77555 The Structural Determinants of External Vulnerability Norman V. Loayza and Claudio Raddatz This article examines empirically how domestic structural characteristics related to openness and product- and factor-market flexibility influence the impact of terms of trade shocks on aggregate output. Applying semistructural vector autoregressions to a panel of 88 countries with annual observations for the period 1974–2000, the analy- sis isolates and standardizes the shocks, estimates their impact on GDP, and examines how this impact depends on the domestic conditions outlined above. The article ï¬?nds that greater trade openness magniï¬?es the output impact of terms of trade shocks, par- ticularly negative ones, while ï¬?nancial openness reduces their impact. Flexibility of labor and ï¬?rm-entry are beneï¬?cial, with labor flexibility dampening the impact of negative shocks and ease of ï¬?rm-entry magnifying positive ones only. Domestic ï¬?nan- cial depth has a more nuanced role in stabilizing the economy. Analysis of interactions across structural determinants reveals complementarities among macroeconomic con- ditions (trade and ï¬?nancial openness and depth) and, separately, among microeco- nomic conditions (flexibility of labor markets and ease of ï¬?rm-entry). Variables across these groups tend to behave as substitutes for each other. JEL codes: F36, F41, F43. Macroeconomic volatility is not only a source of business cycle uncertainty but also a major cause of low economic growth. Ramey and Ramey (1995) were the ï¬?rst to document this ï¬?nding for a cross-section of countries. Fata ´ s (2002) and Hnatkovska and Loayza (2005) show that macroeconomic volatility is particu- larly harmful for developing countries, where volatility is higher and its impact more pronounced. Among the causes of macroeconomic volatility, fluctuations in the terms of trade are important sources of external shocks. Across countries about 10 Norman V. Loayza (corresponding author) is a lead economist at the World Bank; his email address is nloayza@worldbank.org. Claudio Raddatz is an economist at the World Bank; his email address is craddatz@worldbank.org. The paper on which this article was based was prepared for the conference “The Growth and Welfare Effects of Macroeconomic Volatilityâ€? (Barcelona, 17– 18 March 2006), sponsored by the World Bank and Pompeu Fabra University, Barcelona. The authors thank Paolo Mauro (discussant); Luis Serve ´ n; Romain Ranciere and Jaume Ventura (conference organizers); other conference participants; the journal editor; and three anonymous referees for useful comments. They are grateful to Koichi Kume and Naotaka Sugawara for editorial assistance. THE WORLD BANK ECONOMIC REVIEW, VOL. 21, NO. 3, pp. 359 –387 doi:10.1093/wber/lhm018 Advance Access Publication 4 October 2007 # The Author 2007. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org 359 360 THE WORLD BANK ECONOMIC REVIEW percent of the variation in GDP growth and about 25 percent of the variation in growth volatility can be explained by observed differences in the volatility of terms of trade changes (see Easterly and others 1993; Hnatkovska and Loayza 2005). Terms of trade shocks have also been documented to have a signiï¬?cant impact on GDP within countries (see Ahmed 2003; Raddatz 2007), although their preeminence over domestic shocks is subject to debate.1 Going beyond the average impact of external shocks, there is a rich literature suggesting that the impact of an external shock on the real side of the economy may be determined by domestic conditions interacting with the shock to produce macroeconomic stability or volatility. Traditional analysis of the domestic sources of vulnerability has stressed macroeconomic policy responses in monet- ary, foreign exchange, and ï¬?scal areas. A recent example is Broda (2001), who compares the stabilizing properties of different exchange rate regimes in the face of terms of trade shocks. New developments in the study of vulnerability have concentrated on the role of structural characteristics related to the functioning of markets and institutions. Some studies stress the role of factor and product market rigidities in amplifying shocks at the macroeconomic level (see, for example, Kiyotaki and Moore 1997; Bernanke and Gertler 1989; Caballero and Hamour 1994, 1996, 1998; Caballero and Krishnamurty 2001). Others, such as Acemoglu and others (2003) and Rodrik (1999), point to the relevance of insti- tutional development in the control of crises and management of shocks.2 Notwithstanding these contributions, the extent to which domestic struc- tural characteristics can account for the relative instability of aggregate output remains an open question. This article contributes to this literature by examin- ing how certain domestic structural characteristics influence the impact that external shocks may have on aggregate output. The broad issue is whether a country’s vulnerability to shocks is not purely random but linked to structural characteristics that are at least partially policy driven: trade openness, ï¬?nancial depth, capital account openness, labor market flexibility, and ease of ï¬?rm-entry. Not deep institutional variables, these are outcomes through which funda- mentals operate. The purpose in using structural outcome variables is to separ- ate and elucidate the mechanisms that drive external vulnerability. For analysis two aspects of output vulnerability to external shocks can be dis- tinguished: the frequency and strength of shocks affecting a country and the effect that a shock of a given size and frequency can have on the country’s output. This article examines the second aspect, applying an econometric method that, using a panel sample of 88 countries and annual observations for 1974–2000, isolates 1. Mendoza (1995) and Kose and Riezman (2001), using calibrated small open economy models, ï¬?nd terms of trade shocks to account for almost half of economic fluctuations. Hoffmaister, Roldo ´ s, and Wickham (1998), Ahmed (2003), and Raddatz (2007), among others, using time-series analysis, ï¬?nd that external shocks explain a much smaller fraction of output volatility (around 20 percent). 2. Another strand of the literature takes one step back and focuses on explaining the business-cycle behavior of the terms of trade. For instance, Kraay and Ventura (forthcoming) link movements in the terms of trade to countries’ endowments and resulting industrial structure. Loayza and Raddatz 361 and standardizes the shocks, estimates their impact on GDP, and examines how this impact depends on the domestic conditions outlined above.3 From an empirical perspective, the relevant question is whether the differential impact of a given external shock is related to country characteristics. Controlling for the size of the shocks is not easy. Most of the recent literature has relied on either indirect evidence from difference-in-difference estimation (see, for example, Braun and Larrain 2005; Caballero and others 2005; Raddatz 2006) or calibrated macroeconomic models developed to match certain moments of developing countries’ economic performance (for a survey, see Arellano and Mendoza 2002). This article takes a different approach and directly estimates the output impact of external shocks using semistructural vector autoregression analysis, as applied to panel data (cross-country, time-series) of aggregate variables. This method requires the identiï¬?cation assumption that the relevant external variable—the change in the terms of trade—does not respond to domestic output changes or the variables that account for the country’s structural charac- teristics. In practice, this rather uncontroversial assumption amounts to a small open economy condition for the countries included in the analysis. Similar applications of this methodology can be found in Broda (2004), Ahmed (2003), Uribe and Yue (2003), and Raddatz (2007). Controlling for the size of the shock, the analysis accounts for its interaction with the set of country character- istics under analysis and estimates its conditional output impact. Section I presents the econometric method in detail and the speciï¬?cation tests. Section II introduces the data, the variable deï¬?nitions and sources, and the sample of countries and years under analysis. Section III presents the empirical results, including the discussion of symmetric and asymmetric effects, the poten- tial interaction between structural characteristics, and a set of robustness checks. I. ME T H O D O LO GY The impact of exogenous shocks on a country’s economic performance and its relation to the country’s structural characteristics are estimated using a panel vector autoregression. To minimize the need for identiï¬?cation assumptions (see below), the focus is exclusively on terms of trade shocks. Therefore, for a given country i the semistructural model corresponds to: q X ð1Þ Ai;0 xi;t ¼ Ai;j xi;tÀj þ 1it j¼1 3. An analogy illustrates these ideas. How vulnerable people are to disease depends on the seriousness of the disease itself (ï¬?rst aspect) and how well prepared they are to bear a given disease (second aspect). Analyzing the second aspect requires examining how people facing the same disease (in type and strength) react, to shed light on why some people suffer so much from an attack of, say, the flu, while others remain unscathed. 