How did you guys solve the issue of code backwards-compatibility? Are algos guaranteed to continue running without modification in perpetuity?

Hi Simon - we're not ready to make that guarantee, no. For instance, we're going to deprecate batch_transform at some point in the future. It will be a process when we do, but at some point, we will stop running the algorithms that have batch_transforms.

Obviously, we're intent on making our customers as happy as we can. We're going to work very hard to keep algos up, and up with a minimum of fuss. When we can't do that, we'll make sure the customer knows about it promptly.

Hello Dan,

Please elaborate on the "lifetime free subscription". Can the deployed algo be stopped, amended and restarted? Or just stopped and started?

P.

^^ I was actually typing up the same question as Peter when his post came through. Will we be able to edit the algorithm once it has started?

Yes, you can stop and start and edit the algorithm. The offer is intended as a lifetime free subscription and you can put whatever algo you want in that subscription for your lifetime, within the size limits described.

Expanding a bit more: We want to offer something of value to our customers who are helping us out in these early days. We also want to remove the fear that we're trying to bring people in for cheap and then raise the prices outrageously. Obviously we will have to charge in the future, and we intend to set the price at a reasonable level, but there is understandable concern at this stage of the product.

Thanks Dan,

I'm still getting up the (steep) learning curve, and even though you are offering a "lifetime free subscription" to launch a real-money algorithm, the Interactive Brokers costs and account minimums are non-zero. So, to lock in your offer (of unknown value), it appears I'd need:

• $5,000 (IB IRA account)
  
• $10/month (IB minimum activity fee)

Correct? Annualized, we're talking about 100% X $120 / $5,000 = 2.4%, which is very high. The break-even point would seem to be ~$50,000 (assuming the capital is just parked in a short-term government bond ETF, e.g. Vanguard's VGSH).

Sound correct? Or have I missed something?

Grant

Hello Grant,

We don't know the fees yet for running an algo but $100 per algo per month has been mooted. The free subscription could have a $1200 pa value if that was the case.

I don't know whether to start another thread but I would like to ask everyone what is a practical minimum for day trading? A cash account means funds from sales are not available until settled so that suggests a margin account is required. Reg T means IB will block any trades that cause a 90-day pattern day trader ban i.e. if the Net Liquidation Value of the account is below $25,000. Does that mean one could day trade with, say, $30,000? Or is $50,000 a practical starting balance? Or more?

P.

Hi Peter,

Wow! $100 per month per algo! So much for automation keeping costs in check...

Unless Dan objects, this seems like an appropriate thread. On https://www.quantopian.com/posts/what-would-it-take-for-you-to-invest-real-money-with-an-algorithm, Anony comments "I don't know any traders who trade with less than a quarter mil in their account" and provides some additional feedback. I recall Ernie Chan throwing out a number like $50K minimum in his first book. I recall another post on Quantopian with guidance...I'll see if I can pull it up.

It would be good to hear from the Quantopian folks regarding their target market. In my estimation, they basically need professional-level retail traders (i.e. folks making a living at day trading) or multi-millionaires who can allocate a percentage (e.g. ~10%) of their net worth toward active trading. I've had a hard time understanding how they will reach individuals with more modest starting capital.

Grant

Hello Grant,

Don't trust my $100 per algo per month rumour! It was mentioned as a possibility in a webinar once.

I enjoyed the videos of Gary's talk at http://blog.quantopian.com/gary-chan-on-pairs-trading-video-from-nyc-algorithmic-trading-meetup/ where he suggested $30,000 minimum for pairs trading based on Reg T.

P.

(FINRA: Day Trading Margin Requirements: Know the Rules at http://www.finra.org/investors/smartinvesting/advancedinvesting/daytrading/p005906 )

(From Ernie Chan:

"THE BUSINESS CASE FOR QUANTITATIVE TRADING
A lot of us are in the business of quantitative trading because it is
exciting, intellectually stimulating, financially rewarding, or perhaps
it is the only thing we are good at doing. But for others who may have
alternative skills and opportunities, it is worth pondering whether
quantitative trading is the best business for you.
Despite all the talk about untold hedge fund riches and dollars
that are measured in units of billions, in many ways starting a
quantitative trading business is very similar to starting any small
business. We need to start small, with limited investment (perhaps
only a $50,000 initial investment), and gradually scale up the
business as we gain know-how and become profitable." )

Thanks Peter,

On https://www.quantopian.com/posts/testing-on-dailies-doesnt-make-sense, there is more discussion. For example, James Jack states "I've heard several prop traders say there is just no point intraday trading on stocks with less than 100k because there is not enough return."

