http://globalslant.com/2015/06/black-box-trading-why-they-all-blow-up/

Thanks, Anthony, sorry for a mangled link.

See https://www.quantopian.com/posts/quantopians-first-discretionary-capital-allocations#57078ec6dceb72f2ae000398 where it is reported that Q intends to ramp to 6X leverage. If Q does blow up, at least there will be a big bang!

I'm not sure having zero beta fixes anything. It just means that the algo is running independent of the market, but it could still crash, right? It prevents a pure bet that the market is going up or down. The algo could still get confused, and lose money.

My sense, though, is that the long-short factor style "trading" (seems more like investing/asset allocation) that Q is promoting may be more tame that other approaches, but I'm no expert.

The problem that August was more specific. This article is correct to about half way, then makes a wrong assumption; the problem was not that they were all doing generally the same thing, as I understand it. The problem was that everyone was doing literally the same thing; the same factors, the same universes, etc. Everyone was long the same stocks and short the same stocks. This is caused by both the same constraints, which he identified, but also by the fact that it's all the same people writing these models. They talk to one another, they live close to one another, they move between firms.

The promise of Quantopian is that by tapping into a vast green field of people with no exposure to that world, they might find some strategies which are not crowded. Of course, by funneling everyone into the same general class of strategies (factor-based equity long-short), they somewhat ruin that, but they have their reasons I suppose.

My working hypothesis is that Q is not really Q, but Point72, and has been influenced by them for quite awhile. They were handed a script, and funding to put on the show. Fawce's announcement didn't use the term 'acquisition' but it sure sounded that way, and perhaps was the only way for Q to move forward, given their burn rate. It is a huge win for Q the start-up, but it means that Q is no longer a start-up.

Another working hypothesis of mine is that Q is more of a recruiting tool for Point72, than a diversified crowd-sourced hedge fund. If you think about it, what better way to get your foot in the door than to get noticed by Q, maybe rank highly or win a contest, get funded as a manager, participate in their various training opportunities, etc. From the Point72 perspective, worst-case, Q is a self-funding global recruiting tool. If they end up with a handful of rain-makers on their payroll, it'll be a win. I seriously doubt that they bought into the visionary, unproven crowd-sourced concept as the sole rationale for putting up the money.

Regarding the narrow focus on the "me too" approach of factor-based equity long-short, presumably Point72 buys into the concept, and it would seem to be a natural fit for the Q platform (whereas something like HFT of some exotic derivatives would not). I can see how it is focusing the engineering resources on a single, achievable near-term goal. They've put together a high-level plan, assigned tasks, and are charging ahead to victory. It beats the alternative of chaotic sandbox play from which one might hope for a return on some brilliant innovation, but never get it.

One area of hope for the factor-based equity long-short is that although Q is reluctant to discuss it, my sense is that it will stress their infrastructure, and force an examination of where they should invest in improvements. I have an algo that would seem to fit their profile, but I basically can't run a backtest back to 2002, due to a memory error. And even if I make some compromises to get it to run (taking ~ 45 minutes), it bogs down my browser to load transactions, and I can't load the backtest into the research platform, due to memory limitations. Another example is that of the ML discussion, deep learning, model training, etc. Something's gotta give in terms of the infrastructure, or Q won't be competitive against professionals with real (but still commodity) computing power. The fact that Q is looking for users to sort out, point-in-time, how to deal with up to 1500 equities, and various other data inputs is probably a good thing. They'll eventually realize that their least-common-denominator, one-size-fits-all computing paradigm may not be the best. A small subset of the 80,000+ users would benefit from more horsepower.

One unfortunate turn of events is that the factor-based equity long-short strategy and associated workflow are not being adequately justified and debated with users, in my opinion. The worker bees do their technical postings, but the idea of considering if it is a good idea in the first place, and considering possible alternatives are not supported by Q. It is just a waste of time (hence the blog posting by the chief investment office as received knowledge, with commenting turned off). Unfortunate, but I can kinda see the practical rationale. It is business, not academia or coffee shop debate. And you don't won't the Q troops to lose faith and focus. There is a price to pay, though, in user engagement, buy-in, and leveraging the collective intelligence of the crowd in making high-level decisions.

Hi all,
I've been lurking these forums for some time and lately I've been implementing a strategy and some variations of it more seriously. The domain itself is interesting, if only for the intellectual challenge of it.

@Grant I've been following your thoughts here and there on the forums and you're indeed a thoughtful fellow. I can say that I more or less agree with your sentiment regarding the recent Quantopian developments.

My opinion whether Quantopian is or isn't going to shit is so what? For a newbie like me they're offering good enough content and good enough infrastructure to play with. And should I get invested that much into the platform and the platform itself becomes private/paid/goes away completely, they're open sourcing major components of their work, which with some custom coding and help from the community should be workable to get your own trading infrastructure in place. And by then maybe I'll have more knowledge/expertise to research and work with different tools.

Never forget, on the internet "If You're Not Paying For It, You Become The Product". So given that, are you OK with investing your time and effort here?

