Testing Export-Led Growth in Bangladesh: An ARDL Bounds Test Approach

Transcription

1 Testing Exort-Led Growth in Bangladesh: An ARDL Bounds Test Aroach Biru Paksha Paul Abstract Existing literature on exort-led growth for develoing countries is voluminous but inconclusive. The emerging economy of Bangladesh is registering sectacular growth in both exorts and outut in recent decades. This scenario has sarked a rise in research on the exort-outut relationshi for the country. The results are nevertheless ambiguous. In a study over the liberalized regime from 979 to 00, this study engages a relatively new method of the autoregressive distributive lag (ARDL) bounds test aroach, and finds strong evidence on exort-led growth for Bangladesh in both the long run and the short run. Imorts, however, do not show any significant relationshi with outut. The trade reform that began in Bangladesh in the late 970s aears to have benefited the country s economic growth. Index Terms ARDL bounds test, Bangladesh economy, exort-led growth, level relationshi, trade liberalization. JEL Codes: F4, F43, C3, O53 I. INTRODUCTION With the advent of trade liberalization in recent decades, the hyothesis of exort-led growth has drawn widesread attention to research. Bangladesh, as a vibrant South Asian economy, has recently exhibited sectacular growth erformance. The growth rates of its gross domestic roduct (GDP), imorts, and exorts remained above 5, 6, and 0 ercent, resectively, over the last two decades []. This scenario has triggered a number of questions such as: ) Is a significant art of Bangladesh s GDP growth exort-led? ) What is the imort-outut relationshi? 3) Is exort-led growth in Bangladesh, if any, a long-run or short-run henomenon? This aer attemts to address these questions by examining Bangladesh s trade-outut relationshi in an autoregressive distributed lag (ARDL) bounds test aroach, a relatively new method to cointegration. The studies on exort-led growth in Bangladesh are numerous, but the results are still inconclusive. The ositive imact of exorts on outut in the short run is not surrising as exorts enter the aggregate income equation []. In the long run, exorts may affect growth through availing economies of scale, introducing incentives of imroving the quality of the roducts, reducing inefficiencies, and finally, innovating new technology due to cometitive ressure in the world market [3]-[6]. Since the main objective of this study is to examine exort-led growth for Bangladesh, I resent the most relevant studies on the trade-outut The manuscrit received Setember 9, 03; revised October 7, 03. Biru Paksha Paul is with the State University of New York at Cortland (SUNY Cortland) ( biru.aul@cortland.edu). relationshi for the country. While [7] finds evidence of exort-led growth in Bangladesh for the eriod, [8] finds no significant effect of exort growth on GDP growth over the eriod. Reference [9] finds that exort growth significantly increases economic growth through its ositive imact on total factor roductivity over the eriod. Reference [0] finds no evidence of a long-term relationshi between exorts and economic growth over the eriod. Reference [] finds the evidence of exort-led growth for Bangladesh in both the long and the short run over the eriod. Reference [] finds both long-run and short-term causality from income to exorts over the eriod. Reference [3] finds unidirectional causality from exorts to growth for the eriod. Reference [4] shows that trade liberalization fosters exort growth for the eriod. Reference [5] works with 44 develoing countries over the eriod, and finds evidence on exort-led growth for Bangladesh in the long run, not in the short-run. Reference [6] finds limited suort for exortled growth over the eriod. Reference [7] finds long-run evidence on exort-led growth for the eriod. Reference [8] is the only one that uses the ARDL aroach like mine, and examines the contributions of exorts, foreign direct investments, and remittances to GDP of South Asian countries. Based on a samle, this work finds the evidence of cointegration among these variables in Bangladesh. The error correction term, however, is highly insignificant (-value 0.63), suggesting an unstable long-run equilibrium model. When most aers in this resect find evidence on exortled growth in Bangladesh, they have three major