362 THE WORLD BANK ECONOMIC REVIEW where xi,t ¼ (Dtti,t, Dyi,t)0 is a vector that contains the ï¬?rst difference of the log of terms of trade index (Dtti,t) and the log of real GDP per capita (Dyi,t), both as deviations from their country-speciï¬?c means.4 The matrices Ai,j contain the structural coefï¬?cients for the different lags incorporated in the model (includ- ing the contemporaneous one). The structural errors 1i,t are independent and identically distributed, with zero mean and a diagonal variance – covariance matrix S. The identiï¬?cation assumption applied is that terms of trade changes are strictly exogenous for a given country—D tti,t does not respond to Dyi,t at any lags. This assumption is equivalent to imposing the following triangular structure on all the A matrices:   ai;j 0 ð2Þ Ai;j ¼ i;j 11 : a 21 ai;j 22 For the developing and small developed countries in this study this assump- tion should be uncontroversial. In fact, for the sample of countries included in this study a standard Granger causality test cannot reject the hypothesis that output fluctuations do not Granger-cause terms of trade fluctuations.5 The relatively weak assumption required to identify the impact of terms of trade shock is the reason for focusing exclusively on these shocks. It is prefer- able to focus on a reduced set of shocks that can be clearly identiï¬?ed than on a broader set of shocks that would require strong and controversial identiï¬?cation assumptions. This means, however, that the results have to be interpreted with caution. Since the model does not account for all exogenous sources of fluctu- ations, it is semistructural. The terms of trade variable captures all strictly exogenous variables whose fluctuations are correlated with those of the terms of trade. Therefore, statements on the effects of different structural character- istics on the ampliï¬?cation or dampening of shocks apply directly to terms of trade shocks and indirectly to other exogenous contingencies that are correlated with these shocks, such as the world business cycle. The baseline model corresponds to a panel vector autoregression in which part of the coefï¬?cients in the A matrices are assumed to be common across cross- sectional units. Of interest is testing how different structural characteristics of a country affect the impact of terms of trade shocks on output, as captured by the a i ,j 21 coefï¬?cients, so these coefï¬?cients are permitted to vary across countries according to the speciï¬?c characteristics whose role is to be determined. 4. Of course, this is equivalent to including a country ï¬?xed effect in the vector autoregression. 5. When performed on a country by country basis, the test cannot reject the null hypothesis in 72 of the 88 cases. The results are not materially affected by excluding the 16 countries in which the hypothesis is rejected that terms of trade are not Granger-caused by output. The countries are Canada, Chile, Greece, India, Ireland, Jamaica, Kenya, Lesotho, Mali, Malaysia, Niger, Papua New Guinea, Senegal, Singapore, Togo, and Thailand. Loayza and Raddatz 363 In particular, it is assumed that ;j ai21 ¼ bj0 þ bj1  OPENi þ bj2  FDEVi þ bj3  CAOPENi ð3Þ þ bj4  LABORi þ bj5  ENTRYi where OPENi, FDEVi, CAOPENi, LABORi, and ENTRYi are measures of trade openness, ï¬?nancial development, capital account openness, labor flexibility, and ease of ï¬?rm-entry for country i, respectively (described below). The analysis will also allow for the possibility that the influence of these characteristics on the transmission of terms of trade shocks may differ for increases and decreases in log terms of trade (with respect to the mean change). For the notation above, this corresponds to allowing the bj coefï¬?cients to vary with the state of the terms of trade in the following way:  bþ if Dtti;t . 0 b¼ bÀ otherwise where b ¼ (b0, . . . , bj, . . . , bq), bj ¼ (bj j þ 0, . . . , b5 ), and b and b 2 are similarly deï¬?ned. The remaining of the coefï¬?cients that capture the dynamics of the terms of trade (a i,j i,j 11) and the lagged effect of output on itself (a 22) are restricted to be the same for all countries. The use of panel vector autoregressions, with the corresponding restrictions on the parameters, is common in the recent literature estimating the impact of exogenous shocks on macroeconomic variables (Broda 2004; Ahmed 2003; Uribe and Yue 2003), because the limited length of the time-series dimension of the data (around 25 annual observations) makes it difï¬?cult to estimate country- speciï¬?c dynamics. Using a panel vector autoregression approach increases the degrees of freedom of the estimation, and if the common parameter restrictions are correct, provides more efï¬?cient estimators. Of course, the obvious disadvantage is that the model is incorrectly speciï¬?ed if these restrictions are not valid. A concern with this approach, as Pesaran and Smith (1995) note, is that assuming common coefï¬?cients may yield parameters that underestimate the short-run impact of exogenous variables (and overestimate the long-run impact) if the dynamics differ substantially across countries. However, as Pakes and Griliches (1984) demonstrate, if differences in slope coefï¬?cients are uncor- related with the exogenous variables, the estimated parameters would be con- sistent estimators of the average coefï¬?cients. This is an important result for the analysis here as there is no obvious reason why the marginal effect of terms of trade in a country should be determined by the terms of trade itself. Nevertheless, in an additional exercise (and at the cost of reduced precision of the estimates) the vector autoregression is also estimated on a country by country basis, without imposing any restriction on the dynamics. The estimated country-speciï¬?c effects of the shocks are then related to the structural 364 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 . Unit Root Test Augmented Dickey – Augmented Dickey – Fuller by country Fuller by country (cannot reject unit (cannot reject unit Levin – Lin – Chu Variable root, percent) root, percent) p-value (1) (2) (3) Log GDP per capita 72 86 0.987 Log terms of trade index 60 83 0.994 Note: Column 1 reports the percentage of the 88 countries in the sample for which the Augmented Dickey – Fuller test cannot reject the null hypothesis of a unit root when the number of lags augmenting the test is country speciï¬?c, as determined by performing the Hall (1990) pro- cedure on a country by country basis. Column 2 reports the percentage for the case where for all countries the model is augmented using the median number of lags (two) across countries. Column 3 shows the p-value of the Levin – Lin – Chu (2002) test for panel unit roots for the case in which the panel is augmented by two lags. Source: Authors’ analysis based on data described in text. characteristics under study. The results prove to be very similar to those obtained with the panel methodology. As mentioned, the variables in the vector autoregression are the ï¬?rst differences of the log terms of trade and output per capita. The relevant series is modeled as difference–stationary for two reasons. First, standard tests suggest the presence of a unit root in the levels of both series. Columns 1 and 2 of table 1 show the results of the Augmented Dickey–Fuller tests performed on a country by country basis for country-speciï¬?c and common lag structures.6 In most cases the test cannot reject the null hypothesis of a unit root for both series (about 85 percent of the time for both series when the median number of lags is used for all countries). The panel-based unit-root test suggested by Levin, Lin, and Chu (2002), augmented by the median number of lags across countries (two), reaches a similar conclusion: the null hypoth- esis of a unit root cannot be rejected. The second reason for modeling the relevant series as difference–stationary is that previous empirical papers in this literature (for example, Ahmed 2003 and Broda 2004) have done so, giving this speciï¬?cation the advantage of being more directly comparable with previous results.7 The vector autoregression speciï¬?cation uses two annual lags in the bench- mark speciï¬?cation. This lag structure was determined using standard lag selec- tion tests (Akaike information criterion, Schwartz information criterion, and Hannan–Quinn criterion). 6. The number of lags for each country was determined using the methodology of Hall (1990). The common number of lags used in column 2 corresponds to the median across countries (two lags). 7. A Pedroni (1999) panel cointegration test, not reported, does not reject the null hypothesis of no cointegration between log terms of trade and output. The different statistics derived by Pedroni tend to give different results, but most of them cannot reject the null hypothesis of no cointegration. Because the power and size tradeoff of the different tests varies with the cross-sectional and time-series dimension of the panel (see Pedroni 2004), statistics with the largest size (that tend to overreject) and highest power at short time dimensions were emphasized. Those tests, corresponding to the panel and group t-statistics derived by Pedroni (1999), clearly do not reject the null hypothesis of no cointegration. Loayza and Raddatz 365 Under the identiï¬?cation assumptions described above, the parameters of the model are estimated using a two-step procedure: the reduced-form coefï¬?cients are ï¬?rst estimated equation by equation by ordinary least squares (OLS); then the impulse-response functions (IRFs) are computed for each of the structural shocks (using the reduced-form coefï¬?cients and the variance–covariance matrices of the reduced-form errors derived from these coefï¬?cients). The conï¬?