Hi Jamie,

Thank you for this post. A few more questions from me:

• Is there a 'hold-out' period for the FactSet fundamental dataset for Q Community users, and if so, how long is it?
  
• In general, do you believe the FactSet fundamental dataset is of better quality than the Morningstar fundamental dataset?
  
• Does FactSet fundamental dataset contain more data fields (than Morningstar), for example trailing-twelve-month data, rather than requiring us to calculate this ourselves in a CustomFactor?
  
• Will each new market that you roll out also have FactSet fundamental data available from day-1, or 'only' historical price & volume data?
  
• Would you be willing to share what some of the other FactSet datasets are that you may consider rolling out after Fundamentals? For example, earnings surprises, mergers & acquisitions announcements, sentiment data, etc?
  
• How far back will data be available for new markets and FactSet datasets? Since 2002 the earliest, as with US Equities?
  
• Do you reckon there will be a decent 'learning curve' with rolling out new markets and new datasets? For example, if it takes about 6 months for the first market and first new FactSet dataset (fundamentals), it may only take another 3 months for the second market/dataset, and another 2 months for the third, etc?

Thanks again! I'm very much looking forward to both new markets and new datasets!

Thanks Jamie for the post, and great move to enable access to markets outside the US on the Quantopian platform!

As international market data will be available in Pipeline, will Quantopian be curating the "local universe" viz. QTradableStocksXX to kick off research in factors, along with FactSet Fundamentals for the local universe?

In anticipation..

It is sometimes advised that investing in global equities is riskier than US equities. Presently, Q has a minimum 6 month out-of-sample period on all algos, prior to an initial evaluation for Q fund suitability. Will this be extended (e.g. to 1 year) for algos that trade exclusively in global equities, to cover the risk? Or perhaps there would be other ways to mitigate the risk? Or maybe the 6-month OOS is just to cover the over-fitting risk, which would be the same for global equities?

Also, presumably in the end, Q will be looking for algos that, in an integrated fashion, combine US and global equities. Will there be a constraint on the allocation mix? Or maybe authors will be able to write separate algos but pair them for evaluation, and set fixed weightings?

Hi Joakim,

Great questions!

Is there a 'hold-out' period for the FactSet fundamental dataset for Q Community users, and if so, how long is it?

Yes, all FactSet datasets will have a holdout in the community. FactSet fundamental data will have the last 1 year of data held out (on a rolling basis). In addition, pricing data for international markets will be held out 1 year. That said, all contest entries using these datasets will be evaluated by Quantopian with up-to-date data.

In general, do you believe the FactSet fundamental dataset is of better quality than the Morningstar fundamental dataset?

FactSet fundamental data is a very high quality dataset. Personally, I find the data structured in such a way that it's easier to use and understand. For instance, they clearly label the update frequency of each field. In terms of data quality, many investment firms use and rely on this data; the data has therefore had QA for many years, so that - along with FactSet's great reputation - gives me high confidence in the quality of the data. It will be interesting to identify whether there are new opportunities for alpha in a newly avaialble fundamentals dataset. This is getting away from your question a bit, but I think it's still in the realm of "which fundamental dataset should [I] use"?

Does FactSet fundamental dataset contain more data fields (than Morningstar), for example trailing-twelve-month data, rather than requiring us to calculate this ourselves in a CustomFactor?

I haven't analyzed this closely yet. However, I do know that the FactSet fundamental dataset includes trailing-twelve-month data (they call it last-twelve-month, or LTM data for short). Our initial release of the dataset will not include these fields, but we'll be looking to add it in the near term.

Will each new market that you roll out also have FactSet fundamental data available from day-1, or 'only' historical price & volume data?

Each new market will be added with fundamental data in addition to pricing and volume data. FactSet's fundamental data will cover both international and US equities. Morningstar's fundamental data will only cover US equities.

Would you be willing to share what some of the other FactSet datasets are that you may consider rolling out after Fundamentals? For example, earnings surprises, mergers & acquisitions announcements, sentiment data, etc?

After fundamentals, we'll be adding datasets available on Open:FactSet. We haven't yet decided on a priority order, but all of the datasets will come from that list (which is continuously growing!).

How far back will data be available for new markets and FactSet datasets? Since 2002 the earliest, as with US Equities?

Upon initial release, the data for international markets will go back as far as 2004. At some point in the future, we might try to extend this.

Do you reckon there will be a decent 'learning curve' with rolling out new markets and new datasets? For example, if it takes about 6 months for the first market and first new FactSet dataset (fundamentals), it may only take another 3 months for the second market/dataset, and another 2 months for the third, etc?

Great question. I think there's definitely a 'learning curve' to adding new datasets, but the shape of that curve depends on the data that we're talking about. For instance, we expect to increase the rate at which we add pricing and fundamental data for new markets, because the data has a similar look and feel in each new market. Obviously, there are unique properties to almost every new market, but the general shape of the data will be the same. When it comes to new types of data, it will depend on the extent to which structure of the data is new. We will be optimizing our development to learn about new structures early and integrate new datasets as fast as possible.

