Collusion in Customs: Evidence from Madagascar

Size: px

Start display at page:

Download "Collusion in Customs: Evidence from Madagascar"

Augustus French
5 years ago
Views:

1 Collusion in Customs: Evidence from Madagascar Cyril Chalendard, Ana M. Fernandes, Aaditya Mattoo, Gael Raballand, Bob Rijkers The World Bank Annual Bank Conference on Development Economics June 25, 2018 PRELIMINARY

2 Motivation Tax evasion and corruption are well-established development constraints Less is known about who evades and to what extent evasion is facilitated by (which) bureaucrats This paper offers a new methodology to detect collusion in customs, and quantify its prevalence and costs using unique customs transaction data from Madagascar

3 This Paper: A Road Map Step 1: Detect deviations from conditional random assignment of import declarations to inspectors 10.3% of declarations are deviant Step 2: Demonstrate that such deviations are associated with increased risk of evasion Higher risk scores, higher taxes, larger declarations, but lower ex-ante unit prices Step 3: Document differential treatment of deviant declarations by inspectors Cleared more quickly, given priority, less fraud, and less upward value revision => associated with potential tax revenue losses Step 4: Describe how collusion between inspectors and brokers evolved after introduction of performance pay for inspectors (work in progress) Initial findings suggest no drastic change in nature or prevalence of collusion

4 Why customs in Madagascar? Customs revenue account for 44 percent of overall tax revenues in spite of substantial tariff evasion 3/4rd of revenue is made in Toamasina the main port and subject of this study Conditions are conducive to collusion Few players: approximately 15 inspectors (at any given point in time), 52 brokers Repeated interaction High stakes: each inspector is responsible for approximately 2% of the state budget Limited sanctions: culture of impunity and absence of meritocracy Performance pay was introduced in Toamasina in September 2016 Impact is likely to depend on what happened to collusive arrangements

5 Related literature Corruption and public sector performance (Olken and Pande, 2012) Role of bureaucrats: Dincecco and Ravanilla (2017), Best et al. (2017) Tax evasion (Besley and Persson (2013) Performance pay: Kahn et al. (2001); Olken et al. (2016) Tariff evasion: Bhagwati (1964, 1967); Fisman and Wei (2004); Javorcik and Narciso (2008, 2017), Dutt and Traca (2010); Yang (2008); Sequeira and Djankov (2014); Sequeira (2016), Rijkers et al. (2016), Javorcik and Demir (2018) Customs performance Volpe Martincus et al. (2015); Laajaj et al. (2018)

6 Customs clearance process Inspector assignment IT system assigns inspector Customs port manager can over-ride this Registration Risk-analysis GASYNET: Issues risk score Recommends inspection channel Potentially issues valuation advice Comments on risk characteristics Customs risk department Determines inspection channel (Also) comments on risk characteristics Assessment Inspector checks the declaration, potentially changes inspection channel issues a scan physically inspects Determines customs value, taxes and potential infractions Clearance Taxes and potential penalties are paid Goods clear customs

7 Data Madagascar customs Transaction-level import data (ASYCUDA) HS8 product, value, weight, source country, inspector, importer, broker, taxes and tariffs paid, date of registration, assessment and clearance both pre- and post-assessment Inspector allocation and actions (ASYCUDA SPY database) Risk management data (inspection channel, results from scanner exam) Fraud records SGS (Gasynet): risk score, inspection channel & (proprietary) valuation advice Port authority: physical weight COMTRADE: HS6 product export values and quantities recorded by partners (used to construct median unit prices by HS6 product-partner)

8 1 Detecting Deviations from Formal Procedure

9 Who allocates? Allocation of declarations to inspectors Automated allocation via the ASYCUDA customs IT system The customs port manager has discretion and can override the initial allocation in case of absenteeism, illness or unexpected events How? Minimum workload rule: incoming declaration should be allocated to the inspector who has the lowest workload (lowest number of pending declarations) and is active (connected to IT system) Testable implication: Share of declarations that a given inspector handles should not vary systematically across brokers

10 Observed inspector share Detecting deviations from official rules For broker b, inspector i s share in semester t is: SS iiiiii = nn iiiiii kk jj=1 nn jjbbbb = nn iiiiii NN bbbb n = number of import declarations registered by broker & k = number of inspectors active Predicted inspector share according to official rules (conditional random allocation) For inspector i, probability of receiving at least XX declarations from broker b is assumed to follow a binomial distribution with probability pp iiii (= share of total declarations in semester t handled by inspector i): Prob XX; NN bbbb, pp iiii = NN bbbb! XX! NN bbbb XX pp iiii XX (1 pp iiii ) NN bbbb XX

11 Do allocations of declarations follow official rules? Shares computed by semester for brokers importing at least 50 declarations.

