Exchange rate volatility–economic growth nexus in Uganda

(1)

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=raec20

Download by: [203.128.244.130] Date: 14 March 2016, At: 20:29

ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: http://www.tandfonline.com/loi/raec20

Exchange rate volatility–economic growth nexus in Uganda

Lorna Katusiime, Frank W. Agbola & Abul Shamsuddin

To cite this article: Lorna Katusiime, Frank W. Agbola & Abul Shamsuddin (2016) Exchange rate volatility–economic growth nexus in Uganda, Applied Economics, 48:26, 2428-2442, DOI:

10.1080/00036846.2015.1122732

To link to this article: http://dx.doi.org/10.1080/00036846.2015.1122732

Published online: 17 Dec 2015.

Submit your article to this journal

Article views: 70

View related articles

View Crossmark data

(2)

Exchange rate volatility – economic growth nexus in Uganda

Lorna Katusiime, Frank W. Agbola and Abul Shamsuddin Newcastle Business School, University of Newcastle, Callaghan, NSW, Australia

ABSTRACT

The global financial crisis has disrupted trade and capital flows in most developing economies, resulting in an increased volatility of exchange rates. We develop an autoregressive distributed lag model to investigate the effect of exchange rate volatility on economic growth in Uganda.

Using data spanning the period 1960–2011, we find that exchange rate volatility positively affects economic growth in Uganda in both the short run and the long run. However, in the short run, political instability negatively moderates the exchange rate volatility–economic growth nexus.

These results are robust to alternative specifications of the economic growth model.

KEYWORDS

Exchange rate volatility;

economic growth; Uganda JEL CLASSIFICATION C32; E44; F31; F43

I. Introduction

The effect of exchange rate volatility on economic growth has gained considerable attention following the breakdown of the Bretton Woods system. In the present environment of financial deregulation, glo- balization and crises, the importance placed on exchange rate dynamics is unlikely to wane.

Moreover, national economic prosperity is increas- ingly linked to the ability to compete successfully in the global economy. Consequently, exchange rate volatility remains a major concern for national gov- ernments operating in a global economy. This is particularly relevant for developing economies because of their fragile financial systems and high vulnerability to external shocks (Aghion et al.,2009;

Tumusiime-Mutebile2012).

Arguably, excessive exchange rate volatility increases uncertainty, which may adversely affect economic growth. While a plethora of theoretical and empirical studies have investigated the impact of exchange rate volatility on economic growth (Aghion et al., 2009; Arratibel et al.

2011; Schnabl 2008, 2009), the empirical findings are mixed. Notably, the exchange rate literature does not provide a direct link between exchange rate volatility and economic growth. Instead, the debate is framed within the context of economic growth outcomes under different exchange rate regimes. Proponents of a free market economy

(Edwards and Levy Yeyati 2005; Friedman 1953;

Hoffmann 2007) argue that a flexible exchange rate regime allows the domestic economy to adjust to volatile real shocks with minimum output losses. Nonetheless, such an exchange rate regime may be accompanied by excessive exchange rate volatility, leading to poor macroeconomic performance. In contrast, a fixed exchange rate regime can be conducive to macroeconomic stability, which in turn can promote international trade and investment, and ultimately economic growth (Frankel and Rose 2002). However, a fixed exchange rate regime may often encourage protectionist behaviour and thereby lead to inefficient allocation of resources (Obstfeld and Rogoff 1995).

The empirical literature provides mixed results on the effect of exchange rate volatility on economic growth. For example, some empirical studies have found that exchange rate volatility has no impact on economic growth (Bleaney and Greenaway 1998), while others have asserted that an increase in exchange rate volatility reduces economic growth (Arratibel et al. 2011;

Boar 2010; Schnabl2008,2009). It is important to note that studies by Frankel (1999) and Husain, Mody, and Rogoff (2005) have provided empirical evidence to show that the macroeconomic performance of an economy under different exchange rate regimes is influenced by country-specific

CONTACTLorna Katusiime Lorna.Katusiime@uon.edu.au VOL. 48, NO. 26, 2428–2442

http://dx.doi.org/10.1080/00036846.2015.1122732

Downloaded by [203.128.244.130] at 20:29 14 March 2016

(3)

factors. For instance, Aghion et al. (2009) found that the effect of real exchange rate volatility on economic growth is moderated by the level of financial development of the country.

Evidence for the effect of exchange rate volatility on economic growth in Africa is very sparse and the findings are mixed. For instance, in a recent panel study, Adewuyi and Akpokodje (2013) examined the effect of exchange rate volatility on macroeconomic activity in African countries. They provided evidence that exchange rate volatility has a significant positive effect on economic growth. However, their study also found differences in the impact of exchange rate volatility on growth across country groups. In view of the mixed empirical evidence, this study examines the effect of exchange rate volatility on economic growth in Uganda – a developing economy that has received little attention in the extant literature.

Like other small open economies, Uganda’s growth trajectory is sensitive to exchange rate volatility and global economic trends (Kasekende, Atingi-Ego, and Sebudde 2004). In the wake of the global financial crisis (GFC), Uganda has experienced increased exchange rate volatility arising from global shocks, balance of payments deficits and speculative attacks on its currency (Bank of Uganda 2011). This macroeconomic instability is threatening to undermine economic growth gains achieved prior to the GFC.

For instance, economic growth declined from an average of 5% during the period 2005–2008 to an average of 2.3% for the period 2009–2012 (The World Bank 2013). Although an understanding of the exchange rate volatility–economic growth nexus is important for developing effective macroeconomic and exchange rate policies, no previous empirical study explicitly investigated the impact of exchange rate volatility on economic growth in Uganda.

