If they decide to do it organically from their own sources, I'm sure someone in the Q ranks has some inside view of how HF's work and can be guided into the structure of one. I'm also assuming that this structure has already been in place since they have investors already starting to fund it as well.

for 10b fund would it really be worth to go through IB? Only someone that have experience with a professional software like front arena or Blomberg would be able to compare. But I suspect Q is directed to algo trading not much fund optimization. I would say it misses a lot of risk models, and monitoring features.

Startup funding is based on milestone's that show your funders you're able to progress somewhat linearly toward your goal. Each milestone generally results in the next funding round. IB is clearly an example of a proof of concept; you'd be insane to run a $100M hedge fund with those clowns, let alone a $10B one. However it is a convenient platform with an API that lets them show that the more innovative and thus higher risk portion of their concept works, i.e. that they can attract algo writers, that the algos they come up with don't suck... The actual writing of the code and establishing the connections to the market is trivial in comparison from a risk perspective since its been done hundreds of times before. The people who are funding Quantopian will be happy to open the checkbook to do that once everything else has been proven out. The good news for us is that it means the software and market connections, which are already quite good especially considering what we're paying for them, are only going to get better. The challenge for us is to prove to Quantopian's funders that we collectively can write algorithms that don't suck!

So far, I gather that IB will need to be scrapped, and maybe minute bars. Anything else? Will Python scale?

It won't matter if Python scales or not...it can be translated. The cost of such would be trivial for a $1B+ fund...speaking as if I had an ideal algo and a significant interest in the fund.

Re-building all technology if needed wouldn't be a problem and Python is certainly not going to be a bottleneck unless they are planning to go for latency-sensitive strategies (which with their technology seems unlikely). In my opinion, their real limitations stem from two sources. On one hand, it seems that the founders and early employees are smart technologists who don't necessarily have experience in the trading world so it's hard for them to know the right way to do things. On the other hand, they need to be usable by the general public, which places limitations on both the kinds of interfaces they can offer (complexity scares people away and makes it hard to use) and the data it can offer (the minute-bar data without even a bid/ask, let alone a full book, for example, I suspect is a result of wanting to save $ on licensing) .

A while back, I would've said that the lack of an interactive research platform is another critical flaw but they are clearly working on that, even if the current offerings are quite limited. Another thing they could probably use (and can easily build) is an execution engine which can take in alphas and would automatically place trades to monetize them, run portfolio optimization algos to estimate proper positions, work on carefully executing into those positions, etc. Some algos won't fit in this model but most (capital intensive, non-latency sensitive) ones will.

Ultimately though, I think the biggest question is whether they can get enough good algos from the community and whether they can probably identify them. Finding good signal is very hard work even for experts with good research platform. Furthermore, a lot of value stems from using novel sources of data, which the current model doesn't really allow. Will John Q Public be able to come up with anything decent and high capacity? And will Quantopian be able to identify the few pieces of wheat from the mass of chaff? In the space of strategies they are looking at, sharpe 1-2 is quite good -- but those strategies take years to accurately evaluate and backtests are flawed.

Thanks all,

If I'm reading things correctly, the present system wouldn't be competitive at $10B, but I understand the rationale for what is offered today. It's one of the mysteries of the Q how this whole thing is supposed to work as a business. I gather that the present focus is to get to a proof-of-concept level of $25M or so, with $1M-$5M per algo in the fund, and then start to scale from there. But then to keep it truly crowd-sourced, so that each manager doesn't need to become a professional, at $10B, you need about 3000 managers at an average of $3M of capital each. But if a much more sophisticated platform is necessary to be competitive with other billion-dollar hedge funds, you need 3000 sophisticated managers, which sounds unlikely and even if it were possible to recruit that many, it would be seemingly unwieldy (and are there 3000 independent arbitrage strategies out there?). The amount of capital per manager could be increased, to reduce the number of managers, but then the whole concept of crowd-sourcing falls apart, and Q in its present form ends up as a recruiting tool for hiring full-time managers or for collecting algos that could be eventually licensed and fully researched and developed by professionals. Or is my logic flawed?

Grant

My first interpretation of their business model was, trading is becoming more and more automated, markets are efficient and active trading becoming harder and harder, we provide a platform that is the best in algorithmic trading and we become indispensable. But since then, the focus has shifted a few times. Seems that less time is focused on developing the platform and going in other directions.

It's about getting a bunch of good algos which have a collective capacity of upto $100B. Then raising $10b and running it becomes a doable task. HFT kind of stuff is not possible with this infrastructure. But that's okay. The emphasis on US equities alone could limit the scale. A lot more can be done with futures and options than with equities, like stat arb using options. Currency is a big market. Bonds are an opportunity. Some commodities are big markets. Emerging markets offer a lot of tradable inefficiencies.

Equity markets have this problem - When they doing well, they excite investors. But investors don't need alternatives when benchmark SPY does well. When equity markets are in a bear phase, investors are not excited in the first place.

It's the path from a low-level prototype phase ($25M-$100M) to an actual hedge fund business I'm wondering about. At the $100M level, Q might make $1M per year (I'm assuming 10% return, 10% would go to Q and 10% to the crowd-sourced managers), which is not nearly enough to support 30 employees (maybe 3 employees?).

So, as things are laid out now, we have:

1. Crowd-sourced hedge fund, with the opportunity for lots of amateur, but potentially capable managers to participate at the $1M-$5M level each. Novel, cool idea.
   
2. The need to scale to $10B in overall capital, for the fund to be a business.
   
3. A platform that may not make sense at the scale required for the hedge fund to be a Q business.
   
4. Each crowd-sourced stategy is a black box to Q, with code that is only accessible to the owner.

I'm assuming that the folks who provide money to Q are asking similar questions (or maybe I have the business model all wrong). And folks within Q may be wondering, too ("Hey Fawce, how's this thing supposed to work, again? I'm not sure I follow."). It will be the world's first crowd-sourced hedge fund, but if it isn't a scalable, viable business for Q, then it won't be a hedge fund for long. It is just hard for my wee litle brain to take it all in and formulate a coherent picture.

