MODELING GROWTH (AND LIBERALIZATION) USING
MODELING GROWTH (AND LIBERALIZATION) USING SMOOTHT RANS ITI 0NS ANA LYS ISDAVID GREENAWAY, STEPHEN LEYBOURNE,and DAVID SAPSFORD:Economic liberalization has been a pervasive phenomenon over the last twentyyears. Programs have been initiated on the assumption that liberalization promoteseconomic growth, but the empirical evidence for this is limited. This paper takesa novel approach to modeling growth and structural change as smooth transitions.This allows us to model deterministic change without imposing discrete changes.We use smooth transition analysis to reappraise the time-series properties of longrun growth rates in a number of developing countries which have undertakenliberalization. Our results challenge conventional wisdom on both methodologicaland empirical grounds. (JEL F1, C2, 0 4 )I.INTRODUCTIONLiberalization, in the sense of trade reforms which reduce anti-export bias in oneway or another while improving the information content of relative price changes, hasbeen promoted in some eighty or so developing countries in the 1980s and 1990s. Someof these liberalizations are unilateral, most arepolicy conditioned under the aegis of WorldBank Structural Adjustment Loans (SALs).'All have been undertaken on the assumptionthat liberalization will ultimately improve export and growth performance. The evidenceon liberalization and growth, however, is atbest somewhat mixed. On the one hand, onehas the exhaustive study of Papageorgiou,Michaely and Choksi [1991] which claims aclose and direct association. On the otherhand, Greenaway and Sapsford [1994] findonly a limited role for liberalization, and arange of studies of SALs associate adjustmentwith a deterioration in growth performance.2These studies generally rely either on crudedata inspection, simple correlation analysis,or multivariate regression. This paper takes anew approach to the question. It starts fiomthe presumption that any changes in economicperformance following a liberalization may bemore appropriately modeled as a steady transition rather than a discrete change. To modelThere is a large and growing literature onliberalization and its effects, stimulated initially by the very extensive trade reform programs promoted in developing countries overthe last fifteen years or so and sustained bythe attempts to understand what is happeningin the eastern and central European economies. The latter do not feature in this paper,though the techniques demonstrated here mayin time prove to be useful for gaining insightinto the effects of regime changes underway.* An earlier draft of this paper was presented at theWestern Economic Association meetings, Vancouver, July1994,the Economic Society of Australia annual conference,Gold Coast, Australia, September 1994 and at seminars atLa Tmbe University, Monash University, University of Tasmania and the Australian National University. Two anonymous referees and Editor Tom Willett provided a numberof constructive comments. The authors gratehlly acknowledge all of this assistance as well as financial support fromthe Nufield Foundation.Greenaway: Professor, Department of Economics, University of Nottingham, England, Phone 44-115-951-5469Fax 44-115-951-5552E-mail david.greenaway@nottingham.ac.ukLeybourne: Reader, Department of Economics, Universityof Nottingham, England, Phone 44-115-951-5478Fax 44-115-951-4159E-mail stephen.leybourne@nottingham.ac.ukSapsford: Professor, Department of Economics, Universityof Lancaster, England, Phone 44-1524-594-234Fax 44-1524-594-244E-mail d.sapsford@lancaster.ac.ukABBREVIATIONS1. For a review of experience, see Whalley [I9911andPMC: Papageorgiou, Michaely and Choksi [1991JGreenaway and Morrissey [1993].2. See Harrigan and Mosley [I9911 and World Bank[1990].LSTR: Logistic smooth transition regressionDFR Dickey-Fuller regression798Economic Inquiry(ISSN 0095-2583)Vol. XXXV, October 1997,798-814OWestern Economic Association International
