A. Remittances and Macroeconomic Volatility

We assess the correlation between business cycle volatility and remittances on a sample of 27 African countries with yearly data in 1980-2005. Table 1 provides some insightful statistics about business cycle fluctuations and remittances in these economies. Consumption appears to be more volatile than output, a stylized fact established in developing countries literature on business cycle. The growth rate of households’ consumption is 1.6 more volatile than the growth rate of GDP (1.62 for HP filter variables). For all countries, consumption fluctuations are significantly much higher than output fluctuations.⁴ Figure 3 hints at the volatility-remittances nexus and indicates that output and consumption fluctuations are negatively correlated with the size of remittances. The following cross section model that captures the impact of remittances on macroeconomic volatility is estimated

Table 1:

Business cycle fluctuations and remittances

Source: African Development Indicators and Author calculations.

Table 1:

Business cycle fluctuations and remittances

	Real GDP	Consumption	Remittances
	Growth	Growth	(% of GDP)
Mean	3.2	3.0	3.3
Volatility	5.4	8.7	3.7
Relative volatility	…	1.6	0.7
Correlation with Real GDP growth	…	0.39	0.04
	…	(0.00)	(0.2)
Correlation with Remittances	0.0	0.02	…
	(0.2)	(0.5)	…

Source: African Development Indicators and Author calculations.

View Table

Table 1:

Business cycle fluctuations and remittances

	Real GDP	Consumption	Remittances
	Growth	Growth	(% of GDP)
Mean	3.2	3.0	3.3
Volatility	5.4	8.7	3.7
Relative volatility	…	1.6	0.7
Correlation with Real GDP growth	…	0.39	0.04
	…	(0.00)	(0.2)
Correlation with Remittances	0.0	0.02	…
	(0.2)	(0.5)	…

Source: African Development Indicators and Author calculations.

View Full Size

Macroeconomic volatility and Remittances in Africa

Citation: IMF Working Papers 2015, 049; 10.5089/9781498300940.001.A001

Source: Author calculations form African Development Indicators database, 2012.

• Download Figure
• Download figure as PowerPoint slide

where σ_i,X stands for the standard deviation of per capita GDP or consumption using HP filtered data or growth rates. R_i denotes the average ratio of remittances-to-GDP for country i over the sample period. Remittances are measured as the sum of workers’ remittances and compensation of employees in accordance with the World Bank definition. γ is the parameter of interest (the semi-elasticity) that captures the impact of remittances on macroeconomic volatility. X_i is the vector that stacks all the relevant determinants of volatility identified in the literature such as per capita income, financial development, government size, development aid and terms of trade volatility. Financial development is measured by the credit to the private sector to GDP while government size is measured as the ratio of government consumption to GDP. Development aid is an important source of funds for developing economies and particularly for a wide range of African economies. However, the strong variability of aid flows is likely to be harmful for these economies in terms of business cycles because it translates into a high macroeconomic volatility (Pallage and Robe, 2001; Pallage, Robe, and Berube, 2007; Bulir and Hamann, 2003; Arellano and others, 2009; among others). Thus we measure development aid in the regression by its volatility rather than its size. Finally, u_i represents the error terms, assumed to be uncorrelated with the regressors.

A natural issue to account for in this analysis is the potential endogeneity of remittances. There are several reasons supporting the evidence that remittances are not endogenous in our sample of economies. First, correlation test between remittances and growth at leads and lags and at country levels have provided no evidence of a significant relationship. Second, another empirical test of endogeneity (Nakamura and Nakamura (1998)) test shows that remittances are not endogenous in our regressions.⁵ Third, micro studies in African countries have shown that (i) more than half of households in Burkina Faso, Ghana, and Nigeria, and 30 percent of households in Senegal receiving remittances from outside Africa are in the top two consumption quintiles; (ii) Africans also tend to remit more often, and African migrants from poorer African countries are more likely to remit than those from richer African countries. (World Bank, 2011); and (iii) a significant portion of international remittances are spent on land purchases, building a house, business, improving the farm, agricultural equipment, and other investments. These findings imply that remittances could be acyclical at the macro level though they appear to be procyclical (for the richer households) and contracyclical (for poor households) at the micro level. Finally, the endogeneity might have died off by taking country average values over time.

