I suggest an appropriate paper as desired

https://www.federalreserve.gov/pubs/ifdp/2006/869/ifdp869.pdf

What is the idea of your paper, building a model depends on the researcher what to think and how to think? This platform does not give ready research ... Greetings

Below are links to some useful papers on the subject. Having 55 companies data may be good for comparison; however, the sample length of 5 years is quite small. If you intend to analyze the data as time series, you need to expand the number of years since predictive function in time series requires large sample size to get robust forecast. If you are running simple predictive model, i.e. OLS, then 5 years may be adequate, e.g. Anderson-Darling requires n > 5.

LINKS:

https://www.google.com/url?sa=t&source=web&rct=j&url=http://statmath.wu.ac.at/~hauser/LVs/FinEtricsQF/FEtrics_Chp5.pdf&ved=2ahUKEwjRpsCFj9XeAhVPi3AKHcR4DGkQFjAAegQIAxAB&usg=AOvVaw1lNBJcizGBrRQBxJ8cALve

https://www.google.com/url?sa=t&source=web&rct=j&url=https://www.princeton.edu/~otorres/Panel101.pdf&ved=2ahUKEwjRpsCFj9XeAhVPi3AKHcR4DGkQFjAEegQIBhAB&usg=AOvVaw2urGdw2dB635zqtH5x5P0R

https://www.google.com/url?sa=t&source=web&rct=j&url=ftp://www.cemfi.es/wp/00/0016.pdf&ved=2ahUKEwj1oY-xj9XeAhVBrI8KHWE4DOk4ChAWMAF6BAgIEAE&usg=AOvVaw2uh8GhaCUH3CFDX6G5Fon8

Thanks Paul Louangrath ,

as per the data availability, I have scope to increase sample size in time series order. However, that will make my panel data set unbalanced. Can an unbalanced predictive model be developed?

Ajit kumar Roy , Thank you.

This is indeed helpful and thought provoking

The first article below (page 6) gave instruction on efficient test when the data may be unbalanced. Why would increasing sample size lead to unbalance data? Unless you anticipate missing data from come of the 55 companies, in that case then make the best of what is available.

LINK:

http://eprints.lse.ac.uk/2116/1/Nonparametric_Inference_for_Unbalanced_Time_Series_Data.pdf

https://homepage.univie.ac.at/robert.kunst/pan2010_pres_mayer.pdf

Interesting..

Absolutely try to increase the time size. Because, most of the tests in the panel require T> N.