Grant

So Dan, just to understand, if I were to clone this long-only sector ETF rebalance algo and start trading it, and then realize that I could have just bought SPY instead, so I change my one algo to go long SPY, then I realize that I'm actually bearish on the market, so I change it to short the SPY, then I decide to make it a market neutral also with longs vs shorts, this mutable algo is still the one "free lifetime algo?"

"You acknowledge and agree that we may update the Site and Services with or without notifying you."

How are we expected to know about updates, which may impact a live account adversely, without being notified?

"You must provide us with a valid credit card (Visa, MasterCard, or any other issuer accepted by us) or some other acceptable payment method (“Payment Provider”), as a condition to signing up for the Brokerage Connection Services."

I thought it was free.

"Billing disputes should be notified to the following address: Quantopian; Billing Disputes; 77 Summer Street 3rd Floor; Boston MA 02110."

Do you have a fax number and/or email address for billing issues?

"We reserve the right to terminate our provision of any or all of the Brokerage Connection Services with or without notice to you."

Termination WITHOUT notice?

"Your notice should be sent, in writing, to the following address: [insert address for termination notices from users]."

???

Hi Abraham,

A number of your questions look like they are from the legal terms of service, so I'll leave those for Dan to get back to you on - I know he did a lot of work on this and should be able to answer your questions in more detail than I could.

I did want to respond to your question regarding the sample algo Dan included in his post. The purpose of this example is to give folks a functioning algo that works in minute-mode backtests and live trading so they can have a starting point for learning the platform. We would not suggest anyone live trade this example 'as is' with real money, or any other sample that we or anyone else on Quantopian posts for that matter. We are leaving the alpha creation up to you guys - the algorithm writers!

If I'm interpreting the second part of the question correctly, I think you were also asking if we intend to limit this offer to just the first live algo deployed? The answer there is, no! You can stop, edit and redeploy your algo as many times as you'd like. Our goal is to make it as easy as possible for our users to get started and learn about systematic investing.

Best regards, Jess

Thanks Jess.

Hi Jess,

I see that a survey was sent out but I cannot sign the agreement before I get responses to the legal terms of service and therefore cannot respond to the survey.

Running through your questions:

So Dan, just to understand, if I were to clone this long-only sector
ETF rebalance algo and start trading it, and then realize that I could
have just bought SPY instead, so I change my one algo to go long SPY,
then I realize that I'm actually bearish on the market, so I change it
to short the SPY, then I decide to make it a market neutral also with
longs vs shorts, this mutable algo is still the one "free lifetime
algo?"

Correct. That's still "one free lifetime algo." The thing you can't do is run more than one of those ideas at one time - you'd have to pay for the 2nd and 3rd algorithm. If you do them one after the other, that's one algorithm.

"You acknowledge and agree that we may update the Site and Services
with or without notifying you."

How are we expected to know about updates, which may impact a live
account adversely, without being notified?

The Site and Services are updated several times per day. Notification of every change is simply impractical. I think it's helpful to remember that we are trying to align Quantopian's interests with your interests. We want you to succeed; we need you to succeed. That means we're going to do everything we can to keep you aware of changes that might impact your live account. We will work to notify you of every change that impacts your live trading.

"You must provide us with a valid credit card (Visa, MasterCard, or
any other issuer accepted by us) or some other acceptable payment
method (“Payment Provider”), as a condition to signing up for the
Brokerage Connection Services."

I thought it was free.

That phrase in the agreement is definitely leaning forward to later this year for when we start charging for this service. As I noted above, "In the future we plan to charge a flat monthly fee per live algorithm." We're not collecting card information yet, but we will be. That is not in conflict with the current free usage, and with the free lifetime offer.

"Billing disputes should be notified to the following address:
Quantopian; Billing Disputes; 77 Summer Street 3rd Floor; Boston MA
02110."

Do you have a fax number and/or email address for billing issues?

You can email us at [email protected]. We don't have a fax machine. If setting up a fax forwarder is a key requirement, I'm sure we could do it. We don't do a lot of faxing =)

"We reserve the right to terminate our provision of any or all of the
Brokerage Connection Services with or without notice to you."