Results. The model in equation (1) is estimated under Generalized Least Squares (GLS) methods instead of standard Ordinary Least Squares (OLS) to overcome potential heteroskedascity in the errors terms. The results are reported in Table 2. Columns (1) and (2) display the results for output volatility (HP filtered and the growth rate of per capita GDP respectively) while columns (3) and (4) display the results for consumption volatility (HP filtered and the growth rate of per capita consumption respectively). We follow an eclectic approach by choosing the best model on the grounds of its fit (adjusted R-square) and report here only the relevant model. The results show a clear negative correlation between output fluctuations and the size of remittances for both measures of volatility. In our preferred regressions, an increase in workers’ remittances-to-GDP of 1 percentage point reduces the log of the standard deviation of GDP per the cyclical component of per capita GDP by 4 percent while the reduction in per capita output growth is 3.2 percent. While remittances reduce consumption volatility, the impact is somewhat small and statistically insignificant (columns (3) and (4)). This result goes at odds with some empirical papers on developing countries in general (Chami, Hakura, and Montiel, 2009; Combes and Ebeke, 2011; Gupta, Patillo, and Wagh, 2009; among others) but it lines up perfectly with the evidence presented by Craigwell, Jackman, and Moore (2010).

Table 2.

Macroeconomic Fluctuations and Remittances

(GLS Estimation)

Note: t-student are in parenthesis. (*), (**), and (***) denote significance at 5,1, and 0.1 respectively. ŷ (respectively ĉ) is the cyclical component of the log of per capita GDP (respectively per capita consumption) using the HP filter with a smoothing parameter of 6.25. Δy (respectively Δc) is the growth rate of per capita GDP (respectively per capita consumption).⁶

Table 2.

Macroeconomic Fluctuations and Remittances

(GLS Estimation)

	GDP		Consumption
Variables	ŷ	Δy	ĉ	Δc
	(1)	(2)	(3)	(4)
Remittances	−4.03*	−3.17*	−1.61	−1.96
	(-2.73)	(-2.18)	(-0.67)	(-0.81)
Government size	−2.26	−3.07*	3.13	2.63
	(-1.71)	(-2.71)	(1.51)	(1.58)
Aid volatility	4.12*	3.84*	−0.49	−0.75
	(2.13)	(2.14)	(-0.26)	(-0.39)
Financial development	−0.83	−0.77	−1.79*	−1.31*
	(-1.50)	(-1.44)	(-2.60)	(-2.21)
Average GDP			−0.13	−0.16
			(-1.58)	(-1.86)
Constant	−3.36***	−2.70***	−0.44	0.32
	(-12.62)	(-12.78)	(-0.22)	(0.17)
Number of observations	25	25	25	25
adj. R-square	0.42	0.43	0.48	0.47

Note: t-student are in parenthesis. (*), (**), and (***) denote significance at 5,1, and 0.1 respectively. ŷ (respectively ĉ) is the cyclical component of the log of per capita GDP (respectively per capita consumption) using the HP filter with a smoothing parameter of 6.25. Δy (respectively Δc) is the growth rate of per capita GDP (respectively per capita consumption).⁶

View Table

Table 2.

Macroeconomic Fluctuations and Remittances

(GLS Estimation)

	GDP		Consumption
Variables	ŷ	Δy	ĉ	Δc
	(1)	(2)	(3)	(4)
Remittances	−4.03*	−3.17*	−1.61	−1.96
	(-2.73)	(-2.18)	(-0.67)	(-0.81)
Government size	−2.26	−3.07*	3.13	2.63
	(-1.71)	(-2.71)	(1.51)	(1.58)
Aid volatility	4.12*	3.84*	−0.49	−0.75
	(2.13)	(2.14)	(-0.26)	(-0.39)
Financial development	−0.83	−0.77	−1.79*	−1.31*
	(-1.50)	(-1.44)	(-2.60)	(-2.21)
Average GDP			−0.13	−0.16
			(-1.58)	(-1.86)
Constant	−3.36***	−2.70***	−0.44	0.32
	(-12.62)	(-12.78)	(-0.22)	(0.17)
Number of observations	25	25	25	25
adj. R-square	0.42	0.43	0.48	0.47

Note: t-student are in parenthesis. (*), (**), and (***) denote significance at 5,1, and 0.1 respectively. ŷ (respectively ĉ) is the cyclical component of the log of per capita GDP (respectively per capita consumption) using the HP filter with a smoothing parameter of 6.25. Δy (respectively Δc) is the growth rate of per capita GDP (respectively per capita consumption).⁶

