Introduction of Scanner Data into Austrian CPI and HICP – practical implementation experience, with a focus on window length options

After several years of preparation and a two-year transition period, scanner data have been introduced into the Austrian CPI and HICP in January 2022. A significant factor was the amendment of the Austrian national CPI-Regulation in December 2019, which since then regulates the scanner data requirements and ensures the weekly scanner data deliveries by most important retailers, initially by the grocery and drugstore retail trade (NACE classes 47.11 and 47.75). During the implementation of the project, pragmatic decisions had to be taken on a number of issues ranging from the way to establish a good relationship with data providers through the method of data access, to the classification of products, and the choice of the appropriate index calculation and aggregation method. One small, but not insignificant subset of these decisions, is the time window length chosen when adopting a multilateral approach, i.e. based on how many consecutive months of data the index is compiled. Although a two-year transition period in which traditionally collected price data and scanner data can be compared seems to be comfortably long, it is too short to test the commonly used window length of 25 months. That is why Statistics Austria introduced scanner data into production with a 13-month window length. After an extra year, however, we started to study the benefits of possibly more precise data resulting from a longer window length at the overall index level and at lower aggregation levels. We also assessed the additional resource use (computational capacity) that would be required to move from a 13-month window to a 25-month window. On this basis, we have carried out a cost-benefit analysis to determine whether it is more reasonable to choose a shorter or longer window length. On the whole it seems that in most cases the 13-month window length provides similarly good data quality as a 25- month window and saves plenty of resources, however there are specific conditions (e.g. seasonality) in which a longer window length has a positive impact on data quality.