Monetary Variability and Monetary Variables in the Franc Zone · 2017. 2. 7. · Monetary Variability and Monetary Variables in the Franc Zone Simeon Coleman and Michail Karoglou1

Monetary Variability and Monetary Variables in the Franc Zone

Simeon Coleman and Michail Karoglou1

ABSTRACT

Failure to detect or account for structural changes in economic modelling canlead to misleading policy inferences, which can be perilous, especially for themore fragile economies of developing countries. Using three potential monetarypolicy instruments (Money Base, M0, and Reserve Money) for 13 member-statesof the CFA Franc zone over the period 1989:11-2002:09, we investigate themagnitude of information extracted by employing data-driven techniques whenanalyzing breaks in time-series, rather than the simplifying practice of imposingpolicy implementation dates as break dates. The paper also tests Granger's(1980) aggregation theory and highlights some policy implications of the results.

1. INTRODUCTION

THE IMPORTANCE OF MONETARY VARIABLES in macroeconomic modelling is read-ily evident in almost all applied macroeconomic literature. Capturing theidiosyncratic behaviour of their time-series dynamics is essential, par-

ticularly when one considers the impact of monetary variability on inflationuncertainty and decreasing levels of economic activity.2 Although the sourcesof monetary variability can be quite diverse — ranging from differing nationalmonetary policies or money demand disturbances to speculative short-terminternational capital transactions — the association between high monetaryvariability, inflation uncertainty and decreasing levels of economic activityshould be an important policy issue.3

On the one hand, substantial empirical evidence suggests a welfare-reducing property of monetary variability (see for example the rational expec-tations macro-models of Barro, 1976; Gray,1976). On the other hand,Devereux (1987) finds evidence to the contrary. This apparent contradiction of

Economic Issues, Vol. 15, Part 2, 2010

- 17 -

views notwithstanding, the accepted view is that high monetary variabilityisassociated with substantial policy implications.4 It is plausible to infer that theoverall effect, whetherpositive or negative, islikely tobe morepronounced insmall and developing economies, not least, because of less developed financialmarkets and the central role of holding physical cash. Fielding (2004, p.2), inan article on the Franc zone of sub-Saharan Africa, states that ‘ManipulatingM0 is still a potentially effective monetary policy strategy in countries wherewhere financial marketsareunderdeveloped and cash makes up a large frac-tion of the money stock’.5 However, as M0 is the poor person’s financial asset,it is most likely that the poor will bear the brunt of any inflationary spikes.Consequently, the issue of monetary variability, although important to anyeconomy, is paramount to welfare and policy formulation in developing andtransitional economies.

The recent empirical literature has addressed the issue of monetaryvariability by considering breaks in corresponding series. In almost all cases,the breaks are identified using exogenous information such as the announce-ment or the implementation date of a particular monetary policy. Suchmethodologies are relatively simple to implement, but imply not only that theimpact of monetary policy on monetary variability is instantaneous and easi-ly identified, but also that a significant change in monetary policy generatesat most a single break in monetary variability. A more realistic methodologywould consider possible transitional periods between an announcement date,implementation date(s) and the actual date(s) that monetary variabilitychanges are observed. In addition, it would test for the existence of any otherbreaks that might exist in the series. However, the inherent complexity ofimplementing such methods would be unnecessary if their results turned outto be similar to the results of the simpler and less rigorous methodologies.

Along these lines, the first purpose of this paper is to introduce aframework for conducting time-series analysis of monetary variability whichmakes use of recent advances in the non-parametric methods of detectingbreaks in the mean and volatility dynamics. More data becoming readily avail-able guarantees that such methods will continue becoming more popular inempirical work. In addition, because the underlying methods are data-driven,their results depend primarily on the properties of the available data and noton a collection of exogenous information.6

The second purpose of the paper is to investigate, empirically, whetherresults obtained from a methodology drawn from this framework are differentfrom results obtained from the standard methodology of imposing the break-date using exogenous information. Using data from the CommunautéFinancière Africaine (hereafter CFA) Franc Zone of sub-Saharan Africa, we findthat in many cases not only do the results differ, but the evolution of monetaryvariability is substantially richer than thought a priori. In fact, we show that theuse of the standard methodology appears to average out the effects of monetaryvariability. Consequently, moving beyond the standard paradigm is not only an

S Coleman and M Karoglou

- 18 -

academic exercise, but is a crucial challenge for policy formulation.Finally, the third purpose of the paper is to examine the extent to which theuse of different monetary variables can affect inferences based on the evolu-tion of monetary variability. Using three monetary variables, the Money Base(aggregate variable), M0 and Reserve Money (as thedisaggregated variables), we and thatinmostcasesinference about monetaryvariability can be substantially affected by the choice of instrument used tomeasure it.7 Interestingly, when viewed in terms of Granger’s (1980) aggrega-tion hypothesis, the results suggest that the disaggregated data contain muchmore information than the aggregated data. This finding, in the context ofmonetary variability, is as interesting as it is unsettling for those who con-struct the macro-models which inform policy makers’ decisions given theweight it places on the appropriate choice of target instrument.

The rest of the paper is organised as follows. Section 2 provides a briefoverview of monetary variability and relates this to the relevance of a curren-cy devaluation incident as considered for the Franc zone. Section 3 describesthe proposed procedure for modelling monetary variability. Section 4 describesthe data employed and discusses theempirical results. Section5 lays out somepossible policy implications of the findings and Section 6 concludes.

2. MONETARY VARIABILITY AND CURRENCY DEVALUATION IN THE FRANC ZONEOne of the most acknowledged monetary policy issues, which is also directlyrelevant to monetary variability, is the issue of currency devaluation. As a pol-icy issue, it relays key signals about the economy to internal and externalobservers. Consequently, the corresponding behaviour of monetary variables,that are likely to be targeted by monetary authorities in the event of currencydevaluation, should attract a lot of attention from both policy makers and theresearch community. Indeed, since the 1970’s research in this area has grownsteadily, but is still relatively limited.8 For our purposes, this offers a solidplatform on which we can base our analysis. The traditional paradigm sug-gests that the currency devaluation (implementation) date introduces, at most,a single (and known) break. We focus on the CFA Franc zone Africaneconomies and for very good reasons. First, these countries are all classifiedas developing economies. Second, they share a common currency, and theJanuary 1994 currency devaluation date is well known and uncontested. Thisuniformity will prove invaluable for drawing general conclusions.9

3. METHODOLOGYThe statistical procedure used in this paper shows a method for capturing thedynamics of monetary variability. It centers around a ‘Nominating-Awarding’procedure (see Karoglou, 2006a), which can be used to provide robust esti-mates of volatility around policy implementation dates. On the one hand, thetraditional paradigm suggests quite simply that the only breakdate is the offi-


- 19 -

cial policy implementation/announcement date. On the other hand, theNominating-Awarding procedure suggests the use of a method of detecting allthe potential breakdates in the series, and then a further method for deter-mining which of these dates are, indeed, breakdates. In both stages, oncebreakdates have been identified the sample can be split accordingly into con-tiguous segments. Then an estimator of the standard deviation can provide ameasure of the variability in each segment.10

In this paper, apart from the sample standard deviation, we also usethe square-root of a Heteroscedasticity and Autocorrelation Consistent (HAC)estimator of the variance. Consequently, we have not only a second morerobust estimate of monetary variability, but also a rather straightforwardmethod of assessing the impact of heteroscedasticity and autocorrelation onthe simple measures of monetary variability. Specifically, we use the VARHACestimator of den Haan and Levin (1997), which bypasses the issue of selectingfor estimation an appropriate bandwidth.11

The above methodology is applied to three principal monetary variab-less: the Money Base and its components — M0 and Reserve Money. In thisway we have also a direct way of testing Granger’s (1980) aggregation hypoth-esis; in other words we determine whether the disaggregated data containmore information than the aggregate data. If the number and/or timing of thebreaks identified in M0 and Reserve Money do not correspond to the breaks inthe Money Base then the disaggregated data contains more information aboutmonetary variability than the aggregated data and, therefore, should be pre-ferred for macro-modelling. This is particularly important for policy making,when decisions are based on inferences that are sensitive to the underlyingeconometric model. It is instructive at this stage to outline the Nominating-Awarding procedure.

The Nominating breakdates stage

The Nominating breakdates stage detects all the dates that could potentiallybe breakdates.12 A possible method is to use exogenous information, which isexactly what the traditional paradigm on monetary variability suggests(1994:01, in our case). To avoid imposing such a restriction on the informa-tion the data can provide, using a data-driven method is suggested. This papermakes use of several statistical procedures, based on a number of recentlydeveloped non-parametric tests that detect a single break in volatility dynam-ics. Specifically, we use: (i) the test of Inclan and Tiao, 1994 (I&T ); (ii) the twotests of Sansó, Aragó, and Carrion, 2003 (SAC1 and, depending on whetherwe use the Bartlett or V ARHAC kernel variant, and ); (iii) themodified version of the test of Kokoska and Leipus, 2000 (K&L) as proposedby Andreou and Ghysels, 2002 (again, depending on whether we use theBartlett or V ARHAC kernel variant, and ); and (iv) the test ofLee, Tokutsu and Maekawa, 2003 (LTM).