- dence bands for the IRFs are estimated by parametric bootstrapping, assuming normally distributed reduced-form errors.8 I I . D ATA The following are the main variables used in the analysis. Real GDP per capita, in constant 2000 U.S. dollars was obtained from the World Development Indicators Database (World Bank 2005) This series was used instead of GDP adjusted for purchasing power parity (PPP), despite reduced international comparability, because it has more recent coverage than the measures from the Penn World Tables (Heston, Summers, and Aten 2002) and longer coverage than the PPP series produced by the World Bank.9 The terms of trade index is the ratio of export prices to import prices using the current and constant price values of exports and imports from the national accounts component of the Penn World Tables version 6.1 and updated using the terms of trade data from the World Development Indicators Database (World Bank 2005).10 To reduce concerns about structural breaks, data are for the post Bretton-Woods period, 1974–2000. The structural characteristics of countries are captured in the following vari- ables. Trade openness is measured as the log of the ratio of total trade to GDP. Financial development is the log of the ratio of private credit provided by banks and other ï¬?nancial institutions to GDP, obtained from Beck, 8. The procedure can be briefly described as follows. The estimated variance– covariance matrix of the reduced form errors is used to simulate a random realization of the perturbations. The initial values of the different variables, the baseline coefï¬?cients, and the simulated perturbations are used to simulate a new set of observations for the variables in the vector autoregression. These simulated observations are used to estimate a new set of coefï¬?cients. This exercise is repeated 500 times. The IRF is computed for each set of coefï¬?cients obtained from the bootstrapping. A 90 percent conï¬?dence interval is built for the IRF by taking the 5th and 95th percentile of the empirical distribution of the IRF on a point by point basis. 9. In a robustness check presented below using PPP-adjusted GDP, the results remain basically the same. 10. This index is used instead of the more traditional net barter index because of its broader coverage. However, this index includes the service export sector (tourism and ï¬?nancial services), whose prices are not measured as precisely as those of merchandise trade and are much less likely to be exogenous to domestic conditions (the main identiï¬?cation assumption). This is unlikely to be a problem for the average country because of the typically small relevance of the export service sector, but to address any potential concern the index was replaced by the net barter terms of trade index in cases where the correlation between these two indexes (based on the post-1980 data in which both are available) was smaller than 0.5, taken as an indication of the importance of the export service sector (21 cases, or 25 percent of the sample). 366 THE WORLD BANK ECONOMIC REVIEW Demirgu ¸ -Kunt, and Levine (2000) or, if unavailable from that source, from ¨c the World Development Indicators Database. Openness in capital account transactions is captured by the Chinn–Ito index (Chinn and Ito 2002), with a higher value indicating a higher degree of openness.11 The index of labor market flexibility, calculated from data in World Bank (2003), is a weighted average of three indicators (flexibility of hiring, conditions of employment, and flexibility of ï¬?ring) as in Botero and others (2004). The original index was rescaled to range from 0 to 1, with higher values indicating more flexible labor markets. Finally, the index of ease of ï¬?rm entry, calculated from data in World Bank (2003) and O’Driscoll, Feulner, and O’Grady (2003), is a weighted average of four indicators (registration procedures, cost to register, days to register, and burden of entry regulations) as in Chang, Kaltani, and Loayza (2005). This index also ranges from 0 to 1, with higher values indicating fewer restrictions. The sample includes 88 countries representing different regions and income levels (see appendix). Sub-Saharan Africa has the largest share in the sample, at 30 countries, followed by Latin America at 20, East Asia and Paciï¬?c at 11, the Middle East and North Africa and Western Europe at 10 each, South Asia at 4, Eastern and Central Europe at 2, and North America at 1. There are 35 low- income countries, 35 middle-income countries, and 18 high-income countries. The sample includes all countries for which measures of the structural charac- teristics described above and at least 15 continuous observations of both terms of trade and output per capita were available during 1974–2000. The six large industrial countries (the United States, Japan, Germany, United Kingdom, France, and Italy) are excluded because of the possible endogeneity of their terms of trade, as are ï¬?ve developing countries whose terms of trade data exhibited long flat periods (Cape Verde, Grenada, Nepal, St. Lucia, St. Kitts, and Nevis). Summary statistics for these variables for each country are in the appendix. Cross-sectional univariate summary statistics and bivariate correlations for these variables are presented in tables 2 and 3, respectively. Table 3 displays the well documented positive correlations between structural characteristics and output growth and negative correlations between measures of volatility and growth. It also shows that the structural characteristics are positively corre- lated with each other, although the magnitudes of the correlations are not par- ticularly large, except between ï¬?nancial development and ease of ï¬?rm-entry, which reaches 66 percent. These relatively low correlations suggest the possi- bility of sorting out the role of different structural characteristics in the trans- mission of shocks. 11. The Chinn– Ito index corresponds to the ï¬?rst principal components of the following four binary variables reported in the International Monetary Fund’s Annual Report on Exchange Arrangements and Exchange Restrictions (various issues): existence of multiple exchange rates, restrictions on current account, capital account transactions, and the existence of requirements to surrender export proceedings. Loayza and Raddatz 367 T A B L E 2 . Descriptive Statistics of Country Averages, 1974–2000: Univariate (variables reported in appendix) Variable Mean Standard deviation Minimum Maximum Average output growth 1.17 1.94 2 3.35 7.35 (percent) Average terms of trade 2 0.75 1.55 2 5.99 2.57 growth (percent) Standard deviation 4.19 1.98 1.21 10.06 output growth Standard deviation terms 11.82 7.57 1.00 34.08 of trade growth Trade openness 3.79 0.54 2.57 5.67 Financial depth 2 1.43 0.91 2 4.31 0.39 Financial openness 2 0.16 1.15 2 1.64 2.68 Labor market flexibility 0.47 0.14 0.21 0.80 Ease of ï¬?rm entry 0.65 0.15 0.22 0.94 Note: Variables are measured over the period 1974–2000. Average output growth is growth of real GDP per capita. Average terms of trade growth is the average growth of the terms of trade index. Standard deviation output growth is the standard deviation of the growth rates of real GDP per capita. Trade openness ¼ Log (Exports þ Imports)/GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn–Ito measure of capital account openness. Labor market flexi- bility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining information on number of procedures, monetary cost, and time to open a new ï¬?rm. Source: Authors’ analysis based on data described in text. II I. R ES ULT S The basic results are derived from estimating the cumulative output effect of a one standard deviation shock to the terms of trade at different levels of a par- ticular structural characteristic. As explained, this estimation is conducted in the context of a panel (cross-country, time-series) vector autoregression with (demeaned log) GDP changes as the dependent variable and (demeaned log) terms of trade changes as the exogenous variable. The output effect of terms of trade shocks are allowed to vary with ï¬?ve country structural characteristics: trade openness, ï¬?nancial depth, ï¬?nancial (or capital-account) openness, labor market flexibility, and ease of ï¬?rm-entry. To get a sense of how much a given structural factor contributes to amplifying or dampening the external shock, the shock’s cumulative output impacts are compared at relatively low (25th) and high (75th) percentiles of the world distribution of each structural characteristic. The cumulative effect of a one standard deviation shock in the terms of trade on the level of GDP per capita for low and high levels of each country characteristic is displayed in ï¬?gure 1 and table 4. To indicate the accuracy of the estimated impacts, ï¬?gure 1 also presents their 90 percent conï¬?dence bands and table 4 the corresponding (empirical) standard errors.12 To provide a 12. Critical values and corresponding conï¬?dence intervals are obtained from the empirical distribution derived through the parametric bootstrapping procedure already described. 