WPS5515 Policy Research Working Paper 5515 Do Private Inspection Programs Affect Trade Facilitation? Irina Velea Olivier Cadot John S. Wilson The World Bank Development Research Group Trade and Integration Team December 2010 Policy Research Working Paper 5515 Abstract Private inspection of international shipments has positive and significant trade-facilitation effect. These been used over the last half-century for a variety of programs raise import volumes for countries using them purposes. These include prevention of capital flight and by approximately 2 to 10 percent. The findings here also improvement of import duty collection, among others. suggest that the benefit of private inspection of imports The existing literature has failed to find much impact may be associated with reforms and best practices applied of these inspection programs on collected tariff revenue by private inspection firms. Private firms' inspection of or corruption at the border. This paper explores the cargo may promote faster clearance times and process "facilitation" effect of private inspection programs on reliability, rather than improved tax collection. trade. The results indicate that private inspection has a This paper is a product of the Trade and Integration Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Additional information may be accessed at http://econ.worldbank.org/projects/trade_costs olicy. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at ocadot@ worldbank.org, jswilson@worldbank.org, and Irina.Velea@unil.ch. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Do Private Inspection Programs Affect Trade Facilitation?1§ Irina Velea Olivier Cadot John S. Wilson± Keywords: International trade, facilitation, PSI, gravity JEL classification numbers: F13, F15 § This paper is part of a World Bank research project on trade costs and facilitation under the Multidonor Trust Fund on Trade. The assistance of Yassine Cherkaoui in editing this draft is greatly appreciated. Olivier Cadot and Irina Velea would like to thank Cotecna staff for useful discussions while working on a report that lead to this paper. The views expressed in this paper are the authors' and do not necessarily reflect those of the World Bank Group, its Executive Directors, the countries they represent or affiliated institutions. University of Lausanne The World Bank, University of Lausanne, CEPR, CEPREMAP and CERDI. ± The World Bank. 1. Introduction Private inspection of international trade shipments, whether at embarkation or at disembarkation, has been used over the last half-century for a variety of purposes. In the 1970s, when capital controls and exchange-rate regulations were widespread, governments used pre-shipment inspection (PSI) to prevent under-invoicing of exports and over-invoicing of imports, two common forms of capital flight. As exchange-rate misalignment and capital controls progressively receded in the late 1980s, governments turned to PSI to assist weak and corrupt customs to improve import-duty collection.2 In both cases, the vast majority of client governments were in developing countries and PSI was sometimes pressed on less-than- enthusiastic governments by donors. Since the adoption in 2002 of the US Container Security Initiative (CSI), inspection was used for an altogether different purpose -to control the risk of smuggling weapons of mass destruction by international terrorist networks. Client governments for this type of service are typically industrial countries rather than developing ones, and the purpose of the inspection is to check the exact nature of the shipment rather than its value. Both CSI and PSI are elements which impact the transparency of the trading environment of nations and ought to be looked at carefully (on this, see Helble et al., 2009). In all cases, private inspection entails potential costs to traders and thus, ultimately, to final consumers. First, private inspection costs are typically billed to the importing country government and the fees can be substantial (up to 1% of shipment value). Second, inspection can slow down the logistics at embarkation and disembarkation points. The physical form of inspection varies and technology, soft and hard, has greatly improved the way it is done. Best practices today combine scanning (which in itself takes only a few minutes for each container, but often creates bottlenecks in harbor logistics) with random inspection based on risk- assessment techniques (e.g. "profiling" of traders and transit companies). Nevertheless, even if technology has made inspection less burdensome than it used to be, time to export is a significant determinant of comparative advantage for a country (cf. Wilson and Li, 2009) and speed very much defines competitive advantage in the logistics industry - (on this see e.g. Shawdon 2006), and thus concerns have been recently voiced about the nuisance potential of 2 The first PSI program set up with the mission of improving import-tax collection was Indonesia's, in 1985. The mission was then to curb import under-invoicing instead of over- invoicing. Page | 2 mandatory inspection programs.3 Regions like Sub-Saharan Africa suffer from already high trading costs and the apprehension stemming from increased pre-shipment controls are understandable4. Because the CSI is relatively new, it is too early to assess its effect on aggregate trade flows. By contrast, evidence on pre-shipment inspection (PSI) and more recent Destination Inspection (DI) programs is readily available since it has been in use for many years, and the experience should be relevant since the inspection techniques are essentially the same irrespective of what the inspection is done for.5 The idea of this paper is thus to draw from the experience of private inspection programs (PSI and DI) to assess whether inspection affects trade flows and, if so, how. "Private inspection" designates a service provided by a private company to a client government that consists of inspecting ­selectively or systematically--cargoes bound for the client government's country. The inspection can be performed at the embarkation port, prior to loading onto the vessel ("pre-shipment inspection" or PSI), or at the disembarkation port, after unloading ("destination inspection" or DI). The list of programs is given in Appendix 1. The needs underlying the demand for inspection services have evolved over time. Early concerns focused on the manipulation of transfer prices by multinational companies to go around capital controls. This was characteristic for the 1970s when capital controls and exchange-rate regulations were widespread. At the time, the governments used pre-shipment inspection (PSI) to prevent under-invoicing of exports and over-invoicing of imports, two common forms of capital flight. These governments were mainly located in the centre of the African continent, including Congo, Kenya and Tanzania. As exchange-rate misalignment and capital controls progressively receded in the late 1980s, governments from developing countries turned to PSI to assist customs to improve import- duty collection. This time, international donors encouraged adopting inspection programs. Developing countries, in general, comparing to the industrial countries tend to have higher 3 These concerns have been voiced in professional publications lsuch as the Journal of Commerce or Traffic World, in particular in the CSI context. Of particular concern are the 24- hour rule and the Customs-Trade Partnership Against Terrorism program. See also Edmonson (2006). Shawdon (2006) also reports concerned voiced in a series of interviews conducted with 52 intercontinental shippers. Giermanski (2007) however discusses how new technology like "smart containers" can mitigate the slowing-down effect of inspection. 