Levinsohn-Petrin TFP estimation in Stata without proxy I'm trying to compute TFP by using levpet Stata command. I use revenue estimation, yet it requires proxy option which is intermediate input (usually material resources used for goods/services production (as far as I understand)). Unfortunately, I don't have this data as is in my table. And other authors when referring to usage of levpet write that they use hours worked and capital only. The question is whether there is a way in Stata to leave proxy blank (by dummy column or something like that) or may be I can use one of the columns I have as in intermediate input. I have the following fields in the table: various types of liabilities, depts, ebitda, operating expenses, sales, cost goods sold, profits, investment cash flow, various assets, operating revenue turnover, revenue.
I thought may be it has sense to use cost of goods sold or investments as intermediate input for the proxy option of levpet. 
 A: You can use investment as the proxy. Levinsohn & Petrin (2003) was an extension to Olley & Pakes (1996). Olley & Pakes used investment as the proxy to form the control function. A key assumption is that the proxy is a monotonic function of unobserved productivity. A potential problem with using investment is that it is often 0, and so cannot be a strictly monotonic function of productivity.
If you haven't already, you really should read Levinsohn & Petrin (2003), Olley & Pakes (1996), and Ackerberg, Caves, & Fraser (2006). Alternatively, Van Beveren (2012) is a pretty good review article about TFP estimation.
A: The issue that the econometrician is trying to solve is unobserved productivity. In other words when working with data we note a change in productivity say it increases but we don't know why. We could solve the problem by asking firm managers what they did but hey, who has the time and resources. Besides, different managers respond to similar shocks differently - increasing (permanent/temporary) workforce, offering more bonuses..the list is endless. However, OP come along and say wait a moment.. When responding to shocks firm managers realise a change investment. So continuting with our example, if from data we see an increase in investment it is because the manager is responding to a productivity shock. They then provide a semiparametric method of how to do this. 
Later, LP say that at times the demand curve for investments may have a kink . In other words it makes no sense for a manager to increase investment because that will not translate to more market share. Neither will a reduction in investment since that would result to declining sales. Such managers would prefer to increase run-time of machines by for instance by employing two shifts instead of one, leasing more machines etc. Whatever the case that will mean increased energy use but no recorded increases in investment for the period. The intuition here is that LP captures effects that OP fails to capture in addtion to correcting those that OP corrects for. 
So I am afraid you cannot leave the proxy blank precisely because this is the raison d'etre of the levpet. You are better of using OLS. The precise variable to use depends on the composition of firms in your dataset. LP mention electricity which is appropriate for manufacturing firms but this variable may be off in a dataset of transportation companies where the expenditure of fuel is more realistic, investment cash flow would do for banks etc. 