The objective of this study is to empirically investigate the exchange rate volatility–economic growth nexus in Uganda. In our investigation, we control for the effects of fundamental determinants of economic growth as identified in the extant literature, such as domestic investment, human capital, trade openness, financial development and inflation (for a review, see Barro and Lee [1994]; and Durlauf, Kourtellos, and Tan [2008]). We also test the

hypothesis that the impact of exchange rate volatility on economic growth is affected by the political instability of the early 1970s to the mid-1980s and in recent times.

The rest of this article is organized as follows.

Section II provides an overview of the literature on exchange rate volatility and economic growth, highlighting the mixed theoretical predictions and empirical evidence and the importance of other fundamentals of growth. Section III describes the methodology employed in the analyses by describing the models and estimation technique used, namely, the autoregressive distributed lag (ARDL) bounds testing approach introduced by Pesaran, Shin, and Smith (2001). Our measure of exchange rate volatility is generated using a Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model, the standard measure of exchange rate volatility in the extant literature. Section IV presents the empirical results. Section V draws some conclusions and makes policy recommendations for managing exchange rate volatility in order to maintain a stable economic growth path for Uganda.

II. Exchange rate volatility and economic growth: an overview

The two main theoretical foundations underlying empirical studies of economic growth, namely, the neoclassical growth theory pioneered by Solow (1956) and the endogenous growth theory popularized by Romer (1986) and Lucas (1988).

The neoclassical growth theory posits that short- run steady growth is generated through exogen- ous technical progress. Early theoretical work investigating the linkage between exchange rate volatility and economic growth has relied on classical growth theories (Baxter and Stockman 1989). In contrast, endogenous growth theory is based on the argument that steady growth can be generated endogenously. In other words, this growth trajectory could occur without any exo- genous technical progress but rather through external capital accumulation, human capital development or through existing productive designs. Technological innovation makes it possible to introduce new and superior products and processes, and this consequently increases

(4)

productivity and thus economic growth. Based on endogenous growth theory, it could be argued that technological progress is achieved through the implementation of effective economic policies that ensure macroeconomic stability and promote increased investment and productivity.

Since the advent of endogenous growth theory, there have been advances and extensions of the model proposed by Romer (1986) and Lucas (1988), whose recent empirical work has focused on analysis within the endogenous growth theoretical framework. The endogenous growth model is typically augmented with a variable representing exchange rate regime or volatility. Proponents of the flexible exchange rate regime (Edwards and Levy Yeyati 2005; Friedman 1953; Hoffmann 2007) argue that this regime permits an economy to adjust in response to external shocks with minimum output losses. Consequently, these real external shocks have differing impacts on the domestic and foreign economy. In a flexible exchange regime with sticky prices and wages, the exchange rate tends to adjust to correct the discrepancy between domestic and foreign prices in the presence of external shocks. This has the effect of countering the adverse influences on output.

However, a flexible exchange rate regime can be accompanied by excessive exchange rate volatility, which may be detrimental to macroeconomic stability and performance. A fixed exchange rate regime reduces exchange rate uncertainty, which in turn promotes macroeconomic stability and increases international trade–key drivers of economic growth (Frankel and Rose 2002). Nevertheless, as argued by Obstfeld and Rogoff (1995), a fixed exchange rate regime induces protectionist and noncompetitive behaviour. For instance, fixed exchange rate regimes may encourage speculative capital flows, moral hazards and overinvestment in the domestic economy because of the implicit or explicit guarantee of stable exchange rates making economic agents dis- regard potential exchange rate risks (Schnabl2009).

This topic of optimal exchange rate policy continues to generate debate.

In recent times, alternative approaches have emerged exploiting the indirect links between exchange rate volatility and growth. This literature argues that exchange rate volatility can negatively influence some key determinants of economic

growth, such as investment and trade. Excessive exchange rate volatility may deter or delay investments, particularly when investment decisions are irreversible and adjustment costs to exchange rate volatility are high (Goldberg and Kolstad 1994). A number of empirical studies provide evidence of a negative impact of exchange rate volatility on investment (Aghion et al., 2009; Arratibel et al. 2011).

However, other studies have either found no effect of exchange rate volatility on investment (Bleaney and Greenaway1998) or a positive effect on investment (Goldberg and Kolstad 1994). An increase in exchange rate volatility may reduce international trade as market participants direct their resources to less risky economic activities (Clark 1973).

However, the higher risk resulting from exchange rate volatility may provide new opportunities to market participants and thereby increase trade. In general, the literature does not suggest an unequi- vocal link between exchange rate volatility and trade (McKenzie1999).

In the context of the theoretical ambiguity regard- ing the effect of exchange rate volatility on economic growth, several studies attempt to empirically address this issue, but provide mixed results.

Bleaney and Greenaway (1998) find exchange rate volatility is irrelevant in determining economic growth, whereas other studies find that increased exchange rate volatility leads to lower growth (Arratibel et al. 2011; Boar 2010; Schnabl 2008, 2009). Evidence of a positive exchange rate volatility–economic growth relationship is also provided by some (e.g., Mahmood and Ali 2011). However, their findings may be due to the omission of other determinants of economic growth. The mixed results on the relationship between exchange rate volatility and growth may be attributed to the role of country- specific factors, including the level of financial development, human capital, physical capital and institutional settings (Schnabl2008; Frankel1999; Husain, Mody, and Rogoff2005).

Importantly, evidence for the effect of exchange rate volatility on economic growth in Africa is very sparse and the findings we do have are ambiguous.

Using a sample of sub-Saharan African countries, Ghura and Grennes (1993) and Bleaney and Greenaway (2001) find exchange rate volatility has no significant effect on economic growth.

Conversely, Adewuyi and Akpokodje (2013) did

(5)

find a significant positive effect of exchange rate volatility on economic growth for a panel of African countries during the period 1986–2011.

Exchange rate volatility exerted more significant effects in non-francophone countries, including Uganda, compared to francophone countries. Given the inconclusiveness of findings on the effect of exchange rate volatility on economic growth, this study aims to provide new empirical evidence in the context of Uganda, a developing country that has experienced major institutional and economic policy reforms and recurring economic instability.