4 Portugal-Perez and Wilson (2009) provide an overview of the barriers to trade in Africa 5 The technology, however, has changed as scanning has become sufficiently cheap to be used in almost all ports and is used by DI. Page | 3 marginal tax rates on imports (Burgess and Stern, 1993). This is because the imports taxes often represent the main source of the overall tax revenue for this group of countries. High import tax rates lead to large rent-seeking opportunities for Importers but also for Customs well placed to take advantage. As a result, these countries are faced with a relatively large Customs corruption problem. PSI offers a simple solution to this problem with a certain cost (typically between 0.5% and 1% of import value). PSI performs the verification of the price and the classification of goods before their departure and then transmits the correct information to the Customs Service at the point of destination and, sometimes, to the destination country Ministry of Finance in advance of the arrival of goods. Although initially perceived by importers as an intruder, PSI survived. In 1988 the WTO gave an official approval by establishing the Agreement on Preshipment Inspection within its framework. As a result, the program extended to Latin American and South Asia often as part of the "Structural Adjustment Programs". The list of programs adopted simultaneously with IMF programs is given in Appendix 2. Inspired by a new customs valuation method developed by the WTO in the late 1990's, private inspection developed a new package of services provided at destination, i.e., in the client's country. Known as Destination Inspection, the service offers reduced clearance times by combining best practices like scanning (which takes only a few minutes for each container) and computerized risk management system (e.g. "profiling" of traders and transit companies). After the introduction of the new service, many countries (like Ghana and Tanzania) shifted from PSI to DI and quality control. Also late 1990, former Soviet Bloc countries (like Moldova or Uzbekistan) joined the program for a short transition period again as part of the structural adjustment programs. Characterizing the period after 9/11 and the adoption in 2002 of the US Container Security Initiative (CSI), inspection started to be used for a quite different purpose - to control the risk of smuggling of weapons of mass destruction by international terrorist networks. Potential client governments for this type of services are from all countries and the purpose of the inspection is to check the exact nature of the shipment rather than its value. Although private inspection and PSI have always been controversial, the empirical literature on its effects is relatively limited and its results are ambiguous. Low (1995) surveyed the evidence from an institutional point of view and noted that performance varied widely depending on the form of the contract and on the relationship between the surveillance Page | 4 company and the client government.6 However one empirical regularity was that capacity building (the transfer of knowledge by private surveillance companies to customs) was typically the weak point of PSI. Yang (2005) studied the Philippines' PSI program and showed that when a new origin country was covered, imports from that country were deflected to the Philippines' export-processing zones and then fraudulently brought into the domestic market. Thus, the program was generating new forms of fraud rather than curbing it. Anson et al. (2006) studied four PSI programs and showed that they had two effects working at cross- purposes. On one hand, PSI generates valuable information that the client government can use to curb fraud. On the other hand, it de-motivates customs administrations, so that part of the information brought in by the private company simply substitutes for slackening customs effort. Using a simple model they showed that the "perverse" effect (customs de-motivation) can sometimes more than offset the direct effect, resulting in less duty collection rather than more, and verified that this indeed seemed to be the case in Argentina and Indonesia. Yang (2008) was more positive, showing that PSI did have some effect in curbing customs corruption. This literature leaves an important question unanswered. Improving tax collection for budget purposes is one thing; encouraging trade is another. A vast literature suggests that trade openness is correlated with economic growth (for a recent empirical investigation of this link, see Wacziarg and Welsh 2008). Another strand of literature highlights the importance of trade facilitation in encouraging trade, relative to the reduction in traditional tariff and non- tariff barriers (see e.g. Wilson, Mann and Otsuki 2003, Portugal-Perez and Wilson 2010). Does private inspection encourage trade? Private surveillance companies typically bring with them best practices in terms of customs and transit procedures, potentially resulting in shorter and less variable clearance times. This could encourage trade, partly or totally offsetting the slowing-down effect discussed above. In their study analyzing the relationship between trade facilitation, trade flows and GDP per capita in the Asia-Pacific region (for the Asia-Pacific Economic Cooperation's members), Wilson et al,(2003) define and measure trade facilitation using four different indicators for port efficiency, customs environment, regulatory environment, and electronic-business usage. They further introduce these indicators in an augmented gravity equation specification. They estimate their model solely in a cross-section context though, for the Asia-Pacific countries. In the absence of specific empirical data, the authors rely on survey information in order to construct the four indicators. Their sources are as diverse as to include the World Economic 6 Ramirez (1992) and Byrne (1995) also provide brief discussions as to the pros and cons of PSI. Page | 5 Forum Global Competitiveness report as well as the World Competitiveness Yearbook from IMD Lausanne and the Maritime Transport Costs and Port Efficiency from the World Bank Group. This data is not always available for all countries. In addition it can only be used with caution, while resulting mostly from qualitative analysis. Our study differs from the paper of Wilson et al. (2003) as we do not try to measure trade facilitation as such. Instead we set up to isolate and evaluate the causal effect of one program assumed here to affect trade and then interpret its effect given the knowledge we have about the program specific characteristics. In this case the robustness of the method of estimation is more important. We set out to explore the effect of private inspection on trade volumes in this paper, using the most common vehicle to test the effect of policy changes on trade volumes: the gravity equation. We estimate a standard version of the gravity equation with importer and exporter fixed effects on a large panel of countries tracked over twenty-six years. Estimating the effect of private inspection on trade flows is a standard case of "treatment effects". We use a difference-in-differences (DID) estimator which consists, intuitively, of comparing the difference in an outcome variable (here the volume of trade) between before and after the treatment for a "treatment group" (here the countries that had private inspection programs at some point) and for a "control group" (here the countries that never had private inspection programs). The average effect of the treatment on the treated is then the estimate of the coefficient on a dummy variable equal to one during the treatment period (private inspection program) for the treatment group (private inspection-using countries). DID estimation raises two specific issues. First, as discussed in Bertrand, Duflo and Mullainathan (2004), possible serial correlation in the outcome variable (here trade) is exacerbated by very strong autocorrelation in the treatment variable (a binary variable that changes value only once or twice in the sample period). The result, they show, is a high probability of type-I error (rejecting the null hypothesis of no effect when it is true, i.e. when there is no effect).7 We control for this using the two-step method suggested in their paper. In the first stage, we run a standard panel gravity equation for all countries and years and retrieve the residuals. In the second stage, we keep those residuals only for the treatment group (countries that had a private inspection program at some point), take their without- and 7 Performing repeated estimations on "manufactured data" with non-existent (placebo) treatments, they showed that the null was rejected in up to half the cases when it should be no more than 5% if standard errors had been correctly estimated. Page | 6 with-treatment averages, and run a panel regression on the resulting two-period panel with a dummy for the treatment period. Second, when the treatment is a policy -as it is here- its assignment is at least partly the voluntary decision of the treated. Therefore, it depends on observed and unobserved characteristics of the country. If it depended only on observed characteristics ("selection on observables") those could be controlled for in the treatment-effect regression. But if it also depends also on unobservable characteristics correlated with trade volumes, there will be correlation between one regressor (the dummy variable marking private inspection treatment) and the error term. OLS estimates will then be biased. We address this endogeneity bias in several ways. First, following the tradition in the policy-evaluation literature, we use fixed effects to control for unobserved time-invariant country characteristics that might affect both their trade performance and their willingness to adopt a private inspection program. However, as discussed in Besley and Case (2000), this fix may not be sufficient if omitted characteristics are time-variant. They suggest an instrumental-variable approach in which domestic political factors are used as instruments. We follow this approach and instrument PSI by governance indicators and their squares. For robustness, we compare our IV estimates with those obtained from Blundell and Bond's 8 system-GMM. We also check for "fortuitous" policy effects by running our DID regression with fictitious starting points for private inspection programs, five years before and after the program was implemented, and comparing the measured effects with those when the true starting date was used. If effects are apparent several years before private inspection programs are put in place, it is likely that those measured effects are fortuitous. The results so far suggest a positive effect of private inspection on trade volumes, with plausible magnitudes ­roughly between 5% and 10% of trade volumes. The effect survives our various robustness checks and seems, as far as we can tell from aggregate data, to vindicate the anecdotal evidence on improvements in transit environments brought about by private firms as part of inspection contracts. The paper is organized as follows. Section 2 presents the data and estimation issues, section 3 discusses the results, and section 4 concludes. 8 System-GMM runs two equations, one in levels and one in first differences, in which endogenous RHS variables are instrumented in the level equation by first differences and in the differenced equation by their past levels. Page | 7 2. Data & estimation 2.1 Data The sample is a panel including 179 importing countries and 170 exporting countries over 1980-2005, or a notional total of 791,180 one-way trade observations, of which 321,348 have positive trade (41%). The dependent variable is the aggregate one-way trade value reported in the IMF's Direction of Trade Statistics (DOTS). It is customary in trade-volume studies to "mirror" export statistics, i.e. to disregard direct export statistics from the exporting country and instead to use import data from its partners. The reason is that customs typically monitor imports (on which duties are based) better than exports (rarely taxed). However our study purports to measure only the trade-facilitation effect of PSI, not its effect on the capacity of customs to record imports correctly. Mirrored import data would confound these two effects and would thus potentially bias results upward, generating a statistical illusion of increased volumes. In order to avoid this source of bias, we use direct export statistics, at the cost of having noisier data than if we had used mirroring. Standard gravity regressors include GDPs in constant 1995 dollars, taken from the World Bank's World Development Indicators; "great-circle" distances between the main industrial agglomerations of countries in the sample, taken from CEPII,9 and dummy variables for common land borders, common official languages, and formal colonial ties. The "treatment variable" is equal to one when an inspection program is in force in the importing country j of a pair (i,j) at time t. It covers programs run by the largest four firms in the industry, Société Générale de Surveillance (SGS) and Cotecna Inspection SA, both based in Geneva; BIVAC International (a subsidiary of Paris-based Bureau Veritas), and Intertek, based in London. The list of programs is given in Appendix 1. Table 1 shows descriptive statistics. For dummy variables, the mean is the proportion of the sample's observations for which the variable is equal to one, i.e. the incidence of the variable. 