Exchange Rate Pass-Through to Consumer Prices in South Africa: Evidence from Micro-Data

Abstract

A sizeable literature examines exchange rate pass-through to disaggregated import prices, but few micro-studies focus on consumer prices. This article explores exchange rate pass-through to consumer prices in South Africa, for 2002–2007, using a unique data set of highly disaggregated data at the product and outlet level. The empirical approach allows pass-through to be calculated over various horizons for different goods and services. The heterogeneity of pass-through for food sub-components is considerable. Switches between import and export parity pricing of maize are found significant for five out of ten food sub-components. Using actual weights from the CPI basket, overall pass-through to the almost 63 per cent of the CPI covered is about 30 per cent after two years, and higher for food.

Notes

1. Aron, Creamer, et al. (2013) surveys the literature. Of the handful of studies that directly explore pass-through to the sub-components of the CPI, Belaish (2003) examines four CPI components for Brazil; Leung (2003) examines sub-components of Canada’s CPI; Soffer (2006) investigates 31 sub-components of Israel’s CPI; and Rigobon (2007) examines approximately 5,000 items underlying the monthly CPI in SA. Parsley (2012) examines 158 annual consumer prices for SA compiled by the Economist Intelligence Unit (EIU).

2. Rigobon (2007) assumes equal weights for South Africa’s consumer prices; so does Parsley (2012), for the set of goods prices and the set of services prices from the EIU.

3. In South Africa, the exchange rate is defined as foreign currency relative to domestic currency, and increases with appreciation.

4. Systems methods, potentially available in longer samples than the one considered in this article, allow the exchange rate and domestic costs to be endogenous, and hence introduce a role for feedback effects. The resulting measures for pass-through from the two methods will be different.

5. We have also experimented with other specifications that include error correction terms. Given the short time-span of our data, there does not seem to be enough long-run information to estimate these well. Inclusion of these terms does not significantly alter the results, however.

6. An obvious extension is to allow for fixed effects. A fixed effects estimator would allow the constant to vary by product and store (i and k) or product, unit and store (i, j and k). Owing to the time-consuming nature of the estimation process, a full set of results has not been produced controlling for these. Exploratory regressions in some CPI components suggest that the results do not differ fundamentally from those reported here.

7. These are the underlying weights, based on the 2000 Income and Expenditure Survey, which were used to construct the CPI basket from January 2002. In 2009, a new CPI basket was introduced based on weights from the 2005–2006 Survey.

8. Various capture codes are used in collecting the data; we excluded capture codes reflecting out of stock goods and incomparable goods, due to changes in quality.

9. Given that the study is based on six-month price changes, prices collected quarterly, annually or at other non-monthly intervals have been excluded. A comparable approach is adopted in other micro-data studies (see Alvarez and Hernando, 2006).

10. These are the weights in the CPI for South Africa as a whole, as opposed to the weights for ‘Metropolitan areas’ and ‘Metropolitan and other urban areas’ as published by Statistics South Africa.

11. Housing costs were excluded due the fact that certain housing sub-sector price information – including the rents of housing, flats and townhouses – is based on a frequently updated price index rather than on actual pricing conduct. Hence, there is a tendency for small but frequent changes in price. This is as a result of a practice by the statistical authorities during the period under review, but subsequently discontinued, that any observed price change over a quarter was distributed over the three months of the following quarter

12. For the clothing and footwear component, we impose a coefficient of zero on contemporaneous changes in the exchange rate, since the point estimate was positive. Lags of the exchange rate generally have negative signs, but the standard errors are relatively large.

13. The beverages and tobacco component specification includes the change in US PPI for processed foods and feeds as a foreign price control. For beverages and tobacco we impose a coefficient of zero on the current exchange rate term whose point-estimate is positive.

14. We impose a coefficient of zero on contemporaneous changes in the exchange rate.

15. For the transport component we include the current and six-month lagged Brent oil price change to proxy foreign prices.

16. The specification for recreation includes a US price index for computer parts and accessories, since computers form part of this CPI component. However, the aggregated standard error on this is relatively large.

17. Estimations for the furniture sub-component include South African unit labour costs.

18. For the food component in Table 3, foreign prices are proxied with lags in US maize prices up to 12 months, and the dummy for white maize and unit labour costs are included.

19. Results are available on request from the authors.

20. The only foreign price used for fruit was that of bananas, however.

21. As noted above, however, the sample length is not sufficient for robust estimates.

22. This component contains sauces, condiments and baby food, likely to be mainly domestically produced.

23. The overall ERPT of 21 per cent shown in the penultimate column of Table 4 corresponds to these less preferred estimates.