Household debt and consumption revisited

Philip Bunn and May Rostom

The academic literature finds that the build-up of household debt before the 2008 financial crisis is linked to weaker consumption afterwards. But there is wider debate over the mechanisms at play. One strand of literature emphasises debt overhang acting through the level of leverage. Others find it was over-optimism acting through leverage growth. In this post, we revisit our previous analysis on leverage and consumption in the UK using synthetic cohort analysis. The correlation between leverage measures and their link to other macroeconomic variables mean it’s challenging to tease out their effects. Yet we find that whilst both mechanisms played a role, there is evidence that debt overhang linked to a tighter credit constraints was the bigger driver.

In the UK, the ratio of household debt to income rose from about 85% in 1997 to almost 150% in 2007, with much of that increase accounted for by increases in mortgage debt (Chart 1). For most of this period, household consumption growth was close to its historical average – there was no sharp acceleration in spending – and inflation was low and stable. When the crisis hit, consumption fell sharply. How might the build-up in debt have affected households’ consumption response in the wake of the GFC?

Chart 1: Household debt to income ratio rose sharply prior to the financial crisis while consumption growth was close to average

Academics have put several hypotheses forward to explain this relationship. For this post, we examine two of them: pre-crisis overoptimism, and debt overhang. These two hypotheses are not necessarily mutually exclusive. In fact, both are likely relevant, but not all animals are equal, and some are more important than others.

The over-optimism hypothesis

One view argues that households adjusted their expectations downwards. In this view, prior to the crisis, some households were buoyed by looser credit conditions and rapid growth in house prices increased the amount of collateral homeowners could borrow against. Households who felt positive about the future, could have also been optimistic about their future income, and would have been comfortable to increase leverage quickly. The more optimistic households had to revise down their future income expectations and also cut back their spending by more than others during the GFC, and so this correlates with the change in leverage.

In one paper, Andersen et al (2016) find a strong correlation between the increase in pre-crisis leverage and Danish spending patterns during the recession, but less so for the level. Similarly, in an aggregate-level cross-country comparison, Broadbent (2019) shows that growth in debt between 2005 and 2007 was a better predictor of the economic downturn than the level. However, none of these studies specifically refer to the UK.

The debt overhang hypothesis

A second hypothesis relates to the size of outstanding debt. Households who were highly leveraged going into the crisis faced binding borrowing constraints once credit conditions tightened, limiting their ability to refinance or borrow more. For the UK, this was important: during the GFC many mortgagors were on two-year fixed rates that were refinanced often. This ‘debt overhang’ may have caused these households with higher levels of leverage to cut back spending more. Indeed, a number of studies point to the importance of the level of pre-crisis leverage in explaining the weakness of consumption growth (Dynan (2012) and Baker (2018)), with debtors having higher marginal propensities to consume (Mian et al (2013)).

Credit constraints may also play a role here. Prior to the crisis, mortgage products with loan to value (LTV) ratios greater than 90% were common in the UK, and in some cases offered loans greater than the value of the property – the most infamous example being Northern Rock’s ‘Together’ mortgage at 125% LTV. When the crisis hit, those high LTV products disappeared (Chart 2). This reduction in credit availability will also have been amplified by falls in house prices, which will have raised a household’s outstanding LTV ratio for a given level of debt. UK house prices fell by up to 13% between 2007 and 2009. Taking these two facts together, any household going into the crisis with an outstanding LTV above 75% would have struggled to refinance their mortgage or take on additional debt. LTV ratios are primarily associated with the level of debt. We estimate that around 15% of mortgagors were in this position. These were primarily young households: the average age was 35, and five out of six were younger than 45.

Chart 2: There were very few mortgages with LTVs above 90% during the financial crisis

Revisiting the debate: micro evidence for the UK

We revisit our previous analysis (Bunn and Rostom (2015)) on the comovement between leverage and consumption during the GFC, to consider the role of these different measures and associated hypotheses. For the UK, household-level panel data containing both debt and consumption around the GFC are not available, so we track groups of households, or cohorts, over time using the well-established methodology of Deaton (1985). Nevertheless, our results tell a plausible story.

One challenge with this exercise is that all measures of debt we examine are well correlated – this means it’s hard to definitively conclude which leverage measures are driving this effect. For example, those with high levels of debt to income often also saw strong growth in their leverage (Chart 3).

