S. household balance sheets. Moreover, median net worth is only about half as large as in the SCF. The latter discrepancy likely owes in large part to the fact that the PSID collects data on only a limited part of total household wealth. Indeed, Barry Bosworth and Rosanna Smart (2009) present a thorough comparison of the PSID and the SCF and conclude that, once put on a comparable basis, the wealth measures in the two surveys are very similar through the 95th percentile of the wealth distribution.

Another issue raised by table 1 is that median nonhousing consumption seems low relative to median pretax income. Again, the explanation may be the limited scope of the questions. Li and others (2010) find that the information from many of the consumption categories added since 1999 compares favorably with estimates from the Consumer Expenditure Survey. In any event, to the extent that the movements in these partial measures are correlated with movements in more comprehensive measures, the results presented below should generalize to consumption as a whole.

II.D. Household Debt and Spending during the Credit Boom

To lay the groundwork for the analysis of deleveraging, I examine households' experience in the period leading up to the crisis, so as to shed more light on why households accumulated so much debt during the boom. Table 2 compares households in the top quintile of leverage as of 2007 with those with lower leverage. Given that rapidly rising home prices in some parts of the country were thought to have been a key precipitating factor for the credit crisis, I divide households into three groups that might be expected to have very different responses to this trend: nonhomeowners, homeowners residing in states in the top quartile of home price appreciation between 2000 and 2006 (henceforth called "boom states"), and homeowners residing in other states ("non-boom states")." Households with retired heads are excluded from the comparison because I normalize some variables by current household income, which is often hard to interpret for retirees. The sample is also restricted to households that had a complete set of interviews for the 2005, 2007, and 2009 waves of the PSID.

PRECRISIS PATTERNS OF LEVERAGE AMONG NONHOMEOWNERS The first two columns of table 2 present estimates for nonhomeowners. For this group I define leverage as total debt--which effectively means consumer debt such as auto loans, credit card balances, and student loans--divided by total assets. The median amount of consumer debt for highly leveraged nonhomeowners in 2007 was $20,000, whereas the median for other nonhomeowners was zero. Along many other dimensions, the two groups summarized in the first two columns look fairly similar. For example, median levels of pretax income and nonhousing consumption in 2007 were about the same, and both groups had a few hundred dollars of financial assets at the median. Given that the increase between 2005 and 2007 in the median ratio of consumer debt to income (middle panel of table 2) for the highly indebted group was only about one-quarter as large as the median 2007 ratio, it appears that many of the highly indebted have been so for a while. This result could indicate a chronic shortfall of self-control by some households, but it could also reflect fully rational behavior for households that expect their income to be much higher in the future because, for example, the household's primary earner works in a profession where income rises sharply with job tenure. Note that highly indebted nonhomeowners are considerably more likely to have a college degree than other nonhomeowners. More education might increase the use of credit by this group in two ways: first, because it is associated with greater access to credit, and second, because people with more education tend to have steeper age-income profiles, perhaps leading them to smooth their consumption by taking on large amounts of debt early in their careers.

PRECRISIS PATTERNS OF LEVERAGE AMONG HOMEOWNERS LIVING IN NON-BOOM STATES The third and fourth columns of table 2 correspond to homeowners living in states outside the top quartile of states ranked by home price appreciation during the boom. For all homeowners I define leverage as mortgage debt for the household's primary residence divided by the value of that residence, all as of 2007. Using this more restricted measure of leverage allows me to retain more households for the analysis, as there are many cases where full data on other types of assets and liabilities are not available. (14) Highly leveraged homeowners in non-boom states tended to be younger and to have (slightly) lower income and smaller homes than other homeowners in those states. At the median in these states, the consumption of the highly leveraged homeowners was a little higher relative to income than the consumption of other homeowners. Both the mortgage debt and the consumer debt of the highly leveraged group were considerably higher than those of other homeowners: at the median, their mortgage debt amounted to nearly 2 years' worth of income versus 8 months for less leveraged homeowners. For the highly leveraged homeowners, monthly debt obligations represented 28 percent of pretax income at the median, and roughly a fifth of them had become homeowners sometime in the preceding 2 years.

Relative to 2 years earlier, the homeowners in these non-boom states who were not highly leveraged seem to have been in a stronger financial position, with no increase in debt and an increase in median net worth of 26 percent of annual income. In contrast, the highly leveraged group experienced an increase in mortgage debt exceeding 3 months' worth of income at the median. Even for this group, however, the median ratio of net worth to income increased between 2005 and 2007, and the median ratio of consumer debt to income edged down.

