Trade Liberalization, Economic Crises, and Growth

741KB Sizes 3 Downloads 125 Views

Report

PDF Reader
Full Text

World Development Vol. 40, No. 11, pp. 2177–2193, 2012 Ó 2012 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

http://dx.doi.org/10.1016/j.worlddev.2012.03.020

Trade Liberalization, Economic Crises, and Growth ROD FALVEY Bond University, Gold Coast, Australia NEIL FOSTER Vienna Institute for International Economics, Austria

and DAVID GREENAWAY * University of Nottingham, UK Summary. — Many economic reforms are undertaken during an economic crisis, but is a crisis a good time to undertake trade reform? We investigate whether an economic crisis at the time of trade liberalization aﬀects a country’s subsequent growth performance. We employ threshold regression techniques on ﬁve crisis indicators to identify the “crisis values” and to estimate the diﬀerential growth eﬀects in the crisis and non-crisis regimes. Although trade liberalization in both crisis and non-crisis periods raises subsequent growth, we ﬁnd that an internal crisis implies a lower acceleration and an external crisis a higher acceleration relative to the non-crisis regime. Ó 2012 Elsevier Ltd. All rights reserved. Key words — trade liberalization, growth, crises

1. INTRODUCTION

used in the literature (output falls, inﬂation increases, exchange rate depreciations, increased external debt to export ratios, and increased current account deﬁcits), which we are also able to combine into two factors roughly representing the internal and external dimensions of a crisis. We employ threshold regression techniques on our crisis indicators to identify the relevant “crisis values” and the diﬀerential post-liberalization growth eﬀects in the crisis and non-crisis regimes. Our results indicate that an economic crisis at the time of liberalization does aﬀect post-liberalization growth, with the direction of the eﬀect depending on the nature of the crisis. An internal crisis implies lower growth and an external crisis higher growth relative to the non-crisis regime. The remainder of the paper is organized as follows. Section 2 reviews the theoretical and empirical literature linking crises, liberalization, and growth. Section 3 discusses data, methodology, and long-run results, while Section 4 adds in short-run eﬀects. Section 5 extends our analysis and examines robustness of our main results, and Section 6 concludes.

Is an economic crisis a good or a bad time for a country to undertake trade liberalization? This is a question to which policymakers need an answer, since an economic crisis is often a politically convenient time to undertake economic reforms because the policy status quo is clearly unsustainable. But while immediate policy reforms in some areas are clearly called for, it is not obvious that the reform package should include signiﬁcant trade liberalization, though it often does. Here we present evidence that an economic crisis at the time of trade liberalization does aﬀect a country’s post-liberalization growth performance. Furthermore, its eﬀects depend on the characteristics of the crisis. Trade liberalizations have been widespread in the last three decades, particularly among developing and transition countries. The reasons for this include the perceived limitations of import substitution as a development strategy 1; the weight of empirical evidence suggesting a positive relationship between openness and growth 2; and, not least, the inﬂuence of the International Financial Institutions (IFIs—World Bank and IMF) which often required that trade liberalization be included as part of a package of reforms when agreeing to loans. 3 Despite their early promise, recent experience and evidence suggests not all trade reforms have been as successful as anticipated (Singh, 2010). This is partly attributable to weaknesses in reform packages themselves, including inappropriate timing and sequencing of reforms, their lack of credibility to private agents and doubts over commitment shown by some political actors. In many cases it seems a crisis was necessary to trigger the reforms. Could it be, therefore, that an economic crisis is an unfortunate time to undertake trade reforms? In this paper we examine whether the extent and type of economic crisis at the time of liberalization aﬀects post-liberalization growth in a panel of 75 countries using annual data over the period 1960–2003. We consider ﬁve crisis indicators commonly

2. BACKGROUND TRADE LIBERALIZATION AND GROWTH The potential growth eﬀects of trade liberalization are well known. 4 While the immediate impact is likely to be negative as resources become redundant in areas of comparative * The authors gratefully acknowledge helpful comments from two anonymous Referees. Falvey and Greenaway acknowledge ﬁnancial support from The Leverhulme Trust under Program Grant F/00/114/AM. Comments from participants at the Conference on Economic, Social and Environmental Consequences of the Liberalisation of Trade in North Africa and the Middle East in Rabat, Morocco, 2007 are gratefully acknowledged. Final revision accepted: March 27, 2012. 2177

2178

WORLD DEVELOPMENT

disadvantage, their eventual reallocation into areas of comparative advantage will see a rise in the growth rate in the medium run as income moves to a higher steady state level. 5 Longer run gains in the growth rate must come through improvements in factor productivity and these can emerge through a variety of channels. Increased imports of capital and intermediate goods not available domestically may directly raise the productivity of manufacturing production (Lee, 1995) and increased trade (exports and imports) with advanced economies could indirectly raise growth by facilitating knowledge and technology spillovers. Learning by doing may be more rapid in export industries. 6 A liberal trading regime may attract export-platform FDI. The magnitude of these long-run growth eﬀects will vary across countries, depending on their sectors of comparative advantage in particular. While the empirical literature on openness and growth is voluminous (Dollar, 1992; Sachs & Warner, 1995; and Frankel & Romer, 1999 are prominent examples) that on trade liberalization and growth is more limited. Some comparative cross-country studies have been undertaken, including Little, Scitovsky, and Scott (1970), Krueger (1978), Bhagwati (1978) and Papageorgiou, Michaely, and Choksi (1991) (PMC). The latter is the most sanguine, concluding that trade liberalization results in a more rapid growth of exports and GDP, without signiﬁcant transitional costs of unemployment. 7 Other studies ﬁnd liberalization leads to growth in exports and improvement in the current account (although some of this is because of import compression), and that while some countries have increased investment following liberalization, others suﬀer an investment slump. So the impact on growth may be positive or negative, although there seem to be more cases of a positive than negative growth eﬀect (Greenaway, 1998). Econometric studies are relatively more plentiful 8. Greenaway, Leybourne, and Sapsford (1997) use a smooth transition model to test for a transition in the level and trend of real GDP per capita for 13 countries in the PMC sample and relate these to liberalization. While all displayed a transition in level or trend, in the majority it was negative, 9 and where it was positive it generally could not be related to liberalization episodes. 10 Greenaway, Morgan, and Wright (1998, 2002) (GMW) use a dynamic panel model to examine both the short- and long-run impact of liberalization on growth in a large sample of countries. Results using three measures of liberalization suggest a J-curve eﬀect, growth at ﬁrst falls but then increases after liberalization. Wacziarg and Welch (2008) update the Sachs and Warner (1995) indicator of liberalization, and regress per capita output growth on country (and time) ﬁxed eﬀects and their indicator of liberalization. They ﬁnd the diﬀerence in growth between a liberalized and non-liberalized country is 1.53% points. Salinas and Aksoy (2006) use an alternative indicator 11 and ﬁnd trade liberalization increases growth by between 1% and 4%. Although the later empirical evidence provides broad support for the hypothesis that trade liberalization improves economic growth, this support is far from universal and it is clear some liberalizations have been more successful than others. Given the variety of circumstances under which trade liberalizations have occurred this is hardly surprising. Where liberalizations have been the outcome of a speciﬁc policy review process, have had broad political support, and been undertaken in a stable economic and political environment they are likely to be sustained and successful. But in many cases liberalizations have been undertaken as part of a “package” of reforms emerging from an economic or political crisis. Crises appear to facilitate some reforms. 12 Drazen and Grilli (1993) model a “war-of-attrition” in an economy that

has settled into a Pareto–inferior equilibrium, and where reforms are resisted because of uncertainty over who is more willing to bear the costs. An economic crisis may then help to move the economy to a welfare-superior path, as reforms that would be resisted under normal circumstances, may be accepted if the losses from a continuing crisis are large. Such an approach seems particularly promising for explaining macroeconomic stabilizations, where the distribution costs are low and there is likely to be consensus on the policies required, and this is conﬁrmed by the empirical evidence (see for example, Bruno, 1996; Bruno & Easterly, 1996; Drazen & Easterly, 2001; Alesina, Ardagna, & Trebbi, 2006). But with structural reforms (e.g. trade and labor market reforms) the distributional costs are higher and there is a lower likelihood of consensus on the appropriate policies (Rodrik, 1996). The empirical evidence on whether crises facilitate structural reforms is correspondingly less decisive. Lora (1998) ﬁnds empirical support (in Latin America) for the hypothesis that a crisis involving a decline in real income is likely to facilitate trade reforms, although he notes that the eﬀect is quantitatively small. Tornell (1998) presents empirical evidence on the relationships among drastic political change, a major economic crisis (measured by inﬂation and a decline in output) and trade liberalization. Using Probit models explaining the start of liberalization he ﬁnds that the unconditional probability of reform is 2.7%, increasing to 27% with an economic crisis and 60% with both an economic and political crisis. Campos, Cheng, and Nugent (2010), however, ﬁnd that, unlike political crises, economic crises have no signiﬁcant impact on the implementation of reforms. Even if an economic crisis facilitates structural reforms in general, it need not be a good time to undertake trade liberalization; for two reasons. First, trade reform works by correcting distortions in relative prices, but high and variable inﬂation can confound price signals, making it diﬃcult to disentangle relative price changes from changes in the general price level, thereby blunting incentives to reallocate resources (Rodrik, 1989a). Moreover, the slowdown in domestic activity associated with crises can exacerbate transitional unemployment as resources shift between sectors, increasing opposition to reforms and increasing the likelihood they will be reversed. Second, if trade liberalization is to be successful (and sustained), the private sector must respond to changed incentives, and if private agents are sceptical of policymakers’ commitment, they will be slow to incur the (sunk) costs associated with shifting resources between import competing and export sectors. Short-run adjustment will be prolonged and eﬃciency gains delayed. In such a situation there will be few gainers from liberalization, while some will lose due to increased foreign competition. Such an outcome is likely to make it politically diﬃcult to sustain reforms as well as limiting their impact. Thus scepticism on behalf of the private sector may be more likely for liberalizations undertaken in times of crisis. This may be compounded if liberalization is undertaken as part of a package of reforms that countries were obliged to negotiate to secure ﬁnancial support from the IFIs (Rodrik, 1989b). In the absence of a crisis and conditions requiring trade reform laid down by IFIs, it would be clear to the private sector that a government that undertook liberalization would be committed to the reforms. In the presence of intervention from IFIs however, there is an incentive for uncommitted governments to undertake reform temporarily to receive funds. In this situation it is diﬃcult for the private sector to distinguish between a government committed to reform and one undertaking reform for ﬁnancial gain. 13

