April 2022

The impact of changing consumer expenditure patterns at the onset of the COVID-19 pandemic on measures of consumer inflation

This article examines the impact that changes in consumer spending patterns that occurred during the early stages of the COVID-19 pandemic had on measures of consumer inflation. When spending patterns change quickly, fixed-weight price indexes can become biased. The authors focus on the effects that these changes had on the Consumer Price Index for All Urban Consumers (CPI-U). During the first half of 2020, prepandemic weights were used to calculate the CPI-U. The authors use contemporaneous weights to construct several alternative price indexes and compare them with the CPI-U for the same period in order to quantify the effects of the changing expenditure patterns. They also look at which expenditure categories drive the divergence between the fixed-weight CPI-U and the alternative indexes constructed with contemporaneous weights. Lastly, the authors compare the contemporaneous-weight indexes with indexes constructed with real-time expenditure data.

During the study period, the effect on the monthly inflation rate from using fixed weights averaged 0.08 percentage point, which resulted in the CPI-U understating inflation relative to a similar contemporaneous-weight index. In most months, the absolute value of the deviation resulting from the use of fixed weights was 0.1 percentage point or less. The exception is the month of April 2020, for which the monthly change in the CPI-U was more than 0.2 percentage point below a similar index using contemporaneous weights. This deviation is an upper bound of the actual bias from the changing weights because other factors are driving the divergence between the different indexes. To control for these other factors, we look at the year-over-year change in the effect and find that, outside of April 2020, these other factors largely explain the divergence between the indexes.

Using real-time expenditure data from OI, Cavallo finds a much larger impact from the lack of contemporaneous weights in the CPI-U.⁴ The OI expenditure data are based on credit card transactions data from Affinity Solutions.⁵ Somewhat surprisingly, the CPI-U tracks our alternative indexes calculated with contemporaneous monthly CE weights more closely than the indexes calculated earlier by Cavallo that used real-time OI data, despite the fixed-quantity weights in the CPI-U reflecting prepandemic purchasing behavior. One explanation for this discrepancy is that the OI data, while tracking spending well in some categories, are not comprehensive and the OI expenditure categories do not map well to the expenditure categories used in the CPI. The mapping is complicated because the OI data are organized by industry, whereas the CE and CPI data are grouped by type of product. These limitations in the OI data lead to divergences in the monthly CE weights and the weights used by Cavallo, which helps to explain the different results.

Methods

The CPI-U is calculated in two steps. Initially, basic- or elementary-level indexes are calculated at the item-area level. Expenditure categories (items) are groupings of similar goods and services, and there are over 200 expenditure categories in the CPI-U. The second stage of the calculation uses expenditure data from the CE to aggregate these elementary-level indexes. The weights used in this aggregation represent quantities that are fixed for the level of expenditure in the reference period. The reference period for the aggregation weights covers 2 years of CE data, and these quantity weights are updated every 2 years. During the onset of the COVID-19 pandemic in the first half of 2020, the CPI-U weights were based on CE expenditure data for 2017 and 2018. The 2017 and 2018 biennial CE weights were first used to calculate the December 2019 CPI-U weights, which were then used to calculate the January 2020 index values. In this article, we focus on price changes in the first half of 2020, when the quantity weights used in the CPI were constant.

The use of lagged quantity weights, which are fixed over a 2-year period, is a potential limitation of the CPI-U. The lag in the weights is due to the time it takes to collect and process data on household expenditures from the CE. As long as expenditure patterns are not rapidly changing, the lack of real-time weights may not be a major issue. The bias from using fixed weights could potentially be a larger issue in periods when consumer expenditure patterns are changing rapidly, such as occurred with the onset of the COVID-19 pandemic. As a result of the changing expenditure patterns, the basket of goods and services reflected in the CPI during this period may not accurately reflect what consumers were actually purchasing. Several researchers have cited the lack of real-time expenditure weights as a major weakness of the CPI-U as a measure of inflation during this period.⁶

