Chapter

3 Long-Run Determinants of Inflation in WAEMU

Author(s):
Anne Gulde, and Charalambos Tsangarides
Published Date:
April 2008
Share
  • ShareShare
Show Summary Details
Author(s)
Abdoulaye Diop, Gilles Dufrénot and Gilles Sanon 

In this chapter, we propose an empirical model of the long-run determinants of the inflation rate in WAEMU member countries. Understanding the economic processes that drive inflation is important because prices are an element of the competitiveness of economies. Using an econometric test that allows us to select appropriate macroeconomic variables that induce inflationary pressures, we find that in the WAEMU countries money supply is not the sole determinant of inflation in the long run. Other factors such as supply-side constraints and pass-through effects also play a significant role.

We document the impact of such variables on long-run inflation by proposing an empirical model of determinants of inflation in the WAEMU countries. We employ the Pesaran, Shin, and Smith (2001) (PSS) approach, which yields conclusions that are more robust than the traditional Engle-Granger approach, especially when applied to samples as small as ours.

The chapter first presents the background for our arguments. Next, it describes the methodology and data, followed by the main empirical results. The final section presents the conclusions.

The authors are with the West African Economic and Monetary Union (Ouagadougou) and ERUDITE (Paris XII), GREQAM (Marseille), and the WAEMU Commission, respectively. They gratefully acknowledge comments from Bruno Cabrillac, Anne-Marie Gulde, Joachim Ouadraogo, Charalambos Tsangarides, and two referees. The views expressed in this chapter are those of the authors and do not necessarily represent those of their institutions.

Background

Whether or not an increase in the money supply is a major cause of inflation in sub-Saharan African countries is still debated in the literature1.Chhibber and Shafik (1990) and Sowa and Kwakye (1991) find that money expansion is the backbone of long-term price changes in Ghana. Elbadawi (1990) gets the same result for countries in Southern Africa. The findings from other studies are mixed: they also find significant roles for such macroeconomic fundamentals as exchange rates, real income, nominal interest rates, foreign prices, real money, and output gaps (see, for instance, London, 1989; Atta, Jefferis, and Mannathoko, 1996; and Barnichon and Peiris, 2007).

In this chapter we focus on the WAEMU countries. For several reasons we believe that long-run inflation is not solely a monetary phenomenon in these countries:

  • (1) These countries are members of the CFA franc zone.2 Two important characteristics of this zone are the free convertibility of the CFA franc into euros guaranteed by the French Treasury, and an open capital account with no financial repression between the African countries and France. Theoretically, in this context a domestic expansion of credit mainly affects the balance of payments, with a weak effect on money and domestic inflation. This is indeed the case with regard to demand for tradable goods: excess demand spills over into higher demand for foreign manufactured goods. The implication is that in the WAEMU countries the pass-through effects of foreign prices are substantial; they are channeled by the prices of manufactured goods imported from European countries, which account for 60 percent of WAEMU trading volume. Historically, average inflation in the union closely tracks inflation in Europe, especially in France. Moreover the central bank for the union has no way of reacting when the U.S. dollar appreciates or depreciates against the euro, because the exchange rate of the CFA franc is fixed against the euro3. Thus, WAEMU countries also face exchange rate pass-through effects when importing from dollar zones.
  • (2) The CFA franc zone effectively separates monetary from fiscal policy. Indeed, to gain credibility, the central bank has avoided implementing expansive monetary policies in response to national fiscal deficits. The agreements with the French Treasury impose a cap on the amount of credit extended to each country of 20 percent of the country’s public revenue in the preceding year. As a consequence, inflation in the franc zone is not likely to come from monetization of fiscal deficits.
  • (3) Inflation accounts for supply-side constraints in the agriculture sector. This happens not only through supply-side shocks, such as rainfall or locust invasion, but also through the permanent component of food production. Indeed, the trend appreciation in the consumer price index (CPI) is associated with low productivity in the primary sector; food items comprise nearly 70 percent of the total weight of the CPI.

As a consequence of factors like these, a suitable model for long-run inflation in the WAEMU countries must go beyond the quantity theory vision, mainly because inflation reflects the economic structures and the institutional framework in which monetary policy operates. The operations account mechanism and free capital movements in the franc zone imply that domestic credit policy does not necessarily have lasting effects on money supply and thus on inflation4. We expect the following variables to be determinants of the long-run inflation rate, along with nominal money: (1) food production or any variable capturing supply-side constraints in locally produced goods sectors and (2) the exchange rate pass-through and foreign prices.

Temporary deviations from the long-run level may also be observed due to demand or supply-side shocks related to the same variables (for example, droughts, differences in pass-through of imported inflation to nontraded goods prices, wage push).