The inability of remittances to significantly reduce consumption volatility though they decrease output fluctuations is puzzling. However, several explanations can be put forward to solve such a puzzle. First, households’ consumption in developing economies encompasses both consumption of durables and non-durables. It might be that remittances actually reduce the volatility of consumption of non-durables goods but not the volatility of durables. Thus the inability to disentangle these two components makes it harder to observe a significant impact on aggregate consumption fluctuations. Second, surveys at microeconomic levels in some African countries have shown evidence that more than half of households receiving remittances from outside Africa are in the top two consumption quintiles (Mohapatra and Ratha, 2011). In this case, the consumption smoothing effect of remittances may not show up because these households have access to alternative means (e.g., credit markets) to smooth their consumption and may use these remittances for investment instead. Third, remittances can fuel consumption volatility when they are used for other purposes than consumption (e.g., land purchase and housing). As argued by Combes and Ebeke (2011), this contributes to a boom in the real estate sector and then affects the consumption pattern on aggregate. Finally, because of the very limited development of financial and capital markets, consumption volatility may be so high that the increase in remittances would not be enough to alleviate these credit constraints.

B. Remittances and Risk Sharing

To complement the previous section findings, we evaluate the extent of risk sharing through remittances using a methodology developed in Asdrubali, Sorensen, and Yosha (1996). Econometric framework. Consider the following identity, holding for any period t:

where y_it is GDP per capita and R_it denotes the remittances per capita of country i at period t, all in per capita real terms. The mechanism at work in equation (2) is that smoothing of domestic shocks takes place via remittances if the ratio $\frac{y_{it}}{y_{it}+R_{it}}$ varies positively with y_it. That is, an increase in y_it entails a small increase in y_it + R_it. To derive a simple quantitative measure of smoothing from identity (2), first take logs and then differences of both sides of equation (2), and multiply both sides by Δy_it. Then taking expectations leads to:

Finally dividing by Var(Δy_it) gives:

where β_R is the OLS coefficient from the regression of $\mathrm{\Delta }{y}_{it}-\mathrm{\Delta }{\tilde{y}}_{it}$ on Δy_it and β_u the OLS coefficient of the regression of $\mathrm{\Delta }{\tilde{y}}_{it}$ on Δy_it. β_R measures the incremental percentage of smoothing (when positive) or de-smoothing (when negative) achieved via remittances while β_u accounts for the smoothing through other channels and the amount not smoothed at all.

To estimate the parameter β_R, we run the following regression:

where α_i and λ_t are the individual fixed effects and the time effects respectively. Equation (5) shows how risk-sharing from remittances is achieved. The interpretation of this equation is that the smoothing virtues of remittances are stronger when they move countercyclically with the receiving country’s GDP, and remittances could be destabilizing when they are procyclical. In the later case, an increase in GDP elicits a positive reaction from remittances, inducing a stronger increase in $\mathrm{\Delta }{\tilde{y}}_{it}$ than in Δy_it, implying a negative β_R.

The main results. The results are reported in Table 3. The Hausmann test and the variance decomposition of $\mathrm{\Delta }\tilde{\text{y}}$ both show that the random effects model is an appropriate specification for our data. The model in equation (5) with random individual effects is estimated using different estimation methods: Ordinary Least Squares (OLS), Instrumental Variable (IV), and System-GMM of Arellano and Bond (2003). IV and GMM estimation techniques are used to account for potential endogeneity issues where the growth rate of per capita GDP is instrumented by its own first lag.

Table 3.

Remittances and Risk Sharing

Note: Standard errors are in parenthesis.

Table 3.

Remittances and Risk Sharing

	GLS	GLS	IV	IV	GMM	GMM
	(1)	(2)	(3)	(4)	(5)	(6)
βR (%)	3.74	3.70	16.7	21.56	14.81	9.35
	(1.85)	(1.95)	(8.97)	(12.31)	(7.7)	(4.75)
Time fixed effects	No	Yes	No	Yes	No	Yes
Number of observations	644	644	619	619	644	644
Sargan Test Prob > ϰ2					0.81	0.33
Hansen Test Prob > ϰ2					0.77	0.64
AR(1): p–value					0.009	0.008
AR(2): p–value					0.226	0.241

Note: Standard errors are in parenthesis.

View Table

Table 3.

