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

More documents

Recommendations

Info

APPENDIX BB-1.0. Data PropertiesCointegrated Structural VAR ApproachAnalysis of the cointegrated SVAR is carried out on the truncated sample only (1994:03-2005:12) covering the period when Malawi adopted indirect tools of monetary control. Using theAugmented Dickey-Fuller Test, it is established that all variables are integrated of order 1 at 1percent except for reserve money, which is integrated of order 1 at 5% (see table B1). Thisprovides a basis to test for cointegrating relations, defined as deviations from steady staterelations. In line with the exploratory data analysis carried out in Section 3.4 (Data, Data Sourcesand Measurement of Variables), we allow for each endogenous variable a linear deterministictrend with intercept and trend in the cointegrating equation and no trend in the VAR. A laglength of 3 is chosen in conformity with the standard information criteria (AIC, SIC and HIC)while ensuring stability of the system. At this lag length, the system is stationary or stable withall 21 inverse roots of the characteristic autoregressive polynomial showing modulus less thanone and lying inside the unit circle (see Table B2). Results of the Johansen Cointegration Testreveal that the system has a reduced rank 3 . While a trace suggests that there are threecointegrating relations, a maximum eigenvalue test indicates that there is only one cointegratingrelation (see Table B3). The study adopts results of the trace test since it is known that the test ismore robust to both skewness and excess kurtosis in the residuals than the maximal eigenvaluetest (see Harris & Sollis (2003); Cheung & Lai (1993)).3 Applying the Johansen approach to cointegration of multivariate systems, three possible cases are expected in therelationship between the rank of matrix and its characteristic roots. First is a case of full rank i.e. rank(number of variables). All rows (columns) are linearly independent (i.e. all variables are I(0)) and all characteristicroots are inside the unit circle with modulus less than 1. The system is effectively stationary and the levels of thevariables have stationary means. The appropriate modelling strategy, therefore, is to estimate the standard VAR inlevels. Second is a case of zero rank i.e. rank . In this event, , all rows are linearly dependent,there are no cointegrating variables and the system is non-stationary. In this case, is an matrix of zerosand the appropriate model is a VAR in first differences involving no long run elements as given by:The third case is one of reduced rank i.e. rank. The system is nonstationary but there are rcointegrating relations among the variables. The cointegrating relation is determined by and .v | P a g e
B-2.0. Cointegrated SVAR AnalysisEstimation results from the cointegrated structural VAR are generally similar to those from thelevels estimation carried in line with Sims, Stock and Watson (1990). Understandably, a numberof differences also show up. Among the differences, impulse responses from the cointegratedSVAR die off very quickly compared to those from the estimation in levels. In order to retainclear visual images, the forecast horizon is reduced from 60 months in the levels estimation to12-month in the cointegrated SVAR.The cointegrated SVAR confirms the finding in the levels estimation that monetary policy inMalawi employs hybrid operating procedures, with the bank rate and reserve money as operatingtools. Both the bank rate and reserve money respond significantly to shocks in the threeintermediate targets of monetary policy namely exchange rates, aggregate money supply andbank lending (see Figure B1), revealing that the central bank is concerned with movements in thethree targets and to achieve desired levels in these targets, the two policy tools are used.Consistent with the levels estimation, the cointegrated SVAR also shows that the exchange rateand money effect are important channels of monetary transmission in the country, though theimpact is not as pronounced as in the levels estimation. The effect of bank lending in themonetary transmission process, however, is insignificant in the cointegrated SVAR, whichcontradicts the finding in the levels estimation.The observed differences from the two estimation approaches are not unexpected. An importantsource of these differences is the imposition of what may be possibly incorrect cointegratingrestrictions in the process of estimating the cointegrated VAR. Kim and Roubini (2000) andBecklelmans (2005) argue that this is usually the case in cointegrated VARs with the implicationthat the resulting inferences are often incorrect as well. In an attempt to circumvent the problem,some studies opt for a simple differences specification (see, for example, Weitong (2007);Boivin & Giannoni (2002); Kasa & Popper (1997); Kugler et al (2004); Karame & Olmedo(2002); Mihira & Sugihara (2000)). The approach, however, is not persuasive as it yieldsinconsistent estimates if some variables are cointegrated (Bernanke & Mihov, 1997).vi | P a g e
Page 1 and 2: University of Cape TownSchool of ec
Page 3 and 4: Among the few studies undertaken on
Page 5 and 6: Interest Rate (Percent)In line with
Page 7 and 8: 3.2. Identification of Structural S
Page 9 and 10: The non-zero coefficients and in ma
Page 11 and 12: only the price level in the long ru
Page 13: long-run relationships in a linear
Page 16 and 17: Percent(RM-Innovations)Percent (BR-
Page 18 and 19: Figure 4.4 also shows that an expan
Page 20 and 21: Table 4.1 also reveals that the dif
Page 22 and 23: To determine the importance of the
Page 24 and 25: FIGURE 4.7: Impulse Responses for t
Page 26 and 27: Figure 4.9:Responses to Structural
Page 28 and 29: ates) is re-estimated with the samp
Page 30 and 31: TABLE 4.2: Variance Decomposition -
Page 32 and 33: authorities pursue both price stabi
Page 34 and 35: Dabla-Norris, E., & Floerkemeier, H
Page 36 and 37: APPENDIX AInterpolation of GDP usin
Page 38 and 39: GDPMalawi Kwacha (millions)about 10
Page 42 and 43: Table B1: Stationarity ResultsVaria
Page 44 and 45: Table B3: Johansen Cointegration Te
Page 46 and 47: APPENDIX CTABLE C1: Roots of Charac
Page 48 and 49: APPENDIX CTABLE C3: Structural VAR
Page 50 and 51: APPENDIX CTABLE C5: Roots of Charac
Page 52 and 53: APPENDIX DFIGURE D1:Impulse Respons

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

Create successful ePaper yourself

Delete template?

Save as template?