POVERTY AND EQUITY NOTES SEPTEMBER 2023 The value of data: an estimate of the cost of (not) updating Brazil’s consumer price index Authors: Ricardo Vale, Otavio Conceição and Gabriel Lara Ibarra. • Household budget surveys are the basis for adjusting the weights of consumer price indices. In the Brazilian case, both the weighting structure of INPC and IPCA are adjusted by the results found in POF. Given the influence of INPC and IPCA on government budgets, monetary policy and even the readjustment of private-sector contracts, timely survey data is of paramount importance to the design of effective public policy. • In Brazil, the low periodicity with which household consumption data is incorporated into the price index can generate meaningful differences in the magnitude of measured inflation. It is estimated that a one-year delay in the update of the weighting structure of the index in 2020 would have led to a difference of 0.8 percentage points in annual measured inflation. The difference in annual inflation would have reached 1.1 percentage points in 2022. • INPC is the reference index for the adjustment of the minimum wage and social security benefits, among other government policies and instruments. A simple extrapolation of the counterfactual results suggests that spending with pension benefits in 2021 would have been BRL 5.5 billion (USD 1 billion) higher had the INPC weighting structure not been updated in 2020. • Even though this result is context-specific to Brazil and refers to a particular period of time, it demonstrates that POF is an important investment that contributes to the production of several official statistics of the country, including in the preparation of the public budget and in the formulation of the monetary policy. Any cost-benefit analysis of this key element of the national statistical ecosystem should thus consider the consequences of disruptions in its periodicity. Consumer price indices are central to monitoring, guiding, and defining a country's economic development path. By capturing how prices change over time, consumer price indices provide a measure of the evolution of the cost of living for households. The impact of price indices on the economy is very broad, affecting everything from the adjustment of pensions to the monetary policy of the Central Bank, from cash transfer programs to private sector contracts. The measurement of inflation has, therefore, real consequences for the country's evolution. An important fact about price indices is that they are calculated based on the population's average consumption basket – information that comes from national household budget surveys (HBS). Such surveys collect detailed information on the consumption pattern of households at a given moment in time. In particular, they allow to gauge the relative importance of different consumption items through their share in total household expenditures. Indicators on consumption items can then be aggregated (at the local or national level) into a consumer price index (CPI). The relationship between CPIs and HBS demonstrates the importance of the availability of high-quality and up-to-date information on households’ consumption patterns. In fact, CPIs have their weighting structures adjusted with some frequency to reflect the reality of the expenditure patterns of their target populations. POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 This is crucial, as delays in capturing changes in consumption patterns, such as the replacement of landlines, cameras for photography, and VCRs by smartphones and streaming services, can lead to very different cost- of-living estimates. Moreover, families adjust their consumption in response to their income changes, themselves correlated with the economic growth and its cycles, thus highlighting another important reason for closely monitoring the evolution of costs of living. Not without reason, UN (2009, paragraph 4.45) states that, ideally, a HBS should be conducted every year, while IMF et al. (2020, paragraph 3.43) recommends a five-year interval. In this note, we study the Brazilian case and focus on the potential fiscal implications of the unavailability of a household budget survey in a timely manner. The note presents two hypothetical exercises that vary the timing at which the national statistical office incorporates updated information from a household budget survey into the CPI. Varying the timing of adoption of the expenditure information allows to create a counterfactual price index that can be compared to the true CPI at different points in time. Finally, using the actual and counterfactual CPI we answer the following question: what would have been the government expenditures should the CPI update have been delayed? The note focuses on expenditures on pensions (aposentadorias) due to data availability. Recognizing that there are many other government policies that depend on inflation estimates, the estimates presented can be interpreted as a lower bound of the effect of interest. Consumer price indices are based on a representative consumption basket of their target population. In the case of Brazil, the basket composition of the National Consumer Price Index (Índice Nacional de Preços ao Consumidor, INPC) is defined and updated by the average consumption of families with