Awesome!

Is the default slippage model the same quadratic impact/2.5% one?

Hi Dan -

I'm kinda surprised you didn't take this opportunity to re-jigger the contest judging so that the effect of gross leverage is normalized out, with all algos compared at an effective gross leverage ~1.0 (and maybe remove the arbitrary restriction of leverage < 3.0). Or maybe the contest judging criteria already capture this?

Grant

One concern is that this is another step in the direction away from any kind of innovative crowd-sourcing concept, where somehow you could engage and fund all of your 80,000 users. The more capital allocated to individual strategies, and the more herding you do of users to familiar hedge fund territory, the less innovation you'll get, I'd think. You seem to be headed toward 30 or so algos, each at $10M, or $300M AUM. I guess the idea of getting 300 algos at $1M each just doesn't seem feasible?

Grant, this decision is a step in the right direction, if you want to get big. It is well known that you can pretty much replicate the diversification in the S&P500 by choosing about 30 member stocks at random. So....

Quantopian's decision to nudge the algo writers toward highly liquid equities is indirectly doing that - if they choose a minimum of 30 algos, focused on the top 1000 stocks by market cap, they will pretty much be as diversified as the total portfolio of the top 1000....this decision is a hedge against underperformance. The downside is that in the aggregate, it will be tough for those 30 algos to outperform the top 1000 stocks (equal weighted, random weighted, or market cap weighted).

Very cool change, will be interesting indeed.

@ Grant:
You may be right that 300 algos aren't feasible. However, in general, I would suppose it's more attractive to have too many high capacity algos that you can then allocate less than $10M to, rather than having too many low capacity algos that cannot be extended leverage-wise in the manner that Q prefers.

Yeah, I just don't quite get the crowd-sourced fund concept yet, as it migrates toward large portfolios, with high capital levels. Kinda smells like a conservative "me too" effort that might not play out so well for a start-up like Quantopian. The other end of the spectrum would be sorting out how to do $100K (or less) per algo, but engage all 80,000 users--$8B AUM, presto! Guess that ain't really the plan. I don't understand what the remaining 79,970 users will do?

But maybe the projections are that they'll outperform similarly constructed funds? We'll see.

I strongly agree with Grant on this matter.
Given the modest tools Quantopian has made available to its members and the ambitious results they expect to achieve, the requirements and selection criteria are very unreasonable.

If we hope to achieve any measure of success with the current requirements and a limited selection of 30 or so algorithms, we are going to need a considerably better development environment and reliability between backtesting and live trading. This would mean we can't get away with using, for example, crude slippage models and low-frequency data instead of actual simulations based on tick level order-book data nor can we stay limited to trading equities without the much needed diversification from futures and options.

If Quantopian does not want to go that route then if we hope to achieve any success with the limitations of the current tools that are provided, we are going to need to considerably ease up on the requirements and allow for the inclusion of a wider diversity of algorithms with capital allocations that are more suitable.

I strongly agre with Quantopian on this mattter.

Quantopian will need to at least monitor the exposure, leverage, trade fills, etc. of each algo, and doing so for a large number of algos is expensive and impractical, but more importantly, adding more than 30 algos will not have any benefit to the offering as a whole. The number of algos could be even smaller, if they want to have a chance to outperform by a larger margin.

Well, it just seems like a kind of global recruiting effort, but otherwise, kinda-sorta like other funds. I had this picture in mind of a crowd-sourced fund that would somehow be able to accommodate the entire crowd. Joe Six-Pack could get his $100 per month check, along with the hot-shot, experienced coder-traders, making a living at it. But maybe I misinterpreted the crowd-sourced concept from the get-go.

"it just seems like a kind of global recruiting effort"

I agree it is, but we all know that by the law of large numbers, they will have 1% genius-level talent here (FOR FREE), and maybe 2% of the algos will outperform the market, even out of sample. So if they go for that top 1% of the crowd they will have plenty or algos to pick from...I don't think they are looking to hire people, or to increase their expenses to obscene levels so that the average Joe Six-Pack can have a $100 per month check.

I had this picture in mind of a crowd-sourced fund that would somehow be able to accommodate the entire crowd. Joe Six-Pack could get his $100 per month check, along with the hot-shot, experienced coder-traders, making a living at it.

The problem with trading is that there is no middle ground. There are losers (lots of them) and there are winners (not that many) and not that much between. Especially in algo space.

I think this is a very good move from Q and it's good to hear they are thinking of allocating real money to good algos.

I'm not sure. It is easy to say it is a good approach because it is familiar. I guess if it is too weird/difficult/innovative, then it might not work and/or investors won't know what to make of it. Whose lunch is Quantopian gonna eat, and by what means? Not obvious, but maybe the quality and diversity of those 30 algos will be superior, due to the Quantopian Web 3.0 virtual global R&D department.

It'd be a shame not to at least try to figure out if there are trading problems that could be solved by 80,000 people working on them at the same time. That'd be kinda cool.

Yeah that would most definitely be a unique avenue but unfortunately not the one we are seeing here. If you wanted to discuss what types of trading problems might be most suitable for crowdsourcing and how we might be able to integrate such widely dispersed knowledge I would be very happy to discuss that.

The approach being taken here is pushing Quantopian very strongly in the same direction that thousands of investment banks, hedge funds, and proprietary trading firms have taken, except that we are using technology that has been available since the late 1980's. That's not an exaggeration, apart from the fact that the Internet and electronic trading infrastructure itself has improved tremendously, the tools we are using are no better than what was available then.
This is extremely familiar territory and there is no reason to believe that just because we are making this accessible to more people that this makes us any more likely to do well when we are literally trying to do the exact same thing they have been doing for 30 years.

Currently, the only way to meet what Quantipian is expecting with what we know is possible to do on Quantopian is for your submitted algorithm to be a composite of many various strategies that are coalesced intelligently so that you get relatively good performance throughout both the back testing history and the walk forward. This is an unnecessary constraint, as a fund manager I would be much more interested in receiving a bunch of submissions for an algorithm that trades, for example, volatility futures very well during the current market conditions. It would then be my responsibility to figure out how much to allocate and when to shut it off.

I am certain that at this moment there exist 100 algorithms on Quantopian that could each trade up to $100,000 that would outperform any single algorithm currently on here trading $10 million.

"I am certain that at this moment there exist 100 algorithms on Quantopian that could each trade up to $100,000 that would outperform any single algorithm currently on here trading $10 million. "

I have a Contest #20 submission ID=577fbc3eab5ec33077000813 which trades 10M with 2X leverage, but can be scaled to billions of dollars and takes very little maintenance. I bet you can't find an algo with a better annual return on all of Quantopian. I am only half joking, since I just submitted it and it made 1% in one day, so the annualized return figure is way out there. But do check on it, and see how it compares to the rest, both 10M and 1M algos, as we progress.

I hope Quantopian will keep the personal algo trading feature alongside the hedge fund. It's a cool feature!

I think you should consider making the contest more rewarding towards collaborative efforts.

Competing against each other creates incentives not to share your thoughts and algos. If you made it possible to share credit where it is due, like for instance awarding smaller prices to the two most deserving collaborators or something, I think you would help build a stronger community and a better learning platform, which in the end would result in better algos for Q. This was just one small example, I am sure you would be able to come up with better systems for acknowledging collaborative efforts in the contest, and create stronger incentives to share.

Are the entries for contest 20 rolling or do we need to submit new entries for contest 21 as well?