Reading about each entry here: https://www.quantopian.com/about

Quantopian is funded and advised by Bessemer Venture Partners, Khosla Ventures, Spark Capital, GETCO, Blake Darcy, Matthew Granade, and other angel investors.

the three first VCs are primarily tech investors, but GetCo and the others are all financial people. So presumably Quantopian understands the challenges. From what little we know about this gap:

[The Q] -----------| big gap |-----------[Hedge Fund]

We can surmise 'til the turkeys roost, and still not have any clearer picture of what or how the Q will evolve. From our limited viewpoint it looks like that gap is substantial. But for all we know the VCs will just sell off the Q to some big 'ol fund as their exit strategy seeing how the big returns are probably not in the cards. I mean a Quantopian IPO? Not likely. And every VC's dream is to exit their investment in an IPO. Barring that, a sale to one of those $10B+ funds seems likely.

These are the questions I ponder every day, all day. And most nights. I take great comfort knowing I’m not the only one!

Kevin Q said it well, we are betting on the community to produce quality algorithms.

For Quantopian, quality means we can have a high confidence that a strategy will have a positive Sharpe ratio and low correlation, after we allocate capital to it. Think of Quantopian as a boost algorithm in machine learning. As long as we find algos that are marginally better than random, we can combine them into something compelling. Imagine if the community produces, as I expect they will, algorithms that are much higher quality than that bare minimum – the combination will be remarkable.

We are working on guidelines for algo writers, to help translate our fund goals into more concrete direction for the community. The guidelines are a work in progress, but here are some of the key characteristics we want:

• diversified holdings
  
• low net market exposure (balanced short and long exposure)
  
• actively trading
  
• low beta to broad market(s) and risk factors
  
• low correlation to the rest of the portfolio
  
• low leverage (so the fund can control total leverage)
  
• low cash balances

My goal is for the guidelines to be concrete, but not a cook book. I want to be sure that we are leaving room for this community to do the creative work.

All of this depends on Quantopian providing the best algorithmic finance platform in the world.

For the platform, I chose a breadth-first product strategy since inception. I wanted to have the complete workflow from idea to live trading as quickly as possible. Happily, I think we hit that point this quarter — you can now research (beta), develop, backtest, forward test, and real money trade on Q.
Starting now, you’ll see us shifting focus to add depth to the platform. We’ll add more markets, more data, and flesh out our APIs. We’ll support larger portfolios, parameter optimization, and code reuse.

I’d like to address Grant’s concern that we can’t be a $10b hedge fund and a community driven fund at the same time. We will do both.

The key is that Quantopian will manage trading operations. That means our trading desk will cope with algos failing to borrow for shorts, market closures, stock halting, rebalancing among algorithms, and all the myriad daily battles to keep all our fund algos running smoothly.

That means a person can have an asynchronous relationship with the algorithm - pace your work to research and create an algo in a way that fits your life, without facing the prospect of maintaining that algorithm 252 trading days a year. Q can be your side-hustle, second career, hobby, or primary source of income. Scientists, professors, stay-at-home parents, engineers, and professional quants can all work this way. Maybe you have one idea a week, maybe you have one idea a year, but you’ll always have Q ready to back the best ideas.

The platform is there to help you create fully automated strategies. We’ll be taking the responsibility and risk for allocating to your strategy. So once it is written, why couldn’t a community algo writer have $5, $10, or $100M allocated to his/her strategies? The same argument applies to running smaller allocations - with a fully automated strategy why not run $100, $50, or $10k?

As an algo writer, you invest your time and you trust us with your ideas. In exchange, you don’t have any expenses to develop your strategy, you don’t need to take on the risks or effort associated with operating it, you don’t need to raise capital to fund it. But, by virtue of your efforts and creativity, you’ll share in the upside.

Thanks for being a part of Quantopian,
fawce

That's inspiring, growing organically, adapting, the puzzle-solving of working thru unknowns.

A question:
I think there is a certain phenomenon that affects all investors, and it is so common that it might have a name and yet I haven't seen discussions on it, and if I mention this with friends their eyes glaze over. Imagine a sizable 80% profit on $50K after one year so now the portfolio is worth $90K. The investor says, now I'm going to remove my original investment, use that to fund a Startup and only be in the stock market with those current profits, $40K. Their very next thought: Wait a minute, if I remove the original, then in another year the portfolio value might be up from that profit of $40K to only $72K (adding 80% profit again, as $32K) or if I let the $90K ride then it could be $162K instead (80% profit again, second year). The 162 sure looks better than 72 at 225% more. Decisions decisions, something that affects our world a lot. It's merely which investment vehicle/risk/result one goes for, the Startup or stock market, that was background and the question is simpler ...

Now applying that to the fund. A quant joins the hedge fund with an algo of high quality and Quantopian invests $1M into it.
When either Quantopian or the quant decide to take/remove some of the profit, that's money that would have contributed to profits for both of them going forward so I'm wondering whether those transactions might have restrictions (anything interesting to talk about) or if the math is straightforward for easy going, thanks.

Thanks Fawce. At least there's hope for me; if I can just write an algo that is "marginally better than random" I'll be all set! : )

One thing to consider is that it might make sense for there to be a way for an algo writer to license the content of his algo, the kernel of the strategy. I know that the code is sacred at this point, since you are wanting to build up trust. However, $100M is a lot of money to have flowing through a black box, controlled by one person (who, in a crisis, might not be available). And at that level, having a professional re-write the original code per a standard, modular structure would make sense.

The platform has improved dramatically. It would be interesting to benchmark it against what a typical big hedge fund would use, both for research/development and trade execution.