Remittances and Risk Sharing

	GLS	GLS	IV	IV	GMM	GMM
	(1)	(2)	(3)	(4)	(5)	(6)
βR (%)	3.74	3.70	16.7	21.56	14.81	9.35
	(1.85)	(1.95)	(8.97)	(12.31)	(7.7)	(4.75)
Time fixed effects	No	Yes	No	Yes	No	Yes
Number of observations	644	644	619	619	644	644
Sargan Test Prob > ϰ2					0.81	0.33
Hansen Test Prob > ϰ2					0.77	0.64
AR(1): p–value					0.009	0.008
AR(2): p–value					0.226	0.241

Note: Standard errors are in parenthesis.

The results in Table 3 confirm our cross-section evidence: remittances have a positive and significant risk-sharing power. They absorb a fraction worth 3.7 to 22 percent of GDP shocks. The extent to which this number is important depends on the contribution of the other channels (e.g., capital and credit markets) in absorbing GDP shocks in these economies. Yehoue (2005) found that remittances together with standard channels (capital and credit markets) absorbed only 15 percent and 13 percent of GDP shocks in the CEMAC and WAEMU CFA zones in 1980–2000 while development aid especially from France absorbed about 44 percent of shocks.⁷ Therefore, remittances appear to outweigh capital and credit markets channel in dampening the effects of macroeconomic shocks. In addition, given the potential effect of remittances on promoting financial development (Gupta, Patillo, and Wagh, 2009; Combes and Ebeke, 2011; World Bank, 2006, 2011), their cumulated direct and indirect stabilizing effects on business cycle fluctuations could be significantly higher than thought. Moreover, our quantitative findings represent the lower bound values of the effective impact of remittances in African countries as the quality and coverage of data in general and on remittances in particular leave much to be desired. Transfers through informal channels, misreporting in the receiving countries, and misclassification of remittances into other revenues (exports revenues, tourism receipts, nonresident deposits, or FDI, World Bank, 2011) are among the main reasons for the underestimation of remittances in developing countries, particularly African economies.

A. The Theoretical Model

We briefly review the main features of the model. A more detailed description of this model can be found in Jidoud (2012) and references mentioned above.

Preferences. The economy is populated by a representative household whose objective is to maximize its discounted lifetime utility:

where β ϵ [0, 1) is the discount factor, ct is current consumption, U(.) is a period utility function, and Et (.) is the conditional expectation operator on the information available to the household up to period t. The objective of the household is to maximize the above utility subject to the sequence of per period budget constraints given in (11).

Two utility specifications are called to compete in this setup. The first is the KPR preferences (King, Plosser, and Rebelo, 1988) given by

and the second is the so called GHH preferences following Greenwood, Hercowitz, and Huffman (1988)

ω > 1 governs the intertemporal substitution of labor supply while σ ≥ 1 stands for the inverse of the intertemporal elasticity of consumption.

The motivation behind confronting the two preferences specifications is the size of the wealth effect they generate. They indeed represent different levels of constant wealth effect for a given elasticity of labor supply. GHH preferences (8) have the property that labor supply is independent of wealth and is accordingly determined independently from the consumption level of the household. Thus the wealth effect is zero. On the other hand KPR preferences generate a sizeable wealth effect which, everything else equal, would reduce the labor supply of the households in response to a shock that induces an increase in the marginal utility of wealth. In addition, GHH preferences are widely used in the business cycles analysis of developing countries on the grounds that they help the small open economy models reproduce the stylized business cycle facts of open economies, in particular, countercyclical trade balances (Mendoza, 1991; and Garcia-Cicco, Pancrazi, and Uribe, 2010; among others). Therefore comparing the performances of these two specifications is of particular interest for this literature.

Technology. The economy produces one single good using a Cobb–Douglass production function with constant return to scale as follows:

where Yt denotes output in period t, Kt denotes capital in period t, Nt denotes labor input in period t, $A^{-}>0$ is a scale parameter that pins down the level of output. We assume that fluctuations in output are solely driven by a transitory productivity shock at which follows a first-order autoregressive process, that is,

where |ρ_a| < 1 denotes the persistence of the shock.

The resource constraint of the economy is given by

Ct, It, and Ret denote private consumption, private investment, and migrants’ transfers from abroad (remittances) respectively. Dt accounts for the stock of foreign debt (international bonds) held by the country, inherited from the previous period, and qt the price of one unit of bonds (see next subsection).

