Estimating the Size of Fiscal Multipliers in the WAEMU Area

2.1 Definitions of the Main Concepts

The government has two instruments to implement its fiscal policy: taxes and spending. Different spending programs have varying effects on real GDP. Similarly, some types of taxes have a greater effect on GDP than others do. Fiscal multipliers are an economic concept that refers to the effect of a change in fiscal (i.e., taxes) or budgetary policy on a country’s overall economic activity (Spilimbergo et al., 2009). Specifically, fiscal multipliers measure the effect that a change in tax revenues or government spending by a unit can have on an economy’s GDP. There are two main types of fiscal multipliers: the expenditure multiplier and the tax multiplier.

The spending multiplier measures the effect of discretionary fiscal policies on economic activity. The spending multiplier measures how much this initial increase in government spending ripples through the economy as a whole. We then expect positive expenditure multipliers. The idea is that an increase in government spending leads to an increase in overall demand in the economy, as businesses see increased revenue and potentially hire more workers to meet this increased demand. The tax revenue multiplier measures the effect of a change in tax on economic activity. When tax rates rise, households and businesses have less money to spend or invest, which can reduce overall demand in the economy. The tax multiplier measures how much a tax cut or tax increase can affect economic growth. We then expect negative tax revenue multipliers.

Using these concepts, economists and policymakers can assess how changes in fiscal policy may influence economic activity, unemployment, and other key economic indicators. Importantly, however, the actual effects of fiscal policies can be complex and depend on many factors, including the general economic context and how economic agents respond to tax changes.

2.2 Literature Review on Fiscal Multipliers

3.1 Econometric Model Specification

Most of the studies on fiscal multipliers have used VAR (Afonso & Sousa, 2012; Blanchard & Perotti, 2002; Burriel et al., 2010; Mountford & Uhlig, 2009) or VEC (Puonti, 2016) approaches and DSGE modeling (Coenen et al., 2012; Drautzburg & Uhlig, 2015). In this study, we also use the VEC approach on panel data to estimate the size of fiscal multipliers in the UEMOA area. VEC modeling allows accounting for both the short- and long-term relationships between variables that are integrated of order one and that are cointegrated. The VEC approach in estimating fiscal multipliers is in line with the findings of Thiombiano et al. (2022), suggesting the presence of inertia of the fiscal impulse in the WAEMU zone. This finding supports the idea that fiscal decisions are adapted not only to immediate conditions but also to past fiscal trajectories. Consequently, this reinforces the use of dynamic models for the estimation of fiscal multipliers in the WAEMU zone.

The VEC model is an extension of the VAR model of Sims (1980), which is used to analyze long-term relationships between multiple time series. The VEC model is mainly used to model nonstationary series that exhibit cointegration relationships. The general form of the panel data VEC model is obtained from the short-term dynamic equation and the cointegration equation. Following the literature on VEC models (Groen & Kleibergen, 2003; Puonti, 2016), our econometric specification of the panel VEC model of a K × 1 vector of endogenous variables Y i , t is presented in the following reduced form in equation (1):

where Δ denotes the difference operator and p is the lag order for the underlying reduced-form VAR. Γ k , for k = 1 , … , p − 1 , are matrices of parameters of dimension K × K , and α and β are both matrices of parameters of dimension K × r . The column vectors of the matrix β consist of the coefficients of the cointegration relationships or the long-run equilibrium relationships that may exist between the elements of the endogenous vector Y i , t . In a VEC model with r -cointegrating relations, one needs to impose r restrictions, usually on the most endogenous variables of the VEC model, to identify the parameters of the cointegrating vector. The matrix α contains the adjustment coefficients or the speed of convergence toward the long-run equilibrium. The vector v can be rewritten as v = α μ + γ , where μ is a r × 1 vector of parameters, representing the constant in the long-run relationship, and where γ is a K × 1 vector of parameters, representing the constant in the short-run relationship. We assume homogeneous long-run coefficients and adjustment parameters. X t represents a vector of exogenous variables with dimensions of M × 1 , where M denotes the number of exogenous variables. The matrix θ is the corresponding set of coefficients for these exogenous variables, with a dimension of K × M . The vector of reduced-form residuals, e i , t , is a K × 1 zero-mean white process with covariance matrix Σ e . These reduced-form residuals are related to the vector of uncorrelated structural shocks, u i , t , by the short-run covariance relationships e i , t = S u i , t , which is equivalent in structural VAR representation to A e i , t = B u i , t with S = A − 1 B . The structural shocks are assumed to have a unit covariance matrix Σ u = I K , which implies the following restrictions: Σ e = S S ′ . To analyze the effects of the underlying structural shocks, we need to recover the K 2 elements of S from the estimated residual covariance matrix Σ e . The symmetry of Σ e imposes K ( K − 1 ) / 2 additional linearly independent restrictions to identify the elements of S exactly and then to identify the structural shocks exactly. A and B are identified by imposing B to be a diagonal matrix and the diagonal elements of A to be one.

