During the current pandemic, economic variables have moved quickly and by large magnitudes. Given the publication lags for official data this has led to a greater emphasis on higher-frequency and/or more timely measures to track the economic impact of the pandemic and gauge the state of the economy in real time. This post looks at the emerging body of work in this area, with a particular focus on real-time measures of consumer expenditure and activity in the labour market.
Several recent papers have sought to track consumer expenditures by examining the transactions of at least several thousand anonymised users of a bank or fintech company. These granular, timely microdata have been used both to gauge the timing and magnitude of total consumption, as well as to examine which expenditure items and income groups drove the fall in consumption.
Hacioglu Hoke, Känzig and Surico (2020) use data from over 30,000 regular users of a UK bank account collation service to estimate that in April 2020, actual consumer expenditure (excluding imputed rents) fell by around about 40% relative to April 2019. Most of the decline occurred in the weeks immediately preceding the implementation of legally mandated lockdown measures. They also find that high-income groups experienced the largest relative percentage decline in expenditure.
Carvalho et al (2020) utilize a transaction-level dataset from Spanish bank BBVA, which includes 1.4 billion transactions since 2019 conducted using either a BBVA card or sales terminal. Stockpiling is evident in the days preceding the lockdown. Whereas UK expenditure declined in the period leading up to lockdown, Spanish expenditure held steady, but then dropped abruptly once lockdown began. In lockdown, aggregate nominal expenditures were around 50% lower than on the same day in 2019. There was significant heterogeneity across different expenditure categories. Relative to 2019, expenditure on essentials and goods with very low demand elasticity doubled in lockdown. By contrast, expenditure on goods such as food and entertainment away from one’s home almost disappeared completely. This dataset includes purchases using company cards so the figures likely also include some intermediate corporate purchases.
Andersen, Hansen, Johannesen and Sheridan (2020) find similar heterogeneity when examining the Danish economy using data on the card payments of 760,000 anonymised Danske Bank customers. The authors divide the economy into three sectors, ‘open’, ‘constrained’ and ‘closed’. The first category accounts for around half the economy, and saw in increase in expenditure of around 10% in shutdown. The other two sectors, ‘constrained’ and ‘closed’, each accounting for roughly a quarter of the economy, suffered a reduction in expenditure of around 40% and 70% respectively. In aggregate, expenditure fell by around 25%. Similar to the Spanish evidence, aggregate Danish spending remained similar to 2019 until the shutdown, then fell sharply.
Chetty et al (2020) scrutinize transaction data from an aggregator of credit and debit card spending, which covers nearly 10% of all US card spending. Between January and 10 June, high-income households cut spending by 17%, compared to a reduction of 4% by low-income households. As wealthier households constitute a large proportion of expenditure, they account for more than half of the total reduction in card spending since January. The decline in high-income households’ expenditure has severely impacted low wage employees in affluent areas, especially those working in small businesses. In the wealthiest areas, in a two week period, 65% of workers at small businesses were laid off. Incomes are not directly observed, but proxied with the median household income in the cardholders’ ZIP code. Nevertheless, their findings are similar to other studies which use observed income data, such as Farrell et al (2020). Baker et al (2020) find that, between the 26 February and the 11 March, US household spending increased by around 50%, but then plummeted over the next few weeks. This analysis is based on the transactions of around 5,000 users of an online platform that encourages saving. The median user’s payroll income is, unsurprisingly, relatively low.
The patterns from these data point to stockpiling of key goods followed by a sharp fall in expenditure once the virus and lockdown restrictions took hold. As expected, this decline in spending is concentrated in sectors most exposed to the lockdowns. However, transaction-level data on card spending for a subset of goods and services by a selection of the population are by no means a perfect match for the final consumption measure in the national account, because transactions-based datasets may exclude certain categories of expenditure.
