2. Literature overview

There is a vast literature studying the effects of the WTO GPA. In this section, we restrict ourselves to a discussion of the more recent contributions by Shingal (Reference Shingal2011, Reference Shingal2015), Rickard and Kono (Reference Richard and Kono2014), Fronk (Reference Fronk2015), Kutlina-Dimitrova and Lakatos (Reference Kutlina-Dimitrova and Lakatos2016), and Gourden and Messent (Reference Gourdon and Messent2017).

Shingal (Reference Shingal2011) examines whether the WTO GPA has fulfilled its intended purpose of non-discrimination. Noting that governments frequently discriminate against foreign suppliers in favour of domestic ones, the author studies whether the WTO GPA has led to greater market access for foreign suppliers. His analysis reveals that the WTO GPA has not been effective in increasing, or even sustaining, market access for foreigners in the services procurement markets of Japan and Switzerland.Footnote ¹

Rickard and Kono (Reference Richard and Kono2014) – noting that home-biased government procurement is a pervasive phenomenon, and that measuring domestic bias is difficult since it is not directly observable – measure discrimination in government procurement by estimating the impact of procurement spending on the imports of goods and services. According to the authors, an increase in procurement shifts spending from the private to the public sector. Empirical analysis indicates that the WTO GPA does not eliminate discrimination in public procurement.Footnote ²

A different approach was adopted by Fronk (Reference Fronk2015) who is concerned with the estimation of the impact of procurement disciplines in bilateral and multilateral agreements, which he calls national treatment agreements (NTA). He employs a theoretical model incorporating elements from political economy, international trade, and auction theory to derive his estimating equations. Then, he empirically analyses the effects of NTAs on procurement awards. Using a probit selection model, Fronk (Reference Fronk2015) finds that signing national treatment agreements with the US increases countries' procurement awards, both in terms of contracts and in total value.Footnote ³

Shingal (Reference Shingal2015), using the dataset from the WTO's notifications on domestic and foreign purchases by Japanese and Swiss governments at the sector level over the period 1990–2003, undertakes econometric analysis of the determinants of procurement with a home bias using variables inspired by the political economy, trade-macroeconomic, and procurement literatures. The empirical results reveal that the coefficient on the WTO GPA is statistically insignificant, indicating that the disciplining mechanisms of the WTO GPA may not have led to greater foreign access to the governments' goods procurement market.Footnote ⁴

Recently, using a multivariate logit model, Kutlina-Dimitrova and Lakatos (Reference Kutlina-Dimitrova and Lakatos2016) analysed the impact of a set of selected explanatory variables on the probability of awarding public procurement contracts directly cross-border. The explanatory variables include contract value, number of offers, GDP per capita, trade–GDP ratio, the scope of public enterprises, regulatory protection of incumbents, barriers to foreign direct investment (FDI), and the type of tender procedure.Footnote ⁵ Empirical results reveal that the value of a contract has a positive influence on the probability of a cross-border win, implying that high-value contracts are more likely to be awarded to a foreign company. In addition, the results point to the negative influence of the number of offers on the probability of a cross-border win.

Finally, Gourdon and Messent (Reference Gourdon and Messent2017), making use of Tenders Electronic Daily TED data for the period 2009–2014, study the impact of a set of selected explanatory variables on the probability of awarding public procurement contracts directly cross-border. The explanatory variables include gross domestic product (GDP), imports, distance, government procurement share in GDP, common border, common language, colony, and remoteness. In addition, to assess the efficiency of such agreements, the authors include dummies whenever the reporting and partner countries are both in the WTO GPA or have signed an agreement with procurement disciplines. The authors show that a country's membership of the WTO GPA increases the probability of their firms being awarded a procurement contract in the EU through the cross-border mode of supply. Thus, under WTO GPA membership, foreign suppliers are expected to win a greater share of available contracts. However the authors note that the procurement pie is not expanding. Noting that procurement will be discriminatory if foreign firms cannot contest the procurement market through foreign direct investment or if government entities differentiate among firms based on their nationality, the authors investigate the role of investment agreements. Using Rickard and Kono's basic (Reference Richard and Kono2014) framework, the authors show that FDI barriers reduce the effectiveness of the WTO GPA in increasing the chance of a cross-border award, and that to maximize gains from WTO GPA accession, a country should also undertake investment liberalization.Footnote ⁶