With all due respect, the sentence, "That means our trading desk will cope with algos failing to borrow for shorts, market closures, stock halting, rebalancing among algorithms, and all the myriad daily battles to keep all our fund algos running smoothly," doesn't sound as if it were written by someone who's actually done that.

Technology, financial infrastructure and rules are changing rapidly, but I'm still willing to hazard some predictions.

1. This idea is not scalable beyond $100 million as black box trading algorithms for a number of regulatory, technical and investor reasons; and the maximum size is likely to shrink in the future rather than grow.
   
2. The obvious alternative is to build a portfolio construction system that takes the individual algorithm trades as signals. Then you can build all your compliance, risk management, cybersecurity, trading optimizers, model validation, account management, hedging, drawdown control, audit and other functions on the portfolio system without knowing anything about the algorithms.
   
3. Building and maintaining the system in 2 is expensive, tens of millions of dollars per year. The biggest quant investors do it, but smaller shops avoid the necessity by combining their algorithms and portfolio construction in an integrated process, which is much simpler and cheaper. Quantopian can't do that easily if it remains committed to taking ideas from the crowd and leaving ownership of the algorithms with the creators.
   
4. Once you convert algorithms to signals, you will find that some algorithms that trade well on their own do not help in portfolio construction. The nine highest paid players in major league baseball consist of five pitchers, two first basemen, a third baseman and a catcher. That doesn't make a good team to field in a game.
   
5. Similar to (4), the best algorithms will not be optimized to use as signals. For example, rewarding zero Beta is nice for evaluating the quality of the algorithm, but some signals would run better if they took more positive or more negative Beta, and you could combine these to get zero Beta at the portfolio level. Limiting drawdowns is another test of algorithm quality, but if the algorithm hedges the overall portfolio, it might make sense to let losses run as long as they are more than offset by gains in other strategies. If you don't do that, you could be canceling your insurance just before the disaster happens.
   
6. (4) and (5) are going to make it tempting to offer a variety of funds that combine different algorithms in different ways in order to make use of all your profitable algorithms. Some funds might use only a single algorithm, and be used by investors who diversify on their own. Other funds might combine large numbers of algorithms for investors who want a one-stop investment. Some might be tail hedging funds, some might target Beta of one, some might target constant volatility. Moreover investors will ask for funds with special rules like no tobacco or defense stocks, or Sharia compliance, or UCITS, or onshore/offshore, or different volatility levels, or different liquidity terms, or different currencies of account. Offering a variety of products makes it much easier to grow the business, but it then requires a meta-portfolio system, because you can't trade each fund as if the others don't exist.
   
7. If Quantopian goes this route, it would make sense to hire a team of professional analysts who comb through the signals determining how best to combine them. It will also need professional portfolio managers to manage the portfolio-level considerations, like counterparty risk, liquidity risk, redemption risk and so on.
   
8. At a $10 billion size, your trading strategy is dramatically different than at $100,000.
   
9. One alternative to the portfolio construction system would be to maintain a single model portfolio, combining trades from all successful algorithms, and let individual clients trade toward it as they see fit; some rebalancing quickly, some slowly, each one adjusting for their individual constraints, and running at preferred levels of volatility, leverage, Beta and other parameters. Of course, clients don't pay the same fees for that as they do for a fund, but the overhead is far lower.
   
10. Another alternative would be to run each algorithm as a separate fund, letting investors pick and choose which set they wanted to buy. This also has far less overhead; but you pay higher transaction costs and are less cash efficient.
    
11. The final alternative I can think of is the one Grant mentioned, buying the algorithms (probably with some kind of incentive contract that rewards future performance) and folding them in to a centralized system. This is intermediate in overhead between having a full-fledged portfolio construction system and outsourcing the portfolio construction to clients/not constructing portfolios in the first place.
    

Men were sent to the moon by people who had never done that. :) :D

Yeah, but they sure sent a lot of smaller unmanned rockets out to various closer places before they went to the moon! Which is to say I agree with you garyha, audacious goals aren't unreasonable, but you need to take a bunch of small steps to get there. Which is what it appears Quantopian is doing. I do think Aaron is onto something with the "doesn't sound as if it were written by someone who's actually done that." statement though. I too feel Quantopian's staff is short on domain expertise in that area although I know they know that and are working on filling it out. Its pretty typical startup mentality, which I've been guilty of myself, to assume that smart disruptive technocrats can overcome anything and to focus on the fun tech stuff we enjoy while pushing off that boring compliance and back office stuff for later.

Yes, it is possible to get to the Moon. But if someone wrote, "We'll fill a cylinder with explosive propellant, put at astronaut at one end, point it at the moon and ignite," you'd say, "That's the right idea, more or less, but there are a lot of difficult details."

I agree, the spirit to get to the moon required people to make sweeping plans and figure that the engineering details could be addressed one at a time. In that sense, there's nothing wrong with saying the trading desk will take care of everything.

On the other hand, looking at existing quant hedge funds suggests some educated guesses about choices that will have to be made if Quantopian grows to $10 billion assets under management.

So it seems Quantopian is becoming just a fund with cheap high qualified labour?
Besides the open/crowd community, there is not much innovation in the platform. it is actually a bit behind existing system and the technology is not the best to scale. So the community is its biggest resource and their ability to attract good Quants. Hope the winners make some good money because a thousand or two might attract a few but i doubt it will attract the top guys.

The Q Tribe has discussed this topic over multiple posts for half a year or more now. Generally I believe they listen to this discussion with only half an ear as they've got gobs of available resources to tap when the time comes to actually build out a real fund. And no doubt there are dozens of factors that will enter into actually accomplishing this move to becoming a fund (the BIG GAP I mentioned above).