9 So-called "great-circle" distances are simply the shortest routes along the earth's surface, irrespective of actual terrain. Page | 8 Table 1 Descriptive statistics Figure 1 shows the percentage of importing countries with inspection programs in force over the sample period and the percentage of trade flows covered. Figure 1 Percent of importing countries and trade flows covered by inspection programs, 1980-2006 It can be seen that, at about 5%, the proportion of trade covered by inspection is much smaller than the proportion of countries with inspection programs in force (close to a quarter). This is because countries with inspection programs tend to be poor. Both proportions also rose roughly up to the mid-1990s, after which they peaked and showed a timid downward blip in the last year or so. 2.2 Estimation Our basic equation is a Difference-in-Differences (DID) augmented-gravity equation in which the "treatment group" is the set of countries having used inspection programs and the "control group" is the set of countries never having used such programs. That is, let M ijt be the log of j's aggregate imports from i in year t, yit and y jt the log of the exporter's and importer's GDPs in year t, ij the log of their distance, X ij a vector of time-invariant controls (common language, common border, past colonial ties), ci and c j exporter and importer fixed effects, and t a time effect for year t. The equation is M ijt 1 yit 2 y jt 3 ij 4 p jt X ij (1) i j t where 1 if j has an inspection program in force in year t p jt (2) 0 otherwise. Page | 9 The estimate of 4 is the basic estimate of the average treatment effect for the treated. However, as discussed in the introduction, OLS estimates of (1) suffer from several possible sources of bias. First, as discussed in Bertrand, Duflo and Mullainathan (2004), possible serial correlation in the outcome variable (here trade) is exacerbated by very strong autocorrelation in the treatment variable (a binary variable that changes value only once or twice in the sample period). The result, they show, is a high probability of type-I error (rejecting the null hypothesis of no effect when it is true, i.e. when there is actually no effect).10 We control this by using a two-step method suggested in their paper. In the first stage, we run a standard panel gravity equation for all countries and years and retrieve the residuals. In the second stage, we keep those residuals only for the treatment group (countries that had an inspection program at some point), take their without- and with-treatment averages, and run a panel regression on the resulting two-period panel with a dummy for the treatment period. In addition to autocorrelation issues, the coefficient on the treatment variable would provide an unbiased estimate of the effect of the treatment only if that treatment was randomly assigned, so that individuals who did not get it (the control group) were statistically identical to those who did (the treatment group). This is obviously not the case here since the treatment (an inspection program) is chosen by the individuals (the countries). Beyond the observable characteristics that may influence the decision to adopt an inspection program, which can be controlled for in the regression, there may be unobservable characteristics, picked up by the error term, that also influence the decision to adopt inspection programs. In that case the inspection variable would be correlated with the error term and all estimates would be inconsistent. Several fixes can be used, none of which is perfect (see Besley and Case 2001 for a discussion). First, country fixed effects control for time-invariant country characteristics that may simultaneously depress the country's foreign trade and raise its probability of adopting inspection programs. However, fixed country effects do not control for time-variant omitted variables. To control for those, the treatment variable can be instrumented with variables correlated to it but not to the error term. The estimation procedure is implemented by Stata's treatreg command and is akin to a Heckman two-stage procedure. The first stage is a probit regression of the treatment on the instruments and all exogenous variables, and the second stage is an OLS regression of the outcome variable on the treatment, all exogenous variables, 10 Performing repeated estimations on "manufactured data" with non-existent (placebo) treatments, they showed that the null was rejected in up to half the cases when it should be no more than 5% if standard errors had been correctly estimated. Page | 10 and the estimated hazard rate retrieved from the first stage. The two stages can also (and more efficiently) be estimated simultaneously by maximum-likelihood.11 The key issue is of course to find good instruments. We follow the approach suggested by Besley and Case (2001) and use political instruments taken from the World Bank's worldwide governance database. Specifically, we use Kaufman's governance index (the WB- Worldwide Governance Indicators (WGI) project12) and its square, based on the idea that very corrupt and very honest governments would be reluctant to use inspection programs, the first because it would reduce corruption opportunities and the second because inspection would not be needed. This suggests that the probability of using an inspection program (the dependent variable in the first-stage probit) is a concave function of governance, something which can be captured by a second-degree polynomial. In some specifications, we also use IMF structural adjustment programs that were adopted simultaneously with PSI, as well as the official amount of OECD aid per country per year. When no good instruments are available, the alternative route consists of using the Generalized Method of Moments (GMM), and in particular Blundell and Bond's system-GMM (Blundell and Bond 1998). The system-GMM estimator runs two simultaneous equations, one in levels and one in first differences, in which endogenous RHS variables in levels are instrumented by their lagged first differences in the level equation, and by their lagged levels in the differenced equation. Suitable instruments can of course be added in the levels equation if available. The number of lags used as instruments is the experimenter's choice, and it has been argued that this latitude sometimes makes results unstable. There is however a rule of thumb which states that the number of lags should not be larger than the number of individuals in the panel. In our case, this rule of thumb is of course not binding because the number of lags is anyway severely limited by the panel's short duration. We report GMM results and compare them with IV results for robustness. 11 Both options are available in Stata's treatreg command, and results reported in this paper are based on the second method. 12 http://info.worldbank.org/governance/wgi/index.asp Page | 11 3. Results 3.1 Baseline estimates Table 2 presents estimation results using OLS and instrumental-variable estimation. Table 2 Baseline results: OLS and IV Regression (a) gives OLS estimates (with importer, exporter and time effects) for comparison with instrumental-variable regressions. Regressions (b), (c) and (d) are two-stage treatment- effect regressions where PSI is instrumented in three different ways. In (b), the first-stage regression (a probit) is on corruption and its square. As expected, the probability of observing a PSI program is a concave function of corruption control, but the turning point is around an index value of 10, so beyond the relevant range it rises in corruption control. Both instruments combined pass Sargan's overidentification test. In (c) and (d), the first-stage regression is on a variety of instruments including other governance indicators, the amount of OECD aid and several dummy signaling the presence of World Bank Structural Adjustment Programs (identified in Appendix 2), which combined pass Sargan's test in (c). Results are very similar both qualitatively and quantitatively. The coefficient on private inspection is stable across specifications and there is no significant bias in OLS estimation. Given the semi-log specification, the coefficient on private inspection is the percent increase (decrease) in trade associated with a change in the private inspection variable from zero to one. The estimates in Table 2 suggest an increase between 8% and 11.5%. While in a plausible range, these estimates are quite high, suggesting, if true, that with an average tariff rate of, say 10% private inspection would roughly just pay for itself even if it did nothing to tariff avoidance. 