To emphasize the added value of the present paper compared to the existing literature, we note that the paper uses TED micro-level data released recently by the EU, covering more than three million tenders conducted in the European Economic Area (EEA), Switzerland, and Macedonia for the period 2006–2016.Footnote ⁷ A very interesting feature of this data set is the variable B_GPA, which records whether the contract is covered by the WTO GPA. Using this data set, we examine the following research questions:

• • Does the WTO GPA foster openness of government procurement markets by increasing the probability that a foreign firm will win a contract?

• • Does the WTO GPA promote competition by increasing the number of offers submitted?

• • Does the WTO GPA lower corruption risk in government procurement?

• • Does the WTO GPA improve procurement cost-effectiveness by lowering the probability that the procurement price is higher than the estimated cost of the procuring authority?

The TED data set contains detailed information on public procurement contracts that allow us to examine the above research questions. Our empirical analysis provides the following results: First, using a multivariate logit model, we analyse the probability of a foreign firm winning government procurement contracts and show that the WTO GPA reduces barriers for foreign countries. Second, we find that the WTO GPA promotes competition. Third, we find that corruption risks, measured in two different ways, are lower in WTO GPA auctions: the first measure is the probability of a single-bid auction, and the second the number of recurrent wins by a single firm. Finally, we show that the WTO GPA improves procurement cost-effectiveness by lowering the probability of the procurement price being larger than the estimated cost.

3. Main features of the TED data set

The TED data set contains data on 3,562,829 government procurement contracts conducted in 33 countries for the period 2006–2016. As stated in TED (2016), the sources of the data are contracting authorities and entities across Europe. The data are extracted from the contract notice and contract award notice standard forms filled in by the authorities.Footnote ⁸ Public authorities are obliged to publish their tender invitations on TED for all contracts exceeding EU public procurement thresholds. However, as emphasized by Kulina-Dimitrova and Lakatos (Reference Kutlina-Dimitrova and Lakatos2016), contract awards below the threshold are also reported on TED. Authorities are not prevented from announcing a tender on TED even if the tender's value is below the threshold, and since tenders are often awarded to the most economically advantageous bidder, the final award value might be well below the threshold.

To conduct the empirical analysis, we employ the TED award notices data. Of these contracts, 1,936,456 are covered by the WTO GPA and 1,626,373 are non-WTO GPA contracts, indicating that 46% of the procurements in the TED data set are non-WTO GPA contracts. Note that the coverage schedules are a critical part of the WTO GPA.Footnote ⁹ The EU ‘SIMAP’Footnote ¹⁰ form for public procurement contains question IV.1.8, which asks whether the procurement is covered by the WTO GPA. We use this information to examine the effects of WTO GPA on government procurement. Table 1 gives the distribution of the contracts across countries, and we present average contract values for each country in the Online Appendix.Footnote ¹¹ We note that firms from 201 different countries have won at least one contract.Footnote ^12, Footnote ¹³ Also, while Armenia, Canada, EU member countries, Hong Kong, Israel, the Republic of Korea, New Zealand, Chinese Taipei, and the US are parties to the WTO GPA, some of the important trade partners of the EU, such as Australia, Egypt, India, Malaysia, Mexico, Pakistan, the Russian Federation, Thailand, and Turkey, are not.Footnote ¹⁴

Table 1. Authority Country conducting the procurement

Each contract in the data set is identified by a unique contract ID number. The year of the contract, contracting authority name, contracting authority country, Common Procurement Vocabulary (CPV) sector code, winner firm name. and winner firm country are available for each contract. We use this information to examine the likelihood of a foreign firm winning a contract. We identify the sector of procurement using the first two digits of the CPV code, noting that there are 72 major sectors.Footnote ¹⁵

Government institutions implement different procurement procedures. The most common procedure is ‘open tender’ with 1,405,288 WTO GPA covered and 1,234,645 non-WTO GPA contracts awarded using the open procedure. Additionally, authorities implement negotiation, restricted auction, and competitive dialogue procedures.Footnote ¹⁶ The TED data set also provides information about procurement results, namely procurement price, estimated cost determined by the procuring authority, and number of offers received.