Methodology and Data

Model Specification

Inflation dynamics is modeled as an error-correcting mechanism, where it is assumed that the price level returns to its long-run equilibrium when the effects of transitory shocks have dissipated:

Equation (3.1) shows that the price change depends on the transitory shocks affecting prices and their determinants and on an innovation reflecting the discrepancy of the prices from their target level in the preceding period. pt* denotes the long-run price, and we assume that

where ωt is an independently identically distributed (iid) error term; yts represents the volume of locally produced goods and accounts for the impact of supply-side constraints on domestic prices; st is the nominal exchange rate against foreign currencies; pforeignt denotes the foreign prices; and DUMt is a vector of dummies that captures the effects of exogenous shocks (considerable devaluation and sociopolitical factors).

We expect the following signs on the coefficients: β1, >0, β2<0, β3>0, and β4>0. Indeed, the reduced-form equation (3.2) is representative of different approaches to inflation. The choice of money as an explanatory variable is compatible with the monetarist view that inflation is a monetary phenomenon. Usually, the output variable is present in Keynesian models, where inflation is attributable to demand pressures. In this case, output variables measure an output gap, and one would expect a positive sign on the output coefficient. Yet in developing countries, more specifically in small economies with high trade openness, inflationary pressures owing to domestic demand are attenuated by high demand for tradable goods. Supply-side constraints affect domestic prices more significantly because of the low productivity of locally produced goods sectors. We thus expect the coefficient of the output to carry a negative sign, indicating that a decline in the value added of the primary sector would appear to constrain the supply side of the economy and consequently raise the price level. In the WAEMU countries, imported goods account for a sizable share of domestic demand. As a consequence, pass-through effects on inflation must also be considered. These effects are captured by the nominal exchange rate and foreign price variables. An increase in the nominal exchange rate (a depreciation of the local currency) increases the prices of the imported goods. Similarly, higher foreign prices are transmitted to domestic prices.

The Econometric Methodology

Our results draw on the bound-testing approach of level relationships proposed by PSS. This approach has an advantage over the Engle-Granger methodology because it makes it possible to test for the existence of a long-run relationship between a dependent variable and a set of explanatory variables when it is not clear whether the regressors are I (1) or I (0). The Engle-Granger cointegration test assumes that all variables are I (1), but, as is generally known, the unit root often yields mixed results, notably when they are applied to small samples. For purposes of clarity, let us summarize the PSS methodology.

Define Δyt,Δpt,xt=(mt,yts,st,pforeignt) the tth-row of the matrix of the explanatory variables and zt=(yt,xt).

We consider the following conditional error correction model (ECM) equations with unrestricted intercept and trend:

where ut is an error term.5 The procedure amounts to testing two assumptions: (1) the dependent variable has a unit root and (2) there is a relationship in levels between the dependent variable y and the independent variables xt-1. The test is formulated as follows:

The restricted model is tested by computing a Fisher statistic, as is usually done to test restrictions in an econometric regression. However, when the regressors consist of a mixture of I (0) and I (1) variables, the asymptotic distribution of the Fisher statistic is nonstandard under the null. The Monte Carlo simulations done by PSS yield two critical values, say, L1 (for the lower bound) and L2 (for the upper bound) that provide a range covering all possible configurations of the regressors into purely I (0), purely I (1), or mutually cointegrated. The conclusions are as follows:

  • If the computed F-statistic lies below the lower bound, there is no level relationship between the variables.
  • If the computed statistic is higher than the upper bound, the null of no level relationship between the variables is rejected.
  • If the statistic lies between the two bounds, we cannot reach any conclusion about cointegration.

PSS show that the critical bounds vary with the specification of the deterministic components—the constant term and the trend in equation (3.3). In particular, it is necessary to take into account the possibility of level relationships between the deterministic components of the variables. Such relationships yield some restrictions in the regressions. PSS accordingly distinguish five cases: no intercepts and no trends, restricted intercepts and no trends, unrestricted intercepts and no trends, unrestricted intercepts and restricted trends, and intercepts and trends that are both unrestricted.

In practice, the test is implemented in several steps.

Step 1. We estimate equation (3.3) to obtain a parsimonious specification. This means that we need to find the optimal lag p, select the variables that enter the regression with a significant coefficient, and apply some misspecification tests on the residuals. In this effect the optimal lag p is selected according to the usual information criteria (Akaike, Schwarz) and the following misspecification tests are applied on the estimated residuals: the Durbin-Watson test for first-order autocorrelation, the Breusch-Godfrey serial correlation LM test, the Jarque-Bera tests for normality, and the ARCH LM test.

Step 2. We use the model selected in step 1 to apply the PSS test. At this step several versions of the model are estimated by taking into account different specifications of the deterministic components if a constant, a trend, or both are significant in the regressions.

