15 Million fund -- Quantopian is betting on you guys to produce
59 responses

Anony,

What's your concern here? You seem to imply that there is some sort of "too good to be true" at play...

Grant

Hi,

Like any company, we're here to make money. Our success depends on making money for our customers as well as ourselves. I expect successful quants to make more money from their own strategies than we do. I also expect competition for inclusion in the fund to drive team-formation inside our community - this famously drove the final stages of the Netflix Prize.

One small, but important, clarification: our recent raise was venture investment in Quantopian Inc. The money will be used for operations, not as investment capital. We have a lot of work to do together before we launch the investment vehicle. The first step is letting the community know where we are going, and rallying your support. Next up is developing track records with real money trading.

thanks,
fawce

Disclaimer

The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances.

Hi Fawce,

If I read your announcement correctly, to be eligible, an algorithm must be up and running live with real money to be eligible for consideration as Quantopian-fundable. Backtesting and a paper trading track record are not sufficient. I think that this highlights an opportunity to bring these simulation tools in line with actual practice. As a specific example, say someone conjures up an algorithm that would require $250K of capital to perform optimally. It may not make sense to set the algo up live with$25K, since the frequency of trading will be limited by commissions. So, it seems you'd want to partner with algo writers at an earlier stage, and ensure that backtesting and paper trading are predictive of live, real-money trading.

On a somewhat separate note, what is the incentive to increase the number of Quantopian users running live algos with real money? Has there been any discussion with IB or other brokers in reducing commissions or minimums if you can deliver a certain size block of traders?

Grant

Hi Grant,

I love the idea of quantopian staking algorithms and quants earlier in the cycle, but I see that as possible only after we have the fund operating. I think we will need to do something along these lines in the future, in order to find all the talent out there. Given data about successful quants on our platform, the question will be whether we can predict success for quants who have only backtested and simulated. My intuition is yes, but that we need data.

Our focus on increasing the number of live traders is simply to increase the number of track records we can evaluate. We'd like to have as many algorithms with history on quantopian as possible, so that we have as much data as possible to evaluate who to include in the fund. We haven't started negotiating with counter-parties for better rates on commissions and borrowing, but that certainly makes sense at higher scale.

thanks for the questions and thoughts,
fawce

Thanks Fawce,

Are you still planning to charge a maintenance/service/usage fee per live algorithm per unit time (e.g. monthly)? Or has that business model been scrapped in favor of this hedge fund idea?

Grant

How are you going to assess track records - will you look at how the algorithm works (i.e. the code) or simply when it chooses to buy and sell? You would need to know which markets an algo was operating in?

@James: Code is not supposed to be visible to Quantopian unless it is explicitly shared either by a) including a backtest of it in a post in the forum or b) adding one of Quantopian's employees as a collaborator on a particular algorithm. Given, however, that the code obviously does run on a server at some point, I'm not sure I accept this narrative.

Another interesting angle that seems to be open: what is to prevent someone from writing a [thin] algorithm on Quantopian which merely consumes an external CSV containing buy/sell signals that are computed on another platform entirely (or manually entered by hand); i.e., an experienced [human] trader could essentially give the appearance of being a bot/algorithm (or there could be some blending of the two). Would the inclusion of such an "algorithm" in a fund purported to contain only fully automated bots be of concern (to outside investors or others)?

Hmmmm. While I enjoy reading the posts here, and get the occasional "oh, that's how one might implement that in python" idea here (as I suspect most members do), I have just a bit of skepticism about this "fund". Seriously? U.S. equities only, either "daily" or "minutely" with recently discussed limitiations in sending orders on market close? No options (not even retail-available ones), no futures, no FX? Quantopian infrastructure => IB as the only option for order delivery?

I find it hard to believe that someone(s) with professional trading experience put up 15M for such a crippled system. I can imagine a friends/family putting up a few dollars to someone who has done excedingly well at retail trading, but good heavens... one has more opportunities to play with the main retail platforms - TOS, IB, Tradestation, Ninja not to mention more expensive (pro/semi-pro) ones than with this arrangement. It makes me suspicious that something is amiss here - in the sense of sloppy, not sneaky.

Why doesn't the announcement page (https://www.quantopian.com/managers) lay out the exact criteria for selection and "share" of returns?

Sorry to bust on a fun/interesting project, but this "fund" looks all wrong. I doubt I'm the only one thinking the above thoughts, even if I'm the first to post in such blunt/rude language.

Let's see.... A web-based broker-agnostic automated trading system, with custom user-written python code running off a standard data and order API - that might be a compelling product, for a limited but but discerning audiance/customer base. If that's where Quantopian wanted to go, cool; but that's not at all what seems to be happening.