368 T A B L E 3 . Descriptive Statistics of Country Averages, 1974–2000: Bivariate Correlations (cross-sectional correlations between the different variables reported in appendix) Standard Labor Terms of Standard deviation deviation terms Trade Financial Financial market Ease of Variable Output growth trade growth output growth of trade growth openness depth openness flexibility ï¬?rm-entry Output growth 1.00 Terms of trade growth 0.20 1.00 Standard deviation 2 0.41 2 0.13 1.00 output growth Standard deviation terms 2 0.57 2 0.12 0.55 1.00 THE WORLD BANK ECONOMIC REVIEW of trade growth Trade openness 0.25 2 0.03 2 0.13 2 0.29 1.00 Financial depth 0.52 0.30 2 0.49 2 0.65 0.32 1.00 Financial openness 0.25 0.27 2 0.33 2 0.47 0.35 0.56 1.00 Labor market flexibility 0.31 0.09 2 0.32 2 0.29 0.45 0.31 0.28 1.00 Ease of ï¬?rm entry 0.45 0.28 2 0.46 2 0.60 0.40 0.66 0.51 0.56 1.00 Note: Variables are measured over the period 1974– 2000. Average output growth is growth of real GDP per capita. Average terms of trade growth is the average growth of the terms of trade index. Standard deviation output growth is the standard deviation of the growth rates of real GDP per capita. Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining infor- mation on number of procedures, monetary cost, and time to open a new ï¬?rm. Source: Authors’ analysis based on data described in text. Loayza and Raddatz 369 F I G U R E 1. Cumulative Output Impact of Terms of Trade Shock: Symmetric Case Note: See table 2 for deï¬?nitions of variables. Bands are 90 percent conï¬?dence intervals. Source: Author’s analysis based on data described in text. 370 THE WORLD BANK ECONOMIC REVIEW T A B L E 4 . Basic Results under Symmetric Analysis: Cumulative Output Impact of a One Standard Deviation Terms of Trade Shock for Low and High Values of Five Structural Characteristics (percent of GDP) Trade Financial Financial Labor market Ease of openness depth openness flexibility ï¬?rm-entry Low value 0.227 1.178 1.250 1.569 0.908 (25th percentile)a (0.222) (0.234) (0.244) (0.240) (0.262) High value 1.609 1.032 0.819 0.338 1.430 (75th percentile)a (0.266) (0.265) (0.207) (0.260) (0.281) Difference 1.382 2 0.147 2 0.430 2 1.231 0.523 (0.308) (0.312) (0.236) (0.297) (0.373) Test Ho:Diff. ¼ 0 ** ** ** * (one-tail) *Signiï¬?cant at the 10 percent level; **signiï¬?cant at the 5 percent level. Note: Numbers is parentheses are standard errors. Critical values are obtained from empirical distribution (which may have non-Gaussian properties). Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn – Ito measure of capital account openness. Labor market flexibility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm entry is a 0 2 1 index combining information on number of procedures, monetary cost, and time to open a new ï¬?rm. a Percentiles of the world distribution of the respective structural characteristics. Source: Authors’ analysis based on data described in text. benchmark for quantitative comparison, the median cumulative output impact of a one standard deviation terms of trade shock (the impact calculated at the median of all structural characteristics) was estimated. It was approximately 1 percent of GDP. As noted, the size of the shock is made the same for all countries in order to focus on the variation in responsiveness to a uniform shock. The most noticeable result is that the cumulative output impact of terms of trade shocks rises with greater trade openness. This is likely to be a size effect in the sense that a higher volume of trade implies a larger share of economic activities that the terms of trade, as relative prices, can influence. This effect should not be confused with a purely mechanical effect, which applies to the relation between trade prices and nominal GDP (or the price of GDP in terms of importable goods). Since the analysis is based on real GDP, mechanical price effects should not be present.13 The effect of openness is large and signiï¬?- cant: the output impact of the shock is 1.4 percentage points higher at the third quartile of trade openness than at the ï¬?rst quartile. The vulnerability of open economies should not have a normative implication; it merely reflects the extent of real resource shifts in the presence of price signals and, from a 13. Unless output deflators are incorrectly or inconsistently measured, an issue considered in the discussion of robustness. Loayza and Raddatz 371 methodological perspective, highlights the need to control for openness in assessing the impact of other structural characteristics. Conversely, greater ï¬?nancial depth seems to have no effect on the impact of terms of trade shocks. This is surprising, considering that ï¬?nancial depth is usually considered an antidote to external vulnerability. This important issue will be revisited, particularly in the analysis of asymmetric effects of positive and negative shocks and complementary interactions with other structural characteristics. An increase in ï¬?nancial openness reduces the effect of terms of trade shock, signiï¬?cantly but by a moderate margin: the difference in the cumulative output impact between the 25th and 75th percentiles of ï¬?nancial openness is –0.43 percentage point. That access to international ï¬?nancial markets has a stabiliz- ing effect while domestic ï¬?nancial depth does not is puzzling (a possible inter- action between these two ï¬?nancial aspects is considered later). Easing ï¬?rm-entry signiï¬?cantly though moderately ampliï¬?es terms of trade shocks: the output impact of the shock is 0.52 percentage point higher at the 75th percentile than at the 25th percentile of ease of ï¬?rm-entry. Entry of new ï¬?rms is only one side of the ï¬?rm-dynamics process; ï¬?rm exit can also be a reaction to external shocks. Moreover, ï¬?rm dynamics may have different characteristics under negative and positive shocks. The shock- amplifying effect of ease of ï¬?rm-entry is reconsidered in the analysis of asym- metric effects. Finally, of all structural characteristics considered here, improvement in labor market flexibility has the strongest effect on reducing the impact of terms of trade shocks on per capita GDP. The difference in the shock’s cumulative output impact between the 25th and 75th percentiles of labor market flexibility is –1.23 percentage points, in absolute value almost as large as that of trade openness. The ability of ï¬?rms to adjust their activities on the labor margin seems crucial for an economy’s ability to accommodate the shock. Robustness This section examines the robustness of the basic results to changes in measure- ment of the terms of trade shock, in the sample of countries, the application of a longer lag structure in the estimated vector autoregressions, the inclusion of the exchange rate regime as a country characteristic, and implementation of an alternative method of estimating the effects of structural characteristics. The results on several robustness checks using the panel vector autoregression meth- odology are presented in the rows of table 5. The robustness check on the methodology is presented in table 6. The ï¬?rst concern is whether the amplifying effect of trade openness reflects mostly a mechanical effect. Two robustness checks address this issue. The ï¬?rst replaces the simple terms of trade index with one that weighs export and import prices by the size of export and import volumes. When the basic exercise is 372 THE WORLD BANK ECONOMIC REVIEW T A B L E 5 . Robustness: Cumulative Output Impact of a One standard Deviation Terms of Trade Shock for Low and High Values of Five Structural Characteristics (percent of GDP) Labor Ease of Exchange Trade Financial Financial market ï¬?rm rate Robustness testa openness depth openness flexibility entry regime Benchmarkb Low 0.227 1.178 1.250 1.569 0.908 High 1.609 1.032 0.819 0.338 1.430 Difference 1.382 2 0.147 2 0.430 2 1.231 0.523 Trade weighted terms of trade Low 0.526 1.018 0.890 0.996 0.264 High 1.470 0.522 0.435 0.290 1.571 Difference 1.995 2 0.496 2 0.455 2 0.706 1.307 Purchasing power parity GDP Low 0.281 0.948 0.983 1.167 1.000 High 1.308 0.906 0.821 0.539 0.779 Difference 1.027 2 0.042 2 0.162 2 0.627 2 0.221 Developing countries only Low 0.235 1.314 1.370 1.714 1.015 High 1.793 1.138 0.935 0.415 1.569 Difference 1.558 2 0.176 2 0.435 2 1.299 0.553 Excluding 10 percent largest countries Low 2 0.014 1.185 1.133 1.427 0.841 High 1.606 0.873 0.769 0.325 1.290 Difference 1.620 2 0.312 2 0.365 2 1.102 0.448 Excluding mainly manufacturing exporters Low 0.207 1.273 1.275 1.638 0.935 High 1.701 1.052 0.912 0.359 1.522 Difference 1.494 2 0.220 2 0.363 2 1.279 0.587 Three lags in common lag structure Low 0.128 0.982 1.101 1.482 0.589 High 1.407 0.891 0.631 0.056 1.555 Difference 1.279 2 0.092 2 0.470 2 1.425 0.966 Including exchange rate regime Flexible 0.467 1.228 1.411 1.706 1.153 1.296 Fixed 1.794 1.342 1.100 0.651 1.562 1.224 Difference 1.327 0.114 2 0.311 2 1.055 0.410 