Data

Our sample consists of the following WAEMU countries: Benin, Burkina Faso, Côte d’Ivoire, Mali, Niger, Senegal, and Togo. The countries and the years were selected according to data availability. Data are not available for Guinea-Bissau before the mid-1990s.

Data are collected from the WAEMU Commission for 1970 to 2005 and from the IMF’s International Financial Statistics (IFS) database. We consider both supply-side (notably foreign price pass-through and output) and demand-side factors (such as the money growth). The definitions of the different variables used are the following:

  • pt: consumer price index (at 1985 = 100);
  • mt: for robustness, we consider three alternative definitions of money, namely, the monetary base, M0; the money aggregate, M1; and broad money, M2;
  • pforeignt: index of prices of manufacturing goods imported from France. This variable is chosen as a proxy for foreign prices. Indeed, because imports into WAEMU countries from the euro area represent on average 70 percent of their total imports, WAEMU area inflation closely tracks inflation in European countries. Choosing a trade-weighted average of all European manufacturing prices does not modify our results, because France is the main European trading partner for WAEMU countries;
  • st: nominal exchange rate of the CFA franc against the U.S. dollar; and
  • yts: value added in the primary sector (at 1985 constant prices and expressed in billions of CFA francs).

All variables are transformed in logarithmic form6.

Empirical Results

Selection of the Long-Run Models

Table 3.1 gives the F-statistics for testing the existence of a long-run inflation equation under different scenarios for the deterministic components (constant and trends). These statistics should be compared with the critical value bounds, L1 and L2. This step is very important, because it avoids estimating spurious regressions by using the standard Engle-Granger (EG) methodology, especially when there is doubt about the degree of integration of the explanatory variables (in our case, the latter are a mix of I (0) and I (1) variables).7

Table 3.1.Test of the Existence of a Long-Run Relationship Between Inflation and Its Determinants
Money IndicatorCountryModelF-statL1L2ConclusionLong-Run Determinants of the CPI
Monetary baseBenin(5)1.0411.6411.64No long-run relationship
(4)1.166.296.29No long-run relationship
Burkina Faso(3)5.454.945.73Inconclusive
(2)6.943.624.16Long-run relationshipst
Côte d’Ivoire(5)8.854.015.07Long-run relationshipM0t, pforeignt, st
(4)5.173.384.23Long-run relationshipM0t, pforeignt, st
Mali(5)4.064.015.07Inconclusive
(4)3.964.685.15No long-run relationship
Niger(1)2.553.154.10No long-run relationship
Senegal(3)10.593.234.35Long-run relationshipst, pforeignt
(2)5.202.793.67Long-run relationshipst, pforeignt
Togo(3)20.463.234.35Long-run relationshipyt, st, pforeignt
(2)17.322.793.67Long-run relationshipst, pforeignt
M1Benin(5)39.894.015.07Long-run relationshipyt, st, pforeignt
(4)19.723.884.61Long-run relationshipyt, pforeignt
Burkina Faso(5)17.974.015.07Long-run relationshipM1t, yt, st
(4)7.783.384.23Long-run relationshipM1t, yt, st
Côte d’Ivoire(3)13.553.794.85Long-run relationshipM1t, pforeignt
(2)6.693.103.87Long-run relationshipM1t, pforeignt
Mali(1)4.382.453.63Long-run relationshipyt, pforeignt, st
Niger(3)3.602.864.01Inconclusive
(2)1.732.563.49No long-run relationship
Senegal(1)23.582.723.83Long-run relationshipst, pforeignt
Togo(1)20.353.154.11Long-run relationshippforeignt
M2Benin(4)15.833.384.23Long-run relationshipyt, st, pforeignt
(5)25.824.015.07Long-run relationshipyt, st, pforeignt
Burkina Faso(3)7.263.234.35Long-run relationshipM2t, yt, st
(2)7.043.624.16Long-run relationshipM2t, yt, st
Côte d’Ivoire(3)16.374.945.73Long-run relationshipM2t
(2)56.003.634.16Long-run relationshipM2t
Mali(5)6.394.875.85Long-run relationshipM2t, pforeignt
Niger(1)2.552.723.83No long-run relationship
(2)3.103.103.87No long-run relationship
Senegal(1)23.972.453.63Long-run relationshipM2t, pforeignt
Togo(1)18.902.723.85Long-run relationshipyt, pforeignt
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. F-stat = Fisher statistic, L1 = Lower bound, L2 = Upper bound, st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M0t = monetary base. The explanatory variables were selected using the criterion in Table A3.1 and by reestimating the equations until all the regressors were statistically significant.
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. F-stat = Fisher statistic, L1 = Lower bound, L2 = Upper bound, st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M0t = monetary base. The explanatory variables were selected using the criterion in Table A3.1 and by reestimating the equations until all the regressors were statistically significant.

