A0202
Title: Migration choice models: An approximate Bayesian computation alternative to traditional approaches
Authors: Gabriele Romano - University of Chicago Booth School of Business (United States) [presenting]
Abstract: An approximate Bayesian computation (ABC) method is proposed as a flexible alternative to analyzing migration choices across countries. Traditional approaches like multinomial logit and gravity equations rely heavily on restrictive distributional assumptions, particularly the Type-I extreme value distribution for the random utility component. Through Monte Carlo simulations, it is demonstrated that these conventional methods exhibit poor frequentist properties under model misspecification, with coverage probabilities of confidence intervals often dropping to zero. The proposed ABC method offers several advantages: It requires fewer distributional assumptions, accommodates arbitrary correlation structures, and produces full posterior distributions rather than just point estimates. Using simulated migration data calibrated to European countries, it is shown that the ABC approach achieves better mean squared errors for some economic variables and provides more reliable coverage probabilities, though at the cost of wider credibility intervals. While the method's flexibility makes it an interesting alternative for researchers concerned about model misspecification, its computational intensity and sensitivity to tuning parameters present practical challenges that may limit its applicability.