Termination WITHOUT notice?

If we ever did this, if we ever shut down without notice, it would mean that something had gone horribly, horribly wrong. I guess I wish we didn't have to say this in our legal agreement, but it's important, legal, cover-your-butt language. We need it in order to limit our liability. If things ever go horribly, horribly wrong, it would be even worse if we were, in addition, violating our terms of use. Again, we want you to succeed; we need you to succeed. We're not terminating this service unless something terrible happens.

One way of looking at this is to say "They're not guaranteeing that they will succeed, which makes them untrustworthy." Another way of looking at it is to say "These are smart professionals with a healthy respect for real-world challenges that can unexpectedly derail the best of plans." I think it is smarter to trust the person making the realistic assessment than the person who promises something that simply can't be promised.

"Your notice should be sent, in writing, to the following address:
[insert address for termination notices from users]."

???

Thanks for catching that. That is a mistake. The answer is: Quantopian; Termination Notices; 77 Summer Street 3rd Floor; Boston MA 02110. I will make the fix to that document shortly.

I hope that helps.

Dan

Hi Dan,

Regarding real money trading, can we integrate Quantopian with other than IB and trade with real money in live market? For example, can we trade with Merrill or Credit Suisse account with Quantopian source codes?

Thanks.

Hi all, very interested in getting involved with the testing of the algorithms with both paper and real funds in my IB account. Please let me know next steps.

Thanks,
Shaun

Hi Shaun,

Thanks for your interest. You can get started with your IB demo account right away, as Dan says above, we've released this feature to open beta (that means all registered users have access to it by default). To get on the waitlist for a real money account you can fill out this brief survey. We are admitting people to real money trading on a rolling basis so we can make sure that we offer each live money pilot trader a high touch engagement model.

To launch an algorithm to your IB demo account and see how that works follow these steps:

1) Press Clone Algorithm on an existing algo (I'd suggest using the one Dan posted at the top of this thread, or I'm attaching another option here).

2) Quantopian requires you to run at least one backtest to paper trade for best practices. So click Run Full Backtest, this will bring you to the backtest results page.

3) On the backtest results page you should see a button that says: Live Trade Algorithm in the upper right. Once the backtest is complete you can click this button. You'll see two options:

i) Quantopian - that is Quantopian's live trading simulation, in which we feed your algo a 15 minute delayed feed of 1 minute price bars

ii) Broker - this is the option you want to connect with IB. Select Broker and you'll be prompted to configure your account (if you haven't already done so). Once your account is configured (either follow the link on the live trading dialogue box or go directly to your Trading Accounts page) you can deploy the algorithm to the market (navigate back to your list of algorithms here). You will be prompted to enter your IB password every time you deploy an algo, Quantopian doesn't store your IB password for security reasons.

N.B. Once you launch an algo Quantopian will be logged in to your IB account. If you go to IB's TWS or WebTrader interface and login you will create a session conflict that may interfere with your algo's execution. Verify that you aren't logged in to IB before you deploy an algo, and refrain from logging in to your account during market hours (after hours is fine). This limitation is easier to get around with funded accounts as IB will grant you a secondary login that you can use for side by side viewing.

Best wishes,
Jess

Hi Jessica,

I clicked live trade algorithm, broker, and then see a message indicating no accounts found even though I have already added the account. I am then not able to click deploy.

Hi Shaun - can you shoot me a direct message via [email protected] and we will take a look at this for you right away?

Thanks! Jess

Kyu - At this time Quantopian only integrates with Interactive Brokers. We do not have integrations with any other brokers yet.

Hi All,

I was able to implement the algorithm in my interactive broker account today. Kicked it off at 10:47 AM New York time and turned it off around 12:07 PM. Everything occurred in the live market very seamlessly and as expected. After I turned the algo on I was sent to a screen showing my current positions. After a moment the algo initiated a few limit orders at about the current bid price for the industry SPDR ETFs specified in the code. The weights of the positions I established and the fill price were excellent. Within moments I was long the positions and watched the markets moving. This algorithm works great if you are bullish and feel an upside move is coming. Later I turned off the algorithm and liquidated most of the established positions for my first live test run with actual money. Everything went well and on my account had a total P&L of 65$ after commissions. Really excited to work with the algorithm in the future and to run it for longer intervals and days to ride upward price movements.