- 20 -

2BTSAC 2

VHSAC

& BTK L & VHK L

A key property of these tests is that they do not distinguish betweenbreaks in the mean and volatility dynamics (Karoglou,2006b). For our pur-poses, this is a plausible characteristic, since both types of break have to betaken into account. Some other properties of these tests worth mentioninginclude the following: I&T has been found to be most sensitive to the existenceof volatility breaks for independent and identically distributed (iid) data, butsuffers severe size distortions for strongly dependent data. In contrast, thevariants of the K&L and SAC2 tests do not exhibit size distortions for stronglydependent data, but are known to exhibit lower power. The performance ofboth the SAC1 and LTM tests lie somewhere in between. In general, the rela-tive performance of each of the above tests depends on the underlying data-generating process (DGP).13 Therefore, when the DGP is not known, it isinstructive to use all of these tests and, depending on the specific objective of

The Awarding breakdates stage

Following the nomination of the potential breakdates outlined in the preced-ing section, this stage serves as a screening process and robustness check.The method is used to decide whether a nominated breakdate is indeed abreakdate. Initially, we simply assume that all nominations are breakdates. Infact, one can infer that this is what the traditional paradigm on monetary-variability implicitly suggests. A more elaborate, but more robust method is tocompare, statistically, the distributions of each pair of contiguous segments.In this paper, we focus on the mean and/or the varianceof contiguous seg-ments. If both are statistically equal then we can assume that the two seg-ments come from the same distribution,15 in which case, we unite the two seg-ments into one. If either the mean or the variance of each segment is statisti-cally different from the mean or the variance of its contiguous segment thenwe award the breakdate property to the hitherto nominated breakdate.Consequently,application of this method results in a subset of the nominatedbreakdates, which will include only those breakdates that introduce a changein the mean and/or the variance. These are then deemed to be the detectedbreakdates of the Nominating-Awarding procedure.

Finally, equality of the means is tested by the standard t-test and theequality of the variances by the standard F-test, the Siegel-Tukey test withcontinuity correction (Siegel and Tukey, 1960; Sheskin, 1997), the adjustedBartlett test (seeSokal and Rohlf, 1995; Judge et al., 1985), the Levene test(1960) and the Brown-Forsythe test (Brown and Forsythe, 1974).16 It is impor-tant that we test the homogeneity of variances with more than one methodbecause of the variations in the properties of each test. For example, on theone hand, the standard F-test, although quite powerful, requires sample sizesto be equal and is also sensitive to departures from normality. The Siegel-Tukey test, on the other hand, although less sensitive to such departures doesnot require the equality of sample sizes, assumes independence in the sam-ples, and equal medians. The original Bartlett test is also robust when thesample sizes are not equal, however it is sensitive to departures from normal-ity and for that reason we employ its adjusted version which uses a correctionfactor for the critical values and the arc-sine square-root transformation of thedata, in order to conform with the normality assumption. The Levene test, analternative to the Bartlett test, is less sensitive to departures from normalitywhile the Brown-Forsythe test(1974), a variant of the Levene test, uses themedian instead of the mean to improve the performance (i.e. size and power)of the test when the underlying distributions are skewed. Using such a bat-tery of tests requires us to employ a decision rule in order be able to drawcoherent conclusions, especially when these tests produce conflicting results.The decision rule used in this paper is that if any two of these tests reject thenull (i.e. the means and/or variances are not equal) then the segments are notunited — in other words, the starting date of the second segment is identifiedas a breakdate.


- 22 -

4. DATA AND EMPIRICAL RESULTSThe dataset used in this paper comprises monthly data for MoneyBase, M0,and Reserve Money of 13 member states of the Franc zone extracted from theInternational Monetary Fund's International Financial Statistics (IFS) data-base and spanning the period1989:11—2002:09.17 We note that the M0 datafor each country is the value of notes and coins withdrawn from bank branch-es in that country, net of the value deposited. Money withdrawn from countryi can be spent in country j making the tracking of the conventional M0 (vol-ume of notes and coins in circulation) difficult. There have been someattempts by the central banks to track the movement of cash across nationalborders, primarily due to large cross-border movements (particularly, migra-tion and remittances), however these have been largely unsuccessful and any-policy deductions made in this paper should take this into consideration.18

The west-African member states included are Benin (Ben), Burkina-Faso(BFaso), Côte d’Ivoire (Civ), Mali (Mal), Niger (Ner), Senegal (Sen) and Togo(Tog), jointly referred to as UEMOA; while the central-African member statesare Cameroon (Cam), Central African Republic (CAR), Chad (Chd), CongoRepublic (Con), Equatorial Guinea (GEQ), and Gabon (Gab), often jointlyreferred to as CEMAC.19


- 23 -

Δlog M0 Δlog(Reserve Money) Δlog(Money Base)Ben -7.01(1) -11.04(1) -14.78(0) BFaso 9.99(0) 13.68(0) 12.11(0) Cam -13.37(0) -14.43(0) -14.92(0)CAR -10.61(0) -14.96(0) - 9.74(0)

Cgo -16.23(0) -14.42(0) -13.36(0)

Civ - 7.47(2) -11.38(1) - 9.49(0)

Gab -19.26(0) -15.92(0) -17.12(0)

GEQ -11.92(2) -13.63(0) -11.70(0)

Mal -14.72(0) -15.22(0) -13.40(0)

Ner -18.79(0) -17.03(0) -18.48(0)

Sen -11.39(0) -15.49(0) -11.70(0)

Tcd -10.46(0) -14.61(0) -10.78(0)

Tog -7.43(1) -11.40(1) -17.21(0)

Table 1: Augmented Dickey-Fuller (ADF) stationarity tests on Δlog(M0),Δlog(Reserve Money) and Δlog(Money Base) across member states.

Notes: The null hypothesis of a unit-root is rejected in all cases at the 1% level.Numbers in parentheses represent selectedlag order based on the SchwartzInformation Criterion (SIC) with a maximum lag length of 3.


- 24 -

Table 2: Correlation of Money Base Series Across Member States

Note: Entries in the top right half of the matrix have been ignored since they are a mir-ror image of the bottom left section.

Table 3: Correlation of M0 Series Across Member States

BFasoCivMalNerSenTogCamCARTcdCgoGEQGab

Ben0.040.05

-0.09-0.240.26

-0.010.290.08

-0.170.140.070.14

CAR

0.000.000.090.19

Cam

-0.11-0.050.500.060.20

Tog

0.110.070.180.210.090.15

Sen

0.120.360.160.120.390.090.30

Ner

-0.080.06

-0.010.04

-0.040.09

-0.05-0.01

Mal

0.000.120.130.090.120.160.010.030.02

Civ

0.190.090.570.110.410.210.030.350.210.27

BFaso

0.180.040.040.36

-0.120.190.080.050.160.060.09

Cgo

0.100.41

EQ

0.16

Ted

0.010.060.06


Ben0.15

-0.010.00

-0.260.240.030.17

-0.09-0.140.070.050.09

CAR

0.080.070.090.23

Cam

-0.06-0.010.190.120.12

Tog

0.110.010.080.140.140.06

Sen

0.110.170.070.170.24

-0.030.24

Ner

0.070.11

-0.06-0.01-0.040.05

-0.060.03

Mal

0.100.070.210.060.050.130.120.00

-0.02

Civ

0.040.040.260.130.060.130.040.370.160.19

BFaso

0.160.100.030.27

-0.040.040.090.190.140.010.16

Cgo

0.080.24

GEQ

0.13

Ted

-0.010.070.13


Table 1 presents the basic descriptive statistics of the log-differences of thedata. Note also that the data to be analysed in this paper are in log-differences.In addition, Tables 2 - 4 show the cross-country correlations between themonetary variables, all of which the Central Bank monetary authorities maytarget with monetary policy. The monetary variables appear to be uncorrelat-ed, even for countries within each of the sub-unions (UEMOA and CEMAC).This preliminary observation is important and may be interpreted as signallingsub-optimality of a single zonal policy; especially when there is limited scopefor policy formulation at the country level. Monetary variability across mem-ber states may therefore vary significantly — this issue is tested empiricallylater in the paper.

