Abstract
Public procurement can be highly vulnerable to corruption. This paper outlines a methodology and results in assessing corruption risks in public procurement and their impact on relative prices, using large databases on government contracts and tenders. Our primary contribution is to analyze how price differential in public procurement contracts can be explained by corruption risk factor (aggregated in a synthetic corruption risk index). While there are intrinsic limitations to our study (price differentials can come from structural reasons, such as a limited number of potential suppliers) it still provides a guiding tool to assess where corruption risks would have the biggest budgetary impact. Such analysis helps inform mitigating policies owing to the granular data used.
I. Introduction: A Method to Assess Corruption Risks in Public Procurement
1. Public procurement, a crucial way to implement government budgets, can be highly vulnerable to corruption (IMF, 2019). Estimates of losses through procured spending amounts to about 10–20 percent, even in countries with relatively high integrity of their procurement systems in the European Union (Hafner et al., 2016). Consequences for public finances can be dire, as public procurement constitutes about 12 percent of global GDP or 11 trillion USD per year (Bosio et al., 2020). Corruption can lead to higher deficits and lower growth, due to (among others) inadequate quality and/or insufficient level of infrastructure (Schwartz et al., 2020).
2. Does corruption explain higher prices paid for procured public goods or services? This simple but crucial question has, in a nutshell, contrasted answers. There are, for example, policy experiments showing how strengthening–or simply introducing–rules that deter corruption tend to reduce the cost or improve the quality of procured goods and services (Banerjee et al., 2016). There are also microeconomic studies pointing at similar results (see, for example, Di Tella and Schargrodsky, 2003, on the procurement of medical goods in Argentina, or Palguta and Pertold, 2017, on public procurement in the Czech Republic, Coviello and Mariniello, 2014, on the role of transparency and publicity). Empirical studies can also suggest “red flags” of corrupt behaviors, which can subsequently inform the design of procurement rules (see Fazekas, Tóh, and King, 2016, on how to measure corruption risks in public procurement). For example, both Paguta and Pertold, (2017) and Coviello, Guglielmo, and Spagnolo (2018) show that when public officials can exert discretionary powers (usually for procurement contract values below a certain threshold) then there is a concentration of contracts awarded to specific bidders and/or untransparent bidders in terms of their owner structure. However, the fact that corrupt behaviors can lead to higher prices paid doesn’t necessarily mean that either the impact is significant or that reducing the impact is worthwhile economically. In particular, corruption can, in some instances, lead to rather small costs, while other sources (inefficiencies in the procurement process, or costs associated to implement anti-corruption measures) can be a lot more significant (see Bandiera, Prat, and Valletti, 2009, for public procurement in Italy). A common denominator to all these studies is how uncompetitive public procurement processes can help corruption thrive, eventually translating into higher prices paid (Bandiera, Prat, and Valletti, 2009, Fazekas and Tóth, 2017) or lower quality of procured goods and services (Golden and Picci, 2005, Fazekas and Tóth, 2017).
3. The main contribution of our paper is to assess whether or not red flags of corrupt behaviors have an impact on prices of procured goods and services. We do so by estimating the impact of corruption risks (assessed through red flags), on relative prices (i.e., comparing prices paid with reference prices). Our approach has three main novelties and two caveats. First, we rely on very large databases, covering all regulated public procurement contracts available in the countries reviewed. We achieved that by gathering data from government web portals for public procurement. As a result, we can have broader results in terms of coverage than other papers of the literature that usually focus on sub-set of procurement contracts. Second, we focus on corruption risks in a more systematic way, by mapping seven red flags of corrupt behaviors, which correspond to risks identified in the literature (i.e., not only on procurement processes such as competitive procedure types, but also on the part of suppliers, for example by exploring the impact of concentration of procurement contracts on specific bidders). Thus, we assess through our estimations if these red flags have an explanatory power on relative prices. This is important in terms of informing anti-corruption policies, which is our third contribution. Indeed, a red flag can point to very granular policy recommendations, because our methodology enables policy makers to explore which sectors may be more vulnerable to corruption and/or what factors are the most important in explaining price differentials. We explore the policy analysis and recommendations in a companion paper (Basdevant and Fazekas, 2022). The first caveat of our approach is that we focus on corruption risks and not actual instances of corruption. In particular, and like others working in this field, finding that our variables have an impact on relative prices gives an indication of potential corrupt behaviors, but other factors may be at play. For example, single bidding on specific markets may have nothing to do with corruption but more about structural features of these markets. Thus, our results need to be taken with caution by policy makers, which should subsequently further explore corruption vulnerabilities. The second caveat is that by nature of this quantitative exercise we do not explore how corruption risks could affect the quality of procured goods and services. This could nonetheless be the subject of further research but is left aside in the context of this paper.
