When are Stock Factor updates recognized?

Manage
21

Yuval, That behavior matches what I’ve been experiencing. My updates usually happen on Saturdays and are recognized as expected. But the last update was done on Sunday.and hasn’t appeared yet. By that I mean I did a manual rebalance today and the Sunday update is still unrecognized.

Walter

EDIT: unrecognized in Live Strategies. I think the screener is processing the universe as expected.

23

Sorry for the massive delay here. The task fell into backlog for a while, but we finally got around to improving the treatment of stock factor for current and backtest use.

Here’s the breakdown of its expected behavior now:

  • A current universe, ranking system, screen, or rebalance can access today’s data.
  • A historical universe or ranking system can access Sunday data.
  • Historical screen rules should be able to see data up to the as of date itself.
  • A daily screen backtest or simulation should be able to see data one day prior to the weekday (Sunday - Thursday).

Let me know if you experience further difficulties here, @WalterW.

24

Thanks Aaron! I’ve noticed the recent change and it looks good!

25

So I do not understand this with relation to how anyone doing any serious machine learning can use this. Just a question as I have not started rebalancing a machine learning port recently at P123. I think I did it this previously by copying and pasting from the P123 screener page to get more the 500 rows of ranking data from the screener as the downloads were limited then too!!!

P123 would not address any serious requests from people wanting to do machine learning then—including all of the ranking data for a factor in the screener (I asked). Which made serious machine learning ALMOST impossible at the time.

Copy and pasting for the P123 screener was always fraught with formatting issues in Excel. Sometimes you could not stop it from pasting into HTML. But P123 gets a fair amount of revenue from people doing automated quantitative systems now (whether you want to call it machine learning or whatever). I assume things have changed at P123 and DataMiner can be use for this?

I will be developing a machine learning system using a DataMiner download assuming I can get the data I need on the day of the rebalance. I can only assume P123 wants to attract machine learners and already has enough machine learners that P123 has already addressed this.

But to do any serous machine learning with P123 data, I need the most updated set of ranks for each factor on say a Thursday (not Sundays data but the overnight run on Wednesday). I then run the machine learning prediction (using an already trained model) using the most recent factor ranks and then I sort the predicted returns. I buy the 25 stocks with the greatest expected returns buy putting them nto in-list and let P123 rebalance (selling stocks that no longer have high expected returns).

I do not know of any other reasonable way to do machine learning at P123 now other than using the machine learning to develop a ranking system which is what I am funding now.

I assume this has already been addressed this and there is a user-friendly answer. I am not questioning that.

So I guess just a question as to the best way to do this. Is it:

  1. Does the screener not have a 500 stock limit for the most recent ranks (requiring multiple screen runs and download for each factor (but a possible method).

Edit: the 500 ticker limit download limit does seem to be in effect for ta screen run today and not just historical data!!!

  1. can the most recent ranks (i.e., late run on Thursday morning) be accessed thru DataMiner (just one download)? And I assume P123 managed the pricing so the most P123 member do not have to pay more for DataMiner access to run daily ports.

I assume you probably can use DataMiner for this but maybe it is meant for historical weekly data (a question)? Maybe here on August 1st I can test that but I would prefer to use my credits elsewhere this month as I am not rebalancing a machine learning port today.

I am sure P123 already has a simple answer for this simple question. It has probably been addressed already—addressing concerns I have not even considered.

Thank you in advance for addressing my very basic question. I apologize for not having tried it on DataMiner myself already (see above for data limitations involved).

Jim

26

So this (with today’s date) would be ideal in DataMiner as it is what I have been using to download the data for my system. Just continuing it with today’s date would be intuitive. Kind of like turinng a sim into a port (a good idea that P123 already uses).

I already has all of the factors but today’s date did not give any data:

Main:
Operation: Ranks
On Error: Stop # ( [Stop] | Continue )
Precision: 4 # ( [ 2 ] | 3 | 4 )

Default Settings:
PIT Method: Prelim # ( [Complete] | Prelim )
Ranking System: ‘RankSystem’
Ranking Method: NAsNeutral
Start Date: 2023-08-01
** End Date: 2023-08-01**
Frequency: 1Week
Universe: Easy to Trade US
Columns: factor #( [ranks] | composite | factor )
Include Names: true #( true | [false] )
Additional Data:
- Future 1wkRet: Future%Chg(5)
- FutureRel%Chg(5,GetSeries(“$SPALLPV:USA”)) #Relative return vs $SPALLPV:USA

Also 2023-07-31 did not give any data.

Is there something else I can use that gives me all of the factors at once (with current data)?

So probably I can run a screen in DataMiner without the 500 limit now (maybe 150 of them without reaching any limits). But run 150 screens and sort the data to get the tickers lined up? I am sure P123 has been working on this and has another DataMiner download I am not aware of yet.

Thank you in advance for advice on the best method(s) that P123 have developed for people wanting to automate some of this (not just machine learners, I think).

Jim