Abstract
This study examines the asymmetry of the loss function for private forecasters in exchange rate forecasts of the South African rand. It tests rationality under the possibility of an asymmetric loss function. The results indicate less evidence of asymmetry for a horizon of 1 month but considerable evidence of asymmetry for a horizon of 3 months. However, the shapes of the distributions formed by estimated asymmetry parameters of sub-samples for each forecaster are symmetric, regardless of the forecast horizons, which implies that these forecasters do not herd or antiherd. In fact, the results of our empirical herding test show that forecasters neither herd nor antiherd, which is in sharp contrast to recent findings on antiherding for foreign exchange rates in emerging market economies. Our findings provide consistent evidence for a recent suggestion that antiherding might result in the rejection of rationality, even under asymmetric loss functions. Our findings also suggest that central bank transparency might be associated with herding behaviours.
Acknowledgement
We are grateful to two anonymous referees for helpful comments and suggestions. All remaining errors are our own.
Notes
1 The sample is, in fact, an unbalanced panel between July 2001 and October 2009.
2 To examine the consensus as an individual forecast series, we would need to assume that individual forecasters have an identical loss function, although the assumption is unlikely.
3 Unlike some previous studies, the forecasters are not classified as a bank, security firm, insurance company and so on; these days, it is difficult to distinguish forecasters clearly from among these industries. Rather, this article classifies those forecasters in financial institutions as commercial and investment banks.
4 Although not reported in this article, results of the lin–lin loss function are consistent with the main findings.
5 A constant is included as an instrument in all models considered in this study. However, for notational convenience, it is dropped hereafter.
6 In previous studies, the lagged forecast is used extensively as an instrument. Because the sample in this study is not continuous, this instrument cannot be used.
7 Monetary fundamentals (MF) are defined as MF = Δm+Δy + 0.1Δr, where Δm is the log differential of money supply, Δy is the differential in industrial production and Δr is the differential in interest rate. The difference is between South Africa and the US. See Flood and Rose (Citation1995) for details. Note that actual monthly series for the US is from the Federal Reserve System.
8 The interest rates are the repo rate for F1 and Treasury bills (91 days) for F2. The yield spreads are calculated as the difference of each interest rate from the benchmark overnight rate. In addition, we examined the lagged change in the nominal effective rate, the lagged change in the real effective rate and the forward rates, and obtained consistent results, although they are not reported in this article to conserve space.
9 There are a few remarks on SEs of the estimates although they are not reported to conserve spaces. The SEs are typically small and vary mostly between 0.02 and 0.12. This is broadly consistent with the estimates by institutional affiliation presented later.
10 The variables for actual and forecast are defined as the difference from the lagged actual to conduct the F-tests properly. The SEs are estimated using the approach of Newey and West (Citation1987) because the error term is likely to be serially correlated.
11 Although not reported in this article, J-tests with values of asymmetry parameter restricted to 0.5 yield consistent results.
12 See Bernhardt et al. (Citation2006) for more details.
13 Note that failure of rejection may be attributed to the low power of the J-test. Therefore, like the previous studies, our results do not necessarily provide strong evidence of rationality.
14 Indices range from 0 to 15, and the higher those indices, the greater transparency is.