Forums  > General  > Detrending using CAPM  
Page 1 of 1
Display using:  


Total Posts: 8
Joined: Jan 2019
Posted: 2019-07-28 03:55

As some of you helped me before, I am still writing my thesis and I had valuable input before!

I want to try calculating abnormal returns before specific economic events (mergers/transactions) and feed them to a neural network to test if they are recognizable.

Can I use a rolling beta estimation and substract the index effect on a stock? or is this too theoretical? I am planning to filter Beta estimations with kalman filter.

All feedback is welcome! Thanks in advance!


Total Posts: 377
Joined: Jan 2015
Posted: 2019-08-02 17:45
I'd try a variety of approaches, then use cross-validation to see which produces the best fit and how big the magnitude of difference is.

FWIW, usually a simple approach like just rolling the last three month's of linear-regressed beta works pretty well. You're not going to get very different beta estimates even with much fancier state-space models. At least for single-name stocks. The covariance structure tends to be pretty stable over time. For weird pairs like VXX/SPY you do need to be a little more clever.

Good questions outrank easy answers. -Paul Samuelson
Previous Thread :: Next Thread 
Page 1 of 1