Edge/alpha, of course, has greatly lessened over time; Updated run in last 10 years I suspect is 1/2 the edge, and last couple even worse - such is the ever evolving alpha landscape.
This process could of course be extended and improved - love to discuss with you if you have time/interest.
great piece and thank you for sharing. I guess the best way to regularise the unstable results of the Trend Factor approach is to use a Lasso regression on the trend factors, rather than a least-square one. Ordinary regression curve-fits too much and cross sectional returns are full of noise, hopefully the Lasso helps the Trend Factors really stand out
Fantastic work, thanks for sharing.
Edge/alpha, of course, has greatly lessened over time; Updated run in last 10 years I suspect is 1/2 the edge, and last couple even worse - such is the ever evolving alpha landscape.
This process could of course be extended and improved - love to discuss with you if you have time/interest.
Hello,
great piece and thank you for sharing. I guess the best way to regularise the unstable results of the Trend Factor approach is to use a Lasso regression on the trend factors, rather than a least-square one. Ordinary regression curve-fits too much and cross sectional returns are full of noise, hopefully the Lasso helps the Trend Factors really stand out
Thanks!! That's a great idea, will try it out!!
Awesome, do you think it's feasible to code in Real Test?
Thanks! I'm pretty confident there must be a way to do it in RealTest (Marsten is great!), although I'm not a specialist. Maybe ask in the RT forum?
Great post!
Thanks!!!
Great post! I appreciate the time and effort you put in to coding this. It’s an interesting concept.