Currency devaluation and aggregate output in East Africa.
Upadhyaya, Kamal P. ; Rainish, Robert ; Phelan, John 等
Abstract
This paper studies the effect of currency devaluation on aggregate
output level in three East African countries using panel data. An
empirical model that includes monetary, fiscal and exchange rate
variables is developed. Two versions of the model, one with real
exchange rate and another with nominal exchange rate and
foreign-to-domestic price ratio are estimated. Before estimating the
model the time series properties of the panel data is diagnosed. The
estimated results suggest that currency devaluations are neutral to the
economy in short and intermediate run but expansionary in the long run
and the effect emanates from the change in foreign-to-domestic price
ratio not from the change in nominal exchange rate.
JEL classification: F31, F41
I. INTRODUCTION
Devaluation of the real exchange rate is often considered a major
tool in the stabilization of the foreign sector of an economy. It is
argued that a devaluation or depreciation improves the terms of trade by
raising the price of imported goods and service and lowering the price
of exports, leading to an improvement in a country's trade balance.
This improvement in the foreign sector expands aggregate output and
employment in the overall economy. Some argue, however, that currency
devaluation may not necessarily increase the level of output in a
developing country. They even posit that devaluation may contract the
economy. Specifically, it is argued that devaluation may lead to a
negative real balance effect, resulting in lower levels of aggregate
demand and output. At the same time it is also argued that currency
devaluation distributes income from the group with a lower marginal
propensity to save to the group with a higher marginal propensity to
save. This may reduce aggregate demand, leading to a lower level of
output in the economy (Krugman and Taylor 1978; Lizondo and Montiel,
1989). Secondly, a nominal devaluation can decrease aggregate demand
through the negative real balance effect due to a higher price level
which in turn may decrease the level of output. Finally, it is also
possible that if export and import elasticity are very low the trade
balance (measured in terms of domestic currency) may deteriorate leading
to a recessionary effect in the economy. If that is the case then
currency depreciation may lead to a lower level of output and employment
in the economy.
Contractionary effects of exchange rate depreciation can also come
through the supply side. Exchange rate depreciation raises the cost of
imported inputs, leading to a decrease in aggregate supply. In addition,
it may also raise the domestic interest rate and wage level through an
increase in the price level. This may also decrease aggregate supply in
the economy.
The relationships between currency devaluation and output growth
have been investigated by a number of studies however the empirical
findings of the effects of depreciations on the economy are mixed.
Gylfason and Schmid (1983), and Conolly (1983) find that exchange rate
devaluations have a positive effect on the economy. But, Gylfason and
Risager (1984), and Branson (1986) find devaluation to be
contractionary. Edwards (1986), in his widely cited study based on
pooled time-series cross-section data from 12 developing countries,
finds a small contractionary effect in the first year that becomes
expansionary in the second year and neutral in the long run. Using
econometric methodology developed by Wickens and Breusch (1988),
Upadhyaya (1999) finds that devaluation has a neutral effect in the long
run.
Some recent studies have examined the impact of currency
devaluation by incorporating real exchange rate directly and
alternatively by decomposing it into nominal exchange rate and relative
price level (foreign-to-domestic price ratio). For example, Upadhyaya
and Upadhyay (1999) find that devaluation generally did not have any
effect on output over any length of time in the six Asian countries
studied, and any effect uncovered came from changes in relative price
level, and not from nominal devaluation. In another study, Upadhyaya,
Dhakal and Mixon (2000) found currency depreciations were usually
contractionary in selected Latin American countries, and that the
contractionary effect came from nominal exchange rate, not from the
relative price level. Upadhyaya, Mixon and Bhandari (2004) reported
short run expansionary effects on output in Greece and Cyprus between
1969 and 1998 that emanated from both nominal devaluation and changes in
the relative price level.
The present study is based on the panel data from three East
African countries namely Kenya, Tanzania and Uganda. Since currency
devaluation has been one of the elements of the structural adjustment
program of the IMF and World Bank in developing countries, it is
expected that this study will also help to evaluate the success of such
programs.
The organization of the paper is as follows. Section 2 outlines the
methodology of the study and the data. The empirical findings are
discussed in section 3, and the final section presents a summary and
conclusion.
