2. Modelling Monetary Policy in Namibia: A Structural VAR estimation in the new consensus
2.6 Structural Forecast Errors Variance Decomposition (SFEVD) Analysis
Our results in the last section point to the empirical evidence of an effective monetary transmission mechanism through interest rate and credit channel in Namibia. In this section we analyse the relative importance of monetary policy and credit structural shocks at different horizons. For the sake of space we discussed the SFEVD for domestic repo rate and credit shocks. From the monetary policy statement issued by BoN it is clear that these two channels are always closely monitored for the effectiveness of monetary policy actions. The rest of the results for SFEVD attributed to other shocks are given in the Appendix. Tables 2.1 and 2.2 give the SFEVD, which describe the variation attributed to domestic monetary policy (`Ã) and credit (¨J) shocks in equation (2.12). Structural forecast error variance decomposition analysis displays proportions at each point in time as caused by the shock in the variable itself and the variation attributed to others structural shocks in the system. The SFEVD statistic lays bare relevant information about the relative importance of each unanticipated shock as they affect all endogenous variables in the structural model. Hence, we use results from this exercise to compare the relative strength of individual channels of the transmission mechanism in Namibia. This approach allows us to establish the proportions of the variation in output, which is accounted for by credit and monetary policy shocks at different horizons.
In Table 2.1 the first column gives the horizon from period 0 to the eighth quarter. Columns two to six give the SFEVD for commodity price inflation, changes in SA policy rate, QGDP, inflation, private credit growth and changes in domestic repo rate. At each step SFEVD statistics indicate the percentage attributed to monetary policy shock (`Ã) as in Table 2.1 and credit shock (¨J) in Table 2.2. Our main focus is the SFEVD for QGDP, which represents the percentage of variation accounted for by monetary policy shock.98 In column three, the result shows that domestic monetary policy shock accounts for about 5.0% in the second quarter, 7.0% in the fourth quarter, and 8.0% in the seventh quarter after the initial impact. Meanwhile, column four shows that there is a significant influence of domestic monetary policy shock on inflation. Monetary policy shocks account for more
98 The discussion about the SFEVD for CMI and SA repo is not of much interest because they do not form
part of the objective and their impulse response functions to domestic monetary shock are not statistically significant.
than 11.0% of variation in the rate of inflation in the second and fourth quarter after initial impacts. In column six, SFEVD shows that 34.9% of variation in domestic policy rate shock is attributed to itself in the second quarter and less than 20.0% from five to eight quarters ahead. In other results not reported here we observed, SA’s monetary policy shock accounts for more than 2.0% in SFEVD for QGDP over the same horizon. Similarly, Table 2.2 credit shock accounts for less than 1.8% of the variation in QGDP. This evidence indicates that the interest rate channel through Namibia’s repo rate is relatively stronger than the credit channel. This is because the result for SFEVD from the short run SVAR AB-model shows that domestic monetary policy shock (i.e. repo) accounts for more variations in quarterly GDP and inflation in Namibia.
Table 2.1 SFEVD, which describe the variation attributed to domestic monetary policy
(`Ã) shock
Note: (1)∆pO, (1)∆, (3) ∆»kr, (4) Oj¼~, (5) ∆rp and (6) ∆j -i.e. (1) sfevd shows variation in ∆pO attributed to a shock in Namibia repo rate shock from 0 to 8th quarter.
Table 2.2 SFEVD, which describe the variation attributed to private credit (¨J) shock
Note: (1)∆pO, (1)∆, (3) ∆»kr, (4) Oj¼~, (5) ∆rp and (6) ∆j -i.e. (1) sfevd shows variation in ∆pO attributed to a shock in private credit (¨J) shock from 0 to 8th quarter.