4. The rest of this paper focuses on the presentation of the results for the five pilot countries for which the methodology was developed: Georgia, Indonesia, Paraguay, Romania, and Uganda. These countries were chosen as they had publicly available dataset on procurement contracts and provided a diverse representation of continents.6 The dataset for the five pilot countries includes over 1.5 million contracts, capturing from 15 to 55 percent of total procured spending in each country. Our analysis provides a granular distinction between cases where, say, corruption risks could be high with, overall, limited impact on price differentials, versus cases where corruption risks would be small, but with potential large impacts on relative prices. We also develop a Corruption Risk Index (CRI), which can be of particular relevance to track more precisely corruption vulnerabilities (as opposed to one of its component, which, taken individually, may not provide enough information on corruption risks).
II. Assessing Vulnerabilities to Corruption Using Objective Data
5. In this section, we present seven indicators of likely corrupt impediments to open competition in public tenders, which constitute our measurement of corruption risks in procurement using hard data. We first present seven indicators, the “red flags”, that we use to compute a composite indicator, the CRI, based on a simple average of these seven red flags. Then we offer a balanced assessment of the strengths and weaknesses of our approach to measuring corruption risks. These “red flags” do not attempt to identify corruption per se, but instead to measure risks in an objective manner. The indicators used encompass the indicators typically used in the literature on the subject, but in a more comprehensive and consistent way. Indeed, most papers would typically focus on a subset of these seven indicators, as they usually only look at a very specific corrupt behavior (either for a specific behavior, or for a specific country, see Fazekas, Tóth, and King, 2016, or Fazekas and Kocsis, 2020 for a review as well as Table 1). In addition, the seven selected indicators can be consistently calculated across a large sample of publicly available procurement datasets, underpinning a globally standard methodology.
Red Flags Used in the Literature on Corruption in Procurement Systems
Red Flags Used in the Literature on Corruption in Procurement Systems
| Source | Indicator(s) used | Country | Years | Sector | Potential for international compa | Included in the CRI (y/n) | Included in the CCT (y/n) |
|---|---|---|---|---|---|---|---|
| Auriol, Flochel, and Straub (2011) | Number of exceptional procedures | Paraguay | 2004–2007 | general procurement | HIGH >Requires harmonization of definitions |
y | y |
| Bandiera, Prat, and Valletti (2009) | Price differentials between goods either bought locally or procured. | Italy | 2000–2005 | standardized goods (e.g. paper) | LOW Price data is not readily available in most countries, many countries don’t have national procurement agencies, national procurement agencies are likely to be captured in many countries. |
n | y |
| Chong, Klien, and Saussier (2015) | Number of negotiated procedure | EU | 2008–2012 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Coviello and Gagliarducci (2010) | Number of bidders Same firm awarded contracts recurrently Level of competition |
Italy | 2000–2005 | general procurement | HIGH Number of bidders, recurrent contract award, and competitiveness of bids are available in many countries. |
y | y |
| Di Tella and Schargrodsky (2003) | Price differential for standardized products | Argentina | 1996–1997 | health care | MEDIUM Detailed product-level price and quantity information is not readily available across many countries, but can be collected. |
n | n |
| Fazekas and Kocsis (2015) | Composite risk indicator (e.g., singled bidder, period for submitting bids) | EU | 2009–2014 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Ferwerda, Deleanu, and Unger (2016) | Composite indicator (e.g., short advertisement period) | EU | 2006–2010 | general procurement | HIGH Some of the indicators are based on data that is typically not collected centrally. |
y | y |
| Golden and Picci (2005) | Ratio of physical stock of infrastructure to cumulative spending | Italy | 1997 | infrastructure | MEDIUM It is hard to compute comparable value of the stock of physical capital across countries different in the quality of infrastructure and |
n | n |
| Hyytinen, Lundberg, and Toivanen (2009) | Number and type of invited firms Use of restricted procedure |
