What do high-frequency expenditure network data reveal about spending and inflation during COVID‑19?

Introduction

The COVID‑19 pandemic and the containment measures that followed have shifted what Canadians buy. For example, since the pandemic began, Canadians have travelled less and bought more cleaning products and non-perishable foods compared with before. The official consumer price index (CPI) uses a fixed basket of goods and services, based on expenditures reported by Canadians in the Survey of Household Spending (SHS). The CPI basket was last updated in January 2019 using data from the 2017 SHS.1 Quantifying the cost of a fixed basket over time allows for consistent measurement of pure price change. Although there is a small measurement bias in the CPI, the fixed-basket approach has worked relatively well under normal economic conditions.2

Concerns have been raised, however, that focusing on a fixed basket of goods and services might be less useful now since consumers have changed their spending patterns because of COVID‑19. Given these changes, CPI inflation may not fully capture the expenditures that a typical Canadian consumer makes currently. There is some evidence that Canadians’ views of inflation may differ from the official measure: in the Canadian Survey of Consumer Expectations, consumer expectations for one-year-ahead inflation rose slightly in the second quarter of 2020 despite a sharp decline in the officially reported inflation rate.3^, 4

To address this potential gap between Canadians’ perceptions and the official inflation rate, the Bank of Canada partnered with Statistics Canada to construct an adjusted CPI to better gauge the typical basket purchased by consumers during the COVID‑19 pandemic (a real-time basket versus the fixed basket). To construct our index, we combine anonymized and aggregated datasets on consumer credit card purchases with data from Statistics Canada’s Monthly Retail Trade Survey and transaction data from Canadian grocery retailers. Then, we use these data to map changes in the shares, or weights, of consumer spending on the goods and services in the CPI basket. Figure 1 outlines our process.

Figure 1: Constructing the adjusted CPI measure

The rest of this note is organized as follows. Section 1 describes the card payments data, known as high-frequency expenditure network (HFEN) data, collected from major Canadian payment card providers. Section 2 outlines the method used to calculate inputs to the adjusted CPI from these data. Section 3 covers the implied changes in spending shares and CPI weights. Section 4 shows the results of the adjusted CPI measure based on these data along with the implications for CPI inflation, and section 5 offers some concluding remarks.

1. High-frequency expenditure network data

National statistical agencies require timely data from surveys and other data sources on the size and composition of consumer spending. Extreme events, which may introduce unpredictable and large shocks, highlight the need for timely data. While Diewert and Fox (2020) advocate for a continual consumer survey to address this need, the International Monetary Fund (2020) points out that it can take up to a year to update weights after the data have been collected. Alternative data sources may thus offer a more feasible approach. Galbraith and Tkacz (2013) propose using debit card transaction data to investigate the effect of extreme events on total consumer spending. To understand the effect of payment disruptions on the real economy, they study what happened in Canada after two extreme events, the September 11, 2001, terrorist attack and the SARS outbreak in 2002. Similarly, HFEN data on debit card purchases may be useful in tracking the effects of COVID‑19 on the level of household spending. In addition, COVID‑19 has affected credit and debit card spending across industries in an uneven way, in terms of both the direction and the magnitude of the change.5 Thus, HFEN data on individual spending categories could potentially be helpful for determining the shares of spending on consumption that underpin the CPI.6 This potential has also been demonstrated by Cavallo (2020), who uses debit and credit card transaction data to estimate the impact of COVID‑19 on the US CPI.

The Bank has access to HFEN datasets from payment service providers. These datasets provide weekly statistics on transaction values processed in Canada, which includes close to three-quarters of the value of payment card purchases in Canada. For 2019, that amounts to around $600 billion. The data are available on an aggregate level and by merchant type.7

2. Mapping the CPI components

The challenges

Four practical considerations arise when we map the HFEN data to the expenditure categories of the various components that make up the CPI. First, the market segments and merchant categories used by the payment service providers do not align one to one with the CPI components. Second, the datasets do not cover all payment methods. Cash, cheques, bank transfers and some other payment methods are not included in the data, yet they play an important role for some expenditure categories, such as rent, utilities and car purchases. Third, while the HFEN data contain transactions from all major card networks, they may not cover all industries and consumer demographics evenly. Analysis of consumer data by Henry, Huynh and Welte (2018) and of retailer data by Kosse et al. (2017) shows that the adoption of card networks differs across consumer groups and industries. Fourth, growth in card payments may not be the same as growth in spending, and some networks have grown at a faster pace than others. While these considerations are partially addressed by our methodology, they also indicate the need for complementary data and research to calculate the weights.8

The proposed solution

Since HFEN data are available by merchant type, we can map HFEN data to CPI components based on the main type of product sold by merchant type. Since most merchants sell a group of products, HFEN data are useful for mapping larger groups of sub-categories such as food purchased from stores but not for mapping granular breakdowns into specific commodities such as bread or cereal.