shortcomings: omitting imortant variables, selecting inaroriate variables and samles, and indulging methodological deficiencies. Although GDP and exorts are of rimary imortance in testing exort-led growth, excluding imorts can create omitted variable bias or missecification error. Many studies overlook this issue. Some studies use the IIP to roxy GDP an attemt inaroriate for a country like Bangladesh [9]. Of crucial imortance is the issue of samle selection. After its indeendence in 97, Bangladesh embarked on socialist lanning, and the consequential trade controls were massive. After the regime change in 975, the country slowly began to move towards rivatization and the market economy. The exchange rate remained fixed until the late 970s. In 979, Bangladesh launched a managed exchange rate to romote exorts in articular. In the same year, the ruling government won the arliamentary election and that legitimacy emowered the regime to continue its liberalization olicies. Anti-exort bias was reversed. The DOI: /IJTEF.04.V5.33

2 exort romotion bureau was remodeled in 979 to exlore new areas of trade. Hence, the year of 979 can be viewed as a beginning of trade liberalization in Bangladesh (see [0], [] for details). Most studies ignored this oint in samle selection. Many studies suffer from methodological deficiencies such as small samle and omitted variable bias, imroer conversion of data frequency, conflicting results in the Johansen test, and insignificant error-correction model. My study overcomes these shortcomings, and adots an efficient aroach to measure both the long-run and shortrun exort-income relationshi in Bangladesh. In a samle over the liberalized regime from 979 to 00, this study finds evidence on exort-led growth for Bangladesh. The result is robust in that the effect of exorts on outut is ositive and significant in both the long run and the short run. Imorts, however, do not show any significant relationshi with GDP. The exchange rate reform and exort romotion efforts that began in the late 970s aear to have benefited Bangladesh s growth. II. DATA AND METHODOLOGY The 000-constant U.S. dollar values of GDP, exorts, and imorts have been collected from []. These series commence in 979, the beginning of the liberalized regime, and end in 009. Based on [] and [3], I add observations for 00. Although discarding the observations of 00 does not make any difference in findings, I refer to kee them to increase the degrees of freedom. Thus, the samle begins in 979 and ends in 00, as shown in Fig.. I(0), urely, or mutually cointegrated. In contrast, the Johansen test can work only with nonstationary series. Moreover, it is unnecessary that the order of integration of the underlying regressors be ascertained rior to testing the existence of a level relationshi. Therefore, unlike tyical alications of cointegration analysis, this method is not subject to this articular kind of retesting roblem. Thus, a long-run relationshi can be established with this technique irresective of the time series roerties of the variables in the model. Even when some of the model regressors are endogenous, the bounds testing aroach generally rovides unbiased long-run estimates and valid t- statistics [9], [30]. Moreover this aroach rovides a method of simultaneously assessing the short-run and the long-run effects of one variable on the other. At the same time, the ARDL has an aealing searation of short-and long-run effects [3]. The long-run relationshi can be estimated in the following forms: y=α t + βx+β t m+ε () t t xt 3 4mt t () mt 3 5 6xt 3t (3) where y denotes income/gdp/outut, x stands for exorts, m is imorts, α i (i = 3) stands for intercet terms, β i (i = 6) signifies coefficients on resective variables, and finally ε it (i = 3) denotes error terms. To imlement the bounds testing rocedure, the following conditional ARDLerror correction models (ECMs) are resented, which corresond to () through (3), resectively: c xt 3mt i xt mt u t (4) xt c xt 3mt xt mt ut (5) mt c3 mt 3xt xt mt u3 t (6) Fig.. Bangladesh s outut, exorts, and imorts (in log). Source: [], [], and [3]. The studies that sequentially develoed the ARDL bounds test aroach include [4], [5], and [6]. There are numerous advantages in the ARDL method, which makes it more useful than others. With a small samle size, as is the case with mine, this method is more