This Selected Issues paper for the United States discusses the microeconomics of the country—household wealth and savings. Households’ consumption-saving decisions have an important bearing on the U.S. economic outlook. This paper demonstrates how households with consistently lower income, which have shown growth in the years prior to the crisis, experienced larger declines in their saving rates and a larger rise in their indebtedness before the crisis, contributing significantly to the dynamics of the mean saving rate.

Abstract

This Selected Issues paper for the United States discusses the microeconomics of the country—household wealth and savings. Households’ consumption-saving decisions have an important bearing on the U.S. economic outlook. This paper demonstrates how households with consistently lower income, which have shown growth in the years prior to the crisis, experienced larger declines in their saving rates and a larger rise in their indebtedness before the crisis, contributing significantly to the dynamics of the mean saving rate.

I. U.S. Household Wealth and Saving: the Micro Story Behind the Macro Dynamics1

Aggregate savings statistics reveal little about the types of households that drove down the U.S. saving rate before the 2008 crisis and its subsequent recovery. Using PSID micro data, this paper demonstrates that households with consistently lower income growth in the years prior to the crisis experienced larger declines in their saving rates and a larger rise in their indebtedness before the crisis, contributing significantly to the dynamics of the mean saving rate. Households with a larger share of total assets in housing and higher debt-to-income ratios raised their saving rates more sharply after the crisis, from depressed levels. The findings indicate that groups whose balance sheets were more adversely affected by the housing bust have made limited progress in rebuilding their net worth through active savings, suggesting that in the absence of asset price appreciation these households may wish to save more in the future.

A. Introduction

1. Households’ consumption-saving decisions have an important bearing on the U.S. economic outlook. In the years leading up to the 2008–09 recession, U.S. households played an important role in supporting U.S. and global growth by sustaining high levels of consumption. This development was mirrored by a decline in the saving rate of the aggregate household sector from 10 percent in the early 1980s to about 1 percent in 2005. The decline in the saving rate was facilitated by increasing credit availability and surging asset values—the equity price bubble in the second half of the 1990s and the house price bubble in the first half of the 2000s. The saving rate stopped declining in 2006, as the house price bubble began to deflate, and increased significantly during the recession. The pace at which output will recover going forward depends, in part, on the future saving behavior of U.S. households.

uA01fig01

Household Saving Rate and Wealth

Citation: IMF Staff Country Reports 2012, 214; 10.5089/9781475504910.002.A001

Sources: Bureau of Economic Analysis; and Federal Reserve Flow of Funds Accounts.

2. Average wealth and income figures on their own may not be sufficient to draw strong insights into the dynamics of savings and consumption. As of 2012Q1, the ratio of aggregate household net worth to disposable income (DI)—a key driver of the personal saving rate—stood above its pre-bubble historical averages. The recovery of aggregate net worth has however been mostly driven by the return of equity prices toward their pre-crisis levels, benefiting mainly upper-income households (as it is usually the higher-income households that own stocks). By contrast, housing wealth—traditionally the main saving vehicle for middle-income groups—remained almost 30 percent below its peak. Likewise, although the household debt-to-DI ratio has declined significantly, from 134 percent of DI in 2007 to 114 percent in 2012Q1, it remains higher than its levels before the housing bubble. Moreover, evidence suggests that the aggregate reduction in household debt was mainly driven by weak inflows and defaults.2 Thus, for a large share of households who are current on their mortgage debt, net worth could still be below desired levels.3 Recent evidence from the Federal Reserve’s 2010 Survey of Consumer Finances (SCF) also indicates that, despite the recovery in aggregate wealth statistics, the majority of U.S. households continue to struggle with real net worth levels that are below their mid-1990s levels. Accordingly, these households may continue to repair their balance sheets going forward through additional savings.4

3. An important question is how the various heterogeneities across households affect the dynamics of aggregate consumption and savings. The evidence on growing U.S. income and wealth disparities has raised new questions about the importance of differences between households for trends in aggregate data. For example, some analysts have argued that, the sluggish growth in real incomes during the decade preceding the crisis pushed middle-income families to borrow more to sustain their living standards. The increased indebtedness, in turn, was enabled by rising house prices and more valuable housing collateral. Following the crisis, researchers have investigated whether the pre-bubble growth in debt was concentrated at the “bottom” of the wealth and income-growth distributions. For instance, Kumhof and Ranciere (2010) document that the surge in household debt as a share of DI during the 2004–07 period was driven by the bottom 95 percent of the wealth distribution, and Mian and Sufi (2009) show that in the 2002–05 period, mortgage credit expanded more strongly in ZIP codes with lower income growth, while the opposite had been true in previous periods. Dynan (2012a) discusses how heterogeneities across households in terms of income, balance sheets, age, and the degree of liquidity and credit constraints may be affecting aggregate consumption dynamics, including since the 2008–09 recession. Obtaining insights on these issues calls for empirical evidence of how the characteristics of different groups of households have changed over time and the share of aggregate consumption and savings accounted for by the different groups.