Sweden | 1990–1998 | cleaning services | HIGH Both number of bidders and procedure types are readily available in many countries. |
y | y |
| Klasnja (2016) | Single bidder auctions Non-open procedure types | Romania | 2008–2012 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Olken (2006) | Difference between in-kind benefits (rice) officially recorded and those reported in surveys | Indonesia | 1998–1999 | welfare spending | MEDIUM It is possible to design user surveys across a wide range of countries to track actual receipts, although it may be expensive. |
n | n |
| Olken (2007) | Differences between prices and quantities officiallyy reported and independently audited | Indonesia | 2003–2004 | infrastructure (roads) | LOW Auditing large numbers of projects by independent engineers is costly and unlikely to allow for cross-country comparisons. |
n | n |
Red Flags Used in the Literature on Corruption in Procurement Systems
| Source | Indicator(s) used | Country | Years | Sector | Potential for international compa | Included in the CRI (y/n) | Included in the CCT (y/n) |
|---|---|---|---|---|---|---|---|
| Auriol, Flochel, and Straub (2011) | Number of exceptional procedures | Paraguay | 2004–2007 | general procurement | HIGH >Requires harmonization of definitions |
y | y |
| Bandiera, Prat, and Valletti (2009) | Price differentials between goods either bought locally or procured. | Italy | 2000–2005 | standardized goods (e.g. paper) | LOW Price data is not readily available in most countries, many countries don’t have national procurement agencies, national procurement agencies are likely to be captured in many countries. |
n | y |
| Chong, Klien, and Saussier (2015) | Number of negotiated procedure | EU | 2008–2012 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Coviello and Gagliarducci (2010) | Number of bidders Same firm awarded contracts recurrently Level of competition |
Italy | 2000–2005 | general procurement | HIGH Number of bidders, recurrent contract award, and competitiveness of bids are available in many countries. |
y | y |
| Di Tella and Schargrodsky (2003) | Price differential for standardized products | Argentina | 1996–1997 | health care | MEDIUM Detailed product-level price and quantity information is not readily available across many countries, but can be collected. |
n | n |
| Fazekas and Kocsis (2015) | Composite risk indicator (e.g., singled bidder, period for submitting bids) | EU | 2009–2014 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Ferwerda, Deleanu, and Unger (2016) | Composite indicator (e.g., short advertisement period) | EU | 2006–2010 | general procurement | HIGH Some of the indicators are based on data that is typically not collected centrally. |
y | y |
| Golden and Picci (2005) | Ratio of physical stock of infrastructure to cumulative spending | Italy | 1997 | infrastructure | MEDIUM It is hard to compute comparable value of the stock of physical capital across countries different in the quality of infrastructure and |
n | n |
| Hyytinen, Lundberg, and Toivanen (2009) | Number and type of invited firms Use of restricted procedure |
Sweden | 1990–1998 | cleaning services | HIGH Both number of bidders and procedure types are readily available in many countries. |
y | y |
| Klasnja (2016) | Single bidder auctions Non-open procedure types | Romania | 2008–2012 | general procurement | HIGH Requires harmonization of definitions |
y | y |
| Olken (2006) | Difference between in-kind benefits (rice) officially recorded and those reported in surveys | Indonesia | 1998–1999 | welfare spending | MEDIUM It is possible to design user surveys across a wide range of countries to track actual receipts, although it may be expensive. |
n | n |
| Olken (2007) | Differences between prices and quantities officiallyy reported and independently audited | Indonesia | 2003–2004 | infrastructure (roads) | LOW Auditing large numbers of projects by independent engineers is costly and unlikely to allow for cross-country comparisons. |
n | n |
6. Because our indicators are comprehensive and observable for all countries in our study, they allow the development of a synthetic CRI, which can also be used for cross-country comparisons.7 These seven red flags and the underlying corrupt behaviors are as follows (see also Appendix II):
Single bidder contracts, that is contract awarded in a tender where only one bidder participated, represents a straightforward way to gauge limited competition in public tenders. It can be a sign of corrupt practices in public procurement as corruption is more likely to arise and indeed easier to organize when there is only in participating company (Klasnja, 2016, Charron, et al., 2017; Fazekas, Tóth, et al., 2016).