efficient than other techniques. In contrast, the Johansen cointegration technique, as in [7], requires larger samles for the results to be valid [8]. The simlicity of this ARDL bounds test method is aealing. As oosed to other multivariate cointegration techniques, it allows the cointegrating relationshi to be estimated by the OLS method once the lag order of the model is identified. This asymtotic theory rovides a simle univariate framework for testing the existence of a single level relationshi between a variable and its regressors when it is not known with certainty whether the regressors are urely integrated of order one, where Δ denotes the first difference oerator. c i (i = 3) shows constants, π i (i = 3) signifies coefficients on the lagged levels, θ i, i, and δ i (i = ) denote coefficients on the lagged variables, and finally u i (i = 3) stands for error terms. signifies the maximum lag length, which is decided by the user. The researcher usually deends on literature and convention to determine the maximum lag length. The selection criteria such as Schwarz Bayesian Criterion (SBC) and Akaike Information Criterion (AIC) are mainly used to determine the order of the ARDL model. Given my samle size, I decide to use the SBC to determine the lag length of the ARDL, because the SBC chooses the most arsimonious model [3], [33]. Here F and W (Wald) statistics are emloyed to bounds test for the existence of a long-run relationshi where the null hyothesis states that the coefficients on the lagged levels are zero: H 0 : π = π = π 3 = 0. This method rovides lower and uer bound critical values where the lower

3 bound critical values assume all variables are I(0) while the uer bound critical values assume all variables are. If the calculated F or W statistic exceeds the uer bound, the null hyothesis of no cointegration can be rejected. If they fall below the lower bound, the null hyothesis of no longterm relationshi cannot be rejected. However, if they fall within their resective bounds, inference would be inconclusive. Now estimating long-run coefficients and the ECM along with the short-run arameters becomes necessary. The sign of the error correction (EC) coefficient must be negative and significant to ensure convergence of the dynamics to the long-run equilibrium. III. ARDL AND ERROR CORRECTION ESTIMATIONS As discussed before, testing the variables for unit roots before the ARDL estimation is unnecessary, because this estimation can accommodate any variables, which are, I(0), or mutually cointegrated. However, one limitation of the ARDL method is that it cannot be estimated with I() series. Although the series of GDP, exorts, and imorts, as shown in Fig., are most likely I (), I refer to test them to make sure that none of them are I (). TABLE I: PHIILIPS-PERRON UNIT ROOT TESTS WITH BANGLADESH S OUTPUT, EXPORTS, AND IMPORTS: Variables: In levels In first difference Outut 7.39 (.00) Exorts 0.60 (0.99) Imorts (0.9) Model A Model B Model A Model B Integration -.7 (0.90) -.0 (0.47) -.43 (0.36) (0.0) Note: Model A includes intercet, and Model B includes both intercet and trend. The null hyothesis states that the variable has a unit root. - values are shown in the arentheses under each adjusted t-statistic. Source: [], [],and [3]. The Augmented Dickey-Fuller (ADF) test is widely used in this regard, as in [34] and [35]. Reference [36] roosed a modification of the Dickey-Fuller (DF) test and has develoed a comrehensive theory of unit roots. The Phillis-Perron (PP) test has introduced a t-statistic on the unit-root coefficient in a DF regression, corrected for autocorrelation and heteroskedasticity. Formally, the ower of a test is equal to the robability of rejecting a false null hyothesis. Monte Carlo simulations show that the ower of the various DF tests can be very low. Reference [37:07] comments that the DF test does not have serious size distortions, but it is less owerful than the PP test. Reference [38] asserts that for low frequency data like mine the PP test aears to be more owerful than the ADF test. Accordingly, I adot the PP methodology to test unit roots in the variables. Table I resents the results of the Phillis- Perron unit root tests with Bangladesh s GDP, exorts and outut. All the variables exhibit unit roots, whereas they become stationary in first differences. Thus, none of the series