Breakdates in the monetary variables

Our findings suggest that there are no significant breaks in the means of theseries in either the pre- or post-devaluation periods (see Table 5). In fact, forall the series and across all the member states, the null of equal means isrejected only for the two periods in the Money Base series of Cam and only atthe 10 per cent significance level. The overwhelming support for equal meansin the money variables over the pre- and post-devaluation periods may beattributed to the administrative structures of the Zone, which are designed (atleast, in principle) to ‘harmonise’ the effects of monetary policy of memberstates.20


- 25 -

Table 4: Correlation of Reserve Money Across Member States


Ben0.190.18

-0.05-0.050.19

-0.090.040.05

-0.130.00

-0.12-0.04

CAR

0.000.060.040.19

Cam

0.24-0.08-0.010.100.17

Tog

0.050.070.030.080.02

-0.11

Sen

-0.060.00

-0.050.06

-0.08-0.030.10

Ner

0.26-0.080.00

-0.07-0.10-0.040.120.01

Mal

0.020.220.01

-0.100.010.08

-0.020.070.08

Civ

-0.180.18

-0.01-0.040.050.17

-0.140.15

-0.010.10

BFaso

0.120.150.290.170.080.060.030.030.08

-0.090.03

Cgo

-0.03-0.04

GEQ

0.01

Ted

-0.11-0.11-0.04


Tables 6 - 8 report the number and timing of the detected breaks in the vari-ance of the Money Base, M0 and Reserve Money respectively (after theNominating breakdates stage) while Tables 16- 21 report the correspondingresults after the Awarding breakdates stage. Finally, Tables 9 - 15 report thealternative measures of volatility used in this paper to identify the variancesin the pre- and post-official devaluation periods, and also in each of the iden-tified regimes. Although we find no evidence of significant breaks in themean of the series, the results in Tables 6 - 8 underscore the importance ofinvestigating breaks in both the mean and the variance of series used inempirical applications.

Overall, the results confirm that neighbouring (nominated) segments,as reported in the Columns 2 - 7 of Tables 6 - 8, have statistically significant-ly different variances, typically at the 5 per cent level. Except for four coun-tries, (CAR, Civ, Cgo and Tcd), the variances in the pre- and post- official cur-rency devaluation periods are statistically different at the 5 per cent level forboth the Money Base and M0 series. For Gab, there appears to be no statisti-cal difference in the variances in the pre- and post-currency devaluation peri-ods for the Money Base alone. A similar scenario is observed for both Sen andTog when we consider M0 alone. With respect to Reserve Money, exceptingCAR, Civ, Cam and Gab, the variances in the pre- and post-devaluation peri-ods are all found to be statistically different. Therefore, from a broader pointof view, out of the 13 member states, after imposing the breakdate and split-ting the data into the pre- and post-official currency devaluation periods (with50 and 91 observations respectively), we detect a statistically significant break


- 26 -

BenBFasoCivMalNerSenTogCamCARTcdCgoGEQGab

pre-0.01680.00830.00320.0071

-0.0002-0.0037-0.0106-0.00920.0053

-0.00370.0060

-0.01380.0028

post-0.00870.00020.00900.0111

-0.00070.00530.00710.0127

-0.00120.01150.00900.02440.0021

t-test0.491.090.490.280.020.701.30

1.86***1.171.260.220.920.04

pre--0.00050.00780.00390.00580.00390.0051

-0.0152-0.00600.00530.00120.0032

-0.01750.0016

post-0.0099

-0.00280.00670.0094

-0.00250.00250.00760.0069-0.0050.01200.00760.01800.0058

t-test0.451.470.250.230.310.191.141.131.030.860.370.510.37

pre-0.03450.0092

-0.00510.0083

-0.0064-0.0193-0.0092-0.02230.0044

-0.03480.0267

-0.00900.0099

post-0.00480.00690.01680.01540.00510.01130.00580.0214

-0.01500.00940.01310.0389

-0.0064

t-test0.830.090.380.210.300.730.441.510.240.960.350.830.35

Δlog(Money Base) Δlog(M0) Δlog(Reserve Money)

Table 5: Tests for Significantly Different Means in the Pre- and Post Devaluation Period

Notes: In this table, *** indicates rejection of the null of no significant difference inthe means of thetwo periods at the 10% level of significance.

(at the 5% level) in the variance for 8 member states for the Money Base series(Table 6); 7 for the M0 series (Table 7); and 9 for the Reserve Money series(Table 8). When we do not impose the breakdate but instead we detect thechanges non-parametrically (Columns 2 - 7), the evolution of volatility appearscompletely different.


- 27 -


10.050(56)

.

.0.0484(56)0.043(54)0.065(99)0.039(49)0.069(91)0.043(77)0.052(48)0.058(119)0.388(38)

.

20.117(96)

.

.0.0774(43)0.079(44)0.082(44)0.101(22)0.061(61)0.028(75)0.118(24)0.146(15)0.117(66)

.

3...

0.1213(39)0.160(48)

.0.079(77)

.

.0.067(47)0.061(18)0.173(48)

.

4.........

0.029(18)...

5.............

6.............

pre-0.0440.0350.0630.0480.0390.0610.0390.0420.0350.0620.0670.3490.087

post-0.1130.0470.0700.0920.1310.0780.0920.0770.0300.0720.0800.1400.082

Notes: Along a given row, a bold font indicates a detected structural change relative tothe preceding segment. A ‘.’ indicates no break detection using the algorithm.

Table 6: Detected Structural Changes in Δlog(Money Base).

Regimes


10.082(49)0.034(99)

.

.0.026(29)

.,

0.044(58).

0.046(47).

0.660(46)0.080(74)

20.153(52)0.040(49)

.

.0.053(17)

.

.0.076(92)

.0.129(23)

.0.184(12)0.050(77)

30.306(14)

.

.

.0.091(45)

.

.

.

.0.066(48)

.0.139(54)

.

40.116(13)

.

.

.0.177(42)

.

.

.

.0.024(18)

.

.

.

50.055(20)

.

.

.

.

.

.

.

.

.

.

.

.

6.............

pre-0.0880.0310.0610.0720.0500.0780.1250.0430.0340.0640.0700.6360.078

post-0.1510.0460.0660.0960.1420.0710.0920.0750.0320.0740.0670.1360.056

Notes: Along a given row, a bold font indicates a detected structural change relative tothe preceding segment. A ‘.’ indicates no break detection using the algorithm.

Table 7: Detected Structural Changes in Δlog(M0).

Regimes

Overall, the results confirm that neighbouring (nominated) segments, asreported in Columns 2 - 7 of Tables 6 - 8, have statistically significant differ-ent variances, typically at the 5 per cent level. Except for four countries, (CAR,Civ, Cgo and Tcd), the variances in the pre- and post- official currency deval-uation periods are statistically different at the 5 per cent level for both theMoney Base and M0 series. For Gab, there appears to be no statistical differ-ence in the variances in the pre- and post-currency devaluation periods for theMoney Base alone. A similar scenario is observed for both Sen and Tog whenwe consider M0 alone. With respect to Reserve Money, excepting CAR, Civ,Cam and Gab, the variances in the pre- and post-devaluation periods are allfound to be statistically different. Therefore, from a broader point of view, outof the 13 member states, after imposing the breakdate and splitting the datainto the pre- and post-official currency devaluation periods (with 50 and 91observations respectively), we detect a statistically significant break (at the 5per cent level) in the variance for 8 member states for the Money Base series(Table 6); 7 for the M0 series (Table 7); and 9 for the Reserve Money series(Table 8). When we do not impose the breakdate but instead we detect thechanges non-parametrically (Columns 2 - 7), the evolution of volatility appearscompletely different.

In particular, our results confirm the following:

(i ). In the case of the Money Base, there are significant breaks for BFaso,Civ, Sen, Cam and Gab; one break for Ben and CAR; two breaks for Mal, Ner,Tog, Cgo, and GEQ and three breaks for Tcd.


- 28 -


10.069 (21)0.089(58)0.030(78)0.070(54)0.065(40)0.129(54)0.044(40)0.173((54)0.416(99)0.343(54)0.256(33)0.466(32)0.305(66)

20.055(35)0.190(84)0.575(21)0.242(94)0.138(13)0.175(20)0.060(17)0.165(98)0.557(53)0.228(98)0.208(119)0.291(117)0.240(86)

30.257(94)

.0.245(42)

.0.235(18)0.400(17)0.241(41)

.

.

.

.

.

.

4....

0.337(36)0.262(61)0.242(54)

.

.....

5. ...

0.206(13)........

6....

0.182(22)....... .

pre-0.0640.0880.2910.0670.0760.1270.0510.1680.3910.3120.2510.3890.289

post-0.2500.1670.3420.2310.2620.2800/2400.1630.4900.2300.2030.2940.255

Note: Along a given row, a bold font indicates a detected structural change relative tothe preceding segment. A ‘.’ indicates no break detection using the algorithm.

Table 8: Detected Structural Changes in Δlog(Reserve Money).Regimes

(ii). In the case of M0, there are no breaks for BFaso, Civ, Sen, Mal, Ner, Tog,and Cgo; one break for Cam, Gab and GEQ, Ben and CAR; three breaks forNer and Tcd; and four breaks only in Ben.