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

These considerations combine to suggest that a trade liberalization undertaken at a time of crisis may reﬂect weaker commitment from policymakers and higher scepticism from private agents. If so it will be less likely to be sustained and successful, and less likely to have a signiﬁcant growth promoting impact. The nature of the crisis itself may also be important. A severe “internal” crisis (falling output and high and variable inﬂation) will distort price signals and delay growth enhancing beneﬁts. A severe “external” crisis (currency depreciation, growing current account deﬁcit, and high debt to export ratio) will also constrain growth and is more likely to lead a “not otherwise reform minded government” to undertake reforms to obtain support from IFIs. Of course the trade liberalization itself will eventually free up these constraints, particularly if the external crisis occurs in a highly inwardlooking policy regime. In practice an economic crisis will exhibit both internal and external symptoms, which is why we include indicators of both in our analysis. One limitation of our analysis should be made clear at the outset. Ideally we would extend it by employing a measure of the strength or depth of liberalization to investigate empirically how it is related to the presence or absence of an economic crisis at the time it was undertaken. We could then shed further light on why liberalizations might have generated diﬀerent growth outcomes. Unfortunately the multidimensional nature of trade policy, in terms of the numbers of both products traded and types of policies, precludes us from doing this in any systematic fashion. Aggregate measures of even direct trade policy instruments (tariﬀs and quotas) are diﬃcult to formulate (Anderson & Neary, 2005), and even if we could, the time series data on trade policy measures required for quantiﬁcation is not available covering liberalizations for an adequate sample of countries. Given these constraints, our analysis is largely restricted to whether crises at the time of liberalization have implications for subsequent growth, and leaves the investigation of why they might have such implications to future research once data becomes available. 3. DATA, METHODOLOGY, AND INITIAL RESULTS The starting point for our empirical analysis is an equation similar to the initial regression estimated by GMW (2002) 14: D ln y i;t ¼ b1 ln Y i;60 þ b2 SYRi;60 þ b3 D ln POP i;t þ b4 D ln TTI i;t INV þ b5 þ dLIBi;t þ gt þ ei;t ð1Þ GDP i;t where i denotes country and t time; and yi,t is the GDP per capita; yi,60 is the GDP per capita in 1960 (the base period); SYRi,60 is the Average years of secondary schooling in the population over 15 in 1960; TTI is the Terms of trade index; POP is the Population; INV/GDP is the Ratio of gross capital formation to GDP, and LIB is the Dummy variable taking the value one for all years after and including the year of liberalization and zero otherwise. We estimate this using annual data for a panel of (up to) 75 countries within the period 1960–2003. Much of the data are from the World Bank’s World Development Indicators (2005) database; including GDP, population, investment, and the terms of trade. Data on schooling are from Barro and Lee (2001). The indicator of trade liberalization is from Wacziarg and Welch (2008), 15 and is a broad measure. That the average growth experience is likely to be higher post-liberalization is indicated by Figure 1 which plots the average annual growth rates of all countries in our sample for the 10 years pre- and

2179

post-liberalization. Given the diﬀerent timings of liberalizations, it should be clear that these averages are based on a different number of observations for each period. Also reported (the dotted lines) are the average growth rates over the 10 years pre- and post liberalization. 16 Figure 1 shows that average growth following liberalization (around 2%) tends to be higher than that prior to liberalization (around 0.7%), consistent with the econometric results we report below. The results of estimating Eqn. (1) are reported in Table 1. 17 The ﬁrst regression is our base speciﬁcation, excluding the liberalization dummy. The outcomes for the control variables are largely in line with existing results, particularly those reported by GMW (2002). We ﬁnd negative and signiﬁcant coeﬃcients on initial GDP per capita and population growth, and positive and signiﬁcant coeﬃcients on initial schooling, investment, and the terms of trade index. In regression 2 we add the liberalization dummy. This leaves the control variables largely unchanged, with the liberalization dummy itself positive and signiﬁcant. The estimated coeﬃcient indicates that liberalization has a favorable impact on growth of around 2% in the years following it, in line with estimates reported by GMW (1998 and 2002), Wacziarg and Welch (2008) and Salinas and Aksoy (2006). One limitation of estimating (1) is that data constraints mean that only 39 out of the 75 countries (and only 952 observations out of a potential 2767) are included. Three variables are responsible for this: initial output per capita, initial schooling, and the terms of trade index. We therefore drop initial output per capita and schooling in regressions 3 and 4, replacing them with a full set of country dummies. Including country ﬁxed eﬀects allows us to drop time invariant variables, with the country dummies capturing the impact of country-speciﬁc factors on growth, including initial levels of output per capita and schooling. The estimated coeﬃcients on the remaining control variables are largely unaﬀected, as is the liberalization dummy which remains highly signiﬁcant. Finally, in regressions 5 and 6 we drop the terms of trade variable, which increases our sample to 2619, and allows inclusion of all 75 countries. 18 This lowers the coeﬃcient on population, which also becomes insigniﬁcant, but has little impact on investment. That on liberalization increases in size but is still within the range of estimates in the literature, and is again highly signiﬁcant. The regressions in Table 1 give an estimate of the average impact of trade liberalization on growth across all liberalizing countries. Using the ﬁnal regression (6) as a base, we now explore whether these growth eﬀects diﬀer depending on: (a) whether the country faced an economic crisis at the time of liberalization; and (b) if it did, the nature of the crisis. Several variables have become standard indicators of aspects of an economic crisis (Alesina et al., 2006; Campos et al., 2010): the proportional decline in per capita GDP (OUT), the inﬂation rate 19 (INF), the nominal exchange rate (XR), the ratio of debt to exports (DEBT), and the current account deﬁcit (CAD). Data are again taken from the World Development Indicators (2005) database. 20 Each represents a speciﬁc aspect of a crisis. Individually they are informative, but will be even more so if they can be combined in some way. In particular it is of interest whether the internal or external dimensions of an economic crisis at the time of liberalization have diﬀerent implications for a country’s subsequent growth performance. Factor analysis is a method of condensing a number of random variables into a smaller number of uncorrelated variables. 21 The ﬁrst factor accounts for as much of the variability in the data as possible, and each succeeding factor accounts for as much of the remaining variability as possible. We implement the factor analysis procedure using the original

2180

WORLD DEVELOPMENT

Figure 1. Average growth before and after liberalisation.

Table 1. Initial results ln y ln Y60 INV/GDP Dln POP SYR60 D ln TTI

1

2

0.005 (3.18)*** 0.26 (10.42)*** 0.54 (3.39)*** 0.005 (2.07)** 0.02 (1.92)*

0.005 (3.12)*** 0.23 (9.20)*** 0.55 (3.43)*** 0.008 (0.31) 0.02 (1.83)* 0.02 (5.47)*** Yes No 952 310.95*** 0.36

LIB Time dummies Country dummies Observations F-Statistic R2

Yes No 952 26.23*** 0.34

3

4

5

6

0.27 (7.73)*** 0.67 (1.87)*

0.26 (7.45)*** 0.74 (2.03)**

0.21 (7.35)*** 0.36 (1.36)

0.20 (6.95)*** 0.28 (1.06)

0.02 (2.31)**

0.019 (2.22)** 0.018 (4.78)*** Yes Yes 1327 206.43*** 0.36

Yes Yes 2619 13.02*** 0.27

0.028 (7.24)*** Yes Yes 2619 13.57*** 0.29

Yes Yes 1327 15.25*** 0.34

Notes: t-Statistics in brackets. All models estimated using white heteroscedasticity-consistent standard errors. * Signiﬁcance at the 10% level. ** Signiﬁcance at the 5% level. *** Signiﬁcance at the 1% level.

data on our ﬁve crisis variables and employing the maximum likelihood factor method. The results yield two retained factors, with the rotated factor loadings as reported in Table 2. While there cannot be said to be a deﬁnitive separation of variables, the ﬁrst factor (which explains over 80% of the variance in the variables) has its largest positive weightings on OUT, INF and, to a lesser extent DEBT, while the second (which exTable 2. Rotated factor loadings

OUT INF XR DEBT CAD

plains the remainder of the variance) has its largest positive weightings on CAD and XR. In what follows we therefore label the ﬁrst factor INT and interpret it as an indicator of the internal dimension of the crisis, and the second EXT and interpret it as an indicator of its external dimension. For each crisis indicator we calculate a standardized score as: CRISISjit ¼

Factor 1 [INT]

Factor 2 [EXT]

0.497 0.466 0.023 0.216 0.094

0.112 0.077 0.150 0.033 0.282

Combined the two factors account for all of the variance in the crisis variables, with INT accounting for 82% of the variance of the crisis variables, and EXT 18%.