The results show the worth of considering alternative indicators of money: the conclusions vary according to whether M0, M1, or M2 is used. With the monetary base, a long-run inflation equation is found for four countries and money has an impact on long-run inflation only in Côte d’Ivoire. With M1, the test concludes in favor of a long-run relationship for five countries and money has a significant effect on inflation in Burkina Faso and Côte d’Ivoire. With M2, there is a long-run inflation equation for all the countries except Niger and the impact of money is significant for Burkina Faso, Côte d’Ivoire, Mali, and Senegal.

The fact that the effect of money on long-run inflation increases when a broad money aggregate is used indicates that monetary policy affects inflation through the behavior of economic agents. The extent of the effect depends upon the desire of commercial banks to expand credit and upon broad money demand. Care must be taken to avoid choosing a too narrow definition of the monetary aggregate.

What Are the Determinants of Long-Run Inflation?

The conditional ECM regressions associated with the long-run relationships are given in Tables 3.2, 3.3, and 3.4. In Table 3.5, we give the values of the long-run coefficients. They are computed by dividing the elasticities of the variables expressed in level (with one lag) by the absolute value of elasticity of the lagged CPI. Table A3.1 shows the results of misspecification tests on the residuals.

Table 3.2.ECM Inflation Equations with the Monetary Base
Burkina FasoCôte d’IvoireSenegalTogo
(2)(3)(4)(5)(2)(3)(2)(3)
CPIt-1-0.11-0.12-0.47-0.17-0.24-0.25-0.59-0.51
(-3.14)(-3.22)(-4.39)(-3.58)(-3.07)(-3.77)(-7.67)(-7.92)
M0t-10.110.12
(2.89)(2.92)
pforeignt-10.240.220.460.600.62
(3.81)(2.55)(2.74)(6.61)(8.53)
st-10.110.09-0.04-0.050.110.460.170.15
(3.72)(2.95)(-1.89)(-2.19)(4.51)(2.74)(4.32)(4.53)
yt-1-0.17
(-2.63)
dm0t0.080.11
(1.76)(2.25)
dpforeignt3.112.980.971.01
(6.29)(6.01)(9.63)(1.88)
dst0.220.210.140.10.280.150.14
(6.64)(6.47)(2.77)(2.05)(4.68)(3.51)(3.92)
dyt-0.202
(-2.31)
dCPIt-10.690.24
(7.41)(1.94)
dpforeignt-1-2.592.221.22
(-5.50)(11.79)(4.19)
dst-10.060.07
(1.97)(2.21)
dyt-1-0.59
(-2.62)
constant-0.52-0.99-0.59
(-2.80)(-2.51)(-2.62)
trend0.020.01
(1.80)(2.37)
Dummies
19770.090.08
(2.15)(2.37)
1982-93-0.16
(-7.03)
19940.250.25
(6.84)(5.97)
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M0t = monetary base. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M0t = monetary base. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Table 3.3.ECM Inflation Equations with M1
BeninBurkina FasoCôte d’IvoireMaliSenegalTogo
(5)(4)(5)(4)(3)(2)(1)(1)(1)
CPIt-1-0.36-0.30-0.98-0.40-0.25-0.36-0.66-0.59-0.27
(-7.19)(-7.43)(-8.40)(-5.16)(-6.07)(-3.93)(-3.58)(-6.75)(-5.53)
M1t-10.360.140.220.18
(5.65)(2.57)(6.05)(4.30)
pforeignt-10.280.250.220.230.860.190.27
(5.31)(3.72)(3.24)(2.97)(3.27)(2.36)(5.68)
st-10.050.210.110.090.28
(1.94)(6.39)(2.99)(1.79)(7.12)
yt-1-0.44-0.53-1.70-0.29-0.25
(-4.42)(-3.73)(-7.07)(-4.68)(-2.28)
dm1t0.080.131.86
(2.42)(3.50)(3.90)
dpforeignt0.971.241.86
(3.68)(4.67)(3.90)
dst0.140.150.180.070.19
(4.74)(5.39)(4.59)(1.90)(4.25)
dyt-0.39-0.39-0.74-0.32-0.31
(-4.43)(-2.97)(-8.02)(-3.32)(-2.50)
dm1t-1-0.150.06
(-3.62)(1.85)
dpforeignt-11.600.95-1.32-1.140.50
(5.89)(5.81)(-3.44)(-2.35)(3.19)
dst-10.090.070.08
(2.78)(1.69)(1.74)
dyt-10.39
(3.51)
constant1.620.054.90-0.55
(3.22)(3.01)(3.51)(-4.02)
trend0.020.05
(3.98)(6.00)
Dummies
1977-0.15-0.13
(-5.73)(-3.38)
1982-93-0.17-0.19-0.11
(-14.48)(-8.70)(-6.30)
19940.060.160.150.19
(1.78)(4.64)(3.48)(6.11)