4. This paper presents evidence on the balance sheet and saving heterogeneity across U.S. households, with the aim of better understanding the drivers of aggregate savings. We seek to characterize the types of households that depressed the aggregate saving rate in the housing-boom years (that is, between the 1999–2007 surveys) and those that accounted for its surge in the aftermath of the 2008 crisis. We focus on several aspects of heterogeneity. Following the literature that documents the increasingly uneven distribution of income and wealth in the United States, we describe how different segments of the income distribution contributed to the changes in the aggregate saving rate over time. We also document the experiences of households that had a larger share of their wealth in housing and households that entered the crisis with a higher level of debt. To the best of our knowledge, our study is the first that tracks household saving behavior and balance sheets through both the pre- and post- bubble periods, differentiating between income levels, growth rates, and other household attributes.

5. To study longer-term household trends, we make use of a well-established longitudinal dataset on households, the Panel Survey of Income Dynamics (PSID). As a panel survey collecting data from the same households over many years, it allows us to track household behavior over time, and condition our analysis on factors such as a household’s income growth over a given period. Repeated cross sections (if panel data were not available) would not allow for a direct measurement of how a given household responds to changes in its income or assets over time. The data collected also allow us to determine a given household’s saving out of its current income.5 In comparison, the SCF, which contains very comprehensive information on the income and wealth of U.S. households, only asks a qualitative question about whether the surveyed households save or not, but not their amount of saving. However, the PSID does not capture well the top and bottom tails of the income and wealth distributions (in contrast to the SCF that oversamples the wealthy and the Survey of Income and Program Participation of the Census Bureau that oversamples government transfer program participants, who are often poor). Comparing the mean statistics in the PSID to the corresponding aggregate measures from the NIPAs nevertheless allows us to gauge the importance of the tails of the distribution in driving the aggregate saving and balance sheet dynamics.

6. The findings point to significant heterogeneity in saving rates and balance-sheet repair. Households experiencing lower income growth during 1999–2007 saw a sharper decline in their saving rates and a larger rise in their indebtedness before the crisis, contributing significantly to the decline in the overall saving rate. These households were less able to reduce their debt and raise their net worth after the crisis. Households that had a larger share of their wealth saved in their primary residences during 1999–2007 saw their saving rate rebound sharply between 2009 and 2011. Nonetheless, the saving rate of these households during the 1999–2011 period remained well below the saving rates of those less reliant on housing as a store of wealth. The findings suggest that an important share of U.S. households may continue to seek to rebuild their net worth through active savings.

7. The rest of the paper is organized as follows. Section II introduces our dataset and discusses where the mass of our sample lies within the U.S. income and wealth distribution. It also presents the mean saving rates and wealth ratios for the PSID sample, comparing them with aggregate saving and wealth levels for the U.S. household sector. Sections III and IV explore the importance of income growth and housing in explaining the decline in the saving rates during the housing boom years. Section V examines whether the households that entered the recession with higher debt burdens subsequently experienced a sharper correction in their saving rates. Section VI concludes.

B. Data

Structure and Content of the Dataset

8. The paper uses data from seven PSID survey waves between 1999 and 2011. The PSID is a longitudinal survey that follows a sample of households taken in 1968 and their offspring. As such, the PSID does not capture the immigrant population, but is thought to mimic the dynamics of the aggregate population reasonably well. We focus on the 1999 to 2011 sample because there are wealth and saving data available in every wave. The saving data in the 2001, 2003, 2005, 2007, 2009, and 2011 waves cover the preceding two years, while the saving data in 1999 cover the preceding five-year period, 1994–98. The most recent data, from the 2011 survey, are preliminary with the final results expected to be released in late 2012.

9. The PSID includes a module on wealth which allows us to compute active saving—defined as the net purchase of assets. Since 1999, the PSID survey inquires about the active saving of households in its wealth module. The PSID categorizes wealth into eight components: (1) main home equity, (2) other real estate equity, (3) equity in private business or farms, (4) net worth of vehicles, (5) checking and savings accounts, money market mutual fund accounts, certificates of deposits, government saving bonds, treasury bills, including those in investment retirement accounts, (6) equities in publicly traded corporations, mutual funds, investment trusts, and investment retirement accounts, (7) other assets—corporate bonds, rights in a trust or estate, cash value of life insurance, and valuable collections, and (8) total non-collateralized debt which includes credit card debt, student loans, and other unsecured debt. When applicable, households are asked to report net wealth subtracting debt that is collateralized by the specific asset in question (applicable to categories (1), (2), (3) and (4), although households do report their gross assets and debt for main home equity, (1)). The PSID also includes a separate pension module that inquires about saving in private pension accounts and the wealth held in private defined-contribution pension accounts. We do not include pension saving in our analysis since the results of the 2011 pension module remain unpublished. But data available for 1999–2009 suggest mean saving in pensions to be fairly stable over time; hence their inclusion in the analysis is unlikely to affect our main findings. We follow Juster et. al. (2005) in the way we calculate active savings. For wealth categories with potentially large capital gains (i.e., (1),(2),(3), and (6)), active savings are computed as the difference between the amount invested and the amount removed or debt repaid between two periods. For example, active saving in the main home is computed as the value of improvements in the house plus the decrease in the main home mortgage debt for the year(s) a household does not move and the change in net equity in the main home in the year(s) it moves. For wealth categories where capital gains are not important, active savings are computed as simply the change in net wealth between two periods.