We identify firms by their names and base countries. One of the unique features of the TED data set is the availability of firm names and the countries in which they are located. We use this information to calculate the total number of wins by a single firm. There are 414,917 unique firms that have won at least one contract. Ninety percent of the firms have won ten or fewer contracts.Footnote ^17, Footnote ¹⁸ When we consider only the WTO GPA covered procurements, we note that 258,647 unique firms have won on average 7.48 contracts. In comparison, when we consider non-WTO GPA contracts the average number of wins is 8.38 by 193,981 firms.

4. Effect of the WTO GPA on the openness of government procurement markets

To analyse the effects of the WTO GPA on the openness of government procurement markets, we consider three cases. First, we call a firm foreign whenever the country of the procuring authority is different from the country of the firm, and we examine whether the WTO GPA reduces barriers for foreign firms. Next, we consider a different definition of foreign firm, and define a firm as foreign whenever the procuring authority is in an EU Member state and the country of the firm is a non-EU state. We then examine whether the WTO GPA promotes non-EU firms.Footnote ¹⁹ Finally, the third case investigates whether the WTO GPA encourages competitive tendering procedures.

4.1 Effect of WTO GPA on the probability that a foreign firm will win a contract

Using the following logit regression specification, we examine whether the WTO GPA reduces barriers for foreign firms which are attempting to win government procurement contracts:

(1)$${\rm Prob(}C_{irt} = 1{\rm \vert x}) = F\lpar {x_{irt}^, \beta} \rpar $$

where C _irt is a dummy variable, that is 1 if a procurement contract is awarded to a firm when the country of the procuring authority is different from the country of the firm, and 0 otherwise. $F\left( {x_{irt}^, \beta } \right)$ is the logit probability function of $x_{irt}^, \beta $ and $x_{irt}^, $ contains the explanatory variables GPA _irt and control variables, X’. GPA _irt is the dummy variable, which is 1 if the procurement is covered by a WTO GPA and 0 otherwise. The coefficient of GPA _irt indicates the impact of the WTO GPA on the probability that a foreign firm wins a government procurement contract in EU countries.Footnote ²⁰ X′ contains dummy variables for the procurement method (type), and dummy variables for the type of contracting authority. Additionally, X′ includes country-specific factors, such as the ratio of trade-to-GDP and GDP per capita.Footnote ²¹ The fixed-effect of the dummy variables for the years 2007–2016 and sector dummy variables, identified by the first two digits of the CPV codes, are also given. Note that foreign firms have been awarded 61,889 contracts: 37,516 covered by the WTO GPA and 24,373 non-GPA contracts.Footnote ²²

Using a multivariate logit regression and instrumental variable (IV) GMM linear probability model estimation enables us to examine the effect of WTO GPA on the probability that a foreign firm wins a contract (see Table 2). As presented in Table 1, 44% of WTO GPA and 54.79% of non-WTO GPA contracts are conducted in France and Poland. The over-representation of these countries might alter the results. Accordingly, to assess the robustness of the results, we conduct the empirical analysis using the complete data set but excluding France and Poland. In the second column, we focus on the contracts with estimated contract values above the thresholds determined by the EU, namely 1,413,379 contracts. Specifically, we eliminate contracts where the estimated contract values are not known or the contract values are below EU threshold levels.

Table 2. Effect of WTO GPA on the probability of a foreign firm winning a contract

Notes: * p < 0.05; ** p < 0.01

The third column considers that the WTO GPA covered variable might be endogenous. There might be unobserved factors that authorities use to determine whether tenders are covered by the WTO GPA. These unobserved factors might also be related with the probability that a foreign (or non-EU firm) wins a contract. In that case, the WTO GPA covered variable will be correlated with the error term and this endogeneity problem will affect the results. We employ an instrumental variable generalized method of moments (GMM) methodology to consider possible endogeneity of the WTO GPA covered variable. Lewbel (Reference Lewbel2018) shows that a linear probability model can be estimated using the heteroscedasticity based instrumental variables (IV) of Lewbel (Reference Lewbel2012) when the dependent variable is binary and an explanatory variable is potentially endogenous. Accordingly, we correct for possible endogeneity of the WTO GPA covered variable using the IV GMM methodology of Lewbel (Reference Lewbel2012):Footnote ²³

(2)$$C_{irt} = \beta _0 + \beta _1GPA_{irt} + {X}^{\prime}\theta + \varepsilon _{irt}$$

where C _irt is a dummy variable as described in equation (1). In equation (2), β ₁ measures the effect of WTO GPA on the probability that a foreign firm wins a contract. X′ contains the control variables as described above.

