## Section I. Detailed Technical Assessment and Recommendations ## Source details **Canonical URL:** [Section I. Detailed Technical Assessment and Recommendations](https://www.imf.org/-/media/files/publications/tar/2024/english/tarea2024023.pdf) ## Other formats - [Markdown version](/-/media/files/publications/tar/2024/english/tarea2024023.pdf.md) - [Structured JSON version](/-/media/files/publications/tar/2024/english/tarea2024023.pdf.json) --- ### A. Introduction - Field price collection is the main data source for the CPI of Moldova; the basket comprises around 1,300 representative products for which prices are collected in the major cities. - Price collectors use tablets to record observed prices; collected prices are processed to obtain price indices for required aggregates. - The NBS plans to modernize CPI compilation by using Scanner Data, which typically cover all items sold by a retailer at all locations and include quantities sold and revenue. - Advantages of Scanner Data: - Improved product coverage (all products sold by a retailer instead of a sampled few). - Improved regional coverage (all outlets of participating retailers). - Expanded time coverage (large part of the reference month vs. one or two price observations). - Contains quantities as well as prices enabling detailed weighting at the individual item level and use of weighted index formulas. - Use of Scanner Data will reduce field collection costs but increase processing and analysis complexity. - Roadmap for integrating Scanner Data: - Transmission and pre-processing: 2023–2024 - Classification: 2023–2024 - Index compilation: 2024–2025 - Data integration: 2025–2026 - Finalization of production system and publication of the Scanner Data methodology: 2026 - Scanner Data in the published CPI: January 2027 ### B. Transmission and Pre-processing Findings: - NBS market analysis identified five retail chains representing together more than 50 percent of total sales in the food retail sector. - COICOP Division 01 Food and non-alcoholic beverages weight in the CPI basket: 35 percent in 2023. - Theoretically, Scanner Data from the five retailers could represent around 18 percent of the CPI basket for COICOP Division 01. - Practical rollout suggested: start with food products from one or two retail chains, then extend to non-food and other chains. - Initial contacts with major retailers for direct Scanner Data access were not successful; retailers reluctant to share detailed data and already provide similar information to STS. - STS introduced an electronic sales monitoring system in July 2023 recording each receipt line: text item description, number of items purchased, VAT rate, and price. - A test data set from STS covered receipts for two major retailers during the first six days in September 2023; mission examined only the first retailer. - Raw test data: over one million lines (each a receipt); transformed table: over five million lines (each a specific item purchase on a receipt). - Data lacks a formal product identifier (barcode/shop code); item descriptions are free text. - Technical feasibility confirmed to create an item code from the text string (same text string → same code); validity must be monitored. - Suggested temporal aggregation: first three weeks of the month to allow processing time. - Suggested grouping: all outlets of a retail chain in a region. - Aggregation formulas suggested (preserve notation): - quantities_i^t = Σ_r q_{i,r} - sales_i^t = Σ_r p_{i,r} q_{i,r} - price_i^t = sales_i^t / quantities_i^t - Pilot data contained around 19,000 items (food and non-food), a plausible assortment size for a retailer. - Need to discuss regular and secure transmission protocol with STS and consider further pre-processing by STS. - A formal memorandum of understanding between NBS and STS is recommended to define: contents and characteristics of the data set, transmission rules (frequency, technical characteristics), and other provisions. - IT infrastructure required at NBS for storing and transforming large files; target output table variables by region and item: text string describing the item; an item code; total sales; total quantities; the price; the region where the item was sold; the date. Recommended Actions: - Develop a memorandum of understanding with the STS to arrange a regular and secure transmission of the data. - Develop an IT infrastructure at the NBS for storing and pre-processing the STS data. - Use the text string to identify an item and calculate average prices at the level of an item, retailer and region. ### C. Classification Findings: - Scanner Data items must be mapped to the CPI classification; Moldavian CPI uses 5-digit ECOICOP (European variant of COICOP) with an additional 6-digit level in the basket; representative products are selected below the 6-digit level. - Mapping errors cause misclassification and incorrect price indices; in principle all Scanner Data items should be classified. - Mission recommendation: map items to the 6-digit level of COICOP (not to representative products). - Reference table requirements: - Variables: item description; item identifier; 