The ongoing COVID-19 pandemic represents the largest world-wide macroeconomic shock in at least a century (Baldwin and Weder di Mauro 2020a, 2020b), leading to substantial declines in employment (Bartik et al. 2020, Coibion et al. 2020a), consumption (Baker et al. 2020, Coibion et al. 2020b), and output (Guerrieri et al. 2020, Makridis and Hartley 2020). For instance, retail sales and food service in the US decreased in April by 16.4% from the previous month, and 21.6% from the same month last year, according to a recent release by the Census Bureau.
Social networks are now a primary vehicle for obtaining information in the average household (Westerman et al. 2014). Individuals may adjust their consumption in response to information communicated through friends in connected regions, even if their own county has fairly low exposure to the virus. Quantifying how individuals make consumption and savings decisions in response to shocks not only to their own fundamentals but also to those of their connected friends is important for understanding the sources of aggregate fluctuations, particularly during episodes of uncertainty and panic.
Social networks as amplifiers
There is now clear evidence that the pandemic has shifted household expectations about real economic activity. For example, Coibion et al. (2020b) show that households in counties that experienced lockdown earlier anticipate higher future unemployment, lower future inflation, higher uncertainty, and lower mortgage rates over the next 20 years. Similarly, Binder (2020) shows that greater concern about COVID-19 is associated with higher inflation expectations and more pessimistic unemployment expectations. There is also a rich literature on the role of personal experience in belief formation.
But how are individuals' expectations and consumption decisions influenced by their social networks? If my personal exposure to the pandemic is limited, am I more pessimistic about the economy and more concerned about the infection risk if others within my social network are themselves adversely affected? And, how does that affect my consumption activity? The two are likely linked. For example, Bailey et al. (2018) find that individuals with geographically distant friends who experienced larger housing price increases were more likely to transition from renting to owning, and paying more for a given house.
In Makridis and Wang (2020), we apply these ideas to a new setting in which we exploit plausibly exogenous variation in the exposure of counties to other counties (and countries) that vary in their COVID-19 experience: some areas were affected significantly more than others. Using a combination of Facteus’s card-level transaction data and Facebook’s Social Connectedness Index (SCI), we find that being friends with heavily infected regions reduced consumption; i.e. a 10% increase in SCI-weighted cases and deaths is associated with 0.64% and 0.33% declines in consumption, respectively. To put this into perspective, each county block in the map below plots the SCI-weighted cases of infection on a particular date, and shows substantial variations of the social-media exposure to COVID-19 experienced by different counties.
Figure 1 Number of cases per thousand on Facebook (1 April 2020)
To address concerns about time-varying omitted unobservables or potential selection effects in counties that have more friendship ties with others, we condition on local infections and deaths, as well as county and time fixed effects. Moreover, we also exclude connections to counties in the same state based on the potential for mobility and/or correlation in the response of consumption to common state-level policies. Our results are also robust to state-by-time fixed effects, which directly control for potential state-level policies that influence both consumption and other county infection rates.
Heterogeneity and the composition of consumption
Not all consumption responds equally to peer effects. For example, contact-based consumption goods and services could respond more elastically to changes in the infection rates among connected counties, because increases in the perception of risk associated with the pandemic are likely to prompt individuals to self-isolate. We group each of the 982 merchant classification codes (MCCs) into 17 broad categories based on its degree of exposure to infection risks, as well as its demand elasticity.
Using these new categories for consumption, we estimate separate elasticities in response to SCI-weighted infections by category. Consistent with our theory about peer effects, we find that these declines are greater among social-contact-based consumption categories and activities away from home. For instance, each 10% increase in socially-connected cases is associated with a 2% decrease in clothing/footwear/cosmetics, a 1.3% decrease in contract-based service, and a 1.1% decrease in travel. These are twice to three times as large as the drop in average spending. Moreover, our results are consistent with Coibion et al. (2020), who find that the declines are largest in travel and clothing.
Figure 2 Elasticities by category
We also estimate heterogeneous treatment effects across various dimensions of county characteristics. For example, our elasticities are even greater in lower income counties, counties with higher shares of individuals under the age of 35, less populous countries, counties with more digitally-intensive employees (Gallipoli and Makridis 2020), and more teleworking employees (Dingel and Neiman 2020).
Selection effects or social networks?
