Download
Abstract
Corrupt public works are rarely identified, so their cost is hard to measure. Two Peruvian legal records identify them: the Odebrecht bribery case and a 2021 ruling against thirty-two construction groups for fixing road contract prices. Projects with a documented bribe went over budget by 53 percent, the rest by 17. Where firms fixed the price the winner charged 108 percent of the government’s cost estimate, against 90 percent when they competed, and the same firm’s bid was about four points higher against that estimate on a contract it had fixed. Households next to a corrupt road did worse than those next to a comparable road where the firms competed, though across a wider forty-kilometre area we cannot detect a difference. The ruling also lets us test the methods used to catch price fixing. Ranking contracts by the winning price puts a fixed one above a competitively bid one 87 percent of the time. Ranking them by how far apart the bids are does no better than chance.
What the paper does
Prices the distortion. Peruvian road tenders admit bids only inside a band around a published reference value, so the cartel’s gain is not a wider spread but an erased discount. Colluded winners sit near the ceiling of that band, at a median of 108 percent, competitive winners at the 90 percent floor. The raw gap is 10.8 points of the reference value, 7.5 with year, terrain, works-type, size and bidder-count controls, and 4.1 holding the winning firm and the year fixed. That last figure is identified by the fourteen firms that won tenders of both kinds, so the gap is a property of the tender rather than of which firms turned up.
Validates the detection screens against a verdict. Because the competition authority adjudicated each tender, the collusion screens the literature uses can be scored against ground truth rather than defended in the abstract. A level screen on the winning bid reaches an area under the curve of 0.87. The dispersion screen most often used sits at 0.50, no better than chance. Across 34 refits dropping one award year or one winning firm at a time, the level screen ranges 0.854 to 0.901 and beats dispersion in every one, with the two ranges never overlapping.
Locates who bears the cost. A matched design pairs each colluded tender with a comparable competitively procured one and compares the districts around each. Households immediately around a colluded work do worse. Widening the catchment to forty kilometres and collapsing it into one population-weighted area observation, the effect is indistinguishable from zero, which rules out a regional loss of the size the ten-kilometre contrast would imply, though not a much smaller one. The finding is about where the burden falls. Inside the affected radius the shortfall sits in labour earnings, most robustly in earnings from independent work, and opens within two years of the tender rather than waiting for the road to be delivered.
Keywords
Corruption, Procurement, Collusion, Bid-Rigging, Cost Overruns, Collusion-Detection Screens, Spatial Incidence, Peru, Infrastructure
Citation
Chávez Padilla, Carlos César. 2026. “The Price of Corruption: Evidence from Lava Jato and the Construction Club in Peru.” Working paper.