6-digit COICOP code. - No duplicates; each item mapped to a single COICOP code. - Initialization: map all items in the dataset at once. - Recurrent monthly procedure: for each item in a month: - If item already in reference table: do nothing. - If item not included: assign 6-digit COICOP and update reference table. - Once created, reference table is augmented each month and no records should be removed. - Classification methods: - Manual labelling: feasible using item descriptions and e.g., Excel, but resource intensive; treat as last resort. - Attribute-based classification: requires auxiliary classification already labelling items; map auxiliary categories to COICOP to automatically assign COICOP categories; performance depends on mapping quality and may need manual follow-up. - Additional metadata (GTIN, retailer-specific classification, item weight) will facilitate COICOP mapping and package-size adjustments. - NBS should request metadata from retailers; mission examined a retailer web shop whose assortment classification is relatively well suited to map to COICOP. - Additional metadata should be linked to the text descriptions provided to STS. Recommended Actions: - Map Scanner Data items to the 6-digit COICOP level and store mappings in a non-duplicating reference table. - Use attribute-based classification when auxiliary classifications or metadata (e.g., GTIN, retailer classification) are available; rely on manual labelling only when necessary. - Request additional item metadata from retailers and link it to STS text descriptions to support automated classification. Classification and machine learning - Machine Learning methods can create a classifier that learns from already classified items; once estimated, new incoming items can be automatically classified. - A fully classified training data set must be available to design a classifier and evaluate its performance. Recommended Actions: - Reach out to the retailers to access additional metadata of the items in view of applying attribute-based classification. - Classify already available items to the 6-digit COICOP and put in place a regular process for updating the reference table. Transition to COICOP 2018 - The NBS will change to the COICOP 2018. - The correspondence between COICOP 2018 and the COICOP version currently used in the Moldavian CPI is less problematic for food products (the main scope of Scanner Data). - An item mapped to the current COICOP version will have to be re-assigned to the relevant COICOP 2018 category. ### D. Index Compilation and Data Integration Key points from the mission: - Derive average item prices from aggregated sales and quantities (price_i^t = sales_i^t / quantities_i^t). - Process Scanner Data prices separately, potentially using superlative index formulas (Fisher). - Introduce an aggregation structure that distinguishes at the lowest hierarchy level between Scanner Data and field-collected prices. - Plan timeline: index compilation development 2024–2025; data integration 2025–2026; finalize production system and methodology publication in 2026; incorporate Scanner Data in published CPI by January 2027. Scanner Data applications and index compilation: - Can serve as a sampling frame for selecting representative products; review/specification of representative products is conducted at year-end to update the next year’s basket. - Can identify the most popular products to benefit the NBS relatively quickly. - Aim to use all items (not only a sample) and weight each item according to its importance. - Index formulas illustrated: Jevons, Laspeyres, Paasche and Fisher. The Fisher price index is one of the preferred index number formulas. - In Scanner Data contexts, bilateral Fisher price indices are typically embedded into a multilateral method. Quality/weight changes and weight-adjustment techniques: - Matching item codes will not capture pure price changes linked to changes in the weight of an item (e.g., 200g to 180g with unchanged price should record a price increase). - Techniques require connecting “same” items with different weights and adjusting based on weight; text string extraction was explored to access item weight, but preferred option is to obtain weight directly from retailers. Integration with CPI: - Scanner Data will not cover all household purchases; field price collection in shops not covered by Scanner Data cannot be stopped. - Best practice: compile separate price indices with Scanner Data and combine them with field price collection indices using an appropriate aggregation methodology. - Current calculation is at COICOP 6-digit category broken down by region. A target aggregation structure was developed to integrate Scanner Data into the Moldavian CPI. This aggregation structure is for internal NBS calculations only and does not impact published dissemination. Target Aggregation Structure (Index structure — Weights – Data source — Index method): - COICOP 5-digit — HBS — Laspeyres-type - └ COICOP 6-digit — HBS — Laspeyres-type - └ Region — HBS — Laspeyres-type - └ Field price collection — Enterprise statistics (turnover) — Laspeyres-type - └ Repres. Prod. — No weights — Jevons - └ Scanner Data — Retailer Enterprise statistics (turnover) — Laspeyres-type - └ Item*Region*Retailer — Scanner data — Multilateral (Fisher) Progressive introduction of Scanner Data: - With annual weight updates, aggregation structure can be updated each year, adding new retailers or more COICOP 6-digit products covered by Scanner Data. - When integrating Scanner Data from a retailer into the CPI, prices of that retailer must be removed from the field price collection index. IT architecture and software: - CPI software should be adjusted to integrate Scanner Data. - Possible architecture: separate system for processing Scanner Data (e.g., R software) producing indices at retailer-region-COICOP6 level; these indices injected into main CPI software for combination with field price indices. - Current CPI software limitations: combines short-term (current month vs previous month) indices from basic level to higher levels using geometric mean at basic level and fixed weights introduced in January each year. - Technical note: weights would have to be price-updated before combining short-term indices, or short-term indices can be chained at the basic level before aggregation to higher levels. - NBS plans to improve CPI software and review aggregation mechanisms; mission prepared examples to illustrate aggregation methods for IT specifications. Other uses of Scanner Data: - NBS compiles and publishes average prices for a list of socially important retail products; Scanner Data could be used to select specific items meeting product specifications and insert “Scanner Data average price” into calculations. Recommended Actions: - Use Scanner Data for reviewing the basket of representative products. - Improve the aggregation methods in the CPI software. ### E. Owner-Occupied Housing (OOH) Findings and recommendations: - CPI for housing rents can be improved by expanding coverage. - NBS will start collecting rents for houses located in Chisinau from the same real estate listing websites used for apartment rents. - Accuracy of the rent price index can be improved by developing explicit stratification by region and type of dwelling. - An improved rental price index can be the basis for including OOH in the CPI using the imputed rentals approach. - Given the small rental market in Moldova, further analysis and user consultations are needed to assess whether the imputed rentals approach is a viable option. Rental equivalence approach and alternatives: - Rental equivalence: measure change in price of housing services consumed by owner-occupiers by estimating market value of those services. - Acquisitions approach: measures changes in prices of dwellings and goods/services acquired by households as owner-occupiers. - Payment approach: measures changes in payments made by owner-occupiers (e.g., mortgage interest payments). Current practice and market characteristics: - CPI includes rentals collected for different apartment types characterized by number of rooms and vintage (new or old); data collection at end of each month. - Rents for Chisinau are collected from popular real estate listing websites; other regions by price collectors. - Rents labelled in euros are transformed into local currency using appropriate exchange rates. - Rental market in Moldova is small. According to the HBS, the share of owner-occupiers in Moldova is estimated to be 95 percent. - Inclusion of OOH under rental equivalence is expected to increase the combined share of rentals and imputed rentals to more than 15 percent. - Mission conducted an experimental simulation including imputed rentals by assuming the imputed rental index equals the current rental index. Improvements and recommended actions: - Price index for rentals should include both apartments and houses; NBS agreed to collect and include rents for houses in Chisinau from next year on. - Internal weights of the rental price index should be reviewed using HBS expenditure data broken down by region and by type of dwelling. - Opportunity to automatize and expand data collection via web scraping; NBS should reach out to NBM which already scrapes data for the RPPI. - Further analysis and user consultations needed to assess feasibility of imputed rentals approach. Recommended Actions: - Expand the coverage of the rent price index and include rents for houses in the CPI. - Start discussions with the NBM on web scraping rents from real estate listing websites. ### Summary of Mission Outcomes and Priority Recommendations Selected outcomes: - TA mission assisted NBS with introducing alternative data sources for CPI and examined treatment of OOH. - Incorporating Scanner Data will improve CPI quality: regional, product and temporal coverage; availability of price and quantity at item level enables more accurate indices. - STS Scanner Data assessed as suitable with some transformation; STS electronic sales monitoring records each receipt line; sample examined covered one week of receipts for a major retail chain. - IT infrastructure and pre-processing required; raw STS data large and requires item identification (no formal item