CE data from January 2020 to June 2020, a period that covers the onset of the pandemic, allow us to form retrospective 1-month index relatives by using contemporaneous weights for the first half of 2020.⁷ (The formulas used to construct the alternative indexes are presented in the appendix.) Using the index relative for period t – 1 to t at the item-area level and expenditure shares, we can calculate the overall index relative for month t – 1 to month t as the weighted average of the item-area relatives. A 1-month Laspeyres index relative is formed as an arithmetic average by using the period t – 1 expenditure shares as weights. Taking a harmonic average and using period t expenditure shares as weights forms a 1-month Paasche relative. Finally, a 1-month relative for a Tornqvist index is formed by taking a geometric average and using the average of the expenditure shares in periods t – 1 and t as weights. The Tornqvist formula is used to calculate the Chained CPI-U (C-CPI-U) and explicitly captures the effects of item-level substitution, which makes the C-CPI-U a better approximation of a true cost-of-living index.

Chained price indexes are formed by linking these 1-month relatives over time. One issue that arises when calculating chained indexes is the problem of chain drift. The frequent updating of weights can lead the index to drift away from unity even if prices and quantities return to the base-period levels. In order to minimize the impact of chain drift, we focus exclusively on the 1-month index relatives, rather than on longer term divergences in the different indexes. Chain drift causes an increase in divergence between the different index-number formulas and typically generates positive bias in the Laspeyres index.

The final C-CPI-U uses contemporaneous CE monthly expenditure weights; however, differences between the CPI-U and C-CPI-U will not only show the effects of using contemporaneous weights but also the effects of using a different index-number formula. Because the CPI-U uses a formula that is similar to the one used to calculate a Laspeyres index, we consider the comparison with the 1-month Laspeyres index relative that uses contemporaneous weights to be the most relevant for identifying the effect of the fixed and lagged weights in the CPI-U. We present alternative index formulas for comparison.

Price index results

We begin our analysis by looking at the changes in the monthly expenditure weights in order to better understand how consumer purchasing behavior changed at the start of the pandemic. Over this period, some item categories had large changes in their expenditure shares, which could contribute to the bias from using fixed weights. Table 1 shows the CPI item categories with the largest change in monthly expenditure shares from January 2020 to May 2020, as well as the change in the item-level relative importance in the CPI-U.⁸ Changes in the CPI-U relative-importance ratios are driven by changes in the relative price levels and reflect the changes in expenditure shares that would occur if quantities were fixed.

Table 1. Change in item-level monthly expenditure shares and CPI relative importance, January–May 2020 (in percent) 

Item	Change in expenditure share	Change in CPI relative importance
Largest share increases
Owners’ equivalent rent of primary residence	3.80	0.32
New vehicles	1.58	0.02
Outdoor equipment and supplies	1.54	0.01
Sports vehicles, including bicycles	0.68	-0.01
Hospital services	0.43	0.03
Largest share decreases
College tuition and fees	-2.94	0.01
Full-service meals and snacks	-2.08	0.02
Airline fares	-0.85	-0.16
Other lodging away from home, including hotels and motels	-0.61	-0.02
Limited-service meals and snacks	-0.58	0.07
Note: The change in expenditure share is calculated as the Consumer Expenditure Surveys (CE) monthly expenditure share for May 2020 minus the CE monthly expenditure share for January 2020. The Consumer Price Index (CPI) relative importance is the fixed-quantity expenditure share reflecting changes in relative prices from the expenditure reference period. Source: U.S. Bureau of Labor Statistics.

All of the categories with the largest decreases in expenditure shares had much larger decreases than we would expect from price changes alone. For example, the large decrease in expenditures for college tuition and fees partly reflects normal seasonal variation, as tuition payments are commonly made at the start of the semester, but it also reflects declines in college enrollment. The decrease in expenditure shares for the other categories is likely due to the impact of COVID-19 on quantities purchased. Similarly, the largest increases in expenditure shares were driven by changes in quantities rather than by changes in price.

Table 2 shows the 1-month percent change in the various price indexes over the period from January 2020 to June 2020. In most months, the 1-month percent change in the CPI-U is within 0.1 percentage point of the monthly Laspeyres index, although the monthly Laspeyres index generally shows higher inflation (or less deflation) during this period than does the CPI-U.

