Does this mean that the winning algos are no longer required to pass ad hoc pyfolio tests of 'economic prudence' and so forth, since you won't actually be trading them with your capital at all?

When you win the contest, you get a $5000 check

Any chance you could make these rules retroactive? : )

Simon - it does mean that "economic prudence" is a lot less relevant. We'll still remove people trying to "game" the system. But we're not going to sweat overfitting and such quite so much.

Grant - I imagine that we'll offer this prize retroactively and optionally to whomever wins Contests 8, 10, and 11. But for Contest 1, the ship has sailed. =)

...I'm actually shamelessly excited to get a t-shirt .

Will the winning algorithms be selected for the fund? How will it work?

Kinda surprising that in Dan's announcement there was no mention of the Q fund. Presumably the contest has a role in identifying algos for the fund. Has a systematic approach been developed? Is Q looking for 6 months of paper trading, and then you'll roll up your sleeves with pyfolio, etc. and do an assessment of the top candidates from the contest? Or could an algo be selected for the fund within the 6 month period, and additionally get $5000 at the end of 6 months?

Personally, I'd be glad to send Q the code of any algo I enter in the contest, if you want to evaluate it for the fund. Is an entrant notified if you do an evaluation of one of his contest algos?

So does it mean that the selected algo will no longer live trade but just paper trade, right?
Giving just a fixed prize isn't very meritocratic, anyway. Even a % on the actual profits is more meritocratic.
Just my opionion, anyway.

Thanks

My algorithms require historical fundamental data and for the moment there is no way to get them working with live trading, as I noted here:
https://www.quantopian.com/posts/historical-fundamentals-data-problem-with-live-trading

Unfortunately also the new Pipeline API didn't solve the problem, see
https://www.quantopian.com/posts/introducing-the-pipeline-api#562e3bc48d7aef6e630006a6
and
https://www.quantopian.com/posts/how-to-get-historical-fundamental-data-with-the-pipeline-api

I'd like to know if the Quantopian Team is addressing this issue and if there will be fixed in a foreseeable future.

Grant and Pravin - The irony here is that I deliberately avoided talking about the fund exactly because people get confused about the two. They are separate! That said, I'll try to answer the questions.

As has been said before, you do not need to win the contest to be selected for the fund. The fund needs dozens of algos. We don't want to add to the fund, one algo per month - that is an arbitrary and debilitating constraint on the fund. We take algos that are good, as we find them. Sometimes that's more than one a month. Sometimes it's less.

I don't think of the contest as "identifying" algos for the fund. I think of the contest as "aiming" algos for the fund. The contest is the best tool we've found for telling our community what we're looking for. On one hand, we can't specify exactly what we want because we need uncorrelated variety, not mass-replication from a template. On the other hand, we can't say "just write anything, it's all good." It's not all good. The contest (particularly after the summer rules changes) has been doing a good job at guiding the crowd.

We are indeed developing a systematic process for identifying algorithms for the fund. If you track the development of Pyfolio you'll have a good idea of how the exploration and iteration in that process.

I'm working on a page that outlines more of our fund process. It's been a bit of a challenge to write because the process continues to evolve. I'm trying to capture the core principles without putting out a bunch of details that will quickly be obsolete. I'm probably going to resist follow up questions about the process until I have a more full-formed reply. I'm aiming for the next couple weeks.

Emiliano - correct, the winner is only paper trading, not real money. I also loved the "winner gets the real money profit" prize, and I'm sorry to see it go. It's just not practical for us at the $1,000,000 level at this time. Maybe we'll be able to bring it back in the future.

Costantino - I commented on your post here: https://www.quantopian.com/posts/how-to-get-historical-fundamental-data-with-the-pipeline-api#564f4bf44c1eb86fdb0000a5

Apologies, I feel I shouldn't make a comment given a total lack of contribution as a lurker. As an ex-fund manager back in the machine learning world. I am actually surprised at this decision. Keeping a cut of profits is interesting, indeed it can always be scaled, if 100% doesn't work. However, if people develop something 'real', for a while at least then they at least share in the results - a small prize less interesting. Apologies - for my comment, I am not really in this community but I do watch and find it of great interest. The philosophy I agree with greatly, the idea that one will get immediate results from the best crowdsourced algos, should possibly deserves patience.

I always love a good t-shirt!

John - when people develop something "real," as you say, it gets included in the hedge fund, and they do get a cut of the profits. If your intellectual property is valuable, you will be rewarded for it.