Respectfully,
Michalis

Simon's analysis of the blog post is pretty good. If you're reading this far and you skipped Simon's post, go back. As for his concern about us encouraging people to write long-shorts, we're not too worried about that at this point. New authors are joining every day, new data sets are added to the platform, and we check the correlation of strategies. In the future we intend to expand beyond US equities. To steal a line from the linked blog post: “They Are All Doing Different Things”

Grant, your hypotheses are incorrect, and candidly I'm somewhat hurt by the pejorative way you phrased them.

• Quantopian was not acquired by Point72, there is no script, and there is no show. They invested $2 million dollars in our company. It is fair to say that they have influenced us. Show me a company that isn't influenced by it's customers, and I'll show you a great shorting opportunity.
  
• Despite your skepticism, Point72 invested in Quantopian for exactly the stated reasons. There has been skepticism about Quantopian since we started. First it was that anyone would write an algo on the web, then it was whether or not quants would collaborate, then it was whether anyone would trade money, etc. We look forward to proving the doubters wrong again, and look forward to the next set of moving goal posts.
  
• I readily acknowledge that our platform has limitations. The good news is, we love knocking those limitations down and finding new ones. When we started, algos were limited to 5 equities, there were no history() functions, no corporate fundamental data, no live trading, and no research platform. We're not providing enough support for ML? Looking forward to knocking that one down, too.
  
• I'm not exactly sure what debate you are looking for. We are certainly not saying that long/short is the only kind of algorithm to write, or that it's the only correct way to invest. Are you trying to debate whether long/short is a viable class of strategies at all? That seems like a settled question, given the billions of dollars invested this way today. Are you trying to debate what the pros and cons of long/short algorithms are? That discussion has been been going in these forums for years. Do you want to debate Quantopian's choice to give allocations to long/short algorithms? The key question there is "what does the investor market want?", by which I mean family offices, pension funds, endowments, etc. I love the Quantopian community but the community's expertise doesn't cover that area. Still the discussions have certainly happened. I'd suggest that the real issue here isn't that Quantopian refuses to talk about these things, but that you disagree with the conclusions we've reached.
  

Michalis, I'm glad you're enjoying the resources we're putting out there. I like your attitude about it - it's a great thing to learn more and share it back with the community. There is one nuance that I think is worth discussing. At Quantopian we've worked very hard to make sure our business goals and our community members' goals were aligned. As our business has evolved we've kept an eye on that goal the whole way through, tweaking periodically to make them fit. In particular, remember that Quantopian will only make money on strategies that we have licensed and/or pay for. We're glad you're here making the community better, but we will only profit on that work if you do, too.

Hi Dan,

Sorry if I ruffled some feathers with my tone. It is amazing what you've accomplished since the early days, I'd agree. I'm still here, participating, and trying to learn, as time allows. And I'd encourage others to do the same. There are worse hobbies and habits.

I just think, given the commitment you are seeking from the crowd, you ought to approach every decision as if the crowd owns Quantopian. And share a lot more information. I don't think you approach things this way. But if you did, it would be novel and might yield surprising results. You could even go so far as trying to sort out how to give users equity (or provide a means to earn it). Things have gotten a real un-directional feel to them (in my opinion), since the days when Fawce was doing Q&A on the forum. Something has been lost. So it goes.

Regarding Point72 specifically, a Google search shows a $250M figure kicking around, and you have a site that sure looks like an acquisition (https://www.quantopian.com/point72 ) -- even the logos are similar in their squareness. It is not unreasonable to think that things are headed in that direction.

Best regards,

Grant

Interesting discussion

I think opening up the investment to any user would probably help make it more community driven. Say Grant publishes an algorithm and invest 50k of his own money. The algorithm is classified and ranked agains others and is open for investment under a fee schema.
I come and look at it and decide to join grant with 10k and he gets the fee he defined. People could write algorithms and make it open for investment and share the fee, say I write an algorithm but don't have the initial capital, but Mr M has, he can check backtest and decide to open the algorithm with me and share the commission we get from new investments.

There are probably some legal barriers around this, but this would improve the algorithm evaluation process and the diversification. I think there are already some platforms out there like collective2.

One angle is that Q wants to have its cake and eat it to. Operate as the traditional corporate pyramid scheme, worker bees at the bottom, rich fat cats at the top. Crowd-sourcing in this context doesn't include a fundamental flattening of the pyramid. The business has not been crowd-sourced. The crowd is kept at arms length, and doesn't get to decide anything important. There's a price to pay for not entertaining this approach. If one really wants a crowd-sourced effort, then the crowd needs to integral to the business. Power to the people is carefully thought out and applied, so that the power structure is maintained.

@Grant I think you probably have the wrong expectations, Quantopian has evolved and changed its business model. It is mainly an open research platform that provides super expensive data for free. The competition is a recruitment platform, If your algorithm ranks on top 5 on those competitions you would prob get way more money if you just pitch it directly to hedge funds.

I am also not sure if your approach would win any investments, other communities are more member driven, but members need to subscript when new data needs to be added, most of the features must come from members implementing it.