GREENAWAY. LEYBOURNE & SAPSFORD: MODELING GROWTHthe change in this way, we make use of therecent work of Granger and Terasvirta [ 19941,who explore the properties of a variety of nonlinear specifications that facilitate modelingstructural change as a smooth transition between states. We estimate and test the adequacy of a number of such specifications.Our analysis is in two stages. First we takea novel approach to the modeling of growthwhich allows us to model deterministicchange without, as other analysts have done,imposing discrete changes. Not only does thishelp clarify the statistical properties of thetime series, it also challenges the assumptionsthat underpin much growth modeling. Havingidentified the transitions in the growth serieswe are investigating, we then explore the coincidence of these transitions with well-documented episodes of liberalization. We do notformally test whether liberalization results ingrowth. Our results are, however, informativein two respects. Firstly, they place a questionmark against the widely held presumption thatliberalization is a panacea for growth. Secondly, they also point the way to the appropriate econometric modeling of these processes.The remainder of the paper is structured asfollows. Section I1 briefly describes the context against which our analysis is set. SectionI11 sets out the details of our methodology andthe models to be tested. Section IV reports ourresults, contrasts them with those of previouswork and evaluates the implications for policy. Section V concludes and identifies possible extensions.II.LIBERALIZATION, (EXPORTS) ANDGROWTH: PREVIOUS EVIDENCEThere are essentially two strands of the literature which provide relevant background:one which relates exports and growth; a second which relates liberalization to exports andgrowth.The exports and growth literature is an extensive one, which goes back some years.3 Itstarts from the presumption that exports andgrowth are directly related, with causality running from changes in exports to changes in3. A recent survey of results is given in Greenaway andSapsford [1994].799g r w t hAnalysts. have deployed a variety ofempirical methods, usually of growth accounting type models, and covering a rangeof countries and time periods. The consensusfrom this literature can be summarized as follows. First, it seems to be the case that exportsand growth are in general correlated. However, the correlation holds rather morestrongly for cross-section than time-seriesdata. Second, there is substantial evidence infavor of a threshold effect, i.e. industrialization needs to have proceeded beyond somecritical level, (as proxied by GDP per capita),for the export-growth relationship to hold.Thus notwithstanding the qualifications relating to causality which some authors haveraised, there is quite a lot of empirical supportfor the assertion that exports and growth arerelated. This has been taken by some as a basisfor recommending liberalization, the presumption being that liberalization stimulatesexport performance and this in turn stimulatesgrowth performance.More recent are attempts to establish directly the liberalization-exports-growth relationship. In policy terms the most influentialstudy here is Papageorgiou, Michaely andChoksi [1991] (hereafter PMC). This is a massive study of 36 liberalization episodes in 19countries, over the period from the mid 1950sto the mid 1970s. Using essentially informalanalysis which compares average export andgrowth performance across all of the episodesfor the three years before and three years aftereach episode, they conclude, quite unequivocally, that liberalization boosts both exportsand growth. Notwithstanding the reservationsabout the methodological foundation to thiswork articulated in Greenaway [1993] andCollier [ 19931, the results have been widelyreported as conclusive proof of the efficacyof liberalization-most notably of course inthe World Bank itself.5Greenaway and Sapsford [1994] subjectPMC sample (or rather a subset thereof), tomore rigorous statistical scrutiny using slopeand intercept dummies within a standard pro-4. Note that some studies have conducted formal testsand question whether causality unambiguously runs fromexports to growth. See, for instance, Jung and Marshall[I9851 and Darrat [1987].5. See, for example, Thomas and Nash [ 19921.