Uganda’s economy

Like many developing countries, Uganda has under- gone a major economic transformation to develop market-based institutions (Brownbridge, Harvey, and Gockel 1998; Whitworth and Williamson2010;

Katusiime 2015). Prior to the early 1990s, Uganda operated under a system of direct controls on prices and flows of goods and capital. Until 1992, the Bank of Uganda (BOU) controlled the level and structure of interest rates and for most of the 1970s and 1980s nominal interest rates were held below the rate of inflation. For instance, inflation averaged 103% during the period 1981–1990 while nominal lending and time deposit rates averaged 31% and 24%, respectively. The negative real interest rates discouraged the Ugandan public from holding deposits with commercial banks while financial repression and severe mismanagement led to a decline in the quality of financial institutions. In addition, government- owned institutions dominated most banking in Uganda whereby of the 290 commercial bank branches operating in the country in 1970, only 84 remained by 1987. Of these, 58 branches were operated by government-owned banks. Lending prac- tices, although administered through domestically owned banks, were highly influenced by government and predominantly focused on promoting agricul- ture. A case in point is the rural farmers’scheme of the Uganda Commercial Bank. The number of para- statals also increased as the government nationalized previously foreign-owned enterprises. Due to its involvement in domestic production, the government became heavily involved in setting prices for goods such as beer, salt, sugar and soap among others. Further, under the foreign exchange control

act of 1964, the public was forbidden to hold foreign currency. There was an acute shortage of foreign exchange during this period and exporters/importers of commodities were required to deal directly with the central bank which operated under various fixed exchange regimes that were at odds with market conditions. As a result of the controls in the goods and financial markets, parallel/black markets for goods and financial services developed.

The transition from a highly regulated economy to a market-based one can be traced back to the late 1980s and early 1990s when after more than a decade of political instability and economic decline, the government implemented macroeconomic reforms aimed at returning the economy onto a sustainable growth trajectory with assistance from international donors, including the International Monetary Fund and the World Bank (Kasekende, Atingi-Ego, and Sebudde 2004; Whitworth and Williamson 2010;

Katusiime, Shamsuddin, and Agbola 2015a).

These reform efforts focused mainly on introdu- cing market reforms and commitment to prudent macroeconomic management (Kuteesa et al.

2009). Among the key reforms was the introduction of a floating exchange rate regime in 1993 as part of extensive policies aimed at eliminating market controls, thus resulting in the liberaliza- tion of interest rates, current and capital accounts as well as the privatization of state-owned enterprises (Whitworth and Williamson 2010;

Katusiime, Shamsuddin, and Agbola 2015b). The impact of these reforms has been a substantial reduction in poverty and high and sustained economic growth, earning the country recognition as one of the fastest growing economies on the African continent. This in turn has exerted a strong influence on development thinking and international aid architecture in other developing countries in Africa (Kuteesa et al. 2009).

III. Methodology Model specification

The effect of exchange rate volatility on economic growth is examined using an ARDL bounds testing method proposed by Pesaran, Shin, and Smith (2001). The ARDL approach is less restrictive,

(6)

applicable to variables of any order of integration and yields unbiased parameter estimates. Unlike a vector autoregressive model, the ARDL approach does not require all the variables to be integrated of the same order. The bounds test is applicable whether variables have mixed orders of integration (Bahmani-Oskooee, Bolhassani, and Hegerty 2010;

Morley 2006) and as such this approach does not require pretesting the order of integration of the variables, especially when the computed Wald or F-statistic falls outside the critical value bounds.

Therefore, this approach eliminates the uncertainty associated with low power of unit root tests in pretesting the order of integration (2001).

The ARDL model also takes into account endo- geneity (Harris and Sollis 2003; Pesaran and Shin 1998) and performs relatively well in small sam- ples (Narayan and Smyth 2005; Pesaran and Shin 1999). A general ARDL relationship may be spe- cified as follows:

ϕðL;pÞyt ¼β_iðL;qiÞxitþα⁰ztþεt (1) where L is the lag operator; ϕðL;pÞ ¼1ϕ₁L ϕ₂L²ϕ₃L³ ϕ_pL^p and β_iðL;qiÞ ¼β_i0þ β_i1Lþβ_i2L²þ þβ_iqL^qi andzis a vector of deterministic variables comprising the intercept, and exo- genous variables with fixed lags; yt is the dependent variable; xit represents explanatory variables in the cointegrating vector;p and qi are the lag lengths; α⁰ represents coefficient on the deterministic variables and ε is the error term. The error correction repre- sentation of Equation 1 can be expressed as follows:

Δyt¼X^k

i¼1

β_i0Δxitþα⁰ΔztX^{^}^p1

j¼1

θ_jΔytj

X^k

i¼1

X^{^}^qⁱ¹

j¼1

β_ijΔxi;tjθð1;^pÞECTt1þεt

(2)

where Δ is the first difference operator; the error correction term (ECT) is given by ECTt¼

ytP_k

i¼1θixitΨ^_⁰zt

and θð1;^pÞ ¼1Pp i¼1θ measures the quantitative significance of the ECT; θ_j and β_ij are the parameters representing the model’s speed of convergence to equilibrium.