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M1t = money. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Source: Authors’ calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M1t = money. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Table 3.4.ECM Inflation Equations with M2
BeninBurkina FasoCôte d’IvoireMaliSenegalTogo
(5)(4)(5)(4)(3)(2)(1)(1)(1)
CPIt-1-0.40-0.41-0.71-0.79-0.21-0.20-0.45-0.39-0.32
(-6.44)(-5.20)(-5.09)(-4.80)(-5.63)(-8.12)(-4.15)(-3.75)(-6.69)
M2t-10.400.470.170.160.340.11
(5.20)(5.13)(5.16)(9.79)(3.61)(2.20)
pforeignt-10.300.420.220.22
(4.67)(6.93)(3.07)(4.04)(6.57)
st-10.080.100.180.21
(2.52)(2.39)(3.51)(3.51)
yt-1-0.47-0.16-0.56-0.64-0.13
(-4.11)(-3.40)(-3.99)(-4.07)(-4.35)
dm2t0.200.200.220.220.22
(1.81)(1.86)(3.65)(3.92)(2.32)
dpforeignt1.121.571.421.48
(3.44)(4.20)(4.18)(3.63)
dst0.110.130.230.13
(3.55)(3.71)(3.42)(3.12)
dyt-0.36-0.17
(-3.50)(-1.80)
dCPIt-1-0.27-0.330.58
(-2.22)(-2.82)
dm2t-1-0.21-0.25
(-2.00)(-2.47)
dpforeignt-10.65
(4.89)
dst-10.110.090.190.11
(2.04)(1.89)(3.24)(2.85)
dyt-10.290.35-0.26
(2.31)(3.04)(-3.05)
constant1.790.093.26-0.42-0.26
(3.02)(4.50)(4.27)(-5.35)(-2.90)
trend0.02-0.03
(3.76)(-4.04)
Dummies
1977-0.14-0.11
(-4.82)(-3.40)
1982-93-0.17-0.170.19-0.16
(-11.55)(-9.10)(3.49)(-7.49)
19940.190.170.140.140.19
(4.06)(3.88)(5.79)(5.83)(3.71)
Source: Author’s calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M2t = money and quasi-money. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Source: Author’s calculations.Note: (1) model with no intercepts and no trends, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends. st = nominal exchange rate, pforeignt = foreign CPI, yt = value added (primary sector), and M2t = money and quasi-money. The numbers in parentheses are the t-ratios of the estimated coefficients. These regressions make sense for the countries for which we conclude in favor of the existence of a long-run relationship between inflation and its determinants (see Table 3.1). When there is no long-run relationship, estimating an error correction mechanism would yield spurious regressions.
Table 3.5.Long-Run Elasticities-Dependent Variable: Consumer Price Index (CPI)
Money IndicatorBeninBurkina FasoCôte d’IvoireMaliSenegalTogo
(2)(3)(4)(5)(2)(3)(2)(3)
Monetary baseMoney0.230.69
Value added (primary sector)-0.35
Nominal exchange rate1.000.790.080.290.481.820.300.29
Foreign CPI0.510.931.811.021.21
BeninBurkina FasoCôte d’IvoireMaliSenegalTogo
(5)(4)(5)(4)(3)(2)(1)(1)(1)
M1Money0.360.350.900.50
Value added (primary sector)-1.22-1.76-1.73-0.74-0.38
Nominal exchange rate0.130.210.280.140.48
Foreign CPI0.771.200.880.640.331.02
BeninBurkina FasoCôte d’IvoireMaliSenegalTogo
(5)(4)(5)(4)(3)(2)(1)(1)(1)
M2Money0.560.590.800.800.750.29
Value added (primary sector)-1.17-0.39-0.79-0.80-0.39
Nominal exchange rate0.190.240.260.29
Foreign CPI0.741.020.450.561.34
Source: Authors’ calculations.Note: (1) model with no intercept and no trend, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends.
Source: Authors’ calculations.Note: (1) model with no intercept and no trend, (2) model with restricted intercepts and no trends, (3) model with unrestricted intercepts and no trends, (4) model with unrestricted intercepts and restricted trends, and (5) model with unrestricted intercepts and unrestricted trends.

In the restricted models, it is assumed that the constant and trend components of the explanatory variables are colinear; in the unrestricted models, they are not. The significance of the coefficients is generally robust to the assumptions made on the deterministic components, and the long-run coefficients are quite stable when M2 is used as an indicator of money (see Table 3.5).