12 Who manipulates? Distribution of Declarations Across Inspectors All Declarations - Shares by broker Distribution of Declarations Across Inspectors Reassigned Declarations - Shares by broker Frequency Frequency Inspector's share of a given broker's declarations (%) Inspector's share of a given broker's declarations (%) Predicted Final Initial Predicted Final Initial Excess inspection share = SS iiiiii pp iiii Excess interaction: (i) excess inspection share>0.05 & (ii) Prob<0.001 After initial IT assignment: 9.8% After IT assignment and intervention from customs port manager: 10.3%

13 2 Are deviant declarations more risky?

14 Deviant declarations are more risky Risk score Inspector's excess share of a given broker's declarations Excess inspection share is correlated with ex-ante risk scores physical inspections (red channel) mixed shipments differentiated products being subject to valuation advice by Gasynet Fitted 95% CI

15 Excess interaction is associated with higher stakes.higher taxes bigger declarations Tax rate ln initially declared weight Inspector's excess share of a given broker's declarations Inspector's excess share of a given broker's declarations Fitted 95% CI Fitted 95% CI Excess interaction is also associated with higher initially declared value

16 Excess interaction is associated with potential ex-ante revenue losses Higher unit price gaps (=median unit price unit price) Higher potential revenue losses (=tax rate*unit price gap*initial weight) Ex ante unit price gap ln potential tax revenue loss Inspector's excess share of a given broker's declarations Inspector's excess share of a given broker's declarations Fitted 95% CI Fitted 95% CI

17 Determinants of deviations Dependent variable: Excess inspection share Risk score ** (0.0002) Red channel (0.0017) Valuation advice performed ** (0.0069) Tax rate ** (0.0095) Mixed shipment * (0.0013) Contains differentiated products (0.0015) Ex-ante unit price gap *** (0.0010) Initial customs value (0.0003) Observations 29,460 R-squared Source country fixed effects Yes HS2-product fixed effects Yes Month-year fixed effects Yes Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

18 3 Differential Treatment

19 Deviant declarations are treated differently ln assessment time (from allocation to assessment) ln number of other declarations cleared during assessment time Dependent variable Scanned Share of containers scanned Rerouting of declaration to a different inspection channel Fraction of ex-ante value gap adjusted ln (1+tax adjustment) Fraud record (1) (2) (3) (4) (5) (6) (7) (8) Excess Inspection Share (by broker) *** *** ** ** *** *** ** (0.2313) (0.1912) (0.0412) (0.0338) (0.0627) (0.6143) (0.3986) (0.0586) Observations 50,162 49,355 50,232 16,637 50,232 16,992 50,232 50,232 R-squared Risk Controls YES YES YES YES YES YES YES YES Inspector FE YES YES YES YES YES YES YES YES Broker FE YES YES YES YES YES YES YES YES Source country FE YES YES YES YES YES YES YES YES HS2-product FE YES YES YES YES YES YES YES YES Year-month FE YES YES YES YES YES YES YES YES Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

20 Excess interaction is associated with revenue losses Ex-post unit price gap relative to COMTRADE median unit price Dependent Variable Ex-post weight gap Ex-post weight gap relative to port authority weight Ex-post value gap relative to reference value (port authority weight *COMTRADE median unit price) Ex-post value gap relative to SGS value advice Ex-post potential revenue loss (=tax rate *ex-post unit price gap *final weight) (1) (2) (3) (4) (5) (6) Excess Inspection Share (by broker) *** *** *** *** *** (0.1530) (0.0355) (0.0464) (0.1508) (0.1751) (0.1504) Observations 29,460 28,724 29,460 29,460 29,460 29,460 R-squared Risk Controls YES YES YES YES YES YES Inspector FE YES YES YES YES YES YES Broker FE YES YES YES YES YES YES Source country FE YES YES YES YES YES YES HS2-product FE YES YES YES YES YES YES Year-month FE YES YES YES YES YES YES Note: Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

