regression method for data that contain a number of observation for several years

Your data structure is a panel, i.e. repeated observations of the same entities (companies in your case). Pooling all observations in a standard linear regression and ignoring the fact that many observations are repeated will lead to biased estimates. The most straightforward method for your case is a Fixed Effects Estimator. Check your statistical package on how to do this. The idea is to do a linear regression on the company-specific mean values of dependent and independent variables. This way, you implicitly control for all time-constant firm characteristics.

Hang on - so, what is the unit of analysis in your data set? Is it companies, or is it managers? Think carefully about this. How are the "top 5" ranked? Does the cohort change frequently? What variables are you interested in with respect to policy impact?

It's true that from a company view, you've got a panel, but if there's enough churn in the management roster you could justify a pooled approach. In addition, if the missing data is non-random (e.g., cyclic, confined to certain firms/sectors/etc), then it would bias a panel approach, and a pooled regression may be the best you can do.

It's down to the particulars of your data, which only you have access to.