(iii). Finally, for Reserve Money, there are no breaks for Cam; there is onebreak for Ben, BFaso, Mal, Tog, CAR, Tcd, Cgo, GEQ and Gab; two for Civ;and three for Ner and Sen.

The evolution of monetary variability


- 29 -

Money Base

BenRegime 1Regime 2Pre-Post-

BfasoPre-Post-

CivPre-Post-

MalRegime 1Regime 2Regime 3Pre-Post-

NerRegime 1Regime 2Regime 3Pre-Post-

SenRegime 1Regime 2Pre-Post-

TogRegime 1Regime 2Regime 3Pre-Post-

Obs.

15256965091

1525091

1525091

1525643395091

1525444485091

15299445091

1524922775091

Mean

0.01080.00990.01130.01680.00870.00260.00840.00020.00910.00310.00900.00750.0103-0.00120.00570.00700.0110-0.00240.0003-0.01210.0035-0.0003-0.00060.00450.00030.0095-0.00360.00530.0010-0.01040.00270.0027-0.01070.0071

Std. dev.

0.09900.04750.11920.04300.11280.04580.03480.04620.06970.06200.07010.08100.04790.07650.11980.04760.09150.10530.04220.07820.15830.03870.13050.07210.06540.08120.06060.07750.07980.03880.09890.07220.03850.0917

VARHAC

0.08240.04620.11470.04180.11260.04660.03580.04530.05810.08680.05260.04500.03380.05580.05750.03090.06220.07050.04210.04030.09900.03830.08440.04650.06510.08450.05930.07900.05740.06100.04770.05660.06270.0639

Skewness

-0.09-0.42-0.07-0.320.040.70-0.53-0.151.041.171.050.280.93-0.650.521.170.420.710.410.430.55-0.130.590.601.06-0.071.060.500.890.080.710.040.090.61

Kurtosis

5.562.814.152.774.474.743.752.883.823.964.1512.118.953.898.7010.3611.198.243.222.714.702.145.753.734.653.024.913.576.192.443.013.492.484.97

Table 9: Volatility Measures for Δlog(Money Base).

cont.....


- 30 -

Note: Pre- and Post- indicate the periods before and after the devaluation date respectively.

Table 10: Volatility Measures for Δlog(Money Base) ...cont

Money Base

CamRegime 1Regime 2Pre-Post-

CARRegime 1Regime 2Pre-Post-

TcdRegime 1Regime 2Regime 3Regime 4Pre-Post-

CgoRegime 1Regime 2Regime 3Pre-Post-

GEQRegime 1Regime 2Regime 3Pre-Post-

GabPre-Post-

Obs.

15291615091

15277755091

152482447185091

15211915185091

1523866485091

1525091

Mean

0.00680.00280.0126-0.00910.01260.00470.0116-0.00240.0053-0.00120.0059-0.00400.0171-0.0090.0076-0.00360.01150.00940.00600.0599-0.01030.00600.00890.0147-0.02350.03080.0288-0.01390.02440.00760.00270.0021

Std. dev.

0.06580.06900.06020.04160.07620.03660.04260.02740.03450.03010.07050.05130.11510.06640.02800.06110.07140.07310.05830.14110.05870.06600.07920.22890.38340.11640.17140.34570.13930.08600.08600.0811

VARHAC

0.07240.09880.03610.03890.05880.04730.05600.02790.03390.02980.08250.05230.18750.04610.02850.06270.08010.07360.05800.14880.05610.06560.08020.08880.21180.06830.12690.18130.11600.06220.05990.0572

Skewness

-0.180.18-0.57-0.42-0.200.680.61-0.19-0.34-0.070.35-0.530.040.231.55-0.180.692.080.631.620.330.462.640.260.400.320.590.360.630.02-0.160.01

Kurtosis

3.733.853.562.513.256.545.534.993.735.075.423.762.694.665.054.436.1514.014.356.003.274.0315.9712.866.403.183.557.354.623.183.183.37


- 31 -

M0

BenRegime 1Regime 2Regime 3Regime 4Regime 5Pre-Post-

BfasoRegime 1Regime 2Pre-Post-

CivPre-Post-

MalPre-Post-

NerRegime 1Regime 2Regime 3Regime 4Pre-Post-

SenPre-Post-

TogPre-Post-

Obs.

15249521413205091

15299495091

1525091

1525091

152291745425091

1525091

1525091

Mean

0.0115-0.00540.02530.03610.0382-0.0053-0.00050.01000.00220.0119-0.00880.0079-0.00270.00750.00390.00670.00930.00580.0094-0.0009-0.0046-0.00430.0053-0.00820.0038-0.00260.00570.00500.00260.0076-0.01510.0075

Std. dev.

0.14080.08090.15130.29440.11130.05400.08710.14990.04080.03350.04000.03050.04560.06350.06000.06540.09700.07090.09610.11410.02600.05130.09020.17480.04890.14150.07360.07680.07050.13040.12330.1052

VARHAC

0.16480.08140.14880.18670.07340.05670.08830.13340.04700.03370.01680.02990.05540.04400.10010.05100.06960.09970.06970.07520.02570.026170.04880.06150.05010.08920.04130.08600.05290.11240.11780.0669

Skewness

0.27-0.310.20-0.11-0.290.450.02-0.03-0.410.130.16-0.36-0.231.211.271.230.620.750.540.050.67-0.440.220.141.070.051.491.311.651.64-0.380.80

Kurtosis

5.943.184.072.152.552.273.535.724.534.243.743.624.173.883.774.0026.336.4028.458.443.932.333.804.906.655.905.794.836.7312.723.945.56

Table 11: Volatility Measures for Δlog(M0).

cont.....


- 32 -

M0

CamRegime 1Regime 2Pre-Post-

CARPre-Post-

TcdRegime 1Regime 2Regime 3Regime 4Pre-Post-

CgoPre-Post-

GEQRegime 1Regime 2Regime 3Pre-Post-

GabRegime 1Regime 2Pre-Post-

Obs.

15258925091

1525091

152472348185091

1525091

1524612545091

15274775091

Mean

0.0035-0.00470.0083-0.00600.00700.00460.0053-0.00050.00690.0047-0.0043-0.00010.00750.00110.01190.00770.00320.00760.0129-0.01590.05250.0134-0.01740.01790.00700.00760.00580.00150.0058

Std. dev.

0.06490.04370.07530.04270.07450.03400.03550.03170.07160.04610.12660.06500.02320.06330.07380.06740.06900.06700.37880.65250.17660.13730.62970.13530.06590.07960.04970.07760.0561

VARHAC

0.06540.04200.07540.04040.07530.04000.03380.03110.08950.04720.24350.03850.02530.06520.08800.04800.06780.04910.12050.20070.08350.07870.19490.08110.03530.03980.02940.02570.0291

Skewness

0.64-0.220.58-0.250.61-0.17-0.38-0.180.32-0.400.211.000.78-0.470.620.240.240.38-0.25-0.03-0.010.00-0.02-0.120.000.05-0.230.23-0.78

Kurtosis

4.542.923.723.113.904.503.885.535.793.882.486.704.785.266.193.713.623.9011.034.102.566.924.355.825.594.724.334.116.31

Table 12: Volatility Measures for Δlog(M0) ...cont



- 33 -

Reserve moneyBen

Regime 1Regime 2Regime 3Pre-Post-

BfasoRegime 1Regime 2Pre-Post-

CivRegime 1Regime 2Regime 3Pre-Post-

MalRegime 1Regime 2Pre-Post-

Obs.1522135945091

15258845091

1527821425091

15254945091

Mean0.00850.06410.00490.00080.03450.00480.00330.0091-0.00700.00920.00700.01610.00390.02380.0211-0.00500.01690.00470.00720.00250.00830.0153

Std. dev.0.20640.06720.05420.25580.06280.24890.15600.08820.18890.08750.16600.32960.30290.56120.24190.28770.34060.19510.06880.24030.06610.2302

VARHAC0.15480.06950.05670.15480.06250.15490.14090.08950.13290.08900.12320.14480.19280.56040.24790.08240.14420.15960.06350.18870.05980.1623

Skewness-0.67-0.87-0.54-0.48-0.23-0.35-0.630.45-0.570.570.080.170.280.24-0.860.050.180.810.000.720.100.94

Kurtosis4.233.032.542.822.372.566.543.375.163.752.915.103.033.035.732.816.047.472.805.252.965.69

Table 13: Volatility Measures for Δlog(Reserve Money)

Reserve moneyNer

Regime 1Regime 2Regime 3Regime 4Regime 5Regime 6Pre-Post-

SenRegime 1Regime 2Regime 3Regime 4Pre-Post-

TogRegime 1Regime 2Regime 3Regime 4Pre-Post-

Obs.1524013183613225091

152542017615091

152401741545091

Mean-0.00470.0037-0.0328-0.06450.0168-0.02900.0386-0.00650.00500.0030-0.0091-0.0035-0.04250.0286-0.01940.0113-0.0058-0.0020-0.0475-0.01700.0130-0.00920.0058

Std. dev.0.21670.06440.13220.22880.33200.19820.17830.07490.26110.23250.12790.17030.38790.25960.12580.27880.19220.0290.05860.23770.24000.04890.2386

VARHAC0.15790.06060.05810.05490.22450.19220.12560.07350.17890.10490.12400.16650.21790.11780.12120.10360.13020.04300.06160.17960.15520.03110.1487

Skewness0.09-0.890.970.07-0.120.730.22-0.400.080.12-0.410.400.73-0.39-0.400.060.17-0.290.930.23-0.050.020.06

Kurtosis3.973.413.252.162.333.412.512.632.895.444.011.823.703.714.164.254.222.533.882.463.082.622.84

Table 14: Volatility Measures for Δlog(Reserve Money)Note: Pre- and Post- indicate the periods before and after the devaluation date respectively.