21 The results are robust to Controlling for familiarity number of past interactions between the inspector and the broker Excluding declarations registered during weekends or after hours (e.g. after 8 pm and before 7 am) Controlling for congestion number of pending declarations Dropping top 3 inspectors with largest average deviations Excluding outliers Dropping transit countries as source countries

22 Results are not driven by inspector specialization Ex-ante unit price (item-level) Item-level regressions Dependent Variable Change in Ex-post unit price unit price (item-level) (item-level) Ex-ante potential revenue loss (=tax rate *ex-ante unit price gap relative to median *initial weight) (item-level) Change in potential revenue loss (item-level) Ex-post potential revenue loss (=tax rate *ex-post unit price gap relative to median *final weight) (item-level) (1) (2) (3) (4) (5) (6) Excess Inspection Share (by broker) *** *** *** *** *** *** (0.1488) (0.0127) (0.0538) (0.1233) (0.0120) (0.0506) Observations 185, , , , , ,723 R-squared Risk Controls YES YES YES YES YES YES Inspector FE YES NO NO YES NO NO Broker FE YES YES YES YES YES YES Source country*hs8 FE YES YES YES YES YES YES Year-month FE YES YES YES YES YES YES Inspector*HS8 FE NO YES YES NO YES YES Note: Robust standard errors (clustered by inspector in columns (1) and (4)) in parentheses, *** p<0.01, ** p<0.05, * p<0.1

23 (Prioritized) high tax declarations receive especially privileged differential treatment Dependent variable Fraction of Fraud ex-ante value record gap adjusted Ex-post unit price gap Ex-post potential revenue loss (1) (2) (3) (4) Panel A: Interaction EIS with Taxation Rate Excess Inspection Share (by broker) (EIS) (1.7552) (0.0652) (0.4902) (0.4131) Tax rate*eis-b * ** ** *** (4.8671) (0.1685) (1.4425) (1.3001) Tax rate *** *** *** *** (0.2910) (0.0179) (0.0763) (0.0740) Observations 16,992 50,232 29,460 29,460 R-squared Panel B: Interaction EIS with high tax and high speed dummies Excess Inspection Share Share (by broke (1.1575) (0.0805) (0.2451) (0.2235) EIS*High tax -fast (htf) *** *** (1.2850) (0.1026) (0.4185) (0.4001) EIS*High tax -slow (hts) ** * (1.5508) (0.1566) (0.3621) (0.3805) EIS*Low-Tax Fast (ltf) (1.2999) (0.0954) (0.3350) (0.3605) High tax -fast (htf) *** *** *** *** (0.1086) (0.0081) (0.0384) (0.0261) High tax -slow (hts) ** *** *** *** (0.1076) (0.0060) (0.0361) (0.0285) Low-Tax Fast (ltf) *** *** *** * (0.0519) (0.0037) (0.0138) (0.0096) Observations 16,992 50,232 29,460 29,460 R-squared Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

24 4 How did collusion evolve after the introduction of performance pay for inspectors

25 The prevalence of manipulations did not change Dependent Variable: Excess inspection share (1) (2) Reform (post Sept 2016) (0.0021) (0.0034) Risk score ** (0.0002) Red channel (0.0017) Valuation advice performed ** (0.0069) Tax rate ** (0.0095) Mixed shipment * (0.0013) Differentiated products (0.0015) Ex-ante unit price gap (0.0003) Initial customs value *** (0.0010) Observations 51,315 29,460 R-squared Source country fixed effects Yes Yes HS2-product fixed effects Yes Yes Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

