生产率与效率分析lecture8-随机前沿效率1.ppt

Estimating Mean TE Suppose u i.i.d. half-normal E[exp(-ui)] = 2[1 - ?(? ? ? )] exp(- ??2/2). Substitute ML estimates of ? and ?2 into this This only gives mean TE across the sample Often want to know TEi Estimating Firm-Specific TE Jondrow, Materov, Lovell & Schmidt (1982) Battese and Coelli (1988) TEi = [(1- ?(?A+ ?ei / ?A) / (1- ?(?ei / ?A)] x exp(?ei + ?A2 / 2) where ?A = ? [?(1 - ?) ?2 ] and ei = lnyi - xi ? Substitute ML estimates of ?, ? and ?2 into this Testing Hypotheses Important to test for presence of technical inefficiency (TI) Do not impose TI by estimating SF model without testing for TI Options Wald test LR test Testing for TI (cont.) Options Wald test LR test H0: no TI in sample HA: TI in sample Wald test found to reject H0 20% of time when H0 false only 5% of time - Coelli (1995) poor size (prob. of Type I error) properties The Stochastic Frontier Analysis (SFA) The Stochastic Frontier - History Farrell (1957) 2 measures of firm efficiency technical efficiency (TE) obtain maximal output from given set of inputs allocative efficiency (AE) use inputs in optimal proportions, given input prices History (cont.) Recall technical efficiency (TE) allocative efficiency (AE) Focus on TE only For simplicity in explanation of stochastic frontiers (SF) History (cont.) To calculate efficiency must know production function of fully efficient firm Never know it Farrell estimate fully efficient production function History (cont.) Estimate fully efficient production function Two ways 1. non-parametric piece-wise linear technology Charnes, Cooper others DEA (not Drug Enforcement Agency!) ignore in this session History (cont.) Estimate fully efficient production function Two ways 1. non-parametric piece-wise linear technology Charnes, Cooper others DEA (Data Envelopment Analysis) ignore in this session History (cont.) Estimate fully efficient production function Two ways 2. parametric function Aigner,Lovell and Schmidt, others stochast