income between 1 and 5 minimum wages as measured in the household budget survey (Pesquisa de Orçamento Familiar, POF), which is carried out by the national statistical office (Instituto Brasileiro de Geografia e Estatística, IBGE). POF has been conducted with an average periodicity of 6.25 years in the last three decades (i.e., since the POF 1987/1988). The update of the reference consumption basket for INPC takes place only after POF is collected and processed. The sequence of the last two updates of the POF information in the INPC calculation was as follows: (i) the POF conducted between July 2008 and July 2009 was incorporated in January 2012; and (ii) the POF conducted between July 2017 and July 2018 was incorporated in January 2020. In Brazil, changes in the measurement of prices variation generate major economic impacts. Price indices guide increases given to the minimum wage and other wage negotiations, pensions, income transfers, readjustment of prices in service and supply contracts, debt updating, settlement of interest rates and premiums on financial assets. In particular, INPC is the official CPI used as a reference for the adjustment of the minimum wage and as a legal minimum for the adjustment of pensions paid by the federal government (Art. 41-A, Ordinary Law No. 8.213, 1991). The Broad Consumer Price Index (Índice Nacional de Preços ao Consumidor Amplo, IPCA), in turn, is the reference index for the Brazilian inflation targeting regime. Thus, the entire Brazilian monetary policy is designed around the objective of bringing IPCA to the center of the target. Therefore, divergences in the estimation of the typical Brazilian consumer basket have the potential of affecting the basic interest rate of the local economy and, consequently, the level of economic activity and employment. Not conducting POF frequently has the potential of distorting the measurement of consumers' cost of living. Nine years passed between the conclusion of the most recent POF in 2017/2018 and the previous one, conducted in 2008/2009. During this period, the consumption habits of Brazilians have changed. While people used to watch movies on DVD and communication was still largely done through traditional phone calls or SMS in 2008, streaming services (e.g., Netflix) and messaging software (e.g., Whatsapp) were already widely adopted in 2017. Among the changes, we highlight the case of the category “Food and Beverages”, which Page 2 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 accounted for 24.5 percent of the INPC reference consumption basket in December 2019 and whose weight fell to 19.3 percent by the occasion of the weights update in January 2020, i.e., a difference of 5.2 percentage points. In contrast, the relative weight of the category “Communication” rose from 3.4 percent to 5.7 percent over the same period. This note estimates the impacts that not updating the basket with the usual periodicity may have on INPC. Our interest is to understand the consequences of POF being carried out with different periodicity. In particular, we present the results relative to two different counterfactual exercises. The first estimates the impacts of a hypothetical scenario that assumes that the update of the INPC reference basket did not occur in January 2020 or in any subsequent period. Notably, 2020, 2021, and 2022 were atypical years in terms of economics conditions affecting consumption prices due to the pandemic, the Russian invasion in Ukraine, and the presidential elections in Brazil. To test the exercise under different circumstances, we conduct a second hypothetical scenario that assumes that the update of the basket did not occur in January 2012 or in any subsequent period. A priori it is not possible to predict what the impacts of a change in the frequency of incorporation of POF information are. There are several factors that affect the consumption of goods and services, and the impact of these factors can change over time. If the price of a good varies, one must consider both the substitution effect and the income effect on the price elasticity of demand. Not only the price elasticity of the good itself affects its demand, but also the cross elasticity between goods. Furthermore, there is not only price dynamics acting on the demand for a good, but also other factors. If households are getting richer or poorer on average, the income elasticity of demand is another factor to be considered. One must also consider the fact that technological and economic dynamics determine household demand and that cultural preferences and habits are not static. Moreover, families can consume thousands of different goods and services, which makes it challenging, at least with the current knowledge, to accurately update the weights of items consumed without a field survey that could produce updated data. It is important to emphasize that the results presented in this note are context-specific. As predicted by the economic theory, there are multiple factors affecting the demand for goods and services, and therefore there is no pattern of differences in the index that can be established ex-ante with only the knowledge of an earlier budget survey. If inflation is being overestimated, then contracts and prices are being over-adjusted and resources are being wasted. Meanwhile, if inflation is being underestimated, then wage earners who have their salaries adjusted by the inflation measure would observe their purchasing power reducing by a rate equivalent to the underestimation. The main point is that having timely data is necessary to minimize the risk of miscalculating inflation. To make the calculations of this study, we use the monthly INPC data published by IBGE. Specifically, we use the price variation and weight of each category that makes up the INPC consumption basket. The source of data is the IBGE online data warehouse (Sistema IBGE de Recuperação Automática, SIDRA), which consolidates the INPC data at the subitem level. 