...cont

Tables 9 - 15 report the main descriptive statistics and the two volatility meas-ures used in this paper to identify the variances in the pre- and post- devalu-ation periods, and also for each identified regime.

In particular, we can deduce the following from the results in Tables 9- 10 (Money Base). On the one hand, when January 1994 is imposed as thebreakdate, only the results for GEQ suggest a significant decrease in volatilityin the period following the official currency devaluation date. In addition, forCiv, Tcd, Cgo and Gab we find no statistical difference between the volatilitybetween the two periods. On the other hand, when we do not impose a breakdate, the algorithm suggests that multiple regimes actually exist in mostcases.


- 34 -

Reserve moneyCam

Regime 1Regime 2Pre-Post-

CARRegime 1Regime 2Pre-Post-

TcdRegime 1Regime 2Pre-Post-

CgoRegime 1Regime 2Pre-Post-

GEQRegime 1Regime 2Pre-Post-

GabRegime 1Regime 2Pre-Post-

Obs.15254985091

15299535091

15254985091

15233

1195091

15232

1175091

15266865091

Mean0.0117-0.00670.0219-0.02240.02150.01300.01070.01720.0043-0.01490.0017-0.00720.0066-0.03480.00950.01870.01780.01900.02670.01300.0169-0.00840.0284-0.00910.03890.01120.02500.00070.0100-0.0064

Std. dev.0.16880.17570.16390.16590.16180.46670.41420.55140.38660.48730.27250.34010.22660.30860.22930.21750.25220.20690.24860.20210.33140.45840.28980.38470.29210.26840.30260,23830.28580.2540

VARHAC0.14280.14250.13170.11680.12960.34960.47780.17100.30940.27600.22600.34770.17640.31570.17610.15220.19740.13880.25460.20640.15440.46190.29220.38020.28020.20150.30900.14470.17440.1690

Skewness0.220.340.170.300.060.390.440.320.580.230.05-0.020.26-0.400.300.960.381.240.421.510.190.210.210.240.28-0.35-0.37-0.39-0.38-0.41

Kurtosis3.042.983.133.183.123.855.012.684.873.413.472.883.162.463.095.142.626.422.618,125.184.074.545.304.903.673.363.773.823.60

Table 15: Volatility Measures for Δlog(Reserve Money) ...cont


With the exception of BFaso, Civ, Sen, Cam and Gab, where no signif-icant difference in volatility of the series is found, at least two distinct regimesare found for each of the remaining member states. Of the eight remainingcases where significantly different breaks are determined, only Tog and Tcdsuggest an initial break within two months prior (November 1993) to the offi-cial devaluation date. For Mal and Ner, the initial break occurs within sixmonths (by July 1994) following the official devaluation date, while the initialbreak occurs later for the other member states. In most of the cases where asignificant break is detected, it is suggestive of an increase in volatility of theseries, however this is not conclusive as this is not the case for all memberstates.

In Tables 11-12 (M0), imposing the break date suggests that approxi-mately half of the member states (six out of thirteen) show a significant changein the volatility in M0 between the two periods. However, not imposing abreakdate appears to be more informative, as the procedure indicates a rich-er evolution of volatility in both the pre-and post-official devaluation periods,which is not captured otherwise. Preliminary tests suggest that as many asfive regimes are detected in Ben, four in Tcd and Ner, three in GEQ and two inBFaso, Cam and Gab. Robustness tests, however, show some regimes to bestatistically similar (see Tables 7 and 18) at the one per cent level, such as inCiv, Mal, Sen, Tog, CAR and Cgo.

Tables 14-15 suggest that imposing the January 1994 break dateresults in significantly different volatilities in the Reserve Money for all mem-ber states in the two periods, except Civ, CAR, Cam and Gab. As with Tables9 - 12, the identification of volatilities is richer when the (restriction of a) breakdate is not imposed. As many as six regimes are initially detected in Ner, fourin Sen and Tog, three in Ben and Civ and two in each of the remaining mem-ber states. Further tests for statistical difference, however, suggest that someconsecutive regimes are not statistically different, at least at the five per centlevel (see Tables 6-21). For example, the six regimes detected for ReserveMoney in Ner is confirmed to be actually four distinct regimes. Furthermore,the volatility in the earlier detected regimes 5 and 6 (beginning in July 1999)is found to decrease to the level detected in regime 2 (ending in June 1993).


- 35 -


- 36 -

Benseg1 - seg2

BfasoCivMal

seg1-seg2seg1-seg3seg2-seg3

Malseg1-seg2seg1-seg3seg2-seg3

Senseg1-seg2

Togseg1-seg2seg1-seg3seg2-seg3

Camseg1-seg2

Siegel-Tukey

4.58**4.26**2.22**0.46

3.97**3.46**3.05**0.09

5.69**3.65**5.53**3.11**1.481.00

4.33**4.49**3.64**2.11**3.11**0.03

Barlett

43.97**39.65**4.55**0.85

22.22**10.50**36.89**7.86**64.28**17.67**71.74**20.03**3.47*3.11*

36.74**28.76**23.2**2.49

19.39**1.23

Levene

19.39**21.04**5.17**0.658.8**9.34**9.22**0.95

27.3**17.63**34.17**10.64**3.44*2.43

19.41**29.49**16.39**3.14*

12.00**0.38

Browne-Forsythe

19.35**21.04**5.9**0.49

8.47**9.32**8.73**0.84

27.38**14.27**33.76**10.7**2.81*2.59

19.19**28.84**16.43**3.12*

11.71**0.49

Table 16: Robustness Tests for separate segments in Δlog(Money Base) series

...cont

F-test

6.82**5.46**1.74**1.26

3.66**2.56**6.31**2.46**11.26**3.45**14.11**4.09**1.62**1.56

5.62**6.66**3.92**1.70

3.33**1.30

CARseg1 - seg2

Tcdseg1-seg2seg1-seg3seg1-seg4seg2-seg3seg2-seg4seg3-seg4

Cgoseg1-seg2seg1-seg3seg2-seg3

GEQseg1-seg2seg1-seg3seg2-seg3

Gab

Siegel-Tukey

0.101.140.07

2.63**0.861.89*2.13**3.16**3.06**0.48

3.31**0.761.75*4.42**4.67**3.00**1.82*0.28

Barlett

1.2714.09**

1.7122.15**3.04*6.83**10.26**26.72**12.79**

1.9031.87**

0.0310.86**55.06**66.73**25.66**8.37**0.25

Levene

0.636.00**0.31

16.57**1.263.75*8.7**

14.67**5.22**0.25

17.91**0.02

4.59**20.95**23.18**9.39**5.11**

0.1

Browne-Forsythe

0.605.48**0.20

16.3**1.10

4.24**8.6**14.8**5.33**0.14

17.55**0.00

4.63**20.2**22.70**9.30**4.45**0.09

Table 17: Robustness Tests for separate segments in Δlog(Money Base) seriesF-test

1.322.42**1.36

5.15**1.68*3.23**3.07**16.65**5.42**1.43

6.23**1.07

5.84**6.22**10.98**5.03**2.18**1.13

Notes: In this table, *,**,*** indicates rejection of the null of no significant differencein the variances of the two segments at the 10%, 5% and 1% levels of significance respectively.

Notes: In this table, *,**,*** indicates rejection of the null of no significant difference in the variances of the two segments at the 10%, 5% and 1% levels of significance respectively.


