Utility efficiency analysis, MSc · Performance Analytics

Scope My Performance Analytics project, worked individually on 130 publicly owned US electricity distribution utilities.

The problem

Benchmark the operational efficiency of 130 US electricity utilities: which are getting the most out of what they spend, and which are lagging. In a regulated industry this isn't academic, it's how a regulator sets price controls and how a utility justifies its costs.

Approach

Ran two complementary frontier methods. First, Stochastic Frontier Analysis: a Cobb-Douglas cost function (log opex on log units and network length) estimated in Stata's frontier routine, which separates genuine inefficiency from statistical noise. Second, Data Envelopment Analysis, a non-parametric frontier for peer benchmarking. Returns-to-scale tests (CRS versus VRS) justified the model choice, and running both methods shows where the efficiency verdict is robust and where it depends on the approach you take.

Result

Efficiency scores for all 130 utilities, cross-checked across estimators: only 2 came out fully efficient and 15 fell below 50%, a wide spread of performance. The point of a spread like that is what you do with it, flag the laggards for scrutiny, set realistic improvement targets against the frontier, and treat the robust verdicts differently from the ones that shift between methods. It's the same benchmarking approach regulators and risk functions use across utilities, banking and healthcare.

Tools

Stata DEA SFA Cobb-Douglas

Bar chart comparing mean efficiency scores across SFA estimators, COLS, MOLS, Battese-Coelli and JMLS, all around 0.76 to 0.89 — Efficiency scores compared across estimators: the verdict holds up whichever you use.

Distribution of efficiency scores across 130 utilities, clustered toward the high end with a tail of underperformers — The spread: most utilities cluster near the frontier, a tail lags well behind.

Stata stochastic frontier output: cost function on ln_opex, ln_units and ln_length across 130 observations — The model: Stata's stochastic frontier fit of the Cobb-Douglas cost function (130 firms).

All projects