Over the course of the Great Recession, purchases of new vehicles and other consumer durables fell by $153 billion. Purchases of new homes and other forms of residential investments declined by more than $260 billion. All-told, declines in broadly defined durable spending accounted for more than half of the total decline in GDP during the Recession.
Various stimulus programs, such as the ‘First-Time-Home-Buyers-Credit’ and the ‘Cash-for-Clunkers’ programme were implemented in response to these declines in durable spending. How effective are these sorts of fiscal or monetary stimulus at propping up durable demand during recessions?
The simplest, but not free of flaws, way to answer this question is to regress changes in durable spending on changes in government spending. However, this will give a misleading answer because we know that government stimulus is more likely to occur when durable spending is falling, so a simple regression might lead one to the erroneous conclusion that government spending causes durable spending to fall. Thus, the standard approach for estimating the response of durable spending to fiscal or monetary stimulus is to use vector autoregression models (VARs). These methods look at the relationship between economic variables over long periods of time, and typically exploit differences in timing to isolate causal effects. See Bernanke and Gertler (1995) for an example of the effect of cuts in the FFR on durable spending.1
Spending not responsive to stimulus during recessions
A weakness of standard VAR models is that they assume the relationship between consumer spending and government spending is stable across time. In a recent paper, (Berger and Vavra 2014), we call this assumption into question and argue that aggregate durable spending is much less responsive to economic stimulus during recessions. This is because microeconomic frictions lead households to adjust their durable holdings much less frequently during these times. Our results imply that standard VAR approaches substantially overstate the effectiveness of durable stimulus during recessions. Intuitively, if few households are buying or selling cars or houses during recessions, then the stimulus will have a dampened effect relative to the average.
Our paper begins by documenting the large declines in the frequency of durable adjustment during recessions. Figure 1 below shows the frequency of adjustment of different kinds of durable goods in the Panel Study of Income Dynamics (PSID). The green and red line show how often households change houses (the split in 1997 is driven by a change in sampling methodology), and the blue line shows the frequency of adjustment of a broad measure of durables (only available beginning in 1999). The gray bars represent the fraction of each year in recession. The decline in frequencies associated with these recessions is evident. Various formal regressions confirm the economic and statistical significance of these declines: Adjustment frequencies fall by 15-20% during recessions.
Figure 1. Frequencies of durable adjustment
Note: We split the sample at 1999 as the PSID survey questions and sampling frequency changed. Broad measures of durable consumption are only available in the latter half of the sample.
We also show that other measures of durable turnover fall during recessions. Figure 2 displays the fraction of cars or automobiles that change hands each year. Again, this turnover plunges during recessions for both used and new durables.
Figure 2. How frequently does the durable stock change hands?
Notes: "New turnover each year is #Mew Homes (Light Vehicles)/#Homes (Light Vehicles) at start of year.
"New+Existing" turnover adds used sales to the numerator.
Sources for housing data: HUD and Census.
Sources for auto data: CNW. Recession indicator is the fraction of year spent in recession.
Predicting aggregate spending and its response to shocks
After documenting these patterns, we show that such procyclical frequency is exactly what is predicted by a model with realistic household level frictions. In particular, we build a quantitative model where individual households are subject to idiosyncratic labour income shocks, binding borrowing constraints, and purchase both durable and non-durable consumption goods. In the model, durable consumption goods are subject to a fixed adjustment cost. These fixed costs imply that households pay some strictly positive financial cost even when adjusting their durable holdings by tiny amounts. These fixed costs proxy for, e.g., brokers’ fees on housing transactions, transfer tax and titling fees for automobiles.
Fixed costs of adjustment naturally induce a ‘gap’ between a household’s current durable holdings and those it would choose if it faced no adjustment costs. For small values of the gap, it is not worth paying the adjustment cost, so that the durable stock remains constant. In contrast, when the gap between a household’s current durable holdings and its desired durable holdings is large, a household is more likely to adjust. Intuitively, households are more willing to pay adjustment costs when they are far from their optimum than when they are close to it.