Thanks for the clarification Dan

Very interesting change, I am also shamelessly excited about potentially getting a t-shirt. It does scare me as one of the biggest benefits of winning the contest for me was the real money funding. After deploying with real money I learned that the Quantopian default slippage model wasn't right it was actually too aggressive (at least in the case of my algorithm). With paper trading for contest that will be no longer be available for contest winners, and nothing in my opinion speaks louder than real money being put behind an algorithm to see if the algorithm can hold up in the real world. That being said I understand the economic feasibility of not being able to fund millions into each algorithm.

Yeah, getting the commission and execution data from my algo on real money was very valuable.

UPDATE! Entries for Contest 12 are now open! If you enter your algorithm now, it will be on the $1m capital base. Everyone is reset to having 3 new entries to make!

(The leaderboard for Contest 12 is not updated. The leaderboard will be corrected sometime after close on Monday the 23rd.)

I'd like to see more realistic defaults for slippage and commission. Not only would it benefit Qs algorithm selection process, it would greatly improve our personal confidence with regard to live trading our algorithms on our own.

Additionally, tshirts.

Sometimes, as Spencer points out, the slippage model isn't aggressive enough. On this other thread, Kevin Q argues the commission model is too aggressive. And James would like them both to be more realistic.

The challenge is that everyone is right, to some degree. Slippage in particular is very difficult to model because isn't uniform, by stock, by day, or by hour of day. I agree that our default commission is a bit high (though, I'll point out that Robinhood isn't a platform for the hedge fund, so the commission isn't going to zero for us in general).

We don't want to make frequent changes to slippage or commissions because we want backtests to be repeatable. We also want everyone to be on a level playing field - everyone using the same model. That keeps us from iterating freely. We're bound to past decisions, to some degree, and we really don't want to thrash around on it.

We keep collecting data and at some point will find a model we think is solid enough to warrant an update. As always, thanks for the feedback.

P.S. I've seen the preliminary tshirt design, and it is just sexy. You all are very right to want one. I'm actually jealous that as an employee, I can't qualify to wear it.

I'd just like to say that, as a Canadian, I think the default commission is in a slightly more realistic place than the slippage. But now I'm just being bias and only slightly less ambiguous.

What do you mean? I am a Canadian trading with IB, and my algos (which are all currently trading on quite small accounts) are executing trades at the minimum $0.35 @ 0.0035/share roughly. This is for US stocks.

At Scotiabank I end up paying roughly $0.003-$0.005 per share, the default commision here is not far off that mark.

Hmm last I heard, the default commission model was $0.03/share, 10x more than the above. There was a proposal to update it, but I don't believe it ever happened?

This is i presume obvious but to ask the obvious rather than searching around - what are the very 'rough' parameters for a contender algo? given a simple mind important stuff first? What would i be expecting to grasp would be a triangulation - trade cost- some form of risk return parameter ( unknown and very bad is a piory guess !)...i presume usual unicorn stuff - uncorrelated alpha, sustainable, logical, justifiablw...

horrified to see my half baked thoughts published whilst on my iphone - anyway regardless...it was half baked...anyway ignore and apologies...

You're right Simon, that is significantly higher.

I wonder why it doesn't effect my algorithms live performance much.

Perhaps because the slippage model is more generous in its consistency ?

I can't figure out why commissions don't strongly effect my most recent algorithms performance.

I concerned with regard to what this might mean when trading with actual money.

I'm super confused at the moment...

...How is Q's slippage model applied while purchasing ?

Okay, I just polished off my 3 entries for contest 12 fingers crossed for that t-shirt. As I was doing so, 2 potential issues arose that I don't believe were addressed to this point. First, now that entries are not being funded I was wondering will the top person in contest 12 get the 1st place prize even if determined "something that Quantopian doesn't want to be funded." Basically will Quantopian still disqualify entries for not being what they are looking for in a strategy? Secondly, are you allowed 3 new entries in each new contest from this point or are you only 3 entries total at any one time? Just was wondering if I could get some clarification on these 2 points.

I figured it out :D. Fortunately I didn't have to pull my entry. But I'd like to know why quantopian won't be using Robinhood for the fund or contest ?

This seems like a relatively disappointing change. In the past, there was a two-phase process -- first an algorithm was selected and then we saw some true out-of-sample behavior to validate whether the alpha was real. With this scheme, there is no true out-of-sample.