But I look at this (this being Quantopian) and the first question I ask is, how are the VCs and angels going to get their $25 million back? Right? I mean they're the driving force here. What is their exit strategy? Do they think that the Q will achieve a $50M valuation? 100M? 1000M? Will the infrastructure they've built be worth even the $25M someday? Will some tech/finance research arm come along and offer Spark and the rest enough to take off their hands? Little VC dependent companies like the Q get folded into, subsumed, or dissolved everyday -- all depending on whether the funding "partners" are having a good month or not. So is the VCs exit strategy based on valuation, revenue stream or IP? That's the real question.

To some or all:
Quantopian is spending a lot of money and offering the service all free so far, it is a something-for-something system and I believe the goal is fair exchange to benefit both.

If one thinks they are not likely to win the contest (I know the feeling), take heart, there are lots of ways to make money on Quantopian.

All across life, or especially in the bucket called innovation, there's that particular phenomenon known as crab mentality best to avoid doing, yet an interesting read. Thanks.

"they've got gobs of available resources to tap when the time comes to actually build out a real fund."

I don't think this is a wise attitude.

People tend to think that the hard part of building a successful asset management business is getting ideas to beat the market. That's actually easy. There are lots of them, many have been published decades ago and still work. Or you can find your own. The hard part is combining them into a good business.

Some ideas have very high Sharpe ratios, they basically never lose money. But they tend to have low and erratic capacity. You can build all the infrastructure and raise the capital, and then watch for years when no opportunity arises. Even when you get a spate of deals, you can't put your profits to work in further deals to get exponential growth.

Other ideas have large and consistent capacity, but they tend to have lower Sharpes. A Sharpe of 0.5, for example, returns less than the risk-free rate in 31% of years, and 13% of five year periods. It can take a lot of time to build up a saleable track record, and any a number of events can derail things before success.

So you decide you'll combine a whole lot of ideas to get large, predictable capacity and to increase the overall Sharpe. Great, but now you've got a lot of complexity and leverage; and big start-up expenses before you earn any revenue.

Obviously it's not impossible to succeed, but most people who try fail, and rarely because they didn't have good investment ideas.

I see two possible reasons Quantopian might bypass the traditional obstacles. One is if it can identify a regular source of good new investment ideas. If it could generate a good five year track record, even with small capacity ideas, it might mean that investors would trust new ideas coming out of the pipeline with less evidence than they would require for an uncredentialed new idea. It would function more as an incubator than a hedge fund.

The second is if the community of part-time quants can produce ideas outside the existing envelope of capacity/Sharpe. I don't think it would be enough to match existing big quant hedge funds, I think you'd have to do better.

It's an interesting experiment, but I don't think you can just find the good investment ideas and figure you'll decide how to monetize it later. The vision of what Quantopian is trying to build matters for how ideas should be evaluated and incentivized today.

Landing men on the moon was an application of science and engineering. Sure, there was risk, but it was basically a matter of "If I do X, I'll get Y." Q is different. There is the technical aspect, which is not really where the innovation comes in. Above, Fawce says "All of this depends on Q providing the best algorithmic finance platform in the world." The algorithmic finance industry has been around awhile, and my read is that it can be very lucrative. So, I figure your typical big honkin' investment bank/hedge fund/what-have-you has invested lots of money into R&D resources and people to use them productively. And the mechanics of executing trades is nothing new--lots of infrastructure there, as I understand, well beyond what Q offers today. It is not really innovation in my book to offer a slice of those resources to the masses. It is more of a re-packaging. And the basic premise is that if Q could just offer better resources (i.e. the same ones already available within the industry), the whole enterprise would just flourish. We'll see how things play out. It is just not obvious if it might be better to hire 20 or so highly qualified researchers/quants/algo developers, put them in a room with top-notch resources, and see what they can do. Would we have ever made it to the moon if the space program had been crowd-sourced, in a Q fashion? What are the VC's getting that they couldn't just buy directly, going the traditional route? There is the argument that the quality of the R&D will be better if it is crowd-sourced, but we aren't talking about knitting mittens here (which, by the way, I can't do). Math/data science/programming/advanced degrees/trading expertise/on-and-on experience required. Not quite like posting some comments and pictures to Facebook.

The success of Q as a crowd sourced hedge fund is built around premise that "crowd" (which I guess means people like us who don't trade professionally) can generate tons of profitable algorithms. Then assuming that there are tons of these Q can select and choose to invest in a subset of these.

Before Q, an average Joe on the street (not wall street) did not have access to infrastructure let alone data. With Q, people like us have an opportunity to experiment with data and attempt to come up with interesting profitable algorithms. Since many of us are not professionals (else we would not be here), there is a learning curve and I guess it will take some time for "crowd"to grasp the ropes.

I would like to hear from Q on what they have seen since inception. Do they believe that crowd can generate enough profitable algorithms for Q to make a business out of it? What time-frames are they looking at?

“Landing men on the moon was an application of science and engineering. Sure, there was risk, but it was basically a matter of ‘If I do X, I'll get Y.’"

Again, with all due respect, I disagree. This is not the attitude of the people I know who worked on Apollo.

You’re thinking of the moon landing as a project. You know what happened, and it’s easy to fit the narrative into engineering problem/solution/new problem/solution/repeat until success. But that leaves out all the alternative paths that were pursued, not just different strategies, but different versions of the goal.

Think about why we went to the moon. According to President Kennedy, “We choose to go to the Moon in this decade and do the other things, not because they are easy, but because they are hard; because that goal will serve to organize and measure the best of our energies and skills.” That’s not “If I do X, I’ll get Y.” It’s setting out a vision, a challenge, in order to make yourself better. That part was necessary to get the funding, to attract the right people and to keep the program on track after mistakes and failures. It was also necessary to give the project meaning.

The “If X then Y” question is whether Quantopian can produce quantitative strategies cheaper and better than traditional shops. If so, follow-on questions are whether those strategies can be used to build a successful business, and if so, how the rewards will be distributed.

