Description

Using the universe of electronic invoices in Chile, we develop a methodology to classify firms’ transactions into product categories based on a machine learning algo-rithm trained on a labeled corpus of product descriptions. The resulting classification allows us to construct measures of machinery and equipment outlays and goods con-sumption that closely track their national accounts counterparts, while also tracing the effects of shocks on consumption and their implications for consumer price index mea-surement. Using the COVID-19 pandemic as a case study, we document a reallocation toward durable goods, show that mobility restrictions and income support policies do not fully account for these shifts, and quantify the implications of expenditure reallocation for measured inflation.