3.2 Robustness Table 3, Table 4 and Table 5 report several robustness exercises. First, while the baseline results are realized using exports data RHS, we check if they are robust to using mirror data instead. Table 3 presents the results with imports data RHS which are very similar with the results in Table 2. Error! Reference source not found.Table 3 Page | 12 Robustness checks 1 : Mirror data Second, as discussed earlier, we follow the methodology suggested by Bertrand, Duflo and Mulainathan (2004, henceforth BDM) to verify that our diff-in-diff estimates are not driven by serial correlation. In that procedure, the first stage is a standard gravity equation without the treatment variable (private inspection), and the second stage regresses, for treated countries only, the average residuals from the first stage, averaged over pre- and post- treatment periods, on the treatment variable. The sample for the second stage is thus a two- period panel including private inspection countries only (9'884 observations). Results are reported in the first two columns of Table 4, labeled (a). It can be seen that private inspection remains significant at the 1% level with an estimated coefficient (0.104) that is just between the OLS and IV interval estimates (0.0804-0.115). This suggests that our positive and significant results from diff-in-diff estimation are not attributable to the over-rejection of the null hypothesis generated by serial correlation in the error term.Error! Reference source not found. Robustness checks 2 Third, we re-run our treatment-effect regression but we "trick" the estimator by feeding in wrong dates for the start and end of private inspection programs. If we were to find significant estimates with the wrong dates, it would be likely that those estimates (and, by the same argument, those obtained with the right dates) were spurious. In regressions (b) and (c) in Table 4, we move the start and end dates of private inspection programs forward by 5 years and backward by 5 years. It can be seen that the estimates become insignificant, suggesting that the estimates in Table 2 are not spurious. Fourth, we re-estimate (1) by GMM (columns (d) and (e) in Table 4). Estimates are comparable with the exception of the coefficient on GDP exporter that is multiplied by 2 comparing to its equivalent in the OLS estimation. Thus, it is necessary to be cautious about the magnitudes, although the qualitative results seem robust. The test for AR (2) in first differences rejects the null hypothesis of no autocorrelations in levels. Fifth, and perhaps more importantly, it may be argued that private inspection programs have been introduced as part of broader reform packages that sometimes included trade liberalization. If that was the case, the effect of private inspection would be confounded with that of trade liberalization and we would attribute to private inspection effects that do not belong to it. The first columns ((a), (b) and (c)) of Error! Reference source not found. show estimation results with a direct measure of trade liberalization included as an additional regressor. We used the binary trade-liberalization index constructed by Wacziarg and Welsh (2008), which takes into account a broad range of trade-policy measures to precisely identify the timing of trade-liberalization episodes. Wacziarg and Welsh's index bears some similarity Page | 13 with the celebrated Sachs-Warner index but excludes non-trade related components like the black-market premium on foreign exchange, and covers a broad range of years and countries. Including the Wacziarg-Welsh dummy to control for trade liberalization does not affect our central result. Table 5 Robustness checks 3 Finally, in the past, many studies used log GDPs as proxies for the importer and the exporter specific factors, which is currently considered to be the gold medal mistake (Baldwin and Taglioni, 2006). The current practice has been mostly moving towards using fixed effects instead (ex. Feenstra, 2003). As an alternative to fixed effects and also relevant for panel estimation is to calculate the multilateral resistance terms (also called remoteness). In particular Carrere, 2006 and Baier and Bergstrand, 2002 suggest to construct a proxy using values of between 2 and 6, as follows: N Ri X k ( Dik )1 1 i Pi k 1,k i N R j X k ( D jk )1 1 j Pj k 1,k j where N represents the total number of countries in the sample. The panel estimation enables one to estimate remoteness and to identify the country specific fixed effects at the same time and we show these results in columns (d) (e) and (f) of Table 5. The private inspection variable remains significant and comparable with previous results. All in all, at this stage it seems safe to say that private inspection seems to be robustly associated with an increase in bilateral trade volumes in a range centered around 5% and stretching between roughly 2% and 10%. Because our exercise is based on aggregate data, we cannot tell whether this increase is uniformly spread over product categories, so no inference should be drawn on the effect that private inspection can have on collected duties. Indeed, as discussed in the introduction, results so far are inconclusive on that front. Page | 14 3.3 Extensions We add two extensions to the estimations above. First, we re-estimate (1) taking into account zero-trade values.13. The first five columns of Table 6 show results using a Poisson estimator for five-year periods. The estimated treatment effect is positive and significant for the last four periods, so qualitative results are unaffected. However, again the size is smaller. For 2000-2004 it is a small 1.45%; while for earlier periods it is between 2.7% and 6%. For 1980- 84, it is negative, corresponding to a period where inspection programs had a different mission (see the earlier discussion). The last column reports Tobit marginal effects, which are in the same ballpark (5.47%). Thus, by and large the message is unaffected, although exact magnitudes may be closer to 2-5% than to 10%. Table 6 Extension results 1 : Poisson and Tobit Second, we test the persistence of the effect of private inspection programs by preventing the PSI dummy switch to switch back to zero. The results are presented in Table 7. The effect of PSI is not proved persistent. The statistical results are mixed. Thus, the estimated coefficient for the persistent effect is significant for the OLS regression in column (a) and for one Treatreg specification with a longer sample, in column (d). It is insignificant in columns (b) and (c) that show treatreg specifications with each time with a shorter sample. The fact that the effect of PSI may not be always (for all countries and for all years) persistent was already remarked by Law, 1995 that indicates that the benefit of private inspection programs remains visible only when the program is in place and that this program often does not seem to bring long term benefits to the country. Johnson, 2001 also remarks that the positive impact of PSI persists after the program removal only when the Institutions in the country undertake real reforms as well. He mentions the case of Philippines, Mexico and Argentina as positive examples. Table 7 Extention results 2 : Test of persistent effect 13 Zero trade values were also replaced for missing trade values in two different cases ­ first, when trade values appeared missing for all 26 years from 1980 