Dear Mr. Josse

In my opinion you can use DEA window analysis. In some paper they usually using 3-year window analysis method to avoid “no solution” output caused by the few DMUs and sparse data.

Hope this paper will help you:

https://www.sciencedirect.com/science/article/pii/S1876610214030653

Good luck for your research !!!

Thanks a lot for your help, it's very useful!

You can either examine all DMUs in each year (1 frontier per year), in a DEA-window (1 frontier for, says, 3 years), in a pooled-DEA (1 frontier for all DMUs for all years), or as a panel (i.e. Malmquist-DEA).

You can the test for the differences between years (using nonparametric test such as Mann-Whitney-Wilcoxon) or can use Tobit regression with year as a control variable.

Thanks a lot Dr. Thanh, amazing answer!

Stata has a program that can do this. Read this:

https://www.stata.com/meeting/dcconf09/dc09_lee_ji.pdf

You can do the analysis for each year alone. I think that is better.

Apply window analysis

You also can run a annual DEA for each year

DEA can be used for both cross-section data as well as time-series data. The only difference is the interpretation of results. Hope the following link might help you to understand more about DEA

https://www.researchgate.net/publication/321824482_Data_Envelopment_Analysis