X jit X jit sjit

ð2Þ

where Xjit is the value of indicator j in country i in period t, X jit is the average of this indicator over the ﬁve years up to and including t, and sjit is the standard deviation of the indicator over this ﬁve year period. 22 The interpretation of the standardized score is straightforward, and standardized scores can be compared since converting our data to scores results in a distribution with mean 0 and standard deviation 1. A standardized score of 0.5, for example, indicates that, at the

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

time of liberalization, the value of this indicator was half a standard deviation above its recent average. Given the way the indicators are deﬁned, higher values indicate a deeper crisis. Preliminary evidence that a crisis at the time of liberalization may have implications for subsequent growth performance is given in Figure 2. Here we have divided the liberalizing countries into crisis and non-crisis samples according to an arbitrary threshold of 0.5 for each indicator. As in Figure 1, this reports the average growth rate across all countries for each year pre and post liberalization (i.e. 10 years before and 10 years after), along with the average growth rates over the 10 years pre and post liberalization. The solid lines represent non-crisis countries and dashed lines crisis countries. 23 Overall, the econometric results we report below are nicely reﬂected in this alternative framework. For OUT, INF, XR, and INT the growth rate in the post-liberalization period is higher for non-crisis countries, while the reverse is true for the CAD, DEBT, and EXT. The ﬁgures also indicate that growth following liberalization is higher than that prior to liberalization for both regimes. Of course these indicators only signal a “crisis” if their value exceeds some positive threshold, which is itself unknown, a priori. But our interest is not simply in what threshold might be said to indicate a crisis. Rather we are concerned with what threshold indicates a crisis of suﬃcient magnitude that it has implications for the liberalizer’s subsequent growth 24. To determine this we employ the panel threshold regression model of Hansen (1999), and estimate thresholds for our crisis indicators that allow the coeﬃcient on the liberalization dummy to vary discretely depending upon the value of the crisis indicator at the time of liberalization. In general terms, the model for a single threshold can be written as: y i ¼ a þ d1 xi þ ei

qi 6 k1

y i ¼ a þ d2 xi þ ei

qi > k1

where qi is the threshold variable (a crisis indicator in our case) and k1 is the estimated threshold. We can write this as a single regression of the following form: y i ¼ a þ d1 Iðqi 6 k1 Þxi þ d2 ðqi > kÞxi þ ei

ð3Þ

where I(.) is the indicator function, taking the value one if the argument is true and zero otherwise. Here the observations are divided into two regimes depending on whether the threshold variable is smaller or larger than k1. The two regimes are distinguished by diﬀerent regression slopes, d1 and d2. Chan (1993) and Hansen (1999) recommend estimation of k1 by least squares. In practice this involves searching over distinct values of qi for the value of k1 at which the sum of squared errors is smallest, which is then our estimate of the threshold. Once we have an estimate for the threshold it is straightforward to estimate the equation. Having found a threshold it is important to determine whether it is statistically signiﬁcant or not. In essence, this boils down to testing whether the threshold regression model is preferred to the linear model, by testing the null hypothesis that d1 = d2. If we cannot reject this null then the regression model in (2) collapses to the linear model and the threshold is considered not signiﬁcant. 25 A complication with this test is that the threshold is not identiﬁed under the null hypothesis, implying that classical tests do not have standard distributions and critical values cannot be read oﬀ standard distribution tables. Hansen (1996) suggests bootstrapping to simulate the asymptotic distribution of the likelihood ratio test allowing one to obtain a p-value for this test. Firstly, one estimates

2181

the model under the null (i.e. linearity) and alternative (i.e. threshold occurring at k). This allows one to construct the actual value of the likelihood ratio test (F1): F1 ¼

S 0 S 1 ðk1 Þ r2

where r2 ¼

1 S 1 ðk1 Þ nðt 1Þ

Here S0 and S1 are the residual sum of squares from the linear and threshold models, and n and t are the number of crosssection units and time periods respectively. Using a parametric bootstrap (see Cameron & Trivedi, 2005) the model is then estimated under the null and alternative and the likelihood ratio F1 is calculated. This process is repeated a large number of times. The bootstrap estimate of the p-value for F1 under the null is given by the percentage of draws for which the simulated statistic F1 exceeds the actual one. 26 This method can be extended to consider more than one threshold. 27 In terms of our regression speciﬁcation, the single threshold (i.e. two-regime) model can be expressed as: INV D ln y i;t ¼ b3 D ln POP i;t þ b5 GDP i;t þ d1 LIBi;t IðCRIS jiLIB 6 kj Þ þ d2 LIBi;t IðCRIS jiLIB > kj Þ þ ti þ gt þ ei;t ð4Þ Here the observations are divided into two regimes depending upon whether the value of the crisis indicator at the time of liberalization (CRISjiLIB) is smaller or larger than the estimated threshold for that indicator (kj). The impact of liberalization on growth will be given by d1 for observations in the low (“non-crisis”) regime (CRISjiLIB 6 kj) and by d2 for observations in the high (“crisis”) regime (CRISjiLIB > kj). To estimate (3) we ﬁrstly have to estimate the threshold parameter which is taken as the value that minimizes the concentrated sum of squared errors from the least squares regression. To allow us to concentrate on crises we impose the restriction that the threshold must be positive. 28 The results for a single threshold for each indicator are presented in Table 3A, where LIB1 and LIB2 refer to the coeﬃcients on liberalization in the non-crisis and crisis regimes, respectively. In this table k1 refers to the estimated threshold on each of the crisis variables, with the ﬁgure in brackets reporting the percentile of the distribution at which the threshold lies. The row entitled p-value reports the p-value from the bootstrap procedure used to test whether LIB1 = LIB2 (i.e. whether d1 = d2). Despite the variety of indicators used, deﬁnite patterns can be discerned. First, there is at least one significant crisis threshold for all indicators, and in the majority of cases these thresholds are less than unity and all are less than the values (1.5 or 2) commonly imposed in the literature. 29 This suggests that less severe crises may be more important than normally thought. Second, trade liberalization raises growth in both crisis and non-crisis regimes, consistent with the results in Figure 2. Third, the individual indicators fall into two groups in terms of their predictions of the sign of the eﬀect of a crisis on subsequent growth. Liberalizing during a time of crisis involving above threshold falls in output, increases in inﬂation, or depreciations of the exchange rate is associated with lower subsequent growth, while liberalising during a crisis involving above threshold increases in the debt to export ratio or the current account deﬁcit is associated with enhanced subsequent growth. These results lend support to the arguments, noted above, that liberalization at a time of high inﬂation or unemployment will reduce subsequent growth beneﬁts by masking relative price signals and delaying resource reallocations. They also support a view that trade liberalization is more

2182

WORLD DEVELOPMENT

Non-crisis countries Crisis countries

Figure 2. Growth comparison with exogenously chosen crises.

eﬀective when countries are subject to external constraints as indicated by DEBT and CAD crises. These issues are investigated further in the short-run analysis of the next section. Evidence that diﬀerent dimensions of an economic crisis may have diﬀering implications for subsequent growth rates

reinforces our interest in exploring their combined eﬀects through our two estimated factors (INT and EXT). The single threshold results indicate that liberalization during an internal crisis (INT above its threshold) is associated with dampened growth, while liberalization during an external

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

2183

Table 3A. Endogenous threshold results Dln y

INV/GDP Dln POP LIB1 LIB2 k1 p-Value Observations F-Statistic R2

Crisis indicator OUT

INF

XR

DEBT

CAD

INT

EXT

0.19 (6.25)*** 0.28 (0.99) 0.038 (7.87)*** 0.025 (4.60)*** 0.05 (66th) 0.009*** 2494 12.18*** 0.28

0.20 (6.45)*** 0.23 (0.82) 0.043 (7.85)*** 0.022 (4.85)*** 0.09 (54th) 0.00*** 2458 12.34*** 0.29

0.21 (7.15)*** 0.39 (1.46) 0.034 (5.73)*** 0.024 (5.80)*** 0.9 (34th) 0.045** 2384 13.50*** 0.29

0.24 (7.43)*** 0.37 (1.05) 0.023 (4.97)*** 0.50 (4.43)*** 1.34 (88th) 0.00*** 1890 9.30*** 0.24

0.22 (6.38)*** 0.54 (1.78)* 0.030 (5.62)*** 0.051 (4.33)*** 1.09 (90th) 0.047** 1961 8.72*** 0.25

0.25 (7.34)*** 0.52 (1.56) 0.027 (5.50)*** 0.012 (2.00)** 0.51 (75th) 0.005*** 1774 10.25*** 0.28

0.26 (7.40)*** 0.46 (1.34) 0.020 (3.98)*** 0.033 (6.33)*** 0.78 (72nd) 0.018** 1774 10.62*** 0.28

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1996). * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

crisis (EXT above its threshold) is associated with ampliﬁed growth. We next use the two independently estimated thresholds to construct four separate liberalization dummy variables, each reﬂecting one of the four possible situations at the time of liberalization: LIB(N, N) no crisis, LIB(E, N) an external but no internal crisis, LIB(N, I) an internal but no external crisis, and LIB(E, I) a crisis in both dimensions. The results are shown as regression 1 in Table 3B. The strongest growth eﬀects arise when the EXT indicator is above its threshold (the coeﬃcient on LIB(E, N) is signiﬁcantly diﬀerent from the coeﬃcients on LIB(N, N) and LIB(N, I), but not that on LIB(E, I)). Liberalization in the absence of a crisis is also associated with signiﬁcant

Table 3B. Endogenous threshold results

INV/GDP Dln POP LIB(N, N) LIB(N, I) LIB(E, N) LIB(E, I) k1(EXT) k1(INT) p-Value Observations F R2

1

2

3

0.26 (7.40)*** 0.51 (1.53) 0.025 (4.52)*** 0.009 (1.49) 0.033 (6.23)*** 0.029 (2.93)*** 0.78 0.51 N/A 1744 10.42*** 0.28

0.26 (7.42)*** 0.51 (1.52) 0.025 (4.52)*** 0.009 (1.49) 0.033 (6.34)***

0.26 (7.40)*** 0.45 (1.34) 0.020 (3.97)***

0.78 0.51 0.013** 1744 10.52*** 0.28

0.032 (6.13)*** 0.039 (3.35)*** 0.78 0.00 0.584 1744 10.68*** 0.28

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1996). ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

growth eﬀects, but liberalization when there is an internal but no external crisis, has no signiﬁcant implications for subsequent growth. While these results are interesting and suggestive, they are based on dummy variables deﬁned by two thresholds each estimated ignoring the other. Our ﬁnal step therefore is joint estimation of these thresholds. In view of the apparent importance of the EXT indicator, we use the estimated threshold on EXT to divide the sample into two regimes (EXT above and below the threshold at the time of liberalization) and sequentially search for independent thresholds on INT in each regime. The outcomes are shown in the ﬁnal two columns in Table 3B. There is one signiﬁcant second threshold—indicated by the signiﬁcant p-value in Column 2, that on INT in the low (non-crisis) regime for EXT. Its value is the same as the separately estimated threshold for INT, and the results are virtually identical to those in the second column as a consequence. In combination, these results support the view that an economic crisis can have a signiﬁcant impact on post-liberalization growth. In particular, liberalization at a time of internal economic crisis does not appear to yield subsequent growth beneﬁts of the same magnitude as those found with liberalization in the absence of a crisis or where an external crisis is also present. This is consistent with the discussion in Section 2 which suggested that an internal crisis would very likely hamper and obscure the potential beneﬁts of a trade liberalization. 4. SHORT-RUN IMPACTS OF LIBERALIZATION ON GROWTH Given our limited sample sizes, our results are likely to reﬂect a combination of short and longer run inﬂuences. They could therefore be viewed as suggesting that the detrimental eﬀects of an internal crisis at the time of liberalization go beyond the short-run. As mentioned above, GMW (1998, 2002) found evidence of a J-curve eﬀect, whereby growth initially declines or remains stable following liberalization, and then increases after a period. We now modify their approach to consider three issues: ﬁrst, whether a similar short-run relationship holds for our sample; second, whether inclusion of short-run eﬀects disturbs our threshold estimates for the