are (), and they can be used in the ARDL bounds test method. The aroach rovides us with 95 ercent critical bounds for the F and W (Wald) statistics, as shown in Table II. TABLE II: ARDL COINTEGRATION TESTS WITH GDP, EXPORTS, AND IMPORTSIN BANGLADESH: LHS Variable ΔGDP t Δexorts t Δimorts t Forcing Variable exorts, imorts GDP, imorts GDP, exorts F- statis 95% Critical Bounds W- statis 95% Critical Bounds I (0) I () I (0) I () Cointegration Positive Absent Absent Note: The null hyothesis for the F-test W-test is π = π = π 3 = 0. π, π, and π 3 denote coefficients on lagged levels as er (4) through (6). Statistics are bold when significant at the 5 ercent level.δ denotes the first order difference oerator. Source: Same as in Table I. The critical value bounds are comuted by stochastic simulations using 0,000 relications [39]. The long-run level relationshi is evident only when GDP is deendent on exorts and imorts. The F statistic is significant at the 5 ercent level, and so is the W statistic in a consistent manner. Both statistics exceed the 95 ercent uer bound, signifying cointegration among GDP, exorts, and imorts. Cointegration disaears when either exort growth or imort growth is laced in the LHS as er (4) and (5), suggesting no evidence of growth-led exort or growthinduced imort in the long run. The long-term feedback between exorts and imorts is also absent. Since both exorts and imorts act as forcing variables in the exlanation of GDP, we are not sure at this oint whether this imlies exort-led growth or imort-led growth or a combination of both until we get the long run estimates. TABLE III: LONGRUN COEFFICIENTS AND ERROR CORRECTION ESTIMATES FOR ARDL(, 0, 0) MODEL Panel A: Long run coefficients with GDP(t) as deendent variable: Constant exorts(t) imorts(t) Coefficient.983*** 0.88*** Standard error Δexorts (t) Δimorts (t) ECM (t-) Adj. Coefficient 0.045** * 0.63 Standard error Panel C: Diagnostic tests: Regressors Panel B: Error correction model (ECM) estimations with ΔGDP(t) as deendent variable: Regressors ECM(t) = GDP(t) exorts(t) imorts(t) intercet LM-Version F-Version Statistics -value Statistics -value A: Serial Correlation χ()= F(,5)= B: Functional Form χ()= F(,5)= C: Normality χ()= N/A - D: Heteroskedasticity χ()= F(,8)= Note: Δ denotes the first order difference oerator. *, **, and *** indicate that the coefficients are significant at the 0 %, 5%, and % levels, resectively. In Panel C, the nulls are: (A) no serial correlation, (B) no functional-form missecification, (C) no non-normal errors, and (D) no heteroskedasticity. Source: Same as in Table I. Table II guides us to select (4) only for the ARDL. The R 3

4 SBC selects an ARDL (, 0, 0) model, whose results are resented in Table III. Panel A of Table III gives the longrun coefficients of the ARDL estimations. The coefficient on exorts is 0.88, and is highly significant, suggesting that GDP rises by the 88 ercent of the increase in exorts in the long run. In contrast, the long-run imact of imorts on GDP is negative but highly insignificant. While exort-led growth for Bangladesh is quite evident, imorts have no significant effect on outut in the long run. The results of the error-correction model, as resented in Panel B of the same table, are quite consistent with that of Panel A. The short-run imact of exorts on GDP is also ositive and significant at the 5 ercent level. Thus, exorts foster GDP growth in Bangladesh both in the short run and long run. The role of imorts in the error-correction model remains as insignificant as before. The error-correction term is negative as exected, but it is significant at the 0 ercent level, instead of the conventional 5 ercent level. The negative sign before the error-correction term, , suggests that the long-run equilibrium relation comes back to the steady state if the system is ever shocked. The value of the coefficient, however, is very low, suggesting that it will take a long time to restore the steady-state relation if the system is disturbed. The estimated model reflects a fairly high level of goodness of fit, as shown in the adjusted R-squared value, Panel C of the same table resents the diagnostic tests for the estimation. Given the results of the diagnostic tests in both LM and F versions, this estimated model is free of serial correlations, functional-form missecification, nonnormal errors, and heteroskedasticity at the 5 ercent level. IV. CONCLUSION The recent growth erformance of Bangladesh in both outut and exorts has sarked some questions such as: ) Is a significant art of Bangladesh s GDP growth exort-led? ) Is exort-led growth in Bangladesh, if any, a long-run or short-term henomenon? 