a robust semi-parametric estimation of determinants of inefficiency in iranian banking system: a bootstrap method

many empirical papers have applied a semi-parametric two-stage procedure named tobit model to investigate the sources of inefficiency in different industries over the last two decades. using this approach in small samples has recently been criticized for a possible bias in its results. in very recent papers simar and wilson (2007) have tackled this problem by suggesting an alternative bootstrap two-stage model in two algorisms including single and double bootstrap procedures. these algorisms both permit valid inference. moreover, the double bootstrap procedure provides bias corrected efficiency estimates. the current study as one of the first inefficiency determinant studies over banking system applying simar and wilson (2005) procedure is to provide new empirical evidence of this approach. to this aim a dataset of iranian banking system over 2002-2007 is utilized. the outcomes support simar and wilson (2007) in which the results of the conventional two-stage model are likely biased. more precisely, the double bootstrap shows quite different estimates for the parameter and confidence-intervals than the single bootstrap indicating that the bias and autocorrelation are significant problems in the two-stage model. the comparison of the results of double bootstrapped two-stage model and the single- stage panel data model (coelli 1995) performed for the first time in this study exhibits a close similarity in the determinant factors of the technical inefficiencies and their confidence intervals. this finding support coelli (2000) in which the single-stage panel data model provides unbiased and efficient estimates of determinant factors of the inefficiencies and their confidence intervals. the findings also suggest that iranian banks, by and large, would improve their efficiencies by moving toward privatization, holding less risky assets and last but not least focusing on economies of scale and economies of scope through broadening their sizes and products. jel classifications: c01, c02, c12, c19, c23