10. The timing of the variables varies. The families are interviewed early in the year, with all interviews completed by the middle of the year. Income is recorded for the calendar year preceding each survey. Household saving data are reported for roughly the previous two years from the 2001 survey onwards and for the previous five years in the 1999 survey. Wealth stocks are recorded as of the time the survey is conducted—generally at some point in the first half of the survey year. For instance, in the 2005 survey, wealth is recorded as of the interview date, in early 2005, income is recorded for 2004, and total savings are recorded over 2003, 2004, and the part of 2005 that precedes the interview.

Representativeness of the Survey

11. About seventy percent of the households in the PSID sample fall into the middle three quintiles of the U.S. income distribution. After dropping outliers and households that lack key information, we end up with roughly 3500 household observations per year.6 The first panel of Table 1 compares the distribution of income reported in the 2007 PSID survey (that is, 2006 income) with the distribution for the same year in a comprehensive dataset built by the Congressional Budget Office (CBO).7 The comparison is based on before-tax income, which is measured somewhat differently in the two datasets.8 The first row shows the quintile cutoffs from the CBO tables. The second line shows the percentage of households in the PSID that fall within these income brackets. Taken at face value, this exercise suggests that 69 percent of the PSID sample lies in the middle three quintiles of the CBO distribution, with less than 20 percent of the PSID sample falling into the top and bottom quintiles. Given the differences in measurement, which could be particularly important at the lower and upper tails of the distribution, we do not draw precise inferences—but conclude that the bulk of the PSID households are “middle-income” families.

Table 1.

Summary of Income and Wealth Distribution (current dollars, unless otherwise noted)

article image
Sources: Board of Governors of the Federal Reserve System; Congressional Budget Office; Institute for Social Research, Survey Research Center, University of Michigan; and Authors’ estimates.

CBO refers to the Congressional Budget Office; PSID refers to the Panel Survey of Income Dynamics; and SCF refers to the Survey of Consumer Finances.

Censored at top 1 percent. The CBO values are adjusted to reflect current dollars in 2006.

The family income is adjusted by the family size, similarly to the one reported by the CBO, by dividing the income by the square root of the family size.

12. The PSID also represents the wealth holdings of middle- and upper-middle income households reasonably well. A comparison is carried out between the 2007 PSID and the Federal Reserve’s 2007 SCF.9

  • The mean incomes for the bottom 9 deciles of the income distributions are somewhat higher in the PSID than in the SCF, while average income for the top decile is significantly higher in the SCF, indicating that the PSID households mostly fall into the middle three quintiles plus the ninth decile of the SCF income distribution.10,11

  • The bottom four panels of Table 1 compare households’ holdings of specific asset categories by income quintile in the SCF and PSID (with the upper quintile broken into two, in line with the SCF tables). The upper 10th decile of the SCF income distribution has significantly higher average holdings of assets and debt compared with the top 10th decile of the PSID, reflecting that the SCF over samples-the top income groups (which have a lower response rate) to ensure an accurate representation, while the PSID does not capture many households with very high incomes and wealth. The second panel of Table 1 compares direct holdings of stocks (that is, excluding indirect holdings through IRAs and other retirement accounts) between the SCF and the PSID. It shows that stock holdings are modestly lower in the PSID’s bottom 3 quintiles, despite somewhat higher incomes, and significantly lower for the top 2 quintiles. Consistent with higher mean incomes, however, mean value of primary residences and mortgage debt (third and fourth panels) tend to be higher in the PSID for the bottom 4 quintiles of the income distribution. The last panel shows that mean values of non-mortgage debt are very similar for the bottom four quintiles. It is not surprising that the PSID does not match the asset holdings of higher income households when compared with the SCF given that the SCF oversamples high income households, and thus better captures their average balance sheet holdings.

Average Savings and Wealth in the PSID

13. Table 2 presents the means of income, net wealth, and active savings from the PSID wave periods under study. Disposable income is presented for the preceding year; 2011 income is imputed using households’ reported 2008 income and data on state-level income growth from 2008 to 2010, since the preliminary 2011 data release did not include information on income. 12 Saving rates are obtained by scaling savings over the previous two years by twice the income in the previous year. For balance-sheet variables, we present the values from the survey year (that is, as of the date of the interview).

Table 2.

Means of PSID household income, savings, and net worth (current U.S. dollars, unless otherwise indicated)

article image
Notes: The top and bottom 1 percent observations of disposable income and non-housing consumption are trimmed, for net active savings the top and bottom 50 observations are censured. Homeowners that move between surveys and households with missing home ownership information in the current and lagged survey are dropped. Wealth stocks and IRA greater than $ 500 thousands are censured. If home equity for a hosuehold is missing, other housing related wealth variables are ignored.

Calculated as the sum of net active savings in housing, financial assets (cash and deposits, bonds, stocks, and IRAs, other debt), vehicles, other real estate and business.

Calculated as mean net active savings as a percentage of mean disposable income.

Computed as mean of variable to mean DI of the same year multiplied by 100.

Excludes half of consumer durables (part of tangible assets), foreign bonds and deposits, and trade payables to enhance comparability w ith the PSID.

Obtained from the PSID’s Pensions Module. Pension savings are recorded for the survey year, unlike the net active savings in other instruments from the Wealth Module (for the preceding two calendar years from the 2001 surbey onwards, and for the preceding five years in the 1999 survey).