ECONOMIC INQUIRY800duction function growth model.6 The modelwas estimated on data for 13 of the 19 countries in the PMC samplee7The results indicatethat in some two-thirds of all cases, liberalization appears to have no discernable impacton the exports-growth relationship. In threeout of the 12 cases, a significant positive relationship is found, in one case a significantnegative relationship. Given the convictionwith which the PMC results are reported, andthe enthusiasm with which they have been embraced by the key lending agency, these results are unsettling. It could of course be thatliberalization does not affect growth. A secondpossibility is that the results are an artifact ofa misspecified model and, given the reservations of Levine and Renelt [1992] on mostgrowth modeling, this is quite plausible. Another possibility is that liberalization does notaffect economic performance via a discretebreak, but rather by initiating a transition. Ifso, structural break tests which rely’ on theexistence of a discrete break will generallyfail to find one. Most of the case study evidence suggests that trade reforms initiate transitions rather than discrete breaks and we needtherefore to model the process as such, ideallywithout imposing priors on the data. Smoothtransitions analysis offers that facility.111.SMOOTH TRANSITIONS ANALYSISSmooth transition analysis is an approachdeterministic structural change ina time-series regression. Originally proposedby Bacon and Watts [1971] and Maddala[1977, 3961, it has been more recently developed by Lin and Terasvirta [1994] andGranger and Terasvirta [1993, ch. 71. Thebasic idea is quite simple. Rather than attemptto identify any change as a single structuralbreak, one identifies it as a smooth transitionbetween regression regimes over time.Where liberalization is concerned, this isintuitively appealing. Even where liberaliza-tion is implemented in a “big bang,” any subsequent effects on growth will typically begradual rather than automatic, the speed ofadjustment being dependent on the efficiencyof markets in the economy in question. In developing countries, big bang liberalizationsare the exception rather than the rule; sequenced liberalizations are more common. Inthese circumstances it can take time for thereforms to gain credibility and for agents toreact. Either way, modeling the impact ongrowth as a discrete structural break is inappropriate.More concretely, following Granger andTerasvirta [1993, ch. 71, a simple logisticsmooth transition regression (LSTR) trendmodel may be written as: pztsf(y,r) E , , t 1, .) THere S, is the well-known curvilinear logisticfunction that maps t onto the interval (0, 1)and E , is a zero mean disturbance term. Underthis formulation and assuming y 0, the modeltransition occurs smoothly between the initialstateto modeling6. Both intercept and slope dummies were used becauseliberalization has the potential to impact on both the leveland rate of growth of GDP. The former is typically association with the once and for all benefits of improved resource allocation; the latter is attributable to more rapidfactor productivity growth in export-oriented sectors.7. Brazil, Columbia, Greece, Israel, Korea, New Zealand, Pakistan, Peru, Philippines, Spain, Sri Lanka, Turkey,Yugoslavia.and the final statecorresponding to S , 0 and S , 1, respectively. Hence, the mean growth rate of yowhich is the coefficient on the trend variablet, changes fromto (PI PJ through time.Notice the model simultaneously allows the intercept to change from a,to (a, az).Here,z is a location parameter which determines thetiming of the transition. For t z, we havesuch that z identifies the transition midpoint.The velocity of transition is controlled by theparameter y. If y takes a large value then the
GREENAWAY, LEYBOURNE & SAPSFORD: MODELING GROWTHtransition is completed in a short period oftime and as y tends to infinity the model collapses to one with an instantaneous structuralbreak in intercept and trend at time t T. Thus(1) embeds the standard structural break modelas a special case. The parameters a2 and P2determine the direction of transition in the intercept and trend, respectively. If y 0, the initial and final model states are reversed but theinterpretation of the parameters remains thesame.The model (1) is nonlinear in parametersand may be estimated by nonlinear leastsquares (NLS) using a suitable iterative optimization algorithm. As pointed out in Grangerand Terasvirta [1993, ch. 71, while the otherparameter estimates can converge quickly,that for y may converge very slowly, particularly if the true parameter value is large (suchthat the transition occurs quickly). This is because a large set of estimated values of y leadto very similar values of S,, which deviate noticeably from each other only in a local