Introduction to Zipline: A Trading Library for Python • Hacker News Board

Re: Introduction to Zipline: A Trading Library for Python

#131

Post by lordnacho » Tue, Jul 26, 2016, 8:43 AM UTC

hobo_mark wrote: post #15

chollida1 wrote: post #2

I've typed and deleted this post a few times trying to find a way that it doesn't sound kind of pompous but if it helps save one person alot of money then screw it, I'll sound pompous.... I get asked quite a bit on how to start doing algorithmic trading and the first thing I always tell people is don't. I think I've said this many times now but the number of people who come at it with the thinking "I'm a computer sci…

Can't find the original comment but somebody wrote here time ago that some HFs let you run on their infrastructure (keeping 85/90% of any profits) if you can prove to have a valid strategy. Just out of curiosity, I could not find any other information about this, is it a thing that actually happens?

Yep. I am currently doing this, but under an NDA, at an HFT firm.

80% of profits seems a bit rich, though. If they're supplying top spec colo servers, a super fast network to connect them, and people to stare at it, I'd say something a bit lower (but still pretty chunky) would make more sense.

I suppose the key for them is they have a bunch of existing infrastructure that pays for itself, so any extra cream is good. Also they have capital sitting around, so why not?

The strategies have to be more than unusually profitable, though. I don't think I've seen a down day in the 3 months I've been sitting here, so your average trend following sharpe ~1.5 or so ain't gonna impress people.

Re: Introduction to Zipline: A Trading Library for Python

#132

Post by osullivj » Tue, Jul 26, 2016, 8:58 AM UTC

rmah wrote: post #24

unknown_apostle wrote: post #21

Earlier quoted context omitted.

Actually, does anyone know where an amateur could buy/download quarterly balance sheets/income statements for the broad stock market universe?

All US public company quarterly financial reports (and much more) are available in raw form at the SEC EDGAR site ( https://www.sec.gov/edgar/searchedgar/companysearch.html ). Those reports income statements, balance sheets, etc. But beware that companies will often file corrections later. In addition, nearly all finance sites provide summaries of these reports for at least the last few quarters and last few annual r…

Also tiingo.com

Re: Introduction to Zipline: A Trading Library for Python

#133

Post by lordnacho » Tue, Jul 26, 2016, 9:09 AM UTC

I'm sitting at an HFT here, coding.

This zipline thing is quite interesting if you're new, but if you can code, I'm not sure what the advantage is. The idea of a backtest is quite simple, and you can easily fire up something like pandas to do it for you. The equity line is simply your positions x returns, minus costs. To determine your positions, you have to make sure you aren't looking at future prices, but apart from that you are flexible in doing whatever you like.

And this was a question for me. Suppose I want to code a cross-sectional strategy. How would I do that in zipline? It seems to be the kind of thing that gives you one backtest for one time series. Perhaps I just haven't looked into it enough. When we backtest, often we want to do things across the ensemble. We also take positions in a whole universe of instruments, so the backtest needs to be a matrix, rather than just one column.

Incidentally, the example strategy will work quite well for retail traders. You can add a bunch of futures together and get a sharpe well over 1, basically what every CTA does but won't admit to. If you're wondering what all those PhDs do all day, it's adding capacity and researching minor improvements on that MA strategy. A colleague of mine worked at one of these brand names, and another friend owns one.

So, does that mean anyone can simply do this? Well, yes. But you'd have a lot of leg work to do, and you might get discouraged before you start. You need an account from someone like Interactive Brokers. You need a fair bit of money, or you'll have increment problems trading the large contracts. And you'll have to set up all the data feeds and look at it each day.

Re: Introduction to Zipline: A Trading Library for Python

#134

Post by gaius » Tue, Jul 26, 2016, 9:39 AM UTC

selectron wrote: post #44

hellofunk wrote: post #33

Earlier quoted context omitted.

You vastly underestimate the complexity of a discretionary trader's intuition and experience. You cannot just replicate years of human experience with computer code so easily.

This statement is too general. You could of said the same thing about chess, there are chess Grandmasters who devote their lives to studying the game yet computers play chess at a much higher level than any human.

An organisation at the scale of IBM was able to create, after many attempts, and vast investment, a computer that can beat Grandmasters. That insight isn't useful to an individual trying to do the same.