2184

WORLD DEVELOPMENT

long-run growth relationship; third, whether any short-run growth eﬀects of trade liberalization are also crisis dependent. As a ﬁrst step to capturing both the short-run and long-run eﬀects we estimate: INV þ dLRi;t D ln y i;t ¼ b3 D ln POP i;t þ b5 GDP i;t þ /j

3 X

SRðjÞi;t þ ti þ gt þ ei;t

ð5Þ

j¼0

Alongside the long-run (post-) liberalization dummy described above (now relabeled LR), this equation includes four additional liberalization dummies, each corresponding to a single year—the year of liberalization (SR(0)) and each of the subsequent three years (SR(1), SR(2) and SR(3)). The impact on growth in the year of liberalization and in each of the subsequent three years is therefore given by d + uj:j = 0, . . ., 3. The results are shown in the second column of Table 4A. Estimated coeﬃcients on INV/GDP, D ln POP, and LR are very similar to those in the corresponding regression in Table 1. The estimates for the short-run post-liberalization dummies indicate that growth is signiﬁcantly lower than the post-liberalization average in the year of liberalization, is no diﬀerent from this average in the following 2 years, and is suﬃciently higher in the third year to recover what had been lost in the year of liberalization. Our sample thus replicates the type of J-curve eﬀects found previously. To begin the process of examining how these results are affected by a crisis, we initially used a modiﬁed version of Eqn. (4) which estimated common crisis thresholds for all ﬁve postliberalization dummies. The broad pattern of outcomes remained as before, but for three of the indicators we now have a signiﬁcant second threshold. In the light of this evidence that

diﬀerent crisis levels may be applicable to the short and longrun growth eﬀects 30, we proceeded in two steps. First, we estimated crisis thresholds for the long-run dummies in (4) only, applying no thresholds on the short-run dummies. The results are shown in the remaining columns of Table 4A. The estimated thresholds for the crisis indicators are identical to those of the preceding section, and coeﬃcients on the long-run postliberalization dummies are the same or slightly higher in both regimes. For all the single indicators (except DEBT), the estimated coeﬃcients on the short-run dummies show the same J-curve pattern as the linear case. However, there is enough variation in the eﬀects of these individual indicators that when they are aggregated (along with DEBT) into the combined indicators no signiﬁcant short-run eﬀects are evident. Our second step involves estimating crisis-indicator-based thresholds for the short-run post-liberalization dummies, taking as given the estimated thresholds for the long-run dummies. The equation estimated is: INV þ d1 LRi;t IðCRIS ijLIB 6 kLj Þ Dlny i;t ¼ b3 DlnPOP i;t þ b5 GDP i;t ! 3 X þ /1;j SRðjÞi;t IðCRIS ijLIB 6 kSj Þ j¼0

kLj Þ þ d2 LRi;t IðCRIS ijLIB > ! 3 X /2;j SRðjÞi;t IðCRIS ijLIB > kSj Þ þ ti þ gt þ ei;t þ j¼0

ð6Þ where kLj is the long-run threshold for crisis indicator j as reported in Table 4A. The results are shown in Table 4B.

Table 4A. Endogenous threshold results (long-run threshold only) Linear

INV/GDP D ln POP LR1

0.20 (6.75)*** 0.30 (1.12) 0.029 (7.13)***

LR2 SR(0)1 SR(1)1 SR(2)1 SR(3)1 k1 p-Value Observations F R2

0.022 (2.73)*** 0.001 (0.17) 0.003 (0.62) 0.021 (4.32)***

2619 13.31*** 0.30

Crisis indicator OUT

INF

XR

DEBT

CAD

INT

EXT

0.19 (5.95)*** 0.28 (1.01) 0.042 (8.03)*** 0.029 (5.14)*** 0.027 (3.38)*** 0.002 (0.26) 0.001 (0.13) 0.017 (3.16)*** 0.05 0.00*** 2494 11.95*** 0.29

0.19 (6.12)*** 0.24 (0.84) 0.048 (8.29)*** 0.026 (5.29)*** 0.029 (3.47)*** 0.002 (0.34) 0.001 (0.27) 0.015 (2.92)*** 0.09 0.00*** 2458 12.16*** 0.30

0.20 (6.90)*** 0.40 (1.48) 0.036 (5.83)*** 0.026 (5.37)*** 0.016 (2.53)** 0.002 (0.29) 0.001 (0.09) 0.015 (2.79)*** 0.9 0.047** 2384 13.20*** 0.29

0.23 (7.23)*** 0.36 (1.02) 0.025 (4.42)*** 0.051 (4.66)*** 0.009 (1.41) 0.002 (0.22) 0.004 (0.77) 0.009 (1.64) 1.34 0.00*** 1890 9.03*** 0.24

0.22 (6.16)*** 0.54 (1.81)* 0.034 (5.54)*** 0.056 (4.70)*** 0.019 (2.67)*** 0.005 (0.65) 0.0003 (0.04) 0.01 (1.68)* 1.09 0.004*** 1961 8.53*** 0.25

0.25 (7.19)*** 0.53 (1.57) 0.029 (5.02)*** 0.013 (2.01)** 0.010 (1.51) 0.0002 (0.03) 0.002 (0.32) 0.009 (1.63) 0.51 0.007*** 1774 9.92*** 0.28

0.25 (7.25)*** 0.47 (1.39) 0.022 (3.83)*** 0.035 (5.82)*** 0.01 (1.60) 0.0005 (0.07) 0.002 (0.44) 0.009 (1.53) 0.78 0.019** 1774 10.25*** 0.28

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). LRI and SR(J)I refer to the long-run and short-run liberalisation dummies in regime I = 1–2. * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

2185

Table 4B. Endogenous Threshold Results (Short-run Thresholds) Crisis Variables

INV/GDP D ln POP LR1 LR2 SR(0)1 SR(0)2 SR(1)1 SR(1)2 SR(2)1 SR(2)2 SR(3)1 SR(3)2 kLj kS p-Value (kS) Observations F-Statistic R2

0.19 (5.97)*** 0.31 (1.11) 0.041 (7.80)*** 0.033 (5.80)*** 0.013 (1.85)* 0.114 (4.44)*** 0.002 (0.28) 0.021 (1.19) 0.001 (0.24) 0.011 (0.79) 0.013 (2.44)** 0.029 (2.40)** 0.05 0.79 0.00*** 2494 11.91*** 0.30

OUT

INF

XR

DEBT

CAD

INT

0.20 (6.15)*** 0.23 (0.80) 0.048 (8.35)*** 0.026 (5.23)*** 0.026 (2.92)*** 0.053 (2.47)** 0.005 (0.72) 0.017 (0.91) 0.005 (1.00) 0.024 (1.27) 0.012 (2.31)** 0.035 (2.27)*** 0.09 1.6 0.131 2458 11.84*** 0.30

0.21 (6.93)*** 0.38 (1.40) 0.037 (6.00)*** 0.025 (5.24)*** 0.01 (1.56) 0.021 (2.07)** 0.007 (0.93) 0.004 (0.44) 0.011 (1.85)* 0.012 (1.72)* 0.0007 (0.12) 0.029 (3.68)*** 0.9 1.4 0.044** 2384 12.84*** 0.30

0.23 (7.24)*** 0.37 (1.04) 0.025 (4.34)*** 0.055 (5.31)*** 0.007 (1.21) 0.021 (0.83) 0.002 (0.33) 0.022 (0.62) 0.003 (0.54) 0.010 (0.81) 0.012 (1.88)* 0.005 (0.47) 1.34 1.3 0.658 1890 8.95*** 0.24

0.22 (6.17)*** 0.53 (1.76)* 0.034 (5.50)*** 0.056 (4.73)*** 0.023 (2.89)*** 0.005 (0.45) 0.006 (0.78) 0.001 (0.11) 0.004 (0.55) 0.014 (1.75)* 0.014 (2.25)** 0.005 (0.43) 1.09 0.59 0.337 1961 8.26*** 0.25

0.25 (7.19)*** 0.59 (1.73)* 0.028 (4.84)*** 0.017 (2.56)** 0.007 (0.96) 0.040 (5.34)*** 0.005 (0.69) 0.010 (0.79) 0.004 (0.61) 0.002 (0.16) 0.010 (1.64) 0.007 (0.65) 0.51 0.17 0.011** 1774 9.92*** 0.29

0.25 (7.21)*** 0.47 (1.39) 0.022 (3.86)*** 0.034 (5.68)*** 0.019 (2.18)** 0.0003 (0.03) 0.0007 (0.07) 0.002 (0.29) 0.0003 (0.04) 0.006 (0.96) 0.012 (1.42) 0.005 (0.73) 0.78 0.45 0.621 1774 9.89*** 0.28

EXT

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). LRI and SR(J)I refer to the long-run and short-run liberalisation dummies in regime I = 1–2. * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

