Does monetary policy react asymmetrically to exchange rate misalignments? Evidence for South Africa

Clarida et al. (2002) also arrive to the same result in a general equilibrium two-country model, see also Galí and Monacelli (2005).

Along these lines, Proaño (2013) shows in a theoretical model of a small open economy characterized by heterogenous, boundedly rational foreign exchange expectations that PPI targeting is superior to CPI targeting when the nominal exchange rate is driven by trend-extrapolating exchange rate forecasting rules.

In de Mello et al. (2013), the transition variable is the expected inflation gap. They also find policy rate responses to exchange rates for Colombia which are positive and statistically significant in the linear part of the model and negative responses in the nonlinear part model.

Clarida (2001) argues that monetary policy with objectives of stabilizing inflation and the output gap may also be consistent with interest rate responses to an exchange rate when it depreciates and/or when an exchange rate is misaligned even if a central bank does not target the exchange rate.

Clarida et al. (1998) provide reasons why a central bank may smooth interest rates changes, and Clarida et al. (2000) discuss various reasons why purely using the target rate may be too restrictive to describe actual changes in the short-term rate.

We use CPI because the SARB currently targets consumer price inflation using an inflation target range (band) between 3 and 6% for year-on-year inflation.

The composite coincident business cycle indicator is constructed at the monthly frequency by integrating various individual economic time series into a single indicator time series that reflects the movement of the turning points in the business cycle. The time series included in the business cycle coincident indicator displays various aspects of economic activity. This measure is considered to be the best proxy for output for South Africa at the monthly frequency (Naraidoo and Raputsoane 2015). See also the SARB for details about the composite—coincident, lagging and leading—business cycle indicators.

The commodity price series is discontinued from 2008M01. Thus, from 2008M01 onward, we augment these data with a producer price index that accounts among other things for agricultural products, minerals and coal and petroleum products.

The SARB has had numerous monetary policy frameworks which have included adjusting interest rates to achieve their targets, and the repo rate has been regarded as the main monetary policy instrument since 1998. Further, the repo rate has been the rate adjusted by the SARB when adjusting monetary policy since the adoption of inflation targeting on February 2000. Thus, over our total sample period and subsample period, we only use the repo rate as the monetary policy instrument because it is specific with a particular period and it also allows us to use a consistent data series relating to a policy instrument that covers our sample periods. Refer to Baaziz et al. (2013) for more details about the history of the SARB.

More specifically, using the DF-GLS and Ng–Perron tests with the standard AIC and SIC criterion rejects the null of a unit root in annual inflation. Using a modified AIC and SIC provides less evidence toward failing to reject the null of a unit root. However, along with the Ng–Perron MZt test, the MZa, MSB and MPT lean more toward rejecting the null of a unit root in annual inflation. All unit root tests are reported in “Appendix”. Clarida et al. (1998) outline that in their analysis, their econometric approach depends on the assumption that the short-term interest rate and inflation are I(0). However, they note that for the USA and Japan, there is less evidence supporting that short-term interest rates are I(0). One of the problems that they identify is that the Dickey–Fuller test has low power (especially in the context of short samples) and hence they continue their estimation using US and Japan short-term interest rates. In a related manner, Clarida et al. (2000) note that their empirical analysis is based on the assumption that both inflation and the nominal interest rate are stationary. However, they point out that the persistence of both series generally makes it hard to reject the null of a unit root at conventional significance levels. Another factor noted by Clarida et al. (2000) with respect to the stationarity assumption is that many theoretical models that employ a forward-looking monetary policy reaction function feature inflation and nominal interest rates which are stationary.

For estimation purposes and for all the policy rules, the exchange rate enters with a one period lag to avoid simultaneity problems and also to account for the timing convention associated with the real exchange rate being observable for policy authorities when making interest rate decisions. See also Clarida et al. (1998) and Castro (2011).

The set of instruments for the extended model are the same as the instruments used in the baseline model; however, different lag structures are used for the instruments for different policy rule specifications wherever necessary. For the subsample estimates in Table 2, we include a dummy variable for the period 1998:01–1998:12 because of the Asian financial crisis and the associated effects on the South African economy, especially on the repo rate, annual inflation and the exchange rate. Further, we also include a dummy variable for the period 2003:01–2003:12 which captures a rapid reduction in annual inflation by over 10.5 percentage points over the period and also a substantial 550 basis points reduction in the policy rate over the period 2003:01–2003:12.

Svensson (1997, 2002) notes that inflation targeting is an inflation forecast targeting procedure. Thus, inflation and output forecasts are continuously revised on the basis of past and recently available information. As exchange rate depreciations can lead to inflationary pressures and make a conditional inflation forecast higher than its target, this can result in an upward policy rate adjustment to make an inflation forecast consistent with its target over a particular forecast horizon.

South Africa does not have official data on an industrial production index; however, a total mining (seasonally adjusted) and total manufacturing index (seasonally adjusted) are available from the IMF’s IFS and also from Statistics South Africa. Thus, we construct a proxy for the industrial production index with an equally weighted linear combination of manufacturing production index and mining production index. Using our proxy for an industrial production index, we also find consistent results over the total sample and in most of the specifications over the subsample.

Annual output growth in this data refers to annual growth using a composite coincident business cycle indicator. Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index and also with annual quarterly real GDP growth (seasonally adjusted) over 1984:Q2–1987:Q2.

Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index over May 1995–April 1996 and also with annual quarterly real GDP growth (seasonally adjusted) over 1995:Q1–1996:Q1.

Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index over April 1997–October 1999 and also with annual quarterly real GDP growth (seasonally adjusted) over 1997:Q1–1999:Q1.

Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index and also with annual quarterly real GDP growth (seasonally adjusted) over 2001:Q1–2001:Q4.

Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index and also with annual quarterly real GDP growth (seasonally adjusted) over 2008:Q1–2009:Q3.

Similar dynamics are exhibited when examining annual output growth with our constructed industrial production index over August 2013–January 2015 and also with annual quarterly real GDP growth (seasonally adjusted) over 2013:Q3–2014:Q2.