identifier in raw data); NBS should request GTIN or retailer-specific classification metadata from retailers. - Scanner Data should be incorporated into the published CPI by the end of 2026; updated aggregation structure developed; prices from Scanner Data could be processed separately using superlative price index formulas (Fisher). - Improve CPI housing rents coverage and stratification; further work needed to assess OOH inclusion via imputed rentals. Priority recommendations with target dates and responsibilities: - March 2024: Establish a memorandum of understanding with the STS to secure regular access to Scanner Data — Responsible: NBS - July 2024: Assign COICOP categories to the items included in the Scanner Data — Responsible: NBS - September 2024: Improve higher-level aggregation in the CPI software — Responsible: NBS Final recommended near-term actions: - Establish memorandum of understanding with STS for regular access to Scanner Data. - Develop IT infrastructure for storage/pre-processing at NBS. - Assign COICOP categories to Scanner Data items and maintain a non-duplicating reference table. - Improve CPI software aggregation and prepare for staged integration of Scanner Data into indices by January 2027. ### Work plan milestones (October 2023–October 2024) - Result: Transaction data is regularly transmitted from the STS to the NBS. - High — Establish a memorandum of understanding with the STS to secure regular access to Scanner Data. — March 2024 - High — Develop an IT infrastructure at the NBS for storing and pre-processing the STS data. — March 2024 - Result: Scanner Data is classified to COICOP. - High — Reach out to the retailers to access metadata on the products in view of applying attribute-based classification. — January 2024 - High — Classify already available items in the Scanner Data to the 6-digit COICOP. — July 2024 - Result: The CPI software is improved. - High — Improve the aggregation methods in the CPI software. — September 2024 - Result: Scanner Data is used. - Medium — Use Scanner Data for reviewing the basket of representative products. — October 2024 - Result: The rental price index is improved. - Medium — Update the basket and include rents for houses in the CPI. — January 2024 - Medium — Start discussions with the NBM on web scraping prices from the real estate listing websites. — July 2024 ### Appendix I — Higher-level aggregation example (regional to national) Example regions and weights: - Chisinau — 25 - North — 40 - South — 25 - Centre — 55 - TOTAL — 145 Example chaining and combining regional short-term indices to obtain national long-term indices: - January Index = (104.17*25+103.33*40+111.11*25+106.67*55) / 145= 106.08 - February Index = (116.67*25+114.00*40+112.35*25+127.78*55) / 145 = 119.40 - March Index = (121.67*25+116.00*40+114.57*25+136.67*55) / 145 = 124.57 - April Index = (100.00*25+100.00*40+100.00*25+100.00*55) / 145 = 100.00 Alternative method (update weights first, then combine short-term indices) produces identical long-term national indices using updated weights and chaining. *Source: Section I. Detailed Technical Assessment and Recommendations — IMF Technical Assistance Report – Moldova Consumer Price Index (content unit: tarea2024023).* ### Section I. Detailed Technical Assessment and Recommendations ...................................................... 5 ### Section I. Detailed Technical Assessment and Recommendations ### A. Introduction - Field price collection is the main data source for the CPI of Moldova; the basket comprises around 1,300 representative products for which prices are collected in the major cities. - Price collectors use tablets to record observed prices; collected prices are processed to obtain price indices for required aggregates. - The NBS plans to modernize CPI compilation by using Scanner Data, which typically cover all items sold by a retailer at all locations and include quantities sold and revenue. - Advantages of Scanner Data: - Improved product coverage (all products sold by a retailer instead of a sampled few). - Improved regional coverage (all outlets of participating retailers). - Expanded time coverage (large part of the reference month vs. one or two price observations). - Contains quantities as well as prices enabling detailed weighting at the individual item level and use of weighted index formulas. - Use of Scanner Data will reduce field collection costs but increase processing and analysis complexity. - Roadmap (TABLE 2) for integrating Scanner Data: - Transmission and pre-processing: 2023–2024 - Classification: 2023–2024 - Index compilation: 2024–2025 - Data integration: 2025–2026 - Finalization of production system and publication of the Scanner Data methodology: 2026 - Scanner Data in the published CPI: January 2027 - A flowchart of Scanner Data process is referenced (Figure 1). ### B. Transmission and Pre-processing Findings: - NBS market analysis identified five retail chains representing together more than 50 percent of total sales in the food retail sector. - COICOP Division 01 Food and non-alcoholic beverages weight in the CPI basket: 35 percent in 