1 Importantly, the SIDRA data contains the categories’ identifying codes, 1 There are currently two data sources for INPC at the subitem level: (i) the "raw" data from FTP IBGE, which is published as spreadsheets in the .xls format, one file for each year-month in the period between January 2008 and May 2023; and (ii) the consolidated data from SIDRA IBGE in Tables 2938 (for the period between July 2006 and December 2011), 1419 (for the period between January 2012 and December 2019) and 7060 (for the period between Page 3 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 which allows for secure identification of the categories that form the different aggregation levels of INPC over time. The analysis is conducted with the most granular level of INPC, the subitem level, posing specific challenges to the computation of counterfactuals. INPC has four levels of aggregation, in the following order, from the most aggregated to the least aggregated one: (i) group, (ii) subgroup, (iii) item, and (iv) subitem. Given the way that the INPC variation is calculated and published, we are able to compute both the factual and counterfactual INPC at any aggregation level. The main difficulty associated with conducting counterfactual exercises at more disaggregated levels is that new categories typically are incorporated into the new basket while others are removed over time, which implies that the number of elements in compared baskets may not be necessarily the same. Moreover, we do not observe the price variation of goods that were removed from the basket and, therefore, we do not have all the necessary inputs for the calculation of the counterfactual CPI firsthand. Importantly, we show in Appendix A that if counterfactual calculations were performed with more aggregated levels of INPC, we would lose within-category variation and find possibly misleading results. The imputation of a price variation for unobserved goods in the counterfactual scenarios To overcome the missingness of price variation data in the counterfactual building, we implemented a targeted-mean imputation that preserves the weights of the aggregate consumption categories. A straightforward approach to deal with the missingness of price variation of some subitems would be to drop them and reweight the basket based on the remaining ones, or, equivalently, impute them the average price variation of the corresponding item-level category2. In that case, however, we could produce a misleading counterfactual with respect to the original basket that we are trying to replicate. In special, if most of the subitems that were removed in the new weighting structure belonged to a particular item-level category, e.g. communication, our counterfactual would become a representation of a basket in which communication had a lower importance than it actually had in the past. Thus, to maximize the similarity between the basket used for the counterfactual estimates and the one in place before the update in the weighting structure, we impute the item-level average price variation of the remaining subitems to the missing ones (Figure 1). If all the elements at the item level are missing, we impute the subgroup-level average to them, and so forth. Implicitly, our hypothesis is that the expected subitem’s price variation correlates with the one from similar subitems. The price index INPC seeks to approximate a cost-of-living index (IBGE, 2005, p.13). The index is based on a typical consumer basket observed in a given period and which is composed of several categories, whose weights represent the category’s cost as a share of the basket’s total cost. Once the categories’ weights are observed in the first period, IBGE monitors only the variation of prices over time. The variation of the price index is thus calculated by averaging the price variations of each category weighted by the weight of each category in the basket in the base period. This index is equivalent to a Laspeyres Index, in which the quantities consumed in the base period are kept fixed and the basket cost is updated by the new prices over time. Iteratively, the weight of each category is adjusted over time by taking into account only the price change. A category’s weight is updated by multiplying the base weight by a correction factor that considers the ratio of the category's price change to the basket’s price change. IBGE follows this dynamic of estimation of the index variation and update of weights until the estimation of a new typical basket through a new POF is implemented. January 2020 and the last year-month before an update of INPC due to a new POF (i.e, subsequent to the POF 2017/2018). 