Only two of the individual crisis indicators (OUT and XR) have signiﬁcant short-run thresholds—as signiﬁed by p-value (kS)—and both are higher than their long-run values. The estimated coeﬃcients on the long-run liberalization dummies are largely unaﬀected. The estimated coeﬃcients on the short-run dummies again exhibit a similar J-curve pattern. Compared to the post-liberalization long-run, there is lower growth in the year of liberalization and higher growth three years later. Negative growth in the liberalization year is predicted for countries in the high crisis regimes by the OUT and INF indicators, again conﬁrming concerns that high inﬂation or unemployment may mask relative price signals and delay resource reallocation. For the combined indicators, we ﬁnd a signiﬁcant short-run threshold for INT, at a value below its long-run threshold. But the only J-curve eﬀect evident is lower growth in the short-run crisis regime in the year of liberalization. 31 We can now address the three issues noted at the start of this section. First, our results conﬁrm the presence of the short-run J-curve eﬀects found in the earlier literature. Second, the longrun results are essentially unaﬀected by the allowance for short-run eﬀects. The estimated long-run crisis-thresholds are unchanged. The estimated coeﬃcients tend to be slightly higher, but the pattern is unchanged. Third, there is evidence that the short-run growth J-curve is also crisis sensitive. Output,

inﬂation, or exchange rate crises at the time of liberalization imply lower growth in the liberalization year but a stronger recovery three years later. A current account crisis exhibits the opposite pattern. The only signiﬁcant eﬀects for the combined indicators are in the liberalization year, where there is lower growth with an internal crisis or the absence of an external crisis. 5. ROBUSTNESS AND EXTENSIONS For brevity we generally report the robustness results for the long-run model only. (a) Excluding transition economies Transition countries undertook trade liberalization in the same years as they dismantled their socialist economies, making it hard to identify the eﬀects of trade reform. Results when excluding transition countries are remarkably similar to those in Section 3. We again ﬁnd the impact of liberalization tends to be higher for non-crisis countries when crises are measured using OUT, INF, and XR, though the diﬀerence in coeﬃcients is no longer signiﬁcant for XR. We ﬁnd for DEBT and CAD that liberalization again tends to have a stronger impact in

2186

WORLD DEVELOPMENT

crisis countries as opposed to non-crisis countries, though only in the case of DEBT is the diﬀerence signiﬁcant. Finally, the results on the combined crisis indicators, INT and EXT, are consistent with the results above, with liberalization being associated with a stronger impact on growth in non-crisis countries when INT is our crisis indicator, and crisis countries when EXT is our crisis indicator. (b) Allowing for the eﬀects of crises on growth We now examine the possibility that estimated non-linearities in the liberalization variable simply reﬂect omitted effects related to the crisis variables rather than being attributable to liberalization. Bruno and Easterly (1998) examined countries that had high-inﬂation crises (inﬂation above 40% annually for two or more years) and found that growth fell sharply during the high inﬂation crisis. They further showed that growth after the crisis was higher than

before, even though inﬂation had returned to pre-crisis levels. Easterly (1996) found that GDP growth was already positive in the year that high inﬂation declined from its pre-stabilisation peak, with growth becoming stronger in the periods following. Bruno and Easterly (1996) report that broad reforms were the usual outcome of high-inﬂation crises and show a strong association between their measure of stabilization from high-inﬂation and the openness measure of Sachs and Warner (1995). Table 5A reports the results of adding contemporaneous values of our crisis indicators. 32 The coeﬃcients on the liberalization dummy are robust to their inclusion, ranging from 0.018 to 0.022, and remain signiﬁcant. The coeﬃcients on the crisis indicators are as expected, negative and signiﬁcant, except for CAD where a signiﬁcantly positive coeﬃcient is found. Since a large current account deﬁcit is likely to reﬂect relatively high domestic expenditure which is also likely to generate relatively high output, the positive relationship for CAD is not surprising.

Table 5A. Inclusion of crisis indicator linearly Dln y INV/GDP Dln POP LIB CRISIS Observations F-Statistic R2

OUT

INF

XR

DEBT

CAD

0.19 (8.20)*** 0.12 (0.55) 0.019 (5.92)*** 0.035 (34.48)*** 2454 37.99*** 0.60

0.21 (7.91)*** 0.42 (1.53) 0.022 (5.98)*** 0.012 (9.61)*** 2405 15.38*** 0.33

0.21 (8.69)*** 0.49 (1.98)** 0.018 (5.57)*** 0.004 (3.03)*** 2372 16.81*** 0.34

0.24 (7.11)*** 0.80 (2.11)** 0.019 (5.20)*** 0.006 (5.41)*** 1599 27.34*** 0.34

0.21 (5.67)*** 0.68 (2.11)** 0.022 (5.65)** 0.003 (2.89)*** 1675 14.18*** 0.34

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

Table 5B. Threshold results when including the crisis variables Dln y INV/GDP Dln POP LIB1 LIB2 CRISIS k1 p-Value Observations F-Statistic R2

OUT

INF

XR

DEBT

CAD

0.19 (7.47)*** 0.07 (0.31) 0.023 (5.93)*** 0.012 (2.73)*** 0.04 (33.74)*** 0.17 0.005*** 2355 36.24*** 0.59

0.20 (7.35)*** 0.42 (1.46) 0.032 (6.43)*** 0.015 (3.56)*** 0.012 (9.39)*** 0.03 0.00*** 2310 13.80*** 0.32

0.21 (8.01*** 0.53 (2.07)** 0.027 (4.68)*** 0.019 (5.10)*** 0.004 (2.88)*** 0.7 0.186 2225 15.06*** 0.31

0.25 (6.80)*** 0.92 (2.17)** 0.016 (3.84)*** 0.032 (4.56)*** 0.006 (4.93)*** 1.03 0.014** 1387 10.34*** 0.28

0.22 (5.02)*** 0.93 (2.35)** 0.029 (5.35)*** 0.016 (2.41)** 0.004 (3.05)** 0.69 0.04** 1328 8.72*** 0.27

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

Table 5B reports results from estimating thresholds on liberalization. The coeﬃcients on the crisis indicators are consistent with those reported in the previous table. The pattern of threshold results is similar to that in Section 3 (though the positioning of the threshold often diﬀers). For OUT, INF, and XR we ﬁnd the impact of liberalization to be larger in non-crisis countries, and for DEBT we ﬁnd the

2187

opposite. The major diﬀerence is that for a CAD crisis we now ﬁnd the beneﬁts of liberalization to be greater in non-crisis countries. In results not reported 33 we add to the threshold on liberalization a second threshold on the crisis variable, examining whether the relationship between the crisis and growth depends on the level of the crisis indicator. For OUT, INF,

Table 6A. Linear results with crisis variables included Dln y INV/GDP Dln POP LIB1 CRISPEAK CRISPEAK – 1 CRISPEAK – 2 CRISPEAK – 3 Observations F-Statistic R2

OUT

INF

XR

DEBT

CAD

0.17 (5.63)*** 0.24 (0.36) 0.023 (6.03)*** 0.05 (11.53)*** 0.02 (4.94)*** 0.016 (3.74)*** 0.007 (1.55) 2436 22.86*** 0.35

0.18 (6.18)*** 0.32 (1.18) 0.027 (6.89)*** 0.021 (3.85)*** 0.012 (2.43)** 0.0006 (0.14) 0.001 (0.29) 2436 17.53*** 0.31

0.18 (6.24)*** 0.30 (1.08) 0.028 (6.89)*** 0.012 (2.41)** 0.002 (0.38) 0.008 (2.13)** 0.007 (1.78)* 2436 18.00*** 0.30

0.18 (6.11)*** 0.30 (1.08) 0.028 (6.97)*** 0.002 (0.49) 0.001 (0.26) 0.005 (1.14) 0.002 (0.45) 2436 17.57*** 0.30

0.18 (6.01)*** 0.29 (1.05) 0.028 (6.97)*** 0.005 (1.26) 0.005 (1.01) 0.01 (2.48) 0.006 (1.40) 2436 18.36*** 0.24

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

Table 6B. Long-run threshold results with crisis variables included Dln y INV/GDP Dln POP LIB1 LIB2 CRISPEAK CRISPEAK – 1 CRISPEAK – 2 CRISPEAK 3 k1 p-Value Observations F-Statistic R2

OUT

INF

XR

DEBT

CAD

0.16 (5.08)*** 0.29 (1.02) 0.030 (6.51)*** 0.016 (3.05)*** 0.051 (11.29)*** 0.020 (4.79)*** 0.016 (3.64)*** 0.007 (1.56) 0.17 0.30 2337 18.81*** 0.34

0.18 (5.78)*** 0.33 (1.16) 0.038 (7.23)*** 0.019 (4.25)*** 0.02 (3.63)*** 0.011 (2.20)** 0.0002 (0.05) 0.001 (0.34) 0.15 0.00*** 2301 14.65*** 0.30

0.19 (6.74)*** 0.45 (1.69)* 0.035 (5.15)*** 0.022 (5.62)*** 0.014 (2.74)*** 0.003 (0.67) 0.004 (1.06) 0.003 (0.82) 0.70 0.34 2239 17.31*** 0.31

0.24 (7.31)*** 0.42 (1.08) 0.024 (5.01)*** 0.053 (4.69)*** 0.008 (1.53) 0.004 (0.76) 0.011 (2.33)*** 0.003 (0.82) 1.34 0.00*** 1783 10.23*** 0.25

0.21 (6.22)*** 0.64 (2.12)** 0.028 (5.52)*** 0.036 (5.52)*** 0.002 (0.56) 0.004 (0.82) 0.012 (2.92)*** 0.006 (1.49) 1.06 0.018** 1840 8.72*** 0.26

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

2188

WORLD DEVELOPMENT Table 6C. Endogenous threshold results (long-run threshold only)

Dln y INV/GDP D ln POP LR1 LR2 SR(0)1 SR(1)1 SR(2)1 SR(3)1 CRISPEAK CRISPEAK – 1 CRISPEAK – 2 CRISPEAK – 3 k1 p-Value Observations F-Statistic R2

OUT

INF

XR

DEBT

CAD

0.15 (4.76)*** 0.29 (1.02) 0.036 (7.19)*** 0.023 (4.07)*** 0.028 (3.74)*** 0.007 (1.10) 0.004 (0.84) 0.011 (2.32)** 0.051 (11.28)*** 0.020 (4.77)*** 0.015 (3.55)*** 0.005 (1.20) 0.17 0.41 2337 9.47*** 0.35