Re: Introduction to Zipline: A Trading Library for Python

#135

Post by algoisgame » Tue, Jul 26, 2016, 9:46 AM UTC

chollida1 wrote: post #2

I've typed and deleted this post a few times trying to find a way that it doesn't sound kind of pompous but if it helps save one person alot of money then screw it, I'll sound pompous.... I get asked quite a bit on how to start doing algorithmic trading and the first thing I always tell people is don't. I think I've said this many times now but the number of people who come at it with the thinking "I'm a computer sci…

You can have a good starting point here, if you are interested in Algorithmic Trading http://www.quantinsti.com/epat/

Re: Introduction to Zipline: A Trading Library for Python

#136

Post by wslh » Tue, Jul 26, 2016, 11:06 AM UTC

chollida1 wrote: post #2

I've typed and deleted this post a few times trying to find a way that it doesn't sound kind of pompous but if it helps save one person alot of money then screw it, I'll sound pompous.... I get asked quite a bit on how to start doing algorithmic trading and the first thing I always tell people is don't. I think I've said this many times now but the number of people who come at it with the thinking "I'm a computer sci…

What do you think about cryptocurrencies trading? Do you think the same warning applies or there is an advantage because there are few players?

Re: Introduction to Zipline: A Trading Library for Python

#137

Post by flatM » Tue, Jul 26, 2016, 12:31 PM UTC

Fomite wrote: post #87

asdkjasdkj wrote: post #83

Earlier quoted context omitted.

This level of defeatism is the exact reason I got into the markets algo trading. People like you stay out. More to the point, just because a genius mathematician and code breaker started a hedge fund it doesn't at all push out any of the little guys. The market is so large he can't possibly be trading all instruments at once, and "scaling" is a problem for huge hedge funds. Especially ones that have to answer to thei…

In fairness, if someone told me they were going to make a killing on free webmail supported by ads and data mining, I think "How on earth do you intend to complete with Google?" would be a perfectly legitimate question.

Dropbox vs OneDrive or Google Drive could be a better example here.

Re: Introduction to Zipline: A Trading Library for Python

#138

Post by Enzolangellotti » Tue, Jul 26, 2016, 12:32 PM UTC

ori_b wrote: post #102

Enzolangellotti wrote: post #98

Earlier quoted context omitted.

In fairness, by that logic nothing would ever get done. Microsoft? You're never going to win over IBM. Xerox? You'll always be ancillary to Kodak. GM? Ford is already there. The Fugger's Banking company? Good luck against the Venetians and the Florentines. Same old story. At the end of the day, you either try something new or you don't.

> In fairness, by that logic nothing would ever get done. Or it will get done by people who have thought long and hard about how they are going to compete with the giants, rather than someone naïve sap. The fact that the question is being asked does not imply that there are no good answers to it, but it is unlikely that someone who hasn't spent time considering how to improve on the incumbents is going to beat them.

Indeed, but nobody (I certainly wasn't) was advocating to not do your homework and evaluate your risk. Sometimes, however, things just happen, for example James Simons stumbled upon the industry by chance. How many companies were motivated by killer instinct? Did Gates want to kill IBM? Did Xerox plan to kill Kodak (they didn't but they serendipitously started a revolution)?

Re: Introduction to Zipline: A Trading Library for Python

#139

Post by lintiness » Tue, Jul 26, 2016, 1:10 PM UTC

Blackthorn wrote: post #121

blobbers wrote: post #45

Earlier quoted context omitted.

I'm curious what folk's thoughts are on these two platforms: Interactive brokers offers low fee access to their platform ( https://www.interactivebrokers.com/en/index.php?f=13869 ) Quantopian (www.quantopian.com) gives you the ability to trade through robinhood with long trades at 0 commission. There has been some skepticism in the Q forums on how well robinhood's execution is (possibly the effect of a "you get what…

I use Interactive Brokers. They're probably the best retail brokerage out there. They're cheap -- the best rates on commissions and by far the best rates for margin. And they have a reasonably well-documented and supported API which you can write your programs against. The API is actually pretty much industry-standard so if you're interested in a third-party program, they probably support it. If there's any downside,…

use iqfeed or quandl for data (i use iqfeed for quotes as well).

Re: Introduction to Zipline: A Trading Library for Python

#140

Post by infinite8s » Tue, Jul 26, 2016, 2:30 PM UTC

cloudjacker wrote: post #118

sseveran wrote: post #107

Earlier quoted context omitted.

CDS isn't an OTC instruments. Swaps are now traded on SEFs like MarketAxess or TradeWeb. Just because you don't have access to the data don't think that professionals don't. But this isn't OHLC data for listed equities. Expect to pay. Actually high quality realtime equity data (ITCH,PITCH,OpenBook Ultra,etc...) costs real money as well. Checkout CDS and FI market data offerings: https://www.marketaxess.com/data/marke…

That is awesome Yeah i get the promise: professionals do this, pay a premium for that Until they don't because someone provides the data for free or next to nothing In this town we call it disruption

Who's going to provide that data for free when they can charge a LOT of money for access to it. Ever notice why free data on Google or Yahoo are 15 min delayed?