Non-open procedures, leading to uncompetitive tenders (Auriol, Flochel, and Straub, 2016). Lack of open participation in procurement tenders can limit the number of competing bidders, thus opening an avenue for public official to extract an illegal rent from the procurement process. The most straightforward example of such risks is a high-value contract directly awarded to a bidder without any competition or request for quotations. As country regulatory contexts differ from each other and change over time, identifying non-open procedure types requires both observing tendering outcomes and conducting an in-depth analysis of procedural rules, either set in laws or secondary legislations.
Lack of publication of call for tenders, as limited publication can lead to uncompetitive tenders (Coviello and, Mariniello, 2014, Björkman, and Svensson, 2009, Lewis-Faupel et al., 2016, Zamboni Litschig, 2018). Not publishing call for tenders is also a notable deviation from the core principle of transparency of procurement processes established by the OECD (2016).
Period for submitting bids, can also represent vulnerabilities to corruption risks. This red flag is analyzed in two ways. Typically, a short period is associated with unfair competition (less time to prepare adequate bids). However, extensive submission periods can signal legal challenge and lengthy modifications of tendering terms which underpin favoritism towards a single bidder. This red flag is assessed, taking into account country-specific features (see Table 7 in Appendix II) of the degree to which different submission periods are associated with single bidding.
Period for selecting the winning bid can also be a red flag for corruption risks (see also Fazekas and Kocsis, 2020).8 A short period is associated to unfair competition (bids may not all be adequately assessed). In some cases, long decision periods can also signal that a particular bidder was favored because challenging the bid assessment, hence increasing the period for the final award decision, is a hallmark of irregularities.
Spending concentration (by organization, by year) can also be a sign of corrupt practices as corruption could lead to a higher concentration of procured spending in specific bidders. Conversely, dominant market positions can be abused by bidders to extract corrupt rents.
Share of suppliers registered in jurisdictions offering limited company and banking transparency. This indicator is quite critical, as a company registered in such jurisdictions would typically avoid adequate oversight. As a result, it’s easier, especially for trans-national companies, to engage in corrupt activities by facilitating secrecy on illegal payments made to public officials.
7. The CRI is based on a simple average of the seven individual red flags (after being normalized, see Appendix II). We score each contract, on each of the seven red flags, with a discreet score: 0 for lowest corruption risk, 0.5 for medium risk, and 1 for the highest risk. Then, we average the score across all contracts for each category (see Table 2 below). In the case of single bidding, for example, the score is either 0 or 1 at the contract level, and thus the score corresponds to the frequency of single bidding across all procurement contracts (e.g., in Romania 31 percent of contracts were awarded in tenders with a single bidder). Then, we average the scores across the seven red flags for each contract to compute the CRI (Figure 1).9 The CRI provides a more reliable indication of corruption risks because corruption would in general manifests itself through various techniques and strategies. Further, the CRI is primarily based on a cardinal measure of corruption (instead of ordinal), thus avoiding that a country is identified as a high corruption risk simply because its economy would perform less favorably than other countries. This is a particularly helpful contribution of this work, as not only do these indicators avoid the recourse to expert judgments, but also allow for country specificities to be factored in, while still providing an even-handed methodology and thus a common metric to assess corruption risks in any given country.