@Michael Like Mr. Fawcett said, the $15M is going towards operations. I suspect this means the infrastructure that you deem is 'crippling' it. However, maybe you are right, until these upgrades happen, perhaps the the masses will only find Quantopian just a fun and interesting project. Then again, how many projects started as a fun, interesting project in someone's garage, only to grow into billion dollar corporations? :) @Michael F Just want to invite you to consider even further from an angle of the investor's point of view. Say, maybe you and some other potential investors sit down with Fawce. You're aware of this article that Quantopian has over 20,000 users (not all active of course, and yet, a lot) and over$100MM in transactions so far (among only 150 of the users trading real money), and while you can't see ppl's code, here's the key, you're privy to some data showing that the results of some of the algos are impressive. In this hypothetical, as experienced investors much like yourself (I'll bet), they say, wow, those returns already? Even without options or futures or FX etc yet? (As you pointed out). And surely more capabilities to come. They see the rate of increase so they think, by the time we have all of the SEC requirements in place conditions will surely be even better and it looks like we will be able to afford to share handsome profits with the quants for a win-win, I don't see any downside to that except maybe for some folks who might be bothered by the success of others, an unfortunate reality in our world and you might have already been on the receiving end of that phenomenon.

Quantopian has a philosophy of being candid (I like that) so it doesn't surprise me that the criteria haven't been published fully yet as it seems fluid/early, and I'm ok with that part. Could be even a year out, I recall reading.

By the way, on a more general note, while I know Python ok, I'm a newbie with the market and had a misconception, I thought hedge funds were geared to hopefully balance where one set goes up if the other goes down, so I wasn't clear on why this would be described with the "hedge" term, then I read on wikipedia a few minutes ago "a hedge fund is an investment vehicle that pools capital from a number of investors and invests in securities and other instruments", and "hedge funds today do not necessarily hedge".

@Jonathan On that thought it would be possible to write a [thin] algorithm which uses a CSV with malicious commands (i.e. do well for 6 months and then take a large short position or some such). Obviously you wouldn't allocate a significant fraction of the fund to individual algorithms, etc, but the possibility is there (You don't even need a CSV -- just code a future date). The more practical reason to look at the code is to check its robust. Therefore all code in the fund should really be reviewed.... (Im aware the launch page says IP will be protected)

I don't yet follow the logic of this hedge fund idea. The idea is that if one is successful with a small amount of personal money for six months, then Quantopian would be willing to put up a much larger sum. Wouldn't it make sense to backtest, followed by paper trading at IB at scale, and then start working with Quantopian? Are backtesting and paper trading not sufficiently accurate? My understanding (perhaps misinformed) is that equities require a rather large sum of capital (~$250K) to be successful in active trading, so backtesting/paper trading at this scale would make sense. However, if one then puts say$25K toward the strategy, it's just gonna burn money through commissions, right?

Anony,

I guess my gripe is that by requiring 6 months of exceptional real-money performance (details TBD), a lot of potentially talented folks are left out of the game (although I suppose they could still approach Quantopian with backtest/paper trading results, with a plea for funding). For example, to do any kind of active trading, my understanding is that it is $25K minimum capital. How many middle-class brainy college/grad students have that kind of money to put toward a trading algorithm? Basically, it feels like the program is at risk of being slanted toward those with lots of cash sitting around. It'd be nice if there were a ~$0 personal capital threshold, with the burden on Quantopian/IB to provide realistic simulation tools.

