What is respy?#
respy is an open source framework written in Python for the simulation and estimation of some finite-horizon discrete choice dynamic programming (DCDP) models. In comparison to simple reduced-form analysis, these models allow the estimation of structural parameters which reflect agents’ preferences and beliefs by assuming that agents are forward-looking and maximize expected intertemporal payoffs. Over the last decades, finite-horizon DCDP models have become a popular tool to answer research questions in areas of economics such as labor economics, industrial organization, economic demography, health economics, development economics, political economy, and marketing.
What makes respy powerful is that it allows to build and solve structural models in weeks or months whose development previously took years. The design of respy allows the researcher to flexibly add the following components to her model.
Any number of discrete choices (e.g., working alternatives, schooling, home production, retirement) where each choice may yield a wage, may allow for experience accumulation and can be constrained by time, a maximum amount of accumulated experience or other characteristics.
Condition the decision of individuals on its previous choices or their labor market history.
Adding a finite mixture with any number of subgroups to account for unobserved heterogeneity among individuals as developed by Keane and Wolpin (1997, [13]).
Any number of time-constant observed state variables (e.g., ability measures (Bhuller et al., 2020, [5]), race (Keane and Wolpin, 2000, [14]), demographic variables) found in the data.
Correct the estimation for measurement error in wages, either using a Kalman filter in maximum likelihood estimation or by adding the measurement error in simulation based approaches.
As is common with structural economic models, finite-horizon DCDP models oftentimes rely on strong assumptions regarding unobservable state variables and error terms (see Aguirregabiria and Mira, 2010, [1], p. 40 for a list of assumptions used in standard finite-horizon DCDP models). respy focuses on the estimation of so-called Eckstein-Keane-Wolpin (EKW) models. In accordance with Aguirregabiria and Mira (2010, [1]) , we classify a DCDP model as an EKW model if it departures from standard DCDP modeling by relaxing at least one of the following assumptions:
The one-period utility function does not have to be additively separable in its observable and unobservable components but can instead feature different compositions, e. g. multiplicative separability.
Observable payoff variables can be choice-censored and the value of the payoff variable does not have to be independent of the error term \(\epsilon\), conditional on the values of the decision and observable state variables.
Permanent unobserved heterogeneity is allowed to exist, i. e. the unobserved state variables do not have to be independently and identically distributed over agent and over time. As an example, the seminal work of Keane and Wolpin (1997, [13]) introduces permanent unobserved heterogeneity by assigning each individual to one of four types.
Unobservables may be correlated across choice alternatives, i. e. unobserved state variables do not have to be independent across alternatives.
To learn more about DCDP models and related topics, check out Recommended Reading