0.17 (5.45)*** 0.32 (1.14) 0.043 (7.76)*** 0.024 (4.86)*** 0.026 (3.30)*** 0.005 (0.83) 0.004 (0.80) 0.013 (2.75)*** 0.019 (3.45)*** 0.009 (1.80) 0.001 (0.29) 0.002 (0.52) 0.16 0.00*** 2301 7.83*** 0.31

0.18 (6.51)*** 0.45 (1.68)* 0.038 (5.30)*** 0.026 (5.51)*** 0.018 (2.96)*** 0.004 (0.63) 0.003 (0.66) 0.012 (2.36)** 0.013 (2.60)*** 0.002 (0.52) 0.005 (1.17) 0.003 (0.76) 0.69 0.29 2239 8.09*** 0.31

0.23 (6.90)*** 0.43 (1.11) 0.025 (4.23)*** 0.038 (5.05)*** 0.010 (1.43) 0.002 (0.25) 0.003 (0.58) 0.010 (1.69)* 0.008 (1.36) 0.003 (0.71) 0.011 (2.30)** 0.004 (0.92) 1.34 0.00*** 1783 5.30*** 0.25

0.20 (6.00)*** 0.65 (2.14)** 0.034 (5.53)*** 0.042 (5.68)*** 0.021 (3.11)*** 0.006 (0.80) 0.003 (0.49) 0.007 (1.30) 0.002 (0.48) 0.004 (0.85) 0.012 (2.96)*** 0.007 (1.54) 1.11 0.017** 1840 5.91*** 0.27

Notes: All models include a full set of unreported country and time dummies. t-Statistics in brackets based on white heteroscedasticity-consistent standard errors. The p-value of the signiﬁcance of the estimated threshold is calculated using the bootstrap procedure of Hansen (1999). LRI and SR(J)I refer to the long-run and short-run liberalisation dummies in regime I = 1–2. * Indicate signiﬁcance at the 10% level. ** Indicate signiﬁcance at the 5% level. *** Indicate signiﬁcance at the 1% level.

and DEBT the threshold results indicate that the link between crises and growth is negative in both the low and high crisis regime, but the relationship is stronger in the high crisis regime. This result is therefore similar to that found by Bruno and Easterly, albeit using a diﬀerent methodology and diﬀerent variable deﬁnition. Rather than the rate of inﬂation, our indicator measures diﬀerences in the inﬂation rate relative to its average over the recent past. The results for XR are similar, though the coeﬃcient in the low regime is positive (albeit insigniﬁcant). For CAD we obtain positive coeﬃcients in both regimes, though only signiﬁcant in the low regime and we cannot reject the hypothesis that the coeﬃcients are the same. The results for CAD may indicate that our liberalization results are to some degree capturing a relationship between CAD and growth. While allowing our crisis indicators to have a direct eﬀect on growth appears to have a limited impact on our liberalization results (except for CAD), we should still examine whether they are being driven by the response of the economy to a crisis around the time of liberalization. It could be argued that we already account for this by the inclusion of the short-run liberalization variables in Table 4A. This is only true, however, if liberalization is indeed a normal response to a crisis. Our data indicates this is not the case. For OUT 49 of the 75 observations were negative at the time of liberalization, with the number being 40 out of 74 for INF. While the number of liberalizations that would be considered to have taken place in a non-crisis period are smaller for the other indicators (5

out of 68 for XR, 25 out of 53 for DEBT and 33 out of 57 for CAD), they still suggest that many, if not a majority, of liberalizations would have taken place in non-crisis conditions. To examine whether our results are driven by a natural recovery from the crisis rather than a response to liberalization, we construct a variable that is equal to one at the peak of any crisis (for each of the ﬁve crisis indicators) 34 then calculate lags of it for three periods and include it alongside our liberalization variable. Table 6A reports the linear results, where CRISPEAK is the dummy representing the peak of the crisis. The results are largely consistent with those reported in Table 1, with the coeﬃcient on the liberalization dummy taking a value between 0.023 and 0.028. The coeﬃcients on the crisis indicators tend to be negative and signiﬁcant for the ﬁrst couple of periods for OUT, INF, and XR, steadily decreasing in absolute size and becoming insigniﬁcant (or becoming positive and signiﬁcant) in subsequent years. For DEBT and CAD no signiﬁcant eﬀects are found. Table 6B reports the endogenous threshold results when including these crisis variables. The coeﬃcients are similar to those in Table 5A with negative coeﬃcients in the year of the peak of the crisis for OUT, INF, and XR, with the coeﬃcients declining and becoming insigniﬁcant in subsequent years. For DEBT and CAD the coeﬃcients tend to be insigniﬁcant, with the exception of the second year after the crisis when a signiﬁcant negative coeﬃcient is found. The liberalization variables are qualitatively similar to those reported in Table 3A, with a coeﬃcient that is larger in size in non-crisis

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

countries (LIB1) when OUT, INF, and XR are used as indicators, and larger in crisis countries (LIB2) when DEBT and CAD are used. In the case of OUT and XR however, the differences in the coeﬃcients are no longer found to be statistically signiﬁcant (despite being economically important). Finally, Table 6C reports the results when including shortrun liberalization indicators, which allow us to examine whether the results on the short-run eﬀects are simply reﬂecting adjustment to a crisis, or whether they really capture the shortrun eﬀects of liberalization. The results on the (long-run) liberalization variable are consistent with those reported in Table 5B, as are the coeﬃcients on the crisis peak and its lags. The results show that the impact of liberalization on growth is negative in the year of liberalization (and usually signiﬁcant) and in the following two years (though not signiﬁcant). In the third year the impact of liberalization is positive, and often signiﬁcant. Overall therefore, the J-curve is conﬁrmed even when a measure capturing any crisis is included. 6. CONCLUSIONS Our evidence supports earlier results that trade liberalization increases economic growth in the long-run. We also ﬁnd evidence of signiﬁcant crisis thresholds, at levels below those normally assumed in the literature, for all our crisis indicators. While liberalization leads to higher long-run growth whether there is a crisis or not, the characteristics of the crisis appear to inﬂuence the level of post-liberalization growth. If output is declining, inﬂation is increasing or the exchange rate is depreciating at above threshold levels then liberalization has a diminished impact on subsequent growth. But if the debt to export ratio or the current account deﬁcit is increasing at above crisis levels at the time of liberalization, then liberalization’s impact on future growth will be enhanced. Our composite indicators provided some, albeit tentative support for the notion that an internal crisis tends to dampen the growth effects of trade liberalization, while an external crisis tends to amplify them. The explicit allowance for trade liberalizations to have both short and long-run growth eﬀects did not materially aﬀect our long-run conclusions. The same pattern of coeﬃcients remained, with post-liberalization growth rates estimated to be a little higher if anything. The estimated short-run coeﬃcients generally supported the conclusion of a J-curve eﬀect found in the earlier literature. Our robustness checks showed that this was adjustment to the liberalization and not just recovery

2189

from a crisis. Compared to the post-liberalization average, growth is lower in the year of liberalization, and higher three years later. These short-run eﬀects were also found to be crisis sensitive to some degree, exhibiting a similar pattern to the long-run eﬀects with respect to the individual crisis indicators. Output, inﬂation, or exchange rate crises at the time of liberalization imply lower growth in the liberalization year, but a stronger recovery three years later. A current account crisis shows the opposite pattern. So, is an economic crisis a good or a bad time for a country to undertake trade liberalization? Our results suggest that the answer depends on the nature of the crisis. Liberalizations at a time of external (but not internal) crisis can bring additional growth beneﬁts by alleviating the constraints imposed by the crisis. But liberalizations at a time of internal crisis may exacerbate adjustment problems and discourage the resource reallocations which are necessary for trade liberalization to be successful. The timing of trade liberalization is a policy decision. Since most crises have internal and external elements, the safe option for a government committed to trade liberalization is to undertake it during a period of stability. But some governments desiring to undertake trade liberalization may lack the necessary political support in normal circumstances. If a crisis arises, such governments may be tempted to mobilize the general support for action that the crisis generates behind a trade liberalization. Our results suggest that this strategy is likely to be more successful if the crisis is external rather than internal in nature. Other governments have undertaken liberalizations during a crisis, not necessarily because they wished to do so, but because trade liberalization was a condition imposed for access to World Bank or IMF ﬁnancing. In essence, these international institutions have used the need for funding to induce countries to liberalize. This reﬂects a belief that trade liberalization enhances future growth, among other things, a belief which our evidence supports. But our results also suggest that the timing of the liberalization should be sensitive to the nature of the crisis. Conditionality appears to have been particularly successful where the crisis was largely external in nature. But imposing a trade liberalization during a largely internal crisis may yield lower growth prospects than if the liberalization was undertaken at another time. While it might be tempting to view these as minor, short term considerations, outweighed by the imperative of getting the liberalization process started, our results suggest that these crisis-related eﬀects may extend beyond the short-run, although investigation of the precise channels through which these eﬀects take place awaits the availability of the requisite data.

NOTES 1. This view is not uncontroversial. Rodrik (1999) argues that IS policies worked quite well at least until the mid-1970s and that the poor performance of such countries after 1973 was the result of an inability to respond to macro-shocks and not because of IS policies. Moreover defenders of IS policies argue that it has often been misinterpreted and that it never was a rationale for indiscriminate protection. They also cite evidence of successful selective intervention in some of the so-called liberal trading countries of East Asia (Baldwin, 2003; Rodrik, 1995). 2. Again, this statement is not uncontroversial. Rodrı´guez and Rodrik (2000) criticise much of the existing literature on growth and openness. While not arguing that there is a systematic relationship between inward orientation and growth, they argue that the evidence linking outward orientation and growth is overstated.

3. For the period 1980-89, 79% of all loans had conditions in the trade policy area, in excess of those which attached to any other policy (Greenaway, 1998). 4. Dornbusch (1992) and Krueger (1997) provide useful surveys of the gains from trade liberalization. 5. The static gains from trade liberalization need not be limited to such resource allocation gains. Further gains can arise from reductions in rent seeking, corruption, and smuggling. Other gains include those resulting from economies of scale in exporting industries, reduced market power in protected markets, and increased variety and quality of imported goods available to domestic producers and consumers.

