The weights used in the consumer price index (CPI) may not accurately reflect current consumption patterns as the current CPI weights are based on pre-pandemic expenditure survey information. The use of fixed expenditure weights in CPI, may for example, result in understatement of inflation during periods of relative price changes encourage substitution towards cheaper products. Any potential bias in CPI measurement would have implications for the optimal inflation target for an inflation-targeting central bank as it can affect the credibility of the central bank by creating a divergence of the experience of household and firms and official statistics. We showed in an earlier post, for example, that low-income individuals (decile 1) have typically experienced higher inflation—about 2 percentage points more – than the wealthiest group (decile 10) since the pandemic.
We also previously showed that during the COVID-19 pandemic, CPI inflation, which tracks price changes for a fixed basket of consumer goods and services including imports, was substantially below the GDP deflator, which measures inflation for all domestically produced goods and services in a given year.
To estimate possible bias in the measurement of CPI, the chart below calculates headline inflation from underlying 8 digit categories using the 2008 expenditure weights and the 2016 CPI weights (the last set of weights based on expenditure survey results). The estimated bias has been positive, on average, since 2008. In part, this reflects the high weight on food inflation, which at the margin CPI may be slightly overstating costs of living in South Africa. We will update the analysis for the updated weights from the latest 2023 income and expenditure survey once updated 8 digit CPI indices are published later this month. This will allow us to assess whether historical CPI measures have been consistent with underlying economic reality in South Africa.
Note that historical CPI weights back to 1970 are available on our EconData platform to enterprise subscribers. As far as we are aware, EconData is only source of these historical weights for easy download in excel, R or Python.
Compiled by Lisa Martin