Columns 4–6 in Table 2 provide the results for the 3,524,060 contracts for all countries. Column 5 focuses on the estimated contract values above the EU thresholds. Finally, column 6 implements the IV GMM methodology to all contracts.Footnote ²⁴

The coefficient of the WTO GPA-covered variable is significant, with a positive sign for all regression specifications. Accordingly, we conclude that foreign firms are more likely to win government procurement contracts when the contract is covered by the WTO GPA. This result indicates that the WTO GPA is successful in lowering the barriers for foreign firms to win government procurement contracts in EU Member states and affiliated countries.

4.2 Effect of the WTO GPA on the probability that a non-EU firm will win a contract

We now consider a different definition of ‘foreign firm’ and examine whether the WTO GPA promotes non-EU firms. We note that during the years 2006–2016, 13,591 non-EU firms won 29,045 government procurement contracts in the EU. We conduct a multivariate logit regression and an IV-GMM linear probability model estimation to examine the effect of the WTO GPA on the probability that a non-EU firm wins a contract. We estimate the following logit regression specification and linear probability model:

(3)$${\rm Prob(}NEU_{irt} = 1{\rm \vert x}) = F\lpar {x_{irt}^, \beta} \rpar $$

(4)$$NEU_{irt} = \beta _0 + \beta _1GPA_{irt} + {X}^{\prime}\theta + \varepsilon _{irt}$$

where NEU _irt is a dummy variable that is 1 if procurement is awarded to a non-EU foreign firm or 0 otherwise. $F\left( {x_{irt}^, \beta } \right)$ is the logit probability function of $x_{irt}^, \beta $. $x_{irt}^, $ and X′ contains the explanatory variables described in section 4.2. The coefficient of GPA _irt shows the impact of the WTO GPA on the probability that a non-EU firm wins a government procurement contract in the EU Member state and affiliated countries.Footnote ²⁵

Table 3 displays the results of the multivariate logit regression estimation of equation (3) and IV-GMM linear probability model estimation using Lewbel (Reference Lewbel2018) methodology as in equation (4). As in Table 2, the second and fifth columns focus on estimated contract values above the EU thresholds. The coefficient of the WTO GPA covered variable is significant with a positive sign for all regression specifications. Accordingly, we conclude that non-EU firms are more likely to win government procurement contracts when the contract is covered by the WTO GPA. The WTO GPA opens the EU government procurement market to non-EU firms.

Table 3. Effect of WTO GPA on probability of a non-EU firm winning a contract

Notes: * p < 0.05; ** p < 0.01

5. Effect of the WTO GPA on competitive environment

In this section, we examine whether the WTO GPA improves the competitive environment by increasing the number of offers submitted for a contract. We implement a negative-binomial regression methodology, as suggested by Bajari and Hortacsu (Reference Bajari and Hortacsu2003), to assess the determinants of the number of bidders. Specifically, we examine the following regression equation

(5)$$N_{irt} = \alpha + \rho GPA_{irt} + \theta C_{rt} + \beta X_{irt} + \delta FE + \varepsilon _{irt}$$

where N _irt is the number of bids submitted for each contract.Footnote ²⁷ We exclude France and Poland and examine contracts from the remaining EU countries. We present the empirical results for this sample in Table 4 and for all countries in Table OA.7 of the Online Appendix.Footnote ²⁸ Table 4 (second column, consisting of 1,283,658 contracts) studies tenders with estimated contract values above EU thresholds. The coefficient of the WTO GPA-covered variable is significant, with a positive sign for all regression specifications. This result indicates that significantly more firms submit offers to WTO GPA-covered procurements. Hence, we conclude that the WTO GPA improves the level of competition in government procurement auctions.