to 2005 and second when trade values were missing for at least 6 consecutive years after 1992. Page | 15 4. Concluding remarks Our results seem to suggest that the facilitation effect of private inspection is statistically traceable in trade data. Using bilateral export volumes as our variable of interest (in order to avoid picking up the effect of improved bookkeeping on the importing side, which would have no "real" counterpart) we find that private inspection seems to have been associated with an increase in trade between 5% and 10%. Thus, as conjectured at the start of this paper, the burden of inspection and paperwork associated with inspection seems to have been more than offset by improved facilitation at destination. Indeed, anecdotal evidence seems to suggest that at least some surveillance companies have implemented programs to foster the adoption of best practices at destination ports and customs, including container tracking systems, electronic payment of duties, and so on. Technology is clearly part of the story. For instance, recent destination inspection programs have spread the use of scanners in lieu of physical inspection (physical inspection being reserved for suspicious cases). The spread of control systems in international trade may thus be similarly associated with improved logistics and enhanced used of information systems which may, in the end, facilitate transit rather than hampering it. Page | 16 References Anson, Jose, O. Cadot and M. Olarreaga. 2006. "Tariff Evasion and Customs Corruption: Does Pre-Shipment Inspection Help?" Contributions to Economic Analysis & Policy 5, Article 33. Baier, S.L., J.H. Bergstrand, 2001. "The growth of world trade: Tariffs, transport costs, and income similarity. Journal of International Economics 53, 1­27. Baldwin, R., D. Taglioni. 2006. "Gravity for Dummies and Dummies for Gravity Equations", Discussion Paper No. 5850, Centre for Economic Policy Research Bertrand, Marianne, E. Duflo and S. Mullainathan. 2004. "How Much Should We Trust Differences-in-Differences Estimates?" Quarterly Journal of Economics 119, 249-275. Besley, Timothy, and A. Case. 2000. "Unnatural Experiments? Estimating the Incidence of Endogenous Policies." Economic Journal 110, F672-F694. Byrne, Peter. 1995. "An Overview of Privatization in the Area of Tax Administration." Bulletin for International Fiscal Documentation 49, 10-16. Burgess, R. & Stern, N. 1993, "Taxation and Development," Journal of Economic Literature 31: 762-830. Carrere, C. 2006. "Revisiting the effects of Regional Trade Agreements on Trade Flows with Proper Specification of the Gravity Model", European Economic Review 50, 223-247 De Dios, Loreli C. 2009. "Business Views on Trade Facilitation." Economic Research Institute for ASEAN and East Asia, Papers d016. Djankov, S., C. Freund and C.S. Pham. 2010. "Trading on Time." Review of Economics and statistics, 92:1, 166-173. Dreher, Axel. 2006. "IMF and Economic Growth: The Effects of Programs, Loans, and Compliance with Conditionality." World Development 34, 769-788. Eaton Jonathan, and A. Tamura. 1994. "Bilateralism and regionalism in Japanese and U.S. trade and direct foreign investment patterns." Journal of the Japanese and International Economies 8, 478­510. Page | 17 Feenstra, R. 2003. "Advanced International Trade". Princeton University Press, Princeton, NI. Helble, Mathias, Ben Shepherd and John S. Wilson. 2009. "Transparency, Trade Costs, and Regional Integration in the Asia-Pacific", The World Economy, Volume 32, Issue 3, pp. 479- 508. Helpman, Elhanan; M. Mélitz and Y. Rubinstein. 2008. "Estimating Trade Flows: Trading Partners and Trade Volumes." Quarterly Journal of Economics 123, 441-487. Johnson, Noel. 2001. "Committing to Civil Service Reform: The Performance of Pre-Shipment Inspection under Different Institutional Regimes." Policy Research Working Paper No 2594, World Bank, Washington D.C. Lesser, C. and E. Moisé-Leeman. 2009. "Informal Cross-Border Trade and Trade Facilitation Reform in Sub-Saharan Africa." OECD Trade Policy Working Papers No. 86, OECD publishing, OECD. Low, Patrick. 1995. "Preshipment Inspection Services"; World Bank Discussion Paper, World Bank, D.C. Santos Silva, J.M.C., and S. Tenreyro. 2006. "The Log of Gravity." Review of Economics and Statistics, Vol. 88, No. 4, Pages 641-658. Shawdon, C. 2006. "What do global shippers really think?" Supply Chain Management Review, Vol 10, Iss.9, 6-7. Simpson, J. 2009. "Global Enabling Trade Report." World Economic Forum. Yang, Dean. 2008. "Can Enforcement Backfire? Crime Displacement in the Context of Customs Reform in the Philippines." Review of Economics and Statistics 90:1, 1-14. Yang, Dean. 2008. "Integrity for Hire: An Analysis of Widespread Customs Reform." Journal of Law & Economics, 51:1, 25-57. Wacziarg, Romain, and K. Welsh. 2008. "Trade Liberalization and Growth: New Evidence." World Bank Economic Review 22, 187-231. Page | 18 Weerakoon, D. and J. Thennakoon. 2009. "South Asian Yearbook of Trade and Development, Harnessing Gains from Trade: Domestic Challenges and Beyond." Centre for Trade and Development New Delhi, Academic Foundation. Wilson, John S. and Alberto Portugal-Perez. 2010. "Export Performance and Trade Facilitation Reform: Hard and Soft Infrastructure." Policy Research Working Paper No 5261. World Bank Wilson, John S. and Yue Li. 2009. "Time as a Determinant of Comparative Advantage." Policy Research Working Paper No 5128, World Bank, Washington D.C. Wilson, John S. and Alberto Portugal-Perez. 2009. "Why trade facilitation matters to Africa," World Trade Review, Volume 8, Issue 3, pp. 379-416. Wilson, John S., Catherine Mann, and Tsunehiro Otsuki . 2005. "Assessing the Benefits of Trade Facilitation: A Global Perspective." World Economy 28(6): 841-71. Page | 19 Appendix 1 Table 1.1 Programs covered Contract value Country Period Company Type of contract a/ 1979-2001 SGS Exclusive 15 Angola Jan 2002-Jun 2005 Cotecna 2002-2006 BIVAC Argentina 1998-2001 SGS Importer's Choice 25 1994-1997 SGS Geographical 10 Bangladesh Nov 1994-Jan 1999/Sept 2005 - 2006 Cotecna 3 other companies involved Belarus 1997-1999 BIVAC Jan 1991 - Jan 1996 Cotecna Exclusive Benin 2001-2006 BIVAC Bolivia 1986-1987/1990-2003 SGS Importer's Choice 10 1992-2004 SGS Exclusive 5 Burkina Faso July 2004-2005 Cotecna Exclusive Burundi 1978-2008 SGS Exclusive 1 1995-1999/2000-2004 SGS Exclusive 3 Cambodia 2005-2006 BIVAC Cameroon 1988-2006 SGS Exclusive 15 1990-2003 SGS Exclusive 1 Central Afr. Rep. 2004-2006 BIVAC Sept 1994-Jan 1996 Cotecna Exclusive Chad 2004-2006 BIVAC Colombia 95- Feb 1998-Jun 1999 Cotecna 1 other company involved Comoros 1995-2005 Cotecna Exclusive 1993-1998 SGS Exclusive 2 Congo (Brazzaville) 2001-2006 BIVAC/Cotecna 1965-2006 SGS Exclusive 3 Congo (Dem. Rep.) 2006 BIVAC 1975-2000 SGS Exclusive 15 Cote D'ivoire Jul 2000-Mar 2005 Cotecna 1 other company involved 2006 BIVAC Djibouti Jun 1996-Nov 1997 Cotecna Exclusive Dominican Republic Apr 2003-Apr 2005 Cotecna Exclusive 1985-1988/1994-2006 SGS Importer's Choice 12 Ecuador 1994-2005 Cotecna 3 other companies involved Equat. Guinea 1982-1989 SGS Exclusive 0.1 Ethiopia 1993-1995/2000-2004 SGS Exclusive 10 1971-1997 SGS Geographical Split 6 Ghana Aug1994-2005 Cotecna 2 other companies involved Gambia 2001-2002 BIVAC Georgia 1999-2001 Intertek Guatemala 1986-1987 SGS Exclusive 3 Guinea (Conakry) 1996-2004 SGS Exclusive 6 Page | 20 Table 1.1 (continued) Programs covered Contract value Country Period Company Type of