14. The PSID exhibits a boom-bust cycle in net wealth and movements in disposable income that are in line with aggregate statistics. As expected, given the under-representation of very wealthy households, both mean disposable income and net wealth (NW) in the PSID are lower than the comparable measures in the National Income and Product Accounts (NIPAs) and the Flow of Funds accounts (FoF). However, the PSID and NIPA growth rates of household DI are similar in most periods (an exception is the 2003 survey, for which the PSID has significantly lower DI growth compared to the NIPAs).13 Consistent with the decline in house values from mid-2006 onwards and the decline in stock prices in 2008, average net wealth fell by around 34 percentage points of DI between the 2007 and 2009 surveys.14 Average net wealth rebounded by 22 percent of DI between the 2009 and 2011 surveys, driven by a rebound in financial wealth (the timing of the 2009 and 2011 surveys largely coincided with the post crisis trough and subsequent peak of the S&P500 stock price index, which rose by some 55 percent between mid-2009 and mid-2011).

15. Overall, the mean saving rate in the PSID exhibits the decline in the boom years and an increase after the mid-2000s, particularly in the 2011 wave.15 Rows three and four in Table 2 present statistics on household saving. Four observations stand out:

  • The mean PSID saving rate is generally lower than the personal saving rate in the NIPAs. The shortfall is particularly large in the 2003, 2005, and 2009 surveys. 16 Two factors help to explain this. First, employee savings under private pension plans are excluded from PSID active savings but not the NIPAs. Available data from the PSID’s pension module suggests that pension savings fluctuated in a narrow range of 0.9–1.2 percent of DI in 2001–09, thus the pension data can explain some of the difference between the NIPA and the PSID saving rate levels. Second, as documented by Dynan, Skinner, and Zeldes (2004), and consistent with our findings described below, higher income households have higher saving rates, which would drive the aggregate NIPA saving rate higher than the PSID mean. In that sense, the mean PSID saving rate can be seen as an indicator for the saving rate of middle-income households, while aggregate statistics are, to some extent, driven by top-income households.

  • In broad terms, the mean PSID saving rate follows the underlying (U-shaped) time series profile of the NIPA household sector saving rate—with a significant decline during the early boom years (between the 1999 and 2005 waves) and an increase between the 2005 and 2011 waves. However, some of the higher frequency movements in the PSID in the 2007 and 2009 surveys are at odds with those in the NIPAs. Some of the differences are likely due to different measurement of vehicle saving—the drop-off in auto purchases in 2008 would be captured as a higher depreciation of the auto-stock and lower saving in the PSID, whereas it would be captured as lower consumption and higher savings in the NIPAs. The differences are further discussed in the next two bullets.

  • The PSID mean saving rate declined in 2007–08, in contrast to the rise in the aggregate NIPA saving rate over the same period. The decline is particularly sharp for active saving in cash and deposits, which households should be able to report accurately, and vehicles (the details are not reported in Table 2, but are available upon request). This finding suggests that middle-income families dipped into their savings to sustain consumption at the peak of the crisis, perhaps not foreseeing the depth and persistence of the coming decline in incomes. This surprising aspect of the dataset is very robust; many sub-groups of households exhibit this pattern (as revealed later when we look at the saving rate dynamics for different sub-groups). The recently released 2010 SCF has also revealed that the share of households that are able to save out of their income declined to 52 percent in 2010 from 56.4 percent in 2007. Thus, the PSID saving data could be picking up the fact that fewer households were able to save any of their income from 2007 onwards.

  • Based on preliminary data from the 2011 survey, the mean PSID saving rate increased significantly in 2009–10. At the same time, both mean net worth and mean net housing equity in 2011 (as a ratio to mean imputed DI for 2011) remained not only well below their 2007 levels, but also were slightly lower than their 1999 levels (second panel of Table 2). In nominal dollar terms, net worth was only slightly above its level in 2005, while net housing equity was well below its 2005 level, implying a substantial erosion in real terms (consumer prices increased by some 15 percent between early 2005 and 2011). The mean debt-to-DI ratio in 2011 was also higher than in any survey between 1999 and 2007, although in nominal terms, mean debt per household declined between the 2009 and 2011 surveys by about US$1500. Taken together, these findings highlight the extensive damage that the housing bubble and financial crisis have caused for the balance sheets of middle-income households. The rise in the saving rate is consistent with households saving more in response to increased uncertainty about future economic conditions. The increased saving is also consistent with households saving to rebuild their net worth, as the economy improved, given the dramatic loses in wealth due to falling house prices. Whether saving remains high going forward will depend on households’ desire for precautionary saving and continued balance-sheet rebuilding.

C. Saving Behavior Before the Crisis—Did Income Growth Matter?

16. Figure 1 highlights an aspect of the housing boom that has been much discussed but not explored empirically: it represented growing dissaving by not only low-income but also low-income-growth households. Figure 1 compares patterns in saving rates for households ranked by their income growth between 1999 and 2007 and grouped in terciles. Based on this cut, differences in income growth between the terciles are very large. While the lowest tercile saw an average decline in real annual DI by about 8 percent between 1999 and 2007, the upper group experienced an average increase of 16 percent. The top left chart in the panel shows a striking difference in the saving behavior between the tercile with the lowest income growth and the two other terciles between the 2001 and 2005 surveys. The saving rates of the three groups were broadly similar in 1999 (about 3–4 percent). The saving rate of the lowest income-growth tercile declined from around 4 percent in 1999 to about -2 percent in 2005. The saving rates of the middle- and high-income groups also declined, but by much less—from 3–4 percent in 1999 to 1–1.5 percent in 2005. The decline of the saving rate for the low income-growth group was thus roughly twice as large as the declines for the other two groups during the boom years. The qualitative behavior of the saving rate was broadly similar across groups between 2007 and 2011. However, while the saving rates of the two higher- income growth terciles returned to their 1999–2001 levels of 3–4 percent by 2011, the saving rate lower income-growth tercile remained relatively low, closer to 1 percent.