Table 4. Effect of WTO GPA on level of competition and corruption measures excluding France and Poland

Notes: * p < 0.05; ** p < 0.01

6. Corruption risk in government procurement and the WTO GPA

Cost-effective government procurement requires a competitive and transparent procurement system. Corruption limits competition and artificially increases procurement prices above a competitive level. Therefore, to be able to improve procurement cost-effectiveness, authorities should conduct appropriate competition policy actions to deter collusion in public procurement. Hence, collusion prevention is one of the goals of the WTO GPA. In this section, we investigate whether the WTO GPA manages to limit collusion.

We first construct collusion measures to gauge collusive risk in government procurement in the EU. Fazekas et al. (Reference Fazekas, Chvalkovska, Skuhrovec, Tóth, King and Mungiu-Pippidi2014) calculate a proxy indicator of corruption by using signs of limited competition, such as a single bid received or the same company winning recurrent contracts. Ishii (Reference Ishii2009) provides details about the operational structure of bidding rings that manipulate public procurement auctions. Bidding rings determine the winning firm by considering the previous intentional losses of a bid member. Accordingly, ring members that do not win a contract for a certain period of time are more likely to win the next rigged auction. Ishii (Reference Ishii2009) argues that bidding ring members tend to win a higher number of contracts compared to non-ring members. We employ the arguments stated by Ishii (Reference Ishii2009) and use the total number of contracts by a firm as a potential indicator of collusion. We examine the impact of the WTO GPA on corruption by assessing how the WTO GPA affects two red flags of limited competition: contracts with a single bid and firm win ratios.

6.1 Contracts with single bids

Of all procurements in the time period we study, 762,813 (21%) contracts were conducted when there was only one offer. Of this total, 21%, namely 285,905 contracts, were WTO GPA covered and 476,908 were not. We estimate a multivariate logit regression:

(6)$${\rm Prob(}SB_{irt} = 1{\rm \vert x}) = F\lpar {x_{irt}^, \beta} \rpar $$

where SB _irt is the single-offer procurement dummy variable and equals 1 if procurement was conducted with only one offer or 0 otherwise. $x_{irt}^, $ contains the explanatory variables GPA _irt, C _rt, P _irt, and FE, as described above in Section 4.

The last two columns of Table 4 display the regression results using all contracts and using contracts with estimated values above EU thresholds. The table shows that the probability of a one-offer procurement is significantly lower for WTO GPA-covered procurements. The coefficient of the WTO GPA-covered dummy variable is negative and significant at the 1% significance level. In other words, it is more likely that WTO GPA-covered procurements attract more than one offer. Accordingly, we conclude that the WTO GPA helps governments promote a competitive procurement environment that is less susceptible to collusive behavior. Table OA.8 in the Online Appendix estimates equation (6) using the complete data set with all EU countries and employs additional control variables. The WTO GPA covered variable is significant with a negative coefficient. The magnitude of the coefficient is lower (–0.16 compared to –0.27 in Table 4) when we examine all EU countries.

7. Does the WTO GPA improve procurement cost-effectiveness?

Finally, we examine the impact of the WTO GPA on the cost-effectiveness of government procurement. We measure cost-effectiveness by assessing whether the procurement price is lower than the authority's estimated cost.Footnote ²⁹ We can use only 1,116,249 observations since the estimated value is missing for a substantial number of contracts. Additionally, we have eliminated the outliers by implementing Billor et al.’s (Reference Billor, Hadi and Velleman2000) BACON methodology (blocked adaptive computationally efficient outlier nominators) to identify the outliers. Although both the contract price and the estimated cost should have been entered in Euros, some observations have been entered in local currencies. Additionally, the contract price may be for one unit; however, the estimated cost may represent the total amount, which causes the ratio to be unrealistically small.Footnote ³⁰ We identify these entries as outliers and eliminate them.Footnote ³¹ We display the ratio's summary statistics in the Online Appendix Table OA.11. On average, the ratio is equal to 0.89, indicating that the contract price is 89% of the estimated cost.Footnote ³²

We follow the description of OECD (2012) to identify cost-ineffective procurements. As stated by OECD (2012) ‘value for money’ can be assessed by comparing the procurement price and estimated costs. Specifically, procurement prices that are higher than the engineering cost estimates are not cost-effective. OECD (2012) suggests that public authorities should investigate these procurements. We identify the tenders with procurement prices higher than estimated costs (ratio of price and estimate is larger than one). We determine 225,837 (17.45%) procurements that are not cost-effective since their procurement prices are higher than their estimated costs.