3) Is there any feedback between GDP and imorts? My aer addresses these questions in an ARDL bounds test aroach. The results of the studies in this resect are still inconclusive. This aer fills that ga by overcoming the shortcomings of samle selection and methodology that ersisted in most revious aers. In a study over the liberalized regime from 979 to 00, this work finds evidence on exort-led growth for Bangladesh. The result is valid in both the long run and the short run. Imorts, however, do not have any significant relationshi with GDP. The trade reform and exort romotion agenda that commenced in the late 970s aear to have benefited Bangladesh s growth. This finding of exort-led growth for Bangladesh has olicy imlications for other develoing nations that asire to grow fast but confront dilemmas with trade liberalization and economic oenness. This aer raises some additional questions: ) Why is not the feedback between GDP and imort significant? ) How do we estimate the imort demand function for Bangladesh? 3) Is exort-led growth also evident in other South Asian countries? These questions are left for future research. ACKNOWLEDGMENT I am thankful to Christoher Hanes, Daniel Henderson, and Urbashee Paul for their valuable comments. My secial thanks go to Anuam Das and Gazi Salah Uddin for their resource and data suort. Contact: biru.aul@cortland.edu, Tel: REFERENCES [] WB, World Develoment Indicators 00 CD-ROM, World Bank, Washington DC, 00. [] G. Feder, On exorts and economic growth, Journal of Develoment Economics, vol., , 98. [3] B. Balassa, Exorts and Economic Growth: Further Evidence, Journal of Develoment Economics, vol. 5,. 8-89, 978. [4] J. N. Bhagwati and T. N. Srinivasan, Trade olicy and develoment, in International Economic Policy: Theory and Evidence, R. Dornbusch, and J. A. Frenkel, Eds. Johns Hokins University Press, 978, [5] W. G. Tyler, Growth and exort exansion in develoing countries: some emirical evidence, Journal of Develoment Economics, vol. 9,. -30, 98. [6] D. Rodrik, Closing the technology ga: does trade liberalization really hel? NBER WP 654, 988. [7] S. Dodaro, Exorts and growth: A reconsideration of causality, Journal of Develoing Areas, vol. 7, , 993. [8] M. N. Islam and M. Iftekharuzzaman, Exorts-Growth Nexus in a Small Oen Economy: The Case of Bangladesh, in the Economy of Bangladesh: Problems and Prosects, C. E. Weis and A. Wahid, Eds., Praeger Publishing, CT, 996, [9] S. Begum, and A. F. M. Shamsuddin, Exorts and economic growth in Bangladesh, Journal of Develoment Studies, vol. 35,. 89 4, 998. [0] A. Razzaque, B. H. Khondker, N. Ahmed, and M. K. Mujeri, Trade liberalization and economic growth: Emirical evidence on Bangladesh, MAP technical aer, Bangladesh Institute of Develoment Studies, 003. [] M. A. Hossain, and N. D. Karunaratne, Exorts and economic growth in Bangladesh: Has manufacturing exorts become a new engine of exort-led growth? International Trade Journal, vol. 8, no. 4, , 004. [] J. Love and R. Chandra, Exort-led growth in Bangladesh in a multivariate VAR framework, Journal of Asian Economics, vol. 5, , 005. [3] K. A. A. Mamun and H. K. Nath, Exort-led growth in Bangladesh: A time series analysis, Alied Economics Letters, vol., , 005. [4] P. J. Dawson, The exort income relationshi and trade liberalization in Bangladesh, Journal of Policy Modeling, vol. 8, , 006. [5] M. B. Oskooee and M. Oyolola, Exort growth and outut growth: An alication of bounds testing aroach, Journal of Economics and Finance, vol. 3, no.,. -, 007. [6] H. A. Ahmed and M. G. S. Uddin, Exort, imorts, remittance and growth in Bangladesh: An emirical analysis, Trade and Develoment Review, vol., no.,. 79-9, 009. [7] M. A. Hossain, L. Haseen, and N. Jabin, Dynamics and causality among exorts, imorts and income in Bangladesh, Bangladesh Develoment Studies, vol. 3, no.,. 0-3, 009. [8] M. Rahman, Contributions of exorts, FDI, and exatriates remittances to real GDP of Bangladesh, India, Pakistan, and Sri Lanka, Southwestern Economic Review, vol. 36,. 4-53, Sring, 009. [9] P. Dua and A. Banerji, An indicator aroach to business and growth rate cycles: The case of India, Indian Economic Review, vol. 36, no., , 00. [0] N. Aziz. (008). The role of exchange rate in trade balance: Emirics from Bangladesh. Working aer, University of Birmingham. [Online] Available: htt:// [] S. A. Choudhury. (008). Prioritizing develoment: Imact of trade olicy: Bangladesh ersective. [Online]. Available: htt:// f_new%0delhi%0%07_8%0dec%0008.t(4-- 08,04.5).df [] MoF. (00a). Macro Scenario-Bangladesh ministry of finance. [Online]. Available: htt:// 4