Grant

@Grant $25k is possible with an algo that doesn't trade often (e.g. once week on n securities) but I suppose that's long only. But you're right -- the programme is on the side of those with money already. Thats why on the other thread there are so many requests for Forex because it is scalable. Hi Anony, Fawce referenced the Netflix Prize above (http://en.wikipedia.org/wiki/Netflix_Prize), but it seems like an apples-to-oranges comparison in that I gather that the teams competing for the Netflix Prize did not need to put up any money to compete. Maybe once the research environment is up and going something like this could be sponsored? I think for innovation, they're gonna need to figure out how to tap into the young, smart, penniless crowd. In some sense, the Quantopian hedge fund idea is going after an existing market--folks who have cash and proven trading algos who could move their strategies to Quantopian for a shot at more capital. It is a natural approach, but it is not clear how much innovation will result. Grant Hi everyone, Over the past week, we did three public meetups where we discussed the motivation and current thinking on the fund. Here were a few themes that recurred in each meetup, and that appear in this thread as well: • Why are you announcing your plan to be a fund, without details? The reason is we want to design the fund with the community's feedback. I felt that once we had made a firm decision internally to become a fund, we owed our community an explanation. It will take significant time and effort to design and establish the fund, so I didn't want to wait to tell you about the fund until we had everything figured out. We are working on our first draft of the fund design and compensation scheme for participating quants. The plan is to review this privately with our advisors and then to release it to the community for review and feedback. Designing incentives is very difficult, and I'm sure that a community review will result in many optimizations. • Can the fund be successful with just US Equities? Based on what we've seen so far, in an admittedly short run, I think we probably could. However, our vision is to build a multi-asset class platform and fund. I'm fanatical about focus, and completing the workflow from research through trading for US Equities is our current focus. We will add other asset classes. It is a matter of when, not if. • Why require real money trading for prospective quants? Looking at real money track records is a simplifying assumption for us. Like any practical field, there are myriad differences between investing simulation and reality. Investing your own capital demonstrates a level of conviction that I think is a valuable screen on prospective quants. All that said, what makes us totally unique in the hyper-competitive fund world is our scale. One of the things keeping me up at night nowadays is whether our fund design is taking advantage of the full potential of Q's scale. thanks, fawce @fawce - thanks for stepping in with an attempt to clarify. Permit me to suggest, however, that you're communicating the thing in a sub-optimal way. Announcing prior to "having everything figured out" is great, if you're soliciting "community" input, but otherwise creates confusion and (forgive me) looks unprofessional. You might, for all I know, have been more clearly solicitous of input/feedback at the meetups, but the attendees are a small minority of the "community". That said, I'm sympathetic to the difficulties - legal, business, etc. - which you doubtless confront as you try to work this out: not only with potential investors in the fund, but with the core investors (and board?) of Quantopian itself; probably significant overlap between those groups? Not to mention brokers, etc.... And the "community" on Quantopian is often, ahem, inappropriate to invite into those negotiations. :-) Tricky thing you're trying to do, Perhaps a greater degree of openness, to the extent that your backers, etc. can both permit it and find value in it would help. Or just keep it under wraps until it is "worked out". FWIW, I quite agree with the "real money" bit. Also filters out people who've never actually traded a dime. On that note, @gary - "$100MM in transactions" doesn't mean net profit, it means "money changed hands", most likely including exchange and brokerage fees.

@fawce - A username-and-instrument-redacted trade history of all that trading (including volume) would be intesting indeed. If that could be done, keeping the histories separate (by user), but including if the same user was simultaneously running multiple strats... valuable data and inspirational material for all.

Before anything else, Quantopian is a trading technology platform, and it's a young one. There's a lot of attention on the current lack of bells and whistles, and what that means for becoming a competitive fund. But rather than focusing on "what currently isn't," I'd say the more interesting topics are "what could be." Building a full featured trading platform takes time; equities are all that can be traded at the moment, but as Fawce pointed out, that does not mean equities are all that will ever be traded. Quantopian's model of allowing users to 'build, test, invest' means adding new asset classes is more difficult than forwarding orders on to IB because any new features have to be built into backtest simulations.

In my opinion the managers program is a great idea. Any trading platform could have done this, it just hadn't been thought of yet. Bloomberg could have had the foresight to keep statistics on terminal users and started a fund to allocate capital to the best managers on their platform. The model does make more sense for Quantopian because there's already a big focus on community and collaboration.

I believe the potential is there for the managers program to become a competitive fund. The incentive structure is win-win for the parties involved, and the talent pool is potentially much deeper than that of a traditional fund. Another promising aspect is that this type of fund structure will be self diversifying, historically, hedge fund meltdowns have been the result of putting too many eggs in one basket. Quantopian will effectively have the opposite problem of having more baskets than eggs.

Only time will tell how it all pans out. I'm clearly optimistic, but regardless of what the future holds, Quantopian will certainly be one of the more interesting hedge funds ever created and I'm looking forward to seeing how it evolves.

David

@Fawce, I'd be interested in hearing more about scalability and your fund design. Are you concerned that you'll be excluding talent? Or something else?

Grant

@Grant - I meant that I don't want to limit our access to talent. The size and diversity of our quant pool will determine the scalability of our fund. We've managed to attract members from all over the world, and I want to consider each and every one for the fund.

I would be interested to know which FINRA certifications the member quants are likely to be required to obtain, I would think at least Series 65?

I can't attend the meetups now that I've moved, but I think the incentive structure for individual quants will be of paramount importance. Performance sharing must be strictly on personal algo results, not fund results, with high watermarks, robust risk-adjusted-returns requirements (not Sharpe etc!) and no AUM fee paid. The fund can charge one to cover operations, but it should not be distributed. Also, there must be strict rules about halting or modifying algos - ideally, once started, they are in the fund as-is until they are automatically removed for a risk management or required return violation.

Probably the MOST interesting part about all of this for me is how Quantopian will tackle the asset allocation between strategies, whether they'll go with something simple like equal-weighting or Black-Litterman, or something a little more interesting!

Simon.