2 Another potential shortcut to overcome the barrier of missing data would be to estimate the counterfactual at the item level. In Appendix A, we provide examples of how that would mislead the counterfactual estimates. Page 4 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Since May 2018, the index is calculated for 16 regions and the national-level INPC is a weighted average of the region-specific indices. The regional weight is determined by the region’s population. Formally, the index can be defined as follows. Let be a given category, a given region, a given year-month, ,, the price of category in region in period and ,, the quantity of category consumed in region in period . The basket’s total cost in region in period is thus the sum of the expenditures associated with each category considered in the basket, and can be defined as ∑ ,, ,, , with the weight of category in region in period being defined as ,, = ,, ,, ∑ ,, ,, . In turn, the price variation of category in region from period − 1 to period is given by − ,,−1 ,, = ,, . Thus, if the index is called and consists of a base index number, one can define ,,−1 the variation of the of region from period − 1 to period as the following: ∑ ∈ , ,,−1 ∗ ,, (1) , = ∑ ∈ , ,,−1 where , is the INPC variation of region , and , is the set of categories that make up the INPC consumption basket of region in period . The national-level INPC is thus calculated as ∑ ∈ , ∗, (2) , = ∑ ∈ , where , is the national-level INPC variation, , is the weight of region in the national-level INPC in period , and is the set of regions that compose the national-level INPC in period . The weight of category in region , in turn, is updated from period − 1 to period by the following formula: ,,−1 ∗(1+,, ) (3) ,, = (1 +, ) Figure 1. Imputing price variation to no-longer-observable goods to estimate counterfactual inflation Source: elaborated by the authors. Page 5 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 The counterfactuals Over the past 20 years, IBGE has updated the INPC weighting structure three times. The first was in July 2006 as a result of POF 2002/2003 (IBGE, Technical Note 01/2006), the second in January 2012 as a result of POF 2008/2009 (IBGE, 2012), and the third in January 2020 after POF 2017/2018 (IBGE, 2020). In the interval between updates, the weights are updated according to equation (3) above. This note presents two counterfactual exercises whose assumptions are illustrated below (Figure 2). In the first hypothetical situation we simulate a scenario where the weighting structure update that took place in January 2020 (based on POF 2017/2018 information) did not occur. That is, instead of using the published weights, equation (3) is used to update the weights between the published monthly data in December 2019 and January 2020. Then, the national-level INPC variation in January 2020 is calculated using equation (2). Iteratively, it is possible to re-estimate the entire index series. The second counterfactual is based on the same method described above, but for the month of January 2012, when the INPC weighting structure was updated due to POF 2008/2009. Instead of estimating inflation with the weights published in that month, the counterfactual weights are calculated using the monthly data published in December 2011. From then on, a new inflation series is estimated as if no update of the weighting structure had occurred since then. To take into account changes in the geographical coverage of INPC over time, we estimate two versions of the first counterfactual exercise. When IBGE implemented the weights update of January 2012, it used to collect price data in 11 regions to calculate the country-level INPC. Even before the next consumption basket update, IBGE added five regions to the national index. To be sure, we built a counterfactual that keeps the original regional weights from January 2012 constant over time, and another one that weights the counterfactual inflation considering the 16 regions that compose the national-level index since May 2018. There are no significant differences between the two approaches, as demonstrated in Appendix B. For simplicity, in what follows we only show the results that consider the full set of regions in the national-level INPC. Figure 2. POF and the update of the weighting structure over time in different counterfactual designs Source: elaborated by the authors. Page 6 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Not updating the INPC weights in 2020 would have led to a higher annual inflation rate3 in the first sixteen months, but to a lower one between May 2021 and June 2022. Figure 3 shows the evolution of the 12-month INPC variation comparing the factual with the counterfactual one in the period between January 2020 and May 2023. We see in the figure that the counterfactual rate remained higher than the factual one from January 2020 to April 2021, but then the direction of the difference shifts until June 2022. This result has an important consequence for the debate about the periodicity of the POF: the counterfactual annual inflation rate is not always higher or lower than the factual one. Figure 3: Factual and counterfactual annual inflation rate between January 2020 and May 2023 – Exercise 1 Source: elaborated by the authors using SIDRA IBGE data. Had there been no update to the INPC weighting structure in 2020 based on the 2017/2018 POF, measured inflation rate at the end of that year would have been 0.8 percentage points higher. Table 1 shows the factual and counterfactual annual inflation rate in December of each year between 2020 and 2022. As can be seen, at the end of the first year, December 2020, the difference between the counterfactual and factual INPC was around 0.8 percentage points. This result shows that even in a relatively short period of one year, the differences between the actual and the counterfactual can be quite significant. 