Chart 3: Measures of the growth and level of leverage are well correlated

Table A reports the results of our reduced-form regressions for the growth in household spending over the financial crisis on different debt measures as explanatory variables (full details in the technical appendix). We also control for other factors such as income growth, wealth and household composition.

Table A: Regressions for household spending during the financial crisis with different debt metrics

Looking at levels of leverage, column 1 shows that groups of households who went into the crisis with high loan to incomes (LTI) made larger cuts in spending during it. In column 2, the level of LTI is replaced with the change in LTI between 2003/04 and 2006/07. Again this is significant, showing that groups of households who experienced earlier rapid growth in debt also made larger reductions in spending. Putting these two debt measures in together in column 3, the coefficient on both falls, but only the change in LTI remains statistically significant. This result is the same if average LTI of a cohort is replaced by the percentage of high LTI households within each cohort and is consistent with the findings of Andersen et al (2016) for Denmark.

But this picture is incomplete because it abstracts from credit constraints. In column 4, we include a measure of the percentage of households in each cohort with a pre-crisis LTV above 75%. This measure, which is based on the level of leverage, aims to capture credit-constrained households. This metric also has a negative and statistically significant relationship with consumption growth during the financial crisis.

However, when we add the change in LTI in column 5, the coefficients on both the change in LTI and percentage of credit constrained households remain significant. Both coefficients are smaller than when they are included on their own. This relationship during the crisis is not seen in earlier periods when credit conditions were looser (see column 6 for one such example). 

As well as the statistical significance of the estimated coefficients, it is important to also consider their economic significance. Chart 4 shows that the magnitudes of spending cuts associated with debt during the GFC implied by all five equations are similar (the black diamonds), at just under 2% of aggregate private consumption. However, they differ on how to apportion it (the coloured bars). In equations 1, 2 and 4 only one channel is included by definition). In equation 5 – the only equation with two statistically significant debt measures – the percentage of credit constrained households accounts for 60% of the total effect, and the increase in debt in the run up to the crisis accounts for the remaining 40%.

Chart 4: Size of spending cuts associated with debt

Conclusion

What can these empirical results tell us about the co-movement between debt and consumption during the financial crisis in the UK? The strength of the correlation between the different measures of leverage make it challenging to conclude the mechanism at play and to definitively prove causation. Nevertheless, they do support an important role for debt overhang, driven by a tightening in credit conditions, and typically captured by the level of leverage. And the role of credit constraints here is supported by the significant relationship between the percentage of households with an LTV ratio above 75% going into the crisis, and cuts in consumption during it.

There can also be more than one explanation, and we do find some weaker support for overoptimism too, although it is curious there was little sign of a large pre-crisis consumption boom in the macro data. In the run up to the GFC, aggregate consumption growth was close to its historical average and nothing like the boom of the late 1980s, implying that any such pre-crisis effects were probably modest.

Technical appendix

The data used on the regressions in Table A are described in more detail in Bunn and Rostom (2015) but the key points are summarised below:

Sample definition: Equations are estimated using cohort data where cohorts are defined by single birth year of the household head and mortgagor/non-mortgagor status. Only households where the head is aged 21–69 are included. The specification reported in equation 1 differs from the equivalent regression reported in Bunn and Rostom (2015) as cohort cells with insufficient observations, after calculating lagged changes in debt, are dropped.

Data sources: Living Costs and Food Survey for all variables except LTV and measures of wealth. For equations 1 to 5, LTV and wealth are from the Wealth and Assets survey, and from the British Household Panel Survey for equation 6.

Additional controls: All equations also include controls for income growth, changes in household composition and growth in housing and financial wealth and a constant. Equation 6 does not include a control for financial wealth due to data availability. 

Variable definitions: ∆lnC is the log change in non-housing consumption, LTI is the ratio of outstanding mortgage LTI ratio, ∆LTI is the change in the ratio of outstanding mortgage LTI ratio, LTV share>75% is the percentage of households in each cohort with an outstanding LTV ratio of more than 75%. For equations 1 to 5 period t is 2009/10 and t-1 is 2006/07. ∆LTI_t-1 represents the change between 2003/04 and 2006/07. For equation 6 period t is 2006/07 and t-1 is 2003/04. ∆LTI_t-1 represents the change between 2000/01 and 2003/04.

---

Philip Bunn works in the Bank’s Structural Economics Division and May Rostom works in the Bank’s Monetary Policy Outlook Division.

If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.