The law of motion of capital is given by

with δ ϵ (0, 1] being the depreciation rate of capital, and the function Φ(.) represents the capital adjustment function which takes the form

φ ≥ 0 is the elasticity of the capital adjustment costs function. Capital adjustment costs are added in DSGE models because the latter cannot generate realistic moments of investment had these costs being ignored (Koze and Riezman, 2001; among others). Their formulation in equation (13) indicates that changing the stock of capital by a given amount increases with the speed of the desired adjustment and thus gives incentives to the firm to spread out this change over several periods. In addition, capital adjustment costs are zero at the steady state.

B. International Financial Markets and the Role of Remittances

The representative agent has access to international capital markets for non-contingent debt. Then Dt in equation (11) represents the debt or the number of promises from the previous period the agent pays and qt is the price at which he can sell a promise to a unit of goods to be delivered at t+1. Given that the international traded assets are non-contingent bonds, our economy has an incomplete asset markets structure. Furthermore, these markets are assumed to be competitive, and accordingly the gross interest rate, denoted Qt, on foreign borrowing is given by the inverse of the price of bonds:

where R* accounts for the fixed (common) international interest rate. The second component of the bond price is the risk-premium or spread paid by the economy on top of the fixed interest rate on international financial markets. While in the standard SOEM this premium nests on technical necessities, in our setup it acts as a reduced form of a financial friction governing the dynamic effects of aggregate fluctuations and the transmission of financial shocks.⁸ Therefore the parameter ψ reflects the tightness or the sensitivity of these financial frictions to different macroeconomic variables (GDP, debt, remittances). In dealing with this friction, we adopt two modeling assumptions. First we assume the country risk premium is determined by the indebtedness of the economy with respect to its output. So that (14) is given by

where variables with bars are the steady state values. In this specification, the level of the country’s indebtedness has a positive effect on the borrowing cost. The higher the debt relative to GDP the higher is the interest rate paid by the country because investors will fear the probability of default of the country (Garcia–Cicco, Pancrazi, and Uribe, 2010; Ozbilgin, 2010; for a closer formulation). In a second specification we assume the flow of remittances has some impact on the country’s borrowing costs. Formally (14) become

This modeling assumption is underpinned by the empirical evidence that remittances lessen the remittances-dependent countries’ borrowing cost on international capital markets and enhance their creditworthiness. Furthermore, some African countries have used remittances as securitized assets to raise funds on international financial markets at low cost and longer maturities (Ratha, De, and Mohapatra, 2011; Afreximbank, 2005). Evidence about the ability of remittances to ease access to financial markets is a well-established fact in most African economies.

A. The Role of Wealth Effect on Labor Supply

The model is simulated under the world interest rate specification (15) in which we assume the size of remittances have no direct effects on the country’s risk premium. Table 5 reports the results. Column (1) displays all the characteristics of the data for comparison while columns (2)–(4) report the results using KPR preferences. Columns (5)–(7) deal with GHH preferences. It is worth noting that the model replicates well the business cycle facts in these economies. In particular, the consumption to output volatility predicted by the model is the actual one especially with GHH preferences and tight borrowing constraints.

Table 4.

Calibrated Parameters

Table 4.

Calibrated Parameters

Parameters	Description	Value
β	Discount factor	0.95
α	Capital to output ratio	0.33
σ	Inverse of intertemporal elasticity of consumption	5
δ	Depreciation rate of capital	0.15
ω	One plus the inverse of labor supply elasticity	1.5
φ	Capital adjustment costs	5
ρ	Persistence of TFP shock	0.9271
σ	Standard deviation of TFP shock	0.0126
D/Y	Debt-to-GDP ratio	0.80
ψ	The debt price elasticity	{1, 2.5, 7.5}

View Table

Table 4.

Calibrated Parameters

Parameters	Description	Value
β	Discount factor	0.95
α	Capital to output ratio	0.33
σ	Inverse of intertemporal elasticity of consumption	5
δ	Depreciation rate of capital	0.15
ω	One plus the inverse of labor supply elasticity	1.5
φ	Capital adjustment costs	5
ρ	Persistence of TFP shock	0.9271
σ	Standard deviation of TFP shock	0.0126
D/Y	Debt-to-GDP ratio	0.80
ψ	The debt price elasticity	{1, 2.5, 7.5}

Table 5.

Model Predictions and the Wealth Effect

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378 and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady state size of remittances to GDP ratio: 0.0378.

Table 5.

