Florian Guibelin, Loann Desboulets
Océane Piétri : oceane.pietri[at]univ-amu.fr
Morgan Raux : morgan.raux[at]univ-amu.fr
Laura Sénécal : laura.senecal[at]univ-amu.fr
Impact on labor market of replacing unemployment benefit by basic income
In this paper, using an extension of the model developed in HUNGERBUHLER et al. (2006), I compare the steady-states of the labor market with an unemployment benefit (UB) system to a universal basic income (BI) system. Agents are heterogeneous in their skills and choose to look for a job or not. Wages and working time are bargained between firms and workers. Unemployment is caused by matching frictions and the demand side of the labor market reacts to the bargaining’s outcome. Switching from UB to BI makes winners and losers among the population. A larger tax rate in the BI scheme than in the UB scheme combined with a sufficiently high BI leads to a decrease in inequalities of working time, wage and utility. Low-productive workers face higher employment probabilities and utilities while high-productive workers observe a decrease in their employment rates and utilities.
Non-linear variable selection
We propose a new general framework for variable selection using a non-parametric and unsupervised approach. Including or removing certain variables in an econometric model is a standard but difficult task. When theory is lacking then this choice has to be motivated by empirics. This issue has been adressed in countless papers for the most usual regression models but also more recently in non-linear and non-parametric models. A major difference from the literature is that our methodology is applied in an unsupervised manner, with no assumptions on causality. The very interesting aspect of the methodology is that we can handle very large datasets as we will never work on all the observations nor all the candidate variables at the same time. To achieve this goal we will use Random K-Nearest Neighbours and perform linear selection techniques to estimate a sparse local covariance matrix. Based on this any estimator for sparse covariance can be used making our methodology very general.