AVIATION SAFETY AND RISK
Estimating the Target Level of Safety from Public Attention to Aviation Accidents
Abstract: The Target Level of Safety (TLS) guides approval of new aviation procedures, yet the values in regulatory use were established through engineering judgment and analogy rather than estimated from observed behaviour. This paper develops an attention-based calibration linking monthly aviation risk to GDELT coverage, New York Times coverage, and Google search interest. Hansen threshold regressions across alternative accident measures and exposure denominators estimate the rate at which the attention response increases. The fatalities-per-passenger-year specification provides the most stable reporting anchor, at approximately 2.3 x 10^-7, within the HSE/ICAO individual-risk band. Scaled to the midair-collision subset, the estimate is 2.2 x 10^-8 fatal accidents per flight hour, with a 95% interval containing ICAO's en-route target of 1.5 x 10^-8. The results provide an empirical benchmark for acceptable-risk criteria in ultra-safe transport systems.
Empirical Collision Risk Estimation with Rare-Event Corrections and Heterogeneous Uncertainty
Extends the Wu et al. (2024) encounter-based framework with interval-level rather than closest-point-of-approach geometry, a heteroskedastic probit in which position-error scale varies with surveillance type, reporting rate and latency, an empirically estimated Weibull non-intervention term, and an explicit lateral and vertical decomposition that recombines into a single risk figure.
From Estimates to Evidence: A Smoothed Way to Report Operational Risk
Operational risk figures are usually computed from a handful of rare, heavy-tailed events and swing by a factor of two or more between years for reasons that are sampling noise rather than safety performance. This paper factors each rate into an arrival rate and a mean duration, updates both across years by conjugate Bayesian inference, and reports a smoothed central estimate with a credible range that narrows as evidence accumulates. Target Level of Safety compliance becomes a probability statement rather than a yes/no verdict.
A Packing-Geometry Bound on the Number of Close-Neighbour Pairs per Aircraft
System-level collision risk is often obtained by averaging pair-level risk over a reasonable airspace, a step whose definition is acknowledged to be subjective. This paper removes the subjectivity structurally. Defining occupancy as the average number of close-neighbour pairs in which each aircraft participates, it proves an unconditional upper bound of six, the planar kissing number, for any configuration respecting a minimum separation. The bound is scale-invariant and independent of the separation standard, aircraft speed, airspace shape, and heading distribution.
ECONOMICS
Dynamic Matching with Popularity Bias and Stochastic Errors in Anonymous Markets (with Roland Pongou)
Abstract: Can biased agents reach efficient outcomes in decentralized markets? We study this question in a dynamic model of an anonymous bipartite economy with heterogeneous capacities, where agents seek an optimal number of partners and may revise relationships over time. We introduce popularity bias, the tendency to favor already-popular alternatives, through two channels: a rational preference for popular partners, conditional on maintaining one's desired number of links, and a systematic influence on stochastic errors. Our main result is that efficiency, defined as link maximization, is achieved in the long run only when both channels are active. Under rational popularity bias alone, steady states refine toward anti-egalitarian networks, in which a minimal subset of agents concentrates most links, but these need not be efficient. When stochastic errors are also popularity-weighted, stochastic stability selects precisely the networks that are both efficient and anti-egalitarian. The same behavioral bias that drives market concentration therefore also drives efficiency. We discuss implications for platform design, showing that the steepness of popularity signals governs both the strength and the speed of convergence to efficiency, and that larger markets converge more slowly, suggesting a role for algorithmic intervention in large-scale platforms.
Does ESG Rating Coverage Affect Publicly Traded Companies' Performance? Evidence from a Switching Regression Approach (with Ali Ghali and Philippe Kabore)
Abstract: We examine whether ESG rating coverage, a firm's inclusion in a rating agency's scored universe as distinct from doing ESG itself, affects financial performance. Using Refinitiv ESG scores matched with Compustat data for North American firms over 2003-2022, we implement a switching regression with selection correction, instrumented by a survival model that predicts when a firm first enters Refinitiv's scored universe. Average return on assets falls modestly after coverage, while return on capital employed rises persistently. A decomposition shows that the decline in return on assets is denominator growth rather than profit compression: rated firms expand their asset base faster than their net income, consistent with a capital-attraction channel rather than a compliance-cost channel. Among the three rating dimensions, the social pillar consistently enhances performance in both the short and the long term. A cross-rater replication using MSCI ratings reproduces the sign and significance pattern, and a difference-in-differences design exploiting Refinitiv's 2015 expansion to the Russell 2000 yields the same qualitative answer.
Subjective Beliefs and Social Influence in Strategic Voting Behavior: Evidence from Recent Canadian Federal Elections
Abstract: Strategic voting in first-past-the-post elections is typically measured as an aggregate share, but whether the individual decision to defect from a non-viable preferred party is socially contagious has not been investigated. We address this question using survey data from three recent Canadian federal elections (2015-2021), combined with daily constituency-level vote projections used to construct subjective peer-strategic-voting measures. A one-percentage-point rise in the district-aggregate share raises the probability of strategic voting by 0.09 to 0.14 percentage points; the within-party effect is 0.06 to 0.08, and the cross-party effect is -0.05 to -0.10. The pattern matches the within-party coordination and cross-party dispersion predicted by the strategic-voting literature. Effects are stable across elections, parties, and campaign phases.
COVID-19 and Classroom Peer Effects in Ontario (with Christopher Cotton)
Abstract: Using population-level Ontario data, we estimate how classroom peer effects in Grade 6 mathematics changed during the COVID-19 pandemic. Pre-pandemic classroom peer coefficients are stable across the decade preceding the pandemic, at roughly 0.42 standard deviations. The 2022 cohort, whose Grade 4 and Grade 5 years coincided with Ontario's extended school closures, shows a 28 percent lower classroom peer coefficient, statistically significant against a flat pre-pandemic baseline and robust to plausibly-exogenous bounds. The attenuation is larger in absolute magnitude for lower-ability students. We cannot separate a pure weakening of the peer channel from mechanical substitution by Ontario's targeted recovery programs, which were directed at exactly the ability groups where we find the largest attenuation.
Social Interactions and the Demand for Financial Advice (with Roland Pongou)
Abstract: Social interactions - effects of the group on an individual - have been found to impact a wide range of outcomes and decisions; but they have, to this point, not being taken into account in the modeling of the decision to consult a professional on financial matters. Using data from the 2009 version of the Canadian Financial Capability Survey, this study investigates the presence of group effects in the decision to seek financial counsel from a professional advisor and make use of the received advice. Neighborhood trends are found to significantly impact the individual propensity to make use of financial advice, especially when individuals are sorted by the first two digits of their postal code. In contrast, social effects on the decision to consult a professional are not statistically different from 0. Investigation into heterogeneous effects by gender, age, education, and immigrant status point to homophily as the main mechanism behind these positive social interaction effects: individuals simply respond more noticeably to the actions of peers who share similar characteristics.