Back to Community
Live Trading Algos with Real Money


I am interested in trading live with algos and real money. I want the algo to buy and sell for me automatically sequentially multiple times without me having to interact with any applications. I see various posts from people about doing this with things like zipline-live and quantconnect. I see posts that indicate that it is theoretically possible with these types of applications, but can not find very much objective evidence of it practically working in the real-world by individual traders.

1) Have you successfully used your algos to trade live in the real world with real money?

2) I absorb info faster through video than reading. Can anyone point me to a video showing someone trading algos live with real money linking from quantopian?

(I found a video of installing zipline-live locally but I want to see evidence of it really trading real money.)

3) Can zipline-live execute from the cloud also without having to run it locally?

4) Has anyone personally installed zipline live recently and troubleshot the issues with the version of pandas. If you have crossed this bridge in the real world, please share how you did it.

5) Can code be copied from quantopian and pasted into IBpy? Are their any subtle differences with the code?

6) Is there any easy way to take quantopian python code and convert it to easylanguage if I were to use tradestation as a broker?

7) From real world experience, what are the pros and cons between trading algos InteractiveBrokers and TradeStation?

8) What algo platform and brokerage do the "big boys" and girls on wall street typically use in the real world for trading with real money?

Does the individual investors like myself have access to these too?


8 responses

just my 2 cents as I'm one of the people who trades live with zipline for a few years now:

1) yes and plenty of others did as well
2) no I dont make videos of my trades (I sleep when my algo's trade as I'm Down Under)
3) I have it on a virtual headless linux box in "the cloud"
4) I have been involved with maintaining zipline-live but I must admit I have a version that I have modified to my needs (with more guard rails for live trading)
5) not really you still have to modify things a bit, IBpy is a paid thing afaik
6) write an interpreter I guess ;)
7) custom build. A few I know use zipline in their process pipeline but the trading involves often more regulatory hurdles and reporting adjustments which make people often build their own stuff

Thanks for your response Peter! I appreciate your feedback.

Peter and/or anyone,

1) Based on your personal real-world experience, does trading with zipline-live get somewhat correlated to previous test results? I fully understand the past won't give all the answers of the future, but just wanted to hear generalities from real-algo traders.


3a) I don't have any experience with zipline or zipline-live whatsoever. Can zipline-live execute from an online webapp that perhaps is hosted the makers of zipline-live or quantopian? Or perhaps is the only options to use either virtual operating systems or a run it locally on my own machine?

Eventually I want to use a virtual operating system, but I have never used one. I may need to take baby steps.

3b) I suspect that your intention for using a virtual operating systems is to prevent discontinuity of the algo from operating (from internet disconnection or computer crashes etc.). Is this correct? Are there any other reasons? If I have a stable machine and a solid internet connection how risky is it to run zipline-live locally on my mown machine?

3c) I am intimidated by trying to use Linux. I have not used it for about 15 years. I currently have windows 10. Does this put me at a major disadvantage. If Linux has true significant advantages would something like "window10 bash" work?

4) Does the off-the-shelf zipline-live have imminent failure modes, or did you just customize it to have some extra bells and whistles to mitigate your risks?

Does anyone have any recommendations debug this:
"ERROR: statsmodels 0.10.0 has requirement pandas>=0.1, but you'll have pandas 0.18.1 which is incompatible."

Based on what I have been describing does the 2nd release of zipline-live (zipline-live2)seem to be the better option. During my digging, I coincidently came across one of your previous posts:

Are both zipline-live and zipline-live2 safe or are we just giving strangers a free pass to dip into our brokerage accounts? I suspect they're actually safe but what does anyone feedback on this?

99) Does anyone have any unbiased feedback comparing the pros and cons between quantopian vs quantconnect?

Thanks again Peter and anyone else who may chime in!

Hallo Peter,

is there any news regarding your project

just two more question... what hosting service for your linux box in "the cloud" and what source for price and fundamental data are you using?


There's also this option:

I've been trading my own money, but I'm not sure I could properly describe it as "successfully." :/


Thank you for the candid response. This is exactly the type of feedback that I am looking for.

What is the root cause that hinders your success? Is it the algorithm or is it bugs in the pylivetrader?

Thanks again!

This result from ProQuant is possible ?
700% monthly benefit from 100 gbp.
Dynamic position 100%

@Pa Ke -- I think it's mostly dealing with real-world constraints that aren't accounted for in Quantopian's simulation. For example, the brokerage I've been using limits me to "Easy To Borrow" stocks for shorting, which reduces the universe a bit, so I miss a lot of otherwise profitable opportunities. And sometimes I'll get killed on super wide spreads, whereas half the time Quantopian's simulation would have given me a "market" fill without having the cross the spread. Also, real-world buying power calculations are pretty complicated, based on yesterday's overnight margin maintenance levels, so sometimes I can't submit orders I want to place, whereas Quantopian doesn't factor this in at all.

Even with algorithmic trading it is difficult to avoid "human intervention".

You will have a portfolio of systems and the turnover is discretionally decided.

The results should be evaluated only for a single strategy on the single instrument with a big number of trades.