3.2 Data

In this section, we present the dataset we used to estimate the fiscal multipliers. For the estimation of the VEC model with aggregate public expenditure and revenue, we consider a system with three endogenous variables, namely, government expenditure, fiscal revenue, and GDP, as in the studies by Blanchard and Perotti (2002) and Mountford and Uhlig (2009). Public expenditure (G), also known as government expenditure, is expenditure made by the government of a country or administrative jurisdiction to finance various programs, services, investments, and activities that benefit the general population. The GDP, which represents the total value of all goods and services produced in an economy over a given period – typically on an annual basis – measures production. Tax revenues (T) represent the total amount of money collected by the government in the form of taxes, fees, and other fiscal levies to finance public spending.^[1] We further include in the model the quality of institutions (INSQ) and the volatility of country-specific commodity net export prices (commodity terms of trade) as exogenous control variables. Institutional quality, specifically corruption, affects fiscal multipliers (Del Monte & Pennacchio, 2020; Mendoça et al., 2021; Tiganasu et al., 2022). Furthermore, external volatility – particularly the volatility of commodity prices – complicates the implementation of fiscal policies. Indeed, governments often rely on widespread subsidies that are ineffective and financially unsustainable when faced with increases in global commodity prices. We also include in the exogenous regressors two dummy variables, corresponding to the global financial crisis of 2008–2009 (DUM2009) and to the COVID-19 crisis (DUM2020). These time dummies allow us to account for global shocks that affect all countries simultaneously.

To estimate the disaggregated multipliers, we consider two types of disaggregation of government expenditures. In the first disaggregation, government expenditure (G) is broken down into public consumption (GC) and public investment (GK), as by Ilzetzki et al. (2013). Government consumption includes all expenditures made by the government to cover costs linked to the daily operation of government administration, the remuneration of public officials, social transfers, and the purchase of capital goods and services such as education, health, etc. Public investment includes all expenditures made by the government to acquire physical or financial assets that are intended to produce long-term economic benefits. Tax revenues remain aggregated because disaggregated components are unavailable over our estimation period. In a second decomposition, government expenditure is disaggregated into HCE and NHCE, where HCE is the sum of public expenditure on education and health. We indeed assumed that human capital has two components, education and health, as in the study by Bloom et al. (2004).

Data were collected from different sources on an annual basis for a panel of the eight WAEMU member countries, namely, Benin, Burkina Faso, Ivory Coast, Guinea-Bissau, Mali, Niger, Togo, and Senegal, for the period from 1996 to 2023. The starting year of 1996 is motivated by the creation of the WAEMU area in 1994 and the data availability constraints on institutional quality, which are available only from 1996, whereas the ending year of 2023 corresponds to the most updated year. There are gaps in the governance indicators for the years 1997, 1999, and 2001. There are also some gaps in government expenditure on education. To maximize the sample size in the estimations, missing values are imputed via linear interpolation to fill gaps not larger than 3 years.

The GDP, total government expenditure, government consumption, government capital expenditure, and fiscal revenue data are taken from the Economic and Financial Database of the Central Bank for the West African Economic and Monetary Union (BCEAO). Series on government expenditures on health and education are taken from the world development indicators (WDIs) database of the World Bank. All the endogenous variables are expressed in real per capita terms by dividing nominal values by the GDP deflator and then by population size. The series for the population are obtained from the WDI. The index of the quality of institutions is a simple average of 6 indicators of governance obtained from the worldwide governance indicators database. These indicators include voice and accountability, political stability and the absence of violence, government effectiveness, regulatory quality, the rule of law, and the control of corruption. They have a mean of zero and a standard deviation of one and range from approximately −2.5 to 2.5, where higher values correspond to better governance. The volatility of country-specific commodity terms of trade in a given year is measured by the standard deviation of the first log difference year over year of the monthly commodity price index in the year. This latter is constructed via individual commodity prices weighted by the average ratio of net exports to GDP over the preceding 3 years. Data on the commodity price index are taken from the IMF commodity database.