Table 2. Monthly percent change of alternative price indexes, January–June 2020 

Month	CPI-U	Monthly Paasche, current month’s weights	C-CPI-U (monthly Tornqvist)	Monthly Laspeyres, prior month’s weights	Fixed-weight effect (CPI-U minus monthly Laspeyres)
January 2020	0.388	0.309	0.389	0.483	-0.095
February 2020	0.274	0.225	0.263	0.302	-0.028
March 2020	-0.218	-0.225	-0.193	-0.160	-0.058
April 2020	-0.669	-0.603	-0.532	-0.451	-0.218
May 2020	0.002	-0.149	-0.086	-0.027	0.029
June 2020	0.547	0.496	0.576	0.655	-0.108
Note: CPI-U = Consumer Price Index for All Urban Consumers; C-CPI-U = Chained Consumer Price Index for All Urban Consumers. Source: U.S. Bureau of Labor Statistics.

The difference in the 1-month percent change is largest in April 2020. The fixed-weight effect is defined as the difference in the monthly percent change in the CPI-U minus the monthly percent change in the Laspeyres index. In the first half of 2020, the average fixed-weight effect was –0.08 percentage point and the average absolute value of the effect was 0.09 percentage point. In 2019, the average absolute value of the effect was 0.03 percentage point, so pandemic effects do appear to increase the potential bias from the use of fixed-quantity weights in the CPI-U.

A comparison of the various index formulas also provides information about consumer substitution patterns during the January–June 2020 period. Typically, consumers substitute away from items with relative price increases and toward items with relative price declines. The Laspeyres index does not capture the effects of consumer substitution, while the Paasche index tends to overcorrect for substitution bias. Superlative index-number formulas, such as that used in the Tornqvist index, more closely approximate a true cost-of-living index by explicitly accounting for consumer substitution. When preferences are constant, the Paasche index is generally considered the lower bound for a true cost-of-living index, and the Laspeyres index is considered the upper bound.⁹ 

During the early stages of the pandemic, however, demand shifts led to abnormal substitution behavior. For example, households purchased fewer airline fares, even as the relative price of air travel fell substantially. Overall, for the aggregate prices, the monthly change in the Paasche index is less than the monthly change in the Laspeyres index for each month in the first half of 2020. Although some categories displayed abnormal substitution patterns because of pandemic-related demand shocks, household purchasing behavior exhibited, on average, normal substitution patterns during this period. Further evidence in support of normal substitution behavior is the monthly change in the C-CPI-U being less than the monthly change in the Laspeyres index. This indicates that the substitution bias has the typical sign (causing the Laspeyres index to overstate inflation relative to a true cost-of-living index).

Interpreting the fixed-weight effect as a bias is complicated by two issues. First, divergence in the monthly inflation rate of the CPI-U and the monthly Laspeyres index could be due to greater seasonality in the monthly CE weights compared with the weights used in the CPI-U (which are based on expenditures over entire years). Second, even at the monthly level, chain-drift bias could be driving the divergence between the indexes. The Laspeyres formula generally leads to positive chain drift.¹⁰

Table 3 compares the fixed-weight effect for the January–June 2020 period with the effect for the same 6-month period in 2019. In 2019 and 2020, there were similar divergences between the monthly change in the CPI-U and the monthly Laspeyres index in January and June. These divergences likely reflect the effect of normal seasonal variation in the monthly CE weights. Looking at the year-over-year change in the fixed-weight effect removes the effects of seasonality and normal chain drift (in addition to any normal bias from using fixed weights). Therefore, this is our best estimate of the bias associated with the use of fixed-quantity weights in the CPI-U that occurred during the early stages of the COVID-19 pandemic.¹¹ The absolute value of the year-over-year difference is less than 0.1 percentage point in every month except April. The average year-over-year difference outside of April is close to zero (–0.015).