Average Score for Each Red Flag
(Value between 0, lowest risk, and 1, highest risk)
Average Score for Each Red Flag
(Value between 0, lowest risk, and 1, highest risk)
| Single bidding | Procedure type | Submission period | Decision period | No CFT published | Tax heaven | Spending concentration | CRI (average) | |
|---|---|---|---|---|---|---|---|---|
| Uganda | 0.84 | 0.21 | 0.48 | 0.61 | 0.72 | … | 0.37 | 0.54 |
| Georgia | 0.52 | 0.38 | 0.48 | 0.78 | 0.01 | 0.17 | 0.24 | 0.37 |
| Paraguay | 0.47 | 0.80 | 0.04 | 0.57 | 0.00 | … | 0.07 | 0.32 |
| Indonesia | 0.00 | 0.66 | 0.47 | 0.29 | 0.00 | … | 0.27 | 0.28 |
| Romania | 0.31 | 0.02 | 0.05 | 0.13 | 0.03 | 0.04 | 0.21 | 0.11 |
Average Score for Each Red Flag
(Value between 0, lowest risk, and 1, highest risk)
| Single bidding | Procedure type | Submission period | Decision period | No CFT published | Tax heaven | Spending concentration | CRI (average) | |
|---|---|---|---|---|---|---|---|---|
| Uganda | 0.84 | 0.21 | 0.48 | 0.61 | 0.72 | … | 0.37 | 0.54 |
| Georgia | 0.52 | 0.38 | 0.48 | 0.78 | 0.01 | 0.17 | 0.24 | 0.37 |
| Paraguay | 0.47 | 0.80 | 0.04 | 0.57 | 0.00 | … | 0.07 | 0.32 |
| Indonesia | 0.00 | 0.66 | 0.47 | 0.29 | 0.00 | … | 0.27 | 0.28 |
| Romania | 0.31 | 0.02 | 0.05 | 0.13 | 0.03 | 0.04 | 0.21 | 0.11 |
Corruption Risk Index and Its Components
(Indices of CRI components, 1=highest corruption risk)
Citation: IMF Working Papers 2022, 094; 10.5089/9798400207884.001.A001
Sources: Country authorities, and authors computationsCorruption Risk Index and Its Components
(Indices of CRI components, 1=highest corruption risk)
Citation: IMF Working Papers 2022, 094; 10.5089/9798400207884.001.A001
Sources: Country authorities, and authors computationsCorruption Risk Index and Its Components
(Indices of CRI components, 1=highest corruption risk)
Citation: IMF Working Papers 2022, 094; 10.5089/9798400207884.001.A001
Sources: Country authorities, and authors computations8. Strengths and limitations of the CRI. As noted before, the seven red flags are also commonly used in the literature as indications of corruption risks (see IMF, 2018). When developing a composite indicator, we also have the following two advantages, which are, in spirit very similar to the approach proposed by the IMF (2018). First, The CRI is a more robust indicator of vulnerabilities to corruption than its individual components. Across, countries, sectors and over time, corruption typically thrives on various vulnerabilities, as corrupt officials would, especially when corruption is macro-critical, use several strategies to extract illegal rents. Thus, any consistent and reliable indicator of corruption risks, has to be able to gauge a range of such strategies. This feature is a potentially strong value added of the CRI, for both cross-country, cross-sectoral and time series comparisons, but also to assess how corruption may affect relative procurement prices (see next section). Second, The CRI is predominantly data-driven while being informed by established theories of corruption. The validity regressions outlined in Appendix II determine which definitions of risky categories are most closely aligned with the adopted definition of corruption. For example, the selection of high-risk procedure types, such as direct contracts, is driven by their association with single bidding rather than legislative intent or procedural design on paper. Against these two advantages, there are, however, two limitations. First, as discussed previously, in some countries/sectors, uncompetitive markets for specific goods can be related to structural features of the economy/market and not corrupt practices.10 Further, the CRI (and more generally our approach) does not include capacity assessment. The primary reason for not including capacity evaluations is due to data limitations: capacity assessments are not necessarily available for all countries. Further, capacity assessments could potentially be questioned as being non-objective measures since they are derived from experts’ judgements and are sometimes the result of self-evaluations (such as evaluations in the context of the Public Expenditure and Financial Accountability program, PEFA).
