Index | Recent Threads | Who's Online | Search

Posts: 373    Pages: 38    Prev 29 30 31 32 33 34 35 36 37 38
Last Post
This topic has been viewed 19085 times and has 372 replies
Re: FactSet beta site v1.0, NOW LIVE

Jim - I only work with sales surprises. I have totally abandoned earnings.



Thank you for the tip on sales surprises.

My only advice would be to put ALL of the factors a person thinks are good into a node. Perhaps grouped and weighted rationally (by something like principle component analysis).

And if the factor does not have a rational explanation—and you are thinking of using it--the backtest need needs to be really good: highly significant. Well beyond the usual level of significance.

Maybe don’t even use it then. Maybe it depends on how crazy the idea is, how long the backtesting period is, how effective it is (the magnitude of the effect), its effect on turnover etc.

BTW, Renaissance Technologies is said to use a factor (the example was for a technical indicator) that does not have a clearly rational explanation if the p-value is less than 0.01. This is from: The Man Who Solved the Market: How Jim Simons Launched the Quant Revolution. I am not recommending the indiscriminate use of that as a cut-off but it is more significant the usual p-value < 0.05.



From time to time you will encounter Luddites, who are beyond redemption.
--de Prado, Marcos López on the topic of machine learning for financial applications

Jul 19, 2020 11:34:44 AM       
Edit 17 times, last edit by Jrinne at Jul 19, 2020 12:14:58 PM
Re: FactSet beta site v1.0, NOW LIVE

Jim - the earnings growth formula (or earnings estimates) probably worked because it was zero-earnings centric with a bias towards improvement in earnings. This iikely flag stocks ready for explosive growth. This is probably great for cyclical industries. But I have to ask whether the growth formula is the best formula for capturing that phenomenon. I don't know the answer. I do know there are issues with the growth formula that we have pointed out. I use (lowest) Std(NetIncome)/Std(Sales) to accomplish something similar for s/w stocks but it may not be a great general purpose factor.

I have no problem chasing what "works". All factors seem to drift in and out over time and that should really be expected from the markets. The main question is how long they persist and how to recognize when they are not working.

Jul 19, 2020 2:30:30 PM       
Edit 1 times, last edit by InspectorSector at Jul 19, 2020 2:31:52 PM
Re: FactSet beta site v1.0, NOW LIVE

Dear Inspector Sector,

Could you say more, please, about the following statement... "The problem with the previous vendor was that they claimed that it was PIT but they didn't freeze the data until Monday morning, too late for P123 to process. So it may have been historically correct for them but not for P123."

Also, just to be safe, I assume you are referring to CompuStat?

Thank you very much.


Aug 5, 2020 11:35:52 PM       
Posts: 373    Pages: 38    Prev 29 30 31 32 33 34 35 36 37 38
 Last Post