The Solution and Estimation of Discrete Choice Dynamic Programming Models by Simulation and Interpolation: Monte Carlo Evidence

20 Simulation estimation for panel data models with limited dependent variables

“…This is despite the facts that the first important econometric application of simulation estimation was in this area (Pakes, 1986), the area continues to be a fertile one (see, e.g., Berkovec and Stern, 1991;Hotz and Miller, 1991;and Geweke, Slonim and Zarkin, 1992), and that much of my current research is in this area (Keane and Wolpin, 1992;Erdem and Keane, 1992). This omission stems from my desire to focus on the special problems that arise in the simulation of probabilities of sequences of events, excluding those additional problems that arise when the solution of a dynamic programming problem must also be simulated.…”

A practitioner’s guide to Bayesian estimation of discrete choice dynamic programming models

“…In this section, we will compare the IJC algorithm with two other estimation approaches by Keane and Wolpin (1994) and Ackerberg (2009), which also rely on approximating the likelihood.…”

“…Another method to reduce the computational burden for this type of model is by Keane and Wolpin (1994). Their method is more general in the sense that it applies even though all the state variables are discrete.…”

“…In their approach, the expected value functions are evaluated at a subset of state points and some methods of interpolation are used to evaluate the expected value functions at other values of the state space. Keane and Wolpin (1994) provide Monte Carlo evidence that suggests their approximation would converge to the true solution as the subset of the state points that are chosen increases. This method has proven to be very effective in reducing the computational time for finite horizon DDP models with a large state space (e.g., Erdem and Keane 1996;Keane and Wolpin 1997;Ackerberg 2003;Crawford and Shum 2005).…”

“…This is mainly due to two obstacles of estimating this class of model: (i) the curse of dimensionality problem in the state space, putting a constraint on developing models that match the real world applications; (ii) the complexity of the likelihood/GMM objective function, making it difficult to search for the global maximum/minimum when using the classical approach to estimate them. To overcome the hurdle due to the curse of dimensionality problem, several studies have proposed different ways to approximate the dynamic programming solutions (e.g., Keane and Wolpin 1994;Rust 1997;Hotz and Miller 1993;Aguirregabiria and Mira 2002;Ackerberg 2009). 1 Nevertheless, little progress has been made in handling the complexity of the likelihood function resulting from DDP models.…”