AbstractOfficial statistics measuring the cost of living are known to suffer from several biases. This paper shows that the size of the biases can vary with economic conditions. Using homescan data, it is first confirmed that official price indexes can be tracked using such granular data sets. While the often‐acknowledged substitution bias is shown to be relatively small, neglected preference adjustment and product entry/exit result in a 2.6 percentage point bias in the annual inflation rate on average. Furthermore, the bias is particularly large in the aftermath of a shock to relative prices, increasing to 3.7 percentage points.