2190

WORLD DEVELOPMENT

6. Indirect evidence suggestive of the importance of learning by doing in export industries is provided by the recent literature on exporting and productivity (for a review see Greenaway & Kneller 2007). 7. Critiques of these results are provided by Collier (1993) and Greenaway (1993). 8. In an important contribution, Wacziarg (2001) investigates the channels through which trade policy may inﬂuence growth. He ﬁnds that an increased share of investment in GDP is responsible for the bulk of the growth enhancing eﬀects, with a smaller role played by international technology transmission and enhanced macroeconomic policy. This work is extended by Falvey, Foster, and Khalid (2011), who allow the eﬀects of trade liberalization to depend on the years since liberalization and to diﬀer across regions. They conﬁrm the importance of the investment channel in the longer term, but show that changes in the government share of GDP are also important and that there are signiﬁcant regional diﬀerences in the way that trade liberalization aﬀects growth. 9. Maurer (1998) ﬁnds in the majority of cases neither a positive nor a negative impact on growth of the liberalization episodes deﬁned by PMC. 10. Greenaway and Sapsford (1994) model liberalization as a discrete break rather than a smooth transition, and again ﬁnd little evidence of liberalization increasing a country’s growth rate. 11. Though their liberalization dates are generally consistent with those of Wacziarg and Welch (2008). 12. This is not surprising according to Rodrik (1996), who states that “There is a strong element of tautology in the association of reform with crisis. Reform naturally becomes an issue only when current policies are not perceived to be working. A crisis is just an extreme instance of policy failure. That policy reform should follow crisis, then, is no more surprising than smoke following ﬁre.” (pp. 26-27). 13. Support from IFIs cannot be taken as a signal of a lack of local commitment, however, since such support can act as an external anchor strengthening the credibility of reforms and providing short-term ﬁnance that can alleviate the short-term costs for governments committed to reform. 14. There are two major diﬀerences between (1) and the static model estimated by GMW. First, we replace the initial level of secondary school enrollment with the average years of secondary schooling in the population over 15 in 1960 as a measure of initial human capital, since as Pritchett (2001) argues, enrollment ratios are not an ideal measure of the stock of human capital. Second we include a full set of time dummies, gt, to account for time-speciﬁc heterogeneity in growth rates across countries. 15. This is related to Sachs and Warner (1995) who used a dummy variable of openness, with a country being classiﬁed as closed if it displayed at least one of ﬁve criteria, namely; (i) Average tariﬀ rates of 40% or more, (ii) Non-tariﬀ barriers covering 40% or more of trade, (iii) A Black Market exchange rate (BMP) that is depreciated by 20% or more relative to the oﬃcial exchange rate, on average, (iv) A state monopoly on major exports, (v) A socialist economic system. The date of liberalization is then deﬁned as the year in which none of these criteria are met. The measure was heavily criticised by Rodrı´guez and Rodrik (2000), who argued that the information on the BMP and the state monopoly on major exports played the major role in its classiﬁcation. They went on to argue that a high BMP is likely to reﬂect factors other than trade policy, including macroeconomic mismanagement, weak enforcement of the rule of law, and high levels of corruption, while the information on the state

monopoly of exports works like an Africa dummy. In updating this indicator, Wacziarg and Welch (2008) note that the liberalization date is less subject to criticism, and are careful to cross-check their dates against case studies of reforms in developing countries. 16. The ﬁgures for the ﬁve- and twenty-year averages are similar. 17. The regressions estimated in Columns 5 and 6 employ data for 58 countries for the 1960s and 1970s, 70 countries for the 1980s and all 75 countries for the 1990s and 2000s. 18. The results for the restricted sample where we retain the terms of trade variable are very similar to those in the remainder of this section and are available from the authors. 19. The results reported are based on the GDP deﬂator rather than the CPI index, since the GDP deﬂator is available for more countries and more years. Our results are robust to the use of either the CPI or GDP deﬂator. 20. The ratio of debt to exports is calculated as the total value of debt in current US dollars divided by the value of exports in current US dollars. 21. For an introduction to principal components and factor analysis see Kline (1994). Campos et. al. (2010) employ principal components to construct an index of social and political instability. Dreher, Gaston, and Martens (2008) use principal components to construct indices of globalisation. 22. The results using a 10 year period are very similar. 23. Once again, the ﬁgures are similar if we consider the ﬁve- and twentyyear averages instead. 24. Many studies use these standardized scores to create “crisis” dummy variables. Whether a country is viewed to be in crisis is determined by imposing some threshold value (usually 1.5 or 2) on the standardized score. 25. Similar test procedures are used to test for multiple thresholds. In the two threshold model for example the test is deﬁned as a test of two thresholds versus a single threshold. 26. The bootstrap distribution of the test statistic was computed using 1000 replications of the procedure proposed in Hansen (1999). 27. While it is straightforward to search for multiple thresholds, it can be computationally time-consuming. Bai (1997) has shown, however, that sequential estimation is consistent, thus avoiding this computation problem. In the case of a two threshold model this involves ﬁxing the ﬁrst threshold and searching for a second threshold. The estimate of the second threshold is then asymptotically eﬃcient, but not the ﬁrst threshold because it was estimated from a sum of squared errors function that was contaminated by the presence of a neglected regime. Bai (1997) suggests estimating a reﬁned estimator for the ﬁrst threshold, which involves re-estimating the ﬁrst threshold, assuming that the second threshold is ﬁxed. The test of signiﬁcance of the second threshold proceeds along the same lines as described above, with the null and alternative hypotheses being of one and two threshold models respectively. 28. To ensure a reasonable number of observations in each regime we generally impose the restriction that at least ten percent of observations must lie in each.

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH 29. Table 8 in the Appendix reports whether each particular liberalization episode would be classed as having occurred at a time of crisis according to the results in Table 3A. 30. The single thresholds and the lower values of the double thresholds were similar to the thresholds reported in Table 3A (except for XR). The estimated long-run coeﬃcients in the non-crisis regimes were the same or slightly higher than the corresponding coeﬃcients in Table 3A. Of the indicators with two thresholds, only INF has coeﬃcients in its crisis regimes that signiﬁcantly diﬀer from each other, indicating that the second thresholds arise to accommodate the short-run eﬀects for the other indicators at least. 31. Given the mixed bag of short-run results for the individual indicators, it is perhaps unsurprising that consideration of separate short and long-run thresholds tends to generate few signiﬁcant short-run results for our composite indicators. Given this outcome we see little point in pursuing joint thresholds in the short-run.

2191

32. We report, but do not emphasise, the results using per capita income growth as our crisis indicator, because of its deﬁnitional relationship with the annual growth rate of per capita income. The correlation between the two is -0.57. This is not an issue in the main analysis, however, where we only consider the value of the crisis indicator at the date of liberalization, which is generally not correlated with the annual growth rate of per capita income. 33. See Falvey, Foster, and Greenaway (2010). 34. A crisis is considered to occur if we obtain positive values of the crisis indicator for three consecutive periods, with the largest value taken as the peak of the crisis. Note particularly that this variable is deﬁned for any crisis, independently of whether a trade liberalization also occurs.

REFERENCES Alesina, A., Ardagna, S., & Trebbi, F. (2006). Who adjusts and when? On the political economy of reforms. IMF Staﬀ Papers, 53(special issue), 1–29. Anderson, J. E., & Neary, J. P. (2005). Measuring the restrictiveness of international trade policy. Cambridge, MA: MIT Press. Bai, J. (1997). Estimating multiple breaks one at a time. Econometric Theory, 13, 315–352. Baldwin, R. E. (2003). Openness and growth: What’s the empirical relationship? In R. E. Baldwin, & L. A. Winters (Eds.), Challenges to globalization: Analyzing the economics (pp. 499–526). Chicago, Ill: University of Chicago Press. Barro, R. J., & Lee, J.-W. (2001). International data on educational attainment: Updates and implications. Oxford Economic Papers, 53(3), 541–563. Bhagwati, J. (1978). Foreign trade regimes and economic development. Cambridge, MA: Ballinger Press. Bruno, M. (1996). Deep crisis and reform: What have we learned? Washington, DC: World Bank. Bruno, M., & Easterly, W. (1996). Inﬂation’s children: Tales of crises that beget reforms. American Economic Review Papers and Proceedings, 86(May), 213–217. Bruno, M., & Easterly, W. (1998). Inﬂation crises and long-run growth. Journal of Monetary Economics, 41(1), 3–26. Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: Methods and applications. Cambridge, UK: Cambridge University Press. Campos, N., Cheng, F. H., & Nugent, J. B. (2010). Crises, What crises? New evidence on the relative roles of political and economic crises in begetting reforms. Journal of Development Studies, 46(10), 1670–1691. Chan, K. (1993). Consistency and limiting distribution of the least squares estimator of a threshold autoregressive model. The Annals of Statistics, 21, 520–533. Collier, P. (1993). Higgledy–piggledy liberalisation. The World Economy, 16(4), 503–512. Dollar, D. (1992). Outward-oriented developing economies really do grow more rapidly. Economic Development and Cultural Change, 40(3), 523–544. Dornbusch, R. (1992). The case for trade liberalisation in developing countries. Journal of Economic Literature, 6(1), 69–85. Drazen, A., & Easterly, W. (2001). Do crises induce reform? Simple empirical tests of conventional wisdom. Economics and Politics, 13(2), 129–158. Drazen, A., & Grilli, V. (1993). The beneﬁt of crises for economic reforms. American Economic Review, 83(3), 598–607. Dreher, A., Gaston, N., & Martens, P. (2008). Measuring globalisation: Gauging its consequences. New York, NY: Springer. Easterly, W. (1996). When is stabilization expansionary? Evidence from high inﬂation. Economic Policy, 11(22), 65–107. Falvey, R., Foster, N., & Greenaway, D. (2010). Trade liberalisation, economic crises and growth. GDC Working Paper, Bond University.