My back-of-the-envelope here is that say there are 400 managers, each comes up with $25K-$50K, and shows performance over 6 months. Then Quantopian adds $200K to each of their accounts, which adds up to ~$100M total in capital in the fund. And then Q makes $10M-$20M per year in revenue, generously shared with the managers. Isn't it going to be expensive to keep track of so many managers, or is the idea to have it be completely/largely automated? It just sounds pretty unwieldy at first glance, since there end up being a lot of 'cooks in the kitchen' that have to be monitored. Why not just take the 100M (or whatever the number is), and manage it professionally with a smaller number of people? I expect the idea of the fund is to have fully-automated restrictions on the 'pool' of active algos in the fund. I don't think your IB account gets topped up with the extra 200k from Q, its just that they clone your algo and run it in their fund, and at the end of the month they give you a % of the profit (do you also get a % of the losses I wonder?). They dont track managers as such, just run algos. I might be wrong -- we'll have to wait for full details. If you construct a fund with uncorrelated algo returns then this is very interesting (minimal drawdowns?) and uncorrelated with the market. By using lots of algos you can see which perform best in the long-term, and allocate more capital to those algos. This is the potential opportunity of crowd-sourcing algos that you dont get with a few managers using a few algos. I love this concept. Like Simon's earlier comment, I, too, am interested to understand how Q plans to execute asset allocation. To me, it's not so much about asset allocation, as it is about risk management. In other words, if all the managers are trading stocks and equity ETFs, then that's OK if their algos manage risk well. I think it would be incredible if Q had 1,000+ managers specializing in every niche of the stock market (where there's enough liquidity) where each algo was getting great performance, plus some more moderate strategies to help manage overall market risk/volatility. Also, in the longer-term, it will be awesome when Q gets into FX/futures/fixed income. Some questions: 1. Will Q show daily results of the "master fund?" (compliance issue?) It would be great marketing (to attract new investors) if the performance was shared on a on-going basis for total transparency. 2. Assuming the reason for a hedge fund (and not an RIA) is that Q might charge 0% mgmt fee, but 10-20% performance fee only? I think investors would love that. It would be a great setup with a lot of integrity. But, it's too bad regulations are such that they would have to do a hedge fund, because only accredited investors would be able to invest in it (please correct me if I'm wrong). The small investor that probably needs this the most won't be able to invest. But, maybe they could pressure their employer to use it for their retirement fund instead of their 401k mutual funds. 3. How aggressive will Q be at selling their hedge fund? Traders/Managers will appreciate it if Q is out there selling aggressively (pensions, high net worth, retirement plans, endowments, etc), because as assets grow, more revenue for all. I don't get why I live algorithmm is required, if the simulation is good selecting live algo will just exclude a bunch of people. Also who has access to the live trading algos? How do I know the decision was fair? There is no transparency at all? Why not make a public challenge for an algorithm that can perform and publish results and award the funds' Anyone participate in the Dec. 4 webinar? If so, impressions? Additional info. provided? --Grant %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Hedge Fund Webinar Curious about our new crowd-sourced hedge fund? Jess Stauth, VP of Quant Strategy, will answer your questions and give an update of our progress. See statistics from real-money live algorithms. Join us on December 4 at 11AM EST for a preview to the fund! missed it, is there a video or slide? Grant, Yes I attended and was satisfied with what I heard. I also understand where you're coming from on the live trading requirement. Know that you are noticed and read. We need more contributors like you! You've been very active in the community for most of it's existence and that says a lot. I suggest you hang in there and keep learning. Hopefully sooner than later the team approach will catch on. If you continue to demonstrate your growing proficiency, my guess is you'll be in the first draft (penniless or not). Either that or someone will seed you personally. So I encourage you to keep posting your algo.'s. What better way to demonstrate not only what you're learning but more importantly your commitment and willingness to invest yourself in what David E. aptly calls 'one of the more interesting hedge funds ever created' (see his 10/26/2014 post above). For the record: My take on Quantopian coincides with David E.'s 10/26/2014 post above (only he expressed it better than I would have). In the end it's all about trust. That's the key ingredient that keeps our economy healthy. And so it is here. Best, Tom Thanks Tom, So was there any guidance on the vision for the whole thing? I might be interested in working on something, but without knowing what they're looking for, it could be a very futile, frustrating effort. In particular, was there any guidance on how a given algo will be evaluated? Metrics and their relative importance? Also, will Quantopian be providing money to individual managers (e.g. will money magically appear in your IB account)? Or will there be a loan with a kind of lien on the capital? Etc. Or will the algo be licensed to Quantopian, and they'll take it over, applying the capital directly? If the latter, what will be the role of the originator? Etc. The term "crowd-sourced" was applied, but will there be any mutuality in the fund from the standpoint of the individual algo originator? Will it pay dividends or refunds based on the collective performance of all algos? Or will each originator only benefit from the performance of his/her own