2 0.072 Note: Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1 index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining information on number of procedures, monetary cost, and time to open a new ï¬?rm. a “Lowâ€? and “Highâ€? correspond to the 25th and 75th percentiles, respectively, of the world distribution of the respective structural characteristic. b Includes all countries and sets the common lag structure to two lags. Source: Authors’ analysis based on data described in text. repeated using this trade-weighted shock, the amplifying effect of trade openness becomes even larger. This indicates that trade openness, as a mechanism for shock expansion, operates not only through trade volumes but also through domestic Loayza and Raddatz 373 T A B L E 6 . Shock Impact and Structural Characteristics, (Dependent variable: cumulative GDP impact of a one-standard deviation terms of trade shock) Ordinary least squaresa Weighted least squares Constant 2 0.1547 (0.2086) 2 0.2350 (0.2353) Trade openness 0.1286 (0.0502)** 0.1187 (0.0513)** Financial depth 0.0613 (0.0508) 0.0103 (0.0374) Financial openness 2 0.0558 (0.0301)* 2 0.0379 (0.0259) Labor market flexibility 2 0.6658 (0.2164)** 2 0.7498 (0.2190)** Ease of ï¬?rm-entry 0.2808 (0.2176) 0.4098 (0.2463)* R-squared 0.16 — Number of countries 85 88 *Signiï¬?cant at the 10 percent level; ** signiï¬?cant at the 5 percent level. Note: Numbers in parentheses are robust standard errors. The regressions are estimated using a robust procedure that reduces the influence of outliers. Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) GDP. Financial openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1 index obtained from de jure labor regulation. Ease of ï¬?rm entry is a 0 2 1 index combining information on number of pro- cedures, monetary cost, and time to open a new ï¬?rm. a Three outliers are excluded, based on the Hadi method. Source: Authors’ analysis based on data described in text. economic activity more generally. The second robustness check addresses the potential shortcoming of GDP deflators in cleaning out (purely nominal) price effects from real GDP. This should be minimal if the GDP and terms of trade data come from the same source and GDP is adjusted for PPP. Then, at the cost of losing the most recent observations, the analysis is rerun using GDP data from the Penn World Tables. The shock-amplifying effect of trade openness remains large, although somewhat lower than in the benchmark case. A second concern is that the results might derive only from the contrast between developing and developed countries. To consider this possibility, all high-income countries are excluded from the sample, the model is reestimated, and the impact statistics are recomputed. The results are qualitatively the same and quantitatively similar to those obtained using the full sample. This simi- larity indicates that the results can be compared with those of studies that focus only on developing countries. The third concern relates to the exogeneity assumption of the terms of trade shock. Although the largest developed countries are excluded from the sample, the small-country assumption may be problematic for countries like Brazil, China, and India. Excluding the largest 10 percent of countries and repeating the exercise yields results with the same sign and quantitatively similar to those of the benchmark.14 The exogeneity assumption may also be questionable for 14. The large countries are Australia, Brazil, Canada, China, India, Republic of Korea, Mexico, Netherlands, and Spain. 374 THE WORLD BANK ECONOMIC REVIEW countries whose main exports are differentiated manufacturing goods with prices that are likely endogenous. To dispel this doubt, countries that are mainly manufacturing exporters are excluded (manufactured products consti- tute more than half of total exports).15 Again the results are basically the same as those of the benchmark.16 A fourth concern pertains to the correct speciï¬?cation of the vector auto- regression lag structure, which may be relevant in evaluating dynamic effects. To dispel doubts on whether preestimation diagnostics could have indicated a longer lag structure, the shock impacts from vector autoregressions are reesti- mated with three lags for all countries. Little if anything changes: the signs of the effects remain the same as the benchmark, and quantitative differences with the benchmark are mostly small and statistically insigniï¬?cant, except for ease of ï¬?rm-entry, whose shock-amplifying effect seems stronger. The ï¬?nal robustness check concerns the exchange rate regime. This was not included in the set of structural determinants since it is generally associ- ated with standard macroeconomic policy. However, since it has received so much attention in the stabilization literature and could in principle be related to the structural characteristics considered here, an additional exercise includes the exchange rate regime as an interaction variable. The Gosh, Gulde, and Wolf (2002) classiï¬?cation is used to separate country-year obser- vations with a pegged regime from those with intermediate and floating regimes. The results are very similar to those of the benchmark. The effect of the exchange rate regime itself is quite small and statistically insigniï¬?cant. This result is only tentative, however, as a complete analysis of the role of the exchange rate regime requires treatment of measurement issues that is outside the scope of this article. As explained, an alternative to estimating the interactions model using panel data is to estimate the simple model country by country (vector autoregression of output growth on terms of trade growth with free coefï¬?cients) and then to run a cross-country regression of the resulting cumulative impacts on the ï¬?ve structural variables. This method allows for full country heterogeneity in (vector autoregression) parameter estimation, but at the cost of lower estimation efï¬?- ciency and increased noise in the individual country impulse responses. Table 6 presents the results using two methods that eliminate the undue influ- ence of outlying observations. The ï¬?rst column shows the results of OLS estimation where three outliers are previously eliminated using the Hadi method.17 The second column shows weighted least squares (WLS) estimation, with the weights 15. The mostly manufacturing exporting countries are Canada, China, Finland, Hong Kong, China, Hungary, Ireland, Israel, Republic of Korea, Singapore, Sweden, and Switzerland. 16. The results are also unaffected by the exclusion of the 13 countries for which the hypothesis is rejected that terms of trade fluctuations are not Granger-caused by output fluctuations. 17. The method was applied using a p-value of 0.3, which resulted in three observations being tagged as outliers. Loayza and Raddatz 375 inversely proportional to the corresponding squared residual.18 The results are qualitatively similar to those obtained from panel vector autoregressions: the two most important country characteristics affecting the shocks’ impact are trade open- ness (magnifying the impact) and labor market flexibility (reducing the impact). Both carry highly signiï¬?cant coefï¬?cients under OLS and WLS. Financial openness has negative coefï¬?cients under both methods and signiï¬?cantly so under OLS. Similarly, ease of ï¬?rm-entry has positive coefï¬?cients under both methods and sig- niï¬?cantly so under WLS. As in the panel vector autoregression case, ï¬?nancial open- ness appears to stabilize the effect of shocks, whereas ease of ï¬?rm-entry appears to enlarge them. Financial depth is not statistically signiï¬?cant under OLS or WLS, as was the case using the panel vector autoregression methodology. Asymmetric Effects The previous analysis can determine whether structural characteristics have a stabilizing (or destabilizing) effect for all shocks, whether positive or negative. In principle, however, this symmetric treatment could mask important differ- ences in the effects of structural characteristics for positive and negative shocks. For instance, an ideal structural characteristic—one that in reality mag- niï¬?es positive shocks and reduces negative ones—could be found to be ineffec- tual under a symmetric analysis. This section considers separately the output response to negative and positive terms of trade shocks. The results of the asymmetric analysis are presented in table 7. The esti- mation of asymmetric shocks presents larger standard errors as it uses fewer observations and suffers from wide data variations associated with sign tran- sitions. In reading the asymmetric results, for negative shocks a more negative value for the cumulative impact indicates a stronger effect, and for positive shocks a more positive value for the cumulative impact denotes a larger effect. There is some evidence of asymmetric effects. The destabilizing effect of trade openness is strong and statistically signiï¬?cant only in the case of negative shocks. Financial depth has no signiï¬?cant effect on the impact of either positive or negative terms of trade shocks. Therefore, its lack of relevance as a shock stabilizer cannot be explained by asymmetric effects. Financial openness does not seem to have a statistically signiï¬?cant effect either, but for different reasons. An increase in ï¬?nancial openness reduces the (absolute) impact of both negative and positive shocks, and by similar magnitudes, found with sym- metric effects (by around 0.4 percentage point). It is not surprising, then, that assuming symmetry in the case of ï¬?nancial openness produces more efï¬?cient estimates and, thus, signiï¬?cant effects. An improvement in labor market flexibility dampens the effect of both nega- tive and positive shocks in a statistically signiï¬?cant way. However, the stabiliz- ing effect is substantially larger for negative shocks than for positive shocks, meaning that labor market flexibility is particularly important in the face of 18. That is, using the “rregâ€? command in STATA. 