5 mes_en_july_se_0.df?hmyadmin=gqnistr56c5oxdv%cer uqlwwom5andhmyadmin=xrgktgdj7v3tjluz5xtaqmrx9 [3] MoF. (00b). Medium-Term macroeconomic framework: Bangladesh ministry of finance. [Online]. Available: htt:// MyAdmin=GqNisTr56C5oxdV,EruqlWwoM5,. [4] M. H. Pesaran and B. Pesaran, Working with Microfit 4.0: Interactive Econometric Analysis, Oxford University Press, 997. [5] M. H. Pesaran and Y. Shin, An autoregressive distributed lag modeling aroach to cointegration analysis, in Econometrics and Economic Theory in the 0th Century: The Ragnar Frisch Centennial Symosium, S. Strom, Ed. Cambridge University Press, 999, [6] M. H. Pesaran, Y. Shin, and R. J. Smith, Bounds testing aroaches to the analysis of level relationshis, Journal of Alied Econometrics, vol. 6, , 00. [7] S. Johansen and K. Juselius, Maximum likelihood estimation and inference on cointegration with alication to the demand for money, Oxford Bulletin of Economics and Statistics, vol. 5, , 990. [8] S. Ghatak and J. Siddiki, The use of ARDL aroach in estimating virtual exchange rates in India, Journal of Alied Statistics, vol. 8, , 00. [9] P. K. Narayan, The saving and investment nexus for China: evidence from cointegration Tests, Alied Economics, vol. 37, no. 7, , 005. [30] N. M. Odhiambo, Energy consumtion and economic growth nexus in Tanzania: An ARDL bounds testing aroach, Energy Policy, vol. 37, no.,. 67-6, 008. [3] I. Bentzen and T. Engsted, A revival of the autoregressive distributed lag model in estimating energy demand relationshis, Energy, vol. 6, , 00. [3] W. Enders, Alied Econometric Time Series, 3e, Wiley, NJ, 00. [33] H. Lutkeohl, Comarison of criteria for estimating the orders of a vector autoregressive rocess, Journal of Time Series Analysis, vol. 6,. 35-5, 985. [34] D. Dickey, and W. A. Fuller, Distribution of the Estimates for Autoregressive Time Series with a Unit Root, Journal of the American Statistical Association, vol. 74, , 979. [35] D. Dickey and W. A. Fuller, Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica, vol. 49, , 98. [36] P. C. Phillis and P. Perron, Testing for a Unit Root in Time Series Regression, Biometrika, vol. 75, no., , 988. [37] G. S. Maddala and I. M. Kim, Unit Roots, Cointegration, and Structural Change, Cambridge University Press, 998. [38] I. Choi and B. Chung, Samling Frequency and the Power of Tests for a Unit Root: A Simulation Study Economics Letters, vol. 49,. 3-36, 995. [39] B. Pesaran and M. H. Pesaran, Time Series Econometrics using Microfit 5.0, Oxford University Press, 009. Biru Paksha Paul was born in Bangladesh in 963. He received his B.S.S. (Honors) and M.S.S. both in economics from the University of Dhaka in 986 and 989, resectively. He earned his M.B.A. in finance from the University of Technology-Sydney in 999. He received his M.A. in alied economics from the State University of New York at Binghamton in 004. He received his Ph.D. in economics from the same university in 007. He is now serving as an associate rofessor of economics at the State University of New York at Cortland (SUNY Cortland) where he joined in 007. He teaches macroeconomics and finance by relating theories to his twelve-year industry exerience in banking and economic journalism. He worked as a consultant at Bangladesh s Central Bank, Ministry of Finance, and the UNDP. His aers have been acceted in numerous journals such as the Journal of Asian Economics, Journal of Quantitative Economics, Energy Economics, Indian Economic Review, and International Review of Economics. He has ublished one book on Indian business cycles and inflation. His research interest focuses on macroeconomic olicy and economic growth of develoing economies. Dr. Paul is a life member of Bangladesh Economic Association. He received SUNY Cortland Award for excellence in research and scholarshi in