Consumption shifts and inflation measurement during COVID-19
By Anne-Sophie Fraisse (Anne-Sophie.FRAISSE@oecd.org), Annabelle Mourougane (Annabelle.MOUROUGANE@oecd.org), Pierre-Alain Pionnier (Pierre-Alain.PIONNIER@oecd.org) & Beiling Yan, Statistics and Data Directorate (OECD)
The consumer price index tracks the price changes, or inflation, of the goods and services purchased by an average household. It is a key statistic for wage and price setting, for indexing pensions and social welfare payments, for household saving and investment decisions. It is closely monitored by economists and financial markets and in some countries or regions features in the monetary policy target.
National consumer price indices (CPIs), and Harmonised indices of consumer prices (HICPs) in European countries, are constructed as a weighted average of sub-indices covering different products in the consumption basket (e.g. food, clothing, transport), using the expenditure shares of a base year (usually the previous year) as weights. Weights are updated annually in most OECD countries, and, in normal times, are very stable.
The COVID-19 crisis, however, induced large changes in consumption patterns in 2020. This shift poses challenges to inflation measures for 2020 and for 2021, as the pre-COVID expenditure weights became unrepresentative of the average consumption pattern for 2020, while the 2020 COVID weights may differ from the actual weights in 2021. The latter depends on the extent to which these shifts persist as economies recover from the pandemic.
This Statistical Insights piece examines the extent of the changes in the expenditure weights between 2019 and 2020 and the effect of these changes on the measurement of 2020 CPI inflation for individual countries and 2020 HICP inflation for the euro area. Overall, a significant shift in household spending during the pandemic has led to marked revisions in weights. However, when recalculating 2020 inflation using 2020 rather than 2019 expenditure weights, this generates only modestly higher estimated rates of inflation in 2020, of less than 0.1 percentage point on average over the year across a subset of OECD economies.
How have countries adapted their inflation statistics to reflect changes in expenditure patterns?
Expenditure weights measure the consumer expenditure shares in a base year and are updated at the beginning of each year. In January 2021, 28 out of 37 OECD countries updated their 2020 CPI weights for regular CPI compilation. Under normal circumstances, these weights would have been based on 2019 national accounts data or data from an earlier date.
To reflect the large changes in consumption patterns induced by COVID-19, most EU countries adopted Eurostat’s HICP approach (Eurostat, 2020), using additional data sources to estimate the 2020 expenditure weights for the national CPI. These additional data sources include: high-frequency household budget surveys or ad-hoc COVID-oriented surveys; credit card data; scanner data; quarterly national accounts and short-term statistics for wholesale, retail trade and services; and monthly stastistics for transport, tourism and energy.
Non-EU OECD countries, such as Australia (ABS, 2020), have made similar efforts to adjust their expenditure weights for the most heavily affected spending categories by supplementing the usual data sources with more timely data, such as retail trade and scanner data.
In addition, a few national statistical agencies, including Statistics Canada (StatCan, 2021), the UK’s Office for National Statistics (ONS, 2021) and INSEE, the French Statistics Office (INSEE, 2021), have produced and released experimental CPIs with expenditure shares updated in real time since the onset of the pandemic, using high-frequency data sources such as credit card data, transaction data, retail trade surveys or nowcasts of private consumption.
Household consumption shifted markedly during the covid-19 crisis
The 2020 update of the national CPI expenditure weights was larger than in previous years. A “shift” index that captures the extent of changes in the expenditure weights shows that the average size of weight updates between 2019 and 2020 was three times as large as between 2018 and 2019. For example, the 2020 COVID-basket weights for food and non-alcoholic beverages, housing, furnishing, health, communication and miscellaneous goods and services tend to be larger than that in the 2019 pre-COVID-basket because the expenditures on these items were maintained during the pandemic while overall consumption fell. In contrast, the expenditure shares of clothing, transport, recreation and restaurants were smaller in 2020 than in 2019 (Figure 1a).
The most affected spending categories were food, housing, transport and restaurants. For example, compared to the 2019 basket, the average consumer in the 28 OECD countries covered spent 2.8 percentage points less on transport and restaurants, but 2.5 percentage points more on food and housing during the pandemic in 2020. These are sizable and sudden shifts compared to historical slow moving trends (Figure 1b). The spending share on ICT (information and communication technology) goods also rose, but to a lesser degree, by 0.2 percentage point.
Not all the countries were affected to the same extent (Figure 1c). For example, Ireland had the highest overall change in expenditure patterns, more than 13 times as large as that in Belgium. Consumers in Ireland spent 1.6 percentage points less on transport in 2020 compared to 2019, while those in Belgium spent only 0.1 percentage point less. Similarly, the expenditure share on food increased by 3.8 percentage points in 2020 in Ireland, while that in Belgium went up by only 0.3 percentage point (Figure 1d).
These large and abrupt shifts in expenditure shares lead to modest changes in consumer inflation
To quantify the impact of changes in consumption patterns on inflation, official CPI releases for 2020 that use the fixed 2019 weights were compared to a recomputed CPI inflation that uses the 2020 weights. Both are chain-linked indices and everything else was held constant. These computations are not perfect: the former does not account for the substantial change in the consumption basket that occurred during the pandemic, while the latter ignores the consumption pattern in 2019. 1
The revised CPI inflation is found to be on average higher than the official CPI inflation in 18 of 19 OECD countries where data are available (Figure 2a). For the 19 OECD countries as a whole, the gap between the two measures is modest, amounting to an average of 0.05 percentage point in 2020 (Figure 2b). Over time, the gap remained broadly stable from April to December 2020. The main contributors to the difference between the two measures were food and transport. Combined, they explain almost all the differences for 7 of the 19 OECD countries and 40-90% of the differences for 10 of the 19 countries (Figure 2c). This is consistent with findings from the Banque de France (2020).
