Partial Identification of Power Plant Productivity

• Departmental Seminar

Date and Location

Wednesday, January 18, 2017, 10:30 AM - 11:50 AM
ARE Library Conference Room, 4101 Social Sciences and Humanities

Abstract

Traditionally, productivity, the part of output that is not a result of observed input use, is modeled as a controlled or uncontrolled Markov process. I allow it to be a function of unobserved choices; plants differ in their productivity because they do different things, not only because they receive different shocks. If we think giving incentives to plants to be more productive will cause them to be so—if we think policy can affect productivity—then we must have a model in mind where productivity is a choice or else plants can not respond to the incentive. When we allow it to be a choice, the problem of identifying productivity becomes more difficult because productivity and inputs are now chosen on the basis of the same underlying state variables. No instrument can shift input choice without shifting productivity choice. But a broad class of economic models predict productivity and inputs are positively related in a certain statistical sense even when productivity is a function of unobserved inputs or other choices. I show this economic theory can meaningfully partially identify the effect of policy on productivity. I use the theory to bound the effect of restructuring in the electricity industry on power plant productivity, finding the policy lowered power plant productivity and arguing that an approach that allows productivity to be chosen by the power plant is necessary to estimate the effect of restructuring on power plant productivity.