3.3 Descriptive Statistics

Summary descriptive statistics, including the mean, maximum, minimum, and standard deviation, of the main variables used in the study are reported in Table 1. The mean, maximum, and minimum highlight the overall behavior of the variables, whereas the standard deviation reflects the degree of heterogeneity between countries.

The WAEMU countries exhibit significant heterogeneity in terms of economic variables when adjusted per capita. Between 1996 and 2023, the average real GDP per capita stood at 428,396 CFA francs, with a standard deviation of 204,913 CFA francs, reflecting notable disparities across member countries. The values ranged from a minimum of 191,089 CFA francs to a maximum of 1,086,858 CFA francs, highlighting a concentration of wealth in certain states. Similarly, total public expenditure per capita averaged 78,706 CFA francs, with a standard deviation of 42,136 CFA francs, ranging from 20,555 CFA francs to 233,783 CFA francs, underscoring differences in national budgetary priorities.

The public consumption expenditure (GC) per capita averaged 52,358 CFA francs, ranging from 14,122 CFA francs to 159,701 CFA francs, whereas the public investment expenditure (GK) averaged 26,348 CFA francs, with values spanning from 2,216 CFA francs to 79,914 CFA francs. These figures indicate that public consumption tends to outweigh public investment, although there are significant disparities between member states. In terms of tax revenue per capita (T), the average was 49,350 CFA francs, with wide variation ranging from 3,043 CFA francs to 142,874 CFA francs, reflecting unequal fiscal capacities and revenue mobilization efforts across the region.

Public expenditure on human capital also reveals notable inequalities. The average per capita public expenditure on human capital (HCE) was 20,797 CFA francs, with values ranging from 6,161 CFA francs to 55,276 CFA francs, whereas the average NHCE was 63,207 CFA francs, varying from 18,695 CFA francs to 180,264 CFA francs. With respect to institutional performance, the quality of institutions (INSQ) had an average score of −0.614, with values ranging from −1.441 to 0.043, indicating generally low institutional quality perceptions across the region. In addition, Commodity Terms of Trade Volatility (CTOT) averaged 0.053, with a standard deviation of 0.033. The values ranged from 0.012 to 0.175, highlighting the vulnerability of WAEMU countries to fluctuations in international commodity prices. This volatility presents a challenge to macroeconomic stability and fiscal planning, especially for countries heavily reliant on commodity exports.

3.4 Preliminary Tests

In this section, we carry out a number of preliminary tests essential for achieving adequate econometric modeling and to avoid the trap of spurious regression. This involves testing the presence of unit roots for each variable in the model and the presence of a cointegration relationship among the vector of endogenous variables before proceeding to estimate the model.

4.1 Results with Aggregate Public Expenditure

As the cointegration test indicates one cointegrating relation, the cointegrating vector is identified by normalizing to one the coefficient of the variable assumed to be the most endogenous (here, the GDP). The estimates for the VEC model with aggregate public expenditure are presented in Table 4. Although this model includes several equations, only the long-term and short-term coefficients of the first equation, for which real GDP is the dependent variable, are analyzed and interpreted. This equation provides the different elasticities of production with respect to public spending and taxes.

The first part of Table 4 presents the estimates of the long-run relationship among the vector of endogenous variables, whereas the second part presents the short-run estimates from the VEC model. Diagnostic tests were performed to check the validity of the VEC(1) estimation with one cointegrating relation. In particular, the adjusted Ljung–Box portmanteau tests for white noise in the estimated vector of residuals of each individual in the panel result in p values above 10%, indicating that there is no residual autocorrelation until lag 3 at the 5% significance level in the estimated model. As an additional validity check, the estimation results show that the coefficients of the error correction term have the expected signs.

To compute the fiscal multipliers from the estimated VEC model, we follow Blanchard and Perotti’s (2002) identification strategy to recover from structural shocks. The structural shocks u t = ( u t g , u t t , u t y ) ′ are assumed to be related to the reduced-form estimated residuals e t = ( e t g , e t t , e t y ) ′ through equation (2):

where the upper scripts g , t , and y denote shocks or residuals related to government spending, tax revenue, and GDP, respectively. The matrix form of the previous equation is expressed in equation (3):

Furthermore, Blanchard and Perotti (2002) assume that (i) the government does not change spending in response to GDP within the year, which implies that a 1 = 0 , and (ii) government spending decisions are made before those on taxation, which implies that a 2 = 0 and b 2 are estimated.^[3] The parameter b 1 represents the elasticity of taxes with respect to GDP. Following Blanchard and Perotti (2002), this elasticity is estimated outside the system and corresponds to the lag 0 estimated coefficient from a regression of changes in fiscal revenue on the first lead and lags 0 and 1 of changes in GDP. We then estimate the value of b 1 = 2.62 .

