# Context MPEC was introduced for structural models where an equilibrium or dynamic programming condition can be enforced directly inside the estimator. In dynamic discrete choice, it estimates the same primitive payoff parameters as NFXP while treating the value function as an optimization variable. The practical appeal is diagnostics. NFXP hides the Bellman fixed point inside each likelihood evaluation. MPEC writes that fixed point as an equality constraint, so the fitted result can report the final Bellman constraint violation directly. ## Source Ideas {ref}`Su and Judd (2012) ` introduce the constrained-optimization approach to structural estimation: optimize structural parameters and equilibrium objects jointly while imposing the Bellman fixed point as an equality constraint. {ref}`Iskhakov et al. (2016) ` compare MPEC-style and NFXP-style numerical strategies in dynamic discrete choice settings and identify where the constrained problem becomes fragile as the state dimension or discount factor grows. The core identification lesson matches NFXP and CCP. Reward scale and location need a normalization. Transitions need to be separated from payoffs. Reward features need enough action variation to identify structural parameters. MPEC adds one more requirement. The constrained optimization problem must be small and well conditioned enough that the optimizer can jointly move reward parameters and value variables. ## Where MPEC Fits MPEC targets the same structural reward object as NFXP and CCP in finite tabular dynamic discrete choice models. It is useful as a constrained likelihood cross-check when NFXP is the reference and when Bellman constraint diagnostics are important. CCP is usually faster when first-stage choice probabilities are well supported. NNES and TD-CCP become attractive when the value-function constraint is too large for exact tabular constrained optimization.