Resumen 

Instrumental Variable methods play an important role in regression analysis when one or more predictors are endogenous. A variable qualifies as an instrument when certain assumptions are met. In particular, it must be uncorrelated with the error term of the structural equation, correlated with the endogenous variable and should not influence directly over the response variable. We evaluate a classical approach by increasing the number of instruments. Furthermore, a simple alternative to recover the endogenous parameter is proposed based on a two step optimization process using a modified version of the Anderson-Rubin test and the variance of the estimated errors. Finally, a new sensitivity analysis approach is proposed when the exclusion restriction assumption is suspected not to hold. The proposed method is based upon a Nonparametric Bayesian Approach.