- 37 -

Benseg1-seg2seg1-seg3seg1-seg4seg1-seg5seg2-seg3seg2-seg4seg2-seg5seg3-seg4seg3-seg5seg4-seg5

BFasoseg1-seg2

CivMalNer

seg1-seg2seg1-seg3seg1-seg4seg2-seg3seg2-seg4seg3-seg4

Siegel-Tukey

1.96**2.76**4.51**1.181.26

3.52**0.35

2.86**2.98**4.36**1.231.39

2.14**0.420.98

4.46**2.34**2.57**5.42**0.88

3.57**3.11**

Barlett

15.87**17.81**45.31**

2.583.56*

12.09**1.29

19.36**9.65**36.35**7.91**8.98**2.160.41

5.26**51.60**10.07**37.12**70.56**5.61**21.31**17.48**

Levene5.86**9.62**55.3**2.242.62

16.64**0.67

8.97**12.3**34.78**6.50**5.01**1.930.150.04

18.91**9.09**13.36**24.69**

2.029.23**10.6**

Browne-Forsythe5.87**8.77**51.99**

1.992.69

15.82**0.63

8.29**11.46**31.98**6.67**5.11**1.590.190.04

18.82**7.68**11.13**24.70**

1.659.36**10.94**

Table 18: Robustness Tests for separate segments in Δlog(M0) seriesF-test

2.94**3.49**13.96**

2.012.18**4.00**1.74

7.61**6.95**30.39**4.37**2.21**1.441.18

1.82**8.28**4.01**11.92**44.88**2.97**11.18**3.77**

SenTogCam

seg1-seg2CARTcd

seg1-seg2seg1-seg3seg1-seg4seg2-seg3seg2-seg4seg3-seg4

CgoGEQ

seg1-seg2seg1-seg3seg2-seg3

Gabseg1-seg2

Siegel-Tukey1.431.03

2.14**2.30**0.460.11

3.70**0.21

2.18**2.71**3.98**2.76**0.33

5.78**2.27**5.07**1.74*1.521.41

Barlett0.521.73

16.67**18.05**

0.221.34

32.96**5.32**8.62**14.84**35.29**17.34**

0.07142.77**16.34**93.11**

1.627.17**15.9**

Levene0.771.24

8.88**10.39**

0.320.20**24.23**

1.165.56**13.41**18.92**5.33**0.09

51.96**5.87**34.12**

2.253.75**5.43**

Browne-Forsythe0.761.33

8.82**9.99**0.320.20

23.56**1.20

4.86**13.19**18.62**5.29**0.09

50.29**5.85**34.03**

2.253.52*5.43**

F-test1.201.39

3.02**2.95**1.121.35

7.71**1.99**3.83**3.88**29.50**7.61**1.07

21.87**12.80**22.65**

1.771.93**2.56**

Table 19: Robustness Tests for separate segments in Δlog(M0) series ...cont




- 38 -

Benseg1-seg2seg1-seg3seg2-seg3

BFasoseg1-seg2

Civseg1-seg2seg1-seg3seg2-seg3

Malseg1-seg2

Nerseg1-seg2seg1-seg3seg1-seg4seg1-seg5seg1-seg6seg2-seg3seg2-seg4seg2-seg5seg2-seg6seg3-seg4seg3-seg5seg3-seg6seg4-seg5seg4-seg6seg5-seg6

Siegel-Tukey

6.78**1.44

4.42**6.53**4.21**4.04**0.76

2.11**2.08**3.21**5.76**5.96**5.76**2.49**3.89**5.67**2.1**3.66**2.38**2.25**1.080.121.061.141.48

1.91**2.24**0.26

Barlett

78.23**1.29

31.89**66.75**21.47**32.71**

1.6415.44**

2.4220.86**67.03**72.51**67.16**12.03**41.47**74.59**29.47**29.44**3.53*9.93**1.841.102.600.231.203.48*8.34**0.23

Levene

55.69**1.26

20.91**40.04**19.66**8.34*0.01

11.81**3.68*

15.62**26.88**29.61**38.53**8.74**45.37**46.8**18.82**22.17**5.4**8.57**1.271.212.110.872.083.91*7.64**0.06

Browne-Forsythe

46.53**1.19

17.78**34.13**19.64**18.46**

0.017.53**3.43*

10.23**26.91**29.88**37.87**6.54**

35039**46.41**16.87**20.82**4.43**8.75**1.261.332.19.691.583.94*7.48**0.04

Table 20: Robustness Tests for separate segments in Δlog(Reserve Money) seriesF-test

15.56**1.57

13.96**21.90**3.57**4.56**1.39

3.56**1.55*5.52**12.02**12.11**12.06**4.45**13.01**26.61**10.00**7.82**2.93**5.98**2.251.762.05*1.301.66

2.66**3.40**1.28


cont....


- 39 -

Senseg1-seg2seg1-seg3seg1-seg4seg2-seg3seg2-seg4seg3-seg4

Togseg1-seg2seg1-seg3seg1-seg4seg2-seg3seg2-seg4seg3-seg4

Camseg1-seg2

CARseg1-seg2

Tcdseg1-seg2

Cgoseg1-seg2

GEQseg1-seg2

Gabseg1-seg2

Siegel-Tukey

4.13**1.82*3.62..3.35**1.94*0.681.94*7.34**1.25

6.37**6.25**5.18**3.36**0.230.270.080.991.66*2.18**2.74**3.67**2.71**1.32

2.25**0.231.45

Barlett

31.62**2.74-

38.61**25.19**10.91**3.99**5.13**96.83**

2.5981.15**84.47**26.18**26.94**

0.000.060.383.08*5.96**5.99**12.09**2.91*2.35

5.15**12.53**

0.974.28**

Levene

17.06**4.09**23.18**1.58**5.67**1.613.34*

53.98**1.78

58.73**41.96**21.51**15.26**

0.090.010.111.90

5.74**6.92**11.36**7.24**4.50**2.27

8.13**0.283.34*

Browne-Forsythe

17.11**3.53*

18.08**14.62**4.28**1.762.58

52.05**1.37

51.64**41.28**19.07**15.19**

0.060.010.071.83

4.81**5.83**10.49**7.37**4.48**2.28

8.02**0.28

3.34**

Table 21: Robustness Tests for separate segments in Δlog(Reserve Money) seriesF-test

4.87**1.83

9.60**4.11**5.24**2.25**2.33**23.58**

1.9330.64**31.06**15.88**16.1**1.011.061.161.57*1.79**1.83**2.27**1.531.52

1.75**2.56**1.28

1.62**


...cont

CEMACΔlog(M0)Δlog(Money Base)Δlog(Res. Money)

UEMOAΔlog(M0)Δlog(Money Base)Δlog(Res. Money)

SAC1

---

--

1994:9***-

IT

--

1994:6**

-1994:9***1994:9***1997:12**

---

-1994:9**

--

---

----

KLBT

---

-1994:9**

--

KLVH

---

----

LMT

---

--

1994:9***-

2BTSAC 2

VHSAC

Table 22: Detected Structural Change Dates in UEMOA and CEMAC

Notes: Entries marked ***,**indicate detected breakdates at the 1% and 5% levels of significancerespectively.


- 40 -

Δlog(M0)

CEMACPre-Post-

UEMOAPre-Post-

Obs.

1525091

1525091

1525091

18258945091

15255975091

1525839555091

Mean

0.005-0.0010.008

0.0070.0020.009

0.007-0.0030.012

0.0060.0040.0070.0020.008

0.0110.0110.011-0.0160.024

0.0050.002-0.0170.0240.0020.006

Std. dev.

0.0360.0330.037

0.0430.0420.043

0.0380.0330.039

0.0390.0250.0450.0250.044

0.1260.1510.1100.1320.120

0.1220.0450.1860.1180.0410.145

Skewness

0.520.680.43

1.141.041.18

0.37-0.090.42

1.340.551.270.581.24

0.220.040.45-0.160.54

-0.180.080.14-0.410.19-0.19

Kurtosis

4.594.094.82

4.213.824.35

3.283.183.03

5.862.954.903.144.97

3.272.783.342.833.225

6.142.773.534.143.024.49

VARHAC

0.0230.0210.020

0.0260.0390.033

0.0420.0240.030

0.0260.0390.0520.0350.052

0.1410.1520.0940.0970.119

0.1020.0420.1270.1220.0370.121

Δlog(Money Base)

CEMACPre-Post-

UEMOARegime 1Regime 2Pre-Post-

Δlog(Reserve Money)

CEMACRegime 1Regime 2Pre-Post-

UEMOARegime 1Regime 2Regime 3Pre-Post-

Notes: Pre- and Post- indicate the periods before and after the devaluation date respectively.

Table 23: Volatility Measures for Δlog(M0), Δlog(Money Base) and Δlog(Reserve Money) in CEMAC and UEMOA areas.