Labour market activity
Declines in spending are partly a function of shop closures but also are a reaction to the increasingly perilous employment status of many households. Several pieces of work attempt to track aggregate labour market variables as well as the distribution of job losses across regions and income groups following the lockdown using high-frequency data.
Aaronson et al (2020) use Google Trends data to infer the level of initial claims for unemployment in the US. The authors leverage previous work into the relationship between google searches and unemployment claims after major hurricanes. Authors show that this approach correctly predicted 2.9 million initial unemployment insurance claims released on 21 March, and only slightly under-predicted the data released a week later (predicting 5 million vs official data of 5.8 million). Coibon, Gorodnichenko and Weber (2020) instead examine a large-scale household survey in the US, conducted during 2 to 6 April, and detect significant job losses, but only a 2 percentage point increase in the unemployment rate. Many of those who just lost their job state they are not looking for work, and so, are not counted as unemployed. In official data, based on 12-18 of April, the unemployment rate rose by around 10 percentage points. The estimated employment-to-population ratio was similar to the official figure, suggesting those who were not actively searching for work on survey dates returned to the labour market soon after.
Turning to Europe, Doerr and Gambacorta (2020) consider the proportion of jobs that are in severely affected sectors and the employment share of small firms at the regional level. This is used to gauge the risk to the region’s employment from Covid-19 related disruption. A substantial proportion of regions in Southern Europe and France are at high risk to job losses because of the high share of employment both in the sectors exposed to pandemics and in small firms that are more likely to be financially constrained. Central and Eastern Europe is generally deemed at intermediate risk, while risks are lower in Northern Europe.
Pouliakas and Branka (2020) instead consider unemployment risk by occupation. The European Skills and Jobs survey asks nearly 50,000 workers from across the EU to report the skills required in their role. The jobs requiring little ICT skills, but lots of communication, team working and customer handling skills are assumed to be most at risk from Covid-19 disruption. Overall, 45% of total EU-27 employment is faced with either significant or very high risk.
Tracking the wider economy
A final strand of the literature looks to track GDP. A commonly used indicator for GDP is electricity usage. As noted by Chen et al (2020), electricity usage is pervasive, so it captures the overall pace of economic activity, unlike other indicators such as hotel reservations or flight cancellations which focus on the hardest-hit sectors. Chen et al (2020) show median electricity usage across 31 European countries dropped by 10-15 percentage points in April relative to 2019; historically a 1% drop in annual electricity usage is associated with 1.3% to 1.9% drop in annual output.
The US Weekly Economic Index in Lewis, Mertens and Stoc (2020) combines electricity output with six other indicators, including fuel sales, steel production and same-store retail sales growth to gauge the level of real economic activity in weekly frequency. This index is robust to changes in how it is constructed; subtracting or adding individual series has little effect. The current read points to an 11% year-on-year contraction in GDP in Q2.
Another popular gauge of economic activity is mobility data. Huang et al (2020) use such data from Baidu Maps to monitor the impact of Covid-19 related restrictions on the Chinese economy. The authors propose two new indicators. The first counts the number of new venues added to Baidu Maps each week, while the second measures the volume of visits to all venues over the week. Historically, both indicators have a strong correlation with official GDP data. In 2020, the volume of visits declined sharply in January and February, but has now nearly recovered to last year’s levels. The number of new venues added did not drop as sharply. Over the past few months, it has behaved as it did in 2018.
Real-time substitutes for official statistics have their limitations. Any users of such information must be cognisant of the limitations of these data. Samples are unlikely to be perfectly representative of the population, and may over or under represent activity in certain sectors. Informal sectors, where electronic payments are less commonly used for payment of goods, services and wages are particularly hard to monitor with the approaches outlined above. Further, there will be methodological differences between the calculation of any real-time estimate and the official statistics. Nevertheless, they can be of use both for forecasting and tracking developments in real time, ahead of the release of official statistics.
Joel Mundy works in the Bank’s Research Hub.
If you want to get in touch, please email us at firstname.lastname@example.org 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.