contract a/ Haiti 1983-1994/2003-2006 SGS Exclusive 1/3 India 2001-2006 SGS Exporter/Importer Choice n.a. Indonesia 1985-1997/2003-2006 SGS Exclusive 2 Irak Feb 1998-Oct 2004 Cotecna Exclusive Iran 1996-2006 SGS Importer's Choice 3 Jamaica 1985-1988 SGS Exclusive 4 Kazakhstan 1995-1997 SGS Exclusive 11 1972-1990/1994-1999 SGS Geographical Split 3/10 Kenya Jan 1988 ­June 2005 Cotecna 2/1 other companies involved Laos 2001-2002 BIVAC 1985-1997 SGS Exclusive 4 Liberia 1997-2006 BIVAC 1983-1991/2003-2007 SGS Exclusive 12 Madagascar 1993-2002 BIVAC 1989-1997/2001-2003 SGS Exclusive 5 Malawi 2004-2006 Intertek 1989-2003 SGS Exclusive 7 Mali Oct 2003-2005 Cotecna Exclusive 2007 Intertek Mauritania 1994-1995/1999-2001/2004-2005 SGS Exclusive 2 1985-2006 SGS Importer's Choice 1 Mexico Mar 2006- Cotecna 2 other companies involved Moldova 2001-2003 SGS Exclusive 4 1991-1996 SGS Exclusive 5 Mozambique 2001-2006 Intertek Niger 1996-2005 Cotecna Exclusive 1979-1984/1990-1997/1999-2006 SGS Geographical Split 30 Nigeria 1984-1997/1999-2006 Cotecna 3 other companies involved 1995-1997 SGS Geographical Split 45 Pakistan Jul 1990-Nov 1991/Jan 1995-Mar.1997 Cotecna 1 other company involved Paraguay 1983-1988/1996-1999 SGS Importer's Choice 20 1987-1989/1992-2004 SGS Importer's Choice 25 Peru Mar 1992-May 2004 Cotecna 2 other companies involved Philippines 1986-2000 SGS Exclusive 112 Rwanda 1977-2003 SGS Exclusive 2 1991-2001 SGS Importer's Choice 7 Senegal Oct 2001-2005 Cotecna Exclusive 1990-2003 BIVAC Sierra Leone 2004-2006 Intertek Somalia Sept 1990 -1991 Cotecna Exclusive Suriname 1982-1990 SGS Exclusive 0.5 1972-1998 SGS Exclusive 9 Tanzania Sep 1992-April 1995/1999-2005 Cotecna with SGS and then Exclusive 1988-1989 SGS Exclusive 4 Togo 1995-2006 Cotecna Exclusive 1982-1998 SGS Exclusive 4 Uganda 2001 Intertek Uzbekistan 2001-2007 SGS Importer's Choice 2 1986-1989/2003-2005 SGS Importer's Choice 40 Venezuela Sept 2003-Aug 2005 Cotecna 3 other companies involved Zambia 1977-1998 SGS Exclusive 5 Zanzibar 1982-2004 SGS Exclusive 0.1 Notes a/ In million US$ (current) Source: industry data Page | 21 Appendix 2 Table 2.1 Programs adopted simultaneously with IMF Structural Adjustment Programs Country Year IMF Program Notes Argentina 1998 EFF Bangladesh 2003 PRGF Benin 1991 SAF In place since1989 Bolivia 1986 SBA Burkina Faso 1992 SAF In place since 1991 Burundi 1987 SAF Cambodia 1995 PRGF In place since 1994 Cameroon 1989 SBA In place since 1988 Chad 1994 SBA Congo (Roc) 1994 SBA Cote D`Ivoire 1981 EFF Djibouti 1996 SBA Dominican Rep 2004 SBA In place since 2003 Ecuador 1985 SBA Eq Guinea 1985 SBA Ethiopia 1993 SAF In place since 1992 Gambia 2001 PRGF Georgia 1999 PRGF In place since 1995 Ghana 1984 SBA Guinea 1996 PRGF Haiti 1983 SBA Jamaica 1985 SBA Kazakhstan 1995 SBA Kenya 1980 SBA Madagascar 1983 SBA Malawi 1989 PRGF Mali 1989 SAF+SBA Mauritania 1994 PRGF Mexico 1985 EFF In place since 1983 Moldova 2001 PRGF Mozambique 1991 PRGF Niger 1996 PRGF Nigeria 1987 SBA Peru 1993 EFF Philippines 1986 SBA Rwanda 1980 SBA Sierra Leone 1994 PRGF Tanzania 1981 SBA Togo 1988 SAF+SBA Uganda 1982 SBA Venezuela 1989 EFF Zambia 1981 EFF Source: Authors analysis and IMF programs database constructed by Dreher, Axel, IMF and Economic Growth: The Effects of Programs, Loans, and Compliance with Conditionality, World Development 34, 5: 769-788 (2006). IMF Extended Fund Facility (EFF) IMF Structural Adjustment Facility (SAF) IMF Stand-By Arrangement (SBA) IMF Poverty Reduction and Growth Facility (PRGF) Page | 22 Tables and Figure Tables Table 1: Descriptive statistics Values Log Obs Mean Std. Dev. Min Max Mean Std. Dev. Min Max Exports (US$ th.) 538'362 175539.50 2271267.00 0.00 303000000 7.84 3.64 -4.61 19.53 Distance 670'566 7438.38 4382.42 39.43 19904.45 8.67 0.80 3.67 9.90 Importer's GDP 618'121 1.67E+11 7.29E+11 2.06E+07 1.24E+13 2.32E+01 2.38E+00 1.68E+01 3.02E+01 Exporter's GDP 615'315 1.75E+11 7.45E+11 4.08E+07 1.24E+13 2.33E+01 2.34E+00 1.75E+01 3.02E+01 Importer's GDP/ cap 587'844 7272.45 7989.35 336.20 60228.41 8.31 1.13 5.82 11.01 Exporter's GDP/ cap 587'268 7407.10 8108.80 336.20 60228.41 8.31 1.13 5.82 11.01 Private inspection 670'566 0.16 0.36 0.00 1.00 0.16 0.36 0.00 1.00 Common language 670'566 0.16 0.36 0.00 1.00 0.16 0.36 0.00 1.00 Colony 670'566 0.01 0.12 0.00 1.00 0.01 0.12 0.00 1.00 Common border 670'566 0.02 0.14 0.00 1.00 0.02 0.14 0.00 1.00 Control of Corruption 250'399 50.95 29.04 0.00 100.00 50.95 29.04 0.00 100.00 Voice & Accountability 254'308 50.06 28.78 1.00 100.00 50.06 28.78 1.00 100.00 Political Stability 251'790 48.15 28.62 0.00 100.00 48.15 28.62 0.00 100.00 Governance Effectiveness 253'758 51.91 28.84 0.00 100.00 51.91 28.84 0.00 100.00 Regulatory Quality 253'775 51.72 28.31 0.00 100.00 51.72 28.31 0.00 100.00 Rule of Law 252'563 50.03 28.84 0.00 100.00 50.03 28.84 0.00 100.00 IMF SAF 537'000 0.02 0.15 0.00 1.00 0.02 0.15 0.00 1.00 IMF PRGF 537'000 0.12 0.32 0.00 1.00 0.12 0.32 0.00 1.00 IMF EFF 537'000 0.05 0.22 0.00 1.00 0.05 0.22 0.00 1.00 IMF SBA 537'000 0.15 0.35 0.00 1.00 0.15 0.35 0.00 1.00 Aid/cap (US$) 565'500 45.87 71.80 -203.59 1294.42 3.09 1.53 -6.12 7.17 Notes : For dummy variables, the mean is the proportion of observations for which the variable is equal to one, i.e. the variable's sample incidence. Page | 23 Table 2 Baseline estimation results OLS Treatreg (a) (b) (c) (d) 2nd stage 1st stage 2nd stage 1st stage 2nd stage 1st stage Constant -10.83*** 0.444 4.865*** 4.906*** (0.54) (1.14) (1.610) (1.61) Gravity variables Distance -1.567*** -1.648*** -1.701*** -1.693*** (0.0058) (0.0082) (0.0102) (0.010) Importer's GDP 0.565*** 0.618*** 0.667*** 0.669*** (0.014) (0.035) (0.0404) (0.040) Exporter's' GDP 0.655*** 0.281*** 0.105** 0.108** (0.016) (0.036) (0.0461) (0.046) Colony 1.190*** 1.130*** 0.931*** 0.934*** (0.021) (0.031) (0.0393) (0.039) Comlang 0.682*** 0.777*** 0.865*** 0.861*** (0.013) (0.018) (0.0209) (0.021) Contingency 0.500*** 0.623*** 0.795*** 0.812*** (0.025) (0.035) (0.0394) (0.039) Treatment PSI 0.0811*** 0.0894* 0.0976* 0.119** (0.017) (0.049) (0.0591) (0.058) Instruments 1 Control of Corruption 0.0123*** 0.000745 0.0320*** (0.00079) (0.000852) (0.0012) Control of Corruption (square) -0.000576*** -0.000363*** -0.000543* (0.000010) (1.06e-05) (0.000015) Voice & Accountability 0.0751*** (0.0012) Voice & Accountability (square) -0.000809 (0.000015) Political Stability 0.00675*** (0.00092) Political Stability (square) -0.000272* (0.000013) Governance Effectiveness -0.0650*** (0.0015) Governance Effectiveness (square) 0.000496* (0.000019) Regulatory Quality -0.0210*** (0.0014) Regulatory Quality (square) 0.000329* (0.000016) Instruments 2 Aid a\ -0.182*** -0.240*** (0.00347) (0.0043) IMF PRGF c\ 0.956*** 1.024*** (0.0118) (0.013) IMF EFF d\ -0.249*** (0.021) IMF SBA e\ -0.295*** (0.017) Observations 306'223 156'856 156'856 107,005 107,005 106'264 106'264 R-squared 0.71 Sargan's overidentification test (chi2/prob) 0.0679 0.0788 0 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Page | 24 Table 3 Robustness checks 1 ­ Mirror data OLS Treatreg (a) (b) (c) 2nd stage 1st stage 2nd stage 1st stage Constant -7.666 -4.415*** -0.239 (0.55) (1.199) (1.531) Gravity variables Distance -1.511*** -1.565*** -1.598*** (0.00561) (0.00800) (0.00981) Importer's GDP 0.504*** 0.622*** 0.704*** (0.0147) (0.0352) (0.0415) Exporter's' GDP 0.648*** 0.259*** 0.0804* (0.0155) (0.0354) (0.0458) Colony 1.155*** 1.081*** 1.008*** (0.0208) (0.0312) (0.0431) Comlang 0.717*** 0.834*** 