PRECRISIS PATTERNS OF LEVERAGE AMONG HOMEOWNERS LIVING IN BOOM STATES The most striking comparison in table 2 is that between highly leveraged homeowners and other homeowners in the states with the largest housing booms, shown in the fifth and sixth columns. The highly leveraged households again tended to be younger, but they had considerably less pretax income at the median than households in the same group of states who were not highly indebted ($77,000 versus $93,000). Yet median nonhousing consumption for the highly leveraged households was somewhat higher (about $28,000 versus about $25,000 for other households). For both groups of households in these states, housing was a much more important part of the balance sheet than in states that saw lower rates of home price appreciation: median mortgage debt and home values in the fifth and sixth columns are about double the corresponding figures in the third and fourth columns.

Mortgage debt among highly leveraged homeowners in boom states grew sharply between 2005 and 2007, by an amount exceeding a year's worth of income at the median. In part, this pattern reflects new homeowners entering a housing market that was increasingly expensive. However, new homeowners (those who had purchased their homes since 2005) represented only a little more than a quarter of the highly leveraged households in the boom states. The remaining highly leveraged households likely increased their mortgage leverage by extracting equity through home equity lines of credit, by exchanging smaller mortgages for larger ones in "cash-out" refinancing transactions, and by taking on larger mortgages as they turned one home over for the next. (15)

Despite this increase in mortgage debt, the 2007 financial positions of highly indebted households in housing boom states likely seemed solid to those who did not anticipate the housing bust: median net worth had risen by 13 percent of income over the preceding 2 years, and the median ratio of mortgage balance to home value was 0.84. Note, however, that the typical highly indebted household in a boom state in 2007 had few financial assets compared with other households in those states, and their debt-service obligations amounted to 34 percent of pretax income in 2007--much higher than for households with less debt and for highly indebted households in non-boom states. Such a household would likely have trouble making mortgage payments if faced with an unanticipated disruption to income, but appeared to have a sufficient equity cushion to sell the home and pay off the mortgage should such a shock occur.

In boom states, the highly leveraged homeowners were slightly less likely to have a college degree than those with less debt--in contrast with the pattern for homeowners in other states and for nonhomeowners. (16) To the extent that less educated households are more likely to be lured into taking on precariously high levels of debt because of a lack of financial sophistication, one would expect the difference in median education between highly leveraged households and others to be the same for homeowners in boom and non-boom states and for homeowners and nonhomeowners (all else equal). However, credit access was probably higher for homeowners in boom states than for other people, because lenders believed that continued rapid home price appreciation would make it easier for households to meet their debt obligations. Together with a lack of financial sophistication on the part of less educated people on average, this effect would produce the pattern observed in these data.

THE RELATIONSHIP BETWEEN MORTGAGE LEVERAGE AND HOME PRICE APPRECIATION ACROSS STATES As already noted, the results in table 2 suggest that homeowners who did not expect home prices to fall sharply may have viewed themselves as in a fairly solid financial position as of 2007. Even the highly leveraged homeowners in states that had seen the largest home price booms appeared to be in decent financial shape under this assumption. Expectations of stable home prices, or perhaps even further appreciation, may well explain why those homeowners ended up having relatively high consumption and debt-service obligations, as well as low levels of financial assets. However, a cross-state analysis of leverage and home price appreciation illustrates that the financial situation of many homeowners would take a dramatic turn for the worse if home prices were to take back some of their earlier gains.

The three left-hand panels of figure 4 show actual 2007 mortgage leverage at various points in the distribution of households by leverage (median, 80th percentile, and 90th percentile) in different states, plotted against earlier home price appreciation in that state. The size of the circle corresponds to the state's population. I drop states for which I have 30 or fewer observations on the view that these results are less likely to be representative.

[FIGURE 4 OMITTED]

These three panels show a slightly negative relationship between actual mortgage leverage in 2007 and earlier home price appreciation. In other words, excessive mortgage debt appeared to be less of a problem in states that had experienced more pronounced housing booms. Indeed, in all but three of the states that saw home prices increase by more than 50 percent between 2000 and 2006, homeowners at the 90th percentile of reported leverage would have been able to withstand a 10 percent decline in the value of their home without going underwater.

The three fight-hand panels of figure 4 show what the patterns would likely have been in 2007 if home prices, after rising as they did from 2000 through 2006, had then fallen back to where they would have been had they risen only at the rate of consumer nonhousing inflation from 2000 onward. The numerator of this counterfactual loan-to-value ratio remains the 2007 level of mortgage debt. To construct the denominator, I first estimated what the value of each household's home would have been in 2000 if its appreciation between 2000 and 2007 had matched state-average home price appreciation, and then increased the estimated 2000 home value by the rate at which the "all items less shelter" component of the national consumer price index grew between 2000 and 2007. As can be seen, these counterfactual measures of leverage not only are much higher, but also increase strongly with the size of the home price boom in each state, particularly at the upper end of the distribution. Homeowners above the 90th percentile of leverage would have been underwater in most states and would have had leverage ratios exceeding 1.5 in 15 percent of the states.

