Has Quantopian become all you had hoped it would be?
35 responses

My vote was 'met expectations.' I'll believe the 10B hedge fund story when I see it, but at the other end of the spectrum, being able to do DIY algorithmic trading, with all of the free resources is pretty amazing. I may give it a go with real money through Robinhood, if I can cook something up that makes any sense. Definitely humbling that this field is just like any other. No money trees out there. You gotta know what you're doing. Unfortunately, the Q business model is TBD, as far as I can tell. Q team, agree? Disagree? Quantopian is my first foray into algorithmic trading. The fact that I can backtest for free is immensely addicting. I hope Quantopian sticks around for a few more years if not forever. I don't mind if everything falls behind a paywall except for the free backtesting. If they stick with their business plan of a crowd-sourced hedge fund, I don't see how the paywall concept will work out. Who's going to pay to write candidate algos for a hedge fund? That would be a real coup. See https://en.wikipedia.org/wiki/The_Adventures_of_Tom_Sawyer: He skips school to swim and is made to whitewash the fence the next day as punishment. He cleverly persuades his friends to trade him small treasures for the privilege of doing his work. For those who voted, THANKS! I'll say that writing software for any purpose but trading is so much more fulfilling. When you're done -- it works. Whereas one could write hundreds of trading strategies and in the end, none of them work (believe me, I know). The allure of money squeezed from the market system through a strategy of your own design is a siren song difficult to escape. Sites like Quantopian make it sound like a simple tune played on a kazoo. It's not. Yeah, Grant, no money trees out there, no wailing willows or pines whispering their secrets. Market Tech, spot on, I echo your sentiments. Many are called and few are chosen. Even those that are chosen are usually unceremoniously spat out unless they are very cautious. "Trading" is a very dangerous word. The sukkers and new boys usually blow themselves up with leverage relying on the ever ahifting quicksands of back testing. It has been going well for me, though I am perpetually frustrated by Python. I have taken quite a break from writing new algos since November, and while I may go back to it some time this spring, I am more enjoying learning Rust and writing my own infrastructure. I don't particularly find other programming any more fulfilling -- it's all pretty satisfying. I look at systematic trading as a high-risk software startup that happens to have no customers. The end may be uncertain, but at least I won't spend the journey begging for SEO and languishing in some app store. "...languishing in some app store." That's why I've been trying to convince the powers that be that going vertical is the only way out of "languishing." Becoming YASA (yet another social app) is not at all appealing to me. Targeting some information channel that could use consensus information would be perfect. We'll see. I will say, and it's surprising to me at this point, but Windows 10 Mobile may actually give Microsoft something to work with in the years to come. Today almost to the day is my one year Quantopian user anniversary. For me when I discovered Quantopian it was a logical progression from my experiences swing and day trading a really neat avenue to test ideas that I've had that I've never been able to capitalize on due to barriers to entry (basically the extensive coding knowledge required as well as the difficulty and expense of good data). I came with a poor knowledge of C# I had taken one intro to C# class, and understood some of the basics of coding in C#. I was attracted by the allure of the Open and the potential of maybe one day being able win it. I had no knowledge of Python. I was able to learn by playing with the example algos, watching videos, and taking some free classes to learn. It was neat to watch myself becoming more and more competitive in the Open as my Python improved. To me the free flowing ideas and shared algos are invaluable. When I see a shared algo I like go through line by line and understand how it runs, many times learning new more efficient solutions to problems that I was maybe not optimally solving before (leverage, position tracking, position sizing). Considering my expectations were low (just something interesting I found on Reddit) when I first started using Quantopian I'd say they were defiantly exceeded. In addition I've improved my Python coding skill from nonexistent to about a working knowledge of Python and always improving. Also my knowledge of the market has increased immensely I think thing is due both from my own ideas which I tested, other peoples ideas which they posted, and reading every piece of literature that I can get my hands on in regards to trading strategy. In conclusion I'd say all in all I'd fail into the exceeded expectations category. I am quite happy with how Quantopian has been evolving. It would have been impossible to an individual alone to write such a good infrastructure on their own (plus the data and the maintenance cost), so it's just amazing what we can use this platform for free. The choice of python+pandas is simply perfect and the Research environment is a must for any serious algorithm development. The only problem is that once you choose Quantopian you are stuck with it: if you like to move your algorithms (and NBs) to another platform it doesn't come for free, it costs a lot of effort. Also, just because zipline is freely available doesn't means you can run your algorithms in your own server: you would need the data source and to write the broker integration and maybe a GUI if you like to speed up your development. So periodically I ask myself if Quantopian goals are the same as mine to check if I want to keep investing my time in this platform. For example, Quantconnect chose another business model and they charge a fee for their platform. While this seems not so good, it actually means that the platform development is centered around users needs and the users needs dictate the platform development (e.g. many users want Forex so they got Forex support). In Quantopian is very different and so you might ask yourself what is the best choice for you. All in all, I still love Quantopian and the advantages are still far superior than using a different platform. The short version of my thoughts: I lost interest as soon as I realized all they wanted was the same kind of stat arb currently being employed by most HFT shops. Matt, you mean as opposed to ETF arbitrage or volatility term structure stuff etc etc, ie their focus on factor-based equity long-short? As Simon has pointed out elsewhere, there is a misalignment with the kinds of algos individuals might write for investing their own money, and the institutional-grade, scalable algos that Q needs (presumably, someone has told them, show me X and I'll provideY of capital). So, the proposition becomes "Hey, come help us build this enterprise. We'll reward you. Maybe." Reaching out to universities seems like a smart move, in this context, since you have a pool of algo writers with generally no capital of their own and lots of time on their hands. And willing to work for free, just to get the experience, etc.