algo? What if my algo is running under an IRA account? Would it be eligible? I'm not opposed to doing real-money trading, motivated by this program, but it is still very opaque at this point. Cheers, Grant Patience Grant. They said they'd post the presentation. Then you can get it first hand. Best, Tom. O.K. The roll-out has been kinda strange, though. Above, Fawce states "we want to design the fund with the community's feedback" but there have been few details since the original announcement, and no active online forum for interactive feedback (unless I missed something). Maybe the interaction will pick up as the idea takes shape. --Grant Hi, Thanks for the feedback and the prod to keep the community up to date on our process. To that end, I wanted to share our outline for how we will develop the fund structure and how we will engage the community. There are two major components: 1. Live interaction in meetups and webinars 2. Request For Comments (RFC) We plan to do another webinar this week, and then provide monthly updates on our evaluation and results. We will also record the next webinar and post it with the slides. The version 0.1 RFC will be on the blog before the end of the year. The biggest point of internal debate ahead of the RFC stems from the first and most prominent piece of community feedback: Why is real money trading requirement for evaluation? thanks, fawce Grant and anyone else who's interested, We'll be having 2 more webinars about the crowd-sourced hedge fund. They will be offered on: • Dec 10 (Wednesday) at 11:00AM EST. Sign up for this webinar here. • Dec 16 (Tuesday) at 1:00PM EST. Sign up for this webinar here. This will also be recorded (and posted) for anyone who can't make it! We're looking forward to chatting with you soon. Disclaimer The material on this website is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory services by Quantopian. In addition, the material offers no opinion with respect to the suitability of any security or specific investment. No information contained herein should be regarded as a suggestion to engage in or refrain from any investment-related course of action as none of Quantopian nor any of its affiliates is undertaking to provide investment advice, act as an adviser to any plan or entity subject to the Employee Retirement Income Security Act of 1974, as amended, individual retirement account or individual retirement annuity, or give advice in a fiduciary capacity with respect to the materials presented herein. If you are an individual retirement or other investor, contact your financial advisor or other fiduciary unrelated to Quantopian about whether any given investment idea, strategy, product or service described herein may be appropriate for your circumstances. All investments involve risk, including loss of principal. Quantopian makes no guarantees as to the accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances. Thanks Fawce & Alisa, If you put some amount of money toward algos that haven't traded real money, it would be incentive to bring the Quantopian analysis and simulation platform better in line with real-world trading. If you are putting up money, you'll want to minimize the risk of differences between the predicted and actual results. So, it would motivate sorting out in detail how to improve the tools. The fact that you are hesitating and debating suggests that there are gaps. Grant Grant, You're talking about designing incentives that align our interests with improving our product. I'm all for that. But, that's not what we are debating internally. Our selection method will use the simulation platform, whether we allow paper traders or not. So, the incentives you're describing already apply. The debate is whether requiring the quant to trade his or her own money is a useful filter on algos. The two sides of the debate are: 1. Quants are more likely to be careful with risk if they are required to risk their own capital. This is really about incentive alignment for the quant. Very similar to your argument above, but applied to the quant and the algo versus the platform. 2. We are unnecessarily limiting the number of eligible quants. The conclusion I've reached is easy to say: I want both. I want any quant capable of making a great strategy to be eligible. I also want quants' incentives aligned with producing consistent returns, rather than taking excessive risk. I think we can come up with a way to do both. stay tuned, fawce Grant, First off I would like to say that I really respect you and the amount of work that you have put into this community. I have learned a ton from your posts. With that said, I have to agree with John and team with their decision to base it on real money trades. I agree, I put my money in an algo that I really believe in and I'm very careful with that algo while others I let them roll just to see what happens. I do think that there could be more exposure to the community. Here are a few suggestions on how I think that could be done: 1) Have a leader board and where a quant stack up against the leaders. 2) Allow quants to publish the performance of their algos 3) To point 2, allow outside investors to invest directly into algos. This is really about incentive alignment for the quant. This really seems more suited to a prop firm trading model than a hedge fund. If the goal is to enable quants with only say10k in capital to spare to be able to partake in profits on say 500k notional (assuming $15M split 30 ways), the incentives will be skewed. I suppose one could treat it like this: Algo writer gets 100% of her equity profits, 50% of the profits of any pool money which is treated as direct leverage (against which her equity is a first-loss tranche), and pro-rata (stake/pool) percentage of the profits from any pool money for which she has no obligation to absorb losses. The designer gets to choose the direct leverage they want, up to some limit (10x ?). This part is really important: the LLC pool must get first execution allocation on all algo orders. Then the direct leverage gets the next allocation, and the writer's equity gets the last allocation. This