Using DEA efficiency scores as an independent variable and Simar & Wilson (2007)

[Job Koning]

Hi all,

I'm interested in evaluating the determinants and effects of DEA efficiency scores. I encountered a paper (Simar & Wilson 2007) which indicates it is not apppropriate to use efficiency scores obtained using DEA as a dependent variable in regressions, and provides an alternative method to achieve this (implementable in STATA using the simarwilson command from Badunenko & Tauchmann (2019)).

My question is whether this also applies to methods where DEA scores are used as an independent variable? For instance, I'm interested in the effects of bank efficiency (Measured with DEA) on bank risk taking. What method would be appropriate to measure this and how would one implement this in Stata?

Thanks in advance reader!

[Jared Greathouse]

Email my advisor, he's the Big Fish in the DEA field.

[Harald Tauchmann]

Dear Job,
I would say "probably" yes. The point that Simar & Wilson make in their 2007 paper may also imply problems with using DEA scores as right-hand-side variable. Their key argument is that DEA scores are biased estimates of true but unobservable efficiency. Thus, if true efficiency is what you want to have on the right-hand side of your model, using DEA scores as a substitute means using an explanatory variable that is measured with error. Because of the bias that DEA scores suffer from, this is not classical measurement error. So, I am not sure that the argument that the attenuation bias is always towards zero and you still identify a lower bound effect applies in this setting. If you feel more comfortable with a classical measurement setting, using a bias-corrected measure of efficiency (e.g. as suggested by Simar & Wilson [2007] for algorithm #2) may help. However, to be honest, I would be more concerned that bank efficiency - regardless of how it is measured - is as much an outcome as bank risk-taking, making the estimated coefficient capture a correlation rather than a causal effect.
Best wishes,
Harald

[Job Koning]

Dear Harald,

Thank you for your detailed explanation, I really appreciate it. I ended up dropping this approach from my assignment for the reasons you listed and focussed on other subjects.

Regards,

Job