The specific form of our base model for economic growth (Model 1) can be expressed as follows:

ΔLRGDPCt¼α0þXⁿ¹

k¼1

α1kΔLRGDPCðtkÞ

þXⁿ²

k¼1

α2kΔLGKðtkÞþXⁿ³

k¼1

α3kΔLHK

þXⁿ⁴

k¼0

α4kΔLVOL_ðtkÞþXⁿ⁵

k¼0

α5kΔLOPEN_ðtkÞ

þXⁿ⁶

k¼0

α6kΔLPSCðtkÞþXⁿ⁷

k¼0

α7kΔINFðtkÞ

þγ0LRGDPC_ðt1Þþγ1LGK_ðt1Þ

þγ2LHK_ðt1Þþγ3LVOL_ðt1Þþεt

(3) where L denotes natural logarithm, RGDPC denotes economic growth and is derived as real GDP per capita, GK denotes physical capital and is measured as gross capital formation to GDP ratio, HK denotes human capital and is measured by the human capital index, VOL denotes exchange rate volatility and is derived using a GARCH (1, 1) model as presented in Equation 6, OPEN is trade openness and is derived as the sum of total value of exports and imports to GDP ratio, PSC denotes financial development and is measured by private sector credit to GDP ratio and INF denotes domestic inflation and derived using the GDP deflator.

The capital, both human and physical, and technological progress are widely regarded as fundamental determinants of economic growth. An increase in investment represents an increase in the stock of physical capital and is expected to result in higher economic growth. Further increases in human capital, often assumed to occur by increasing the population’s years of schooling, promote economic growth by increasing the productivity of labour.

The important role of technological innovation is due to the ability to introduce new and superior products and processes, including institutions and policies which enhance the productivity of factor inputs. It is impossible to include all the other potential variables identified in the literature to capture the effects of technological innovation, not least because data do not exist for some potentially important variables for the period of analysis cov- ered in this study. Thus our empirical model inves- tigates the impact of exchange rate volatility on growth, controlling for variables suggested by the theoretical literature based on data availability.

(7)

According to endogenous growth theory, the effects of trade openness on economic growth are ambiguous and depend on the magnitude and dom- inance of the different channels through which openness impacts on growth. This includes, for example, facilitating the transmission of technolo- gies, allowing for specialization and access to markets (Eriṣ and Ulaṣan 2013). In endogenous growth models, the effect of inflation on economic growth is nonlinear, whereby the relationship between growth and inflation may be positive at low inflation rates and negative at higher inflation rates (López- Villavicencio and Mignon 2011). Low inflation rates preserve business optimism and thus increase investment which boosts economic growth while high inflation rates discourage investment and consequently economic growth. A sound and well- developed financial system is essential for growth.

The degree of financial development of an economy may influence economic growth through its effects on capital accumulation, particularly through facilitating savings mobilization and risk management.

Thus, it is expected that α2k > 0; γ₁ > 0; α3k > 0;

γ₂ >0; α5k > 0; α6k > 0 andα7k < 0.

An increase in exchange rate volatility may increase exposure of domestic firms to transaction, translation and economic risks. However, while exchange rate volatility may lead to variability in international competitiveness of domestic firms, its ultimate effect on economic growth depends on the size of the nontradable sector, the extent to which investments are irreversible in the tradable sector and whether or not institutional settings are conducive to take advantage of exchange rate volatility.

Since the relationship between exchange rate volatility and growth is a priori indeterminate, it is expected thatα4k > or< 0 and γ₃ > or< 0.

Two augmented versions of Equation 3 are also estimated. First, an interaction term, the product of exchange rate volatility and a dummy variable for political instability, is added to Equation 3 to obtain Model 2, which allows us to determine whether the effect of exchange rate volatility on growth is moderated by domestic political instability. We expect political instability to negatively influence the relationship between growth and exchange rate volatility. This is because political instability increases uncertainty and risk, discourages physical and human capital accumulation and engenders less

efficient resource allocation by firms and govern- ments (López-Villavicencio and Mignon 2011).

Second, Equation 3 is augmented by including the real trade balance (referred to as Model 3). An improvement in trade balance is expected to positively influence economic growth. This is because an improved trade balance increases inflow of foreign currency which stimulates enterprises and economic growth. Conversely, a trade deficit may reduce growth due to a decline in reserves and high interest rates which discourage investment. The positive impact of a surplus/deficit may also be intensified through multiplier effects on consumption and investment. In addition, where a deficit is unsustain- able, it may lead to a currency crisis which may also further dampen growth.

In Equation 3,αik represents the short-run effect and γ_ik represents the long-run effect, which are normalized byα0. The joint F-statistic proposed by Pesaran, Shin, and Smith (2001) is used to test for cointegration. The null hypothesis of no cointegration is tested as follows:

H0:γ₀¼γ₁¼γ₂¼γ₃¼0 (4a) H1:γ₀Þγ₁Þγ₂Þγ₃Þ0 (4b) The null hypothesis is rejected if the computed F-statistic is greater than the upper level of the bound. The null hypothesis cannot be rejected if the computed F-statistic lies below the lower bound. The test is regarded as being inconclusive if the F-statistic falls within the band.

Data and measurement of key variables

The data used for analysis are compiled from the World Bank, the BOU and the Penn World Table 8.0 (Feenstra, Inklaar, and Timmer 2013).

The analysis is based on annual data for the period 1960–2011. The sample consists of 52 yearly observations. The choice of the sample period and data frequency is guided by data availability. Details of the data and sources are summarized in Table 1 while Table 2 provides a summary of descriptive statistics and the results of unit root tests.

In order to identify the effect of exchange rate volatility on macroeconomic performance, it is important to identify a suitable measure of exchange

(8)

rate volatility. The main contentions in the debate concerning the measurement of volatility arise from the lack of a definitive theoretical approach to quan- tifying exchange rate risk. As a result, various tech- niques are adopted in the literature (for a review, see

McKenzie [1999]). In general, the suitability of any measure adopted is informed by the scope of the analysis.

Two types of exchange rate volatility measures stand out in the literature: ex ante and ex post volatility. The ex ante measures of volatility, also known as implied volatility, are based on market estimates of future volatility while ex postmeasures use observed historical price information to calculate volatility. The most common measure of ex ante exchange rate volatility which captures market expectations of how volatile prices will be in the future is derived from traded foreign exchange option prices (Bonser-Neal and Tanner 1996).

