Does Central Bank transparency impact financial markets? A cross-country econometric analysis

Southern Economic Journal, Jan, 2007 by Marc Tomljanovich

Table 3 reports the estimation of Equation 5 using the TARCH(1,1) model. The qualitative results are roughly the same for all countries. However, use of the TARCH model has made rejection of the null hypothesis more likely for all countries. For example, in the United Kingdom, the null hypothesis is rejected for 6 (9) of the bond maturity pairs in the pre-break (post-break) sample. Estimates of the GARCH and ARCH terms are roughly the same for all countries across most bond maturity pairs, indicating that the volatility process itself was largely unaffected by the central banking procedural shifts. (16) For all countries, the asymmetry term [delta] is statistically insignificant, suggesting that good and bad market information have identical effects on bond market volatility. Finally, it is worth noting the estimated conditional variance has consistently dropped across all countries for almost all bond maturity pairs.

As a whole, the results from Table 3 support Tabellini (1987), and appear inconsistent with the theoretical arguments made by Dotsey (1987) and Rudin (1988) that increased transparency makes interest rates more unstable. This point is underscored by considering that the results for New Zealand are not markedly different than for the central banks that underwent a transparency shift, while for Germany increased instability actually appears to have transpired in the latter part of the time span. (17) However, these findings do not prove that transparency was necessarily the cause of improved efficiency of markets, as for the United Kingdom the move to inflation targeting and for the case of Japan the move to central bank independence occurred simultaneously.

To understand further the impact of transparency on bond markets, we conduct Chow Breakpoint tests to ascertain whether a structural break occurred within our data span. First, assuming the known break dates reported in Table 2 for each country, the month that central bank policies became more transparent, we find that the null hypothesis of no structural change is rejected at the 5% significance level for most bond maturity pairs. (18) Second, we let the data choose the break date, analogous to Andrews (1993), under the premise that events other than increased central bank transparency may have changed the predictability of financial markets during our span.

We estimate Equation 5 for break dates in the range [T.sub.b.sup.*], [T.sub.b.sup.*] 1, ..., T - [T.sub.b.sup.*], where [T.sub.b.sup.*] = 0.1 T. Therefore, we do not consider break dates near the end points of the sample. This is called trimming. We choose the break date maximizing the Wald Statistic that tests the null hypothesis of no structural change. Results can be found in Table 4. We find that for almost every country, the break dates are relatively close to the transparency break dates chosen in this analysis. The exceptions seem to be Australia, in which the estimated break dates occur around 1992 (as opposed to a hypothesized break date of 1996:8), and Japan, in which the estimated break dates vary substantially across bond maturity pairs between 1992:9 and 1995:8 (as opposed to a hypothesized break date of 1998:4).