For the euro area, estimates suggest that the official HICP release was lower than a recomputed fixed-weighted index by 0.12 percentage point on average from April to December 2020.
Admittedly, this counterfactual exercise provides a lower bound of differences in inflation when using fixed or actual weights, as only average annual weights are considered. Actual weights when stringent lockdown measures were implemented, i.e. in Spring 2020, could temporarily be much more different from the 2019 weights than the annual 2020 weights show. Nevertheless, such short-term fluctuations in weights can only have a short-term effect on inflation.
These estimates appear to be generally lower than the experimental estimates based on real-time weights for the United Kingdom (ONS, 2021), France (INSEE, 2021), Canada (StatCan, 2021), and the United States (Cavallo, 2020). Indeed, these studies found that the gap between the experimental estimates and the official CPI inflation was 0.06 percentage point for the United Kingdom, 0.30 for Canada and 0.58 for the United States on average from April to December 2020, and 0.6 for France from April to September 2020.
In the case of the United States, about half of the difference in the gap estimates between Cavallo (2020) and this analysis stems from the very different assumptions on the extent to which expenditure weights were altered by the COVID crisis. In particular, Cavallo’s estimates point to a more pronounced change in the weights of certain components (drop in the transport, recreation and restaurant weights and increase in food and housing weights) than the official Bureau of Labor Statistics (BLS) weights used in this analysis. Differences in the methodology used to compute aggregate CPI inflation (including product classification and the use of monthly weights as opposed to average annual weights) explains the rest.
Looking forward, the pandemic is also expected to affect the 2021 measure of inflation. The example of the United States demonstrates that the official weights to be used in the computation of the 2021 inflation rates, as published by the BLS, can differ markedly from real-time estimates as computed by Cavallo (2020). For instance the fall in the housing weight is more muted in official than in real-time weights. Conversely, real-time estimates point to a more pronounced increase in the weights of recreation and restaurants than official estimates. While it is too early to assess the extent to which changes in consumption patterns are going to be permanent or revert to pre-crisis levels, the on-going reflection on the best way to estimate expenditure weights and measure inflation, including in times of large fluctuations, needs to be pursued.
The measure explained
The CPI weights are updated annually in all OECD countries, except Canada (2 years), Chile (5 years), Colombia (10 years), Czech Republic (2 years), Germany (5 years), Israel (2 years), Japan (5 years), Korea (5 years), Mexico (no defined frequency) and New Zealand (2 years). Among these countries, only Israel updated its CPI weights in January 2021. The US BLS updates expenditure weights every 2 years based on new data from its consumer expenditure survey. For intermediate years, weights are price-updated to reflect changes in relative prices.
European countries compile two price indices: the Harmonised Index of Consumer Prices (HICP), and the CPI. Often, EU countries adopt the same approach in updating the expenditure weights for both indicators. However, the methodology is not always the same, as in the case of Germany where expenditure weights for the HICP are updated annually but every five years for the national CPI.
To reflect the impact of the COVID-19 crisis on expenditure patterns, Eurostat (2020) recommended a mix of additional data sources be employed to compile the best possible estimates for the 2020 expenditure weights for use in the 2021 HICP.
Most EU countries adopted Eurostat’s HICP approach, using additional data sources to estimate 2020 expenditure weights for the CPI. These countries include Finland, France, Greece, Hungary, Ireland, Latvia, Luxembourg, Netherlands, Portugal, Spain, Sweden, Slovenia, and Italy. Turkey and United Kingdom also followed this approach. Austria and Norway used different procedures in updating the national CPI and HICP weights, with CPI weights being less influenced by the changes in consumption patterns caused by the pandemic.
Non-EU OECD countries, such as Australia (ABS, 2020), have also adjusted their expenditure weights for the most heavily affected spending categories by supplementing the usual data sources with more timely data, such as retail trade and scanner data.
Where to find the underlying data?
- Australian Bureau of Statistics (ABS), 2020, “The 2020 annual re-weight of the Australian Consumer Price Index”, Canberra.
- Banque de France, 2020, “L’inflation en France durant les confinements”, Le Bulletin de la Banque de France n°234 : Article 3, Paris.
- Cavallo, Alberto, 2020, “Inflation with Covid Consumption Baskets”, NBER, Working paper 27352.
- Diewert Erwin, 2021, Consumer Price Index Theory, Chapter 2: Basic Index Number Theory, University of British Columbia.
- Eurostat, 2020, “Guidance on the compilation of hicp weights in case of large changes in consumer expenditures”, HICP Methodological note, Luxembourg.
- INSEE, 2021, “Comment calculer un indice des prix à la consommation dans le contexte de la crise de la Covid-19 ? ”, Paris.
- Office for National Statistics (ONS), 2021, “Effect of reweighting the consumer prices basket during the coronavirus (COVID-19) pandemic: October to December 2020”, Newport.
- StatCan, 2021, “Consumer prices: The adjusted price index and basket weights”, Statistics Canada catalogue no. 11-001-X, Ottawa.
- Broadly speaking, this calculation compares a Laspeyre index (with beginning-of-period weights) with a Paasche index (with end-of period weights). Both indices suffer from the ‘substitution item bias’. While in theory, the former is expected to display an upward bias and the latter a downward bias (Diewert, 2021), the magnitude and the sign of the bias remain an empirical question.