When aggregate productivity over-diagnoses deterioration

Using a quarterly panel of seven Codelco divisions (2000–2021), this paper demonstrates that the apparent corporate productivity deterioration is partially over-diagnosed: when separating geology, technical inefficiency, and operational regime persistence, the diagnosis changes substantially. The architecture combines translog stochastic frontier, monotone boosting with economic monotonicity constraints, a three-state Gaussian Hidden Markov Model, and FHK decomposition. The central finding: in divisions with effective deterioration, 91.9% of the weighted drag comes from geology and operation; technical inefficiency explains only 3.2%.