III. Estimating the Impact of Corruption Risks on Price Differentials
9. We assess the budget implications of corruption risks, by estimating the impact of CRI (and other individual indicators) on prices in awarded contracts. The regressions link the size of discounts offered by the winning firm compared to the auction reference price (which is based on standard market prices, and usually corresponds to the maximum budgetary allocation for a given purchase defined prior to the tender) based on corruption risks while controlling for year, contract value, main market, buyer location, and buyer type on the contract level (Fazekas and Tóth, 2018). Relative prices are calculated as actual contract values divided by the initially estimated contract value of the tender11 (or through savings, if available directly in the dataset). One critical interest of such regressions is that they pave the way for bridging our large-scale micro-level dataset with macro aggregates such as budget deficit and to offer different macro spending estimates based on different risk levels in each country and sector. Naturally, and as noted previously, the fact that the CRI (or single indicators as shown in the rest of this section) is linked to higher relative prices is not sufficient to imply a causal relationship from corruption risks to prices. Indeed, higher prices can be explained by specific structural circumstances in some market or countries (please note that the regressions control for market specificities using a wide set of product market fixed effects).
10. We estimate five main regression models to estimate the relationship between the CRI and the relative prices of procured goods, works and services (see Table 3 for the case of Georgia, Appendix III for the other countries, and descriptive statistics and histograms of relative prices are in Appendix I). Model 1 has CRI as the only independent variable with the other models (2–5) including a battery of control variables accounting for variation by product market (CPV division, location, and contract value), organizational framework (buyer type) and time-dependent shocks (year). Models 1–3 restrict relative prices to between 0.5 and 1.5 because those few extremely low relative prices (winning bid below 50 percent of the reference price) or extremely high relative prices (winning bid more than 50 percent above the reference) are most likely erroneous records that would bias our estimates. In addition, we also look at a more conservative set of regressions with relative prices restricted to 0.5–1, in essence cropping the upper end of the distribution (i.e., a few percent of the total sample depending on the country). The rationale behind removing relative prices above one is that in our country sample, just like in most other countries, awarding a contract above the reference price is unusual, requiring special circumstances and bureaucratic approvals hence represent a special case. Finally, model 5 also allows for a quadratic specification for CRI to capture non-linearities in the data. Crucially, for our subsequent discussion of corruption costs, the coefficients remain significant and largely the same size across the different specifications. Model 4 is chosen as the main prediction model for all countries as it is considered most robust with the widest range of control variables, and it has typically the highest explanatory power. While the non-linear models add to explanatory power, the improvement is little which we consider insufficient advantage in return for upping model complexity.
Georgia – Main Results
Defined as the ratio of actual contract value and normal value (i.e. the value underpined by the reference price, based on standard market prices).
Georgia – Main Results
| 1 | 2 | 3 | 4 | 5 | |
|---|---|---|---|---|---|
| Dependent variable: relative price (RP) 1/ | (0.5<RP<1.5) | (0.5<RP<1.5) | (0.5<RP<1.5) | (0.5<RP≤1) | (0.5<RP≤1) |
| CRI | 0.276*** (0.004) |
0.315*** (0.004) |
0.312*** (0.004) |
0.312*** (0.004) |
0.222*** (0.012) |
| (CRI)2 | 0.116*** (0.016) |
||||
| Year controls | ✓ | ✓ | ✓ | ✓ | |
| Contract Value (100 quantiles) | ✓ | ✓ | ✓ | ✓ | |
| CPV division | ✓ | ✓ | ✓ | ✓ | |
| Buyer type | ✓ | ✓ | ✓ | ||
| Buyer location | ✓ | ✓ | ✓ | ||
| Observations | 188,472 | 188,472 | 188,472 | 188,414 | 188,414 |
| R-squared | 0.15 | 0.20 | 0.21 | 0.21 | 0.21 |
Defined as the ratio of actual contract value and normal value (i.e. the value underpined by the reference price, based on standard market prices).
