This is a dynamic panel estimator for a production function in which
- Inputs are correlated with unobserved productivity
- Productivity evolves as a nonlinear (quadratic in current version) Markov process
- No input is 1-to-1 with productivity, unconditionally or conditional on other inputs (i.e. traditional approaches to inverting out productivity are misspecified)
For now, download _panelpf.ado. For output y, inputs x, the basic syntax is panelpf y x, start(name) where start() contains a row matrix of starting values for the production function and quadratic Markov process. This syntax assumes you want to use lags of x as instruments and simple default gmm options. You can also add gmm_options() which contains a different set of instruments for x and other options to pass to Stata’s gmm command. Either start() or gmm_options() with an initial value must be specified.
Note: as described in the paper, the estimating equation for this command is in terms of f.y and f.x (i.e. everything is shifted forward a period). If using custom IVs, shift them accordingly (see example file).
cobbDouglasExample.do: This example simulates a model of production in which
- all inputs choices are correlated with productivity
- some inputs are chosen based on current productivity
- these relationships are not invertible, and
- productivity follows a quadratic Markov process
Then I estimate the production function according to my procedure, OLS, and IV regressions. For comparisons to the production function estimator by Ackerberg, Caves, and Frazer (2015), see my paper titled “Estimating Productivity and Markups Under Imperfect Competition”.
- Near future: Allow for higher polynomial orders in productivity Markov process