Table 3. Impact of fixed weights by year, January–June 2019 and January–June 2020 (in percent) 

Month	2019	2020	Year-over-year difference
January	-0.123	-0.095	0.028
February	-0.009	-0.028	-0.019
March	0.013	-0.058	-0.070
April	0.011	-0.218	-0.229
May	0.008	0.029	0.021
June	-0.074	-0.108	-0.034
Note: The impact of fixed weights is calculated as the difference between the 1-month change in the Consumer Price Index for All Urban Consumers (CPI-U) and that of the monthly Laspeyres index. Source: U.S. Bureau of Labor Statistics.

Table 4 shows the item categories that are driving the differences (both positive and negative) between the CPI-U and the monthly Laspeyres indexes in April 2020, the month in which the divergence in the indexes was greatest. The item-level contribution to the difference in the 1-month price change is calculated as the difference at the item-area level between the CPI-U relative importance and the monthly CE weight, which is then multiplied by the monthly percent change in the item-area price index and aggregated up to the national level for each item category. The item contribution to the difference in the indexes depends on the difference in the weights and the change in the price. If the weights are the same in both indexes, then the item contributes equally to the 1-month change in both indexes and thus the contribution to the difference would be zero.

Table 4. Decomposing the April 2020 1-month change in the CPI-U versus the monthly Laspeyres index 

Item category	Contribution to the difference (CPI-U minus Laspeyres index)	March CPI-U relative importance	March CE monthly expenditure share
All items	-0.218	100.0	100.0
Top contributors driving the Laspeyres index to be greater than the CPI-U
Gasoline (all types)	-0.071	2.957	2.505
Airline fares	-0.032	0.719	0.489
Women’s suits and separates	-0.015	0.520	0.211
Other lodging away from home, including hotels and motels	-0.015	0.865	0.772
Men’s suits, sport coats, and outerwear	-0.011	0.112	0.033
Top contributors driving the the Laspeyres index to be less than the CPI-U
Motor vehicle insurance	0.093	1.713	2.914
Telephone hardware, calculators, and other consumer information items	0.005	0.090	0.537
Fuel oil	0.004	0.084	0.118
Limited-service meals and snacks	0.003	2.682	2.313
Utility (piped) gas service	0.003	0.670	0.893
Note: The relative-importance values and expenditure shares do not sum to 100 because the table presents only selected categories. The CPI-U relative importance is the fixed-quantity expenditure share reflecting changes in relative prices from the expenditure reference period. CPI-U = Consumer Price Index for All Urban Consumers; CE = Consumer Expenditure Surveys. Source: U.S. Bureau of Labor Statistics.

The difference in the monthly inflation rates as measured by the CPI-U and the monthly Laspeyres index is 0.218 percentage point in April 2020, with the CPI-U showing a larger price decline. The items that contribute the most to this difference are gasoline and airline fares, which combined explain about half of the gap. These are categories with large price declines for which the CPI-U relative-importance ratios overstate the monthly expenditure shares (as the pandemic led households to purchase less gasoline and fewer airline fares). Table 4 also lists the item categories that drive the CPI-U to be greater than the monthly Laspeyres index. The largest contributor in this category is motor vehicle insurance, which experienced large price declines. However, unlike airline fares and gasoline, motor vehicle insurance has a CPI-U relative importance that understates the monthly expenditure shares.¹²

Comparison with indexes that use real-time alternative expenditure data

Cavallo uses real-time expenditure data from OI in an attempt to produce price indexes that better capture consumer expenditure behavior during the COVID-19 pandemic. The methodology starts at the level of the major-product categories and their corresponding CPIs. These major-category indexes are aggregated into an all-items index by using weights calculated with OI data. The OI data capture the change in expenditures relative to the first 4 weeks of January 2020, compared with the change in expenditures relative to the first 4 weeks of January 2019. The OI expenditure categories are mapped to CE categories, and updated CPI-U weights are calculated by using the December 2019 CPI relative-importance ratios to reflect the change in expenditures from the OI data.

OI obtains consumer expenditure data from Affinity Solutions, which tracks consumer credit and debit card spending. The data from Affinity Solutions have some features that may complicate their comparison with CE data. First, the Affinity Solutions data include the merchant code of the seller as opposed to the type of good or service purchased. For example, if a consumer buys groceries at a certain retail store, the name of the store will be recorded, but the type of product will not. Expenditure data in the CE and CPI are grouped by the type of product sold. Second, small businesses and nonprofits may use credit and debit cards for routine spending, which would be included in the Affinity Solutions data, whereas the CE data only include household expenditures. Finally, the Affinity Solutions data are not comprehensive because they only cover card transactions (cash transactions are not collected).