26 The reform improved customs performance yet relative differential treatment did not change Dependent variable ln Fraction of assessment ex-ante time (from value allocation gap to adjusted assessment) Fraud record Ex-post unit price gap Ex-post potential revenue loss (1) (2) (3) (4) (5) Excess Inspection Share (by broker) *** *** *** *** *** (0.3442) (1.0042) (0.1042) (0.1488) (0.1383) Excess Inspection Share (by broker)*reform (0.4356) (1.9065) (0.1722) (0.2485) (0.2504) Reform *** *** *** *** *** (0.0344) (0.0740) (0.0052) (0.0130) (0.0120) Observations 50,162 16,992 50,232 29,460 29,460 R-squared Risk Controls YES YES YES YES YES Inspector FE YES YES YES YES YES Broker FE YES YES YES YES YES Source country FE YES YES YES YES YES HS2-product FE YES YES YES YES YES Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

27 Conclusion We offer a new method for detecting collusion in customs exploiting deviations from the minimum workload rule 10.3% of declarations exhibit deviations from this rule Deviations are shown to be associated with an increased risk of evasion and differential treatment by inspectors Revenue losses associated with this type of collusion are equivalent to 6% of total revenue Thus far, initial evidence suggests performance pay contracts have not dramatically altered the prevalence and nature of collusive arrangements but attendant improvements in monitoring helped direct audits

28 Implications IT is not a panacea for curbing corruption manipulation of IT customs system can enable corruption Instead, we conjecture that curbing collusion requires focusing on incentives for all customs staff (not just frontline inspectors!) and brokers (Underpowered) performance pay contracts may not suffice to curb collusion Results call for more research on how bureaucrats performance shapes public good provision

29 Circumstantial Evidence Interviews with inspectors I have been here 7 months, but there are certain brokers whose declarations I have never handled I never get the good declarations Interviews with port manager Workload is not evenly divided across inspectors I don t know why; it s the IT department s job [to divide declarations evenly] UNCTAD Officials: There is over-reliance on IT administrator account suspiciously long queues outside the office of the head of the IT department, who does not normally directly interacts customers Customs audits Sanctioning or voluntary departure of inspectors identified by our method

30 Back-up slides

31 Reallocations are associated with improved customs outcomes especially when they rectify non-random assignment Fraction Fraud Ex-post Ex-post of exante records average potential ln assessment time (from allocation to assessment) value gap price gap total tax revenue loss adjusted VARIABLES (1) (2) (3) (4) (5) Excess Inspection Share (by broker) *** *** ***0.1294*** (0.0374) (0.0984) (0.0089) (0.0237) (0.0230) Reallocation *** ** *** ** ** (0.0483) (0.1224) (0.0045) (0.0244) (0.0210) R excess -> excess (0.3300) (1.0478) (0.0910) (0.0827) (0.0802) R no excess -> excess (0.2554) (0.4549) (0.0246) (0.0785) (0.0889) R excess-> no excess ** * ** (0.1498) (0.5372) (0.0224) (0.1499) (0.1455) Observations 48,029 16,350 48,094 28,309 28,309 R-squared Risk characteristics YES YES YES YES YES Inspector FE YES YES YES YES YES Broker FE YES YES YES YES YES Source country FE YES YES YES YES YES HS2-product FE YES YES YES YES YES Year-month FE YES YES YES YES YES Standard errors clustered by inspector. *** p<0.01, ** p<0.05, * p<0.1.

32 Results are robust to controlling for familiarity Dependent Variable ln assessment Fraction of time (from ex-ante value allocation to gap adjusted assessment) Fraud record Ex-post unit price gap Ex-post potential revenue loss (1) (2) (3) (4) (5) Excess Inspection Share (by broker *** *** *** *** *** (0.2792) (0.6075) (0.0668) (0.1738) (0.1648) Ln(1+interactions in past 6 months ** ** * (0.0244) (0.0749) (0.0047) (0.0184) (0.0169) Observations 50,133 16,962 50,170 29,413 29,413 R-squared Risk Controls YES YES YES YES YES Inspector FE YES YES YES YES YES Broker FE YES YES YES YES YES Source country FE YES YES YES YES YES HS2-product FE YES YES YES YES YES Year-month FE YES YES YES YES YES Robust standard errors clustered by inspector in parentheses, *** p<0.01, ** p<0.05, * p<0.1

Combatting collusion: early lessons from Madagascar Customs

Combatting collusion: early lessons from Madagascar Customs Cyril Chalendard Ana Fernandes Aaditya Mattoo Gael Raballand Bob Rijkers PICARD Tunisia 2017 Introduction Although corruption (in customs) is