Time Series Data for Business Performance: Regression Model?

Thanks in advance for the help. I'm gathering data describing company marketing activities, and trying to model how this impacts subsequent company performance. There are about 10 data points for each company, all in a time-series (month to month tracking data points). For example, I might have "Media Coverage" as one variable, which would be a count of company media coverage that changes month-to-month.

The dependent variables (company performance indicators) will also be time-series data (say, stock price or earning report data).

In total, I expect to have data for at least 25 companies. My plan is to try and create a regression model for this, so see which variables have a meaningful impact and to what extent.

I'm familiar with basic linear regression models. Can you point me to an approach for this problem?


Fortran must die
I have tried to learn time series regression for 8 years and the best I can say is that there is no easy way to do this and no book that I would recommend to someone new to it. None of the forms of regression developed for it are simple.

You might look for regression with ARMA error which is probably the simplest approach. You need at least 50 periods of data (and if seasonality is part of your model that means 48 months at least).
Thanks Noetsi. Glad to know that it's not just me that finds this difficult!

I only have 18 periods of data, covering 18 months, so ARMA might not work. I'm going to start with something much simpler, and do some analysis of cohorts as groups. Thanks again.


Not a robit
Are you pooling data from companies or running multiple models? Can you post some comparable toy data, so we can have an idea before proposing anything.


Fortran must die
If you mean you have only 18 independent months (that is 18 data points) than you don't want to do time series period. Among other problems its impossible to capture seasonality that way.