Affinity Solutions provides OI with data on daily aggregate transactions by county and industry category.¹³ OI divides the transactions among 27 different industries, but not all of them are reported on the OI website. The coverage of the data changes as card providers enter or exit the Affinity Solutions dataset, so OI attempts to identify and correct for any structural breaks at the county level.¹⁴

In addition to the limitations imposed by using a different source of expenditure data, the Cavallo index differs from our indexes for other reasons. First, Cavallo starts at the modified major-group level, which corresponds to the major-product groups in the CPI, except for food, for which he uses the subcategories food at home, food away from home, and alcohol.¹⁵ The major-group CPIs use fixed weights to aggregate the various item indexes within the product category, so starting the analysis at this level will miss the effects of changing expenditures within product categories. Second, Cavallo uses a different index-number formula to aggregate to an all-items index. In order to isolate the effects of the different sources of weighting data, we construct a version of the monthly Laspeyres index that uses the Cavallo weights to aggregate from the modified major-group level to the all-items level. This index uses the monthly CE data to aggregate from the item-area level to the modified major-group level. The only difference between the Laspeyres index with Cavallo weights and the monthly Laspeyres index is the weights used at the major-product-category level.

Table 5 shows the monthly percent changes in the various indexes. The Laspeyres index that uses the Cavallo weights to aggregate the modified major groups tracks the changes in the monthly Laspeyres alternative index closely in January, February, and March. It shows much less of a price decline in April compared with the monthly Laspeyres alternative index, but it shows a larger decline than that in the original Cavallo index. The Laspeyres index with the Cavallo weights tracks the original Cavallo index closely in May and June.

Table 5. Monthly percent change in the CPI-U and monthly Laspeyres index compared with Cavallo indexes, January–June 2020 

Month	CPI-U	Monthly Laspeyres index, prior month’s weights	Monthly Laspeyres index, Cavallo weights	Cavallo index
January 2020	0.388	0.483	0.464	0.39
February 2020	0.274	0.302	0.317	0.28
March 2020	-0.218	-0.160	-0.163	-0.12
April 2020	-0.669	-0.451	-0.233	-0.09
May 2020	0.002	-0.027	0.114	0.11
June 2020	0.547	0.655	0.487	0.47
Note: For more information on the Cavallo index and weights, see Alberto Cavallo, “Inflation with Covid consumption baskets,” NBER Working Paper 27352 (Cambridge, MA: National Bureau of Economic Research, June 2020; revised July 2020), https://www.nber.org/system/files/working_papers/w27352/w27352.pdf. CPI-U = Consumer Price Index for All Urban Consumers. Source: U.S. Bureau of Labor Statistics.

The differences between the monthly Laspeyres alternative index and the Laspeyres index with Cavallo weights are due to differences in the expenditure weights at the modified major-group level. Table 6 compares the category weights for the monthly CE data, the CPI-U, and Cavallo in May 2020. Although June CE data are available, we focus on changes through May because economic restrictions began to be relaxed by June. The weights in Cavallo are much higher than the monthly CE or the CPI-U for housing and for education and communication, and much lower for transportation, recreation, and medical care. For food at home, the CPI-U has a lower weight than the contemporaneous measures, while for food away from home, the CPI-U has a much higher weight. Surprisingly, the CPI-U relative-importance ratios generally track the monthly CE expenditure shares more closely than the Cavallo weights, despite reflecting prepandemic consumer behavior.