Finally, we compare the structural fraction of mean squared errors SFMSE due to monetary shocks as derived from the short run SVAR. Figure 2-5 show the structural fractions of mean squared errors due to domestic and SA monetary policy shocks. These results show that significant
8 .058366 .027867 .079295 .110615 .028687 .147334 7 .058502 .027471 .080094 .111336 .028747 .148205 6 .058837 .027577 .071547 .111108 .028133 .147544 5 .058064 .024765 .065775 .11204 .028189 .148046 4 .057226 .02008 .079942 .114872 .027485 .145864 3 .03499 .014847 .082808 .105897 .024583 .149243 2 .037129 .000083 .050886 .113791 .026839 .152976 1 0 0 0 0 0 .349961 0 0 0 0 0 0 0 step sfevd sfevd sfevd sfevd sfevd sfevd (1) (2) (3) (4) (5) (6) 8 .026306 .048292 .018708 .030724 .740231 .049112 7 .025636 .046964 .01889 .03099 .742482 .046975 6 .021613 .04476 .018909 .029523 .744434 .046299 5 .019995 .045124 .011626 .029764 .756299 .04701 4 .015892 .046427 .012998 .021803 .767231 .048378 3 .003247 .048261 .013178 .020413 .78642 .048693 2 .001655 .049298 .011664 .02213 .852541 .048419 1 0 0 0 0 .97626 .003486 0 0 0 0 0 0 0 step sfevd sfevd sfevd sfevd sfevd sfevd (1) (2) (3) (4) (5) (6)
large fractions in the variation of output are attributed to domestic monetary policy while foreign monetary policy shock only accounts for less than 5% over four quarters. Thus, changing the level of domestic repo rate will result in more significant effects on real economic activity as compared to transmission effects from SA monetary policy rate.
Figure 2-5. Structural Fraction of Mean Squared Errors (SFMSE) due to monetary policy shocks ∆j and ∆.
In all, this evidence from SFEVD and SFMSE show that domestic monetary policy shock repo rate produced consistent significant results regarding monetary policy effects in Namibia. Both structural impulse response functions and forecast error variance decomposition show that repo rate shocks have negative impacts on output, inflation and private credit. Although SA’s monetary policy shocks have significant effects in Namibia the empirical evidence obtained is small relative to effects generated by domestic monetary policy shock.
2.5.3 Robustness Checks
In order to assess the internal validity of our results, we carried out the following three robustness checks. We made three main alternative estimations to the short run SVAR model in (2.12). We estimate the SVAR model with A-model identification restrictions. The main aim of this alternative estimate is to check whether our identification restrictions have shaped the pattern of behaviours portrayed by the structural impulse
0 .05 .1 .15 .2 0 2 4 6 8 0 2 4 6 8
Fraction of MSE in QGDP attributed to domestic mp shock Fraction of MSE in QGDP attributed to SA's shock
90% CI (structural) fraction of mse due to impulse
step
response functions. In the second alternative estimation, we estimated the SVAR AB- model excluding the foreign variables commodity price inflation and SA’s repo rate. We reduced the economic model by trimming the structural representation in (2.12) to three domestic variables: quarterly GDP, inflation and the repo rate. Finally, we present the trimmed model with SA’s policy as the policy instrument. This alteration is necessary to assess the view that Namibia’s policy rate is redundant; therefore, we need to model the transmission mechanism only with SA policy rate as the policy instrument.
In the first alternative specification, we start with exact identification of the three variables SVAR-AB model: QGDP, inflation and domestic monetary policy rate. This makes a lower triangular identification of the type AB-model. Figure B.2-3 present results from this alternative specification, which shows that all SIRFS are statistically significant with evidence of negative impact on inflation while QGDP remains volatile after the second quarter. The impulse responses show that monetary shock produces similar responses on output, inflation and repo rate. Meanwhile, demand shocks produce significant positive responses on inflation and the repo rate (i.e. the monetary policy rate). These structural impulse responses are consistent with results from six variable short run SVARs, therefore the size and order of the system do not significantly influence the responses from QGDP, inflation and policy rate.
Our second robustness check involved trimming the three variables in the SVAR by excluding the commodity price inflation and private credit, and replacing Namibia’s repo rate with SA monetary policy rate. This robustness is aimed to test SA monetary policy effects in the model without monetary policy rate. Impulse responses from this alternative specification show that SA’s monetary policy shock barely produced any statistically significant impulse response functions in quarterly real GDP and inflation. These results show that whether we exclude commodity price inflation and private credit the impacts from SA monetary policy shock are smaller compared to domestic monetary policy shock. Finally, although we used five lags in the structural estimation instead of the three lags suggested by the Akaike Information Criteria (AIC), this switch from four to two lags has not considerably changed the statistical significance of the impulse response functions. Impulse responses to SA’s monetary policy are the same as in the six variables SVAR in equation (2.12). These findings thus indicate that the results from the SVAR models used are robust. Therefore, these results reflect the efficacy and transmission mechanism of monetary policy in Namibia. Finally, these empirical evidences are consistent with stylized facts as found in many studies both from developed and developing countries. In the new
consensus, an increase in monetary policy shock should always lead to a rise in policy rate, lower prices, and reduce real output.