3 The annual inflation rate is the movement of INPC from one month to the same month of the previous year, expressed as a percentage. Page 7 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Table 1: Annual inflation rates between 2019 and 2022 – Exercise 1 Factual Counterfactual Difference Year-month (%) (%) (p.p.) 2019m12 4.48 4.48 0.00 2020m12 5.45 6.26 0.82 2021m12 10.16 9.11 -1.05 2022m12 5.93 7.03 1.10 Source: elaborated by the authors using SIDRA IBGE data. A detailed analysis suggests that the net result between actual and counterfactual inflation depends on the combination of the difference in weights and the observed price variation for each item in each specific month. Table 2 shows the comparison of the factual and counterfactual INPC in each month of 2022, highlighting the month in which we see the largest difference between factual and counterfactual variation, and when the counterfactual annual inflation rate crosses back towards a positive difference in respect to the factual. This change in direction was the month in which an exemption of fuels and energy services taxes was implemented. As the share of the expenditures on gasoline and other exempted goods is lower in the outdated basket than in the updated one, the counterfactual underestimates the fall in prices in that period. In Appendix C, we show a more detailed analysis of the differences relative to this period. Table 2: Factual and counterfactual INPC in the period between January and December 2020 – Exercise 1 CPI monthly variation CPI 12-month variation Year- Factual Counterfactual Diff. month Factual (%) Counterfactual (%) Diff. (p.p) (%) (%) (p.p) 2022m1 0.67 0.67 0.00 10.60 9.45 -1.15 2022m2 1 0.91 -0.09 10.80 9.72 -1.08 2022m3 1.71 1.83 0.12 11.73 10.97 -0.76 2022m4 1.04 1.22 0.18 12.47 11.84 -0.62 2022m5 0.45 0.36 -0.09 11.90 11.23 -0.67 2022m6 0.62 0.64 0.02 11.92 11.29 -0.63 2022m7 -0.6 -0.09 0.51 10.12 10.22 0.10 2022m8 -0.31 -0.07 0.24 8.83 9.23 0.40 2022m9 -0.32 -0.19 0.13 7.19 7.88 0.69 2022m10 0.47 0.46 -0.01 6.46 7.22 0.76 2022m11 0.38 0.41 0.03 5.97 7.02 1.04 2022m12 0.69 0.68 -0.01 5.93 7.03 1.10 Source: elaborated by the authors using SIDRA IBGE data. In the hypothetical scenario in which the country would have postponed POF in the earlier decade, inflation would have been also mismeasured. Figure 4 shows the comparison of the factual annual inflation rate with the counterfactual one in the period between January 2012 and December 2014 considering the second exercise . As in the first exercise, the counterfactual 12-month inflation is higher in the first period of analysis but then crosses down the factual inflation from the last quarter of 2013. The difference is 0.8 percentage point in 2012’s inflation rate (Table 3), similar in magnitude to the first year of analysis in Exercise 1. As discussed before, the results are driven by the combination between the weights difference and the price variation associated to the goods that lost or gained more importance. Thus, nothing would warranty a 0.8 p.p. difference for every counterfactual first year. Nonetheless, this second exercise reinforces that delaying CPI’s Page 8 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 weighting structure update due to POF absence does have important consequences on measured inflation rates. Figure 4: Factual and counterfactual annual inflation rate between January 2012 and December 2014 – Exercise 2 Source: elaborated by the authors using SIDRA IBGE data. Table 3: Annual inflation rates between 2011 and 2014 – Exercise 2 Counterfactual Difference Year-month Factual (%) (%) (p.p.) 2011m12 6.08 6.08 0.00 2012m12 6.20 6.98 0.79 2013m12 5.56 5.36 -0.20 2014m12 6.23 6.00 -0.23 Source: elaborated by the authors using SIDRA IBGE data. The consequences of not frequently conducting the POF on official inflation measured by consumer price indices may be huge when it comes to the public budget. Many budget items are directly affected by inflation, as well as the monetary policy through the definition of the basic interest rate in Brazil, the SELIC. This implies that the impacts on the Brazilian economy can be substantial. For instance, the share of pensions and compensations of employees were, respectively, 30 percent and 13 percent of the central government of Brazil’s primary expenditures in 2022 (STN, 2023). They are directly affected by Page 9 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 the minimum wage adjustment derived from INPC measurement. In addition, some public debt securities of the federal government are indexed by the IPCA, which implies that the cost of the debt itself is also impacted by the measurement of official inflation in the country. If