Falvey, R., Foster, N., & Khalid, A. (2011). Trade liberalisation and growth: A threshold exploration. GDC Working Paper, Bond University. Frankel, J., & Romer, D. (1999). Does trade cause growth? American Economic Review, 89(3), 379–399. Greenaway, D. (1993). Liberalising foreign trade through rose tinted glasses. Economic Journal, 103(January), 208–223. Greenaway, D. (1998). Does trade liberalisation promote economic development? Scottish Journal of Political Economy, 45(5), 491–511. Greenaway, D., & Kneller, R. (2007). Firm heterogeneity, exporting and FDI. Economic Journal, 117(February), F134–161. Greenaway, D., Leybourne, S., & Sapsford, D. (1997). Modelling growth (and liberalisation) using smooth transitions analysis. Economic Inquiry, 35(4), 798–814. Greenaway, D., Morgan, C. W., & Wright, P. (1998). Trade reform, adjustment and growth: What does the evidence tell us? Economic Journal, 108(September), 1547–1561. Greenaway, D., Morgan, C. W., & Wright, P. (2002). Trade liberalisation and growth in developing countries. Journal of Development Economics, 67, 229–244. Greenaway, D., & Sapsford, D. (1994). What does liberalisation do for exports and growth? Weltwirtschaftliches Archiv, 130(1), 152–174. Hansen, B. E. (1996). Inference when a nuisance parameter is not identiﬁed under the null hypothesis. Econometrica, 64, 413–430. Hansen, B. E. (1999). Threshold eﬀects in non-dynamic panels: Estimation, testing and inference. Journal of Econometrics, 93(2), 345–368. Kline, P. (1994). An easy guide to factor analysis. London, UK: Routledge. Krueger, A. O. (1978). Foreign trade regimes and economic liberalisation. Lexington, MA: Ballinger. Krueger, A. O. (1997). Trade policy and economic development: How we learn. American Economic Review, 87(1), 2–21. Lee, J.-W. (1995). Capital goods imports and long-run growth. Journal of Development Economics, 48(1), 19–110. Little, I., Scitovsky, T., & Scott, M. (1970). Industry and trade in some developing countries: A comparative study. Cambridge, MA: Oxford University Press. Lora, E. (1998). What makes reforms likely? Timing and sequencing of structural reforms in Latin America. Working Paper No. 424, InterAmerican Development Bank. Maurer, R. (1998). Economic growth and international trade with capital goods: Theories and empirical evidence. Tubingen, Germany: Kiel University Press. Papageorgiou, D., Michaely, M., & Choksi, A. (1991). Liberalising foreign trade. Oxford, UK: Basil Blackwell. Pritchett, L. (2001). Where has all the education gone? World Bank Economic Review, 15(3), 367–391. Rodrı´guez, F., & Rodrik, D. (2000). Trade policy and economic growth: A skeptic’s guide to the cross-national evidence. In B. Bernanke, & K. S. Rogoﬀ (Eds.), NBER Macroeconomics Annual 2000 (pp. 261–338). Cambridge, MA: MIT Press.

2192

WORLD DEVELOPMENT

Rodrik, D. (1989a). Credibility of trade reform. The World Economy, 12(1), 1–16. Rodrik, D. (1989b). Promises, promises: Credible policy reform via signalling. Economic Journal, 99(September), 756–772. Rodrik, D. (1995). Getting interventions right: How South Korea and Taiwan grew rich. Economic Policy, 20(1), 55–107. Rodrik, D. (1996). Understanding economic policy reform. Journal of Economic Literature, 34(March), 9–41. Rodrik, D. (1999). The new global economy and developing countries: Making openness work. Washington, DC: Overseas Development Council. Sachs, J., & Warner, A. (1995). Economic reform and the process of global integration. Brookings Papers on Economic Activity, 1, 1–118. Salinas, G., & Aksoy, A. (2006). Growth before and after trade liberalisation. World Bank Policy Research Working Paper No. 4062, Washington, DC.

Singh, T. (2010). Does international trade cause economic growth? A Survey. The World Economy, 33(11), 1517–1564. Tornell, A. (1998). Reform from within. NBER Working Paper 6497. Wacziarg, R. (2001). Measuring the dynamic gains from trade. World Bank Economic Review, 15(3), 393–429. Wacziarg, R., & Welch, K. (2008). Trade liberalisation and growth: New evidence. World Bank Economic Review, 22(2), 187–231.

APPENDIX A

Table 7. Summary statistics Variable Dln y y60 INV/GDP Dln POP Dln TTI SYR60 CRISLIB OUT INF XR DEBT CAD

Observations

Mean

Standard deviation

Minimum

Maximum

2766 2194 2620 2766 1364 2148

0.014 6.79 0.21 0.020 0.010 0.540

0.055 1.19 0.08 0.011 0.147 0.626

-0.593 4.52 0.06 0.028 1.844 0.003

0.221 9.28 0.60 0.060 1.986 2.69

2633 2597 2500 1967 2042

0.210 0.160 1.04 0.063 0.2518

0.829 0.872 0.744 1.054 0.954

1.764 1.475 1.271 1.773 1.735

1.739 1.760 1.789 1.735 1.586

Notes: While the mean and standard deviations of the crisis variables are zero and one respectively, there is no reason to suppose that the mean of the variables at the time of liberalization should be zero. Interestingly, for three of the ﬁve crisis variables (per capita output growth, the ratio of debt to exports and the current account balance) the mean of the crisis variable at liberalization is negative, indicating that performance according to these measures was better than average.

Table 8. Liberalizing countries and episodes undertaken in crisis according to Table 3A Country

Lib Year

ln y at Lib

Albania Argentina Armenia Australia Azerbaijan Bangladesh Barbados Benin Bolivia Botswana Brazil Bulgaria Burkina Faso Burundi Cameroon Cape Verde Chile Colombia Costa Rica Cote d’Ivoire Dominican Republic Ecuador Egypt El Salvador Ethiopia

1992 1991 1995 1964 1995 1996 1966 1990 1995 1979 1991 1991 1998 1999 1993 1991 1976 1986 1986 1994 1992 1991 1995 1991 1996

0.07492 0.10856 0.08343 0.049161 0.13708 0.02772 0.035858 0.00014 0.03852 0.079099 0.00325 0.07832 0.01427 0.02942 0.06053 0.00544 0.018003 0.036367 0.024531 0.02364 0.060575 0.028394 0.02622 0.00613 0.073819

OUT (0.05)

INF (0.09) p

XR (0.9)

DEBT (1.34)

N/A p

N/A p

CAD (1.09)

INT (0.51)

EXT (0.78)

N/A

N/A

p p p p

p p p p

N/A

N/A

p p

p

p p

N/A N/A

N/A N/A p

N/A N/A

N/A N/A p

N/A

N/A

N/A

N/A

p

p p

N/A N/A

N/A N/A

p

p

N/A p p p p

N/A

p

N/A

N/A N/A p

p

p

p p p

p p p p p

N/A

N/A

p p

p p

N/A

p

p p

p p p p

TRADE LIBERALIZATION, ECONOMIC CRISES, AND GROWTH

2193

Table 8 (continued) Country

Lib Year

ln y at Lib

OUT (0.05) p

INF (0.09) p

XR (0.9) p

DEBT (1.34)

CAD (1.09)

INT (0.51) p

Gambia 1985 0.0449 p Georgia 1996 0.109529 N/A N/A N/A p Ghana 1985 0.011366 p Guatemala 1988 0.013508 p p Guinea-Bissau 1987 0.00348 p p p Guyana 1988 0.03144 p p p Honduras 1991 0.001985 p p p p p Hungary 1990 0.03241 Indonesia 1970 0.054622 N/A N/A N/A p Ireland 1966 0.006114 N/A N/A N/A p p Israel 1985 0.016261 N/A N/A Jamaica 1962 0.00324 N/A N/A N/A N/A N/A N/A p Jamaica 1989 0.060218 Japan 1964 0.100021 N/A N/A N/A Kenya 1963 0.052643 N/A N/A N/A N/A N/A N/A p p Kenya 1993 0.02314 Republic of Korea 1968 0.08828 N/A N/A N/A p Kyrgyz Republic 1994 0.22355 N/A N/A N/A Latvia 1993 0.03348 N/A N/A N/A p Lithuania 1993 0.17245 N/A N/A N/A N/A Macedonia 1994 0.02495 N/A N/A N/A N/A p Madagascar 1996 -0.00965 Mali 1988 0.01237 p Mauritania 1995 0.023755 p p p p Mexico 1986 0.05861 p Moldova 1994 0.36962 N/A N/A N/A N/A p Morocco 1984 0.020437 p p Mozambique 1995 0.016492 p p Nepal 1991 0.040674 p p New Zealand 1986 0.019049 N/A N/A p p Nicaragua 1991 0.03131 p p Niger 1994 0.00416 p p p Pakistan 2001 0.00569 p p p Panama 1996 0.011507 p p Paraguay 1989 0.028816 p Peru 1991 0.001776 N/A N/A Philippines 1988 0.041637 p Poland 1992 N/A N/A N/A N/A p p p Romania 1992 0.07533 Sierra Leone 2001 0.024827 Singapore 1965 0.086483 N/A N/A N/A p p Slovak Republic 1991 -0.15752 N/A N/A N/A N/A p p p South Africa 1991 0.03085 N/A N/A p p Sri Lanka 1977 0.033611 p p p Sri Lanka 1991 0.033908 p Tajikistan 1996 0.19837 N/A N/A p Tanzania 1995 0.005887 p Trinidad and Tobago 1992 0.02491 p p Tunisia 1989 0.004536 p p p p Turkey 1989 0.01919 p p p Uganda 1988 0.044017 p p p p Venezuela 1989 0.11348 p p p p Venezuela 1996 0.02259 p Zambia 1993 0.039416 N/A N/A p Notes: A indicates that the country would have been classed as in a crisis according to the relevant crisis indicator and the results of Table indicates that the country was not included in the regression, usually due to a lack of data on the crisis variable.

EXT (0.78) N/A p

p N/A N/A N/A N/A p N/A N/A N/A N/A N/A N/A N/A

N/A p N/A p

N/A N/A

N/A N/A N/A p N/A

p

N/A 3A; N/A

Trade Liberalization, Economic Crises, and Growth

Trade Liberalization, Economic Crises, and Growth

Recommend Documents