II. METHODOLOGY AND DATA
Economic activity in a developing country is affected by a number
of fiscal and monetary variables, particularly by the level of fiscal
expenditure and the rate of change of the money supply (Edwards 1986,
Khan and Knight 1981). Following this argument we include government
expenditures and the rate of change of the money supply as fiscal and
monetary variables that explain the level of aggregate output. Following
Upadhyaya and Upadhyay (1999) we use two approaches to examine the
effect of a change in the exchange rate on output. The first one
includes the change in the real exchange rate directly, which is
consistent with the idea that a nominal change in the exchange rate
influences output only if it leads to a change in the real exchange
rate. This approach considers only the effect of a movement in real
exchange rates and disregards the combination of nominal exchange rates
and foreign-to-domestic price ratios that generate such a movement. If
the price level changes at the same rate as the nominal exchange rate
then the real exchange rate remains constant (and has no effect on the
output level). This method, however, ignores any asymmetric influence
that an initial jump in the exchange rate may have on output vis-a-vis
the effect of a gradual rise in the price level. Hence, an alternative
model, including the nominal exchange rate (instead of the real exchange
rate) and the relative price level is also estimated. This approach
enables us to find out whether any effect originates from a change in
the nominal exchange rate or from the relative price ratio.
Based on the above discussion the following empirical models are
developed.
[Y.sub.t] = [b.sub.0] + [b.sub.1] + [G.sub.t] + [b.sub.2]
[M.sub.t] + [b.sub.3] [RE.sub.t] + [b.sub.4][RE.sub.t-1] + [b.sub.5]
[RE.sub.t-2] + [u.sub.1] (1)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
In equations (1) and (2), Y is the aggregate output, G is
government expenditure, M is the money supply (all in 1995 prices), E is
the nominal exchange rate of domestic currency to U.S. dollars,
[P.sup.*] is the foreign price level, and P is the domestic price level.
G and M respectively represent the fiscal and monetary policy variables.
World price is used as the foreign price level. The real exchange rate
RE is defined as ([EP.sup.*]/P) and ([P.sup.*]/P) is the foreign to
domestic price ratio (relative price level). Finally, u is the random
error term.
Since an increase in government expenditure is assumed to be
expansionary, the coefficient of G is expected to be positive. Likewise,
the coefficient of M is also expected to be positive as an increase in
the money supply is considered to be expansionary to the economy. The
coefficient of the real exchange rate, RE, captures the effect of a
change in the exchange rate on output and is the main concern of the
present study. If it is negative, and statistically significant, ceteris
paribus, any change in exchange rates negatively affect the real output.
In that case, devaluations (exchange rate depreciations) are
contractionary to the economy. However, if the coefficient of RE is
positive and significant any exchange rate depreciation is expansionary
to the economy. If it is insignificant then devaluation is neutral to
the output growth. Since a change in the exchange rate in the current
period can affect output with a lag, lagged values of the real exchange
rates are also included in the estimation of the regression model.
Inclusion of lagged values is also important because any devaluation can
have different effect for different time horizons. For example, Edwards
(1986) finds that exchange rate depreciations are contractionary in the
short run, expansionary in the medium run, and neutral in the long run.
As mentioned earlier, this study is based on panel data from three
East African countries namely, Kenya, Tanzania and Uganda. The panel
data series is constructed using the annual time series data from 1972
to 2006 for both Kenya and Tanzania and from 1992 to 2006 for Uganda. We
could not use balanced sample for all three countries because some data
for Uganda were missing before 1992. All the variables are measured in
1995 price and are in billion shillings. They are all collected from the
International Financial Statistics, which is published by the
International Monetary Fund.
III. ESTIMATION AND EMPIRICAL RESULTS
As indicated above this study uses panel data from Kenya and
Tanzania for a period from 1972 to 2006 and Uganda for a period from
1992 to 2006. Since the use of non-stationary data can produce spurious
results it is important to test the stationarity of the data series. To
ensure the stationarity of the panel data Levin, Lin and Chu (2002),
Breitung (2000), and Im, Pesaran and Shin (2003) unit root tests are
conducted. As reported in Table 1, the data series are found to be
nonstationary at level hut are found stationary at the first difference
level.
After establishing the stationarity of the data series a
cointegration test is conducted. Since we have used panel data in our
study we conducted Pedroni's panel cointegration test (Pedroni,
1999 and 2004). The test results are reported in Tables 2 and 3. As seen
in the table, in both versions of equation, all the test statistics show
that the null hypothesis of no cointegration can not be rejected at the
conventional level of significance. This suggests an absence of long run
relationship among the variables in both equations. Therefore equations
(1) and (2) are estimated in the following first difference form without
any error correction terms (Engle and Granger, 1987).
[DELTA][Y.sub.t] = [b.sub.0] + [b.sub.1] [DELTA]G +
[b.sub.2][DELTA]M + [b.sub.3] [DELTA] [RE.sub.t] + [b.sub.5]
[DELTA][RE.sub.t-2] + [v.sub.1] (3)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)
The lagged values of RE, E, and [P.sup.*]/P are included to capture
the short, medium, and long run effects of a change in the exchange rate
on the aggregate output.
