Abstract
Frontier benchmarking is central to incentive regulation in electricity distribution. Yet it is typically implemented using nonparametric cost frontiers that impose economic assumptions, such as convexity, thereby abstracting away from sophisticated nonlinear alternating-current power-flow models. This paper shows that, although physically faithful, mathematically accurate specifications of power-flow systems are difficult to deploy for constructing regulatory cost frontiers at scale, while the adoption of nonparametric frontiers can itself give rise to structural misspecifications. We derive a production possibility set and an associated cost function from a stylised power-flow model and compare the resulting engineering benchmark with convex and nonconvex nonparametric estimators. We find that convexification can generate systematic deviations from the engineering cost frontier, whereas nonconvex counterparts provide substantially closer approximations. Evidence from Finnish and Iranian electricity distributors points in the same direction: nonconvex cost frontiers fit the data better and imply materially different efficiency assessments. In this sense, while economists can be wrong in acritically adopting a convex paradigm, engineering models, despite not directly delivering regulatory objects, can be accurately approximated by nonconvex cost frontiers.