the update of the INPC weighting structure had not occurred in January 2020, the federal government expenditure with pensions benefits in 2021 would have increased by BRL 5.45 billion. To show an estimate of the economic costs of not performing POF, we present the effects on the monetary updating of the federal government's expenditure with social security benefits. According to data from the Central Government Primary Results, expenses with pensions benefits in Brazil in 2021 were of the order of BRL 709.5 billion. In turn, the results relative to our first counterfactual exercise indicate that the factual annual inflation rate was 5.45% in December 2020, while the counterfactual one was 6.26% (see Table 1)- a difference of 0.81 percentage points. If we consider that the pension expenses in 2021 had been updated by the counterfactual INPC of December 2020 (6.26%), the amount would change from BRL 709.5 billion to BRL 715.0 billion. If inflation is being overestimated, then government is wasting resources that could be used to finance other programs. For example, the savings from the update of the INPC weighting structure in 2020 represents 8.8 percent of the cost of the Continuous Benefit Program (Benefício de Prestação Continuada, BPC) in 2021 (Table 4). The estimate also represents 9.6 percent of the cost of the Emergency Cash Program (Auxílio Emergencial) in 2021 and 17.5 percent of the cost of the Brazilian flagship cash transfer program, the Bolsa Família, in 2019. Such results show that the difference between updating expenditures with social security by the counterfactual and factual INPC variation is substantial even over a one-year period. Table 4: Comparing the estimate with the cost of some important social programs in Brazil Difference as a Program Cost (BRL billion) share of the programs’ costs BPC (2021) 61.8 8.8% Auxílio Emergencial (2021) 56.8 9.6% Bolsa Família (2019) 31.1 17.5% Seguro-desemprego (2021) 30.8 17.7% Bolsa Família (2021) 27.2 20.0% Auxílio Brasil (2021) 6.4 85.2% Source: elaborated by the authors using data from the Open Data Portal of the federal government. This note presents the results of two counterfactual exercises that simulate hypothetical situations in which it is possible to observe the variation of INPC had POF not been carried out with a certain periodicity. The first exercise simulates a situation in which the consumption basket and the INPC weights were not updated in January 2020 or in any subsequent period, and the second exercise assumes that they were not updated in January 2012. The results show that the counterfactual INPC can be significantly different from the factual ones in the two hypothetical scenarios, especially if specific prices suffer a shock, as in the case of tax exemptions for fuels in 2022 These results suggest that the untimely update of the CPI consumption basket in Brazil can have large economic consequences, not only on households’ welfare, but also on public spending. Currently, the P a g e 10 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 update of the CPI weighting structure is done only when a new HBS is carried out so that not frequently conducting the HBS implies not frequently updating the CPI structure. Therefore, the evidence highlights the importance of frequently conducting HBS due to its link with the official price indices4. It should be noted that the results presented here are context-specific and reflect particular changes in the consumption patterns in Brazil during the study period, i.e., January 2020 to May 2023 in the first exercise and January 2012 to December 2014 in the second one. One of these changes was the emergence of the COVID-19 pandemic and the period of high inflation in the country, which directly impacts the estimates of the first counterfactual exercise. Thus, there is no guarantee that these results will be the same in the future, what reinforces the necessity of frequent and representative data on household consumption habits over time. The case of price indices in Brazil is illustrative of the reality of potentially other several countries in which HBS are the main source of data for updating the consumption basket of CPIs and in which such surveys are conducted with low periodicity. Our results show that given the estimated cost of such surveys, the economic costs may be substantially higher in the absence of up-to-date data on consumption patterns of the population. 4 One may note that CPI weights structure could be updated using more frequent national accounts estimates. However, national accounts’ expenditures estimates are partially based on HBS estimates, thus also depending on the latter (UN, 2009, paragraph 4.40). P a g e 11 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Hill, Peter (2010). Lowe indices. In: 2008 World Congress on National Accounts and Economic Performance Measures for Nations. Washington, DC. IBGE (2005). Sistema Nacional de Índices de Preços ao Consumidor: estruturas de ponderação a partir da Pesquisa de Orçamentos Familiares 2002-2003. Séries Relatórios Metodológicos v.34, Rio de Janeiro. IBGE (2020). Sistema Nacional de Índices de Preços ao Consumidor: métodos de cálculo. Séries Relatórios Metodológicos v.14, 8ª edição, Rio de Janeiro. IMF, ILO, OECD, EU, UN, and WB. (2020). Consumer price index