AI Factor launches in Beta

Latest Release
22

Please add me to the test group. Thanks.

23

I had an error for some features (factors), where the message that for certain time periods all values (for all stocks) were NA. Had to exclude those from the list.

For example, I get this with FCFEstCY / EV.

I see that when I run this in a ranking system I do get an output, but also an error message saying to contact support.

24

This is designed to avoid training with an overly preprocessed feature. FCFEstCY only started getting populated around 2006. If you want to include it, you could start training from a more recent date, such as 2006-2007.

2 Likes
25

Sign me up.

26

I would like to be added if there is still room.

Thanks,

Daniel

27

I would like to request access to the ML beta on P123. I am very curious to see how it works.

Thank you in advance.

28

I'd like to be a beta user!

29

All set for latest requests. Please logout then back in

1 Like
30

I logged out and logged back in, but I still don't have access.

31

I'm interested!!

32

Want to try the beta.

33

Hello P123 Team,

I can see the great work behind the new release. In the next few days I will be playing with the tool so I can get used to it.

In my particular case, I prefer to make the models on my own and adjust the parameters "a posteriori" of the models that went best. I also prefer to do the predictor filtering on my console with the data. Is there any way to download the dataset used for training the models?

Here is an example:
Let's say I have "MktCap" as a factor and normalized it using a Z-Score. It would be very useful for me to download a dataset with the following data:

  • Company
  • Date
  • Sector or Industry
  • Normalized factor (in this case "MktCap")
  • Target variable (or even its quantile)

This would be incredible for research in the area of factor investing and even factor timing, especially because it could be used for academic studies.

Thank you in advance.

Best regards

34

Latest requests all set. Please note that an Ultimate membership will be needed at some point to continue using AI Factor.

Did you try TOOLS→Download Factors ? You can only download normalized values w/o your own FactSet license. This makes it tricky to use for prediction with Z-Score normalization at the dataset scope level (instead of by date) since you do not have to access to the mean & stddev of each feature.

So, if you must have Z-Score normalized at the dataset level, there are few workarounds:

  • Obtain a FactSet license (probably ~15K/year) and download raw data
  • Download entire dataset every time when you are doing predictions.
  • Use a workaround mentioned here where you can get away with just downloading with one week overlap

Hope this helps

35

Please add me to the test group. Thanks.

36

Thanks!! I found the way to dowload the data!

37

All set

38

I'll give it try - not confident I will understand though :weary:

39

I'm interested in the beta as well. Do you need to be an ultimate subscriber to try it? (thinking try before buy)

40

Very interesting. Could you also add me for beta testing? thanks

41

I would also like to test it. Thanks