Following the usual practice in the literature (Ilzetzki et al., 2013; Mountford & Uhlig, 2009; Puonti, 2016; Woldu & Szakálné Kanó, 2023), the fiscal multipliers are defined as the ratio of a change in GDP to a discretionary and unanticipated change in the government expenditure of tax revenues. More precisely, we first compute the response of the GDP, ΔGDP t + h , government expenditure, Δ G t + h , and government fiscal revenue, ΔGDP t + h , h periods ahead from t to a given fiscal shock (government expenditure shock or government revenue shock). Then, for each fiscal variable s t = G t , T t , we compute the cumulative multipliers over h -year horizons as follows in equation (4):

The impact multiplier is obtained when h = 0 . As the variables used in the estimations are expressed in logarithms, these initial ratios, which are elasticities, are then rescaled by the sample average of the level of output to the level of government expenditure (revenue) to convert percentage changes in currency units. Table 5 reports the impact multipliers and the cumulative multipliers computed from 1- to 5-year horizons. The short-term multipliers refer to fiscal multipliers upon the impact and over a 1-year horizon, whereas the long-term multipliers refer to multipliers in 3-year and longer horizons. The multipliers were computed via the average point estimates of the impulse responses.

The estimated sizes of the fiscal multipliers are presented in Table 5. These results show that a positive shock to government spending has a positive effect on GDP in both the short (upon the impact and 1-year horizon after the shock) and long run (3-year and longer horizons after the shock). The government expenditure multipliers in the short and long run are 0.545 and 0.461, respectively. In other words, an unanticipated increase of one additional unit of government spending in WAEMU countries leads, on average, to increases in GDP of 0.545 units and 0.461 units, respectively, in the short and long terms. These results consistently indicate that our research that government spending shocks have a positive effect on output is validated. Our results are qualitatively consistent with the studies of Blanchard and Perotti (2002) and Mountford and Uhlig (2009), since the response of GDP has a similar pattern. However, the size of the government expenditure multiplier is less than the one and is smaller than that in Blanchard and Perotti (2002) in the case of the United States.

With respect to the estimated fiscal revenue multiplier, positive tax shocks have a negative effect on GDP, with a multiplier of −0.465. The effects remain negative over the long run, with a 3-year multiplier equal to −0.639. This implies that a deficit-financed tax cut policy will have a positive effect on output. The peak of the multiplier is reached in the first year after the initial shock. The results also support our research hypothesis that positive tax shocks have a negative effect on GDP. These findings corroborate the crowding-out effect of tax increases reported by Afonso and Kazemi (2005), Alesina and Ardagna (2005), and Tanzi and Schuknecht (1997). Indeed, tax increases could reduce household consumption and private business investment, which is an important driver of economic activity. In this specification, there is no clear evidence that the spending multiplier is larger than the tax multiplier, as suggested by traditional Keynesian theory.

Our results are consistent with those of other studies conducted in OECD areas and Africa. For instance, Perotti (2004) and Ramey (2011) show that positive public spending shocks have a significant effect on output in OECD countries, although the size of the multipliers may exceed unity in these economies. In Africa, the work of Ilzetzki et al. (2013) indicates that public spending multipliers are generally weaker due to structural constraints such as economic informality and institutional limitations. Our findings on tax revenues corroborate those of Tanzi and Schuknecht (1997), who find that tax increases have a negative effect on output, especially in developing economies, due to their impact on household consumption and private investment.

We check the robustness of the previous results to alternative values of tax elasticity with respect to GDP. Following Blanchard and Perotti (2002), we re-estimate the multipliers for two other different values of tax elasticity b 1 : the baseline value (2.62) minus 0.5 and the baseline value (2.62) plus 0.5. The resulting multipliers have the same signs as the previous one and are not substantially different in magnitude.

To provide better policy implications, we also perform estimations with disaggregated public expenditure to obtain the size of the multipliers associated with the components of public expenditure, such as consumption and investment.

4.2 Results with Disaggregated Government Expenditure

This section presents the estimation results with disaggregated government expenditures. The analysis is divided into two parts to capture different dimensions of fiscal policy effects. First, we examine the results by distinguishing government expenditures into consumption and investment categories. Second, we explore the effects by further disaggregating government spending into human capital and non-human capital.