manual: concepts and methods. (International Monetary Fund, International Labour Office, Organisation for Economic Co-operation and Development, European Union, United Nations, World Bank). Washington, DC : International Monetary Fund STN (2023). Estatísticas Fiscais do Governo Central Orçamentário. Secretaria do Tesouro Nacional. Brasília- DF, Brasil. UN (2009). Practical Guide to Producing Consumer Price Indices. United Nations, Geneva, Switzerland. P a g e 12 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 IBGE publishes price variation data of consumption goods at different aggregation levels. IBGE releases the data on the price variation at the lowest level of aggregation (subitem), but also at other levels that aggregate subitems by their similarity (i.e., items, subgroups, groups). For example, rice is a subitem, that is grouped into the item “Cereals, Pulses, and Oilseeds”, that is aggregated at the “Home food” subgroup, which, by its turn, belongs to the more aggregated group “Food and beverages”. When the consumption baskets of the consumer price indices are updated, several subitems may leave the baskets, at the same time as many or more may join it. These replacements reflect the change of consumption habits over time, which can be explained by technological changes and cultural dynamics. Importantly, however, the categories of item, subgroup and group rarely change. A shortcut to perform the counterfactual exercises conducted in this note would be working with higher levels of aggregation. As mentioned in the note, a barrier to the counterfactual computations is the fact that we do not observe the price variation of the subitems that were removed from the CPI basket. A straightforward way to overcome it would be working with the item-level data of price variation and weights, ignoring the relative changes that happen within the item category. Working at upper levels of aggregation level (e.g., items) has some consequences for our results. For example, we do not capture the substitution effects between categories belonging to the same level of aggregation, as is the case of black beans (feijão-preto) and brown beans (feijão-carioca). Thus, if brown beans used to have a higher share of the households’ expenditures and for some reason its price increased relatively more than the black beans’ one (e.g., due to a specific plague affecting brown beans), we would underestimate price increases in the counterfactual scenario because price variation published at the item-level already takes into account the factual weighting of subitems. The upper aggregations mislead the counterfactual because they do not take into account subitem rebalancing at the price average variation. We provide a numeric example and illustration of this fact using the aggregated categories below (Figure A.1). Before January 2020, rice used to represent 0.72 percent of the expenses made by the typical consumer of Sao Paulo, while this share was 0.49 percent for brown beans. This means that rice had 60 percent of the total 1.21 percent spent at the item relative to cereals (Cereais, leguminosas e oleaginosas), while brown beans got 40 percent. After the update of the weighting structure, the shares of rice and brown beans became 68 percent and 32 percent within cereals, respectively, whereas cereals got a lower share of the total expenditures (0.91 percent). Rice had a higher price variation than brown beans in the recent period. As a consequence, in the analysis with subitem-level data, cereals have a higher share of the expenditures (1.21% vs. 0.91%), but its price variation is smoothed by a relatively lower share of rice within the item level (60% vs. 68%). However, if we had worked with item-level data, we would get the same higher share of cereals without compensating for the price variation. Overall, we would have overestimated the counterfactual INPC had we worked with higher levels of aggregation. Figure A.2 shows the trajectory of the factual annual inflation at the item level and the counterfactual one at the item, subgroup, and group levels. P a g e 13 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Figure A.1. How the item-level aggregation can mislead the counterfactual price variation a. Factual and counterfactual data for “Cereals, pulses and oilseeds” at different levels Factual Subitem W var_p Item W var_p 1101.Cereais, 1101002.Arroz 0.62 leguminosas e 0.91 1.33 2.62 oleaginosas 1101051.Feijão - mulatinho - - 1101052.Feijão - preto - - 1101053.Feijão - macáçar (fradinho) - - 1101073.Feijão - carioca (rajado) 0.29 -1.4 Data CF at the item level Item W var_p Item W var_p 1101.Cereais, 1101.Cereais, leguminosas e oleaginosas 1.21 1.33 leguminosas e 1.21 1.33 oleaginosas Data CF at the subitem level Subitem W var_p Item W var_p 1101.Cereais, 1101002.Arroz 0.72 2.62 leguminosas e 1.21 0.99 oleaginosas 1101051.Feijão - mulatinho - - 1101052.Feijão - preto - - 1101053.Feijão - macáçar (fradinho) - - 1101073.Feijão - carioca (rajado) 0.49 -1.4 1101075.Feijão - branco - - 1101079. Milho em grão - - b. Price variation of rice and brown beans c. Cereals price variation: factual vs. counterfactual at the item level Source: elaborated by the authors using SIDRA IBGE data. P a g e 14 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Figure A.2. Comparing the