Table 6. Expenditure shares at the modified major-group level, May 2020 

Modified major group	Monthly CE expenditure share	CPI-U relative importance	Cavallo weights
Apparel	1.9	2.7	2.6
Education and communication	5.7	6.8	8.4
Food at home	9.4	8.0	10.8
Food away from home	3.1	6.3	3.9
Alcohol	0.8	1.0	1.5
Other goods and services	2.3	3.2	3.3
Housing	46.9	42.6	52.2
Medical care	9.2	9.0	7.4
Recreation	5.1	5.9	2.7
Transportation	15.8	14.5	7.3
Note: CE = Consumer Expenditure Surveys; CPI-U = Consumer Price Index for All Urban Consumers. The CPI-U relative importance is the fixed-quantity expenditure share reflecting changes in relative prices from the expenditure reference period. For more information on the Cavallo weights, see Alberto Cavallo, “Inflation with Covid consumption baskets,” NBER Working Paper 27352 (Cambridge, MA: National Bureau of Economic Research, June 2020; revised July 2020), https://www.nber.org/system/files/working_papers/w27352/w27352.pdf. Source: U.S. Bureau of Labor Statistics.

One complication is that the weights used in Cavallo are seasonally adjusted, while the monthly CE weights are not. Another complication is that Cavallo uses the change in expenditures by category to update the December 2019 CPI-U relative-importance ratios, which are based on expenditures in 2017 and 2018. Hence, the differences between the Cavallo weights and the monthly CE weights could also be the result of changes in expenditure patterns prior to the pandemic.

We next look at changes in expenditures for the different sources from December 2019 to May 2020. Looking at these changes allows us to avoid the issue of the differences in the 2020 monthly weights possibly reflecting prepandemic changes in expenditure patterns. However, the comparison is still complicated by the OI data being seasonally adjusted. In order to make the data more comparable, we seasonally adjusted the CE data by using methods similar to those used in the OI data. The procedure used for the OI data involves dividing the level of spending in a given period in 2020 by the prior year’s value. Then, this annual spending ratio is indexed to the first 4 weeks of January 2020. We follow a similar procedure to seasonally adjust the monthly CE expenditure weights. The monthly expenditures in 2020 are divided by the 2019 values and indexed to the January 2020 year-over-year value.

Table 7 shows the changes in expenditures from December 2019 to May 2020, by modified major-product group. The OI data used by Cavallo do not cover all product categories, so his estimates hold education and housing expenditures unchanged during this period.¹⁶ Housing expenditures in the CE data are relatively flat during this period, but education and communication expenditures fell by nearly 21 percent, seasonally adjusted. The OI data show much larger declines in medical care, recreation, and transportation than the CE data show. In some of these categories, the discrepancy can be explained by OI data not having complete coverage within the category. For example, the OI data do not include vehicle purchases within the transportation category. Similarly, the OI data do not include payroll-deducted health insurance premiums in the medical care category. Because vehicle purchases and spending on health insurance premiums held up better than other items in these product categories, Cavallo’s estimates overstate the decline in expenditures in medical care and transportation. Transportation less new and used vehicles declined by about 38 percent in the CE, so the lack of vehicles in the OI data only partially explains the discrepancy. Interestingly, the monthly CE food-at-home category is unchanged over this period, which may be due to stockpiling earlier in the pandemic.¹⁷

Table 7. Percent change in expenditures, by major-product group, December 2019–May 2020 

Modified major group	Monthly CE (seasonally adjusted)	Cavallo expenditures
Apparel	-57.6	-25.4
Education and communication	-20.5	0.0
Food at home	0.0	14.5
Food away from home	-56.1	-49.6
Alcohol	-12.1	14.5
Other goods and services	-37.5	-15.3
Housing	-0.9	0.0
Medical care	-7.0	-32.4
Recreation	-12.7	-62.9
Transportation	-33.2	-62.5
Note: The monthly Consumer Expenditure Surveys (CE) aggregate expenditures are seasonally adjusted by dividing 2020 monthly expenditures at the major-product-group level by the prior year’s values, which are then indexed to the January 2020 year-over-year values. For more information on the Cavallo expenditures, see Alberto Cavallo, “Inflation with Covid consumption baskets,” NBER Working Paper 27352 (Cambridge, MA: National Bureau of Economic Research, June 2020; revised July 2020), https://www.nber.org/system/files/working_papers/w27352/w27352.pdf. Source: U.S. Bureau of Labor Statistics.