The CPI basket is made up of eight major categories (see Mitchell 2019). Of these, we exclude shelter because it is not well represented in the HFEN data.9 The remaining categories are food; household operations, furnishings and equipment; clothing and footwear; transportation; health and personal care; recreation, education and reading; and alcoholic beverages, tobacco products and recreational cannabis.

Some of these major categories are broken down further into intermediate categories. For example, food includes both food purchased from stores and food from restaurants. Purchases of motor vehicles are included in transportation; however, since these retail purchases are commonly made with cheques or through financing, mapping the weights for that CPI component is done using alternative sources of data.10

We grouped the categories and their sub-categories according to the type of business that sells those products and services. For example, food from restaurants was grouped with alcohol in restaurants.

This proposed solution addresses the first consideration noted above by setting up a concordance between CPI components and merchant types in the HFEN data. To address the second consideration, we exclude categories where most transactions are likely made with non-card payment methods. These include shelter, and purchase and leasing of passenger vehicles; alternative sources of data for these categories are described below. Further research is planned to understand the impact of COVID‑19 on the potential composition changes in other payment methods, in particular cash, and to understand the role of heterogeneity in payment choice across demographics and across product and industry categories.11

The coverage of HFEN and other sources of data

We calculate final HFEN weights for 64 percent of the CPI basket. The sum of the final HFEN weights for January to May 2020 is equal to a total weight within 1 percent of the sum of the original CPI basket weights (64 percent). For the months of January and February 2020, the HFEN weights of individual components are close to the original basket weights.

To map the remaining 36 percent of the CPI basket and to help overcome the four challenges highlighted above, we use additional sources of data to complement the HFEN data in the calculation of real-time weights. These include Statistics Canada’s monthly retail trade survey and transaction data from Canadian grocery retailers, along with Statistics Canada’s expertise.

Estimating the adjusted CPI weights

We compute the relative changes in the adjusted CPI weights from relative changes in the HFEN data. For example, the adjusted CPI expenditure weight of a component x would increase by a factor of 2 if the share in the HFEN data doubles. We focus on changes in year-over-year values to control for seasonality.

The CPI basket weight of a given component \(x_{t}\) in 2019 is denoted by \(w_{t}\). We denote \(g_{t}\) to be the share of the component x in HFEN data, where the time index is measured in months.

In step 1 of our procedure, the alternative CPI expenditure shares (hereafter HFEN weights) are estimated as \(\hat{w}_{t+12}\) \(=\,\frac{g_{t+12}}{g_{t}}\) \(w_{t}\), anchoring the HFEN weights to the 2019 CPI basket weights. With this estimation, the relative increase of the alternative CPI expenditure share matches the relative increase of the share in the data provided by payment service providers.12 To estimate the overall change in spending, the basket weights are scaled by the observed expenditure changes in the HFEN data. Taking the sum of the scaled weights gives an estimate of the overall change in spending.

For step 2, Statistics Canada requires dollar values as inputs. Combining the total value \(PQ_{t}\) of the mapped basket at time \(t\)13 with the HFEN weights \(\hat{w}_{t+12}\) and the estimate of the overall change in spending, we compute the estimated values \(PQ_{t+12}^{x}\) for each mapped component.14

3. Estimated changes in spending patterns

The HFEN data indicate a significant contraction in spending from March to May 2020 compared with the same period in 2019: for the mapped component, spending overall was down 12 percent in March, 31 percent in April and 14 percent in May. As the economy reopens, the HFEN data can also be used to monitor how spending resumes, overall and by market segment.

Using weekly data from January to May 2020, Chart 1 and Chart 2 illustrate that the proposed methodology suggests the alternative weights (\(\hat{w}_{t}\)) for food, health and personal care goods, gasoline, and clothing were close to the fixed CPI basked weights before COVID‑19. This stability can also be observed for other expenditure categories. However, sharp changes occur for several expenditure categories after the outbreak of the pandemic.15 We notice an increase in the spending shares for food purchased from stores and personal and health care goods, while the spending shares for food from restaurants, clothing and footwear, and gasoline declined.

Chart 1: HFEN weights for some CPI components have increased in weight

Note: CPI is consumer price index; HFEN is high-frequency expenditure network.
Sources: Statistics Canada (Table 18-10-0007-01) and Bank of Canada calculations Last observation: July 26, 2020

Chart 2: HFEN weights for other CPI components have declined in weight

Note: CPI is consumer price index; HFEN is high-frequency expenditure network.
Sources: Statistics Canada (Table 18-10-0007-01) and Bank of Canada calculations Last observation: July 26, 2020

4. Adjusted CPI measure suggests slightly higher inflation

Using the adjusted national weights listed in Table 1, Statistics Canada compiled the adjusted CPI measure based on the chained Laspeyres method.16 This adjusted CPI measure shows slightly less downward pressure than the official CPI in the early months of the pandemic. The year-over-year growth of the adjusted CPI measure is about 1.0 percent in March, 0.0 percent in April and -0.1 percent in May, compared with 0.9 percent, -0.2 percent and -0.4 percent for the official CPI (Chart 3). This is because the weights of components with higher inflation, such as food purchased from stores, have increased, while weights for components with lower inflation, such as transportation, have declined.