All told, the results in this section show that the rise in household debt was concentrated both geographically and, within geographic areas, among a subset of homeowners. The rapid rate of home price appreciation in some parts of the country appears to have been centrally related to this increase in debt. The results are also consistent with the view that the most-indebted households may have ended up in a vulnerable situation because they did not appreciate the risk that home prices might take back some of their earlier gains.

III. Deleveraging and Its Consequences

I now turn to the question of what happened to highly leveraged households following the financial crisis and the onset of the recession. I begin by exploring how summary statistics for highly leveraged households com pare with those for households with less leverage, and then formalize the results with regression analysis.

III.A. The Postcrisis Experience of Highly Leveraged Households

The households in each column of table 3 are the same as those in the corresponding column in table 2. Households that were highly leveraged in 2007 are again compared with households that had less leverage in 2007, and I again show separate comparisons for nonhomeowners, homeowners in states that were in the top quartile of home price appreciation during the housing boom, and homeowners in other states, all as of 2007.

The top panel of the table shows the changes experienced by the different groups of households between the 2007 and the 2009 waves of the PSID. One feature that stands out is the greater decline in nonhousing consumption seen by the highly leveraged homeowners relative to their counterparts with less debt. This pattern is particularly evident in the housing boom states, where the consumption of the median household in the highly leveraged group fell by almost 15 percent--about twice as much as the median for other households. Notably, these larger declines occurred despite the highly leveraged homeowners seeing more income growth and smaller wealth losses than the less leveraged homeowners.

A more refined take on the question comes from comparing the relative movements of the ratios of nonhousing consumption (C) and net worth (W) to income (Y)--the rows shown in italic. For the less leveraged homeowners in housing boom states, at the median, C/Y declined by 0.04 and W/Y declined by 0.83. These figures suggest a marginal propensity to consume out of housing wealth of 0.04/0.83 = 0.05, in line with the estimates often cited by analysts and policymakers. For highly leveraged homeowners in housing boom states, conventional wealth effects would imply a decline in C/Y equal to the loss in wealth (0.67) multiplied by a typical estimate of the marginal propensity to consume out of housing wealth (0.05), or 0.03. But in fact, the median C/Y of these households declined by 0.07. It would thus appear that high mortgage loan-to-value ratios might have an additional, independent damping effect on consumption. To draw strong conclusions on this point, however, one should control for the various ways in which households that have a lot of leverage might be different from other households; I do so in the next section using regression analysis.

The results in the second panel of the table speak to how mortgage payment problems varied with 2007 leverage. Not surprisingly, highly leveraged households were much more likely by 2009 to have had problems or to anticipate having problems making their mortgage payments: in housing boom states, 19 percent of such homeowners were behind on their mortgage payments, versus 3 percent of other homeowners in those states. The comparable figures for states that saw less home price appreciation during the boom were 11 percent and 2 percent. Highly indebted households were also more likely to have experienced a foreclosure filing, to have had their mortgage modified, and to report being very or somewhat likely to fall behind on their mortgage payments over the coming year. In both boom states and non-boom states, more than a fifth of highly leveraged homeowners as of 2007 moved between 2007 and 2009, more than double the rate for homeowners with less leverage; 8 and 10 percent of highly leveraged homeowners in boom and non-boom states, respectively, had exited homeownership altogether. These figures suggest that an important way by which some highly indebted households reduced their debt was by downsizing or defaulting.

I noted earlier that, for some households, the strains of a heavy debt burden may manifest themselves primarily through high debt-service obligations relative to their incomes. In a similar analysis comparing such homeowners with other homeowners (results not shown), I found that the former also saw more pronounced declines in their consumption between 2007 and 2009. The prevalence of mortgage payment problems was the same or lower for households with high debt-service burdens as for highly leveraged households. However, households with high debt service in boom states seemed more likely to anticipate distress than households with high mortgage-to-home value ratios: nearly a third of the former reported being somewhat or very likely to fall behind on their mortgage payments over the coming year.

III.B. Formalizing the Results

The central question of interest is whether the overhang of housing debt is holding back consumption growth. In particular, I seek to answer whether consumption has shown more weakness than would be expected given the movements in its other fundamental determinants, including the loss in wealth, weak income, and pessimism or uncertainty about future income. The italicized results in table 3 support the notion that excessive leverage has had an important additional depressing effect on the consumption of some households. In this section I test the hypothesis more formally using regression analysis.