factual 12-month INPC variation at the item level with the counterfactual ones at the item, subgroup and group levels – Exercise 1 30.0 25.0 20.0 15.0 10.0 5.0 0.0 Feb-20 Sep-20 Oct-20 Jan-21 Feb-21 Sep-21 Oct-21 Feb-22 Sep-22 Oct-22 Jan-20 Mar-20 Nov-20 Mar-21 Nov-21 Jan-22 Mar-22 Nov-22 May-20 Jun-20 May-21 Jun-21 May-22 Jun-22 Apr-22 Apr-20 Jul-20 Apr-21 Jul-21 Jul-22 Aug-20 Dec-20 Aug-21 Dec-21 Aug-22 Dec-22 Factual (item level) Item Subgroup Group Source: elaborated by the authors using SIDRA IBGE data. P a g e 15 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 The number of regions considered in the calculation of the national-level INPC has changed over time. In 1979, when the index started to be calculated, prices were collected only in Rio de Janeiro-RJ, Porto Alegre-RS, Belo Horizonte-MG, and Recife-PE. In 1980, Sao Paulo-SP, Brasilia-DF, and Belem-PA were included. Relevant for our period of analysis, Campo Grande-MS and Vitoria-ES were added in 2014, while Rio Branco-AC, Sao Luis-MA, and Aracaju-SE were added in May 2018. This implies that theoretically we could consider either the counterfactual situation in which nothing has changed since the former weighting structure update or the one in which new regions have been introduced even, but the consumption basket of reference is the same. In order to test the sensitivity of the counterfactual INPC to different regional weights specifications, we estimate the national-level counterfactual INPC by averaging the regional estimates in both counterfactual situations (Table B1). As can be seen, the differences between the two hypothetical scenarios are not substantive. In particular, the regions that were introduced over time have a minor share in the national index, thus not contributing to significant changes in the counterfactual estimates (Figure B1). As explained in the main text, the regional weights are determined by the regions’ populations. The same was replicated for the second counterfactual exercise by estimating the counterfactual inflation using i) the original weighting structure, and ii) considering the introduction of new regions in 2014, and the differences are minor. Table B.1. Different regional weighting structures for the national-level counterfactual INPC a. 16 regions b. 11 regions Region INPC weight (%) Region INPC weight (%) Brasil 100 Brasil 100 Rio de Janeiro 9.38 Rio de Janeiro 9.91 Porto Alegre 7.15 Porto Alegre 7.38 Belo Horizonte 10.35 Belo Horizonte 11.04 Recife 5.6 Recife 7.17 São Paulo 24.6 São Paulo 25.24 Brasília 1.97 Brasília 2.39 Belém 6.95 Belém 7.03 Fortaleza 5.16 Fortaleza 6.61 Salvador 7.92 Salvador 10.67 Curitiba 7.37 Curitiba 7.29 Goiânia 4.43 Goiânia 5.27 Vitória 1.91 Campo Grande 1.73 Rio Branco 0.72 São Luís 3.47 Aracaju 1.29 Source: IBGE (2020). P a g e 16 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Figure B.1. Comparing the counterfactual 12-month INPC variation with the factual one in the period between January 2020 and May 2023 – Exercise 1 Source: elaborated by the authors using SIDRA IBGE data. Notes: “Counterfactual a”=first way of calculating the counterfactual inflation referred in this Appendix, “Counterfactual b”=second way of calculating the counterfactual inflation referred in this Appendix. The 12-month inflation in year-month t is calculated as the cumulative inflation that takes into account the inflation in year-month t and the eleven year-months prior to t. P a g e 17 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 The tax exemption for gasoline and some selected services approved in 2022 sheds light on the differences between factual and counterfactual weighting structures. When comparing the counterfactual with the factual data, we observe that the biggest difference materializes in July 2022. Not coincidently, this is the month right after the approval of Complementary Law No. 194/2022, which exempted taxes on fuels, electricity, telecommunication services, and public transportation. The exemption resulted in an approximate 15-percent reduction of gasoline prices in all regions in which prices for INPC are observed. Importantly, gasoline has currently the top share in the INPC basket of most regions. But, before the last weighting structure update, it used to have a smaller share (Figure C.1). Not without reason, July 2022 is the tipping point when the counterfactual inflation accumulated since the start of the calculations gets irreversibly higher than the factual one (Figure C.2). Figure C.1. Gasoline shares in the INPC consumption basket of different regions (Counterfactual x Factual) Source: elaborated by the authors using SIDRA IBGE data. Notes: BH (MG)=Belo Horizonte-MG, Bel (PA)=Belem-PA, Bra (DF)=Brasilia- DF, For (CE)=Fortaleza-CE, PoA=Porto Alegre-RS, and SP (SP)=Sao Paulo-SP. The blue bars represent the counterfactual shares, while the red ones represent the factual ones. Selected regions used for simplicity. P a g e 18 POVERTY AND EQUITY NOTES Budget surveys and price indices: the case of Brazil SEPTEMBER 2023 Figure C.2. Comparing the counterfactual cumulative INPC variation with the factual one since January 2020 Source: elaborated by the authors using SIDRA IBGE data. Notes: Cumulative price variation (%) calculated since the start of the counterfactual exercise in January 2020. P a g e 19