5. POLICY IMPLICATIONS

Monetary variability and monetary policy

The main objective of monetary policy, price level stability, has not changedover the years. However, increasing internationalisation of financial marketsimplies that both real and nominal shocks to money stock have broadened thescope of issues that must be considered when pursuing policies gearedtowards domestic price stabilisation. In addition, it has widened (and compli-cated, somewhat) the scope of policy transmission channels that may apply.20

In the Franc zone, as in any other credible monetary union, financialintegration diminishes the ability of policy makers, at the national level, topursue domestic stabilisation policies independently. The central banksdetermine interest rates for the countries within their jurisdiction, therebygenerating a limited scope for independent monetary policy by any individualcountry (see Figure 1). While there is some level of commonality in the breaksdetected for each given series in member states, it is also readily evident thatin some cases we detect breaks that are only particular to individual member


- 41 -

CEMACΔlog(M0)Pre-PostΔlog(Money Base)Pre-PostΔlog(Reserve Money)Pre-PostSeg1-Seg2

UEMOAΔlog(M0)Pre-PostΔlog(Money Base)Pre-PostSeg1-Seg2Δlog(Reserve Money)Pre-PostSeg1-Seg2Seg2-Seg3

F-test

--

1.19-

1.43-

1.221.91**

--

1.05-

3.05***3.08***

-12.06*

17.02***2.48***

Siegel-Tukey

--

0.84-

0.80-

0.992.21**

--

0.193-

1.73*1.58

-5.04***4.62***2.32**

Barlett.

--

0.49-

2.00-

0.677.57*

--

0.047-

17.073***19.47***

-66.89***81.62***9.32***

Browne-Forsythe

--

0.0004-

1.3992-

0.63596.8628*

--

0.880-

4.9067**5.0305**

-23.8856***30.0182***4.5357**

t-test

-1.527

-2.300**

-1.851***0.030

-0.8627

-0.8633-0.5352

-0.21680.774

-1.3351

Levene

--

0.002-

1.608-

0.5847.031*

--

0.153-

6.55**7.346***

-24.25**32.24***5.004**

Notes: Columns 2-5 re[resnt tests for difference in variance. Column 6 (t-test) is a test fordif-ference in means. *,**,*** indicates rejection of the null of no significant differences in thevariance (or mean) of the two indicated segments at the 10%, 5% and 1% levels of significancerespectively.

Table 24: Robustness Tests for separate segments in CEMAC and UEMOA

states. This result suggests that for some countries the centralised monetarypolicy may, indeed, be sub-optimal since the possibility of having a tailoredresponse for individual member states is limited.

Policy recommendations from analyses that ignore structural changescan be perilous (Hansen, 2001). However, there is no shortage of empiricalarticles that analyse monetary variables under the assumption of a break datedictated by some exogenous policy implementation date. The applicationshown in this paper has highlighted the degree of information loss that aresearcher can face by making such a simplifying assumption. For the Franczone, the 1994 currency devaluation is arguably the most important monetarypolicy decision in its history, and it is of little surprise that almost all articleson the macroeconomics of the zone make mention of it, in one form or anoth-er. It is also true that it is not uncommon for an a priori assumption of a sin-gle break date in the money stock of the zone to be made using the official cur-rency devaluation date. Specifically, our results have shown that substantialinformation is lost by making such an assumption. In fact, when there is noprovision for the possibility of multiple breaks, any conclusions drawn arelikely to be severely biased. Although it may be argued that the imposition


- 42 -

r_bceao

r_beac

Figure 1: BCEAO and BEAC Rediscount Rates, 1989:11-2002:09

of the breakdate averages out the structural breaks that are extracted usingdata-driven methods, the period being considered then becomes an importantinfluencing factor. In short, the scope for erroneous conclusions about thedegree of integration of the monetary series is increased by the implicit sup-pression of some breakdates.

The important implications of appropriate monetary policy, hence pricestability, requires a correct assessment of the order of integration of the realmoney balances. An inadequate assessment of the money variable willinevitably result in estimation errors, and hence inappropriate policies.21 Thebias induced by the number of breakdates that go undetected through impos-ing a single one a priori should be adequate justification for employing data-driven methods of break detection.

Monetary variability and aggregation

An important conclusion drawn from our results is the importance of theappropriate choice of target variable in monetary policy formulation. This isdue to the heterogeneity in the number and timings of detected breakdates inthe three monetary variables.

In this paper, the Money Base is an aggregation of M0 and ReserveMoney. While significant breaks (in the variance) are detected in the MoneyBase series (Table 6), it is clear from the results reported in Tables 7 - 8 thatwhen one considers only the aggregated series (i.e. Money Base) alone, thereis significant loss of information, a result readily observable for each countryin the sample. The results, presented in Tables 6 - 8, therefore offer some sup-port for Granger’s (1980) aggregation hypothesis. Hence, in the determinationof which variables to consider in empirical analysis, the choice of variablesthat are used in research for policy making should be of primary concern forits pragmatic implementation. Choice of a given variable may imply that someinherent information may not be captured, and may not therefore attract theattention of policymakers, even though it should. For example, the use ofeither M0 or Reserve Money has uncovered more breakdates than the aggre-gated Money Base variable, suggesting that the latter variable actually ‘masks’the changes of the former ones. However, the specific purpose of the investi-gation is paramount and should be given more weight in the decision-makingprocess. In particular, when one assesses the aggregated data across the zone(i.e. UEMOA and CEMAC in Tables 22 - 24), the extent of ‘missing’ informa-tion becomes even more obvious. A casual assessment of the literature showsthat there is no shortage of studies that aggregate across the Franc zone, andwhich present policy suggestions based on their estimates. Our estimatesreveal some significant loss of information, which is particularly importantwhen one considers for example zone-wide policy impact on individualmember states.


- 43 -

6. CONCLUSIONSApplication of methods that ignore structural changes can lead to inaccurateinferences being made about economic relationships, hence in misleading pol-icy recommendations. Even when the existence of a certain break is consid-ered an unambiguous fact, it is essential for the researcher to condition analy-sis, modelling and policy recommendations on the potential existence of mul-tiple breaks of unknown timing. This paper employs a method that couldaddress such an issue, and has shows how misleading inferences can be whenbased on an imposed (known) single breakdate (in this case, an official cur-rency devaluation date) and not on data-detected breakdates. Our findingsalso highlight the importance of scrutinising for the most appropriate mone-tary variable for macroecomomic modelling since aggregate variables are like-ly to mask the true dynamics of their components.

Under the assumption that developing countries are more likely to suf-fer from misleading policy recommendations regarding monetary variables, wehave analysed three important potential monetary policy variables for thirteenFranc zone countries. We propose that break detection and timing methods,such as the one applied in this paper, are vital for correct model specification,improved analyses and policy formulation. Our results also offer some supportfor Granger’s (1980) aggregation hypothesis, as we are able to extract moreinformation on structural breaks from the disaggregated data compared to theaggregated data — irrespective of whether aggregation is done across the mon-etary variables or across member states. In view of our findings furtherresearch is warranted, in that there is a need for a country by country analy-sis in order to capture the overall effect of zonal monetary policy and drawrobust conclusions about the soundness of such decisions.

Accepted for publication: 7 May 2009

APPENDIX

A. The Franc zone

The CFA Franc zone is the oldest monetary union currently in existence, pre-dating the eurozone by many decades with the 14 member states’ currenciespegged to the euro (formerly the French Franc). The CFA Franc zone is madeup of two regions. The West-African group is referred to as the UnionÉconomique et Monétaire Ouest Africaine (hereafter UEMOA), and uses a com-mon currency Franc de la Communuaté Financière de l’Afrique (CFAF) issuedby their regional central bank (BCEAO). The Central-African group, oftenreferred to as the Communauté Économique et Monétaire de l’Afrique Centrale(hereafter CEMAC), uses a common currency called the Franc de laCoopération Financière Africaine (CFAF) issued by their regional central bank(BEAC). (The acronyms BCEAO and BEAC refer to Banque Central des Etats


- 44 -

de l’Afrique d’Ouest and Banque des Etats de l’Afrique Central respectively.)Some features of the Zone set it apart from the rest of Africa’s sub-region, par-ticularly the fixed parity with the euro (formerly the French Franc), the guar-anteed convertibility of the CFA Franc by the French Treasury, the pooled for-eign exchange reserves and the limits to credit to governments aimed at curb-ing government over-spending. Of relevance to this paper is the fact that themonetary and exchange rate policies are set by the two central banks, and notnational authorities. Hence, policy decisions should reflect the fact that thecollective welfare of the union supersedes the welfare concerns of an individ-ual member state.

B. The currency devaluation of 1994

Arguably, the most important policy issue in the history of the Franc zone hasbeen the one-time currency devaluation in January of 1994. Especially overthe mid-1980s through 1993, the economies of the Franc zone showed littleeconomic growth (if any) and goods produced by these countries were largelypriced out of the world market as the exchange rate for the CFA franc was arti-ficially high. As a means of rectifying the situation, after consulting with oneanother and with both the IMF and France, the member states made the bolddecision to devalue the CFA Franc (CFAF) by 50 pr cent. Indeed, on the 12thof January 1994 it became official that a 100 CFAF exchanged for 1 FrenchFranc (FF), instead of the previous rate of 50 CFAF: 1 FF. However, there is nofinal verdict yet as to whether this strategy has offered long-term gains to theeconomies of the member states.