Some of the remaining discrepancies can be explained by imperfect category matching between the CE and CPI product categories and those used in the OI data. Some of these discrepancies arise because OI categorizes expenditures on the basis of merchant type rather than product type, which can be problematic when retailers sell multiple types of products.¹⁸ Cavallo’s estimates use OI grocery expenditures for the food-at-home and alcohol categories, and they use OI expenditure data for restaurants and hotels for the food-away-from-home CPI category. In the CE and the CPI, the alcohol category includes alcohol consumed away from home (at bars and restaurants), but in the OI data, alcohol is in the restaurant category and thus is part of the larger category food away from home. The decline in alcohol expenditures in the CE reflects a decline in alcohol consumed away from home. Cavallo’s estimates show an increase in alcohol expenditures because they are part of grocery expenditures.

Imperfect matching between industry and product classifications can explain the discrepancies in nonvehicle transportation and recreation categories. Many of the remaining nonvehicle items in the CPI transportation category are sold by firms outside of the transportation industry. Some examples include vehicle leasing (financial services industry), motor vehicle insurance (financial services), motor vehicle parts (retail), and motor vehicle maintenance and repair (retail). The largest component of the CPI recreation category is cable TV, which is purchased from firms in the telecommunications industry. The pandemic had minimal effects on cable TV expenditures. As in the transportation industry, expenditures at firms in the recreation industry will generally capture services. Expenditures on recreation commodities (for instance TVs, audio equipment, or sports equipment) will usually be from firms in the retail industry.

Grocery stores are a good example of merchants in a given industry that sell multiple types of products, which complicates the mapping between industry-based and product-based classification systems. Nonfood categories purchased at grocery stores are not included in the food categories of the CPI. For example, nonprescription drugs are part of medical care in the CPI. Cleaning supplies and paper products are part of housing, and personal beauty products are part of other goods and services. Finally, hotels are part of the OI expenditure category food away from home, but they are part of the housing category in the CE and CPI.

Conclusion

BLS recognizes that the use of lagged and fixed weights is a potential source of bias in the CPI-U, which could be particularly problematic for periods in which expenditure patterns are changing rapidly. In the first half of 2020, which covers the onset of the COVID-19 pandemic, the effect of using fixed weights in calculating the monthly consumer inflation rate averaged 0.08 percentage point. This means that if contemporaneous weights had been used, monthly inflation would have been, on average, 0.08 percentage point higher than that reported by the official CPI-U. The actual bias from using fixed weights is probably lower because some of the divergence is due to greater seasonal variation in the contemporaneous monthly weight data as well as to chain drift in the monthly Laspeyres index. Controlling for these other factors by looking at the year-over-year change in the effect shows that, outside of April, there is minimal bias from the pandemic-induced changes in expenditure weights. The divergence between the monthly change in the CPI-U and the monthly change in a similar index using contemporaneous weights was largest in April 2020, with airline fares and gasoline being the categories that contributed most to the divergence.

Given the inherent delay in survey-based data collection and estimation, many people have turned to alternative sources of real-time data for information on the economy during the pandemic. Cavallo provides valuable economic data by applying real-time expenditure data from Opportunity Insights (OI) to produce a contemporaneous weight CPI, a practice that continues. However, Cavallo’s results imply a much larger fixed-weight bias than the one we calculated by using the monthly CE data, although our results look at a short period during the early part of the pandemic. The discrepancy between our results and Cavallo’s can largely be explained by incomplete coverage of expenditures in the OI data and imperfect mapping between the OI expenditure categories and those in the CE and CPI. Cavallo’s work demonstrates that applying alternative sources of expenditure data to improve the timeliness of the weights in the CPI-U is a promising area of future research for BLS, which is already taking steps to improve the timeliness of the CE data.

Appendix: Monthly index number formulas using contemporaneous weights

Item-area index relative for item and area (i, a) from month t – 1 to month t:

,

where IX is the item-area price index.

Expenditure share in period t:

.

Price index formulas:

Monthly Laspeyres formula:

.

Monthly Paasche formula:

Monthly Tornqvist (C-CPI-U) formula:

.

 

top Back to Top