Figure 1.
Figure 1.

Saving Rates by Income Growth over 1999-2007

Citation: IMF Staff Country Reports 2012, 214; 10.5089/9781475504910.002.A001

Sources: PSID; and Authors’ estimates.Notes: Low-, Medium-, and High-growth refer to the bottom, middle, and top terciles of income growth between 1999-2007. Saving rates are the mean savings for the tercilescaled by the mean disposable income of the tercile, Mortgage debt is mean debt scaled by mean disposable income for the tercile. Contributions to mean savings are calculated by multiplying the mean saving rate of the tercile by the share of the tercile in total income. (for the overall sample). House price growth is mean house price growth for the tercile.

17. The declining saving rate among households that experienced low income growth also had a material impact on the overall mean saving rate. The top-right chart in Figure 1 shows the contribution of each group to the overall mean saving rate for the households that were included in this analysis, to gauge how much the group with lowest income growth mattered to the overall dynamics. Interestingly, the contributions of the three groups to the mean saving rate were similar in 1999. While the two groups with higher income growth each contributed around 1 percentage point to the 4 percentage point drop in the overall saving rate between 1999 and 2005, the group with the lowest income-growth contributed around 2 percentage points—half of the overall decline. This evidence suggests that households experiencing sluggish income growth made a meaningful contribution to the decline in the U.S. personal saving rate during the housing-boom years.

18. Although low-income growth households experienced similar house price trends as other groups, they experienced a greater increase in mortgage debt as a share of disposable income. The bottom-left chart in Figure 1 shows that (self-reported) house price growth was not stronger for the low-income growth households, suggesting that differential house price appreciation was not a driver behind the saving rate differences. The bottom-right chart compares mortgage debt as a percentage of disposable income and shows that it followed different trends across the groups. Between the 1999 and 2005 surveys, mortgage debt decreased as percentage of disposable income for the top two terciles with higher income growth, with a prominent decrease in the debt of the highest tercile. By contrast, mortgage debt increased as a share of disposable income for households with the lowest income growth.17 Since 2007, mortgage debt relative to income has declined for the group of low income growth households, while it has risen somewhat for the other two groups.

19. Households with lower average income levels during the 1999–2007 period also had lower saving rates, but their contribution to the overall mean saving rate was small (not shown).18 Unlike households experiencing low income growth, households that had lower average income levels did not experience a marked decline in their saving rates between 1999 and 2007—their saving rates were consistently lower than those of higher income households in every period.19 For instance, in the 2005 survey (when the mean PSID saving rate bottoms out), the saving rate of the bottom tercile of the distribution of average 1999–2007 income levels was -3.1 percent, while the tercile with highest average income had a mean saving rate of 1.8 percent. As expected, households in the bottom tercile play a limited role in influencing the overall mean saving rate, given their relatively lower income and saving levels.

D. The Role of Housing

20. Changes in active net savings in primary residences explain a significant share of the dynamics of the mean PSID saving rate over 1999–2011. Saving in households’ main home contributed 1 percentage point of the 4 percentage points decline in the saving rate between the 1999 and 2003 surveys, and nearly 3 percentage points of the roughly 4 percentage point increase in the saving rate between 2005 and 2011. This pattern is consistent with households tapping home equity loans for consumption during the housing boom and losing access to new mortgage credit—both for new homes and for home equity loans on existing homes—in the aftermath of the bubble (see Cooper, 2011, and Bhutta, 2012). Other large contributors to the decline and subsequent rise in the overall mean savings rate were non-mortgage debt and gross financial assets. In particular, the increase in non-mortgage debt contributed about ¾ percentage point to the nearly 4 percentage points decline in the saving rate between the 1999 and 2005 surveys, and ½ percentage point to the 4 percentage points increase between 2005 and 2011 surveys (not shown separately). This pattern is in line with the generally tighter credit conditions facing the household sector since the financial crisis.

uA01fig02

Changes in the Saving Rate: Contributions by Instrument

Citation: IMF Staff Country Reports 2012, 214; 10.5089/9781475504910.002.A001

Sources: PSID; and Authors’ estimates.
Figure 2.
Figure 2.

Saving Rates for Households with Varying Dependence on Housing Wealth

Citation: IMF Staff Country Reports 2012, 214; 10.5089/9781475504910.002.A001

Source: PSID; and Authors’ estimates.Notes: Saving rates for the terciles of the distribution ranking households by the average share of the value of their house in their total assets in 1999–2007, and the contributions of the terciles to the mean PSID saving rate.