0.941*** (0.0123) (0.0177) (0.0211) Contingency 0.456*** 0.621*** 0.819*** (0.0249) (0.0358) (0.0404) Treatment PSI 0.121*** 0.109*** 0.0919* (0.0167) (0.0360) (0.0489) Instruments 1 Control of Corruption 0.0139*** 0.0218*** (0.000784) (0.00137) Control of Corruption (square) -0.000585*** -0.000284* (1.00e-05) (1.82e-05) Voice & Accountability 0.0707*** (0.00120) Voice & Accountability (square) -0.000713* (1.46e-05) Political Stability 0.00554*** (0.000982) Political Stability (square) -0.000231* (1.31e-05) Governance Effectiveness -0.0696*** (0.00157) Governance Effectiveness (square) 0.000565** (1.93e-05) Regulatory Quality -0.0171*** (0.00145) Regulatory Quality (square) 0.000304* (1.67e-05) Instruments 2 Aid a\ -0.240*** (0.00435) IMF PRGF c\ 1.040*** (0.0132) IMF EFF d\ -0.329*** (0.0203) IMF SBA e\ -0.343*** (0.0164) Observations 322,813 165,226 165,226 110,897 110,897 R-squared 0.715 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Page | 25 Table 4 Robustness checks 2 Treatreg BDM 5 years after 5 years before GMM GMM (a) (b) (c) (d) (e) 1st stage 2nd stage 1st stage 2nd stage 1st stage 2nd stage Constant -10.70*** 0.193 0.185 -34.29*** -33.93*** (0.54) (1.14) (1.14) (0.24) (0.23) Gravity variables Distance -1.567*** -1.648*** -1.648*** -1.384*** -1.364*** (0.0058) (0.0082) (0.0082) (0.0038) (0.0038) Importer's GDP 0.557*** 0.633*** 0.633*** 0.851*** 0.813*** (0.014) (0.035) (0.035) (0.0061) (0.0061) Exporter's' GDP 0.655*** 0.280*** 0.280*** 1.376*** 1.375*** (0.016) (0.036) (0.036) (0.0065) (0.0067) Colony 1.190*** 1.130*** 1.130*** 0.862*** 0.839*** (0.021) (0.031) (0.031) (0.022) (0.020) Common language 0.682*** 0.778*** 0.778*** 0.943*** 0.923*** (0.013) (0.018) (0.018) (0.010) (0.0099) Common border 0.500*** 0.623*** 0.623*** 0.811*** 0.936*** (0.025) (0.035) (0.035) (0.015) (0.016) Treatment PSI 0.113*** 0.191*** 0.232*** (0.025) (0.018) (0.018) PSI +5years 0.00685 (0.049) PSI -5years -0.0533 (0.052) Instruments Control of Corruption 0.0129*** 0.00666*** 0.00663*** (0.00081) (0.00083) (0.00040) Control of Corruption (square) -0.000572*** -0.000410*** 0.0000200 (0.000010) (0.000010) (0.000004 Observations 306'223 12'079 156'856 156'856 156'856 156'856 159'226 156'856 R-squared AR2 (Pr > z) 0.71 0 0 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Page | 26 Table 5 Robustness checks 3 With Trade Liberalisation With Remoteness OLS Treatreg BDM OLS Treatreg (a) (b) (c) (d) (e) (f) 2nd stage 1st stage 2nd stage 1st stage 2nd stage 1st stage 2nd stage 1st stage Constant -4.666*** -1.206 -4.633*** -30.08*** 3.368 3.339 -0.93 -1.29 -0.93 (2.941) (6.185) (6.186) Gravity variables -1.567*** -1.648*** -1.648*** Distance -1.534*** -1.572*** -1.534*** (0.00584) (0.00820) (0.00820) -0.0076 -0.0091 -0.0076 0.559*** 0.589*** 0.590*** Importer's GDP 0.559*** 0.613*** 0.562*** (0.0149) (0.0367) (0.0366) -0.026 -0.039 -0.026 0.627*** 0.316*** 0.316*** Exporter's' GDP 0.473*** 0.289*** 0.473*** (0.0170) (0.0386) (0.0386) -0.027 -0.039 -0.027 1.189*** 1.130*** 1.130*** Colony 1.171*** 1.159*** 1.171*** (0.0207) (0.0307) (0.0307) -0.027 -0.033 -0.027 0.681*** 0.778*** 0.778*** Common language 0.767*** 0.776*** 0.767*** (0.0126) (0.0178) (0.0178) -0.016 -0.02 -0.016 0.500*** 0.622*** 0.622*** Common border 0.696*** 0.710*** 0.697*** (0.0248) (0.0352) (0.0352) -0.031 -0.038 -0.031 Trade openess (Wacziarg-Welsh) 1.466*** 1.842*** 1.472*** -0.16 -0.28 -0.16 Remoteness exporter 0.772*** -0.708*** -0.708*** (0.0950) (0.187) (0.187) Remoteness importer 0.198* 0.604*** 0.604*** (0.104) (0.210) (0.210) Treatment PSI 0.110*** 0.0855* 0.129*** 0.0804*** 0.0898* 0.112** -0.023 -0.046 -0.027 (0.0167) (0.0494) (0.0517) Instruments Control of Corruption 0.000624 0.0123*** -0.000621 -0.00092 (0.000791) (0.000846) Control of Corruption (square) -0.000325*** -0.000576*** -0.000368* -0.000012 (1.01e-05) (1.06e-05) Voice & Accountability 0.0269*** 0.0436*** -0.0009 (0.000797) Voice & Accountability (square) -0.000416*** -0.000598* -0.000011 (9.94e-06) Observations 188555 131739 131739 10146 188555 306,223 156,856 156,856 156,856 156,856 R-squared 0.72 0.72 0.705 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Page | 27 Table 6 Extension results 1 Poisson 2000- Poisson 1995- Poisson 1990- Poisson 1985- Poisson 1980- 2004 1999 1994 1990 1984 TOBIT (a) (b) (c (d) (e) (f) Constant -36.01 -42.55 -48.27 -43.87 -9.951*** -2.152*** (2560) (1625) (6730) (3326) (0.28) (0.63) Gravity variables Distance -0.772*** -0.740*** -0.725*** -0.715*** -0.726*** -1.560*** (0.00027) (0.00031) (0.00039) (0.00050) (0.00057) (0.0073) Importer's GDP 0.605*** 0.708*** 0.582*** 0.517*** 0.511*** 0.880*** (0.0023) (0.0023) (0.0030) (0.0031) (0.0037) (0.026) Exporter's' GDP 0.469*** 0.328*** 0.407*** 0.426*** 0.516*** 0.337*** (0.0022) (0.0022) (0.0030) (0.0031) (0.0037) (0.026) Colony 0.0452*** 0.0220*** -0.0174*** 0.0233*** 0.309*** 1.058*** (0.00079) (0.00091) (0.0011) (0.0015) (0.0017) (0.033) Common language 0.154*** 0.184*** 0.343*** 0.359*** 0.182*** 0.744*** (0.00070) (0.00085) (0.0011) (0.0014) (0.0015) (0.016) Common border 0.556*** 0.614*** 0.500*** 0.380*** 0.325*** 0.574*** (0.00072) (0.00085) (0.0011) (0.0014) (0.0016) (0.029) Treatment PSI 0.0145*** 0.0271*** 0.0611*** 0.0564*** -0.217*** 0.0547** (0.0025) (0.0032) (0.0064) (0.0069) (0.019) (0.026) Observations 123'569 123'254 100'502 82'150 70'892 223'163 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Notes: Marginal effects for Tobit Page | 28 Table 7 Extension results 2 OLS Treatreg (a) (b) (c) (d) Second First stage Second First stage Second First stage Constant -8.228*** -3.714*** 0.0776 -9.486*** (0.509) (1.366) (1.485) (0.645) Gravity variables Distance -1.567*** -1.648*** -1.693*** -1.640*** (0.00583) (0.00820) (0.0102) (0.00733) Importer's GDP 0.573*** 0.635*** 0.683*** 0.601*** (0.0147) (0.0350) (0.0408) (0.0168) Exporter's' GDP 0.655*** 0.280*** 0.107** 0.505*** (0.0164) (0.0365) (0.0461) (0.0211) Colony 1.190*** 1.130*** 0.934*** 1.040*** (0.0207) (0.0307) (0.0394) (0.0268) Comlang 0.682*** 0.778*** 0.861*** 0.716*** (0.0126) (0.0178) (0.0210) (0.0146) Contingency 0.500*** 0.623*** 0.813*** 0.660*** (0.0248) (0.0352) (0.0394) (0.0279) Treatment PSI persistent - dummy equals once PSI adopted 0.0938*** 0.0626 0.0290 0.182*** (0.0190) (0.0640) (0.0668) (0.0351) Instruments 1 Control of Corruption 0.0105*** -0.0144*** (0.000808) (0.00147) Control of Corruption (square) -0.000646*** 0.000172*** (1.06e-05) (1.88e-05) Voice & Accountability 0.0864*** (0.00124) Voice & Accountability (square) -0.000849*** (1.54e-05) Political Stability -0.0154*** (0.000988) Political Stability (square) 0.000117*** (1.22e-05) Governance Effectiveness -0.0683*** (0.00167) Governance Effectiveness (square) 0.000711*** (2.05e-05) Regulatory Quality -0.00769*** (0.00143) Regulatory Quality (square) 8.59e-05*** (1.67e-05) Instruments 2 Aid a\ -0.298*** -0.108*** (0.00446) (0.00191) IMF PRGF c\ 0.977*** 1.356*** (0.0139) (0.00841) IMF EFF d\ -0.113*** 0.346*** (0.0211) (0.0120) IMF SBA e\ -0.392*** 0.157*** (0.0165) (0.00838) Observations 306,223 156,856 156,856 106,264 106,264 212,357 212,357 R-squared 0.705 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Page | 29 Figure Figure 1 Percent of importing countries and trade flows covered by private inspection, 1980-2006 25.00 PSI-covered trade 20.00 (%) PSI-using countries (%) 15.00 10.00 5.00 0.00 80 00 88 84 86 04 06 82 02 90 98 94 96 92 19 19 19 19 19 19 19 19 19 19 20 20 20 20 Source: Authors' calculations, DOTS, Industry. Page | 30