Importing your own data

Research
21

Yuval, just to add my $0.02, which sounds like a minority view. I’d probably not use it much, moreso if it came with additional cost as my subscription is pretty expensive for me already, and I presently don’t have data I’d consider uploading except perhaps FRED and other macro oriented datasets that I keep in excel - but that certainly wouldn’t be point in time as those get restated all the time and many come with huge delay. Maybe I’d just upload a combined macro “index” that I score in Excel.

Thinking about how I might play with the feature if available: I’d think maybe a feature like that would be useful for specialization. If maybe I was working within an industry I knew pretty well and if I maintained my own databases I could begin to do some things. For instance in the property and casualty insurance there are internal metrics like combined ratio, frequencies and severities, policy growth, new business growth, retention ratios, reserve development trends that companies report (often inconsistently) that are important for analyzing the business that are not in the financial statements. But that would be for a limited set of companies I was able to follow. For retail something like same store sales is useful, store opening plans, traffic, avg ticket, and if I had access to credit card data (which I don’t) that would be useful for retail. For banking things like Net Interest Margin, efficiency ratio, non-performing loan ratios, capital ratios, loan mix, loan amount, deposit amount. I suspect most industries have key internal metrics independent of public financial statements. I don’t know if I could build and upload anything useful for anything close to comprehensive back-testing though. It’d be more niche application due to my own personal limitations and need for timely maintenance.

Anyhow, just adding my $0.02. For prioritization, probably the thing that’s taking up mental space right now that I have difficulty with is proper or useful specification of pattern recognition tasks related to technical analysis. If there were algos that reliably could define support, resistance, trend breaks, and particular technical chart formations for testing - those would be interesting flags to add. I generally find those types of flags difficult to specify in code, but suspect they might have value as part of buy/sell rules.

edit: for example, I find Ian McMillan’s charting approach/insights interesting https://twitter.com/the_chart_life

22

To add my $2c as well, this is probably not something I would use in the short term.

More important on the list for me are things like:

  1. Hedge: for live stock ports and sims the possibility to have
  • A rule-based dynamic hedge ratio (currently only fixed %) e.g. weak market → 30% but “strong” bear market → 70%
  • The possibility to have more than one ETF as hedge (say 2 to 4?)
  • The possibility to use most / any ETFs we have data (or extended data) for in the database

  1. Simple / current hedge module added to ETF ports / sims (see Georg’s post above)

  2. API to access key pages for weekly rebalance (e.g. holdings, transactions etc)

  3. European / Asian data

  4. Active weighting of books (i.e. programatic / dynamic weighting of the ports in a book)

Thank you,

Jerome

23

What I would like to see is a feature where I can import my own ENTRY / EXIT signals (coming from work outside P123) via a simple Excel file for Sims and Books.
That should be easy to implement.
Any chance this will be realized?

Werner

24

The negativity toward machine learning in the forum is mind boggling. But Marco really has not joined in on that negativity.

Indeed, P123 has never made a mistake in the machine learning tools that it provides: PIT sims using ranks, quality rank performance testing with well scrubbed data which is basically a binned non-linear regression, etc

P123 has been on the cutting edge.

Looks like they will remain on the cutting edge—and eventually drag some of us along.

-Jim

25

George -

Because P123 uses only PIT data, incorporating FRED is problematic. The data you see from FRED has been revised, so it’s not point-in-time. That’s why it’s not on our to-do list. We are offering you the chance to incorporate FRED data yourself, but you won’t be able to design point-in-time models using it.

  • Yuval