Arguably, given the absence of‘options on US dollar (USD)/Uganda shilling’, it is not possible to calculate the implied volatility of the ‘Ugandan foreign exchange rate’. Ex post exchange rate volatility can be measured in terms of either realized volatility (standard deviation of historical exchange rate returns) or conditional volatility from a GARCH model. The realized standard deviation of exchange rate volatility is calculated from the moving subsam- ples of exchange rates.

Another challenge in the literature is choosing an appropriate exchange rate variable to represent the uncertainty component of the exchange rate.

Throughout the floating period, nominal and real exchange rates appear to have moved together.

This co-movement is the result of the stickiness of domestic prices. Thus, the distinction between real or nominal measures of exchange rate volatility is unlikely to significantly change the assessment of the effects of exchange rate volatility (McKenzie1999).

Financial time series such as exchange rate have certain stylized characteristics such as volatility clus- tering and leptokurtic properties that the OLS esti- mator is unable to adequately capture. In this study, the conditional volatility of the nominal exchange rate is calculated from a GARCH (1, 1) model of the following form:

Rt¼μ₀þμ₁ðPOLSÞ þεt where εtjΩt1,Nð0;htÞ (5) ht¼θ0þθ1ε²_t₁þθ2h_t1 (6) where Rt¼ ln _P^P_t1^t

h i

, and where Pt denotes the nominal Uganda shilling/USD exchange rate in Table 1.Data description and sources.

Variable Definition Source

LRGDPC Natural logarithm of real GDP per capita at constant 2005 USD

Penn World Table 8.0 and World Bank WDI databases LGK Natural logarithm of gross

capital formation (as % of GDP).

World Bank WDI database

LHK Natural logarithm of index of human capital per person, based on years of schooling (Barro and Lee2013) and returns to education (Psacharopoulos1994).

Penn World Table 8.0

LVOL GARCH measure of exchange rate volatility and derived using natural log of monthly nominal exchange rate

Bank of Uganda

LOPEN Natural logarithm of trade openness–derived as the total value of sum of exports and imports of goods and services (as % of GDP)

LPSC A measure of financial sector development, calculated as natural logarithm of domestic credit to the private sector (as % of GDP)

INF Inflation rate, measured in terms of GDP deflator in percentage (2005 = 100); for the period 1983–2011 obtained from WDI and that for the period 1960–1982 obtained from Bigsten and Kayizzi-Mugerwa (1999)

World Bank WDI database and Bigsten and Kayizzi- Mugerwa (1999)

LTB Natural logarithm of trade balance–derived as real external trade balance on goods and services (as % of GDP)

World Bank WDI database and Penn World Table 8.0

LVOL × POLS The interaction term is the product of natural logarithm of exchange rate volatility (VOL) and political instability dummy (POLS). The dummy variable for political instability takes a value of 1 during the political instability of 1971–1986 and 0 otherwise

Own calculations

Table 2.Summary statistics.

Variable Mean SD Maximum Minimum Observations

LRGDPC 6.66 0.23 7.13 6.28 52

LGK 2.55 0.41 3.20 1.72 52

LHK 0.40 0.18 0.68 0.16 52

LVOL 0.03 0.10 0.70 0.00 52

LOPEN 3.58 0.30 4.08 2.83 52

LPSC 1.81 0.49 2.88 0.96 52

INF 35.71 54.19 216.00 −11.00 52

LTB −0.35 0.40 0.26 −1.10 52

(9)

periodt;μ₀is the expected exchange rate return in a period of political stability; POLS is a dummy variable for political instability; μ₀þμ₁ is the expected exchange rate return in times of political instability;

ht is the conditional variance of exchange rate returns andθ0; θ1 andθ2 are GARCH model parameters. The annual estimate of exchange rate volatility is derived as an average of the monthly GARCH exchange rate volatility estimates.¹ We apply White’s heteroscedasticity test to examine the presence of heteroscedasticity. The White test statistic of 9.861 with a p-value of 0.002 indicates the presence of heteroscedasticity and thus justifies the use of a GARCH model for modelling exchange rate volatility.

The nominal exchange rate data used are taken from the BOU database. Before 1980, the Uganda shilling was pegged to major currencies including the USD, pound and special drawing rights before 1980 (Atingi-Ego and Sebudde 2004). In the latter half of the 1970s, the domestic economy deteriorated and there was a corresponding prolonged shortage of foreign currency, which gave rise to a parallel unofficial market with an overvalued Uganda shilling. In the decade 1980–1990, various exchange rate regimes were implemented with the aim of restoring macroeconomic stability. These were an independent float, a dual exchange rate regime, auction system, adjustable independent peg and the discre- tionary crawl. The early 1990s saw the emergence of the bureaux market, which helped to narrow the gap between the official exchange rate and the bureau rate (Kasekende and Ssemogerere 1994; Whitworth and Williamson 2010). In 1993, Uganda adopted a flexible exchange rate regime, and since then the Uganda shilling has persistently depreciated against the USD. The study uses data spanning the period 1960–2013 and includes the pre-float and floating exchange rate regime eras. This is justified by the frequent regime changes in the pre-float period.

In order to construct a continuous inflation series, it was necessary to link the old GDP deflator series obtained from Bigsten and Kayizzi-Mugerwa’s (1999) for the period 1960–1982 with the current index for the period 1983–2011 obtained from the World Development Indicators (WDI). We create a

new series of inflation based on the GDP deflator with 2005 = 100 as the base year. This was achieved by splicing the old series on to the new series at a common point of time via a link factor. The index was then rescaled to the new base year of 2005.

The measure of economic growth used in this study is calculated using real GDP data from the Penn World tables and population data from the World Bank database as the natural log of real GDP per capita at constant 2005 prices (in million 2005 USD). The interaction term combines the natural logarithm of exchange rate volatility and a political instability dummy, where by the dummy variable for political instability takes on the value of 1 during the political instability (1971–1986) and zero if otherwise.