neighborhood of the location parameter T. The practical consequence of this is that standard errors of the NLS estimate of y may appear artificially large and should not, therefore, betaken necessarily to indicate insignificance ofthe estimate.The logistic function S, as specified heredoes impose certain restrictions, in that thetransition path is monotonic and symmetricaround its midpoint. More flexible specifications could also be considered which, for example, could allow for non-monotonic andnon-symmetric transition paths. This is facilitated by including a higher order polynomialin t in the exponential term of S,. In addition,we constrain the transitions in intercept andtrend to occur once only, simultaneously, andwith the same velocity. Clearly, a specification which does nor impose these restrictionscould also be entertained. However, a particular advantage our specification has overmore complex specifications is that all the parameters have very straightforward interpretations; in more heavily parameterized versionsthis is no longer the case. Moreover, since thenumber of observations available in this studyis relatively small, degrees of freedom problems would also quickly arise. For these reasons we do not attempt any such extensionshere.80 1In sharp contrast to conventional approaches to modeling structural change, no apriori information is used to fix the date of atransition since the midpoint of the transitionis determined endogenously via the parameterT (with the parameter y then effectively identifying the start and end points). From thestandpoint of modeling liberalization episodes, what this means is that the data areallowed to determine all the pertinent featuresof any transition in the real growth rate-itstiming, duration and direction. If any suchtransition is found, and it need not be, one canthen refer back to the dating of a liberalizationepisode, as established from policy accounts,to see whether or not there is any apparentcoincidence of timing. Our central focus is thePMC study where a number of distinct episodes are identified by the authors. The specific questions we are interested in are: Isthere any evidence of a transition in growthrate in the countries in question over the period to which the PMC study applies? Is thereany connection in terms of timing between thetransition and liberalization as identified byPMC?IV. MODEL ESTIMATIONThe LSTR model was estimated using annual time series data for 13 countries takenfrom the PMC sample: Brazil, Colombia,Greece, Korea, Israel, New Zealand, Pakistan,Portugal, Spain, Sri Lanka, Argentina, Yugoslavia and Indonesia. The dependent variabley , was real GDP per capita at 1980 purchasingpower parity prices. To account for the possibility of stochastic dynamics, the estimatedmodel was augmented to include a lagged dependent variable term asIncluding (at most) a single lagged term inlnb,) was found to be sufficient to yield serially uncorrelated residuals for each series. Notice that with this augmented version of themodel, the mean growth rates ofy, corresponding to the initial and final model states are nowgiven by PI /(1 - cp) and (PI Pz) /(1 - cp), respectively.
802ECONOMIC INQUIRYThe Berndt-Hall-Hall-Hausman optimization algorithm in GAUSSX 3.2 was used tocompute the NLS estimates of the seven unknown parameters in model (2) for each country. Since the model is linear in the parametersai,a2,PI, p2, and cp, considerable economy inestimation is possible as these can be “concentrated out” of the sum of squares functionusing OLS. The estimation results are givenin Table I under LSTR. Asymptotic t-ratios forthe parameter estimates are given in parentheses. Where particular parameter estimates arenot present, they were found to be insignificantly different to zero (at the 5% level), andthe results therefore refer to the model re-estimated assuming these parameters are equalto zero. We also estimate model (2) imposingthe restriction S, O such that the interceptand trend coefficients are constant over time(no transition between regimes occurs). Giventhe inclusion of the lagged dependent variableterm, (2) then specializes to the well-knownDickey-Fuller regression (DFR), which allows the possibility of testing for a unit autoregressive root, or stochastic trend, in realGDP (1nyJ against the alternative of stationarity around a linear deterministic trend. We report these regression results under DFR.The standard likelihood ratio test for therestriction y O does not provide us with avalid test of the null hypothesis of constancyof the intercept and trend against the smoothtransition alternative. This is because underthis null the parameters a2, pz and T are nolonger identified (i.e., they may assume anypossible value). However, a valid Lagrangemultiplier test of this hypothesis has been