Georgia – Main Results
| 1 | 2 | 3 | 4 | 5 | |
|---|---|---|---|---|---|
| Dependent variable: relative price (RP) 1/ | (0.5<RP<1.5) | (0.5<RP<1.5) | (0.5<RP<1.5) | (0.5<RP≤1) | (0.5<RP≤1) |
| CRI | 0.276*** (0.004) |
0.315*** (0.004) |
0.312*** (0.004) |
0.312*** (0.004) |
0.222*** (0.012) |
| (CRI)2 | 0.116*** (0.016) |
||||
| Year controls | ✓ | ✓ | ✓ | ✓ | |
| Contract Value (100 quantiles) | ✓ | ✓ | ✓ | ✓ | |
| CPV division | ✓ | ✓ | ✓ | ✓ | |
| Buyer type | ✓ | ✓ | ✓ | ||
| Buyer location | ✓ | ✓ | ✓ | ||
| Observations | 188,472 | 188,472 | 188,472 | 188,414 | 188,414 |
| R-squared | 0.15 | 0.20 | 0.21 | 0.21 | 0.21 |
Defined as the ratio of actual contract value and normal value (i.e. the value underpined by the reference price, based on standard market prices).
11. To illustrate the outcome of our preferred specification (model 4), we show below the impacts of the CRI on relative prices across the five countries (Table 10). All fiver response functions are upward sloping demonstrating the expected price increasing effect of corruption risks across the board. Interestingly, some of the curves are steeper (e.g., Romania) than others (e.g., Uganda).12 For example, in Romania, an additional red flag (1/7 points increase on the CRI score) is predicted to increase prices by 4.4 percentage points (0.307*(1/7)*100=4.4).
Estimated Price impact of CRI Increase by Country
Elasticities taken from model 4. Significance of the estimates shown in tables of “main results” regression in appendix II, except for Georgia where results are shown in Table 8.
Everything equal, an additional red flag (1/7 points increase on the CRI score) increase prices by 1/7 times the elestacity.
Estimated Price impact of CRI Increase by Country
| Price elasticity to CRI 1/ | Price impact of CRI increase 2/ | |
| (Unit) | (Percent) | |
| Paraguay | 0.39 | 5.5 |
| Georgia | 0.31 | 4.5 |
| Romania | 0.31 | 4.4 |
| Uganda | 0.10 | 1.4 |
| Indonesia | 0.07 | 1.0 |
Elasticities taken from model 4. Significance of the estimates shown in tables of “main results” regression in appendix II, except for Georgia where results are shown in Table 8.
Everything equal, an additional red flag (1/7 points increase on the CRI score) increase prices by 1/7 times the elestacity.
Estimated Price impact of CRI Increase by Country
| Price elasticity to CRI 1/ | Price impact of CRI increase 2/ | |
| (Unit) | (Percent) | |
| Paraguay | 0.39 | 5.5 |
| Georgia | 0.31 | 4.5 |
| Romania | 0.31 | 4.4 |
| Uganda | 0.10 | 1.4 |
| Indonesia | 0.07 | 1.0 |
Elasticities taken from model 4. Significance of the estimates shown in tables of “main results” regression in appendix II, except for Georgia where results are shown in Table 8.
Everything equal, an additional red flag (1/7 points increase on the CRI score) increase prices by 1/7 times the elestacity.
12. We also run alternative specifications, to explore how individual indicators affect relative prices. These regressions can also be of particular use for investigating with more scrutiny what specific factor(s) in the CRI can contribute to overpricing. One could even go further by letting data dictate the relative weights of the CRI components. However, while doing so would bring a better fit, it would lose the ability of the CRI to be used as an easy-to-interpret and even-handed indicator of corruption risks. Further, corruption is likely to thrive (as noted by IMF, 2018 and 2019) when vulnerabilities are widespread across many potential channels.
Georgia – Alternative Specifications
Defined as the ratio of actual contract value and normal value.
Georgia – Alternative Specifications
| 1 | 2 | 3 | |
|---|---|---|---|
| Dependent variable: relative price (RP) 1/ | (0.5<RP<=1) | (0.5<RP≤1) | (0.5<RP≤1) (> 4 contracts per year) |
| 1.singleb | 0.149*** (0.00111) |
||
| 1.nocft | -0.0192*** (0.00311) |
||
| w_ycsh4 | 0.0567*** (0.00364) |
||
| Observations | 188,414 | 188,414 | 130,722 |
| R-squared | 0.40 | 0.06 | 0.07 |
Defined as the ratio of actual contract value and normal value.