IV. Results and discussion

Unit root and cointegration tests results

The ARDL model does not require testing of the orders of integration of variables. Nevertheless, for bounds testing the dependent variable should be I(1) and the regressors should be I(0), I(1) or fractionally integrated (Pesaran, Shin, and Smith 2001). Table 3 reports the Augmented Dickey–Fuller Test (ADF) (Dickey and Fuller 1979) and Phillip–Perron (PP) (Phillips and Perron1988) unit root test results for all the variables employed in the analyses. The results show that there is a mixture of I(1) and I(0) variables employed in the analyses. Since the dependent Table 3.Unit root test results.

Variables

ADF PP

Levels First difference Levels First difference

LRGDPC −0.52 −3.73*** −0.07 −3.77***

LGK −1.06 −9.05*** −1.06 −9.12***

LHK −1.47 −1.85 0.85 −1.85

LVOL −5.82*** −10.86*** −5.82*** −37.06***

LOPEN −1.76 −6.74*** −1.82 −6.93***

LPSC 0.11 −3.44** −0.25 −7.68***

INF −2.73 −7.72*** −2.75* −7.97***

LTB −2.26 −8.90*** −2.07 −10.30***

ADF PP

1% Level 5% Level 10% Level 1% Level 5% Level 10% Level

−3.568 −2.921 −2.599 −3.565 −2.92 −2.598

***, ** and * denote the series is stationary at the 1%, 5% and 10%

significance level, respectively.

1The results based on a measure of realized volatility (LVOL) where the annual volatility is obtained as the sum of log squared monthly returns are fairly similar and hence are not reported here. They are available from the corresponding author on request.

(10)

variable is I(1) and none of the independent variables appears to be integrated at an order higher than one, we use the ARDL technique (see Bahmani-Oskooee, Bolhassani, and Hegerty2010; Morley2006).

The Pesaran bounds test for cointegration was carried out using the econometric software Microfit 5.0 based on Equation 3, referred to as Model 1. The results are reported in Table 4. Given the small sample size, the study draws on asymptotic critical values provided by Narayan (2005) for small sample sizes of 30–80 observations to assess the statistical significance of the test statistics. Since the study is based on annual data, we follow Pesaran and Pesaran (2010) to estimate the conditional ARDL model in Equation 3 with a maximum lag length of 1. The F-test statistic of 7.2 is greater than the upper bounds of the critical values suggested by Narayan (2005) at the 5% significance level. Thus, the null hypothesis of no cointegration is rejected at the 5%

significance level, providing evidence of a long-run equilibrium relationship between economic growth and explanatory variables.

We extend the analysis in Model 2 by including an interaction term in Equation 3 to capture the combined effects of political instability and exchange rate volatility and in Model 3 by including the real trade balance. We perform the bounds test for cointegration to confirm the presence of a cointegrating relationship in Models 2 and 3. The results, presented in Table 4, also yield conclusive evidence of cointegration as indicated by the estimated F-statistics of 6.8 and 7.0 for Model 2 and Model 3, respectively. These are higher than the critical values reported by Narayan (2005) at the 5% significance level. Therefore, a cointegrating relationship exists between economic growth and explanatory variables.

Given the conclusive evidence of cointegration for Models 1–3, we proceed to estimate their long- and short-run dynamics, applying the Schwarz Bayesian criterion (SBC) for selecting the optimal lag length. The results are presented in Table 5.

The F-tests for the presence of a long-run level relationship between economic growth and the explanatory variables in the models are presented in Panel B of Table 5 under model diagnostics. In addition, critical values of those test statistics at the 95% confidence level are presented. These critical values are applicable even when intercept dummies are included in the model as regressors (Pesaran and Pesaran 2010). The results indicate conclusive evidence of cointegration for both Model 1 and Model 2. The calculated F-statistics of 12.1 and 16.6 for Model 1 and Model 2, respectively, are greater than their respective simulated critical values at 95% confidence level. Thus, the null hypothesis of no long-run equilibrium relationship between economic growth and the explanatory long-run variables in Models 1 and 2 is rejected at the 5% significance level. In contrast, for Model 3 the F-statistic of 5.8 lies between the critical values of 4.9 and 6.1 at the 95% confidence level, yielding an inconclusive cointegration test result.

Discussion of results

Table 5 reports the results for alternative ARDL models of economic growth. In view of the significant ECT, we report the results for Models 1–3 in line with Kremers, Ericsson, and Dolado (1992).

Table 4.ARDL bounds cointegration test results.

Dependent variable^a

F-statistic for Case II Intercept

no Trend F(4,136)^b Conclusion Model 1

LRGDPC 7.23 Cointegration

LGK 1.98 No cointegration

LHK 0.62 No cointegration

LVOL 3.80 No cointegration

Model 2

Model 3

Notes:

aThe cointegrating vector includes real GDP per capita (LRGDPC), gross capital formation as a percent of GDP (LGK), human capital (LHK) and exchange rate volatility (LVOL), while trade openness (LOPEN), financial sector development (LPSC) and inflation (INF) are excluded from the cointegrating vector, but included in the short-run dynamic models. In Model 2, the interaction term (LVOL × POLS) is also included in the short- run dynamics while the real trade balance (LTB) is included in the short- run dynamics of Model 3. The F-test statistic indicates which variable should be normalized when a long-run relationship exists between the lagged level variables in the cointegrating vector. For each model, four alternative cointegrating relationships are examined with different dependent variables.

bThe relevant critical values are obtained from Table B.1 Case II: Intercept no Trend when k = 3. They are 3.22 and 4.38 for the lower and upper bound, respectively, at 95% significance level. Narayan (2005) provides a set of critical values for small sample sizes ranging from 30 to 80 observations. WhenN= 55, the critical values are 3.41 and 4.62 for the lower and upper bound, respectively, at 95% significance level.

