Costin Protopopescu

Faculty ,  Aix-Marseille Université , Faculté d'économie et de gestion (FEG)

Econometrics, Finance and mathematical methods
Protopopescu
Status
Assistant professor
Research domain(s)
Econometrics, Environmental economics, Finance
Thesis
1997, Université Toulouse I
Contact
costin.protopopescu[at]univ-amu.fr
+33(0)4 13 55 25 68
Address

AMU - AMSE
5-9 Boulevard Maurice Bourdet, CS 50498
​13205 Marseille Cedex 1

Abstract Recurring statistical issues such as censoring, non-random selection and heteroskedasticity often impact the analysis of observational data from natural and human processes. We investigate the potential advantages of models based on quantile regression (QR) for addressing these issues, with a particular focus on non-market valuation data. First, we provide analytical arguments showing how QR can tackle these issues. Second, we show by means of a Monte Carlo experiment how censored QR (CQR)-based methods perform compared to standard models with selection both accounted for and not accounted for in the modeling. Incidentally, we propose an alternative to the standard estimation procedure for the CQR model with selection, which divides computation time by about 100. Third, we apply these four models to a French contingent valuation survey on flood risk. Our findings suggest that selection-censored models are useful for simultaneously tackling issues often present in observational and human data. In addition, the CQR models give a better picture of the heterogeneity of the coefficients, but the computational complexity of the CQR-selection model does not seem to be offset by better performance.
Keywords Selection model censored quantile regression Monte Carlo experiment nonmarket valuation flood
Publication HAL Google Scholar
Abstract Recurring statistical issues such as censoring, selection and heteroskedasticity often impact the analysis of observational data. We investigate the potential advantages of models based on quantile regression (QR) for addressing these issues, with a particular focus on willingness to pay-type data. We gather analytical arguments showing how QR can tackle these issues. We show by means of a Monte Carlo experiment how censored QR (CQR)-based methods perform compared to standard models. We empirically contrast four models on flood risk data. Our findings confirm that selection-censored models based on QR are useful for simultaneously tackling issues often present in observational data.
Keywords Censored Quantile Regression, Contingent valuation, Flood, Monte Carlo Experiment, Quantile regression, Selection Model, Willingness to pay
Download HAL Google Scholar
Abstract We propose a new correlation measure for functionally correlated variables based on local linear dependence. It is able to detect non-linear, non-monotonic and even implicit relationships. Applying the classical linear correlation in a local framework combined with tools from Principal Components Analysis the statistic is capable of detecting very complex dependences among the data. In a first part we prove that it meets the properties of independence, similarity invariance and dependence and the axiom of continuity. In a second part we run a numerical simulation over a variety of dependences and compare it to other dependence measures in the literature. The results indicate that we outperform existing coefficients. We also show better stability and robustness to noise.
Keywords Pearson Coefficient, Non-Parametric Statistic, PCA, Implicit Dependence, Non-Monotonic, Local Correlation, Non-Linear
Download HAL Google Scholar