Model Predictions and the Wealth Effect

	Data	KPR preferences			GHH preferences
		ψ = 1	ψ = 2.5	ψ = 7.5	ψ = 1	ψ = 2.5	ψ = 7.5
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Output
εσŷ	−4.03	0.39	0.36	0.45	0.03	−0.03	−0.30
εσΔy	−3.17	0.34	0.33	0.36	−0.07	−0.20	−0.53
Consumption
εσĉ	−1.60	−0.59	−0.55	−0.53	−0.80	−0.56	−0.24
εσΔc	−1.96	−0.60	−0.56	−0.53	−0.90	−0.70	−0.47
Relative volatility
σĉ/σŷ	1.57	1.34	1.76	2.27	1.25	1.46	1.54
σΔc/σΔy	1.62	1.27	1.64	2.10	1.17	1.31	1.37
Auto correlation
ρŷ	0.37	0.18	0.34	0.54	0.22	0.39	0.58
ρΔŷ	0.06	0.21	0.43	0.70	0.32	0.56	0.77
ρĉ	0.42	0.07	0.05	0.01	0.11	0.14	0.17
ρΔc	0.00	−0.03	−0.07	−0.12	0.10	0.12	0.18
Correlation
ρŷ,ĉ	0.45	0.97	0.86	0.46	0.98	0.90	0.60
ρΔy,Δc	0.39	0.96	0.85	0.49	0.97	0.90	0.67

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378 and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady state size of remittances to GDP ratio: 0.0378.

View Table

Table 5.

Model Predictions and the Wealth Effect

	Data	KPR preferences			GHH preferences
		ψ = 1	ψ = 2.5	ψ = 7.5	ψ = 1	ψ = 2.5	ψ = 7.5
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Output
εσŷ	−4.03	0.39	0.36	0.45	0.03	−0.03	−0.30
εσΔy	−3.17	0.34	0.33	0.36	−0.07	−0.20	−0.53
Consumption
εσĉ	−1.60	−0.59	−0.55	−0.53	−0.80	−0.56	−0.24
εσΔc	−1.96	−0.60	−0.56	−0.53	−0.90	−0.70	−0.47
Relative volatility
σĉ/σŷ	1.57	1.34	1.76	2.27	1.25	1.46	1.54
σΔc/σΔy	1.62	1.27	1.64	2.10	1.17	1.31	1.37
Auto correlation
ρŷ	0.37	0.18	0.34	0.54	0.22	0.39	0.58
ρΔŷ	0.06	0.21	0.43	0.70	0.32	0.56	0.77
ρĉ	0.42	0.07	0.05	0.01	0.11	0.14	0.17
ρΔc	0.00	−0.03	−0.07	−0.12	0.10	0.12	0.18
Correlation
ρŷ,ĉ	0.45	0.97	0.86	0.46	0.98	0.90	0.60
ρΔy,Δc	0.39	0.96	0.85	0.49	0.97	0.90	0.67

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378 and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady state size of remittances to GDP ratio: 0.0378.

How does the model perform in terms of the effects of remittances on macroeconomic fluctuations? In columns (2)–(4), consumption volatility decreases, but output volatility increases as the size of remittances increases. Hence remittances have stabilizing effects on consumption and destabilizing effects on output. Unlike KPR preferences, GHH preferences give negative elasticities for both consumption and output. The stabilizing impact of remittances on output becomes stronger (higher elasticity in absolute value) as the risk-premium sensitivity ψ increases but the elasticity of consumption volatility decreases (in absolute value) though it remains negative for all degree of financial tightness. For example, when ψ = 1, a 1 percent increase in the remittances-to-GDP ratio increases (decreases) output (consumption) volatility by 0.39 (0.59) percentage point with KPR preferences. These numbers are 0.03 and 0.80 in the case of GHH preferences. But when the risk premium strongly responds to the indebtedness ratio, an increase of 1 percent in the remittances size reduces both output and consumption volatility by 0.3 and 0.24 percentage points. In summary, besides their ability to generate plausible business cycle facts, GHH preferences are more consistent with our empirical finding: increasing the size of remittances cushions both output and consumption instabilities with the effect on output fluctuations being much more pronounced.