(11)

Table 5.ARDL model results.

Model 1 Model 2 Model 3

Regressors ARDL (1,0,0,1) Long run Short run ARDL (1,0,0,1) Long run Short run ARDL (1,0,0,1) Long run Short run Panel A

Intercept 1.512*** 5.598*** 1.521*** 5.629*** 1.628*** 5.635***

(3.79) (25.78) (4.02) (27.54) (3.76) (27.18)

LRGDPC (−1) 0.730*** 0.730*** 0.711***

(10.86) (11.42) (9.80)

GK 0.071*** 0.264** 0.037 0.139 0.082*** 0.285**

(2.96) (2.15) (1.39) (1.25) (2.86) (2.33)

ΔGK 0.071*** 0.037 0.082***

(2.96) (1.39) (2.86)

HK 0.038 0.139 0.080 0.294* 0.057 0.199

(0.80) (0.87) (1.65) (1.85) (1.05) (1.19)

ΔHK 0.038 0.080 0.057

(0.80) (1.65) (1.05)

LVOL 0.045 0.485* 0.275** 1.339** 0.052 0.502*

(0.95) (1.70) (2.56) (2.37) (1.07) (1.85)

ΔLVOL 0.045 0.275** 0.052

(0.95) (2.56) (1.07)

LVOL(−1) 0.086** 0.087** 0.093**

(2.12) (2.25) (2.21)

LOPEN −0.024 −0.089 −0.006 −0.024 −0.030 −0.104

(−1.18) (−1.11) (−0.31) (−0.31) (−1.35) (−1.31)

ΔLOPEN −0.024 −0.006 −0.030

(−1.18) (−0.31) (−1.35)

LPSC 0.103*** 0.382*** 0.105*** 0.388*** 0.103*** 0.357***

(3.21) (7.50) (3.43) (8.06) (3.19) (6.01)

ΔLPSC 0.103*** 0.105*** 0.103***

(3.21) (3.43) (3.19)

INF −0.0002** −0.001* −0.0004*** −0.001** −0.0002** −0.001*

(−2.11) (−1.81) (−3.19) (−2.47) (−2.14) (−1.85)

ΔINF −0.0002** −0.0004*** −0.0002**

(−2.11) (−3.19) (−2.14)

LVOL POLS −0.281** −1.039**

(−2.35) (−2.06)

ΔLVOL POLS −0.281**

(−2.35)

LTB 0.018 0.064

(0.71) (0.75)

ΔLTB 0.018

(0.71)

ECT(−1) −0.270*** −0.270*** −0.289***

(−4.02) (−4.23) (−3.98)

PANEL B Model diagnostics

F-statistics⁵ 12.10 16.58 5.85

95% Lower bound 4.34 4.31 4.87

95% Upper bound 5.79 5.78 6.05

AdjustedR² 0.99 0.61 0.99 0.64 0.99 0.60

SE of regression 0.03 0.03 0.03

SBC 97.95 99.21 96.30

Durbin–Watson statistic 2.17 1.78 1.62

Residual diagnostics

Serial correlation¹ 2.147 [0.143] 0.404 [0.525] 1.863 [0.172]

Functional form² 0.361 [0.548] 2.363 [0.124] 0.554 [0.457]

Normality³ 2.551 [0.279] 1.805 [0.406] 2.142 [0.343]

Heteroscedasticity⁴ 1.102 [0.294] 0.052 [0.820] 1.508 [0.219]

F-statistics 443.953 [0.000] 437.810 [0.000] 390.080 [0.000]

Notes: The values in parentheses aret-ratios while probabilities are brackets.

*, ** and *** denote statistical significance at 10%, 5% and 1% significance levels, respectively.

The critical value bounds are computed by stochastic simulations using 2000 replications.

1Breusch-Godfrey Lagrange multiplier test of residual serial correlation.

2Ramsey’s RESET test for omitted variables/functional form.

3Jarque–Bera normality test based on a test of skewness and kurtosis of residuals.

4White’s test for heteroscedasticity based on the regression of squared residuals on squared fitted values.

5If the F-statistic lies between the bounds, the test is inconclusive. If it is above the upper bound, the null hypothesis of no level effect is rejected. If it is below the lower bound, the null hypothesis of no level effect can’t be rejected.

(12)

Model 1 corresponds to specification of Equation 3, Model 2 is Model 1 augmented with the interaction term for exchange rate volatility and political instability and Model 3 is Model 1 augmented with the real trade balance. The results for all three models are qualitatively similar. In addition, the signs of the coefficients and magnitudes are similar. For instance in all models, both the short- and the long-run coefficients have the expected signs with the exception of the trade openness variable. The signs of the short- and long-run coefficients are similar, but the short-run coefficients are less than long-run coefficients in abso- lute magnitude across all models.

In all three models, exchange rate volatility positively affects long-run economic growth and also positively impacts on short-run economic growth in Model 2. In addition, financial sector development positively affects economic growth in the short and long run in all three models. Further, in the short run, economic growth in Models 1 and 3 is explained by gross capital formation. The coefficient of ECT in all three models is negative and significant at the 1% level. For example, the coefficient of ECT is −0.27 in Model 1, implying that 27% of any deviation from equilibrium is corrected within a year.

Table 5 reports the short- and long-run elasti- cities for Models 1–3. In line with the predictions of economic growth theories, the coefficient on gross capital formation (as a percentage of GDP) in Models 1 and 3 carries the expected positive sign and is statistically significant at the 5% level in both the short run and the long run. The corresponding elasticity coefficient indicates that, in the short run, a 1% rise in the investment-to-GDP ratio results in an increase in economic growth of about 0.07 percentage points in the short run, while a 1% increase in the investment-to-GDP ratio increases economic growth by approximately 0.26 percentage points in the long run. Our results are qualitatively similar to those of Gylfason and Herbertsson (2001), Baldacci et al. (2008), Oketch (2006), Arratibel et al. (2011) and Herrerias and Orts (2011). Importantly, Model 2, which is augmented by the exchange rate and political instability interaction term, yields a coefficient on gross capital formation as a percentage of GDP that carries the expected sign. It is, however, statistically

insignificant in both the short-run and the long- run model specifications.