376 T A B L E 7 . Asymmetric Effects Cumulative Output Impact of One standard Deviation Terms of Trade Negative and Positive Shocks for Low and High Values of Five Structural Characteristics (percent of GDP) Labor market Trade openness Financial depth Financial openness flexibility Ease of ï¬?rm entry Negative Positive Negative Positive Negative Positive Negative Positive Negative Positive Low 2 0.262 0.427 2 1.598 0.853 2 1.791 0.724 2 2.306 0.877 2 1.461 0.210 (0.437) (0.422) (0.418) (0.400) (0.452) (0.436) (0.439) (0.422) (0.460) (0.442) High 2 2.494 0.661 2 1.697 0.386 2 1.429 0.317 2 0.637 0.097 2 2.020 1.249 THE WORLD BANK ECONOMIC REVIEW (0.474) (0.453) (0.484) (0.475) (0.375) (0.369) (0.468) (0.452) (0.481) (0.476) Difference 2 2.231 0.234 2 0.099 2 0.466 0.362 2 0.407 1.669 2 0.781 2 0.559 1.039 (0.523) (0.492) (0.509) (0.504) (0.373) (0.370) (0.481) (0.459) (0.557) (0.548) Test ** ** ** ** Ho:Diff. ¼ 0 (one-tail) **Signiï¬?cant at the 5 percent level. Note: Numbers in parentheses are standard errors of corresponding cumulative output impact. Critical values are obtained from empirical distribution (which may have non-Gaussian properties). Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining information on number of procedures, monetary cost, and time to open a new ï¬?rm. Source: Authors’ analysis based on data described in text. Loayza and Raddatz 377 adverse shocks. Finally, ease of ï¬?rm-entry shows interesting evidence of an asymmetric effect: easing ï¬?rm-entry signiï¬?cantly increases the consequences of positive shocks only. This provides an upbeat spin to the shock-magnifying effect found for ease of ï¬?rm-entry in the benchmark case. Interaction Effects Up to now the output response to terms of trade shock has been allowed to vary linearly with the ï¬?ve structural characteristics. The relevance of each characteristic has been assessed while holding the rest constant. This section examines the output response to terms of trade shocks when the effect of each structural variable is allowed to depend on the rest. This is akin to allowing for multiplicative interactions in a regular regression context, and the focus is on interpreting the equivalent of interaction coefï¬?cients in that context. Allowing for multiplicative interactions is complex, so the analysis is restricted to the case of symmetric effects (of positive and negative shocks). Following the presentation used previously, for a given pair of structural determinants the ï¬?rst is set at its 25th percentile, then the second is varied from its 25th to its 75th percentile and the difference in cumulative output impact is computed. Then the ï¬?rst structural determinant is set at its 75th percentile and the second is again varied from its 25th to its 75th percentile and the corresponding difference in the cumulative output impact is computed. Finally, the difference of the previously computed differences in cumulative output impacts is computed (always high minus low). This difference-in- difference value is the statistic of interest (as mentioned above, it carries analo- gous information to the coefï¬?cient on a regular multiplicative interaction). A negative sign for this difference-in-difference value reveals that the two struc- tural determinants under consideration are complements in dampening the effects of terms of trade shocks: an increase in either one leads to a lower shock impact when the other one is at a high value. Conversely, a positive sign for the difference-in-difference value indicates that they are substitutes: an increase in either brings about a smaller output response to a shock when the other one is at a low value. Table 8 summarizes the results of the interactions model, presenting only the difference-in-difference value for each pair of structural determinants, along with its standard error and its test of statistical signiï¬?cance. The following patterns emerge among the statistically signiï¬?cant results. Financial depth behaves as a complement to trade openness and ï¬?nancial openness. Likewise, labor market flexibility and ease of ï¬?rm-entry are complements. In contrast to the basic case, the interactions model indicates a relevant though nuanced role for ï¬?nancial depth in affecting the impact of external shocks: deepening domestic ï¬?nancial markets can reduce the impact of external shocks when international trade and ï¬?nancial markets are open. This result is consistent with the literature that emphasizes the complementarity between reforms in domestic and international ï¬?nancial markets (see Caballero and Krishnamurthy 2001; Edwards 2001, 378 T A B L E 8 . Complementarities among Structural Characteristics: Differential Cumulative Output Impact of a One Standard Deviation Terms of Trade Shock for Low and High Values Between Pairs of Structural Characteristics (percent of GDP) Trade openness Financial depth Financial openness Labor Market flexibility Ease of ï¬?rm-entry Trade openness Difference-in-difference — 2 1.679 (0.383) 2 0.489 (0.409) 2 0.468 (0.388) 2 0.074 (0.548) Test Ho ¼ Diff-diff ¼ 0 ** — — — Financial depth Difference-in-difference — 2 1.820 (0.474) 0.245 (0.455) 1.353 (0.558) Test Ho ¼ Diff-diff ¼ 0 ** — ** Financial openness Difference-in-difference — 0.993 (0.452) 0.936 (0.436) Test Ho ¼ Diff-diff ¼ 0 ** ** THE WORLD BANK ECONOMIC REVIEW Labor market flexibility Difference-in-difference — 2 2.518 (0.644) Test Ho ¼ Diff-diff ¼ 0 ** Ease of ï¬?rm entry Difference-in-difference — Test Ho ¼ Diff-diff ¼ 0 **Signiï¬?cant at the 5 percent level. Note: Numbers in parentheses are standard deviations of corresponding output impact. Impacts are given in percentage points of GDP. They are the difference between a given pair of structural characteristics of the difference in the cumulative output impact of their corresponding low and high values. This is analogous to the effect of an interaction between a pair of variables. Critical values are obtained from empirical distribution (which may have non-Gaussian properties). Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining information on number of procedures, monetary cost, and time to open a new ï¬?rm. Source: Authors’ analysis based on data described in text. Loayza and Raddatz 379 among others).19 In turn, labor flexibility is more effective in reducing the impact of the shock if ease of ï¬?rm-entry is high, a result that highlights the importance of complementary reforms (see Eslava and others 2005). Three other pairs of variables behave as substitutes. They are ease of ï¬?rm-entry with both ï¬?nancial depth and ï¬?nancial openness, and labor market flexibility with ï¬?nancial openness. Thus, deepening ï¬?nancial markets and opening the capital account reduce the output effect of the shock, particularly when there are impediments to ï¬?rm flexibility. Likewise, labor market flexibility has a larger role in reducing the impact of terms of trade shock when the capital account is closed. It is interesting to note that while ease of ï¬?rm-entry and labor market flexibility are complements, they are substitutes for ï¬?nancial market depth and openness. I V. C O N C L U D I N G R E M A R K S What underlies a country’s vulnerability to external shocks? Why do some countries suffer so much from terms of trade shocks while others remain unscathed? This article examined how certain domestic characteristics influence the impact that terms of trade shocks can have on aggregate output. It has an empirical objective, but the analysis was motivated by the recent literature that emphasizes the role of product- and factor-market rigidities as the source of macroeconomic vulnerability. The results indicate that the two most important country characteristics affecting the output impact of shocks are trade openness and labor market flexibility, with trade openness magnifying the impact and labor market flexi- bility reducing it. Financial openness also shows a signiï¬?cant but smaller stabi- lizing effect. Ease of ï¬?rm-entry magniï¬?es the shocks, but mainly the positive ones, as revealed by an examination of asymmetric effects. Financial depth does not seem to directly affect the impact of terms of trade shocks, but it affects how other structural characteristics amplify or dampen these shocks, as exposed by the analysis of interaction effects. These results are robust to check- ing for mechanical interpretations of the trade-related results, placing stricter restrictions to guarantee shock exogeneity, concentrating exclusively on devel- oping countries, using a longer lag structure for the vector autoregressions, controlling in addition for the exchange rate regime and allowing full hetero- geneity in the estimation of country impulse responses. When the possibility of asymmetric effects (from negative and positive shocks) is considered, trade openness ampliï¬?es negative shocks, whereas ease of ï¬?rm-entry 19. An alternative reading of the previous results may help to clarify the positive role of ï¬?nancial development. First, although trade openness always increases the impact of a shock, this is considerably smaller when the expansion in openness occurs in a country with well developed local ï¬?nancial markets. Similarly, the ï¬?ndings indicate that greater ï¬?nancial openness in an environment of underdeveloped local ï¬?nancial markets may result in an increase in the impact of external shocks. In contrast, when ï¬?nancial openness occurs in a country with well developed ï¬?nancial markets, the impact of the shocks is reduced. 