The Credit Channel of Public Procurement
What this paper finds — and why it matters
Layer 1: Overview
Research question and motivation. Public procurement accounts for roughly one-third of government spending (12.6% of GDP and 30% of total government expenditures in OECD countries in 2019). The standard view is that procurement helps firms grow by raising their revenues. Gabriel asks whether procurement also operates through a previously underexplored credit channel: if a procurement contract is a secure future cash-flow stream, firms can pledge it as collateral to obtain more credit. This matters especially in bank-dependent economies (in Portugal and several OECD countries, >80% of nonfinancial corporate debt is bank loans; <1% of Portuguese firms access capital markets), and for small/financially constrained firms.
Data and strategy. The author web-scrapes >1 million Portuguese electronic procurement contracts (2009-2019) from the official BASE registry, matching winners’ tax IDs to firm balance-sheet/income data (IES via BPLIM) and to the monthly Credit Registry (CRC) with loan-level collateral types. Focusing on contracts awarded via public contests (a silent sealed-bid first-price-auction-like setting) for quasi-exogenous variation yields 138,561 contract-winner pairings and 35,675 unique winner-year observations. Average contract award is ~€202,170 (median ~€33,762-34,762), average duration ~297 days, ~3.6 contestants. Identification uses Jordà (2005) local projections (Eq. 1) regressing credit growth (scaled by lagged assets) on the award amount (scaled by lagged assets), with firm and industry×year fixed effects, SEs clustered at the firm level. The identifying assumption is that winning via public contest is not systematically correlated with firm characteristics; conditional on fixed effects, winner/non-winner differences largely disappear (except total assets, which is controlled).
Main findings (with magnitudes). Winning an additional €1 of procurement raises total firm credit by up to €0.07 (3.3 cents drawn credit on impact, plus ~4 cents in potential/undrawn credit lines; total ~7 cents in the award year), and raises cash and bank deposits by ~6 cents. Interest rates fall by over 0.3 percentage points on impact, indicating the increase is supply-driven (winners’ average implicit rate ~6.9%, median ~5.1%). A back-of-envelope calculation gives ~2.5 pp credit growth one year out (vs. ~5 pp in Spain per di Giovanni et al. 2024). The credit increase is almost entirely collateralized; in monthly data, firm personal guarantees (which include future procurement cash flows) account for >66% of the credit increase at month 4, and adding state guarantees, cash-flow-based lending explains ~75%. On the real side: +6 cents of non-current assets/investment (mostly PPE) per euro, persistent employment gains, ~70% rise in sales income one year post-award, positive net income of ~5 cents per euro. cash-flow-based lending is ~44% of firm credit in the sample.
Heterogeneity and aggregate. Investment responses are concentrated in small/constrained firms (β ≈ €7.3 for small/micro vs. −€1.2 for big firms 2 years out; difference significant at 1%); credit responses do not differ significantly by size. Regionally (Eq. 2, NUTS-III, region+year FE, clustered at region), €1 of procurement raises regional GVA by ~€1.3 (€1.32 on impact), implying ~€0.32 crowding-in of private production; the credit channel accounts for ~5% (5.5%) of this. Procurement boosts private R&D but not TFP, with only modest, short-lived inflation and no broad regional credit expansion (suggesting credit redistribution toward winners).
Q1. What is the identification strategy and what are the main threats to it?
The author exploits public contests, which resemble a silent sealed-bid first-price auction with a costly single bid: the hiring entity does not know who bids and firms do not know their competitors or how many there are, so the winner is not ex-ante predictable. He estimates Jordà (2005) local projections (Eq. 1) of credit growth on the award amount, both scaled by lagged total assets, with firm and industry×year fixed effects and firm-clustered SEs. The key identifying assumption is that winning via public contest is not systematically correlated with other firm characteristics. Threats: (i) selection if contracts go to more productive firms (would overstate effects) or displace private opportunities (would understate); (ii) anticipation, if firms foresee winning and adjust early. He addresses anticipation by including pre-event horizons h=-2, h=-3 (annual) and pre-months (monthly), finding no significant pre-trends, and by focusing on contests (where outcomes are unknown, unlike direct awards) and using yearly aggregation (the announce-to-decision gap was ~4 months in 2020). Figure C.1 shows unconditional winner/non-winner differences mostly vanish once fixed effects are included, except total assets (which is controlled). Appendix C.1 adds a local-projections difference-in-differences robustness check following Dube et al. (2023).