The key channel driving the difference between the two cases is the wealth effect generated by the two preferences. In the GHH preferences, the wealth effect on labor supply is nil whereas it is strictly positive for KPR for a given level of labor elasticity. Because of this wealth effect, labor supply decreases in the case of KPR preferences following an increase in the size of remittances. Households consume more leisure and then the steady state level of employment is reduced. The reduction in the steady state employment increases the elasticity of labor supply and deepens the response of hours and output following a positive TFP shock.¹⁰ This is why output volatility increases following a positive TFP shock for example (see Chami and others, 2008, for a graphical illustration of this effect: technology shocks plus income effect). When households’ preferences are for GHH, their labor supply decision is no longer affected by their income level but only by their wages. In response to a positive TFP shock labor supply increases but does not affect the elasticity of labor supply contrary to the case with KPR preferences. The constant elasticity of labor supply avoids output to respond strongly to TFP shocks. The observed decline in output volatility, although small, comes from a composition effect. Because consumption becomes less and less volatile, GDP fluctuations are moderated as the size of remittances becomes larger.

Figures 4 and 6 plot the impulse responses of different aggregates after a TFP shock for different levels of remittances-to-GDP ratio for KPR and GHH preferences, respectively. It turns out that an increase in remittances does not significantly affect the reaction of output, nor of consumption, in response to a TFP shock in KPR. However, it makes both output and consumption responses less strong for GHH.

View Full Size

IRF to TFP Shocks with KPR Preferences, Without Remittances–Credit Frictions Interaction

Citation: IMF Working Papers 2015, 049; 10.5089/9781498300940.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

IRF with KPR Preferences and Remittances-Credit Frictions Interaction

Citation: IMF Working Papers 2015, 049; 10.5089/9781498300940.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

IRF to TFP Shocks with GHH Preferences Without Remittances–Credit Frictions Interaction

Citation: IMF Working Papers 2015, 049; 10.5089/9781498300940.001.A001

• Download Figure
• Download figure as PowerPoint slide

Why does consumption volatility decrease in both cases? Such a result stems from the reduction of the private consumption multiplier. Indeed an increase of remittances increases the steady state consumption-output ratio. This decouples consumption from output dynamics as consumption responds less to TFP shocks and income fluctuations. Yet this mechanism is not the only one at work. An additional channel driving the reduction of consumption volatility is the composition effect. Because remittances are not volatile themselves, and we are increasing the size of the non-volatile component of consumption, the volatility of consumption decreases. The above investigation has shed light on the wealth effect on labor supply as the main mechanism through which remittances deepen or dampen macroeconomic fluctuations. Furthermore, these effects are stronger in economies in which households have very limited access to financial markets. Unfortunately, this channel fails to generate a strong reduction in macroeconomic volatility as observed in our empirical investigation. For such a channel to generate sizeable smoothing effects, the model needs a strong propagation mechanism.

B. Remittances and Financial Frictions

In the previous subsection, increasing the size of remittances buffered macroeconomic fluctuations in an environment where households’ preferences exhibit no wealth effect and they face strong financial frictions. To improve the model’s performances in generating a sizeable reduction in macroeconomic volatility, we now allow remittances to have negative effects on the risk premium. We thus simulate the model using the specification (16) for the risk premium, which entails a loosening effect of remittances on financial constraints. The results are displayed in Table 6. The predictions of the model are qualitatively identical to those of the baseline setup. Remittances have a negative effect on both output and consumption volatility when the wealth effect in the preferences is killed off (GHH case) while they deepen output volatility when the income effect is allowed (KPR case). However, unlike the previous case, the semi-elasticities in Table 5 are higher in absolute value. That is, the reduction in both output and consumption fluctuations is more substantial. For example, when the risk-premium sensitivity is 7.5, a 1 percent increase in the size of remittances reduces output (consumption) instability by 1.66(1.56) percentage point when preferences are GHH type. In addition, the impact on volatility is stronger as financial constraints become tighter. The labor supply mechanism at work in the previous case still prevails. But the mechanism is much stronger now as more remittances imply greater possibilities to accumulate debt and decouple both consumption and output from TFP shocks. As can be seen in Figures 5 and 7, an increase in remittances entails a smaller response of output and consumption after a TFP shock. The dampening in these reactions is much stronger in GHH preferences. Moreover, the increase in output after a shock lasts longer in this case, affording households more income over time. Topping up the risk premium with remittances proves to be a powerful propagation mechanism, inducing the smoothing effect of remittances. In this new setup, an increase in the size of remittances implies a reduction in the costs of borrowing on international financial markets. This effect transmits into a more stable debt level and accordingly a reduction in the trade balance fluctuations. Hence reducing the risk premium remittances contribute to the stabilization of output and consumption as households get easier access to financial markets.

Table 6.

Model Predictions and Financial Development

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378, and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady-state size of remittances to GDP ratio: 0.0378.