21. Households more dependent on housing wealth in 1999–2007 had lower saving rates in general and raised their saving rates sharply in 2011. Lovenheim (2011) and Dynan (2012b) show that the changes in average house values in the PSID line up reasonably well with the aggregate house price data (see footnote 14). Even if households measure the value of their house with error though, they are likely to judge reasonably accurately the relative importance of the value of their house in their total wealth. The tercile of households for which housing wealth accounted for the highest share of total assets in 1999–2007—typically the households with lower wealth and income—had a mean saving rate close to 2 percent in 1999, and lowered it sharply to about -3.5 percent in 2003, where it stayed until 2009. In 2011, the saving rate of this group rebounded to about 2 percent. The group with medium dependency on housing wealth also saw a sharp decline in 1999–2005 but had already raised its saving rate by 2007. In contrast, the households with the lowest housing-dependency lowered their saving rates more modestly in 1999–2005 and actually lowered their saving rates in 2009, possibly reflecting their greater ability to smooth consumption during the crisis. The tercile of households most dependent on housing assets, and thus most vulnerable to a house price downturn, did make a meaningful contribution to the downward trend in the mean saving rate during the boom years.20 With its active saving rate returning to its 1999 level in 2011 and house prices stagnant, this group has made modest progress in rebuilding its net worth.

E. Did Households with Higher Debt Burdens in 2007 Become More Thrifty After the Crisis?

22. Households that entered the recession with the highest debt ratios experienced a sharper rise in their saving rate in its aftermath. Some analysts have argued that high debt stocks—and not just wealth losses—may have been weighing on the growth of private consumption during the recovery from the recession (see, e.g., Dynan, 2012b). Figure 3 shows that households with the top quintile of debt-to-DI ratios as of 2007 had a very sharp rebound in their saving rates of some 7 percentage points, on average, between the 2009 and 2011 surveys (after lowering their saving rates further between 2007 and 2009). The contribution this group made to the increase in the saving rate between the 2007 and 2011 surveys was also substantial, at close to 1.5 percentage points. Figure 3 highlights that highly-indebted households, as a group, contributed more to the increase in the saving rate between 2009 and 2011 than each of the bottom four quintiles. The question remains whether these households raised their saving rates voluntarily due to increased economic uncertainty and/or to pay down their debt, or whether they were forced to raise their savings due to tightened credit availability. The exact mechanisms behind this dynamic could have important implications for the future behavior of savings and consumption growth.

Figure 3.
Figure 3.

Saving Rates and Indebtedness in 2007

Citation: IMF Staff Country Reports 2012, 214; 10.5089/9781475504910.002.A001

Sources: PSID; and Authors’ estimates.

F. Conclusion

23. Households with lower income growth, higher dependence on housing, and high debt levels prior to the recession exhibited different saving behavior than other groups. Households with low income growth over 1999-2007 experienced steep declines in their saving rate in the years leading up to the crisis and more volatile saving rates in the years surrounding the crisis, contributing significantly to the changes in the overall saving rate. Households that were relatively more dependent on housing in 1999-2007 had lower saving rates over the entire 1999-2011 period, and did not raise their saving rate until 2011. Households that had the highest debt-to-DI ratios in 2007 experienced a substantial decrease in their saving rates until 2009, and a sharp increase in 2011. Such households have made relatively limited progress in rebuilding their net worth through actively saving part of their incomes since the crisis. To the extent these households wish to improve their balance sheets going forward, they need to increase their saving rates from their current levels if the valuations of their assets do not rebound and/or their income growth does not pick up. At the same time, any increase in saving by the lower and middle-income households could potentially be offset by reduced savings by higher-income households, if the latter start saving less out of their incomes as financial asset prices recover further and macroeconomic uncertainty diminishes. Our results suggest that the share of saving and private consumption contributed by top-income versus middle-income households is an important area for future data collection efforts and empirical research.

Appendix I. Distribution of Households Across the Income, Housing-Dependency, and Debt-to-Income Ratio Categories

1. Table A1 presents the bivariate frequency distributions of the households across the terciles used in the exercises shown in Figure 1 and Figure 2, and the quintiles shown in Figure 3.1 There is a fair amount of variation between the types of households that fall into the various groups—that is, we don’t find a strong correlation between the household characteristics we focus on. For instance, low income growth households as well as the groups with higher income growth are all quite uniformly distributed in terms of their housing-dependency—households with low income growth are not necessarily more dependent on housing. Low income growth households tend to have higher debt-to-DI ratios in 2007 than the households with higher income growth, which is not surprising given their lower mean saving rate. But the differences are modest; 24 percent of low-income growth households are in the top quintile of debt-to-income ratios, as compared to 17–19 percent of medium- and high-income growth households.

Table A1.

Distribution of the households across the income-growth, income-level, housing-dependency, and debt-to-income ratio groups

article image
Source: Authors’ calculations.Notes: The table shows the share of households that fall into the categories denoted in a given row and column. The shares in each box add to 1. The sum of the shares across a row (column) gives the overall share of households in the group denoted by the row (column). For instance, in the sample underlying the second box from the left on the top row, 14 percent of the households have both low income levels and low income growth, w hile 33 percent have low income growth (14+12+7). The row and column totals are 33 percent in all boxes except the ones associated w ith housing-dependency, which is calculated only for households that own their main home. Homeownership differs across income groups and hence the housing-dependency related rows (columns) do not always add to 33 percent of households.