I have been delighted with Q, but then my agenda is different from most. Love Python, learning some excellent stuff, and don't care twopence about the Q hedge fund so.....erm....

We'll see how things play out, but if the hedge fund falls flat, then what? I have not heard any discussion of a Plan B.

But Grant, I just don't get it. What do you want from these guys? If Quantopian closes you still have the open source Zipline you can use to develop strategies in.

Or are you mostly wanting to get AUM from these guys? In which case it is not at all wise to rely on one source of funds. Its a real pain but if your object is to manage OPM surely there are other avenues open to you? For instance a direct approach to suitable hedge funds.

The Quantopian model is one particularly seductive form of financial pornography.

Last year, when the contest first launched the viral press, some through paid content, promoted headlines with "Finance novice beats hedge fund pros" and other such nonsense. The reality has been, the first three contest winners were stopped out at 10%, the next three winners are mixed and the performance of contestants became difficult to monitor with no new winner trading these past few months.

Remember, the contest winners were the best of the best leading me to wonder how poorly the average Quantopian user is performing. Certainly, Quantopian's partners at Interactive Brokers know. My guess would be the average Quantopian user loses money in excess of just forking it over to Vanguard in a 60/40 stock/bond allocation.

Enabling newcomers to financial markets to trade using algorithms is akin to putting a brightly colored night light in a socket where a baby crawls about. Although there is more light, the danger to the unsuspecting baby is immense.

Unplug the night light, listen to John Bogle and engage in areas of social finance that adds value to society. That way, you'll have something to show for the money you lose.

Hilarious Sally. I would not have put it quite as bluntly as that but you are right about novises and algos. Dead right. Curve fitting both the portfolio, and the parameters are the main danger of course. And then time frame tested - look at some of the insanely short test periods used here.

Most of these guys should start by reading Talebs Fooled by Randomness. God knows I have been fooled often enough myself.

Having said all of that algorithmic quantitative investment is the right way to go. Look at the history of weather forecasting.

AI will eventually win the day.

And when you mention Bogle let's not forget that index tracking IS quantitative investing. A form of trend following.

Sally - I beg to differ. To succeed here, one needs a combination of things: to find or invent a trading strategy; choose, or find or invent a method to choose, securities suitable for the strategy within the given one- or six-month time frame; have some luck with the market; and correctly implement the strategy in Python using the Quantopian API. Some users have more skills and experience one one side, some on the other, few on both. The Quantopian platform provides a good venue to improve, and the contest (and the hedge fund, for those who believe it will materialize before we die of old age) an incentive. Nothing here stops anybody from writing algorithms that "fork money over to Vanguard in a 60/40 stock/bond allocation". In fact, my first Quantopian algorithm bought and held the Vanguard Total Stock Market ETF (VTI). It came in the top quartile in the May 2015 contest - #62 out of 436 entries. That encouraged me to write and submit a two-asset algo. My algorithms and skills evolved, until my latest one was ranked #3 a few days ago. (I used a year-old list of well-chosen stocks a friend had sent me; the rest is mine.) All without paying (much less losing) a single penny or farthing.

Quantopian, with Interactive Brokers and Robinhood, also provides an opportunity to lose real money fast. I hope those attracted by colored lights will get burnt, but not broken for life, and realize that it takes real, patient work over time, and some affinity for numerate sciences, to make money slowly, if at all. Then welcome to the learning-while-doing club.