We find an economically and statistically significant decline in consumption associated with increases in COVID-19 infections in socially connected counties, even after controlling for time invariant characteristics across space and time, as well as time-varying shocks to local health outcomes (e.g. infections and deaths). However, one concern is that these results are plagued by other time-varying omitted variables that jointly affect connected counties and local consumption outcomes.
One of the ways that we test for selection effects is by taking these insights to an international setting. In particular, we restrict our sample to 15 February – 15 March, before the pandemic became the centre of attention in the US. We focus on county exposure to four countries—South Korea, Italy, Spain, and France—although our results hold for a broader set of countries exposed early on. This allows us to purge variation that may be correlated with time-varying shocks in the US.
We exploit variation along two dimensions. First, counties vary cross-sectionally in their exposure to these countries. For example, whereas Maricopa County in Arizona has an SCI of 142,771 with France, San Francisco has an SCI of 258,825. Second, countries vary in their intensity of COVID-19 shocks. We find that a 10% rise in infections (deaths) in Italy for counties that are more closely connected to Italy is associated with a 0.07% (0.52%) decline in consumption. We see broadly similar treatment effects for each country, although they are smaller for France, perhaps because the US had already witnessed the experience of Asian countries, such as South Korea, Spain, and Italy earlier in March. Moreover, the fact that deaths generate a larger effect on consumption is consistent with the saliency of deaths in the early period of the pandemic, when it was not as large a priority for many US citizens.
Concluding remarks
While the emerging empirical literature on the pandemic has focused on the direct effects of specific policies and/or the spread of the virus, this paper focuses on the role that social networks play in propagating the effects on consumption. Using real-time data on consumption expenditures based on the transactions of 5.18 million debit card users, coupled with data on social connectivity across geographies from Facebook, we quantify the response of consumption to changes in a county's COVID-19 exposure based on its social networks. Our results suggest that these effects from social networks are significantly larger than the direct effects of the virus on consumption.
References
Bailey, M, R Cao, T Kuchler and J Stroebel (2018a), “The economic effects of social networks: Evidence from the housing market”, Journal of Political Economy, 126(6): 2224–2276.
Baldwin, R and B Weder di Mauro (2020), “Economics in the Time of COVID-19”, London: CEPR Press, 6 March.
Baldwin, R and B Weder di Mauro (2020b), Mitigating the COVID Economic Crisis: Act Fast and Do Whatever it Takes, London: CEPR Press, 18 March.
Bartik, A W, M Betrand, F Lin, J Rothstein and M Unrath (2020), “Labor market impacts of COVID-19 on hourly workers in small- and medium- size businesses: Four facts from HomeBase data”, Chicago Booth Rustandy Center, Working Paper.
Baker, S R, R A Farrokhnia, S Meyer, M Pagel and C Yannelis (2020b), “How Does Household Spending Respond to an Epidemic? Consumption During the 2020 COVID-19 Pandemic”, Working Paper.
Binder, C (2020), “Coronavirus Fears and Macroeconomic Expectations”, Review of Economics and Statistics, forthcoming.
Coibion, O, Y Gorodnichenko and M Weber (2020), “Labour markets during the Covid-19 crisis: A preliminary view”, VoxEU.org, 14 April.
Coibion, O, Y Gorodnichenko and M Weber (2020b), “The cost of the COVID-19 crisis: Lockdowns, macroeconomic expectations, and consumer spending”, VoxEU.org, 12 May.
Dingel, J I and B Neiman (2020), “How many jobs can be done at home?”, VoxEU.org, 07 April.
Gallipoli, G and C Makridis (2018), “Structural Transformation and the Rise of Information Technology”, Journal of Monetary Economics, 97: 91–110.
Guerrieri, V, G Lorenzoni, L Straub and I Werning (2020), “Viral recessions: Lack of demand during the coronavirus crisis”, VoxEU.org, 06 May.
Makridis, C A and J Hartley (2020), “The cost of COVID-19: A rough estimate of the 2020 GDP impact. Mercatus Center”, Policy Brief Special Edition.
Makridis, C and T Wang (2020), “Learning from Friends in a Pandemic: Social Networks and the Macroeconomic Response of Consumption”, Available at SSRN 3601500.
Westerman, D, P R Spence and B Van Der Heide (2014), “Social media as information source: Recency of updates and credibility of information”, Journal of Computer-Mediated Communication, 19(2): 171–183.
