Theses#
Here is a list of Master’s theses submitted by students of economics from the University of Bonn.
Badini, S. (2021). On the empirical identification of time preferences in discrete choice dynamic programming models.
The thesis assesses the practical identification of the time preference parameters in a discrete choice dynamic model of occupational choice, after introducing empirically-motivated exclusion restrictions that influence the size of the agents’ choice set. In particular, agents may or may not face future employment restrictions that depend deterministically on their educational choices: Future-oriented agents take the restrictions into account when deciding on their level of education. The thesis sheds light on time preference parameters, which are underidentified in structural models of dynamic discrete choices. Underidentification is especially problematic for counterfactual analysis, since information on time preferences are needed to make any statement about the behavioral response of the agents to a policy intervention. In the literature, both theoretical arguments for identification and empirical identification strategies exploit, with mixed results, variables that leave the per-period utility function unaffected,This while being relevant to the agents’ decisions. The intuition is that comparing the behavioral response of similar agents to different (expected) futures may reveal information on their time preferences.
Contact: @SofiaBadini
Kauf, B. (2020). Uncertainty propagation in the predictions of economic models.
The thesis reconstructs and extends the original data of Keane and Wolpin (1997), compares its predictions to actual values and investigates the impact of uncertainty in the parameters on relevant quantities of interest. While some of the decision rules regarding the creation of the data set have to be infered, I manage to reasonably replicate the original data set up to 1987. In contrast, the predictions of the paper compare less well to the self-constructed data set as they overpredict both the share of individuals in the blue-collar sector and the average annual incomes of blue-collar workers and underpredict the share of individuals choosing the home option. A potential shortcoming of the model which leads to these differences is the missing incorporation of demand-side factors which have been shown to substantially affect occupational employment shares and incomes. As many of the individuals choosing the home option spend an extensive amount of time out of the labor force, factors that force individuals out of the labor force but are not included in the model such as health problems and incarceration have been identified as potential extensions of the model which could improve its predictive performance. Using a simulation approach, I find the estimated effects of an annual $2000 college tuition subsidy on total years of schooling and high school graduation rates to be robust to the introduction of uncertainty in the parameters whereas the robustness of the estimated effect on college graduation rates appears less clear and might be underestimated.
Contact: @bekauf
Maokomatanda, L. (2020). Sensitivity Analysis for Structural Microeconomic Models using Shapley Values.
This thesis computes the Shapely values for the estimated model parameters in Keane and Wolpin (1994) to assess their relative quantitative importance for the model’s counterfactual predictions. The parametric uncertainties in the analysis of structural microeconometric models are ubiquitous. However, sensitivity analysis is rare and particularly challenging in this setting due to many correlated parameters. Shapely values with their foundation in game theory appear particularly suited in this case.
Contact: @lindamaok899
Mensinger, T. (2020). On the Use of Surrogate Models in Structural Microeconometrics.
The thesis applies the ideas of surrogate modeling to the human capital model by Keane and Wolpin (1994) to facilitate the application of uncertainty propagation and sensitivity analysis. The analysis demonstrates that surrogate models do not outperform a brute force approach when propagating the model’s parametric uncertainty to selected quantities of interest to economists. However, surrogate models are essential to also conduct a subsequent sensitivity analysis and machine learning methods are useful to calibrate a surrogate to high precision.
Contact: @timmens
Gehlen, A. (2020). Simulation-Based Estimation of Discrete Choice Dynamic Life-Cycle Models.
The thesis revisits the models by Keane and Wolpin (1994, 1997) to explore Method of Simulated Moments (MSM) estimation as an alternative to the Simulated Maximum Likelihood (SML) approach used by the authors. The thesis discusses the various calibration choices needed to construct an appropriate MSM criterion function for estimation, as well as the challenges that come with optimization of the criterion. The analysis demonstrates that the MSM can be effectively employed for model estimation but simultaneously shows that results are very sensitive to calibration choices.
Contact: @amageh
Suchy, R. (2020). Robust Human Capital Investment.
The thesis enriches the treatment of uncertainty in discrete choice dynamic programming models applied to human capital investment decisions. It extends the Keane and Wolpin (1997) life cycle model by a deliberate account of ambiguity. Individuals fear model misspecification and seek robust decision rules that work well over a whole range of models about their economic environment. The literature on robust optimization informs the implementation of a robust value function iteration procedure that induces robust policies. The resulting robust human capital model is calibrated to a sample of young men from the National Longitudinal Survey. The empirical validation of the model testifies its capability to appropriately match behavioral patterns in the data, to predict occupational choice decisions, and to improve the out-of-sample fit compared to a risk-only model. It carries novel implications for the ex-ante evaluation of educational policies. In case of a tuition subsidy, predictions from a misspecified risk-only model may embellish their efficacy. Given a particular tuition subsidy, a targeted intervention that reduces the level of ambiguity may increase the demand for education.
Contact: @rafaelsuchy
Stenzel, T. (2020). Uncertainty Quantification for an Eckstein-Keane-Wolpin model with correlated input parameters.
The thesis analyzes the uncertainty of the effect of a 500 USD subsidy on annual tuition costs for higher education on the average years of education caused by the parametric uncertainty in the model of occupational choice by Keane and Wolpin (1994). This model output is called a quantity of interest (QoI). The uncertainty quantification (UQ) has two stages. The first stage is an uncertainty analysis, and the second stage is a quantitative global sensitivity analysis (GSA). The uncertainty analysis finds that the tuition subsidy has a mean effect of an increase of 1.5 years and a standard deviation of 0.1 years in education. For the qualitative GSA, I develop redesigned Elementary Effects based on Ge and Menendez (2017) for a model with correlated input parameters. Based on these Elementary Effects, I compute multiple aggregate statistics to quantify the impact of the uncertainty in one parameter on uncertainty in the QoI. However, the analysis does not lead to clear results as there is no consensus about how to interpret the aggregate statistics in this context - even for uncorrelated parameters.
Contact: @tostenzel
Massner, P. (2019). Modeling Wage Uncertainty in Dynamic Life-cycle Models.
The thesis sheds light on the validity of the modeling assumptions of wage uncertainty in dynamic life-cycle models of human capital investment. Since the pioneering work of Keane and Wolpin (1997), a majority of studies in this field followed the approach of assuming serially independent productivity shocks to wages. In the case of Keane and Wolpin (1997), the independence assumption indeed simplifies the numerical solution of the model compared to more complex specifications, such as serial correlation. However, the assumption of i.i.d. productivity shocks seems to be quite narrow in light of findings of the reduced-form literature stream on wage dynamics.
Contact: @PatriziaMassner
Raabe, T. (2019). A unified estimation framework for some discrete choice dynamic programming models.
The thesis lays the foundation for respy to become a general framework for the estimation of discrete choice dynamic programming models in labor economics. It showcases the ability to represent Keane and Wolpin (1994), Keane and Wolpin (1997), and Keane and Wolpin (2000) within a single implementation.
Contact: @tobiasraabe