The Economics of Spatial Coordination in Critical Infrastructure Investment

Abstract: We develop a hybrid approach to estimate spatial coordination mechanisms in structural dynamic discrete choice models by combining nested fixed-point (NFXP) dynamic programming with method of simulated moments (MSM), achieving computational tractability in spatial settings while preserving structural interpretation. Applying this framework to GPU replacement data from 12,915 equipment locations in Oak Ridge National Laboratory's Titan supercomputer, we identify two distinct coordination mechanisms: sequential replacement cascades (gamma_lag = -0.793) and contemporaneous failure batching (gamma_fail = -0.265). Sequential coordination dominates - approximately three times stronger than failure batching - indicating that operators engage in deliberate strategic behavior rather than purely reactive responses. Spatial interdependencies account for 5.3% of variation unexplained by independent-decision models, with coordination concentrated in high-risk thermal environments exhibiting effects more than 10 times stronger than cool zones. Formal tests decisively reject spatial independence (chi-squared(2) = 685.38, p < 0.001), demonstrating that infrastructure policies ignoring spatial coordination will systematically mistime interventions and forgo available coordination gains.