I have recently spent much time chasing rainbows on both Quantopian and Quantconnect.
I got very excited about an algorithm posted by someone else which sifted a given index down to a few stocks based on a ranking mechanism. By way of example, take the Russell 3000, filter it down to the top 50 stocks in terms of ROE and then take the best 20 of those in terms of momentum over a six month period.
Not surprisingly it was the smaller stocks which provided the most bang for the buck but the obvious danger is that greater than expected slippage on low volume will make an optimistic back test meaningless.
Both Quantopian and Quantconnect provide remarkable benefits for the beginner: an online integrated development environment and free stock data, both as to price and fundamentals. Minute data to boot – no messing around with daily data and the resultant clumsy estimations on entry/exit prices and slippage.
So far so good, and all credit to these two outfits for providing such remarkable facilities for free.
The problem comes if you ever want to make your experience pay you out in real currency rather than the funny money of back tested fantasy.
Both outfits seek to sell your ideas to their clients – a more or less anonymous list of hedge funds who are apparently looking for ideas. At one stage Quantopian were open for almost any reasonable money making algo and then they turned incredibly anal, thanks presumably to the demands of their biggest client or client. Big Steve Billions apparently.
What Big Steve seemed to want was so incredibly restrictive that the whole junket became an incredible bore and Quantopian narrowed every aspect of their offering down to pursue this dream. Neutral everything you can imagine and then leveraged to the eyeballs. Hmm….seems to me I have heard that sort of story before – it never seems to work out quite that way.
Quantconnect seems to have a more eclectic client base and so there is rather more scope to be inventive.
Does the hapless geek ever really make any money by presenting algos to these two outfits? There are some people on the forums of both outfits who seems determined to try but it would be very interesting to see some real statistics as to how much people have actually earned.
How many thousands of hours have been clocked up in back testing? Divide that total by the total payout on algos leased by Big Steve and his mates.
For each geek who has had an algo accepted, divide his total payout by the number of hours he spent back testing or designing in the online IDE.
On an hourly basis has said geek make a good living? Would said geek be better devoting his time to the Mechanical Turk or shelf filling at his local supermarket? I don’t know the answer to these questions but somebody at Q and Q sure does.
Leaving all that aside, how do these two providers line up if “all” you want to do is to make money by trading your own book?
The big disadvantage of both platforms is that using an online IDE without the facility to download the results of your back tests and research is like trying to write code while viewing your computer screen through a keyhole.
Sure, Quantconnect allows some downloads and both platforms have limited logging facilities and online debuggers. Gosh that is an amusing word.
But for those who have been used to doing their own thing in Python or some other language, the online IDE is an incredible drag. It is very difficult to learn somebody else’s back tester while peering through a keyhole in the semi darkness.
If you want to use Quantopian to paper trade or live trade with your broker, fuggedaboudit….they changed course and stopped all that. And that emphasizes a big problem – you are laying yourself open to the changing whims and fortunes of a business model which can and does change day by day.
Quantopian’s advantage for back testing is twofold: it is lightening quick when compared to Quantconnect and your can boost your performance exponentially by running seemingly infinite serial backtests all at the same time.
With Quantconnect it is a different and very painful story. The free back tester allows only one back test at a time and the paid for versions don’t get so very much better.
At least with Quantconnect you can live trade your algos – you can connect up to your brokerage account and trade away to your heart’s content.
Or can you?
I spent some weeks learning Quantconnect hoping to do just this. It was an uphill learning curve and Sisyphus would have applauded my valiant efforts. I recreated my Quantopian algo on Quantconnect and pressed the button to paper trade. In no time I would be trading for real and would discover the error of my ways (the algo is too good to be true).
Bang. Thud. Crash. It seems that filtering 3000 stocks down to 50 or 60 with the “best” fundamentals blew up the 512Mb server allocated to free paper trading.
So, onto considering the $20 pcm Quantconnect offering which would have allowed me to live trade. Thud, gurgle, bollox. Same problem – for $20 a month you get the same size CPU.
For $250 a month you get something rather larger but since I intend to trade very small, this was not a serious option for me.
I wish both these Houses well in their endeavors but there are lessons to be learnt here. Leaving aside the questionable aim of “democratizing Wall Street” (yeah, ha ha I’ll do the jokes) the honest truth is that you really need to go it alone. If you don’t have the skills, time or money then don’t bother. Put on your trades manually or buy an index tracker.
Interesting article – I also hate “online” backtesting software due to the lack of debug capabilities.
So what platform/environment do you use now?
Kind of you to say so, thank you. I use Pandas exclusively and tend to write my own stand alone systems as the need arises. Although I have been in contact with somebody who is planning to launch a SAAS using zipline. I have simply been placing orders manually in recent years.
I was actually planning to use Quantconnect for the system I describe above but as you can see that turned out to be impracticable.
You probably just need to optimize your code or write it differently if you are running out of memory.
Hopeless. The code is slim as it can be and the $20 pm server can not cope. I just can’t be bothered with it and have moved on.
It can not cope with thousands of stocks. Period. The $250 version might.