suggested by Lin and Terasvirta [1994]. This testis based on a two-step approach proposed byDavies [ 19771. Briefly, the test procedure firstassumes that the logistic function S, can beadequately approximated by a polynomialfunction of t up to some order k, say, via aTaylor series expansion. Next, the residualsfrom the DFR model (which assumes constantintercept and trend) are constructed, togetherwith the residual sum of squares which wedenote SSR,. These residuals are then regressed on the same DFR regressors togetherwith additional regressors which are polynomial terms in t up to order k 1. Denoting thesum of squared residuals from this second re-gression as SSR,, the Lagrange multiplier testhas the formL M (SSR, - SSRJ / (SSR,/ 2‘).Given standard regularity conditions LA4 hasan asymptotic x2(k)distribution under the nullhypothesis of constancy of the intercept andtrend. The degrees of freedom of the limitingdistribution is k and not k 1 because t itselfappears as a regressor in the null DFR model.For our purposes, we assume a third-orderTaylor-series expansion of S,(y, r) is adequate,requiring that polynomial terms in t up to thefourth order are included in the second-stageregression. The reported LM statistic therefore3 )under the null hypothhas a ( distributionesis of constancy. Based on the 5% signifi3 )(criticalcance level of a ( distributionvalue 7.81), the results of the LA4 tests inTable I suggest that for all 13 countries evidence of a transition in intercept or trend (orboth) is present. Looking at the estimatedmodels more closely, we see that only for Brazil is the transition restricted to the interceptterm; all the other countries display evidenceof a transition in trend, and hence in meangrowth rates (often in addition to a transitionin the intercept term).For each estimated model we also report thevalue of the Box-Pierce Q(3) statistic for residual autocorrelation. Under the null of zeroautocorrelation this statistic has an approximate
DAVID GREENAWAY, STEPHEN LEYBOURNE, and DAVID SAPSFORD: Economic liberalization has been a pervasive phenomenon over the last twenty years. Programs have been initiated on the assumption that liberalization promotes economic growth, but the
of changes in emerging stock market volatility that occurred after the implementation of financial liberalization policies. The empirical results proved that liberalization did not in any way lead to an increase in stock market volatility. He also showed that market volatility did not react in the same way to different types of liberalization.
Trade Liberalization and Economic Growth: The Nigerian Experience (1971-2012) Echekoba F.N.1, Okonkwo V.I.2, Adigwe P.K3 1,2,3 Department of Banking and Finance Nnamdi Azikiwe University, Awka Abstract: Trade liberalization is an essential component of
2002 and Garrido and Peres 1998). On the particular issue of trade liberalization,1 Krueger (1998), Ben-David and Loewy (1998), and Greenaway et al. (1998) continue to argue in favour of the positive impact of trade liberalization on growth and industrialization. Greenaway et al. (
LABOUR-MARKET ISSUES UNDER TRADE LIBERALIZATION: IMPLICATIONS FOR THAI WORKERS Piriya Pholphirul* This paper analyses the impact of trade liberalization on the labour market in Thailand. The impacts on wages, employment, gender roles, labour standards and protection, human development and unionization are investigated.
trade liberalization episodes may influence the composition of dirty versus clean production, the composition of production may in turn influence trade policy. In the case of India's trade liberalization of 1991, studies suggest that the Indian regulators' choice of which industries to liberalize was driven purely by economic considerations,
Structural equation modeling Item response theory analysis Growth modeling Latent class analysis Latent transition analysis (Hidden Markov modeling) Growth mixture modeling Survival analysis Missing data modeling Multilevel analysis Complex survey data analysis Bayesian analysis Causal inference Bengt Muthen & Linda Muth en Mplus Modeling 9 .
14 D Unit 5.1 Geometric Relationships - Forms and Shapes 15 C Unit 6.4 Modeling - Mathematical 16 B Unit 6.5 Modeling - Computer 17 A Unit 6.1 Modeling - Conceptual 18 D Unit 6.5 Modeling - Computer 19 C Unit 6.5 Modeling - Computer 20 B Unit 6.1 Modeling - Conceptual 21 D Unit 6.3 Modeling - Physical 22 A Unit 6.5 Modeling - Computer
Artificial Intelligence – A European approach to excellence and trust. It outlines the main principles of a future EU regulatory framework for AI in Europe. The White Paper notes that it is vital that such a framework is grounded in the EU’s fundamental values, including respect for human rights – Article 2 of the Treaty on European Union (TEU). This report supports that goal by .