Q2. What is the credit channel mechanism and how is it distinguished from a demand story?
The mechanism is cash-flow-based lending: procurement contracts represent secure future cash flows that firms pledge as collateral (personal/firm guarantees), easing borrowing constraints. It is distinguished from a credit-demand story by the price of credit: a demand-driven increase would raise interest rates, but rates fall by >0.3 pp on impact, consistent with a supply-driven expansion. Two micro-mechanisms raise perceived creditworthiness: (i) collateral value of the contract itself, and (ii) a signaling/certification effect where government endorsement reduces bank information asymmetry. Monthly collateral decomposition (Figure 5) shows the credit increase is overwhelmingly backed by firm personal guarantees (>66% at month 4; ~75% including state guarantees), with asset-based collateral mostly insignificant, directly supporting the cash-flow collateral channel.
Q3. How is the signaling/certification mechanism tested separately?
In Appendix Table C.3 (discussed in Section 3.5) the author compares first-time award recipients to firms with previous awards. First-time winners enjoy significantly higher and more persistent responses in credit, employment, and investment, which he interprets as a reputation/certification effect that partially resolves a banking information-asymmetry problem (banks learn the firm has government demand). This is distinct from the pure collateral mechanism, which is tested with the monthly collateral-type decomposition.
Q4. What heterogeneity is documented?
By firm size (Commission Recommendation 2003/361/CE: small = headcount <50 and turnover/balance-sheet <€10m): credit responses do not differ significantly between small and big firms, but investment and employment responses are much larger and more persistent for small/constrained firms (investment β ≈ €7.3 small vs. −€1.2 big at 2 years, difference significant at 1% and growing with horizon; HAC p-values for employment differences are 0.05 at 1yr and 0.00 at 2yr). This is rationalized via the financial-accelerator hypothesis (Bernanke et al. 1999) and investment-cash-flow sensitivity literature (Fazzari et al. 1988). Employment heterogeneity mirrors Giroud and Mueller (2017). By sector: Construction and Medical Equipment (~60% of 2019 procurement value) account for much of the credit response but show no significant persistent differences in investment/employment. By award history: first-time winners respond more strongly (reputation effect).
Q5. What does the monthly analysis add over the annual analysis?
Using monthly credit/collateral data within the first year (relevant since the median contract lasts <1 year), the credit increase begins at award inception, rises sharply in the first month, and peaks ~3 months after the award (aligning with the annual ~3+ cents/euro). The increase is almost entirely collateralized (unsecured credit shows a muted response) and of sound quality (non-performing credit barely moves). Both long- and short-maturity credit rise, with long-term credit responding more strongly. Crucially, no significant credit movement appears up to three months before signing, reinforcing the no-anticipation conclusion.
Q6. What are the aggregate/regional results and how are they estimated?
The author aggregates procurement by spending location to NUTS-III regions and estimates local-projection multipliers (Eq. 2) with region and year fixed effects, SEs clustered at region, sample matched 2010-2016 (25 regions × 6 years), procurement winsorized at the 95th percentile. A €1 increase in regional procurement raises GVA by ~€1.3 (€1.32 on impact, interpreted as an open-economy relative multiplier à la Nakamura-Steinsson 2014), implying €0.32 crowding-in of private production. Eq. 3 interacts procurement with winners’ credit (following Basso and Rachedi 2021): the positive significant interaction means credit amplifies the multiplier; a 1% credit-to-GVA increase raises the multiplier by 11% on impact, and since winners’ credit is ~0.5% of GVA, the credit channel adds ~(0.11×0.5)% ≈ 5.5% (5%). National-accounts regressions (Table 4) show procurement raises private value added (€1.2 on impact), private investment, private R&D (innovation), and modest short-lived inflation, but not TFP; aggregate nonfinancial-firm credit is subdued, suggesting credit redistribution toward winners rather than broad expansion.
Q7. What robustness checks and caveats are noted?
Robustness: anticipation tests at multiple pre-horizons (annual and monthly); a local-projections diff-in-diff specification (Dube et al. 2023) in Appendix C.1; fixed-effects conditioning that removes most winner/non-winner differences; winsorizing the regional regressor at the 95th percentile (results sensitive to outliers). Caveats explicitly acknowledged: (i) no loan-level data, so the implicit interest rate is total interest expense / lagged effective credit, and financial covenants cannot be observed (if present, estimates would be conservative); (ii) under Portugal’s Public Procurement Code (Ch. IX), contracts above ~€500k may require a guarantee up to 5% of value, often a bank guarantee that appears as firm-guaranteed credit—but the central message still holds; (iii) procurement coverage is incomplete (web-scraped data ≈ one-third of total procurement, ~3% of GDP), so regional coefficients should be read with caution; (iv) the regional credit measure may not capture the full cumulative credit response and credit increases could partly reflect non-procurement factors; (v) collateral values are not market-adjusted and are often capped at the loan amount.
Q8. How does this paper relate to and differ from closely related prior work?
It contributes to three literatures. (1) Firm-level effects of fiscal policy/procurement (Barrot-Nanda 2020; Goldman 2020; Cox et al. 2024; Ferraz et al. 2021; Lee 2021): prior work emphasizes revenues as the driver; Gabriel adds a new credit/collateral transmission mechanism across all industries. The closest contemporaneous work is di Giovanni et al. (2024) for Spain, who document a positive procurement-credit correlation; relative to them, this paper provides detailed evidence on the credit-supply channel and its investment implications, measures contract heterogeneity, and—unlike their welfare/allocation-system focus—provides the first local procurement multiplier estimates with the credit channel’s share. (2) Government spending and fiscal multipliers, including stronger fiscal effects under tight credit (Ferraresi et al. 2015; Aghion et al. 2014). (3) Financial frictions and collateral type, shifting from asset/liquidation-value collateral (Kiyotaki-Moore 1997) to cash-flow-based collateral (Lian-Ma 2021; Ivashina et al. 2022; Drechsel 2022; Caglio et al. 2022); the novelty is cash flows from sales to the government as collateral. Notably his investment elasticity for small firms (~5 cents/euro cumulative at one year) is smaller than Hebous and Zimmermann’s (2021) ~13 cents.
Q9. What are the policy implications and their scope conditions?
Two implications: (1) Targeting design—because small/financially constrained firms respond more strongly and persistently in investment and employment, targeting procurement to such firms (as pushed by the European Commission/Parliament for SMEs) likely raises aggregate investment and employment, not just efficiency. (2) Financial stability—letting firms pledge procurement contracts as collateral diversifies collateral away from real-estate/asset-based booms (which deplete project information and lead to deep downturns, Asriyan et al. 2022), so procurement could temper collateral-induced financial fluctuations. Scope conditions: external validity is greatest for countries where procurement is a large GDP share and firms rely heavily on bank credit (true for many developed and developing economies, e.g., Portugal where <1% of firms access capital markets); the effect grows more important the more bank-dependent firms are. The interest-rate decline is a firm-level result and should not be read as procurement lowering equilibrium interest rates economy-wide; a procurement shock can be a reallocation of spending rather than higher total spending/deficit.
Q10. What is the nature of the real-side response and why is the sales response not larger?
Winning raises non-current assets by ~6 cents per euro (mostly PPE/tangible, not intangibles or financial investments), comparable to Hebous-Zimmermann’s ~10 cents and to real-estate-collateral elasticities (~6 cents, Chaney et al. 2012; Catherine et al. 2022). Employment rises persistently beyond the first year (Ferraz et al. 2021), though without a matching rise in value added. Sales income rises ~70% one year post-award—less than a one-for-one mapping of public demand to sales—for two reasons: a ‘duration effect’ (contracts spread revenue over years; some last up to a decade) and a ‘capacity constraint effect’ (firms prioritize government contracts, diverting other business to competitors, which also shows up in regional GVA), potentially mitigated by sub-contracting. Despite higher costs of goods sold, net income stays positive at ~5 cents per euro, so contracts are profitable.