QUANTILE REGRESSIONs with Endogeneity: Simulations and Empirical Application

• Main Authors: Wen-Yin Chen, 陳玟吟
• Other Authors: 管中閔
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/73308224483243339143

博士 === 國立臺灣大學 === 經濟學研究所 === 99 === The analysis of regression with endogeneity has been an important research direction in econometrics. There exist the fitted value approach, the control function approach, and the instrumental variable approach to regression with endogeneity. Yet, there exists little simulation evidence for the three approaches. In the thesis, we first compare the three approaches to quantile regression with endogeneity using sumulations. Next, we apply the instrumental variable approach to analysis of the houshold''s saving behavior in Taiwan. In Chapter 1, we introduce the two-stage quantile regression, the control function, and the instrumental variable quantle regression estimation , and compare these three approaches by extensive simulations. From the simulation resutls, it is found that the performance of the two-stage quantile regression and control function estimators depends on the method used for estimating the reduced-form equation of the endogenous variable. In contrast, the performance of the instrumental variable quantile regression estiamtor is more robust to the method used for estmating the instrument. In a homoskedastic regression model with a continuous endogenous regressor, when the reduced-form equantion of the endogenous variable is estimated using a more efficient method, it is shown that the control function estimator has smaller standard errors and mean squared errors than the two-stage quantile regression and instrumental variable quantile regression estimators. Also, it is observed that when the correlation of the endogenous variable and the instrument is lower, these estimators have larger standard errors and mean squared errors. Moreover, the simulation results of the heteroskedastic regression model imply that a higher degree of heteroskedasticity yields larger standard errors and mean squared errors of the control function and instrumental variable quantile regression estimators. In Chapter 2, we provide new evidence on the relationship between the household''s saving behavior and housing price in Taiwan through the instrumental variable quantion regression approach. The correlation between household wealth accumulation and homeownership results in the simultaneous bias in the conventional mean regression or quantile regression estimation of the household''s saving function. Most of the existing literature of household saving behavior ignores the endogeneity of homeownership. The object of this chapter is to employ an estimation approach to quantile regression with endogeneity to consistently estimate the household''s saving function for both the consideration of endogeneity and heterogeneity. In the instrumental variable quantile regression estimation of the houshold''s saving function, the family structure variables are used as instrumental variables for the determinants of homeownership. The estimation results provide strong evidence on the endogeneity of homeownership decision. From the instrumental variable quantile regression estimation, when the renter has a higher saving rate than the homeowner, it is implied that the renter intends to save more for a housing purchase plan. When the renter is more inclinded to own a house, it is found that a higher housing price produces a larger discouragement effect on saving. Furthermore, the high housing price produces the wealth effect on the homeowner''s saving rate. In addition, when the renter saves more than the homeowner, it is found that the renter''s discouragement effect is larger than the homeowner''s wealth effect, and vice verse. From the constant effect test and exogeneity test, the results confirm the heterogeneity of the household''s saving behavior and endogeneity of homeownership.