Simon - not quite. What I mean is that by the contest's stated criteria, the only sorts of algorithms that are likely to win AND continue to be successful under the same criteria are the sort of mathematically complex, high frequency statistical arbitrage algorithms that are generally the purview of sophisticated quant shops. I personally believe that the likelihood of individuals "out-sciencing the scientists" is very low. Occasionally it will happen, but ask yourself if you would bet on it happening frequently? The professional shops will have more data, more people, more processing power, more experience, etc. Just "more better" across the board.

You don't compete with experts by just trying to be more expert-y.

Individual investors (quant driven or otherwise) ARE set up to outperform with a certain set of strategies and tactics. In order for individuals to be successful, they have to compete in ways that professional/institutional investors are UNABLE to compete for reasons which are behavioral, structural, regulatory, etc. Individual investors can take up larger positions in smaller stocks. Individuals can take up more concentrated positions in fewer stocks. Individuals don't have career risk (at least with respect to their investing) so they can stomach larger drawdowns as long as they remain confident in their strategy. Individuals can have long periods of "doing nothing" without their clients wondering "what the hell they're paying for". And so on.

It seems to me Quantopian could build a really interesting fund out of a collection of strategies which individual investors are likely to succeed with, and in doing so, get close to their stated fund objectives re: beta, drawdown, etc when all of those strategies are combined. But that doesn't seem to be what's going on. What's going on is that they're just asking us to be the amateur version of the experts (entirely because of the way in which the contest criteria are set), and I just don't see how that is ever going to work out well. If an institution wants a long/short factor driven model that scales well, is based on substantial research, is market neutral, etc, they can already go give their money to AQR (or whoever) and be done.

So as an individual, I am not particularly interested in investing my time playing a game that I am not set up to win.

Very well said, and a big part of the reason why I am not participating in the contests anymore; none of the algos that I am trading and researching would qualify (anymore).

I've been with Q a long time (since early 2012, when Fawce and Dan D. were fielding a lot of the help requests and active on the forum). I suppose that I was one of the relatively early beta users. I started with no Python experience (but some Fortran/C/C++/MATLAB) and no trading knowledge and experience. To date, my only real-money active trading has been with $100K of Quantopian's money, which I eventually drained to$90K. I also got invited to QuantCon 2015, which was incredibly cool, and the good attendance definitely gave the impression that Quantopian has momentum as an enterprise. It was a pleasure meeting many of the employees, as well. The upshot is that I have a connection, and I'd like to see them succeed. So, when it seems like there are poor moves, I tend to stir the pot a bit on the forum (and sometimes with direct e-mails).

I still don't get the Q fund concept, and tend to agree with Matt's comments above. Will Q and their money backers end up saying "Why didn't we just hire 20 of the best in the field, rather than trying this crowd-sourcing thing, which turned out to just be a global talent search anyway?" Seems like one could hire some decent talent for the $10's of millions put into Q so far. Grant - Do you think Q "gets" the Q fund concept? That is the 10%-of-the-profit-on-$100,000 question.

Andre,

I'm not sure what you're asking. The problem is that they are doing a "me too" move as a start-up, as I see it. They are going after existing markets, with existing ideas, it seems (as Matt points out, they are looking to compete with AQR and the like). What will be the unique offering?

Grant

Grant - I'm asking if, in your opinion, Q has a clear and well-defined idea of what it wants its hedge fund to be like.

Andre,

Well, I don't know. I'll leave that question for the Q principals. Lots of white space for them to fill in on this forum.

Grant

With a stick as sharp and persistent as yours, Grant, I'd think the Q would be leaking by now... poke, poke.

With the poll nearly up I'd like to thank, again, everyone who voted. Here are the results:

My opinion is that quantopian has created a great backtesting tool that is definitely useful for majority of users.

I think what kind of got lost is that, the hedgefund industry is like really really fword competitive. A lot of what I want to do in quantopian just isnt possible yet given computing constraints and that imo is hurting the top contributors.

I'd say it has failed to meet my expectations. I was more interested in the backtester than in the hedge fund. I initially thought the trading engine was well designed (ie, model reality quite well), but then discovered arbitrary limitations (eg, skipping the first minute of every trading day, cancelling orders at 4 pm, can't handle market-on-open orders, can't make overnight decisions). They just don't make sense to me and since the backtester is also painfully slow, I've simply lost interest and moved on. Still checking the website from time to time though hoping to see improvements. Who knows!

