Subsidising the spread of COVID-19: Evidence from the UK’S Eat-Out-to-Help-Out Scheme*

Abstract This paper documents that a large-scale government subsidy aimed at encouraging people to eat out in restaurants in the wake of the first 2020 COVID-19 wave in the United Kingdom has had a significant causal impact on new cases, accelerating the subsequent second COVID-19 wave. The scheme subsidised 50% off the cost of food and non-alcoholic drinks for an unlimited number of visits in participating restaurants on Mondays–Wednesdays from 3–31 August 2020. Areas with higher take-up saw both a notable increase in new COVID-19 infection clusters within a week of the scheme starting and a deceleration in infections within two weeks of the program ending. Similarly, areas that exhibited notable rainfall during the prime lunch and dinner hours on the days the scheme was active record lower infection incidence—a pattern that is also measurable in mobility data—and non-detectable on days during which the discount was not available or for rainfall outside the core lunch and dinner hours.


A Take-up evidence from anonymized individual-level transaction data
To corroborate the evidence of take-up of EOHO this paper leverages additional data capturing consumer demand more closely. Specifically, individual level anonymised transaction data from the UK Fintech Fable Data is leveraged in this appendix to document how EOHO increased demand for restaurant visits around the days and during weeks in which the subsidy was available.
This will directly corroborate the evidence from the Google Mobility data analysis and also shed some light suggesting that the results are unlikely to be confounded by other changes in mobility. Further, it corroborates the work of ? suggesting both little persistence and no notable spillover effects on other types of consumer spending -with the exception of reduced grocery shopping. As EOHO was anecdotally motivated by the wish to encourage wider economic activity, such as visiting of other retail outlets, this is suggestive that it may have failed to do so in delivering on this expectation. Data and analysis The anonymized individual level transaction level data is collapsed to an unbalanced individual-level daily panel data set measuring the number of transaction across different types of vendors. Fable has classified the vendors into broad categories with restaurant and hospitality venues being classed as "Food and Beverage", clothing retailers being classified as "Clothing & Apparel" and general online marketplaces being classified as "General Merchandise", while grocery stores and chains would be labelled as such. The temporal granularity along with the granularity of the transaction types enables us to document more sharply the impacts that the scheme had on consumer behavior in a quite demanding empirical design.
Specifically, we estimate the following empirical specification: Here, y i,t measures number of transactions on a date t in a specific spending category. The regression exploits within individual variation by absorbing individual 1 level fixed effects ν i along with controlling for local authority by week fixed effects γ l(i),t . The coefficient of interest is η, which captures the differential changes in y i,t on Monday-Wednesday during which the scheme was available during calendar weeks 32 to 36.
As with the main estimation we can pool the effect across weeks 32 to 36 to present results in tabular format, we can, however, also present results in visual format to provide additional evidence in support of the common trends assumption.

Results
Appendix Table A2 documents that individual card use on days during which the subsidy was available saw a notable increase in transactions in Food and Beverage outlets -with an increase of activity by 8.8% relative to the baseline mean. There are no noticeable other effects on consumer activity with the exception of there being a significant decline in transactions for groceries, suggesting that EOHO-induced restaurant visits were substituting -not surprisingly -for spending on groceries. This suggests that there have been no other general patterns suggesting changes in consumer demand that may have shifted the risk exposure profile.
To complement the mobility analysis I have also incorporated the event study in Appendix Figure A5. This figure highlights that individual card transactions in Food and Beverage outlets strongly increased over the EOHO period and then declined again with the program ending, with no discernible pre-trends, again following the main results in the paper.  Figure A2: Share of COVID19 infections that have been contact traced to restaurants Notes: Figure plots data provided by the weekly COVID19 situation reports from Public Health England. During weeks 32 to 36, PHE identified 739 COVID19 infection incidents that were traced back to a specific origin. The figure plots the share of these incidents attributable to Food outlet/restaurants. The share drastically increases from around 5% to around 20% in week 36 and subsequently declining again after the EOHO program ended. The PHE data is very incomplete. During calendar weeks 32 to calendar week 36 more than 50,000 COVID19 cases were detected highlighting that PHE was able to identify only a small share of the infections. Figure A3: Number of restaurant premises registered to participate in the Eat-Out-To-Help-Out scheme across England Notes: Time series plots the number of restaurants premises that are registered in the scheme at different points in time in England. The bulk of registrations is from HMRC's public github repository. Chain restaurant premises are added separately for completeness. Their inclusion does not affect the results substantially. The program started on Aug 3, 2020 and lasted until Aug 31, 2020. Dots indicate points where a flat file with the restaurants was downloadable from the HMRC Github repository track changes. The data in between is interpolated.              Notes: Table presents regression estimates studying the impact of inclement weather on Google mobility proxies capturing visits to Restaurants and Cafes within local authority districts over time. Column (1) and (3) exploit intra-day variation in rainfall falling during core lunch and dinner hours and outside these hours to study its impact on mobility to restaurants on Mondays to Wednesdays during which the EOHO scheme would have been available during calendar weeks 32 to 36. Column (2) explores the impact of rainfall falling during core lunch and dinner hours on restaurant visits occurring from Thursdays to Sundays -days during which the EOHO discount would not have been available. Panel A focuses on the calendar weeks 32 to 36 when the EOHO was available, while Panel B and Panel C can be thought of as placebo exercises studying the rainfall and mobility relationships during times when the EOHO scheme was not available. All regressions control for district fixed effects and NUTS2 area by date fixed effects. Standard errors are clustered at the district level with starts indicating *** p< 0.01, ** p< 0.05, * p< 0.1.