As the tables show, money is not the sole determinant of long-run inflation. In the WAEMU countries, it appears that long-run inflation is also the result of supply-side effects, the nominal exchange rate, and the long-term dynamics of foreign prices. In terms of magnitude, we find that the effects of a monetary expansion outweigh the impact of nominal exchange rate variations, but they can be as important as foreign price effects, though not always. In all cases, money supply has a less significant effect on long-run prices than supply-side constraints, reflected here in production in the primary sector.

Inflation is partly a monetary phenomenon. Indeed, both short- and long-run coefficients are statistically significant in the regressions for Burkina Faso, Côte d’Ivoire, Mali, and Senegal—half the countries in WAEMU. We do not find a unitary relationship between money supply and the price level, as monetarist theory would predict. A 1 percent increase in broad nominal money supply leads at most to a 0.47 percent increase in price (in Burkina Faso) and an average 0.20 percent increase in short-run inflation. This partial effect of money on long-run inflation may have several elements. In some countries, other factors are at play, as we discuss below. Another explanation is the institutional environment described above.

Pass-through effects are also significant in determining domestic prices in the long run. We find significant coefficients for the nominal exchange rate of the CFA franc against the U.S. dollar. The inflationary impact is channeled through the prices of energy and manufactured goods, which in turn affect overall consumer prices. An appreciation of the CFA franc against the dollar tends to bring long-run inflation down. It first reduces the bill for imported goods and then the appreciation leads to a loss of competitiveness, especially in exports of raw materials and primary goods. This causes a decline in activity and a fall in productivity utilization, which in turn pushes down inflation.

Foreign prices show the strongest influence in many countries; their impact is statistically significant. In the long run, 50 to 70 percent of the increase in domestic prices is explained by increases in foreign prices. We even find coefficients that are statistically near 1. The influence of French prices is not surprising because France is a major WAEMU trading partner, but other foreign prices can also affect the domestic price. Their effect is channeled to long-run prices through the nominal exchange rate.

The estimate of the total pass-through effect (by combining nominal exchange rate and foreign price variations) shows definite heterogeneity across countries. When M2 is used as the money indicator, the estimated long-run elasticities for the pass-through vary between 0.22 for Mali and Senegal and 0.52 for Benin. We find stronger effects for coastal Benin and Togo than for landlocked Mali.

The volume of production in the primary sector in some countries tends to have an effect on the long-term dynamics of the CPI. As expected, supply-side constraints have no influence in the countries that have the highest capacity to expand local production, namely Côte d’Ivoire and Senegal.

Our results can be compared to other previous studies on the determinants of inflation in the WAEMU countries. To our best knowledge, the most complete study is BCEAO (2002). Their results are based on the Engle-Granger methodology. As indicated before, this methodology yields spurious regressions when there is a suspicion that the explanatory variables are not all I(1). Despite this, we compare their long-run elasticities with ours. The authors find four variables to have a significant impact on long-run inflation:

  • The money aggregate M2 has an impact on the price level in four countries—Côte d’Ivoire, Niger, Senegal, and Togo—with an average elasticity of 0.27 in WAEMU. Our estimates are higher than those of BCEAO (2002). For instance, the BCEAO finds a long-run elasticity of 0.23 for Côte d’Ivoire, whereas our regressions yield a coefficient of 0.80 for this country. Similarly, we find high significant values for Burkina Faso (0.59) and Mali (0.75), whereas the elasticities are zero in the BCEAO study. These differences may come from specification problems in their regressions, as indicated for instance by the Durbin-Watson statistics obtained for Mali and Togo (1.58 and 1.61, respectively), suggesting the presence of negative autocorrelation in the residuals).
  • Production in the agriculture sector is significant in two countries only, namely Burkina Faso and Mali, with an average elasticity of -0.12 for WAEMU. Again, our elasticities are higher than the BCEAO finds. For instance, for Burkina Faso, we find a coefficient of -0.80.
  • The price of oil is found to have an impact in five countries (the exceptions being Côte d’Ivoire and Senegal), but the average elasticity for WAEMU is small and equal to 0.04.
  • The general level of prices in France is found to have the more significant impact in six countries, with the highest average elasticity equal to 0.47.

We compute structural inflation using the estimated long-run coefficients and the trend in the explanatory variables as obtained from a Hodrick-Prescott filter. The actual and fitted structural inflations are shown graphically when M2 is the monetary aggregate used in the regressions (see Figures 3.1-3.6). The graphs describe a downward trend for prices, which is in line with the idea that the anchor to the euro area (either through the nominal exchange rate regime or the trading relationship with France) has helped the WAEMU countries to import disinflation. The domestic prices have been close to prices prevailing in Europe, though the euro and WAEMU areas have had different long-term dynamics in terms of macroeconomic fundamentals (fiscal and external deficits, indebtedness, production capacities). This is in sharp contrast with the situation in West African countries where the currency regime is characterized by floating and nonconvertible currencies, as in Ghana and Nigeria, and where episodes of upward-trending inflation have been observed historically, as might be expected from the dynamics of their fundamentals.