But I don’t think the questions should be considered in linear order. The main cost in Quantopian’s model is the time and effort of talented people who create algorithms. How do you price that? It depends on the attitudes and beliefs of the participants. If they’re having fun and learning, satisfied with some bragging rights and maybe a little cash, then the cost is close to zero, or conceivably even negative. But if they’re laboring in the hopes of large financial rewards, working on algorithms at the expense of other useful projects, then the costs are high. Therefore, the nature of the long-term business plan affects the current costs, you can’t postpone consideration of the former and expect to make good decisions about the latter.

Think about Wikipedia versus Huffington Post. Both gave volunteer users ways to create things. But Wikipedia was always conceived of as not-for-profit, meaning it attracted a certain kind of editor, and built a certain kind of site. Huffington Post was a for-profit venture, which attracted a different kind of contributor, with different goals and led to a different kind of site. I doubt that either one could have achieved what it did without a clear initial vision. If the idea had been, let’s get some good articles/columns posted and if we get good traffic we’ll figure out what to do next, I think both ventures would have flopped.

My original comment that touched off this subthread, that a comment didn't sound like it was from someone who had ever built a quant hedge fund trading desk, was not meant as a criticism; just as an observation that converting market-beating algorithms to viable businesses is not a bunch of engineering challenges to be approached in "If X then Y" format, but a set of questions that require a vision and a reason in addition to programming skills and financial knowledge.

converting market-beating algorithms to viable businesses is not a bunch of engineering challenges to be approached

But that's exactly what this is: Money + software expertise = software solution

And Quantopian and its Q-Fund are nothing if not software solutions. The high end tech VCs involved here know who to pull in and add to the Q team to solve any of their engineering problems. And the fund creation / financial side is just as easy to solve, $ + talent = fund. And none of this is even CLOSE to rocket science. Nearly every piece of this construction puzzle has been done before, many times. And the only value add, The Q's "first mover" proposition, is that there is IP to be mined in the minds of this "crowd" as Pravin is talking about. And yes, that remains to be seen. Are 50 thousand monkeys beating on a snake better than 20 physics PHDs trapped in a room full of data and pizza and CodeRed?

I'm beginning to think that it will be the platform IP that Quantopian is building that may be of value as a group source algorithmic discovery framework; not necessarily financial. I don't know how robust Python is with regards to creating algos that could discover or test proteins, or structural defects in bridges or the national electrical grid, but maybe that's where it will best be leveraged. Maybe algos to solve chemical or material science problems. A generic crowd-sourced algo discovery system.

In general I don't see how scraping a few basis points off of a few tens of million dollar managed portfolios is going to add up to the $25M they owe their partners. Can their fund, however they manage to build it, really be profitable enough to make the Q worth their investment? Or more? Or is what they're building as infrastructure really what will be eventually billed as their value proposition, their valuation?

But that's exactly what this is: Money + software expertise = software solution

I'm not trying to cut off discussion by claiming you have to have built a successful large quant hedge fund trading desk to have an opinion on this, but I believe that if you talk to people who have done that (or gone to the moon) you will find most disagree. At least that's my experience. They would tell you that the vision of the solution attracts money and expertise, and that it's not additive because expertise attracts money, and money attracts expertise, and it all happens in a ecosystem with important external effects, not a project-planning vacuum.

There are plenty of people who have money and are willing to hire expertise: sovereign wealth funds, endowments, family offices. Some are successful, some not so much. But none have built anything like what you'd need to take Quantopian up to $10 billion assets under management. And if money plus expertise worked liked mathematics, then 40% of large IT projects wouldn't fail so badly they are never implemented (with nearly half of those, 17% of the total, failing so badly they threaten the existence of the organization); and the successful ones wouldn't average 45% over money budget and 7% over time budget, returning 56% less value than predicted).

When you say it's not "even close to rocket science," what metric do you have in mind? It uses considerably more processing power than the Apollo program did, and more sophisticated mathematics. It may not be as important or exciting, but it is challenging. Of course, I'm not arguing that it can't be built, only that the straightforward way to preserve and expand Quantopian's business model would be extremely expensive, and would require massive scale to make economic sense. You can save a lot of money by adjusting the model, but that affects decisions people are making today. So I don't think it's wise to see how things go at small scale and figure that you can always spend and hire later to build it into a large business.

If you want to let a lot of people experiment with financial algorithms cheaply, then you don't need a long-term vision. If the algorithms are successful, there's a lot of good things that could follow from that. That's like wandering on paths in the woods to enjoy the scenery. But if you want to build a $10 billion AUM hedge fund business with the results of your experiments, that's like climbing a mountain. You don't succeed at the latter unless you have a driving vision and prepare for the goal at the beginning. It may mean taking difficult paths early in order to be in tenable positions later.

You write, "nearly every piece of this construction puzzle has been done before." That's true, but it's saying nearly every individual challenge you're going to face climbing the mountain has been solved by some other climber on some other mountain at some time in the past. That proves mountain climbing in general in possible, but not that you can wait until you get to the glacier to think maybe you should have brought an ice ax.

I've got no opinion on Quantopian's business model, I don't know what it is. I just like the idea of crowdsourcing investment algorithms, and I like the way they've gone about doing it. But if anyone reading this board is participating in the hopes of getting a piece of, or royalties from, a large quant hedge fund at some point in the future, there are hard questions about how things are being run that he or she should ask today. It's not "if you build it, they will come," or even, "if you build it, you can figure out later how to get them to come." You have to build it with them in mind from the beginning.

What is needed?

1. Forex, Options and Futures are necessary to be competitive. Asset allocation with equities is not enough; you need to actually hedge your positions. How is Q going to address persistent storage and cloud computing?
   
