How Much Does Household Consumption Impact Business Cycles?

We identify shocks to household consumption using cross-sectoral information. We find that those shocks have accounted for close to 40 percent of pre-pandemic business cycle fluctuations in the U.S. Such shocks have the characteristics of demand shocks: They increase (or decrease) output, inflation and interest rates. The results imply that one might be able to significantly stabilize business cycles by stabilizing consumption fluctuations.

Household consumption is the largest expenditure component of GDP. Accordingly, policymakers and business economists pay a great deal of attention to its ebbs and flows. So when recessions hit, it isn’t surprising that such attention translates into policy in the form of tax rebates or other transfers to stimulate consumption, as was the case in 2001 and 2008/2009.

Most recently, the large cash transfers to families mandated by the American Rescue Plan Act have generated debate on whether they would be too stimulative and “overheat” the economy. Despite the attention of policymakers, until recently, macroeconomic theory has mostly not considered household consumption as an independent driver of economic fluctuations.1

Drivers of Business Cycles

Fluctuations in household demand as an important source of business cycles gained prominence after the 2007-09 recession.2 Newly available evidence suggested that steeply declining housing prices led to destruction of household wealth and reduced consumption.3

Even then, debate remained over the extent to which the household demand channel was the most relevant one, as compared to losses of collateral for entrepreneurs and general curtailment of credit by the banking system.4

In our recently updated paper, “The Consumption Origins of Business Cycles: Lessons from Sectoral Dynamics,” we use information available in the cross-section of industries to provide evidence that macroeconomic disturbances (or “shocks”) that initially impacted household consumption were a key driver of GDP contractions and expansions prior to the pandemic. Apart from a loss of housing wealth that mainly affects consumption, such shocks might include fluctuations in consumer sentiments, consumer credit access or employment uncertainty. In our study, we find that such shocks combine to account for as much as 40 percent of output fluctuations since the mid-1970s.

We also find that those shocks behave like prototypical “demand” shocks, impacting not only aggregate consumption and output, but also inflation and interest rates. At the same time, consumption shocks had little impact on corporate credit spreads and measured total factor productivity, implying that they are distinct from shocks to corporate credit or productivity.

How We Examined Effects of Household Consumption Shocks

To identify how shocks to consumption impact business cycles, we use information available in the cross-section of industries. Intuitively, a negative shock to household consumption should have most of its initial impact on sectors heavily oriented toward consumer goods production (such as the apparel sector), rather than on sectors also geared toward businesses (such as the software sector). Also, being demand shocks, they should lead to greater price changes in those consumption-oriented sectors.5

The strategy we use is designed to avoid a few potential pitfalls.

Sector Sensitivity

The sensitivity of different sectoral prices and quantities to shocks is measured relative to their sensitivity to all shocks. This ensures that our methodology does not capture just the greater cyclical sensitivity of durable or luxury goods but also the increased sensitivity of particular sectors to particular business cycle shocks.

Categorizing Shocks

Our procedure identifies shocks correctly even if there are other shocks that may have similar sectoral impact. For example, a generalized shock to the financial sector would affect household consumption as well as financing to firms. Our methodology allows us to exclude such possibilities through the common assumption that, being exogenous, the time-series behavior of different shocks is uncorrelated.

We further sharpen our results by explicitly identifying other candidate drivers of economic fluctuations using analogous schemes. Thus, for example, shocks to technological progress affects sectors that are more intensive in research and development, shocks to government expenditures affects mostly those that sell most of their output to the government, and so on.

Using Multiple Assumptions and Averaging the Results

Our methodology explicitly considers that our identification assumption is imprecise and incorporates the resulting uncertainty in our estimation procedure. For example, a consumption shock may affect sectors differentially depending on their precise position on production networks.

We accommodate the possibility of model misspecification by identifying the consumption shock several times. In each case, we impose an identification assumption that is a little bit different from our preferred one. Our results are then (weighted) averages of those possibilities, and we describe the uncertainty surrounding those results incorporating those variants. (Said another way, we imposed our identification assumptions through Bayesian priors.) Because we use extensive cross-sectoral data, we can obtain fairly precise estimates.

Correlations with GDP

Figure 1 below validates our identification assumptions. It shows the correlation between various time-series and leads and lags of GDP:

The gray line shows the autocorrelations for GDP. Its value is 1 at 0 lags and declines symmetrically around it.
C shows consumption as measured by BEA aggregate consumption.
HML IP shows the difference between high and low consumption share sectors in the FRB Industrial Production Index. HML π and HML C show the same difference for inflation and consumption growth among Bureau of Economic Analysis personal consumption expenditure categories, respectively.