is vital to prevent people writing algos that front-run their own LLC pool. Besides, there is deep inherit risk in all of these thousands of lines of python code and somebody had to have analyzed them for holes and trojans. Yeah this would be my fear, but this really must be solved by incentives. There aren't really any technical solutions for malicious or naive algorithms, whether or not Python was the best choice. Simon. w.r.t python code and risk, it is interesting that test vectors and scenarios tests aren't a readily accessible to others. If others have thought about this or would be willing to share their test scenarios, I'm always interested in finding strange trading cycles and stocks to throw at my algs. I went to your Meetup earlier tonight. I had some thoughts on the drive home, and offer them up now in the hope that they might be fair recompense for the pizza I ate. :) For your fund to be successful, you need to do several things: One, build tools to serve your individual quants; two, get good at picking quant strategies to include in your fund (manager selection); and three, run your own fund and investment portfolio well. You've been working on number one for years, but have probably just gotten started on two and three. For picking quants, Marcos Lopez de Prado had several papers that are basically on firing managers, quantitatively. E.g., if you observe a Sharpe ratio of 1.5, how much longer should you wait and watch before concluding that the true Sharpe ratio is actually <= 0, or >= 0.5, or the like? Nothing earth shaking, but definitely useful for informing your approach to manager selection: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2254668 The High Cost of Simplified Math: Overcoming the 'IID Normal' Assumption in Performance Evaluation Marcos Lopez de Prado; April 21, 2013 http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2150879 The Sharp Razor: Deflating the Sharpe Ratio by Asking for a Minimum Track Record Length Marcos Lopez de Prado; September 23, 2012 http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2201302 Drawdown-Based Stop-Outs and the 'Triple Penance' Rule David H. Bailey, Marcos Lopez de Prado; April 2013 Next on to running your overall fund's portfolio. (Maybe you've heard all of this before, but in case not...) I was thinking that your hedge fund would be pretty much like a fund-of-funds. However (as I suspect Jess realized), that's probably not quite right. It's more like a proprietary trading firm with multiple trading desks, or perhaps a large buy-side investor who hires multiple independent advisors. (But unusual in that all your portfolio managers trade the same broad asset class, and use a shockingly unified set of tools, although perhaps entirely different key algorithms.) Since each desk (quant algorithm) is totally independent, they can trade against each other, and certainly aren't going to do what the optimal single manager would do if he knew everything that each of them knows. If you're a fund-of-funds, you don't know what's inside each trader's black box, so all you can do is try to pick your cash allocations among the traders as well as possible, you can't do anything about making their trades more efficient. But as a prop firm with multiple desks, you can do better; assuming you're willing to write code to implement it (maybe a lot of code). The standard Grinold and Kahn advice goes something like this: http://www.amazon.com/Active-Portfolio-Management-Quantitative-Controlling-ebook/dp/B005C3WULW/ First pick a risk model you're happy with. E.g., the Northfield US Equity model would likely be fine (check though if they include all those weird ETFs some of your users seem to like), and they're also in Boston. Or write a simpler version yourself if you have the time to figure it out. Then take all the trade decisions from each of your quants, but do not execute them. Instead, using your risk model, take each quant's trades (actually, his portfolio weights) and back out their implied alphas using a reverse optimization. Now you know each of your quants' alpha ratings of different securities. (With a different risk model, they would come out different.) http://www.portfolioprobe.com/2013/05/20/implied-alpha-and-minimum-variance/ Next, feed both those implied alphas and your risk model into a portfolio optimizer. Ideally, your optimizer should handle liquidity and trading costs, plus whatever constraints you think are necessary. If you get fancy there, you will need a specialized nonlinear optimizer, but a simpler quadratic optimizer like de Prado's Open-Source Critical-Line Algorithm MIGHT be enough to be useful (I'm not sure). Finally, actually do the portfolio trades emitted by your optimizer. You should get better performance than any naive black-box weighting. (How big, I'm not sure, I don't have any numbers handy.) You'll certainly save in trading costs, and likely benefit from the risk model's rough understanding of the covariance between securities. You may even find that your master synthetic strategy is largely uncorrelated with its component quant strategies. It should work fine if the average holding periods of all your quants' portfolios are similar. If they are dramatically different (intraday vs. months), well, you have multiple time periods so you're supposed to use dynamic programming instead of a single-period optimizer. How far the holding periods need to diverge before it matters, I'm not sure. I have seen versions of this work in the real world. Whether it would really be worth the effort for you, well... I don't know. It's probably worth reading more about, at least. A practical problem though, especially given your transparency goals, might be how to then allocate gains back to your component quants! I don't recall if any authors directly address that, but even if they did it certainly adds yet another layer of complexity. A few thoughts: • What are the tax considerations here? Is there any way that the managers/algo originators/fund participants/trading desk operators (whatever they end up being) can gain any tax benefit? • Would an IRA account at IB be