The coefficient of the measure of human capital carries the expected sign in all models in both the short-run and the long-run horizons, but the effect is statistically significant at the 10% level only in the long run for Model 2. The result indicates that a higher stock of human capital leads to higher economic growth. In particular, a 1% increase in the level of human capital results in an increase in economic growth of approximately 0.3 percentage points. This finding is similar to the findings of Baldacci et al. (2008), Aghion et al. (2009) and Herrerias and Orts (2011) who found a positive and statistically significant effect of education on economic growth.

In general, exchange rate volatility has a positive effect on economic growth. In the short run, the elasticity of economic growth with respect to exchange rate volatility is only significant in Model 2 at the 5% level, while in the long run, exchange rate volatility exerts a statistically significant effect on economic growth in all models. In the short run, a 1% increase in exchange rate volatility results in 0.3% increase in economic growth, while in the long run, a 1% increase in the level of exchange rate volatility increases economic growth by approximately 0.5–1.3 percentage points. Our finding is similar to that of Adewuyi and Akpokodje (2013) for African countries, but contrasts with the negative relationship between exchange rate volatility and economic growth as reported by Ghura and Grennes (1993), Schnabl (2008,2009) and Arratibel et al. (2011). The positive association between exchange rate volatility and economic growth may be due to the flexible exchange rate regime introduced in Uganda in 1993. Arguably, the flexible exchange rate regime insulates the economy against economic shocks resulting in lower output losses (Edwards and Levy Yeyati 2005; Friedman 1953;

Hoffmann2007).

Table 5 shows that the effect of exchange rate volatility on economic growth is negatively moderated by political instability but the impact is only marginal. The elasticity of exchange rate volatility during periods of political instability is calculated as the sum of the coefficient on foreign exchange volatility and the coefficient on the interaction variable. In times of political instability, a 1% increase in

(13)

exchange rate volatility increases economic growth by only 0.3 (= 1.339 – 1.039) percentage points in the long run. The corresponding short-run elasticity is only−0.006 (= 0.275–0.281) during an episode of political instability. This finding is consistent with the findings of Guillaumont, Guillaumont Jeanneney, and Brun (1999), Gyimah-Brempong and Corley (2005) and Aisen and Veiga (2013) who provide evidence that political instability has an adverse influence on economic growth. In addition, Guillaumont, Guillaumont Jeanneney, and Brun (1999) argue that political instability is often accompanied by ‘bad’ economic policies in African countries.

The coefficient on trade openness variable has an unexpected sign but is statistically insignificant at the 5% level in all models. While a number of studies have found that trade openness has a positive effect on economic growth (Adewuyi and Akpokodje 2013; Aghion et al., 2009; Arratibel et al. 2011), other studies have found no effect at all (Eriṣ and Ulaṣan 2013) or a negative effect (Montalbano 2011). The findings of this study suggest that trade openness may not have stimu- lated productivity growth and therefore economic growth in Uganda.

The level of financial development has a positive and statistically significant effect on economic growth in both the short and the long run. This finding holds in all models. In the short run, a 1% increase in financial development is associated with a 0.1% rise in economic growth, while in the long run, it leads to a 0.4%

rise in economic growth. These results are similar to those of Aghion et al. (2009) and Huang and Lin (2009).

Table 5 shows a negative and statistically significant effect of inflation rate on economic growth. In the short run, a 1% increase in inflation reduces economic growth by approximately 0.01 percentage points, while in the long run, a 1%

increase in the inflation rate results in a decline in economic growth of between 0.03 and 0.05 percentage points. This implies that price instability endangers economic growth in Uganda. Our results are comparable to those of Gylfason and Herbertsson (2001), Schnabl (2009), Gillman and Harris (2010) and Eriṣ and Ulaṣan (2013).

Inflation acts as a tax on physical and human

capital, thus decreasing the rate of return on capital and ultimately lowering economic growth (Gylfason and Herbertsson 2001). From Table 5, we find a positive association between the real trade balance and economic growth in both the short and the long run, but these effects are statistically insignificant.

The coefficient of ECT is negative and statistically significant at the 5% level in all models.

Thus, the results demonstrate that Uganda’s economic growth trajectory moves towards a long- run steady state, although the speed of adjustment is very slow. It is estimated to be approximately 27–29 percentage points per year with the full adjustment to equilibrium expected to take about 3–4 years.

V. Conclusion and policy implications

Both the theoretical predictions and the empirical evidence on the effect of exchange rate volatility on economic growth are mixed. In addition, there is a paucity of research on the experience of African countries. This study provides new empirical evidence on the effect of exchange rate volatility on economic growth in a developing country in transition, Uganda, within the ARDL cointegration theoretic framework. We find a long-run equilibrium relationship between exchange rate volatility and economic growth. Our empirical results suggest that exchange rate volatility has positive short- and long-run effects on economic growth. This finding implies that the flexible exchange rate regime which is accompanied by increased volatility may have served as a buffer, and thereby outweighing the adverse impacts of exchange rate volatility on economic growth in Uganda. We find that during periods of political instability exchange rate volatility appears to exert an adverse effect on economic growth largely due to the increased uncertainty over policy decisions and enforcement in the short run, although the effect is reversed in the long run. In addition, capital stock, human capital and the level of financial development positively affect economic growth. Uganda’s government should promote financial market development and formation of human and physical capital to accelerate economic growth.