Table 6.

Model Predictions and Financial Development

	Data	KPR Preferences			GHH Preferences
		ψ = 1	ψ = 2. 5	ψ = 7.5	ψ = 1	ψ = 2.5	ψ = 7.5
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Output
εσŷ	−4.03	0.57	0.60	0.25	0.00	−0.36	−1.66
εσΔy	−3.17	0.51	0.48	0.13	−0.23	−0.71	−1.93
Consumption
εσĉ	−1.60	−0.98	−1.07	−1.09	−1.25	−1.34	−1.56
εσΔc	−1.96	−0.97	−1.07	−1.08	−1.36	−1.51	−1.20
Relative volatility
σĉ/σŷ	1.57	1.30	1.71	2.24	1.23	1.43	1.55
σΔc/σΔy	1.62	1.25	1.60	2.08	1.16	1.30	1.38
Autocorrelation
ρŷ	0.37	0.17	0.32	0.54	0.21	0.37	0.58
ρΔŷ	0.06	0.20	0.41	0.69	0.31	0.54	0.76
ρĉ	0.42	0.07	0.05	0.02	0.11	0.14	0.17
ρΔc	0.06	−0.03	−0.06	−0.11	0.10	0.13	0.17
Correlation
ρŷ,ĉ	0.45	0.97	0.88	0.50	0.98	0.91	0.63
ρΔy,Δc	0.39	0.97	0.87	0.52	0.97	0.91	0.69

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378, and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady-state size of remittances to GDP ratio: 0.0378.

View Table

Table 6.

Model Predictions and Financial Development

	Data	KPR Preferences			GHH Preferences
		ψ = 1	ψ = 2. 5	ψ = 7.5	ψ = 1	ψ = 2.5	ψ = 7.5
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Output
εσŷ	−4.03	0.57	0.60	0.25	0.00	−0.36	−1.66
εσΔy	−3.17	0.51	0.48	0.13	−0.23	−0.71	−1.93
Consumption
εσĉ	−1.60	−0.98	−1.07	−1.09	−1.25	−1.34	−1.56
εσΔc	−1.96	−0.97	−1.07	−1.08	−1.36	−1.51	−1.20
Relative volatility
σĉ/σŷ	1.57	1.30	1.71	2.24	1.23	1.43	1.55
σΔc/σΔy	1.62	1.25	1.60	2.08	1.16	1.30	1.38
Autocorrelation
ρŷ	0.37	0.17	0.32	0.54	0.21	0.37	0.58
ρΔŷ	0.06	0.20	0.41	0.69	0.31	0.54	0.76
ρĉ	0.42	0.07	0.05	0.02	0.11	0.14	0.17
ρΔc	0.06	−0.03	−0.06	−0.11	0.10	0.13	0.17
Correlation
ρŷ,ĉ	0.45	0.97	0.88	0.50	0.98	0.91	0.63
ρΔy,Δc	0.39	0.97	0.87	0.52	0.97	0.91	0.69

Note: ŷ and ĉ account for the cyclical component of the HP filtered logs of per capita GDP and private consumption while Δy and Δc are the growth rates of the same variables. The model is simulated with completely exogenous remittances. That is, Ȓ_t = 0. The steady-state level of remittances takes the values 0, 0.0378, and 0.0756. The autocorrelations and the relative volatilities from the model are taken when the model is simulated at the empirical average of the steady-state size of remittances to GDP ratio: 0.0378.

View Full Size

IRF to TFP shocks with GHH Preferences and Remittances–Credit Frictions Interaction

Citation: IMF Working Papers 2015, 049; 10.5089/9781498300940.001.A001

• Download Figure
• Download figure as PowerPoint slide

The bottom line of this analysis is that the stabilizing effect of remittances is stronger in an environment where financial frictions are tight, that is, borrowing constraints are permanently binding. This finding is quite intuitive. In an economy in which individuals can easily save and/or borrow, the marginal value of a unit of remittances is small; and remittances may have no effect on macroeconomic volatility. However, in an economy in which the options for saving and borrowing are limited, remittances constitute a direct source of income but also an asset that helps alleviate these borrowing constraints. This finding has some empirical support in the papers of Gupta, Patillo, and Wagh (2009) and Combes and Ebeke (2011), among others. This result also lines up with Ahamada and Coulibaly (2011) who find the smoothing effect of remittances in sub-Saharan African countries to be nonlinear and higher depending on the level of financial development.