eligible? To Simon's point above, how much cash do folks have sitting around to put toward this sort of thing, unless retirement funds are included? • Along the same lines, would a more modest asset allocation type algo be of interest, targeted at the long term (20-30 years)? After 6 months, re-allocating every month/quarter, there won't be much data to mine, so supposing it qualifies, it'll still be matter of relying on offline analysis/backtesting/optimization. Or is this program geared toward the shorter term? Grant Why not incorporate some of the ideas of covestor? To have funds most of the time managed by an algorithm without human intervention is very limiting. GK, To address some of your questions: From a US perspective, I don't think there is any additional tax advantage unless you are already offshore. I am assuming that the hedge fund will be set up as a Master-Feeder, in order to issue both taxable US securities as well as non-taxable offshore securities, depending on the location of the investor. IRA's are a little tricky. They have to be self-directed (no third-party custodian) and there can't be any self dealing, meaning that you can't invest your IRA in a hedge if you are an active manager, like a quant that is scripting algos for the fund. Regardless, in order to invest in the first place, you need to be an accredited investor, which means that you make over$200,000 per year for each of the last two years, or have a net worth of over $1M. I think it is safe to say that most of us would not qualify under that criteria. @Grant, here's my 2cents on your questions: • On tax, I'm not a tax expert, but I'd say no...managers would not get the tax benefits of being a hedge fund manager unless Quantopian requires each manager to have a series 66?? • I'm not exactly sure how Q will set up the fund, but I don't believe IRAs will be eligible. Also, Hedge Funds are for accredited investors only. • From what I understand, if your "more modest asset allocation type algo" has a good track record then it could qualify. From what I understand, Q is not just looking for "short term" strategies. @Lucas, I know personally as an investor, I want 100% algo because I want a disciplined process managing my money. I think a lot of investors want this, too. With Q's ability to use fundamental and technical analysis together in algos...that's pretty awesome...with the right algos together, I'd put my money on the algos vs. using a collection of human wealth managers/traders. Just my opinion, but let's see how it evolves over time. @Lucas - there is nothing as of current preventing bot authors from coding in the ability to be manually controlled/directed. In fact, this is something that may need to be explicitly disallowed by technical means if Quantopian wishes to have 100% algos only. I listened to the recorded webinar, and it did not exclude IRA accounts (but they weren't mentioned explicitly as eligible, either). Guess this detail will eventually come out. --Grant I think it would be great if IRAs were eligible, but it would probably require Q setting up some type of feeder fund where the IRA would hold equivalent shares representing their investment. At the large wealth management firms that offer managed furtures hedge funds, for example, that's how they do it. For non-accredited investors saving for retirement, I think it would be awesome if they had the choice to invest in Q's fund through their employer sponsored retirement plan. Q's fund would be costlier than Fidelity/Vanguard/TRowe plans, but I think Q's fund would represent excellent value because of excellent risk managed, consistent performance that retirement investors covet. More thoughts: • The webinar included data from live Quantopian algos trading real money. Did the traders need to give permission for their data to be analyzed and presented? Or is it in the terms of use that Quantopian has the right to mine trade data and present it publicly? It sorta took me aback that although there was no attribution, the time series trade data were presented. • If Quantopian can mine trade data, then presumably IB (and other brokers) can do the same, and could play the middle man of connecting high-performing traders with potential investors (or maybe they already do this?). • For potential investors in the Quantopian fund, there will be some sort of prospectus, fund document, analyses, etc. Will these documents and data be available to the individual Quantopian managers (algo writers)? To the Quantopian community at large? Anony (with another post evaporating, yet again...huh?), As Jess presented it, at some point Quantopian was engaged by one or more investors interested in putting up money, if they could get access to the data. Quantopian did the math and realized that the best business decision would be to get in between individual traders and the outside investors (and be able to charge the typical 20% on profits, I'm guessing, with a portion going to the algo writer). The message I got is that there is a pot of money out there wanting to put money into a portfolio of algos, and if they pull together a basket of 5-10, Quantopian can make some money. Out of 28,000 current Quantopian members, at most 10 will participate near-term (0.04%)--insignificant, I'm realizing. And the sense I got is that this train has left the station--Quantopian is gonna do it basically the way they described on the webinar, if they can figure out the legal end of things and the procedural mechanics. To me, the other end of the spectrum is more interesting. How does someone start out on Quantopian with$0 personal capital (either unavailable or unwilling to invest) and get off the ground? How can the majority of the 28,000 users benefit from Quantopian's access to investors and the financial community? I think that the argument that one has to have skin in the game is not necessarily valid. I'm sure that there are lots of researcher/analysts, algo developers, and programmers in the industry that do a great job, and just get paid a salary (or charge a consulting fee). There ought to be a way to monetize such skills on Quantopian.