There's always a big question of whether a system like Quantopian, combined with the 'crowd' can actually generate algos which beat the market. With the usual incentive scheme, the people who know aren't talking and the people are talking don't know, so there was little in the way of public evidence that profitability is even possible. The original contest seemed like a great way to incentivise people to actually share some track records, though I suppose the actual outcomes weren't encouraging. The new contest, however, won't even do that.

Actually, can we get an update on how Quantopian's financing / fund raising is coming along? My biggest concern with using this platform (as opposed to building my own or using something more established) is that at some point, Q will go under and a lot of my time invested into building systems will be wasted. The change of the contest from out-of-sample with real money to in-sample with fake money does very little to convince me that things are going well for Q. In some sense, Quantopian has two possible business models -- either to align its interest with the users and require profitable algos to make money or go the brokerage / service provider route and take fees and commissions from quants without needing them to be successful. From what I can tell, Quantopian has chosen the former path but has shown very little evidence that enough users can build profitable strategies to actually pay for Q's expenses which makes me nervous.

Alex S I agree 100% with what you said.

My bet is that during or before QuantCon 2016 in April, Q will announce $X million of capital applied to their crowd-sourced hedge fund. They will have funded N managers who probably won't be named. At the same time, they may provide guidance on their plans to fund more managers, increasing the size of the fund. There will probably be guidance on the payout to the managers, as well. My guess is it'll be something like $25M in capital and 10 managers, so that's $2.5M each. Assuming a 10% annual return, it'd amount to $25K per manager (assuming Q gets 20% of the profit and splits it with the managers). So, my hunch is that a story like this will play out in the coming months, to incentivize the crowd, beyond the contest.

The goal, as I've understood it, is for the Q fund to get to $10B in capital, with institutional investors. It'll be interesting to see if they can sort out how to get there with a pure crowd-sourced approach. It seems far-fetched, but maybe I'm too uninformed and skeptical.

I sincerely hope what you say comes true Grant. Q has been somewhat silent about the fund so far. We do not know anything about it.

In lieu of these changes, is there any possibility of allowing fetch_csv() again for the competition? My algorithm is more or less dependent on it, and I can prove its uptime privately...

I'm also a bit sad about the rules changes, but I do understand the need for them. They added new tools, we can now browse bigger universes, and algos managing 1M make more sense for the fund. They are learning everyday, and I think poor performance from combined first contests winners plus current market volatility is actually a blessing. It will make everything more solid.

However, if the goal is really to make contest algos better at "aiming" for the fund, I think a 6-month contest is a little short for not checking for "economic
prudence" with 10-years+ backtests. Also, a 5000$ prize is a lot less sexier than the "manage our money" model, in my opinion. I hope the t-shirts compensate for that!

Note: I will not be in a position to accept any Boston Bruins-theme inspired design.

I don't think Quantopian is looking for algos that return much more than 10% per year anyway (no "sexy" volatile algos allowed), so $5K guaranteed after 6 months of simulated live trading is a nice prize. And they're running a monthly contest, so it seems like a good deal for the crowd.

Side note:
I just noticed -about 15 minutes after the deadline- that my three algos (that have been rolling-entering for a few months) aren't in Contest 12. Turns out I misread the "you will need to re-enter it", partially tricked by the previous "you can make 3 new entries" in bold font that led me believing it wasn't mandatory.

I understand it's totally my fault but is there any chance to still enter the contest, even if with a penalty for late entry?
(I'm writing this about 35 minutes after deadline)

I think I'm not the only in this boat since the applicants to Contest 12 are about forty, compared to the about eight hundreds of previous ones. I guess something in the communication chain didn't work too well…

edit:
I stand corrected, I now see a much larger leaderboard of ~150 people.

I'll enter the algos in the next starting January next year.

@Dan: just a thought. Why not making optional to the winner to have the price of $5.000 or run the algorithm live with $100.000 (as before, not $1m ) ? An algorithm could scale well and perform similarly at both $1m and $100.000. In that case, I believe, the winner would be more interested in running the algo live than getting the price in cash. For algorithms that are design to run with $1m minimum, there is still the price in cash.

I believe that since the contest winners are now hidden, or at least, difficult to follow the contest has lost its momentum.

You will get more interest if you go back to a leader board that lists past contest winners and how their system has played out. You might add footnotes if the rules have changed from month to month.

You might also consider an additional 'bonus' for a winner to make his system public. In this way winners could earn $5000 cash for winning the contest; and another $2000 (?) for making the system public. This would slowly improve the quality of future contest entries.

Good luck to Quantopian and all algo developers in 2016!!!