ENDNOTES

1. Department of Economics, Newcastle University Business School, Newcastle uponTyne, NE1 4JF. Corresponding author: Simeon Coleman, Division of Economics,Nottingham Business School, Nottingham Trent University, Nottingham NG1 4BU.Email: [email protected] We gratefully acknowledge valuable comments onearlier versions of this paper from colleagues, anonymous referees and editors of thisjournal for helpful comments. All remaining errors are ours.

2. An idea popularised in the famous Nobel lecture by Friedman (1977).

3. Changes in the political environment, the Terms of Trade, perceptions of non-inde-pendence in the judicial system and institutions, the security of investments, and thecredibility of policy announcements may well lead to huge capital inflows/flight.

4. See Caporale (1993) and references therein for a review of some studies on bothsides of the debate.

5. For economies with such characteristics, manipulating M0, which although not


- 45 -

under the direct control of Central Banks can be influenced through conventional pol-icy instruments, remains a potentially effective monetary policy strategy.

5. Nevertheless, it is relatively straightforward to condition the analysis on observablesas well, although beyond the scope of this paper.

7. Reserve Money being a result of policy on Reserve requirements. All three variablesare, at least directly or indirectly, available to the Central Banks as policy instrumentsfor each of the countries in our sample.

8. Some early research articles include Dornbusch, 1973; Obstfield, 1986; Hossain,2005.

9. For the interested reader, Appendices A and B provide a brief background to the CFAFranc zone and the 1994 currency devaluation.

10. A potential limitation of the procedure is that it does not cater explicitly for possi-ble cross country dependency in the variables. Moreover, our use of differenced datamay ignore common trends in the data. However, given that we use monthly data inthis paper these potential limitations are unlikely to introduce any significant bias, ifat all. We thank an anonymous referee for a comment on this.

11. With the large number of HAC estimators available to the researcher, this stageallows for some flexibility.

12. Note that in order to comply with the relevant econometric literature, we have usedthe terms date and breakdate to refer to the month and break-month that correspondto our sample frequency.

13. Sansó, Aragó, and Carrion, (2003) derive some of the asymptotic properties of I&T,SAC1, and SAC2 for the various levels of kurtosis while Andreou and Ghysels (2002)provide some simulation evidence for I&T and K&L using a number of GARCH(1,1)DGPs. See also Karoglou (2006c).

14. This algorithm has been found to be more robust to the existence of transitionalperiods, which are particularly relevant to the series of developing economies, whencompared to the ICSS algorithm of Inclan and Tiao (1994), since it imposes the detec-tion of breakdates to be ordered in time (and therefore it can avoid possible maskingeffects due to the presence of transitional periods).

15. Naturally, an even more robust procedure would require testing the statisticalequivalence of higher moments as well. However, the focus of this paper is the first twomoments only and therefore this assumption is actually weak.

16. All tests have been carried out using EViews.

17. This period is particularly significant, as it includes the two main periods likely tointroduce significant breaks in the series - the 1994 devaluation and the pegging of theCFAF to the euro in 2001.

18. Although this measure makes the link between M0 and prices in country i lessstraightforward, the analyses and policy deductions made in this paper are still validwith this measure of M0.


- 46 -

19. See Appendix A for a brief description of the background of the monetary groupsand their central banks, often referred to by the acronyms BCEAO and BEAC.

20. A thorough review of the administrative and constitutional arrangements thatunderpin the entire Franc zone can be found in some previously published literatureincluding Fielding (2002).

21. The well documented events of the Asian crisis of 1998/9, and subsequent effectsof contagion, underscore the potential deleterious consequences of inadequate adjust-ment of monetary policy in such situations.

22. Of course, the implications of this result is by no means restricted to the Franczones. However, the limitations on country-level monetary policy formulation in themonetary union, and the typical imposition of the breakdate of January 1994 in empir-ical work, make it particularly relevant.

REFERENCES

Andreou E and Ghysels E (2002) ‘Detecting Multiple Breaks in Financial MarketVolatility Dynamics’, Journal of Applied Econometrics. 17, 579-600.

Barro R J (1976) ‘Rational Expectations and the Role of Monetary Policy’, Journal ofMonetary Economics. 2, l-32.

Brown M B and Forsythe A B (1974), ‘Robust Tests for the Equality of Variances’,Journal of the American Statistical Association, 69, 364-367.

Caporale T (1993) ‘Monetary Variability and Real Output’, Economics Letters, 42, 219-222.

Devereux M (1987) ‘The Effects of Monetary Variability on Welfare in a SimpleMacroeconomic Model’, Journal of Monetary Economics, 19, 427-435.

Dornbusch R (1973) ‘Devaluation, Money, and Nontraded Goods’, American EconomicReview, 63(5), 871-880.

Fielding D (2002) The Macroeconomics of the Monetary Union, An Analysis of the CFAFranc Zone, London: Routledge.

Fielding D (2004) ‘How Does Monetary Policy Affect the Poor? Evidence From the WestAfrican Economic and Monetary Union’, Journal of African Economies, 13(4), 563-593.

Friedman M (1977) ‘Nobel Lecture: Inflation and Unemployment’, Journal of PoliticalEconomy, 85, 451-472.

Granger C W J (1980), ‘Long Memory Relationships and the Aggregation of DynamicModels’, Journal of Econometrics, 25, 227-238.

Gray J A (1976) ‘Wage Indexation: A Macroeconomic Approach’, Journal of MontaryEconomics, 2, 221-235.

den Haan W J and Levin A (1997), ‘Vector Autoregressive Covariance MatrixEstimation’. Manuscript, University of California, San Diego.


- 47 -

Hansen B E (2001) ‘The New Econometrics of Structural Change: Dating Breaks in USLabor Productivity’, Journal of Economic Perspectives, 4, 117-128.

Hossain A (2005) ‘Granger-Causality Between Inflation, Money Growth, CurrencyDevaluation and Economic Growth in Indonesia, 1951-2002’, International Journal ofApplied Econometrics and Quantitative Studies. Euro-American Association of EconomicDevelopment, 2(3), 45-68.

Inclan C and Tiao G C (1994) ‘Use of Cumulative Sums of Squares for RetrospectiveDetection of Changes of Variance’ Journal of the American Statistical Association, 89,913-923.

Judge G G, Griffiths W E, Hill R C, Lutkepohl H, and Lee, T C(1985) The Theory andPractice of Econometrics, 2e, New York: John Wiley and Sons.

Karoglou M (2006a) ‘Breaking down the non-normality of stock-returns’, Mimeograph,University of Leicester, UK.

Karoglou M (2006b) ‘On the Detection of Structural Changes in Volatility Dynamicswith Applications’, PhD. Thesis, University of Leicester, Leicester, UK.

Karoglou M (2006c) ‘The Size and Power of the CUSUM-type Tests in DetectingStructural Changes in Financial Markets Volatility Dynamics’. Mimeograph,University of Leicester, UK.

Kokoska P and Leipus R (2000) ‘Change-point Estimation in ARCH Models’, Bernoulli6.

Lee S, Tokutsu Y and Maekawa K (2003) ‘The Residual Cusum Test for the Constancyof Parameters in GARCH (1,1) Models’. Working papers at Hiroshima University.

Levene H (1960) ‘Robust Tests for Equality of Variances’, in I Olkin et al (eds.)Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling.Palo Alto,Cal: Stanford U P, 278-292.

Obstfeld M (1986) ‘Capital Controls, the Dual Exchange Rate, and Devaluation’,Journal of International Economics, 20, 1-20.

Perron P (1990) ‘Testing for a Unit Root in a Time Series with a Changing Mean’,Journal of Business and Economic Statistics, 8(2), 153-162.

Sansó A, Aragó V and Carrion J L (2003) ‘Testing for Changes in the UnconditionalVariance of Financial Time Series’. Department d’Economia Aplicada Working Paper,Universitat de les illes Balears, Spain.

Sheskin D J (1997) Handbook of Parametric and Nonparametric Statistical Procedures,Boca Raton, FL: CRC Press.

Siegel S and Tukey J W (1960) ‘A Nonparametric Sum of Ranks Procedures for RelativeSpread in Unpaired Samples’, Journal of the American Statistical Association. 55, 429-444.

Sokal R R and Rohlf F J (1995) Biometry: The Principles and Practice of Statistics inBiological Research, 3e. New York: W. H. Freeman and Co.


- 48 -

Documents

Monetary Variability and Monetary Variables in the Franc Zone · 2017. 2. 7. · Monetary Variability and Monetary Variables in the Franc Zone Simeon Coleman and Michail Karoglou1