The estimation of the model using panel data from different
countries requires that the unobserved country-specific variables are
not correlated with the included right hand side variables. If they are
correlated the model can generate misleading results. In order to
address this problem we use fixed effects estimation (Pradhan, et. al.
2008). The fixed effects estimations of equations (3) and (4) are as
follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)
Note: Figures in the parentheses are the t-values for the
corresponding coefficient. ***, **, *, # significant respectively at 1%,
5%, 10 %, and 15% critical level.
Equations (5) and (6) are estimated results of equations (3) and
(4) respectively. The overall results of the estimation seem to be fine
in terms of the coefficient of determinant, the F-statistics and the
direction of the coefficients. The Durbin-Watson values in both
estimations, however, are found to be in the inconclusive range. To
ensure that there is not an autocorrelation problem in our estimation we
re-ran the regression with the AR(1) term and found the coefficient of
this term to be statistically insignificant. This ensures that the
estimations are not suffering from the problem of first order
autocorrelation. In both form equations the estimation of the
coefficient of government expenditure is positive and statistically
significant. This indicates the government expenditures are expansionary
in these countries. Unlike Upadhyaya and Upadhyay (1999) we do not find
any crowding out effect of increased government spending on private
sector output and aggregate output in these countries. The monetary
variable (a change in money supply) is positive and statistically
significant at the conventional level of significance indicating the
effectiveness of monetary policy in these countries.
The focus of this study are the coefficients of the real exchange
rate (RE), the nominal exchange rate (E) and the foreign to domestic
price ratio ([P.sup.*]/P). The contemporaneous effect of a change in
real exchange rate (RE) is negative though not significant. The effect
of the lag of RE changes to positive and again is not significant. The
two year lag of this variable is positive and statistically significant.
This finding suggests that real exchange rate is either neutral or at
best a little contractionary in the short run; a little expansionary (or
even neutral) in the medium run; and expansionary in the long run (as
indicated by the statistically significant coefficient of the two year
lag of RE).
Equation (6) presents the regression result that decomposes the
real exchange rate into the nominal exchange rate and the foreign to
domestic price ratio. The results show that the nominal devaluation has
no significant effect on the aggregate output level even at twenty
percent critical level. The contemporaneous and one year lag effects of
a change in the relative price ratio on the real output is not found to
be significant. The coefficient of two year lag, however, is positive
and statistically significant. Based on this finding it can be argued
that whatever effect of devaluation (change in real exchange rate) has
on output level, it comes from the change in the foreign-to-domestic
price ratio and not from the nominal devaluation.
IV. SUMMARY AND CONCLUSION
This paper studies the effect of currency devaluation on aggregate
output level in three East African countries namely, Kenya, Tanzania and
Uganda. For the analysis an empirical model is developed in which fiscal
and monetary variables are included in addition to the real exchange
rate. An alternative model which decomposes the real exchange rate into
the nominal exchange rate and the foreign to domestic price ratio is
also developed in order to identify whether any changes in the aggregate
output is coming from nominal devaluation, relative price ratio (foreign
to domestic price ratio) or both. Panel data comprising of annual time
series data from 1972 to 2006 for both Kenya and Tanzania and from 1992
to 2006 for Uganda is used. In both versions of the model the time
series properties of the panel data are diagnosed using panel unit root
and panel cointegration tests prior to the estimations. Since the data
series are found to be stationary at the first difference level and the
hypothesis of no cointegration could not be rejected the model is
estimated in the first difference form without an error correction term.
The model is estimated using a fixed effect estimator in order to
account for the country specific effect. The estimated results suggest
that the currency devaluation is neutral in the short and intermediate
run and the expansionary in the long run. The results show the
expansionary long run effect comes from the foreign to domestic price
ratio and not from the nominal devaluation.
References
Branson, H. William (1986), Stabilization, Stagflation and
Investment Incentives: The Case of Kenya 1975-80 in, Economic Adjustment
and Exchange Rates in Developing Countries (Eds) S. Edwards and L.
Ahamed, University of Chicago Press, Chicago.
Breitung, J. (2000), The Local Power of Some Unit Root Tests for
Panel Data, in B. Baltagi (ed.), Nonstationary Panels, Panel
Cointegration, and Dynamic Panels, Advances in Econometrics, Vol. 15,
JAI: Amsterdam, 2000, p. 161-178.
Connolly, M. (1983), Exchange Rates, Real Economic Activity and the
Balance of Payments: Evidence from the 1960s in, Recent Issues in the
Theory of the Flexible Exchange Rates (Eds) E. Classen and P. Salin,
Amsterdam, North Holland.
Cooper, R. (1971), Currency Devaluation in Developing Countries, in
Government and Economic Development (Ed) G. Ranis, New Haven, Yale
University Press. Edwards, S. (1986), Are Devaluations Contractionary?