Lucas - I may be unrealistic, and Q may be getting plenty of participation and good algos under their current way of doing things. Something is missing from the formula, though. Too unidirectional of a relationship, in my opinion. More information could be shared, and more decisions could be collective.

@ Dan -

The key question there is "what does the investor market want?", by which I mean family offices, pension funds, endowments, etc. I love the Quantopian community but the community's expertise doesn't cover that area. Still the discussions have certainly happened. I'd suggest that the real issue here isn't that Quantopian refuses to talk about these things, but that you disagree with the conclusions we've reached.

I don't necessarily disagree. You need to understand that users aren't privy to your internal decision making processes, and don't have a substantive part in them, as best I can tell (rightly, you could argue that they aren't entitled to a say, since they don't own the business). The process of landing on the long-short approach, from the outside looking in, was organic. You didn't kick things off by publishing a white paper on your market research, discussions with prospective customers, and a description of how your business plan would allow you to tap into a certain market segment. The initial contest had relatively few restrictions, allowing long-only (or short-only) entries, for example. Eventually, you found your way to a set of criteria that presumably match with your target market (including the recent move to $10M in capital). We've gone from a seat-of-the-pants effort to A Professional Quant Equity Workflow which presumably aligns with what Point72 wants to do with their potential $250M investment. A good thing from my perspective, if my assumptions are correct.

@ Dan -

On the topic of the platform limitations, it would be nice if you empowered your help desk folks to explain in full the reason behind certain limitations. For example, there is a complete, rational explanation for why the research platform has 4 GB of RAM, that limits the number of transactions that can be loaded. For those of us who are engineers, it is very frustrating not to get a full explanation. And if something can't be explained, because you can't reveal it, since it is viewed as internal proprietary information, then just say so. The point is that this isn't Facebook. You have technical users, who may be more engaged if you provide technical responses (including business/strategic considerations, if applicable).

Hi Dan,

Again, sorry for coming across the way I did above, since I'm sure y'all are doing your absolute best. Regarding the use of the term "script" I don't think it was that far off the mark, since you do have "benchmarks" per http://pitchbook.com/newsletter/steven-a-cohens-point72-commits-250m-to-vc-backed-quantopian :

The entire $250 million from Cohen will not be immediately available, but instead disbursed based upon certain benchmarks.

My guess is that you have a hand-shake agreement or formal NDA not to reveal details of your business arrangement. But you could have said to Point72, "Hey, we consider our user base as an integral part of the team. How 'bout we publish a bunch of details of our agreement, including the benchmarks?" Or maybe you tried this, and they said no?

The other thing that is sorta odd is that you started to publish the performance of your fund, but then stopped. I see:

Quantopian’s unaudited 1.93 per cent first full quarter returns are net after the fees paid to the authors, an assumed 2 per cent management fee and the firm’s own 10 per cent performance fee — a structure that mimics the common “two and 20” hedge fund fee structure. The S&P 500 rose 0.8 per cent during the first quarter.

(Ref.: https://www.ft.com/content/0808729a-faa6-11e5-b3f6-11d5706b613b)

Will you be reporting on an on-going basis?

Again, I'm not meaning to disparage your work (and certainly not hurt any feelings). But I sincerely think that you need to evaluate what it means to be a "crowd-sourced hedge fund" and the implications of that business model vis-a-vis your approach toward the crowd upon which you are relying. If you took a fundamentally different approach (e.g. greater user status and influence and radical transparency), would it be beneficial to your crowd-sourced effort?

Dan -

Another quick comment is, if you haven't read it already, I recommend "The Innovator's Dilemma" (I know Fawce is familiar with it). The risk I see is that success for a start-up may have nothing to do with following the pack, creating products that plug into big markets and big customers like Point72. It is more a matter of what are you doing that is fundamentally and radically different, will you be considered an industry disrupter over the next 5 years? Presumably, professional quants at major firms have all of the data you are offering, plus more. And infinite computing power. Etc. And they are working on the same style of long-short algos as you are requesting from users. I think Simon has a valid point:

The promise of Quantopian is that by tapping into a vast green field of people with no exposure to that world, they might find some strategies which are not crowded. Of course, by funneling everyone into the same general class of strategies (factor-based equity long-short), they somewhat ruin that, but they have their reasons I suppose.

How do you know you are not making a fatal mistake, because you are unknowingly caught up in the innovator's dilemma?

@ Dan -

Regarding your comment "we check the correlation of strategies" it seems that in light of the black-box nature of algos, and the prevalence of over-fitting (ref.: https://www.quantopian.com/posts/q-paper-all-that-glitters-is-not-gold-comparing-backtest-and-out-of-sample-performance-on-a-large-cohort-of-trading-algorithms ), you must be talking about out-of-sample data for your check (either paper trading or real-money). Making any definite conclusion would seem to be pretty hard, given the paucity of data. I don't have a feel for it, but for algos that trade daily/weekly, at most, how long would it take to actually perform the correlation check? And how would you account for market cycles? I'm just skeptical that you'd be able to accumulate enough signal to overcome the noise to make a statistically valid assessment of the correlation of strategies. The "error bars" on the correlation measurement could be pretty large, no?