Georgia – Alternative Specifications
| 1 | 2 | 3 | |
|---|---|---|---|
| Dependent variable: relative price (RP) 1/ | (0.5<RP<=1) | (0.5<RP≤1) | (0.5<RP≤1) (> 4 contracts per year) |
| 1.singleb | 0.149*** (0.00111) |
||
| 1.nocft | -0.0192*** (0.00311) |
||
| w_ycsh4 | 0.0567*** (0.00364) |
||
| Observations | 188,414 | 188,414 | 130,722 |
| R-squared | 0.40 | 0.06 | 0.07 |
Defined as the ratio of actual contract value and normal value.
13. To illustrate the global applicability of our approach, we applied it to infrastructure overpricing in the European Union. A prior study by Fazekas and Tóth (2017) looks at relative prices and a similar corruption risk index in infrastructure projects of 27 EU Member States (including the U.K.; but excluding Malta due to its small size). They find comparably diverse, albeit positive impacts of CRI on infrastructure prices across countries with most low corruption countries such as the Netherlands showing a muted, in some cases even insignificant, cost impact of corruption risks. Reassuringly, our impact estimates appear largely consistent with such prior research with the impact estimated falling on the upper end of the EU distribution which is hardly surprising given the five countries in our pilot sample fall in the high corruption spectrum. When comparing Table 9 (right hand-side column) with Figure 2, please note that the latter depicts the full impact of CRI increasing from 0 to 1, while the former shows the marginal effect of an additional red flag, that is 1/7th CRI increase. Thus, the partial impacts depicted in Table 9 are much lower than the overall impact illustrated in Figure 2.
IV. Conclusion
14. The methodology and its results presented in this study shows considerable potential to assist policy makers in identifying corruption risks in procurement systems and their costs. In particular, the methodology leads to a very user-friendly output (see Basdevant and Fazekas, 2022), while the inputs are all shared in a transparent way and can be easily customized by users. Our methodology is also a helpful tool for policy makers and stakeholders to identify (i) sources of corruption risks (e.g., by sector, region, or public organization type) and (ii) reform measures to address these risks (by tackling the identified red flags).
15. Our analysis provides useful guidance on potential anti-corruption measures in public procurement in the five countries covered. First, as noted in Table 4, the impact of corruption risks is significant, with an additional red flag leading to a price increase of 1 to 5 percent. Beyond this broad result, a further granular observation of red flags can inform the direction of potential anti-corruption measures. Appendix II presents a more detailed analysis, and to illustrate our point, further scrutiny could be given to submission periods (among other red flags). For instance, countries can not only assess the rational for the length of their submission periods based on our red flag analysis, but also consider how they fare against other countries. This could lead to fruitful discussions on whether submission periods reflect some country-specific factors, or if they should consider revisiting submission periods to reduce corruption risks.
16. Our analysis comes at an opportune time since the fight against the COVID-19 pandemic has put enhance scrutiny on public procurement contracts across the world. Because of its granular approach, the methodology can be used to assess corruption risks (and their impact) in procured medical supplies. It can also be used to track in real time if these corruption risks are declining as a result of measures taken to strengthen the oversight of public spending. In particular, once established at a country level, the CCT can benefit from rapid updates as countries publish more procurement contracts.
17. To support the broad goal of curbing corruption in public procurement, the development of web portals and databases for public procurement is essential. The introduction of an e-procurement―an important step for increased transparency, lower transaction costs, and reduced discretion in decision-making―would be, in many countries, a critical step, but yet just a first step. Developing formal web portals for procurement should also go hand-in-hand with the development of databases for public procurement contracts, ideally as a machine-readable database. In particular, dedicated efforts could be considered to develop e-procurement systems in low-income countries (LIC).13 Rolling out such systems for LIC could prove particularly helpful in curbing vulnerabilities to corruption in procurement systems, especially if using the tool presented in this paper, in addition to improving economic efficiency and market access more broadly (Fazekas and Blum, 2021).
18. The authors are also working on adding further cost types such as cost overruns and also higher quality pricing information wherever available (i.e., unit prices of standardized goods). This addition would enable a more granular assessment of how corruption may affect procurement costs. Additionally, the dynamic between the components of the CRI could be quite complex at a country level. As such, it may be helpful to further explore the interaction between them, for example through a machine learning approach.