2023. - Theoretically, Scanner Data from the five retailers could represent around 18 percent of the CPI basket for COICOP Division 01. - Practical rollout suggested: start with food products from one or two retail chains, then extend to non-food and other chains. - Initial contacts with major retailers for direct Scanner Data access were not successful; retailers reluctant to share detailed data and already provide similar information to STS. - STS introduced an electronic sales monitoring system in July 2023 recording each receipt line: text item description, number of items purchased, VAT rate, and price. - A test data set from STS covered receipts for two major retailers during the first six days in September 2023; mission examined only the first retailer. - Raw test data: over one million lines (each a receipt); transformed table: over five million lines (each a specific item purchase on a receipt). - Data lacks a formal product identifier (barcode/shop code); item descriptions are free text. - Technical feasibility confirmed to create an item code from the text string (same text string → same code); validity must be monitored. - Suggested temporal aggregation: first three weeks of the month to allow processing time. - Suggested grouping: all outlets of a retail chain in a region. - Aggregation formulas suggested (preserve notation): - quantities_i^t = Σ_r q_{i,r} - sales_i^t = Σ_r p_{i,r} q_{i,r} - price_i^t = sales_i^t / quantities_i^t - Pilot data contained around 19,000 items (food and non-food), a plausible assortment size for a retailer. - Need to discuss regular and secure transmission protocol with STS and consider further pre-processing by STS. - A formal memorandum of understanding between NBS and STS is recommended to define: contents and characteristics of the data set, transmission rules (frequency, technical characteristics), and other provisions. - IT infrastructure required at NBS for storing and transforming large files; target output table variables by region and item: text string describing the item; an item code; total sales; total quantities; the price; the region where the item was sold; the date. Recommended Actions: - Develop a memorandum of understanding with the STS to arrange a regular and secure transmission of the data. - Develop an IT infrastructure at the NBS for storing and pre-processing the STS data. - Use the text string to identify an item and calculate average prices at the level of an item, retailer and region. ### C. Classification Findings: - Scanner Data items must be mapped to the CPI classification; Moldavian CPI uses 5-digit ECOICOP (European variant of COICOP) with an additional 6-digit level in the basket; representative products are selected below the 6-digit level. - Mapping errors cause misclassification and incorrect price indices; in principle all Scanner Data items should be classified. - Mission recommendation: map items to the 6-digit level of COICOP (not to representative products). - Reference table requirements: - Variables: item description; item identifier; 6-digit COICOP code. - No duplicates; each item mapped to a single COICOP code. - Initialization: map all items in the dataset at once. - Recurrent monthly procedure: for each item in a month: - If item already in reference table: do nothing. - If item not included: assign 6-digit COICOP and update reference table. - Once created, reference table is augmented each month and no records should be removed. - Classification methods: - Manual labelling: feasible using item descriptions and e.g., Excel, but resource intensive; treat as last resort. - Attribute-based classification: requires auxiliary classification already labelling items; map auxiliary categories to COICOP to automatically assign COICOP categories; performance depends on mapping quality and may need manual follow-up. - Additional metadata (GTIN, retailer-specific classification, item weight) will facilitate COICOP mapping and package-size adjustments. - NBS should request metadata from retailers; mission examined a retailer web shop whose assortment classification is relatively well suited to map to COICOP. - Additional metadata should be linked to the text descriptions provided to STS. Recommended Actions (implied from findings): - Map Scanner Data items to the 6-digit COICOP level and store mappings in a non-duplicating reference table. - Use attribute-based classification when auxiliary classifications or metadata (e.g., GTIN, retailer classification) are available; rely on manual labelling only when necessary. - Request additional item metadata from retailers and link it to STS text descriptions to support automated classification. ### D. Index Compilation and Data Integration - (Section D is referenced in the table of contents and in report structure; detailed technical guidance on index compilation and data integration is included elsewhere in the Section I report. The mission noted that Scanner Data allows weighted index formulas using item-level quantities and sales and suggested processing Scanner Data separately using superlative price index formulas such as the preferred Fisher price