2. More info from Q on what is working and what is not. Educate the developers to create systems that Fawce listed as goals earlier in this thread. Quantcon was excellent but needs to continue to school python programmers on becoming financial system developers. Monthly seminars focused on trading/market rather than API and Q hygiene.
   
3. Does Q have trading domain knowledge inhouse? Develop and enter an inhouse system in the Contest and share the code. The best growth years of Apple and Microsoft was when apps were developed inhouse to learn how to design system APIs and to show developers the 'best of the art'. Also, an inhouse trading group could help bridge the chicken/egg fund raising gap if the inhouse team could develop a trading system using Q. This system could be proprietary and used to trade Q money daily.
   
4. Publish a schedule of when features will be available. Get developers thinking about the next generation of systems and skills needed going forward.
   
5. What is the relationship between developers and Q? Is Q going to be a broker/dealer or is it going to license code from developers? Is there another option? I can imagine a relationship what would be very attractive; I can also imagine relationships that I would not consider under any circumstances.

Today the goal seems to be very 'computer science' oriented instead of 'hedge fund' oriented. The competition is going to be hedge funds...

@Aaron B., I think we both agree, you through your list of algo-to-fund concerns and me through my BigGap, that the transition of Quantopian into a fund of size will require efforts heretofore unmentioned by Quantopian. My belief is that such efforts are generally architectural, bolt part A open source algo generation to part B large AUM fund. And that this construction is no mystery and requires little original discovery or rocket science level challenges. Especially in this day and age of old-school, business as usual, disruption being done daily; that folks these days are practically built for such shake-it-up type work. Perhaps the Q is setting itself up as some type of Elon Musk type disruptive agent. We all know that the financial world could use a little upsetting of its applecart. I think however, that the Q's business model is exactly what is under discussion here. It claims it is now a hedge fund. I wonder how such a leap will be profitable. I think the leap is primarily the application of software technology, but is there money in it?

You obviously have experience in these matters and allude to knowledge of how such funds might be established. How often does a group of technology focused venture capitalists come together to build a hedge fund? From what I know of the business (little) that combination sounds rare.

The trouble with answering how often technology-focused VCs come together to build a hedge fund is (a) definitions and (b) a lot of the efforts remain small and obscure. I would say none of the large asset managers generally recognized as quant hedge funds today were started by people who focused more on technology than on finance. On the other hand, firms like Citadel moved heavily into technology after succeeding in investment, firms like ICE started with disruptive technology ideas, and firms like Silver Lake are run by VCs. People who start high-frequency trading firms are usually technology guys who got a little experience at a financial institution, and there are plenty of quant ideas that require sophisticated technological innovation. None of these is exactly what I think you mean. But collectively they prove it's possible to do this.

I also think we mostly agree, I just don't like the "bolt open source algo to vanilla fund" concept. That strikes me like Bertelsmann's idea 15 years ago that it could bolt Napster onto a bricks-and-mortar record company, or Time Warner's concept of bolting America Online to a traditional media company. Those aren't exact analogies, both of those ideas failed due to culture clash and incoherent business logic. Using crowdsourced algos in a quant hedge fund doesn't have those specific problems, but it does face some serious challenges.

Running a trading desk for black box algos is very difficult. For one thing, a large hedge fund would be held to a much stricter compliance standard than individuals trading small amounts. Some of those standards require internal information about the algos. Also, there are many regulations that apply at the overall portfolio level, and that's true even if each algo is a separate legal entity. You need to figure out how to adjust for these constraints, and that requires either knowledge of the algos or complicated bilateral communication. And it's not just regulation, market impact of trading happens at the portfolio level, also things like counterparty risk, financing charges and drawdowns; among others.

The usual way people handle these is to build the algos into the trading system as an integrated whole. The second common approach for the largest firms is to build algos that generate signals rather than trading decisions. A separate portfolio construction routine takes the signals and computes a trade. It's not just adding up signals, in many cases you accentuate the impact if different algos agree, two weak buy signals might translate to a very strong buy; and there is more complex logic as well. But this approach requires at least as much work in the portfolio construction system as the signal generation system. And both approaches make it hard to assign profit back to individual algos, and mean that good stand-alone algos might not be the best constituents of the trading system.

If I wanted to build a $10 billion hedge fund from crowdsourced algos, I think I'd do something like what Quantopian has achieved, but I'd buy a non-exclusive license on the winning algorithms. If your algo achieves some level of stand-alone success, I get the right to look at the code and incorporate it in my big multistrategy fund any way I choose, in exchange for running it at some fixed investment (say $1 million) for five years and delivering any profits to you at the end of that time (with losses borne by me). You're free to market your algo yourself, or run your own money with it. Not as revolutionary an idea as funding the black boxes directly, but a lot less overhead, and a lot fewer headaches.

Regarding the moon landing analogy, the difference is that at this point, Q has no firm model of the pool of quants who would get them across Market Tech's big gap. I don't know much history of the space program, but I'm guessing someone was able to do some back-of-the-envelope estimates to understand that a bottle rocket wouldn't work. In the case of Q's pool of potential quants, what's the ultimate number? How many will generate decent algos in a given year? Will they get poached by the competition? Etc. Judging from the contest participation, it's a few hundred, at most, at this point. And if each one is a rock star, that would be just enough to get to $10B -- just give each one $50M-100M in capital! I'm not saying Q won't get there, but it's just not obvious at this point (which, I suppose, should be the case for a good start-up idea--it should appear almost impossible).

What if the objective wasn't a $10B fund but 1,000 $10M funds? Would that change the challenges in a better and more practical direction? I think Q is doing a great job and it is their prerogative of which direction to go but maybe its easier to scale horizontally. I think many of us has found themselves in the same position as myself, sitting around a computer and showing friends some pretty good returns with real money that match pretty good backtests and they want in. What if we could open those smaller markets? I know going in this direction creates its own regulatory issues but their must be a way to democratize this a little and also profit from it for both the quants and Q. I think the main thing that is needed to get there is the ability to track and assign money deposited into an algo to a specific client. I know its not that simple but it could be a better direction to head. Just a thought

What if the objective wasn't a $10B fund but 1,000 $10M funds? Would that change the challenges in a better and more practical direction?