@Karl: Unfortunately, we won't have built-in tradable universe for international markets built at launch. However, in addition to pricing and fundamental data, we're looking into adding certain meta data like asset types (share, ETF, ADR, etc.) to try to make it easier for community members to build their own. At some point, we'll likely visit the idea of built-in universes in non-US markets.

all FactSet datasets will have a holdout in the community. FactSet fundamental data will have the last 1 year of data held out (on a rolling basis). In addition, pricing data for international markets will be held out 1 year. That said, all contest entries using these datasets will be evaluated by Quantopian with up-to-date data.

Curious about the mechanics of this. I guess the idea is the Quantopian Community would develop algos up to present minus one year. Then, they would be launched into the contest and full backtests would be run daily, up to the present. Would the entrant still be able to pull the backtest results into the research platform up to the present, or would they be clipped at present minus one year? Etc. Generally, it sounds like a great way to discourage over-fitting, but if it just means entering the contest to get at the 1-year hold-out period, then it just becomes awkward.

Generally, will you apply this approach to all of your data sets for which there is a hold-out period? Will we be able to enter any algo into the contest, regardless of the hold-out period? Presently, in my factor exploration, I'm avoiding data sets that don't run up to the present, but if the contest rules will be changing, then I might look at some.

Hi Jamie,
So if we use Quantopian Enterprise service, we would have the newest fundamental data. But because stock pricing data comes from Q , we still get the 1 year holdout period outside the US market pricing data?

@Jamie,

Awesome answers, thank you!

Is there anything the Q Community can do to help prepare for the new markets/datasets?

For example, since you said there won’t be any built in Universe filters for the new market on day 1, would it make sense for us to start making a generic one (I’ve seen great examples already in the forums), or would we need access to the new price/volume field before we can start creating a generic universe filter?

@ Jamie -

Some feedback:

Universe definition:

Unfortunately, we won't have built-in tradable universe for international markets built at launch. However, in addition to pricing and fundamental data, we're looking into adding certain meta data like asset types (share, ETF, ADR, etc.) to try to make it easier for community members to build their own. At some point, we'll likely visit the idea of built-in universes in non-US markets.

I'm not following, with respect to the approach that was taken with contest/fund algos and the QTradableStocksUS. I would think that if Q wants community members to be up and running quickly with relevant algos, you'd at least put up a trial universe. If the idea is that community members would need to build their own universe(s), I suppose that's fine, but how will they know what point-in-time stocks would be compatible with what you'd allow into the contest/fund? I'd also note that the QTradableStocksUS is not open source. Presumably, it would be handy if the community could use it as a code base, and adapt it for global equities, if is expected that it'll be a roll-your-own universe approach for global equities.

Data integrity

On Quantopian Partner Data - How is it Collected, Processed, and Surfaced? I'd asked a question, which included how the FactSet (and now Open:FactSet) data will be handled by Quantopian. As we saw with the presumably well-intentioned folks at Alpha Vertex, even professionals can over-fit when generating signal feeds. For a given data set, does FactSet make it clear which data a are consider in-sample, and which are out-of-sample (e.g. a flag included with the data set)?

Community vs. Enterprise

Will Enterprise users be eligible to enter the Community contest, have algos in the Q fund, etc.? Or will Enterprise be solely a SaaS (per the description on https://www.quantopian.com/factset, there is no mention of the crowd-sourced fund)?

Morningstar data

Will the Morningstar data eventually be deprecated/removed? This would seem to make sense, versus carrying two overlapping data sets. If so, this would imply that all algos needing fundamental data would have a 1-year hold out, right?

1-year hold out and user classes

It wouldn't seem to make sense to impose the 1-year hold-out of data for Community users who have gotten Q fund allocations. Generally, I would think that you would want a cadre of select Community users to have access to data up to the present. In his announcement, Fawce makes a distinction between Enterprise and Community users, and states clearly "Quantopian Community access to the new FactSet data will be subject to a delay." Is there a third class of user (e.g. "Community - Under Contract") who would not be subject to the hold-out, but would not be considered an Enterprise user?

Will the Quantopian risk model cover the non US stocks?

@Nick: To start, the risk model will not cover non-US stocks. We will likely look to extend the risk model to cover more markets at some point, but I don't yet have a timeline.

@all: There were some important questions asked by Joakim in another thread about the upcoming addition of global data and its use in the contest:

Will the the FactSet Fundamentals and new market be available for use in the existing contest, or will there be a separate contest for the new market? E.g. a Q US Open & a Q [new market] Open? Will the new market be based in the local currency or how will currency hedging work? (I can’t think of a USD denominated equity market outside of the US)

FactSet Fundamentals for US equities will be usable in the contest. International market data will be available in Pipeline only (i.e. you can build and test factors in research, but you won't be able to backtest with it yet). As such, international market data won't be usable in the contest right away. I hope to have more info on the scope of the international market data once it's been added to the platform.

Regarding currency, the data will be based in the listing currency of each asset (this is usually the same as the local currency). At some point, we will probably look to add an option to convert price-based values to a base currency, but it's not yet at the top of our list.

@Jamie,

This is just a suggestion, but once the new FactSet Fundamental dataset for US Equities is available, would you consider allowing an additional contest entry, so existing participants aren't 'forced' to withdraw one or more strategies, in order to enter a new strategy using the new dataset?

Joakim, that's an interesting request. I'll pass it along internally and see what we can do.