index. The mission developed an updated aggregation structure introducing a first-stage distinction between prices from Scanner Data and field-collected prices; outlets not covered by Scanner Data should continue field collection and processing under current methodology.) Key implementation points from earlier sections relevant to index compilation and integration: - Derive average item prices from aggregated sales and quantities (price_i^t = sales_i^t / quantities_i^t). - Process Scanner Data prices separately, potentially using superlative index formulas (Fisher). - Introduce an aggregation structure that distinguishes at the lowest hierarchy level between Scanner Data and field-collected prices. - Plan timeline: index compilation development 2024–2025; data integration 2025–2026; finalize production system and methodology publication in 2026; incorporate Scanner Data in published CPI by January 2027. ### E. Owner-Occupied Housing (OOH) Findings and recommendations: - CPI for housing rents can be improved by expanding coverage. - NBS will start collecting rents for houses located in Chisinau from the same real estate listing websites used for apartment rents. - Accuracy of the rent price index can be improved by developing explicit stratification by region and type of dwelling. - An improved rental price index can be the basis for including OOH in the CPI using the imputed rentals approach. - Given the small rental market in Moldova, further analysis and user consultations are needed to assess whether the imputed rentals approach is a viable option. ### F. Officials Met During the Mission - (Section listed in table of contents; specific names and positions are contained in the source document's Officials Met section.) Summary of Mission Outcomes and Priority Recommendations (selected items preserved): - 1: TA mission assisted NBS with introducing alternative data sources for CPI and examined treatment of OOH. - 2: Incorporating Scanner Data will improve CPI quality: regional, product and temporal coverage; availability of price and quantity at item level enables more accurate indices. - 3: STS Scanner Data assessed as suitable with some transformation; STS electronic sales monitoring records each receipt line; sample examined covered one week of receipts for a major retail chain. - 4: IT infrastructure and pre-processing required; raw STS data large and requires item identification (no formal item identifier in raw data); NBS should request GTIN or retailer-specific classification metadata from retailers. - 5: Scanner Data should be incorporated into the published CPI by the end of 2026; updated aggregation structure developed; prices from Scanner Data could be processed separately using superlative price index formulas (Fisher). - 6: Improve CPI housing rents coverage and stratification; further work needed to assess OOH inclusion via imputed rentals. - 7: Priority recommendations table (TABLE 1) target dates and responsibilities: - March 2024: Establish a memorandum of understanding with the STS to secure regular access to Scanner Data — Responsible: NBS - July 2024: Assign COICOP categories to the items included in the Scanner Data — Responsible: NBS - September 2024: Improve higher-level aggregation in the CPI software — Responsible: NBS Final recommended near-term actions (from report): - Establish memorandum of understanding with STS for regular access to Scanner Data. - Develop IT infrastructure for storage/pre-processing at NBS. - Assign COICOP categories to Scanner Data items and maintain a non-duplicating reference table. - Improve CPI software aggregation and prepare for staged integration of Scanner Data into indices by January 2027. *Source: Section I. Detailed Technical Assessment and Recommendations — IMF Technical Assistance Report – Moldova Consumer Price Index (content unit: tarea2024023).* ### 27. At a later stage, the classification process could be further optimized by using Machine ### 27. At a later stage, the classification process could be further optimized by using Machine ### Classification and machine learning - Machine Learning methods can create a classifier that learns from already classified items; once estimated, new incoming items can be automatically classified. - A fully classified training data set must be available to design a classifier and evaluate its performance. - Recommended Actions: - Reach out to the retailers to access additional metadata of the items in view of applying attribute-based classification. - Classify already available items to the 6-digit COICOP and put in place a regular process for updating the reference table. ### Transition to COICOP 2018 - The NBS will change to the COICOP 2018. - The correspondence between COICOP 2018 and the COICOP version currently used in the Moldavian CPI is less problematic for food products (the main scope of Scanner Data). - An item mapped to the current COICOP version will have to be re-assigned to the relevant COICOP 2018 category. ### Index compilation and data integration - Scanner Data applications: - Can serve as a sampling frame for selecting representative