# Rolling autoregression coefficient

I was reading a paper and I saw that they run a 3-year rolling autoregression for 20 years (using for example 2013-2016 as trailing and 2016-2019 as forwarding) and got only one beta coefficient and one t-value. I've been looking for a way to do that, but haven't found anything. So, my question is how it's possible to do that? For reference, it's a pooled regression if it makes any difference.

Will appreciate any help!

Edit: For one variable, it's possible to do the rolling regression recursively and apply a filter (Kalman filter for example) and get one coefficient that describes the connection between the variables. However, I couldn't figure out how I could do the same thing for pooled regressions. I think I need to find some mix between the fixed effects method and the recursive rolling method.

• Can you reference/link to that paper? Jun 22 at 18:18
• Of course, here it is. My question is particularly about exhibits 18-21. I actually found a way to get one coefficient in the case of one fund, using recursive rolling regression and applying some filter to that. I am still not sure how I can apply the same idea to the pooled funds. alliancebernstein.com/content/dam/global/insights/… Jun 24 at 1:44