We estimate the extent of these transaction costs in the data, and show that our estimated model is able to explain well the actual durable adjustment in the PSID data. Figure 3 shows the probability of durable adjustment in the estimated model as a function of a household’s durable gap, and compares this adjustment hazard to that in the data. Overall, more than 90% of the variation in empirical adjustment in the data is explained by our quantitative model.
Figure 3. Adjustment probabilities as functions of gaps
After estimating the quantitative model, we explore its macroeconomic implications. By introducing business cycle shocks into the model, we are able to explore how the response of aggregate spending to various policy shocks changes over the business cycle. Figure 4 shows the response of durable spending to various policy shocks in the model.
Figure 4. Response of durable spending to aggregate policy changes
The response of durable spending to these shocks is strongly procyclical. The response of durable spending to a given change in policy is 40-50% smaller if the policy shock occurs in an NBER recession than if exactly the same policy is implemented during a boom. That is, our model implies that the response of durable spending to stimulus is not constant across time and is instead highly dependent on the state of the business cycle. This means that the assumption underlying simple VARs – that the response of durable spending to stimulus is stable across time – is strongly violated.
Why is the response of durable spending to changes in policy so procyclical? During recessions, households’ desired level of durable holdings falls. In general, households can lower their durable holdings by either selling durables, or by waiting around and letting them depreciate. In contrast, if a household wants to increase its durable holdings, it can only do so by actually purchasing durables. In a recession, fewer households want to purchase durables, and more households want to sell durables. However, the asymmetry induced by depreciation means that the frequency of purchases falls by much more than the frequency of sales rises, so that the total frequency of adjustment falls. As fewer households adjust their durable holdings in recessions, the response of durables spending to changes in policy falls. Intuitively, if no household purchases durables, then there is no margin along which to respond to policy.
Figure 5 illustrates how the distribution of gaps and adjustment probabilities varies across time in the PSID data. During the boom, the distribution of durable gaps shifts to the right as households desire larger durables, and adjustment becomes more likely.
Figure 5. Adjustment hazard and distribution of gaps at different dates
Figure 6 shows the implications of these shifts in the data for how durable spending responds to policy shocks. In contrast to Figure 4, these implications come directly from PSID micro-data rather than from the quantitative model. However, the implications are nearly identical:
- During booms (such as 1999 or 2005), the response of durable spending to shocks is substantially larger than during recessions (such as 2001 or 2009).
Figure 6. Response to policy implied by PSID data
While standard analysis of stimulus policy abstracts from non-linearities and microeconomic frictions, infrequent and lumpy durable adjustment is an obvious feature of micro-data. Taking these microeconomic features of the world seriously shows that this abstraction matters for what one concludes about policy. Analysis of the effects of durable stimulus such as ‘Cash-for-Clunkers’ that is based on simple linear VARs is likely to be systematically biased. While this does not imply that such policies will not work, it does imply that traditional evidence for their effectiveness should be viewed with scepticism. In general, policies aimed to prop up durable spending are likely to have a relatively low ‘Bang-for-the-Buck’ during recessions.
Barro and Redlick (2011), “Macroeconomic Effects from Government Purchases and Taxes”, Quarterly Journal of Economics, Vol. 126, Issue 1.
Berger, D and J Vavra (2014), “Consumption Dynamics During Recessions”, NBER Working Paper 20175.
Bernanke and Gertler (1995), “Inside the Black Box: The Credit Channel of Monetary Policy Transmission”, Journal of Economic Perspectives, Vol. 9, Number 4.
Boivin, Kiley and Mishkin (2010), “How Has the Monetary Transmission Mechanism Evolved Over Time?” Handbook of Monetary Economics, Vol. 3.
1 See Barro and Redlick (2011) for similar evidence for government spending or Bovin et al.(2011) for evidence on the response of residential investment.