2. Income levels are somewhat correlated with income growth and the dependency on housing, but not necessarily with debt levels. Not surprisingly, households with higher income growth over 1999–2007 tend also to have higher income levels over the same period. Housing tends to account for a larger share of assets among lower income households; more than half of low income households are in the top tercile of housing dependency, compared to about 20 percent for high-income households. The middle-income households account for a larger share of the top-quintile of debt-to-DI in 2007 compared with the higher- and, in particular, lower-income households, but the differences are not large.

References

  • Bhutta, Neil, 2012. “Mortgage Debt and Household Deleveraging: Accounting for the Decline in Mortgage Debt using Consumer Credit Record Data.” Finance and Economics Discussion Series 2012–14, Board of Governors of the Federal Reserve System.

    • Search Google Scholar
    • Export Citation
  • Bricker, J., Kennickel, A., March, T., Moore, K., and J. Sabelhaus, 2009. Changes in U.S. Family Finances from 2007 to 201: Evidence from the Survey of Consumer Finances. Federal Reserve Board Bulletin, Vol. 98.

    • Search Google Scholar
    • Export Citation
  • Bucks, B., Kennickel, A., March, T., and K. Moore, 2009. Changes in U.S. Family Finances from 2004 to 2007: Evidence from the Survey of Consumer Finances. Federal Reserve Board Bulletin, Vol. 95.

    • Search Google Scholar
    • Export Citation
  • Congressional Budget Office, 2011. “Trends in the Distribution of Household Income Between 1979 and 2007.”

  • Cooper, Daniel, 2009. “Impending U.S. Spending Bust? The Role of Housing Wealth as Borrowing Collateral.” FRB Boston Public Policy Discussion Paper Series, No. 099.

    • Search Google Scholar
    • Export Citation
  • Dynan, Karen, 2012a. “What’s Been Weighing on Consumption? An Overview of the Recent Experiences of Different Types of Households.” Brookings Institution, mimeo.

    • Search Google Scholar
    • Export Citation
  • Dynan, Karen, 2012b. “Is A Household Debt Overhang Holding Back Consumption?” Brookings Papers on Economic Activity.

  • Dynan, Karen E., Jonathan Skinner and Stephen P. Zeldes, 2004. “Do The Rich Save More?,” Journal of Political Economy, 2004, Vol. 112 (2, April), pp. 397444.

    • Search Google Scholar
    • Export Citation
  • Kennedy, James, 2010. “Charge-offs of One-to-Four Mortgage Debt in the Fourth Quarter of 2010,” March 2010, unpublished memo, Board of Governors of the Federal Reserve.

    • Search Google Scholar
    • Export Citation
  • Kumhof, Michael, and Romain Ranciere, 2010. “Inequality, Leverage, and Crises”. IMF Working Paper, 10/268.

  • Kochhar, Rakesh, Richard Fry, and Paul Taylor, 2011. “Wealth Gaps Rise to Record Highs Between Whites, Blacks, and Hispanics,” Pew Research Center, Social and Demographic Trends, July 2011.

    • Search Google Scholar
    • Export Citation
  • Lovenheim, Michael, 2011. “The Effect of Liquid Housing Wealth on College Enrollment,” Journal of Labor Economics, Vol. 29, No. 4, October 2011.

    • Search Google Scholar
    • Export Citation
  • Mian, Atif, and Amir Sufi, 2009. “The Consequences of Mortgage Credit Expansion: Evidence from the U.S. Mortgage Default Crisis”. Quarterly Journal of Economics, Vol. 124, No. 4.

    • Search Google Scholar
    • Export Citation
  • Sommer, Martin and Jirka Slacalek, 2012. “What Drives Household Saving? Examining the Role of Target Wealth”. Mimeo.

1

Prepared by Oya Celasun (WHD), Daniel Cooper (Federal Reserve Bank of Boston), Jihad Dagher (WHD), and Rahul Giri (ITAM, Mexico). The views expressed in this paper do not necessarily indicate concurrence by members of the research staff or principals of the Federal Reserve Bank of Boston, or the Federal Reserve System.

2

Kennedy (2010) finds, using aggregate data, that roughly two-thirds of the reduction in household debt between 2008 and 2010 had been through charge-offs. Bhutta (2012) uses data on individual credit records to show that the decline in mortgage debt was to a large extent driven by weak inflows (given historically weak first-time home-buying) rather than outflows (such as through pay-downs and foreclosures). However, he also documents that on the outflows side, borrowers generally are not paying down their mortgage balances more aggressively than in the past, suggesting limited decreases in mortgage debt for households with an existing mortgage balance.

5

The Survey of Consumer Finances released by the Federal Reserve in June 2012 confirms that all income deciles except the highest had mean net worth levels in 2010 that were below their 2004 levels in current dollar terms (Kennickell et. al., 2012, Table 4).

4

Using aggregate data and state-space methods, Sommer and Slacalek (2012) estimate target net wealth at 525 percent of disposable income at end-2009. Actual net wealth was about 510 percent of disposable income at end-2009, and stood slightly below 500 percent of disposable income at the end of 2011.

5

The PSID also contains information on household consumption in addition to data on household wealth and income. By contrast, the BLS’ Current Population Survey (CPS) collects data on income for individuals; the Consumer Expenditure Survey (CEX) collects data on consumption, income, and some information on asset holdings for households; and the SCF collects data on wealth and income for households. Thus, the PSID is the only survey that combines information on household saving and wealth.