Figure 3.1.Actual and Structural Inflation in Benin

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Figure 3.2.Actual and Structural Inflation in Burkina Faso

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Figure 3.3.Actual and Structural Inflation in Côte d’Ivoire

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Figure 3.4.Actual and Structural Inflation in Mali

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Figure 3.5.Actual and Structural Inflation in Senegal

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Figure 3.6.Actual and Structural Inflation in Togo

(In percent)

Sources: Authors’ calculations; and IMF, International Financial Statistics.

Conclusion

In this chapter, we provide new estimates of the determinants of long-run inflation in the WAEMU countries. The short period defined by our data call imposes a need to be very careful about the econometric methodology used. In particular, regressions based on the Engle-Granger approach may not be robust if there is uncertainty about how well integrated the explanatory variables are. For this reason, we rely on a newer time series approach, the PSS approach.

We present evidence that money is one determinant of price changes in the long run, but not the only one. There are other variables, such as pass-through effects. Even the impact of supply-side constraints is larger than that of money.

This finding has several policy implications. Not only can monetary policy be used to keep inflation under control, countries also have other instruments available. For instance, they can make their economies more competitive through deflation-adjustment policies.

This study could be extended in at least two directions. First, it would be interesting to do a similar exercise for the ECOWAS countries for purposes of comparison. This would raise questions not addressed in this chapter, such as the possibility of nonlinear effects from the study of larger changes in how monetary aggregates affect long-run inflation.

Second, it would be useful to conduct similar regressions by replacing the CPI variable with a core inflation variable. The core inflation measure recently suggested by the WAEMU Commission would not be easy to apply because obtaining historical values for core inflation is difficult, but measures of structural inflation can be obtained, for instance by using filters such as the Hodrick-Prescott filter or moving average filters.

Appendix. Models Selected Using Misspecification Tests on the Residuals
Money IndicatorCountryMaximum LagDWAICSBCGB(4)JBARCH (4)Trend
Monetary baseBenin01.96-3.00-2.420.183.570.48yes
(0.94)(0.16)(0.74)
Burkina Faso12.03-4.35-3.521.370.390.73no
(0.36)(0.82)(0.58)
Côte d’Ivoire12.14-3.96-3.180.500.721.50yes
(0.73)(0.69)(0.23)
Mali12.10-3.01-2.202.120.860.27yes
(0.15)(0.64)(0.89)
Niger02.00-2.74-2.190.251.380.67no
(0.90)(0.50)(0.61)
Senegal01.93-4.26-3.710.130.290.19no
(0.96)(0.86)(0.94)
Togo11.84-4.07-3.302.201.741.41no
(0.13)(0.41)(0.26)
M1Benin02.08-4.36-3.771.700.270.58yes
(0.19)(0.87)(0.67)
Burkina Faso11.98-4.62-4.061.200.491.58yes
(0.35)(0.78)(0.22)
Côte d’Ivoire11.96-3.69-3.770.931.301.01no
(0.46)(0.52)(0.42)
Mali01.94-2.21-1.720.871.560.85no
(0.49)(0.45)(0.50)
Niger01.72-3.19-2.830.382.420.42no
(0.81)(0.29)(0.79)
Senegal01.80-2.89-2.630.691.220.10no
(0.60)(0.76)(0.17)
Togo11.91-3.61-3.290.921.221.79no
(0.47)(0.54)(0.16)
M2Benin02.03-4.28-3.731.041.170.39yes
(0.41)(0.55)(0.81)
Burkina Faso11.62-3.46-2.920.673.230.60no
(0.62)(0.19)(0.66)
Côte d’Ivoire02.11-3.80-3.480.241.480.35no
(0.91)(0.47)(0.84)
Mali12.11-2.66-2.261.300.910.66yes
(0.30)(0.64)(0.62)
Niger01.83-3.07-2.580.730.430.96no
(0.91)(0.80)(0.44)
Senegal11.90-3.38-2.980.260.940.64no
(0.90)(0.62)(0.64)
Togo12.02-3.69-3.301.512.331.78no
(0.23)(0.32)(0.16)
Note: DW = Durbin-Watson, AIC = Akaike, SBC = Schwarz, GB = Godfrey-Breusch, and JB = Jarque-Bera.
Note: DW = Durbin-Watson, AIC = Akaike, SBC = Schwarz, GB = Godfrey-Breusch, and JB = Jarque-Bera.
References

    AttaJ.K.K.R.Jefferis and I.Mannathoko1996“Small Country Experiences with Exchange Rates and Inflation: The Case of Botswana,”Journal of African EconomiesVol. 5 (June) pp. 293326.