Pros and cons. If the funds are deemed to be controlled by one manager, then portfolio-level rules get applied, even if each fund is a separate legal entity. In order to avoid that, there couldn't be any combining of orders, or shared use of cash. That makes everything less efficient. And you have to police carefully to make sure there is no collusion among funds.

In that case, you'd probably set Quantopian up as a prime broker, or partner with a prime broker, providing execution, financing and "capital introduction" services. Then each fund is just a client. There is some red tape involved in that, but a lot less than running a $10 billion fund. And there would be much less concern about the black box algorithms.

I think to work you would need investors to take the Quantopian brand as a quality guarantee, meaning they would invest in funds with shorter track records and less-impressive financial resumes than independent funds.

A) recruit sophisticated managers
& B) create sophisticated managers

Aaron, thanks for the reply. You obviously know much more about setting this up than I do:) I appreciate your thoughts on the subject.

Obviously keeping every fund separate would make it less efficient and more costly. With an idea like I propose, I don't see a way around that. What if Q doesn't set up as a prime broker and just enables execution through brokers like it does now but with better broker choices? Would that eliminate the red tape and potential collusion issues? They would just act as an enabler of hedge funds by supplying what they do now (always up live trading through other brokers, testing and data) but in addition fund/client tracking software and probably a little guidance of how to set up a fund. They could still observe what is successful and invest and have their own fund.

This is such a unique network why go after what has already been done (a massive hedge fund)? It seems like there is a huge market of people that can't get into good funds but aren't satisfied with the standard financial products and we collectively could serve that market well. Including acquiring clients.

I am fascinated with what everyone here has created and really enjoy developing algorithms. I just want to figure out how I can continue to do it forever. In this network, it seems like developing a platform that caters to a new market seems like a more viable solution. That is if its a legal option.

John O.,

I am fascinated with what everyone here has created and really enjoy developing algorithms. I just want to figure out how I can continue to do it forever.

wouldn't that be boss!

Yeah, I concur with you on Aaron B's insights. He sounds more like an administrator than a quant though (while I'm still just a poser.)

I've brought up, in prior posts, this notion you mention regarding this middle tier of money looking for a better management scheme. It's definitely there. Low millions wealthy individuals unenamored with the current offering of how money is worked on the street; seeking some means to avoid the big guns, stay actively involved, yet not versed in the algorithmic means to run their own money. So maybe this whole premise of $10B is a redherring. Maybe what ends up being created and offered is nothing like a traditional large fund. Maybe we should just shove that whole premise right off the table... Where did that assumption come from anyway? (Grant?)

The $10B figure came from Fawce in a discussion thread (I can't put my fingers on it). It has to be a big number. Say year-over-year 10% returns, Q gets 10% of the profit ($100M) and the managers get the other 10% ($100M), of the typical 20% cut. Then Q could support a staff and infrastructure. Otherwise, if it is a $20M business for Q, it won't be worth the trouble--only enough for a small operation in the financial world. At least that's my read.

What if Q doesn't set up as a prime broker and just enables execution through brokers like it does now but with better broker choices?

That's a decent model. Then they're like a hedge fund incubator. An important additional service is help in raising money. An investor would only have to do operational due diligence once for the network, and for each individual fund he or she would only have to assess the algorithm and its creator. Moreover, with standardized track records and documents it would be relatively easy to put (say) $1,000 into each of 25 funds, for the diversification of a big multistrategy hedge fund.

A lot of people believe that newer, smaller funds have better performance. Some of that is survivorship bias, the small funds that didn't do well went away, so if you look at existing funds they usually were hotter in their early years. But there's some evidence that it's more than that. Also a lot of people would trust a quantitative professional doing this as a sideline more than a slick Wall Street veteran. A quant with a career and a reputation wouldn't likely sacrifice it for the chance to eke a few more dollars in fees out of a sideline fund, while a guy whose entire reputation and net worth and friends and family net worth and future job prospects and feelings of personal adequacy are tied up in his fund might cut some corners. Also an outsider is less likely to get arrogant and fool himself about his abilities (there are counterarguments to this, an experienced financial professional has learned some lessons than an amateur has yet to learn, and there are some ethics--yes, it's true--that are drummed into most successful financial professionals that aren't obvious to outsiders). Perhaps the most persuasive argument for the outsider is the financial guy knows exactly what his investing prowess is worth, and has no incentive to leave anything on the table for the investor; while most professionals are used to a kinder world in which benefits are split equally among participants in a deal (there are counterarguments here as well). Counterarguments or no, there are certainly people who would prefer the crowdsource model if it could be rendered easy and safe.

The $10B figure came from Fawce in a discussion thread (I can't put my fingers on it). It has to be a big number. Say year-over-year 10% returns, Q gets 10% of the profit ($100M) and the managers get the other 10% ($100M), of the typical 20% cut.

I think that's excessive. The operational costs of running a large quant fund, assuming you don't have to pay for researchers or expensive business development people, are on the order of $10 million per year. So $1 billion at 2/20 would produce enough management fees to cover the operational costs (investors wouldn't like it if that weren't true). If you split that up into 100 $10 million funds, you save on some infrastructure, but ratchet up things like audit and administration expenses.

Whether the performance fees would be enough to interest Quantopian's owners or algorithm providers, I couldn't say.

Grant, well, I'd like to get some of what Fawce is smokin'. Hey, wait, I'm in Oregon, I can get my own stuff to smoke (soon).