380 THE WORLD BANK ECONOMIC REVIEW magniï¬?es only positive ones. Labor market flexibility dampens both shocks, but especially negative ones, and ï¬?nancial openness seems to reduce both shocks in a similar way. Analysis of the interactions among the structural determinants of the impacts of shocks reveals an interesting pattern. Macroeconomic outcomes (in trade and in ï¬?nancial openness and depth) tend to complement each other: an improvement in any of these dimensions leads to a larger reduction in the output impact of terms of trade shocks in a country that is advanced in the other related dimensions. The same happens for microeconomic conditions (in ease of ï¬?rm-entry and in labor market flexibility), which also tend to be complements. However, macroeconomic and microeconomic conditions seem to behave as sub- stitutes, compensating for each other’s deï¬?ciencies. The article opened by pointing out two aspects of output vulnerability to external shocks: the strength of the shock and the sensitivity to the shock. As the article looked only at sensitivity to the shock, it is only fair to ask whether this is indeed quantitatively relevant in assessing a country’s degree of vulner- ability. The empirical model can answer that question by decomposing the var- iance of total predicted volatility into the portion due to the countries’ sensitivity to a homogeneous terms of trade shock and the fraction due to the variation of these shocks across countries. A conservative estimate of the importance of the portion due to the sensitivity to the shock is 30 percent. The estimate is conservative because it is based on homogeneous parameters across countries, as derived from the panel vector autoregression methodology.20 A more liberal estimate—one based on country-speciï¬?c vector autoregression parameter—would assign an importance more than twice as large. In any case, the relevance of domestic structural characteristics in dealing with external vul- nerability cannot be ignored. A ï¬?nal caveat is in order. The analysis focused on the role of structural characteristics on the ampliï¬?cation of terms of trade shocks only. This is a rele- vant exercise because of the importance typically attributed to these shocks and the advantages it offers for identiï¬?cation purposes. To the extent that the response to other types of external shocks is similar, the results convey infor- mation about the general influence that structural characteristics have on exter- nal vulnerability. However, the possibility that their role in the transmission of other external shocks may differ from the one documented here cannot be 20. This fraction is estimated as follows. For each country in the sample, the long-run output variance is computed in response to its own shock (ai s2 i ), ï¬?rst, by estimating the response to a common shock and then by simulating the effect of its own shock. Actual values of the country’s structural characteristics are used to estimate its long-run output variance in response to a common shock (ais2), and the country’s own data are applied to estimate the variance of its terms of trade shocks (s2 i ). The log of the output variance [log (ai s2i )] is then decomposed into the sum of the log of vulnerability [log (ai)] and the log of terms of trade variance [log (s2 i )]. The cross-country variance of the log of output volatility then corresponds to the cross-country variance of the log of vulnerability, the log of terms of trade variance, and the covariance between them. The ï¬?gure reported in the text corresponds to the contribution of the log of vulnerability to this variance when the covariance term is imputed in proportion to the standard deviation of each component (assuming constant correlation). Loayza and Raddatz 381 dismissed. In particular, ï¬?nancial development, which plays a secondary role in the results of this study, may have a more prominent job in dampening ï¬?nancial shocks. Since an appropriate analysis of this possibility would require more complex and controversial identiï¬?cation assumptions, it awaits future research. APPENDIX COUNTRY SAMPLE AND S U M M A RY S TAT I S T I C List of sample countries and summary statistics is given in table A-1. T A B L E A - 1 . List of Sample Countries and Summary Statistics 382 Average Standard Standard terms of deviation deviation Labor Average output trade growth output terms of Trade Financial Financial market Ease of Country growth (%) (%) growth trade growth openness depth openness flexibility ï¬?rm-entry name (1) (2) (3) (4) (5) (6) (7) (8) (9) Algeria 0.46 2 0.43 2.87 23.14 3.81 2 1.12 2 1.41 0.54 0.66 Angola 2 2.26 2 3.66 9.28 18.08 4.16 2 4.31 2 1.55 0.22 0.22 Argentina 0.27 2 0.01 5.78 8.17 2.70 2 1.78 2 0.13 0.34 0.69 Australia 1.86 2 0.66 1.92 5.06 3.28 2 0.77 1.32 0.64 0.89 Austria 2.20 2 0.31 1.56 1.34 3.97 2 0.25 1.68 0.70 0.74 Bangladesh 1.62 2 2.08 2.27 15.78 2.91 2 1.60 2 1.40 0.50 0.62 Belgium 1.95 2 0.12 1.66 1.59 4.83 2 0.96 1.56 0.52 0.80 Benim 0.55 2 1.69 3.63 14.17 3.71 2 2.37 2 0.24 0.48 0.69 Bolivia 2 0.11 2 3.12 3.00 11.29 3.73 2 1.40 0.68 0.34 0.52 THE WORLD BANK ECONOMIC REVIEW Botswana 5.26 1.49 3.57 8.34 4.63 2 2.06 2 0.21 0.65 0.62 Brazil 1.21 2 1.73 3.68 9.83 2.75 2 1.30 2 1.64 0.22 0.45 Burkina Faso 1.19 0.77 3.43 12.52 3.17 2 2.01 2 0.36 0.47 0.45 Burundi 2 0.61 2 2.78 5.11 33.79 3.28 2 2.48 2 1.09 0.38 0.25 Cameroon 0.61 0.00 7.03 22.39 3.45 2 1.63 2 0.47 0.56 0.59 Canada 1.76 0.18 2.28 3.05 3.90 2 0.37 2.68 0.66 0.94 Central 2 1.42 2 1.27 4.61 16.26 3.22 2 2.60 2 0.66 0.38 0.25 African Republic Chad 2 0.56 2 2.94 9.06 13.46 3.26 2 2.57 2 0.76 0.34 0.41 Chile 3.18 2 2.54 5.75 14.51 3.73 2 0.85 2 1.25 0.50 0.78 China 7.35 2 0.93 3.44 5.74 3.16 2 0.13 2 1.24 0.53 0.61 Colombia 1.34 0.54 2.30 10.19 3.21 2 1.33 2 1.53 0.41 0.65 Congo, Rep. 0.37 2 0.79 7.02 22.26 4.27 2 2.28 2 0.91 0.40 0.58 Costa Rica 1.29 0.14 3.73 9.45 4.12 2 1.69 2 0.56 0.37 0.64 Coˆ te d’lvoire 2 1.14 2 1.95 4.94 16.36 4.05 2 1.16 2 0.53 0.47 0.59 Denmark 1.65 0.40 1.93 2.43 3.97 2 0.89 1.13 0.75 0.91 Dominican 2.27 2 2.49 3.31 11.72 4.02 2 1.37 2 1.46 0.51 0.60 Republic Ecuador 0.40 2 1.73 3.18 13.45 3.71 2 1.53 0.04 0.45 0.51 Egypt, Arab 3.55 2 2.80 2.86 11.33 3.57 2 1.24 2 1.05 0.41 0.59 Rep. El Salvador 0.01 0.07 4.83 17.84 3.90 2 2.71 2 0.64 0.31 0.59 Ethiopia 2 0.09 0.29 7.67 19.72 3.02 2 1.82 2 1.14 0.49 0.69 Finland 2.13 2 0.08 3.05 3.09 3.89 2 0.55 1.54 0.45 0.85 Ghana 2 0.60 2 2.01 5.06 15.93 3.89 2 3.26 2 1.39 0.65 0.55 Greece 1.42 2 1.11 2.46 4.60 3.47 2 0.99 2 0.54 0.33 0.63 Guatemala 0.48 2 1.42 2.59 25.42 3.52 2 1.90 0.63 0.35 0.56 Guinea 1.38 2 3.96 1.42 8.91 3.71 2 3.18 2 1.07 0.40 0.56 Haiti 2 1.59 2 4.08 4.82 12.03 3.34 2 2.2 0.44 0.40 0.32 Honduras 0.53 2 0.59 3.25 13.45 4.07 2 1.23 0.17 0.44 0.56 Hong Kong, 4.56 0.38 4.50 1.75 5.22 0.39 2.68 0.73 0.94 China Hungary 1.69 2 0.88 3.91 3.18 4.36 2 1.25 2 0.68 0.46 0.76 India 3.12 1.64 2.92 10.60 2.57 2 1.45 2 1.03 0.49 0.56 Indonesia 3.87 1.46 4.46 10.94 3.72 2 1.20 2.05 0.43 0.45 Iran, Islamic 2 0.64 2 1.18 7.73 24.31 3.31 2 1.25 2 0.90 0.48 0.63 Rep. Ireland 4.35 2 0.46 3.15 2.55 4.57 2 0.56 0.58 0.51 0.88 Israel 1.86 0.89 1.96 4.16 4.09 2 0.65 2 0.39 0.62 0.83 Jamaica 2 0.21 2 1.60 4.19 8.67 4.17 2 1.33 2 0.36 0.66 0.76 Jordan 1.73 0.68 7.52 7.06 4.31 2 0.49 2 0.18 0.40 0.69 Kenya 0.23 2 0.44 2.33 10.48 3.80 2 1.24 2 0.74 0.66 0.60 Korea, Rep. 5.82 2 0.73 3.79 5.29 4.03 2 0.30 2 0.63 0.49 0.70 Lesotho 2.85 2 0.98 6.64 15.82 4.79 2 2.01 2 0.54 0.55 0.59 Madagascar 2 1.57 0.86 3.67 11.23 3.32 2 1.86 2 0.92 0.39 0.65 Malawi 0.56 2 2.13 5.34 10.94 3.97 2 2.25 2 1.03 0.48 0.63 Malaysia 3.92 2 0.14 4.08 6.99 4.71 2 0.29 1.63 0.75 0.77 Loayza and Raddatz Mali 0.65 0.01 5.93 8.07 3.64 2 2.03 2 0.24 0.46 0.62 (Continued ) 383 384 TABLE A-1. Continued Average Standard Standard terms of deviation deviation Labor Average output trade growth output terms of Trade Financial Financial market Ease of Country growth (%) (%) growth trade growth openness depth openness flexibility ï¬?rm-entry name (1) (2) (3) (4) (5) (6) (7) (8) (9) Mauritania 0.10 0.46 3.36 9.42 4.29 2 1.16 2 1.08 0.41 0.55 Mexico 1.50 2 0.38 3.74 9.90 3.28 2 1.67 0.92 0.23 0.66 Morocco 1.57 1.12 4.98 9.09 3.74 2 1.25 2 1.26 0.49 0.82 Mozambique 0.88 2 3.59 7.90 10.57 3.45 2 2.19 2 1.32 0.26 0.40 THE WORLD BANK ECONOMIC REVIEW Namibia 2 0.47 2 1.94 2.72 11.23 4.58 