Grant

@Grant, regarding your comment "I think that the argument that one has to have skin in the game is not necessarily valid."

I disagree. I think for reasons of integrity and marketing, Q needs to be able to say to potential outside investors that all quants' algos being used are actually used on quant's real money. I think in the webinar Jess eluded to maybe Q doing a prop trading situation (at a later date) to help quants establish a real track record...but, I got the impression that in the near term, if a quant wants to be considered for the fund, then they need to be trading an algo with real money.

Also, I agree with your concern about only ~10 out of 28,000 being in the fund. However, I got they impression that they want that number to be much higher. I think their ideal situation one day would be to have several funds (based on different risk tolerances) where thousands of quants' algos are being used in the funds, with thousands more trying to get in.

Hi Jeff,

I agree, for what Quantopian is trying to do with the fund, it makes sense for individual managers to have capital invested. From the webinar, I gathered that Quantopian is basically starting to put together a story for potential investors that they have every angle covered, including aligned incentives for the managers.

What I think Fawce may be referring to above is the scalability problem at the low end. How many people out of the current 28,000 have ~$50K (or even some number larger than a few hundred bucks) to spare? The scale could be much larger if the poor, clever, ambitious folks could be hooked up with the folks with cash. This is where automation, the scale and reach of the internet and modern computing power can do something really different. Grant Hello Anony, So what's your motivation in deleting your posts? You tend to have insights and a unique flair. If you are wanting to participate in a communal conversation, it would be helpful if your posts would remain visible, from my perspective. Grant Well, I think I get the gist of your comment (I have some concerns, possibly shared by you, over Quantopian which I will direct to Fawce privately). My assumption is that you are the one doing the deleting, although it is reasonable that Quantopian would do some level of screening. Do whatever you want--I can always refer to my e-mail inbox, but it makes for a very awkward discussion thread, especially for anyone new wanting to learn and to contribute. --Grant There is likely to always be some tension for those folks that work, have worked or will again work in the industry. It's why so many do their social media under an alias. There are some highly credentialed lurkers here. My 2 cents. The universe of strategies that have great backtests is infinite. The subset of those strategies that will be perform in the future is comparably tiny. Learning how to develop a strategy that can give you a good backtest is a separate skill from categorizing it as production-testable. Note that I didn't say you can prove it'll work in the future. A lot of leveling up is killing your good backtests without having to try them live. Even with best intents and due diligence, your strat can be lackluster once live. You kill way more ideas than you try. This is a separate issue from backtesting accuracy and production shortfall. Even with real trading results, whether someone/something will keep performing in the future is an area of constant research. Having real money trading isn't just a signaler of dedication, but it's generally the tuition required to learn the real hard lessons; like how illusory backtests can be. That tuition is paid by failing and losing money. It's like the adage where you need to have your heart broken once before you're ready for real relationships. Not an absolute rule, but mostly true. That being said, some of this could be sidestepped by having better automated tools to sanity test algos. Things like automatically testing against other instruments, resampling, etc. My intuition is that any strategy worth investing in can find capital, it's just that most strategies are not. There are a variety of use cases here. Personally, I'm interested in a fully self-funding approach, starting from ~$0 (o.k., I could skip going out to lunch a few times, to chip in $20). It's a hacker/maker/tinkerer approach, versus what Quantopian is pursuing as their primary path, which has an inherent financial barrier to entry (e.g.$Nx10^4, where N = 1-25, which I recognize is still considered small potatoes in the financial world). The problem is that the whole system doesn't scale as it could theoretically, I'm figuring both because of regulation and reporting requirements, and transactional costs that have not been automated away.

The trouble is that if there is no quant skin in the game, there is huge incentive for duplicity and the investors have an principal-agent problem.

As it is, it's pretty tempting to shoot for the best backtest, which is unknowably uncorrelated with future performance. Finding potential strategies should be viewed as the losing end of a long and arduous battle to prove the strategy is bunkum. Instead of looking for diamonds that stand out from coal, it (to me) feels more like polishing every bit of coal down to dust, and hoping that there is something left on the table at the end.

Hello Simon,

I don't have a problem with having skin in the game (the argument makes sense). If I could make \$1 tinkering around on Quantopian, I'd gladly set up an account at IB, and if someone wanted to match my dollar, great! We'd be matched dollar for dollar, with our interests aligned (although after one trade, we'd each have 50 cents left!). For me at this point, it just doesn't make sense to use money that I've obtained via employment, etc.

Grant

You can unfollow the entire thread with a button at the top-right of the page.

Hello Jonathan,

Apologies if you've found me overbearing (your deleted post followed mine above). I've tried to stay on-topic for the thread, but perhaps discussion around 'trading for the masses' is pointless, since (although I could be wrong), it's not where Quantopian is headed. They're moving upmarket with the fund idea, from the way it was pitched. Bummer.

Grant