Just a a different thought.... I see Quantopian as a different version of Covestor. In other words would I put 50k to work in the top algorithms knowing they are vetted by Quantopian as reliable in some way? Yes I would. Just like I put money into the top performers of Covestor. If there are just 2000 people like me we have a billion. Rinse and repeat.

Is this a hedge fund? Dunno, Would it have hedge fund like properties? probably. Would it perform better then the average mutual funds people invest these amounts in now? Hope so. But above all it becomes something that is accessible for sub million dollar investors who are excluded from the elusive world of hedge funds and I'm convinced a crowd sourced set of algos would do better....

On top of that I can learn and play with algorithms and sustain the illusion that I'll create something that is worthy not only my own money but also other people's money.

I see Quantopian as a different version of Covestor.

Without endorsing or criticizing either site, as I have only casual acquaintance with both, Quantopian seems far more likely to me to attract and refine quality algorithms. At the very least, the tools and rules are better suited for the type of quant strategies I have faith in, and for the people who can build those strategies. A successful site has to filter out the coin-flippers and hype artists.

But I see the similarities. The simplest business model for Quantopian would seem to be the Covestor one. My impression is that Covestor isn't doing very well, however.

Q is scalable to $100B+
Thousands of quants.
Risk management can be automated.
Compliance can be outsourced.
Account management can be totally automated.
Part of Q's vision is aggressive waste and fat elimination vs large traditional asset management companies.

My impression is that Covestor isn't doing very well, however.

Because they are not performance based but perc of capital based. My investment increased by 25% per year and I pay less then the dividends the held stock generates. I pay something like 16$ a month....

Most of that money goes to the manager so the take of Covestor will be very low.

To be honest most of the managers are crap and the ones with high sharp accomplish that with high betas. I only invest in people who seem to have edge because of they industry background.

Anyway, I do think that Q can do way better than Covestor as they open up things in the industry now closed to smaller investor. And I love disruptors, but disruptors need to become of a certain size to be disruptive and I'm happy to support Q with that journey.

In my opinion what would be needed to run 10b fund.
o Platform related
- Direct access, no intermediary brokers
- Support to more products, futures, options
- Better alert system, (Email, SMS, Rule based alerts)
- Better Risk models Volatility predictors, Betas, correlations
- Built-in complex orders. It should be easy to place bracket orders covered calls etc..
- Performance (Probably a Native application that lets people develop off-line)
- Multiple files support
- Parameter optimization and batch testing
Nice to have:
- Built-in Portfolio selection algorithm, Heuristic optimizers
- System to manage and combine algorithms/ money allocation

o A larger pool of Quants

With the right investment I am sure Quantopian can be a disruptor and change how investment is done. My feeling is that the platform is
evolving slowly and the focus is moving around. But Quantopian's Idea is exposed.

Just have a look at www.sig.com for an idea of what's required both technically and intellectually.

Here's an eye opener. Take any strategy you've got and ramp it up to $25M and see how it does. Any changes? If you trade anything besides the top 300 market cap securities you probably will see an affect, probably detrimental, on your P&L. My point is that one can't just take any of these strategies, however they get into The Fund, and ratchet them up as massive money movers and expect them to behave just as they do on $100k. The Big Gap is bigger than I thought.

@ Market Tech,

Regarding slippage, Thomas W. put up a relevant blog post, http://blog.quantopian.com/capacity-limitations-of-trading-algorithms/.

Interesting point that contest participants backtest and paper trade with only $100K. Since the rules are evolving, stay tuned for a capacity screen to be added.

The default slippage rule has to be changed before we can even start to talk about the impact of scaling beyond $100K. The current slippage model is quite frankly ridiculous, especially with stocks that don't trade a huge amount but have robust order books (which I believe comprise the vast majority of traded stocks). I'd have to search to find some research on average book depth versus traded volume, but anecdotally I see depth far in excess of the traded volume on the majority of stocks. If you add in dark pools and the kind of fills you get when you place a real limit order, the slippage model becomes even more absurd.
The current slippage model really only works on a SPY or APPL. Take even an S&P 500 component like TEG which average just 240,000 shares a day in trading. There are 420 minutes in the trading day, which means TEG averages only 571 shares traded/minute, or at their current price of around $72/share, $41,000. Using the default slippage model, you could only order around 138 shares, or $10,000 in an average minute and would see a .625% slippage for that. The reality of the model is even worse, since the volume is lumpy and some minutes only a few lots trade, which could very well be the minutes you put your order in. Is it even remotely realistic that an S&P component is limited to trading on average fewer than 2 lots a minute and that just doing that will skew the bid/ask that much? In reality there are several market makers for every S&P component stock, the order book is probably hundreds of lots deep, and you could put a limit order in .01 above/below the mid during any minute of the day and get filled for at least $100K at that price. I know the slippage model seems minor in the grand scheme, but Q has to pay some attention to it if they want a realistic look at scaling.

Grant, thanks for posting Thomas W.'s blog post. Good stuff.
I believe Q will need to DQ algos using low cap securities. Or, only allow fixed allocations to algos using low cap stocks. If these algos using low cap stocks are getting the best returns, then give those managers a higher payout.

For the 10B fund, maybe 500 mangers using some low cap stocks will be given 10M each where they are generating higher returns and getting higher payout percentages. Then, the other 5B will be managed by 100 managers using high cap stocks generating relatively lower returns and getting lower payout percentages.

You're making an assumption that capitalization and tradable volume are proportional. True when you're talking the S&P 100 maybe, not necessarily true at all for the remaining 4,900 or so.

I'm running live money and have never seen slippage issues. Has anyone out there trading real money experienced slippage issues? Can't IB present to Q regarding slippage concerns? I think IB's technology can go a long way towards alleviating slippage paranoia, correct?

Minor correction... There are 390 minutes in a trading day, not 420.

So right you are, thanks.