Unobserved Heterogeneity in CCP Estimation

Описание: What happens when you can't observe all the relevant state variables in your dynamic discrete choice model? In this video, we tackle one of the most common challenges in structural estimation: unobserved heterogeneity. We show how to extend the Hotz-Miller CCP estimation framework to handle unobserved states using finite mixture models and the EM algorithm.

Slides used in the video are available here: https://raw.githack.com/tyleransom/st...
Source code for the slides is here: https://github.com/tyleransom/structu...

Scientific reference: Arcidiacono, P. and Miller, R.A. (2011), "Conditional Choice Probability Estimation of Dynamic Discrete Choice Models With Unobserved Heterogeneity." Econometrica, 79: 1823–1867. https://doi.org/10.3982/ECTA7743

The key insight, drawn from Arcidiacono and Miller (2011, Econometrica), is that if we assume the unobserved state $s$ takes on a finite number of values (i.e., there are R discrete "types" in the population), we can use posterior type probabilities from the EM algorithm as weights when forming CCPs. This allows us to estimate the parameters of our utility function by effectively treating the unobserved type as observed through the lens of Bayesian updating.

We walk through the E-step and M-step of the EM algorithm in this two-stage CCP context, discuss options for updating CCPs (structural model vs. nonparametric vs. hybrid approaches), and flag practical considerations like bootstrapping for correct standard errors in multi-stage estimators.

Tyler Ransom is an Associate Professor of Economics at the University of Oklahoma. Subscribe for more videos on data science, econometrics, and research methods!

#dynamicdiscretechoicemodels #conditionalchoiceprobability #CCPestimation #HotzMillerestimator #HotzMillerCCP #unobservedheterogeneity #unobservedstatevariables #finitemixturemodels #EMalgorithm #expectationmaximizationalgorithm #EstepMstep #ArcidiaconoMiller #ArcidiaconoMiller2011 #Econometrica2011 #structuraleconometrics #structuralestimation #dynamicstructuralmodels #finitedependence #renewalactions #terminalactions #backwardrecursion #dynamicprogramming #Bellmanequation #valuefunction #futurevalueterms #flowutility #utilityparameters #likelihoodfunction #posteriorprobability #Bayesianupdating #priorprobability #typeprobabilities #populationheterogeneity #discretetypes #unobservedtypes #latenttypes #finitetypes #seriallycorrelatedunobservables #persistentpreferences #dynamicselection #outcomeequation #choiceequation #additiveseparability #twostageestimation #multistageestimation #firststageestimation #secondstageestimation #bootstrapping #bootstrapstandarderrors #estimationerror #nonparametricestimation #nonparametricCCP #weightedestimation #weightedlikelihood #empiricalfrequencies #structuralCCPupdate #hybridCCPestimator #machinelearningeconometrics #machinelearningstructuralmodels #discretechoicemodels #randomutilitymodels #binarychoice #multinomialchoice #paneldata #statespace #statevariables #observedstates #unobservedstates #laboreconomics #industrialorganization #publiceconomics #appliedmicroeconomics #appliedeconometrics #graduateeconometrics #PhDeconometrics #econometricslecture #economicslecture #structuralmodeling #reducedform #identification #computationaleconometrics #computationalmethods #numericalmethods #Rustmodel #busenginereplacement #KeaneWolpin #PeterArcidiacono #RobertMiller #KennethTrain #econometricseducation #economicseducation #structuralecon #econometricstutorial #graduateeconomics #researchmethods #datascienceeconomics #Stata #R #Python #Julia #MATLAB #reproducibleresearch #opensourceeconomics