For those of you who are relatively new:

http://blog.quantopian.com/quantopian-manifesto/

Aside from the hedge fund business model (which was not mentioned), I will credit Q for sticking to the ideas laid out in this blog post.

In the post, there is a flavor of collective effort, the little guy versus the Wall Street titans, the raised fist imagery and all. Counterculture, hacking, openness...you get the picture. I think Q has lost its way in this regard (if it was ever on this track to begin with). For example, it is just not cool to be basically asking for help in building their hedge fund business but not providing any substantive information (see https://www.quantopian.com/posts/quantopian-hedge-fund-hows-it-going). Also, with regard to the Robinhood integration, there was no information about how Q might profit from the relationship (and how it fits with the hedge fund business model, for that matter). And what do the stats look like for profitability of Q users, overall? Has there been any actual collective hacking of Wall Street?

What would constitute a "hacking of Wall Street"? Aren't those just buzz words? Marketing speak? Probably.

If I were to take on the task of hacking Wall Street today I would need access to the tick stream and the L2 book. Now, 10+ years ago being given access to free minutely data, both for back test and execution, would have been a stunning coup over the exchanges' draconian data plans. But today? Although it's still a value that cannot be sneezed at by the retail tinker, trader, SPY, it's not enough; not nearly enough to provide the basis for hacking the vampires who tank up at the blood bar of order flow ringing the exchanges.

In 2014 I visited a private hedge fund on the central East Coast. This group, run and owned by a "hands-on" guy, was, they alluded to, responsible for sometimes 10-25% of execution flow on the bottom of the R2k. They had every possible aspect of analysis and execution setup as a set of self referencing digital dials and levers. It was a brilliant basket ballet. They would seem to be the type of traders that the Q platform might be designed to hack. But hardly. That group probably had a 1/4 million man hours in their code. There would be no way to compete against such a team.

How else could Quantopian's cohort quantitatively combat, given the tools available?

Well, I guess you gotta start somewhere. Personally, if Q had launched with a complicated, specialized platform for professionals, I would never have gotten involved. I suppose there is always the possibility to do fancier stuff once they build up a following of quants who can write strategies that make money.

Very few startups have a clear idea of what their final business model will look like. For most, it is a journey. Quantopian are presumably building the tech with this view in mind, and of course, competitions garner interest and foster a growing community that is both great PR, great for networking, and a great source of ideas for the future. I think Quantopian is pretty cool, and have had some fun hacking around with the backtester, although ultimately we have built our own in-house one. Still, at the very least, I'll be keeping an eye on what goes on here. There are lots of ideas out there, but reading a 50 page academic paper for each one just isn't feasible (although Wes Gray's alpha architect is also a wonderful site for this!). With quantopian, you get an overview of the idea, backtest results on the data source, and even the code to replicate it, play around with it, and quickly understand some of the more tricky implementation issues. Basically, it is a fantastic platform for crowd sourcing a broad spectrum of investment ideas, and testing and tweaking your own. I do hope it is financially successful in the fullness of time.

I came to Quantopian with a few ideas and a solid understanding of Python and Pandas. After a few weeks, I threw in the towel. All I was doing was trying to understand the API, which I find hugely unwieldy. I never did get to test out my ideas.

I think there are a lot of good points made above...about the push for strategies that fit with an institutional investor and if these are appropriate for the individual. There is no discussion of stop losses, for example, which I imagine would be key for the individual wanting to preserve his capital.

Also, this "maybe, might" of you might make money with us if you write what we need really rubs me the wrong way. And, if they are also not teaching people what they need to know about managing their own trading (risk management, position sizing), that leads to a dangerous situation for people who would blindly throw their own money into an algo.

Lastly, I question anyone who is testing algos on only six years of data. For four of those, the market went up and the last two things have been flat. Really, I'd rather test on 2000 and 2008. So, for the newbie that comes here and see all those upward sloping graphs, well...

Anyway, just some thoughts on things that could be addressed.

I read through some of the earlier responses, and just thought I'd throw in my \$00.02. As someone who has absolutely ZERO financial or programming background, Quantopian has met all of my expectations. I'm sure it's different for those of you who have a background in either programming or finance, but for me, the biggest draw is that every little bit of progress feels like an enormous achievement. I'm still on the level of only writing very basic algorithms, or modifying source code that has been copied from other backtests, but I've definitely been inspired to start learning more python on my own, independent of Quantopian, and to think more critically about the trades I make in real-life, and how my strategy could be converted into an algorithm.

As a side note, I'm currently trading an algorithm based on rebalancing ETFs based on a specified weight. It's nice avoiding robo-advisor fees!