2 0.97 2 1.18 0.57 0.64 Netherlands 1.83 2 0.14 1.50 1.00 4.52 0.08 2.53 0.46 0.80 New 0.68 0.37 2.39 5.05 3.79 2 0.64 1.70 0.68 0.93 Zealand Nicaragua 2 2.91 2 2.46 7.85 18.30 4.01 2 1.28 0.11 0.39 0.62 Niger 2 1.71 0.10 6.05 17.16 3.47 2 2.13 2 0.53 0.41 0.57 Nigeria 2 0.96 2 0.03 5.59 27.87 4.11 2 2.11 2 1.19 0.57 0.62 Norway 3.04 2 0.48 1.76 7.94 3.95 2 0.22 0.54 0.59 0.82 Pakistan 2.47 2 1.26 1.93 9.80 3.39 2 1.48 2 1.09 0.42 0.65 Panama 1.12 2 0.74 4.92 10.46 3.67 2 0.59 2.68 0.21 0.78 Papua New 0.07 2 1.47 5.43 12.24 4.36 2 1.67 2 0.23 0.74 0.67 Guinea Paraguay 1.32 1.94 4.12 17.48 3.36 2 1.73 2 0.70 0.27 0.50 Peru 2 0.42 2 1.56 6.10 12.96 3.23 2 1.97 0.12 0.27 0.59 Philippines 0.72 0.04 3.76 11.79 3.86 2 1.13 2 0.57 0.40 0.63 Portugal 2.47 0.02 3.23 4.65 3.87 2 0.32 0.09 0.21 0.65 Rwanda 0.32 1.98 10.06 30.39 3.08 2 2.67 2 1.00 0.40 0.55 Senegal 0.19 2 0.85 4.34 6.69 3.90 2 1.29 2 0.24 0.46 0.62 Sierra Leone 2 3.35 2.57 7.15 34.08 3.52 2 3.12 2 0.85 0.33 0.46 Singapore 5.32 2 1.24 2.56 2.08 5.67 2 0.17 2.00 0.80 0.92 South Africa 2 0.38 2 0.79 2.30 5.61 3.75 2 0.69 2 1.12 0.64 0.77 Spain 1.98 0.52 1.76 5.11 3.36 2 0.24 0.36 0.30 0.68 Sri Lanka 3.48 1.15 1.21 14.27 4.02 2 1.73 2 0.52 0.58 0.74 Sweden 1.62 2 0.47 2.00 2.82 3.94 0.01 1.58 0.58 0.84 Switzerland 0.80 1.25 2.38 3.86 3.99 0.31 2.68 0.64 0.79 Syrian Arab 1.57 2 3.00 6.00 13.70 3.80 2 2.68 2 1.64 0.55 0.65 Republic Thailand 4.66 2 1.99 4.37 5.57 3.99 2 0.51 2 0.04 0.39 0.75 Togo 2 0.49 2 2.48 7.08 23.49 4.05 2 1.51 2 0.87 0.43 0.44 Tunisia 2.44 2 1.18 2.65 4.71 4.14 2 0.53 2 0.92 0.43 0.78 Turkey 1.93 2 0.45 4.11 6.89 3.11 2 1.86 2 0.95 0.45 0.77 Uganda 2.05 2 0.91 3.44 20.64 3.11 2 3.69 2 0.47 0.58 0.6 Uruguay 1.59 2 0.26 4.89 6.40 3.35 2 1.29 0.87 0.61 0.68 Venezuela 2 0.94 2.30 4.53 22.09 3.71 2 1.07 0.64 0.25 0.55 Zambia 2 2.23 2 5.99 3.98 26.15 4.08 2 2.78 2 0.71 0.54 0.71 Note: Variables are measured over the period 1974– 2000. Average output growth is growth of real GDP per capita. Average terms of trade growth is the average growth of the terms of trade index. Standard deviation output growth is the standard deviation of the growth rates of real GDP per capita. Trade openness ¼ Log (Exports þ Imports) / GDP. Financial depth ¼ Log (Private credit) / GDP. Financial Openness is the Chinn– Ito measure of capital account openness. Labor market flexibility is a 0 2 1index obtained from de jure labor regulation. Ease of ï¬?rm-entry is a 0 2 1 index combining infor- mation on number of procedures, monetary cost, and time to open a new ï¬?rm. Source: Authors’ analysis based on data described in text Loayza and Raddatz 385 386 THE WORLD BANK ECONOMIC REVIEW REFERENCES Acemoglu, D., S. Johnson, J. Robinson, and Y. Thaicharoen. 2003. “Institutional Causes, Macroeconomic Symptoms, Volatility, Crises and Growth.â€? Journal of Monetary Economics 50(1):49– 123. Ahmed, S. 2003. “Sources of Macroeconomic Fluctuations in Latin America and Implications for Choice of Exchange Rate Regime.â€? Journal of Development Economics 72(1):181– 202. Arellano, C., and E.G. Mendoza. 2002. “Credit Frictions and ‘Sudden Stops’ in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises.â€? NBER Working Paper 8880. Cambridge, Mass.: National Bureau of Economic Research. Beck, T., A. Demirg-Kunt, and R. Levine. 2000. “A New Database on Financial Development and Structure.â€? World Bank Economic Review 14(3):597–605. Bernanke, B., and M. Gertler. 1989. “Agency Costs, Net Worth, and Business Fluctuations.â€? American Economic Review 79(1):14– 31. Botero, J., S. Djankov, R. La Porta, F. Lopez-de-Silanes, and A. Shleifer. 2004. “The Regulation of Labor.â€? Quarterly Journal of Economics 119(4):1339 –82. Braun, M., and B. Larran. 2005. “Finance and the Business Cycle: International Inter-Industry Evidence.â€? Journal of Finance 60(3):1097–128. Broda, C. 2001. “Coping with Terms-of-Trade Shocks: Pegs versus Floats.â€? American Economic Review 91(2):376 – 80. ———. 2004. “Terms-of-Trade and Exchange Rate Regimes in Developing Countries.â€? Journal of International Economics 63(1):31– 58. Caballero, R.J., K. Cowan, E.M.R.A. Engel, and A. Micco. 2005. “Effective Labor Regulation and Microeconomic Flexibility.â€? NBER Working Paper 10744. Cambridge, Mass.: National Bureau of Economic Research. Caballero, R.J., and M.L Hammour. 1994. “The Cleansing Effect of Recessions.â€? American Economic Review 84(5):1350 –68. ———. 1996. “On the Timing and Efï¬?ciency of Creative Destruction.â€? Quarterly Journal of Economics 111(3):805 –52. ———. 1998. “The Macroeconomics of Speciï¬?city.â€? Journal of Political Economy 106(4):724 –67. Caballero, R.J., and A. Krishnamurthy. 2001. “International and Domestic Collateral Constraints in a Model of Emerging Market Crises.â€? Journal of Monetary Economics 48(3):513– 48. Chang, R., L. Kaltani, and N. Loayza. 2005. “Openness Can Be Good for Growth: The Role of Policy Complementarities.â€? NBER Working Paper 11787. Cambridge, Mass.: National Bureau of Economic Research. Chinn, M., and H. Ito. 2002. “Capital Account Liberalization, Institutions and Financial Development: Cross Country Evidence.â€? NBER Working Paper 8697. Cambridge, Mass.: National Bureau of Economic Research. Easterly, W., M. Kremer, L. Pritchett, and L. Summers. 1993. “Good Policy or Good Luck?: Country Growth Performance and Temporary Shocks.â€? Journal of Monetary Economics 32(3):459– 83. Edwards, S. 2001. “Capital Mobility and Economic Performance: Are Emerging Economies Different?â€? NBER Working Paper 8076. Cambridge, Mass.: National Bureau of Economic Research. Eslava, M., J. Haltiwanger, A. Kugler, and M. Kugler. 2005. “Factor Adjustments after Deregulation: Panel Evidence from Colombian Plants.â€? NBER Working Paper 11656. Cambridge, Mass.: National Bureau of Economic Research. Fats, A. 2002. “The Effects of Business Cycles on Growth.â€? In N. Loayza and R. Soto eds., Economic Growth: Sources, Trends, and Cycles. Santiago, Chile: Central Bank of Chile. Gosh, A., A.M. Gulde, and H. Wolf. 2002. Exchange Rate Regimes: Choices and Consequences. Cambridge, Mass.: MIT Press. Loayza and Raddatz 387 Hall, A. 1990. “Testing for A Unit Root in Time Series with Pre-test Data-based Model Selection.â€? Working Paper. North Carolina State University, Department of Economics, Raleigh. Heston, Alan, Robert Summers, and Bettina Aten. 2002. “Penn World Table Version 6.1.â€? Center for International Comparisons at the University of Pennsylvania, Philadelphia. Hnatkovska, V., and N. Loayza. 2005. “Volatility and Growth.â€? In Joshua Aizenmann, and Brian Pinto eds., Managing Volatility. Cambridge, UK: Cambridge University Press. Hoffmaister, A., J. Roldos, and P. Wickham. 1998. “Macroeconomic Fluctuations in Sub-Saharan Africa.â€? International Monetary Fund Staff Paper 45(1):132–60. International Monetary Fund. Various issues. Annual Report on Exchange Arrangements and Exchange Restrictions. Washington, D.C. Kiyotaki, N., and J. Moore. 1997. “Credit Cycles.â€? Journal of Political Economy 211– 48. Kose, M., and R. Riezman. 2001. “Trade Shocks and Macroeconomic Fluctuations in Africa.â€? Journal of Development Economics 65(1):55– 80. Kraay, A., and J Ventura. Forthcoming. “Comparative Advantage and the Cross-Section of Business Cycles.â€? Journal of the European Economic Association. Levin, A., Lin, Chien-Fu, and Chu, Chia-Shang, J. 2002. “Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties.â€? Journal of Econometrics 108(1):1–24. Mendoza, E.G. 1995. “The Terms of Trade, the Real Exchange Rate, and Economic Fluctuations.â€? International Economic Review 36(1):101–37. O’Driscoll, G., E. Feulner, and M.A. O’Grady. 2003. 2003 Index of Economic Freedom. Washington, D.C.: The Heritage Foundation and the Wall Street Journal. Pakes, A., and Z. Griliches. 1984. “Estimating Distributed Lags in Short Panels with an Application to the Speciï¬?cation of Depreciation Patterns and Capital Stock Constructs.â€? Review of Economic Studies 51(2):243 –62. Pedroni, P. 1999. “Critical Values for Cointegration Tests in Heterogeneous Panel with Multiple Regressors.â€? Oxford Bulletin of Economics and Statistics 61(S1):653–70. ———. 2004. “Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series with an Application to the PPP Hypothesis.â€? Econometric Theory 20(3):597–635. Pesaran, M., and R. Smith. 1995. “Estimating Long-run Relationships from Dynamic Heterogeneous Panels.â€? Journal of Econometrics 68(1):79– 113. Raddatz, C. 2006. “Liquidity Needs and Vulnerability to Financial Underdevelopment.â€? Journal of Financial Economics 80(2):677 –722. ———. 2007. “Are external shocks responsible for the instability of output in low-income-countries?â€? Journal of Development Economics 84(1):155– 87. Ramey, G., and V. Ramey. 1995. “Cross-Country Evidence on the Link between Volatility and Growth.â€? American Economic Review 85(5):1138–50. Rodrik, D. 1999. “Where Did All the Growth Go? External Shocks, Social Conflict, and Growth Collapses.â€? Journal of Economic Growth 4(4):385–412. Uribe, M., and V. Yue. 2003. “Country Spreads and Emerging Countries: Who Drives Whom?â€? NBER Working Paper 10018. Cambridge, Mass.: National Bureau of Economic Research. World Bank. 2003. Doing Business. Washington, D.C.: World Bank. ———. 2005. “World Development Indicators Database.â€? World Bank, Washington, D.C.