6

Following Juster et. al (2005) and Cooper (2011), the top and bottom 1 percent of the income and consumption distributions are dropped to limit outliers. So are the households with missing homeownership information in the current and previous surveys (only current survey for 1999), homeowners that have moved since the last survey(as they could potentially misreport passive capital gains/losses as active saving/dissaving—as appears to be the case in the raw data for 2009), and homeowners whose mortgage debt is more than twice as large as the value of their house.

7

The CBO dataset combines data from the Statistics of Income, a nationally representative sample of individual income federal tax returns collected by the Internal Revenue Service, with data from the Census Bureau’s Current Population Survey (CPS), which ensures coverage for the lower end of the income distribution.

8

The CBO’s income measure is before transfers and taxes. It includes all cash income (both taxable and tax-exempt) and the value of income received in-kind from sources such as employer-paid health insurance premiums. It also includes taxes paid by businesses: corporate income taxes are imputed to households with capital income and the employer’s share of payroll taxes (which are considered to be part of labor income) are imputed to employees. It does not include federal transfers and social security income. The PSID’s before-tax income measure includes taxable income, transfer income, and social security Income. Thus, the PSID’s before-tax income measures would tend be higher than the CBO’s for households in the lower income groups (since only the PSID measure includes transfers) but it could be lower for higher-income groups (given its exclusion of imputed corporate taxes). At the same time, the inclusion of the employer-share of payroll taxes in the CBO measure but not the PSID measure would boost the incomes of the employed households in the CBO dataset.

9

Given their lower response rate, the SCF over-samples top-income households to ensure that they are well represented.

10

PSID households reported their 2006 income in the 2007 wave. For comparability with the SCF, we impute their 2007 income based on the growth rate of nominal disposable income in the NIPAs.

11

Federal Reserve Board staff summarize the main findings of the SCF surveys in working papers; for the 2007 survey, see http://www.federalreserve.gov/pubs/bulletin/2009/pdf/scf09.pdf. We use the findings from this paper to draw a comparison between the PSID and SCF. The paper reports average wealth and income by income quintile. The components of the before-income tax in the SCF are wages, self-employment and business income, taxable and tax-exempt interest, dividends, realized capital gains, other support provided by the government, pension and withdrawals from retirement accounts, social security, alimony, other support payments, and miscellaneous sources of income for all members of the household.

12

The before-tax income data recorded in the PSID are converted into disposable income (that is, after taxes and transfers) using the NBER’s TAXSIM software.

13

We don’t draw inferences on income growth in 2011, since the PSID income data are imputed for that year.

14

Dynan (2012b) compares the growth rates of aggregated house values reported by the PSID respondents with those of the CoreLogic National index (CNI) and finds that the PSID data shows a boom-bust pattern broadly in line with the CNI, with some differences in levels and timing that could be due to households being too optimistic about their house value or reporting values with a lag. Lovenheim (2011) reports that the mean and median house prices in the PSID track the FHFA National House Price Index (HPI) quite closely, with some differences in the recent years that are likely due to the fact that the PSID captures new houses while the HPI excludes them.

15

The mean saving rate is calculated as mean saving divided by mean income, rather than the mean of the household-level saving rates. This method gives a higher weight to higher income households as do aggregate statistics (which report total savings divided by total income). It also reduces the contribution of outliers.

16

Under fully consistent measurement the gap would actually have been larger since the PSID captures only the depreciation of a vehicle as dissaving whereas the NIPAs record the entire purchase as consumption.

17

This finding is similar to the one reported in Mian and Sufi (2009), which shows that mortgage credit grew more strongly in ZIP codes that were experiencing negative income growth over 2002–05, unlike in the 1991–2001 and 2005–07 periods, when the correlation was positive.

18

The Appendix documents the bivariate frequency distribution of households across the income-growth and income level terciles, as well as the groups examined later in the paper. The results show that the households are distributed fairly uniformly across the groups, for instance, income-growth and housing-dependency (which we examine later) are not highly correlated across households.

19

We sort the households according to average income over 1999–2007 rather than by income in any given year, as the latter could be tainted by the impact of temporary income shocks. The finding of low saving rates for the lowest income households in any given survey is likely to reflect, in part, the consumption smoothing of households that faced temporary income losses.

20

Kochhar et. al. (2011) find that the housing bust and the 2008–09 recession led to a much greater reduction (in percentage terms) in the net worth of minorities than of whites. The disproportionate impact reflected in turn the fact that home equity is much more important to the wealth of Hispanic and black households than to white households, and that a disproportionate share of Hispanics live in states that saw the largest boom-bust cycle in house prices.

21

The samples used in these exercises are not exactly the same, as the samples that sort households based on income growth and levels are restricted to households that are present with non-missing income data in each survey between 1999 and 2007, while the sample based on debt-to-income ratios is restricted to households present with non-missing debt and income data in 2007. The sample based on housing-dependency requires households to be present in 1999–2007 with non-missing house values and assets. We used the intersection of the samples for the bivariate frequency tabulations (the saving rates for these more restricted samples are similar to those shown in Figures 1–3).

United States: Selected Issues
Author: International Monetary Fund