Dynamic Effects of Monetary Policy Shocks in Malawi* - African ...

Selection of the optimal lag length is guided by established criteria 1 . Akaike and Hannan-QuinnInformation Criteria (AIC and HIC, respectively) suggest a lag length of order 42 whileSchwartz Information Criterion (SIC) suggests a lag length of order two. The problem with theformer is that it uses up all degrees of freedom and renders the estimated VAR unstable with 83inverse roots of the characteristic autoregressive (AR) polynomial lying outside the unit circle.The latter, on the other hand, performs well in a robustness check and it is adopted. At thechosen lag length (of order two), all the eight inverse roots of the characteristic AR polynomialhave modulus less than one and lie inside the unit circle, indicating that the estimated VAR isstationary or stable (see Table C1 in Appendix C). A VAR lag exclusion Wald test also revealsthat all endogenous variables in the model are jointly significant at each lag length for allequations collectively. Separately, all equations are significant at lag length of order one while atlag length of order two, only consumer price and output equations are insignificant (see Table C2in Appendix C). Since the SIC imposes a harsher penalty for adding more regressors than AIC,the criterion is deemed more appropriate for determining lag length in a VAR.Departures from established criteria for the determination of optimal lag length are notuncommon in the literature. Piffanelli (2001) argued that 12 lags are appropriate with monthlydata to avoid seasonality problems. She, however, included six lags in a study of Germanywithout rationalisation. Becklelmans (2005) used a lag length of order three in a study ofAustralia „because it provides reasonable dynamics without shortening the estimation sample toomuch.‟ Vonnak (2005) used three lags in a study of Hungary „because they were enough toproduce uncorrelated residuals, based on the evidence of the multivariate-LM test.‟ Usingquarterly data in a study of Thailand, Disyatat and Vongsinsirikul (2003) settled for two lagswhile the optimal lag length under various criteria appeared to be one because they felt onequarter is too short a period to capture the dynamics of a system. This study, however, finds tojustification to depart from the established criteria.Before making inferences on the structural shocks in the model, we analyse correlations betweenmovements in the bank rate and reserve money and their corresponding recovered structuralshocks to ascertain if the monetary policy shocks are reasonable. Figures 4.1 and 4.2 presentplots of recovered bank rate and reserve money structural innovations against month-on-monthgrowth rates of the bank rate and the logarithm of reserve money, respectively, with therecovered structural shocks plotted on the primary vertical axis and the monetary policyoperating targets on the secondary vertical axis. Positive structural innovations of the bank rateand negative structural innovations of reserve money are associated with monetary policytightening while negative structural innovations of the bank rate and positive structuralinnovations of reserve money are associated with monetary policy loosening.The charts reveal that there is some correlation in the movements of the monetary policyoperating targets and their respective recovered innovations. The correlations are, however, morepronounced between the bank rate and its recovered structural shocks compared to reservemoney and its recovered structural shocks. Reliability of the structural shocks is also ascertainedby assessing the efficiency of the structural coefficients estimated in the SVAR. Table C3 in1 This approach is applied in all subsequent models.15 | P a g e