    • Search Google Scholar
    • Export Citation

    Banque Centrale des Etats d’Afrique de l’Ouest (BCEAO)2002“Evolution des Prix dans l’Union Economique et Monétaire Ouest Africaine” Note d’information (September).

    • Search Google Scholar
    • Export Citation

    BarnichonR. and S.Peiris2007“Sources of Inflation in Sub-Saharan Africa,”IMF Working Paper No. 07/32 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    BoccaraB. and S.Devarajan1993“Determinants of Inflation Among Franc Zone Countries in Africa,”Policy Research Working Paper No. 1197 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    CanettiE. and J.Greene1991“Monetary Growth and Exchange Rate Depreciation as Causes of Inflation in African Countries: An Empirical Analysis,”IMF Working Paper No. 91/67 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    ChhibberA.J.CottaniD.Firuzabadi and M.Walton1989“Inflation, Price Controls, and Fiscal Adjustment in Zimbabwe,”Policy Research Working Paper No. 192 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    ChhibberA. and N.Shafik1990“Exchange Reform, Parallel Markets, and Inflation in Africa: The Case of Ghana,”Policy Research and External Affairs Working Paper No. 427 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    ElbadawiI.A.1990Inflationary Process Stabilisation and the Role of Public Expenditure in Uganda (unpublished; Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    HonohanP.1990“Price and Monetary Convergence in Currency Unions: The Franc and Rand Zones,”Policy Research and External Affairs Working Paper No. 390 (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    LondonA.1989“Money, Inflation and Adjustment Policy in Africa: Some Further Evidence,”African Development ReviewVol. 1No. 1 pp. 87111.

    • Search Google Scholar
    • Export Citation

    LowreyM.P.1995“France and the CFA: Price and Money Supply Growth Determination: 1969-88,”Journal of African EconomiesVol. 4 (October) pp. 22542.

    • Search Google Scholar
    • Export Citation

    OdedokunM.O.1995“An Econometric Explanation of Inflation in Sub-Saharan Africa,”Journal of African EconomiesVol. 4 (December) pp. 43651.

    • Search Google Scholar
    • Export Citation

    PesaranH.M.Y.Shin and R.J.Smith2001“Bounds Testing Approaches to the Analysis of Level Relationships,”Journal of Applied EconometricsVol. 16No. 3 pp. 289326.

    • Search Google Scholar
    • Export Citation

    SowaN.K. and J.K.Kwakye1991“Inflationary Trends and Control in Ghana,”AERC Research Report No. 22 (Nairobi: AfricanEconomic Research Consortium).

    • Search Google Scholar
    • Export Citation
2The franc zone is characterized by monetary agreements between France and 15 African countries that make the CFA franc freely convertible into euros guaranteed by the French Treasury at a fixed parity. The amount that can be withdrawn is limited: the central bank must keep 50 percent of its foreign assets in an operations account and a cap is imposed on the amount of credit extended to each country (credits are equivalent to 20 percent of the country’s public revenue in the preceding year).
3The CFA franc moves up and down against the U.S. dollar in the same proportion as the euro.
4For an empirical analysis, see Honohan (1990).
5The deterministic components are introduced in several ways: unrestricted intercepts and trends and cointegrated intercepts and/or trends.
6We also tried other explanatory variables that proved not to be significant in our regressions. For instance, the price of oil was not statistically significant, which can be explained by the fact that the energy problems encountered by the WAEMU countries are recent in comparison to the 36 years under examination. The French CPI is used as a proxy of foreign prices given the structure of trade of the WAEMU countries with France (their first trade partner during the period under examination). A 1994 dummy was also included in some regressions to account for the devaluation shock. However, such a dummy was not needed across all countries because the increase in the domestic prices that followed the devaluation came from an increase in the prices of imported goods and mark-up behaviors in the domestic markets. When the foreign price channel appears to be the main cause of higher domestic inflation, the 1994 dummy variable appears to be nonsignificant in the regression because its effect is already captured by the foreign price variable.
7We tested the null of unit root using a battery of tests, including the Augmented Dickey-Fuller (ADF); Phillips and Perron (PP); the Kwiatkowski, Phillips, Schmidt, and Shin (KPSS); and Zivot and Andrews. To avoid a glut of tables, the unit root test results are not reported here but are available upon request from the authors. The results were contradictory: the KPSS tests usually led to accepting the null of no unit root, whereas the ADF and PP tests led to accepting the unit root hypothesis. To investigate the presence of structural breaks in the independent variable, we also used Zivot and Andrews’ modification of Perron’s procedure, taking potential breakpoints as endogenous. The conclusions were mixed, because the unit root hypothesis was rejected for some countries and not for others. Facing these conflicting results we cannot apply the EG methodology to test the null of no cointegration between inflation and its determinants.

    Other Resources Citing This Publication