Section I.A reviewed the traditional determinants of consumer spending and highlighted why debt, leverage, and perhaps debt-service obligations might have an independent influence on spending. All told, those considerations suggest estimating the following equation:

(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [DELTA][C.sub.it] is the change in the consumption of household i in period t, [DELTA][W.sub.it] is the change in its wealth, [DELTA][Y.sub.it] is the change in its current income (relevant for households that are liquidity constrained or myopic), [(D/A).sub.i,t-1] is its leverage in the preceding period, and [(DS/Y).sub.i,t-1] is the fraction of its income going toward debt-service payments in that period. [X.sub.it] is a vector of other variables that might influence household consumption growth, such as the interest rate, economic conditions in the state, and demographic factors (which might be correlated with time preference, the risk of job loss, and revisions to expected future income). As in table 3, the changes represent differences between the 2007 and 2009 waves of the PSID. Both leverage and debt service are measured as of the beginning of the period over which the change is calculated, because presumably it is the household's ex ante level of balance sheet distress that is relevant for its consumption. If the debt overhang did hold back consumption growth between 2007 and 2009 above and beyond what would be typical given movements in the other determinants of consumption, one would expect to see negative coefficients on the debt-related variables.

Several complications present themselves. First, there are models that could produce a negative [[beta].sub.lev] or [[beta].sub.dsr] even in the absence of a separate channel related to the degree of leverage or debt burden. Most notably, if time preference rates vary across households, marginal propensities to consume might tend to be higher for low-wealth households because these households are likely to be more impatient. In this case one would expect low-wealth homeowners to show a larger consumption response to the home price bust. Since debt and, especially, leverage are highly correlated with wealth, the coefficients on these variables would then be biased downward. To shed light on whether my results are being biased by such effects, I also estimate regressions for the period 2005-07. Given that home prices rose, on net, over this period, one would expect to see positive coefficients on the debt variables in these specifications if households with low wealth simply have higher marginal propensities to consume; if the coefficients continue to be negative, the results are consistent with the view that high debt tends to damp consumption.

Second, the timing of the data is not ideal for estimating equation 2. Although home prices at the national level had been falling for about a year by the time the 2007 wave of the PSID was launched, they continued to decline rapidly for much of the period between the 2007 and 2009 waves. As a result, the ex ante measures likely understate the degree of debt overhang that may have induced some households to pare back their consumption between 2007 and 2009. However, it is undesirable to simply use ex post (2009) levels of debt variables in the regression: these levels may be correlated with the consumption change simply because debt is often used to finance consumption and (relatedly) because the debt variables are endogenous with respect to any deleveraging the household has done. For this reason I try splitting the sample according to whether the household resided in a housing boom state or not, because those states also tended to see the largest housing busts, such that leverage saw a sharper increase. I also try instrumenting 2009 levels of leverage with households' 2007 leverage and 2007-09 home price growth in their state.

Third, given the noisiness of household data, the small size of the PSID sample is likely to make the estimates imprecise, particularly in cases where I focus on just a subset of the sample. Using broad measures of leverage and debt burden would reduce the sample size considerably, because a number of households do not report all of the information needed to calculate total debt, total assets, or total debt-service obligations. Hence, I focus on mortgage-related measures of debt and assets, which are available for most households.

Finally, I follow a long tradition in the empirical literature on household-level consumption and finances by using a transformation that downweights large values; Carroll, Dynan, and Spencer Krane (2003) provide a formal justification for doing so by showing that the residuals from a linear regression using household data are far from normally distributed. Using log differences in equation 2 is not desirable, however, because it would require dropping households with negative wealth, a group highly relevant to the question at hand. Instead, I take the inverse hyperbolic sine of consumption, income, and wealth before differencing. For a variable [x.sub.it], the inverse hyperbolic sine is defined as

(3) log [[x.sub.it] + [([x.sup.2.sub.it] + 1).sup.1/2].

Except in the case of very small values, the transformed variable can be interpreted in the same way as a logarithmic variable (see Woolley 2011 and Pence 2006 for further discussion). The drawback to moving away from a linear specification, however, is that one cannot interpret the coefficients on the income and wealth changes as marginal propensities to consume. For this reason I also estimate some specifications that are more in the spirit of the calculation done for table 3, dividing the first difference of consumption, income, and wealth by average family income across the 2005, 2007, and 2009 waves of the PSID.
COPYRIGHT 2012 Brookings Institution
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2012 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:p. 299-328
Author:Dynan, Karen
Publication:Brookings Papers on Economic Activity
Article Type:Report
Geographic Code:1USA
Date:Mar 22, 2012
Words:9383
Previous Article:Fiscal policy in a depressed economy.
Next Article:Is a household debt overhang holding back consumption?
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters