Corruption, anti-corruption, and economic development

The paper addresses how governments should balance the social losses from corruption against the costs of anti-corruption supervision in order to maximize government benefits. It motivates the study by noting that corruption is pervasive, undermines welfare, and that many existing studies treat monitoring and its costs as exogenous. The authors argue that governments actually choose supervision levels and related costs endogenously, making the optimal anti-corruption policy a key decision problem. Building on principal–agent reasoning, they posit that officials engage in corruption when discretionary power and rents are high relative to wages and detection risks are low, while governments must weigh the benefits of reducing corruption against the fiscal costs of hiring supervisors and paying wages. The study therefore develops a theoretical model that internalizes (endogenizes) the supervision level and its costs to derive optimal policies under varying economic conditions and institutional environments, including both homogeneous and heterogeneous officials.

The review outlines three main perspectives on corruption's impact on growth: (1) Harmful corruption hypothesis, which argues corruption distorts resource allocation, deters investment and FDI, reduces human capital, creates political instability, and impedes growth (e.g., Krueger, 1974; Shleifer and Vishny, 1993; Mauro, 1995; Mo, 2001; Wei, 1997, 2000; Pellegrini and Gerlagh, 2004; Blackburn et al., 2006; Kunieda et al., 2014; Gründler and Potrafke, 2019). (2) Efficiency-enhancing view, which contends that in rigid or dysfunctional regulatory environments, corruption can grease the wheels by easing bottlenecks and improving efficiency (e.g., Leff, 1964; Lui, 1985; Beck and Maher, 1986; Acemoglu and Verdier, 1998/2000; Dzhumashev, 2014; Egger and Winner, 2005; Wang and You, 2012; Jiang and Nie, 2014). (3) Context-dependent view, which holds that corruption's net effect depends on institutional quality and other conditions; in weak systems, limited corruption may appear less harmful or even offset inefficiencies, while in stronger systems it is detrimental (e.g., Méndez and Sepúlveda, 2006; Aidt et al., 2008; Méon and Weill, 2010; Aidt, 2009; Dong and Torgler, 2010; Zheng, 2015; Alfada, 2019; Petersen, 2021). The review also notes data limitations due to corruption's secrecy and recent experimental approaches to study behavior (e.g., Banerjee, 2016; Banerjee et al., 2022). It concludes that internalizing anti-corruption costs within models is crucial and underexplored, especially given substantial anti-corruption expenditures (e.g., in China).

The study develops a theoretical principal–agent framework that endogenizes the level and cost of government anti-corruption supervision. Economy: A static setting with citizens (mass λ, per capita income y, average tax rate t), a government that maximizes a benefit function, and government employees divided into supervisors (share n) and officials (share 1−n). Government employees earn wage w subject to a minimum wage constraint w ≥ y to ensure participation. Fiscal revenue is f = t λ y. Supervisors: Homogeneous; if they work normally, each detects corrupt or abnormal official behavior with probability p(n), defined as p(n) = n (the share of supervisors working normally). Due to information asymmetry, the government cannot observe supervisor effort; it incentivizes normal work by confiscating detected officials’ wages and distributing them equally among supervisors, ensuring supervisors choose to work (so the effective number of active supervisors equals n). Officials (homogeneous case): Each official has a corruption space b/(1−n), capturing the rent officials can extract if not detected; an undetected corrupt official gains b/(1−n) while reducing effort devoted to public projects by the same amount. Officials choose between normal work (utility w) and corruption (expected utility (1−p(n)) [w + b/(1−n)]). Given p(n)=n, the corruption expected payoff simplifies to (1−n) w + b. Officials choose normal work if w ≥ b/n; otherwise they corrupt. Government objective: G(n, w) = α [ f − w − (1−p(n)) B(n, w) ] + β (1−n), where α > 0 is the benefit from one unit of net resources invested in infrastructure/public goods, β > 0 is the direct benefit per official (reflecting professional contribution regardless of corruption), and B(n, w) is total undetected corruption income. The government chooses (n, w) to maximize G(n, w) subject to 0 ≤ n ≤ 1 and w ≥ y. Analytical cases (homogeneous officials): • Low wage region w < b/n: all officials corrupt; then B(n, w) = b and G simplifies to a linear function in n and w. • High wage region w ≥ b/n: no official corrupts; B(n, w) = 0 and G simplifies accordingly. • Considering only the minimum wage constraint w ≥ y, the optimal policy is derived by comparing these regions and boundary conditions. Key parameter restriction: Assumption 1 requires b < β/α to avoid unbounded corruption space dominating decisions. Heterogeneous officials: A fraction m ∈ (0,1) are Group A (less honest, corruption preference coefficient 1), and fraction 1−m are Group B (more honest), with a corruption preference coefficient σ ∈ (0,1) that scales their effective corruption space to σ b/(1−n). Group B’s corruption incentive condition becomes w ≥ σ b / n to deter corruption. The analysis proceeds in three wage regions: (i) low wage w < σ b / n: all officials corrupt; (ii) medium wage σ b ≤ w < b / n: Group A corrupts, Group B works (partial corruption); (iii) high wage w ≥ b / n: no one corrupts. For each region, B(n, w) and G(n, w) are derived, and optimal (n, w) solved subject to w ≥ y. The heterogeneous case introduces thresholds in per capita income that determine whether the optimal policy yields zero supervision, partial corruption with positive supervision, or comprehensive anti-corruption. The authors define δ = β/α and show that under certain combinations of (m, σ) there exist two income thresholds: a lower supervision-investment threshold N(m,σ) and an upper threshold L(m,σ) for comprehensive supervision, with partial corruption emerging when per capita income lies between these thresholds. Propositions and corollaries formalize the optimal choices in each region and how they depend on parameters y, b, α, β, m, and σ.

• With homogeneous officials and only a minimum wage constraint (w ≥ y):
  • When per capita income y is low (below a threshold proportional to β/α), the optimal supervisory input n is zero and the government sets w = y. Budget constraints make anti-corruption spending costlier than tolerating corruption; officials' professional contributions (β term) dominate.
  • If the government adopts a high-pay policy ensuring w ≥ b/n (so corruption is not individually rational), then the optimal (n, w) depends on y: when y is small, the interior optimum occurs where the government’s indifference curve is tangent to w = b/n (implying positive n and w > y); when y is large, the corner solution is at the intersection w = y and w = b/n, with n = b / y. As y increases further, the optimal n declines (anti-corruption becomes less labor-intensive as higher wages reduce incentives to corrupt).
  • Considering the minimum wage constraint only and optimizing across wage regimes, the optimal n as a function of y exhibits a non-monotonic pattern: starting at n = 0 for low y, jumping up at a critical income y = β/α, peaking at n = b / y at that threshold, and then declining toward zero as y → ∞. This rationalizes why poorer countries often have limited supervision despite higher corruption, while richer countries sustain low corruption with relatively low supervision input.
• Zero-tolerance via high pay is feasible: If the government pays sufficiently high wages (relative to corruption space and supervision), rational officials will all choose normal work; corruption can be eliminated in equilibrium.
• With heterogeneous officials (fractions m of less honest Group A and 1−m of more honest Group B with corruption preference σ ∈ (0,1)):
  • There are three wage regions: low (all corrupt), medium (partial corruption where Group A corrupts but Group B does not), and high (no corruption). The government’s optimal (n, w) depends on per capita income and parameters (b, α, β, m, σ).
  • If per capita income is very low, the optimal choice mirrors the homogeneous case: n = 0 and w = y (no supervision). As income rises past a first threshold N(m,σ), the optimal policy features positive supervision and wages sufficient to deter Group B but not Group A, generating partial corruption. When income exceeds a higher threshold L(m,σ), the optimal choice moves to comprehensive anti-corruption (no corruption), with supervision and wage settings that deter all officials.
  • The existence and breadth of the partial-corruption income range depend on (m, σ). When m and σ are small (more honest societies), partial corruption is more likely and persists over a wider middle-income range. When m and σ are large (fewer honest officials and/or weaker intrinsic honesty), the partial-corruption range shrinks or disappears, reverting to the homogeneous-like pattern without a partial-corruption phase.
• Policy implication: At low incomes, budgets should prioritize productive capacity rather than costly supervision; at middle incomes, targeted supervision and adequate wages can constrain corruption among more honest officials even if some corruption persists; at high incomes, sufficiently high wages and calibrated supervision can sustain zero-corruption equilibria.

The model directly addresses the policy problem of choosing optimal anti-corruption supervision when both the social costs of corruption and the fiscal costs of supervision are internal to government decision-making. By endogenizing the monitoring level, the analysis shows why governments in low-income contexts rationally underinvest in supervision despite high corruption: the opportunity cost of scarce fiscal resources and the value of officials’ professional contributions dominate. As fiscal capacity grows, optimal supervision rises sharply at a threshold and then gradually declines with further income growth as higher wages themselves deter corruption. Introducing heterogeneity in officials’ honesty explains observed partial corruption in many middle-income settings: the optimal policy deters corruption among more honest officials but does not fully eliminate corruption by less honest types because the marginal cost of doing so exceeds the marginal benefit. These results help reconcile empirical observations of high growth coexisting with nontrivial corruption and offer a structured rationale for stage-contingent anti-corruption strategies. The framework highlights how institutional and economic parameters (e.g., the relative contribution of officials versus resource inputs, the size of corruption opportunities, and the distribution of honesty) shape optimal policy and the evolution of corruption over development.

The paper contributes a theoretical framework that internalizes anti-corruption supervision choices, linking optimal supervision and wage policies to economic development and institutional parameters. It shows: (1) at low income levels, optimal supervision is essentially zero; (2) after an income threshold, optimal supervision jumps and then declines with further development, with sufficiently high wages capable of eliminating corruption; and (3) with heterogeneous officials, a middle-income range can arise where the optimal policy results in partial corruption, offering an explanation for persistent corruption in some middle-income countries. These insights suggest a stage-based approach to anti-corruption: prioritize productive capacity at low incomes, deploy targeted supervision and wage policies at middle incomes, and sustain high-wage, calibrated supervision at high incomes. Future research should relax idealized assumptions, incorporate political and cultural factors, and develop dynamic models and experimental validations to assess how changing social norms, political trade-offs, and temporal dynamics affect the optimal balance between corruption losses and supervision costs.

The model relies on idealized assumptions and abstracts from other determinants of corruption such as cultural norms, political environment, and legal institutions, which may limit external validity. It is theory-driven and lacks empirical calibration or validation, introducing potential bias in quantitative implications. Government budgeting is assumed to be guided purely by economic efficiency, omitting political trade-offs and societal pressures that influence real-world decisions. The framework is static and does not incorporate temporal dynamics (e.g., evolving public tolerance of corruption or changes in institutional quality), suggesting the need for dynamic extensions and empirical or experimental tests.