Review of Economics and Statistics, Vol. 68, 501-508.
Engle, R. and Granger, C. (1987), Cointegration and Error
Correction: Representation, Estimation, and Testing, Econometrica, Vol.
55, 251-76.
Gylfason, T. and Schmid, M. (1983), Does Devaluation Cause
Stagflation?, Canadian Journal of Economics, Vol. 16, 641-54.
Gylfason, T. and Risager, O. (1984), Does Devaluation Improve the
Current Account? European Economic Review, Vol. 25, 37-64.
Im, K. S., Pesaran, M. H. and Shin, Y. (2003), "Testing for
Unit Roots in Heterogeneous Panels", Journal of Econometrics, Vol.
115, pp. 53-74.
International Monetary Fund (various issues) International
Financial Statistics, Washington, DC.
Khan, S., Mohsin, S. and Knight, M. D. (1981), Stabilization
Program in Developing Countries: A Formal Framework, IMF Staff Papers,
Vol. 28, 1-53.
Krugman, P. and Taylor, L. (1978), Contractionary Effect of
Devaluation, Journal of International Economics, Vol. 8, 445-56.
Levin, A., Lin, C. F., and Chu, C. (2002), "Unit Root Tests in
Panel Data: Asymptotic and FiniteSample Properties", Journal of
Econometrics, Vol. 108, pp. 1-24.
Lizondo, J. and P. Montiel (1989), Contractionary Devaluation in
Developing Countries: An Analytical Overview, IMF Staff Papers, Vol. 36,
March, pp. 182-227.
Pedroni, P. (2004), "Panel Cointegration; Asymptotic and
Finite Sample Properties of Pooled Time Series Tests with an Application
to the Purchasing Power Parity Hypothesis," Econometric Theory,
Vol. 20, 597-325.
Pedroni, P. (1999), "Critical Values for Cointegration Tests
in Heterogeneous Panels with Multiple Regressors," Oxford Bulletin
of Economics and Statistics, Vol. 61, 653-670.
Pradhan, G., Upadhyay, M. and Upadhyaya, K. (2008), Remittances and
Economic Growth in Developing Countries" European Journal of
Development Research, Vol. 20, 497-506.
Upadhyaya, K. P. and Upadhyay, M. P. (1999), Output Effects of
Devaluation: Evidence from Asia, The Journal of Development Studies,
Vol. 35, 89-103.
Upadhyaya K. P. (1999), Currency Devaluation, Aggregate Output, and
the Long Run: An Empirical Study, Economics Letters, Vol. 64, 197-202.
Upadhyaya K. P., Dhakal, D. and Mixon, F. G. Jr. (2000), Exchange
Rate Adjustment and Output in Selected Latin American Countries,
Economia Internazionale, Vol. 53, 107-17.
Upadhyaya, K. P., Mixon, D. and Bhandari, R. (2004), Exchange Rate
Adjustment and Output in Greece and Cyprus: Evidence from Panel Data,
Applied Financial Economics, Vol. 14, 1181-1185.
Wickens, R. W. and Breusch, T. S. (1988), Dynamic Specification,
the Long Run and the Estimation of Transformed Regression Models,
Economic Journal, Supplement, 189-205.
KAMAL P. UPADHYAYA, ROBERT RAINISH AND JOHN PHELAN
University of New Haven, USA
Table 1
Panel Unit Root Test
Levin, Lin & Chu Breitung t-stat
Variable Level FD Level FD
E -0.16 -3.36 *** 2.800 -1.43 *
RE -4.42 *** -7.46 *** -4.99 *** -10.18 ***
(P * /P) -3.68 *** -6.99 *** -2.93 *** -5.64 ***
G 2.22 -9.76 *** 2.99 -1.90 **
M 2.83 -3.95 *** -0.17 -2.30 **
Y 0.76 -1.55 * 0.83 -3.94 ***
Ira, Pesaran & Shin
Variable Level FD
E 1.05 -2.96 ***
RE -3.19 *** -7.63 ***
(P * /P) -4.56 *** -7.03 ***
G -1.81 -3.42 ***
M 1.32 -5.33 ***
Y 0.60 -4.24 ***
***, **, * significant respectively at
1%, 5%, and 10% critical level.
Table 2
Pedroni's Panel Cointegration Test
(variables: Y, RE, G, M)
statistics probability
v-statistics 0.62 0.33
rho-statistics 1.71 0.09
PP-statistics 0.27 0.38
ADF-statistics 1.32 0.17
Table 3
Pedroni's Panel Cointegration Test
(variables: Y, E, P* /P, G, M)
statistics probability
v-statistics -0.02 0.40
rho-statistics 1.67 0.10
PP-statistics -0.65 0.32
ADF-statistics 1.93 0.06