Table 1: CPI component weights based on high-frequency data

Percent

Note: The adjusted measure is calculated based on the chained Laspeyres method and uses the one-month lagged weights as inputs. For example, the adjusted measure for May 2020 uses the real-time weights from April 2020. See Mitchell et al. (2020) for Statistics Canada’s version.

Chart 3: The adjusted CPI measure suggests a slightly higher inflation rate

Note: CPI is consumer price index.
Sources: Statistics Canada and Bank of Canada calculations Last observation: May 2020

Because the Bank uses total CPI inflation as the target in its monetary policy framework, these results provide useful insights on the measurement risk around that target.17 However, even when the changes in spending patterns during the pandemic have been accounted for, it is evident that the COVID‑19 shock is disinflationary in the short run. This is because despite consumers spending more on components with a relatively higher inflation rate, the price declines from components with lower demand, such as transportation, more than offset upward price pressures from components such as food purchased from stores.

5. Concluding remarks

Our results suggest that inflation adjusted for changes in spending patterns during the pandemic is only slightly higher than the official CPI inflation so far in 2020. It is evident that the COVID‑19 shock is disinflationary in the short run even when changes in consumption patterns are accounted for. Statistics Canada and the Bank will continue to update this new index to calculate an adjusted CPI inflation measure during the recovery period. The difference between the two measures of inflation may dissipate if the changes in consumption patterns reverse.

Endnotes

References

1. Cavallo, A. 2020. “Inflation with Covid Consumption Baskets.” NBER Working Paper No. 27352.
2. Chen, H., W. Engert, K. Huynh, G. Nicholls, M. Nicholson and J. Zhu. 2020. “Cash and COVID‑19: The Impact of the Pandemic on Demand for and Use of Cash.” Bank of Canada Staff Discussion Paper No. 2020-6.
3. Diewert, W. E. and K. J. Fox. 2020. “Measuring Real Consumption and CPI Bias under Lockdown Conditions.” NBER Working Paper No. 27144.
4. Galbraith, J. and G. Tkacz. 2013. “Analyzing Economic Effects of September 11 and Other Extreme Events Using Debit and Payments System Data.” Canadian Public Policy 39 (1): 119–134.
5. Henry, C., K. Huynh and A. Welte. 2018. “2017 Methods-of-Payment Survey Report.” Bank of Canada Staff Discussion Paper No. 2018-17.
6. International Monetary Fund. 2020. “Consumer Price Index Manual: Concepts and Methods.” (Draft). Inter-Secretariat Working Group on Price Statistics.
7. Jaravel, X. and M. O'Connell. 2020. “Inflation Spike and Falling Product Variety During the Great Lockdown.” CEPR Discussion Paper No. DP14880.
8. Kaplan, G. and S. Schulhofer-Wohl. 2017. “Inflation at the Household Level.” Journal of Monetary Economics 91: 19–38.
9. Kosse, A., H. Chen, M.-H. Felt, V. Dongmo Jiongo, K. Nield and A. Welte. 2017. “The Costs of Point-of-Sale Payments in Canada. Bank of Canada Staff Discussion Paper No. 2017-4.
10. Kryvtsov, O. 2016. “Is There a Quality Bias in the Canadian CPI? Evidence from Microdata.” The Canadian Journal of Economics 49 (4): 1401–1424.
11. Sabourin, P. 2012. “Measurement Bias in the Canadian Consumer Price Index: An Update.” Bank of Canada Review (Summer): 1–11.
12. Statistics Canada. 2019. “The Canadian Consumer Price Index Reference Paper.”
13. Mitchell, T. 2019. “An Analysis of the 2019 Consumer Price Index Basket Update, Based on 2017 Expenditures.”
14. Mitchell, T., G. O’Donnell, R. Taves, Z. Weselake-George and A. Xu. 2020. “Consumer Expenditures During COVID‑19: An Exploratory Analysis of the Effects of Changing Consumption Patterns on Consumer Price Indexes. Statistics Canada.

Acknowledgments

We would like to thank Russell Barnett, Erik Ens, Marc-André Gosselin and Oleksiy Kryvtsov for helpful comments. We would like to acknowledge the collaboration of colleagues at Statistics Canada’s Consumer Price Division. We express our gratitude to Michele Sura and her colleagues in Knowledge and Information Services, Scott Jones, Alison Layng, Olga Mkhitarova and Katherine Shrives, for support in data acquisition and curation. We also thank Carole Hubbard and Meredith Fraser-Ohman for editorial assistance and Vivian Chu, April Dang and Ceciline Steyn for technical assistance.