products; review/specification of representative products is conducted at year-end to update the next year’s basket. - Can identify the most popular products to benefit the NBS relatively quickly. - Price index compilation from pre-processed and classified Scanner Data: - Aim to use all items (not only a sample) and weight each item according to its importance. - Index formulas illustrated: Jevons, Laspeyres, Paasche and Fisher. The Fisher price index is one of the preferred index number formulas. - In Scanner Data contexts, bilateral Fisher price indices are typically embedded into a multilateral method. - Quality/weight changes and weight-adjustment techniques: - Matching item codes will not capture pure price changes linked to changes in the weight of an item (e.g., 200g to 180g with unchanged price should record a price increase). - Techniques require connecting “same” items with different weights and adjusting based on weight; text string extraction was explored to access item weight, but preferred option is to obtain weight directly from retailers. - Integration with CPI: - Scanner Data will not cover all household purchases; field price collection in shops not covered by Scanner Data cannot be stopped. - Best practice: compile separate price indices with Scanner Data and combine them with field price collection indices using an appropriate aggregation methodology. - Current calculation is at COICOP 6-digit category broken down by region. A target aggregation structure was developed to integrate Scanner Data into the Moldavian CPI. This aggregation structure is for internal NBS calculations only and does not impact published dissemination. - Target Aggregation Structure (Index structure — Weights – Data source — Index method): - COICOP 5-digit — HBS — Laspeyres-type - └ COICOP 6-digit — HBS — Laspeyres-type - └ Region — HBS — Laspeyres-type - └ Field price collection — Enterprise statistics (turnover) — Laspeyres-type - └ Repres. Prod. — No weights — Jevons - └ Scanner Data — Retailer Enterprise statistics (turnover) — Laspeyres-type - └ Item*Region*Retailer — Scanner data — Multilateral (Fisher) - Progressive introduction of Scanner Data: - With annual weight updates, aggregation structure can be updated each year, adding new retailers or more COICOP 6-digit products covered by Scanner Data. - When integrating Scanner Data from a retailer into the CPI, prices of that retailer must be removed from the field price collection index. - IT architecture and software: - CPI software should be adjusted to integrate Scanner Data. - Possible architecture: separate system for processing Scanner Data (e.g., R software) producing indices at retailer-region-COICOP6 level; these indices injected into main CPI software for combination with field price indices (illustrated in a flowchart in Figure 1). - Current CPI software limitations: combines short-term (current month vs previous month) indices from basic level to higher levels using geometric mean at basic level and fixed weights introduced in January each year. - Technical note: weights would have to be price-updated before combining short-term indices, or short-term indices can be chained at the basic level before aggregation to higher levels. - NBS plans to improve CPI software and review aggregation mechanisms; mission prepared examples (see Appendix 1) to illustrate aggregation methods for IT specifications. - Other uses of Scanner Data: - NBS compiles and publishes average prices for a list of socially important retail products; Scanner Data could be used to select specific items meeting product specifications and insert “Scanner Data average price” into calculations. - Recommended Actions: - Use Scanner Data for reviewing the basket of representative products. - Improve the aggregation methods in the CPI software. ### Owner-occupied housing (OOH) - Rental equivalence approach: - Measures change in price of housing services consumed by owner-occupiers by estimating market value of those services (how much owner-occupiers would pay to rent their dwelling). - Price indicator for imputed rents can be sourced from rental price series, weighted to reflect the stock of owner-occupiers. - Easier to implement if appropriate rent data are available, but may be criticized for being imputations and poses challenges if the rental market is small. - An expenditure weight for imputed rentals would be available from national accounts using the user cost method. The HBS also asks owner-occupiers about a hypothetical rent for their dwelling. - Other approaches: - Acquisitions approach: measures changes in prices of dwellings and goods/services acquired by households as owner-occupiers; requires monthly/timely price indices for acquisition/construction of new dwellings. - Payment approach: measures changes in payments made by owner-occupiers (e.g., mortgage interest payments); can introduce circularity in CPI used for monetary policy. - Current practice in Moldova: - CPI includes rentals collected for different apartment types characterized by number of rooms and vintage (new or old); data collection at end of each month. - Rents for Chisinau