We conduct several regression analyses to investigate whether WTO GPA lowers the probability that a tender is cost-ineffective (the ratio of price and estimate is larger than one). We also examine the impact of collusive behaviour on procurement cost-effectiveness by analysing the effects of single-offer procurements. Specifically, we estimate the following regression specification:

(7)$${\rm Prob(}cost - ineffective_{irt} = 1{\rm \vert x}) = F\lpar {x_{irt}^, \beta} \rpar $$

where cost–ineffective _irt is a dummy variable that is 1 if the procurement price is larger than the estimated cost and 0 otherwise. $F\left( {x_{irt}^, \beta } \right)$ is a logit probability function of $x_{irt}^, \beta $. $x_{irt}^, $ and X′ contains the explanatory variables described in Section 4.

We consider the endogeneity of the number of bidders and of the GPA dummy variable when conducting the regression analysis.Footnote ³³ As stated by Estache and Iimi (Reference Estache and Iimi2010) and Onur et al. (Reference Onur, Ozcan and Tas2012), there exist factors that might simultaneously influence bidders’ participation decisions and the winning bid might cause the OLS estimates to be inconsistent. Ohashi (Reference Ohashi2009) argues that unobserved attributes of the procurement process are represented in the error term, and that bidders’ participation decisions are likely to be correlated with these unobserved procurement attributes. Accordingly, the potential correlation between the error term and number of bidders might cause an endogeneity problem. Unobserved attributes that affect the number of bidders and the WTO GPA dummy variables are represented in the error term. If the empirical model of equation (7) does not incorporate these attributes then both the number of bidders and the WTO GPA dummy variables might be endogenous. We consider potential endogeneity of the number of bidders and the WTO GPA dummy variables and estimate the IV-GMM linear probability model estimation using Lewbel's (Reference Lewbel2018) methodology as in equation (8).

(8)$$cost-ineffective_{irt} = \beta _0 + \beta _1GPA_{irt} + {X}^{\prime}\theta + \varepsilon _{irt}$$

Table 5 displays the results of the multivariate logit regression estimation of equation (7) and IV-GMM linear probability model estimation of equation (8). Columns 1–4 in Table 5 give the estimation results excluding France and Poland. Columns 5–8 analyse all EU countries. Columns 2, 4, 6, and 8 additionally contain the single bidder dummy variable as an explanatory variable to investigate the impact of potential corruption (single bidder procurement) on cost-effectiveness.

Table 5. Effect of the WTO GPA and Corruption on procurement cost-effectiveness

Notes: * p < 0.05; ** p < 0.01

All regression specifications in Table 5 show that the WTO GPA-covered dummy variable is significant with a negative coefficient. The WTO GPA lowers the probability that the procurement price is larger than estimated cost. Accordingly, WTO GPA covered tenders are significantly more likely to be cost-effective.

In addition to the impact of the WTO GPA on procurement cost-effectiveness, Table 5 provides consequential results about the effect of competition and potential corruption. We find that competition plays a consequential role for improving procurement cost-effectiveness. The coefficient of the number of offers is negative and significant in all regression specifications. An increase in the number of bidders significantly lowers the probability that the procurement price is higher than the estimated cost.

Columns 2, 4, 6, and 8 of Table 5 examine the impact of potential corruption on procurement cost-effectiveness. We examine the effect of the proxy for potential corruption, namely single-bidder procurement as described in section 6.1. The single-bidder dummy variables in columns 2, 4, 6, and 8 have significant positive coefficients. Hence, single-bidder contracts are more likely to have procurement prices higher than their estimated costs. Accordingly, potential corruption proxied by single bidder procurement significantly lowers the cost-effectiveness of government procurement.Footnote ³⁴ These results display the importance of competition as well as the importance of eliminating corruption in achieving government procurement cost-effectiveness.