are collected from popular real estate listing websites; other regions by price collectors. - Rents labelled in euros are transformed into local currency using appropriate exchange rates. - Market characteristics and experimental simulation: - Rental market in Moldova is small. According to the HBS, the share of owner-occupiers in Moldova is estimated to be 95 percent. - Inclusion of OOH under rental equivalence is expected to increase the combined share of rentals and imputed rentals to more than 15 percent. - Mission conducted an experimental simulation including imputed rentals by assuming the imputed rental index equals the current rental index. - Improvements and data expansion: - Price index for rentals should include both apartments and houses; currently only apartments included. - Mission recommended starting to collect prices for houses in Chisinau from real estate listing websites; NBS agreed to collect and include rents for houses in Chisinau from next year on. - Internal weights of the rental price index should be reviewed using HBS expenditure data broken down by region and by type of dwelling (houses and apartments). - HBS collects occupancy type, type of dwelling, region, size; estimated proportions multiplied by an appropriate average price could develop expenditure shares for more detailed stratification. A weighting structure for imputed rental index could be developed similarly. - Opportunity to automatize and expand data collection via web scraping; NBS should reach out to NBM which already scrapes data for the RPPI. Similar approach may collect rental data and dwelling characteristics to expand coverage and improve quality adjustment. - Feasibility and further analysis: - OOH inclusion via rental equivalence requires further investigation given concentration of rental market in Chisinau and small rental markets elsewhere; majority of dwellings are houses with no rent data yet collected. - Further analysis and user consultations needed to assess feasibility of imputed rentals approach. - Recommended Actions: - Expand the coverage of the rent price index and include rents for houses in the CPI. - Start discussions with the NBM on web scraping rents from real estate listing websites. ### Work plan milestones (October 2023–October 2024) - Result: Transaction data is regularly transmitted from the STS to the NBS. - High — Establish a memorandum of understanding with the STS to secure regular access to Scanner Data. — March 2024 - High — Develop an IT infrastructure at the NBS for storing and pre-processing the STS data. — March 2024 - Result: Scanner Data is classified to COICOP. - High — Reach out to the retailers to access metadata on the products in view of applying attribute-based classification. — January 2024 - High — Classify already available items in the Scanner Data to the 6-digit COICOP. — July 2024 - Result: The CPI software is improved. - High — Improve the aggregation methods in the CPI software. — September 2024 - Result: Scanner Data is used. - Medium — Use Scanner Data for reviewing the basket of representative products. — October 2024 - Result: The rental price index is improved. - Medium — Update the basket and include rents for houses in the CPI. — January 2024 - Medium — Start discussions with the NBM on web scraping prices from the real estate listing websites. — July 2024 ### Appendix I — Higher-level aggregation example (regional to national) - Example regions and weights: - Chisinau — 25 - North — 40 - South — 25 - Centre — 55 - TOTAL — 145 - Previous-month (short-term) indices for a given 6-digit COICOP category (regional): Dec, Jan, Feb, March, April shown in example tables. - Chaining regional indices then combining using fixed weights yields national (long-term) indices: - January Index = (104.17*25+103.33*40+111.11*25+106.67*55) / 145= 106.08 - February Index = (116.67*25+114.00*40+112.35*25+127.78*55) / 145 = 119.40 - March Index = (121.67*25+116.00*40+114.57*25+136.67*55) / 145 = 124.57 - April Index = (100.00*25+100.00*40+100.00*25+100.00*55) / 145 = 100.00 - Alternative method (update weights first, then combine short-term indices) produces identical long-term national indices: - Updated weights example totals: Total 145.00, 153.82, 173.13, 180.63, 145.00 for Dec–April. - Short-term national indices using updated weights: - January Index = (104.17*25+103.33*40+111.11*25+106.67*55) / 145.00 = 106.08 - February Index = (112.00*26.04+110.32*41.33+101.11*27.78+119.79*58.67) / 153.82 = 112.55 - March Index = (104.29*29.17+101.75*45.60+101.98*28.09+106.96*70.28) / 173.13 = 104.33 - April Index = (82.19*30.42+86.21*46.40+87.28*28.64+73.17*75.17) / 180.63 = 80.28 - Long-term national indices derived: - January Index = 106.08 - February Index = 106.08*112.55/100.00 = 119.40 - March Index = 104.33 *119.40/100.00 = 124.57 - April Index = 80.28 * 124.57/100.00 = 100.00 - The same procedure applies for combining 6-digit COICOP indices to obtain higher-level aggregates. *IMF | Technical Assistance Report – Moldova Consumer Price Index* --- _Source: https://www.imf.org/-/media/files/publications/tar/2024/english/tarea2024023.pdf_