Thanks, Shaun

I noticed with a Quantopian live trading, you can specify the starting capital. For brokerage (IB) live trading, is there (or will there be) a way to modify the percentage of your IB account to trade with? Alternatively, is it best to just modify the algorithm to use scaled weights:

instead of:
context.weights = 0.99/len(context.secs)

use something like:
percent_of_account = 0.5
context.weights = percent_of_account * 0.99/len(context.secs)

So that the 99% becomes 49.5% of the account's balance.

It might not work if the other 50% is already fully invested and grows to more than 50% of the account. I guess you would have to query the market value of the algorithm's positions and the overall account balance to re-adjust the weights each time? (if that's possible with your API).

Hi Joe,

Good idea. I can try to create the variable and add to context.weights calculation next time I see upside move coming in equities. Will let you know how it works. Would be very useful to specify portfolio size available to work with up front. Also if we could have stops/trailing stops added based on overall market movement. Perhaps we add spy with logic against it based on moving averages etc.

Shaun

Hi!
I thought I'd add my thoughts here too. Excellent setup by the way...I think the backtesting is really the primo-centerpiece of Quantopian and the seamless deployment to live environment is pretty great. Looking forward to more instruments, sub-1min bars and a little more control over order entry ;).

I haven't yet used the live trading setup too extensively but below is a few of my notes. I haven't perused a whole lot on the forums so I apologize if much of this already exists, has already been addressed or just generally deemed to be of minimal value but here it is....
If anyone has general insight or input I'd appreciate it.
-General risk caps at strategy deployment level (perhaps I have low confidence in my python writing): max orders, max shares and more relevantly assign max capital to a strategy. Something to avoid both runaway order placement and more importantly the rather vicious IB management of account margin violation.
-TWS access; I assume things are running in a vm somewhere and you are using the TWS api for orders etc. Perhaps a way to run multiple TWS instances with same account has already been addressed. If not it is definitely difficult not being able to use TWS. (...ah just read above that there is a way to run 2 tws clients ... I'll look into this).
-Get flat (perhaps less relevant if one can connect a TWS client); ability to both click-stop strat and click-close open positions if desired.
-Schedule functionality; stop, get flat or pause entries at some time or over some interval: FOMC, EOD, etc.
-Detailed audit log of order placement; time and inside book market data for each order placement, modification and fill. Perhaps already available by requesting logs via TWS api, either way definitely critical for one to determine efficiency of execution.
-Alerts; email or some external msg on trade or some event.

As mentioned, I am sure you've addressed much of this and more but those were my first impressions.

One additional thing I think would be great to see is a general anonymously aggregated performance of all live Quantopian client algos. I think most retail brokers are required to publish this kind of info but typically burry it as it doesn't tend to be too flattering. It would be just nice to know that the retail investor is getting ahead somewhere as I think that's why most of us are here.

Hello David,

Details here http://ibkb.interactivebrokers.com/article/1004 of adding an additional user to your IB account. The second one is used for API access.

P.

Some feedback:

• Is there any way to apply constraints to the system, on the IB side? For example, say I wanted to have a long-only portfolio of specific ETFs, so that even if the Quantopian code went completely haywire I could only be in those ETFs.
  
• What sort of notifications and records will be generated by Quantopian and IB? How will I access them? Will there be any reconciliation done on the part of Quantopian/IB, to verify that everything is working?
  
• Are there human-in-the-loop elements to the Quantopian/IB system? Or is it fully automated (i.e. either running autonomously or disabled)? Is there a mode that would allow manual review and confirmation of trades?
  
• I've noticed that some of the Quantopian staff have started using the term "systematic investing" versus quantitative trading (e.g. see Jess' comment above). This is confusing, since in my mind, systematic investing does not involve the kind of frequent, active buying and selling that the Quantopian/IB system is designed to support (and it is not yet clear if the costs are in line with industry leaders, e.g. Vanguard, etc.). Am I missing something, or can Quantopian/IB be used to manage a retirement portfolio, for example?
  
• Is there a contact available at IB to get their perspective on Quantopian? Could we get some feedback here from them?
  
• Overall, my sense is that the barrier to entry is still very high for trading, both from the standpoint of the capital that is required, and also the technical expertise and time commitment hurdles.

Grant

• I'm not familiar enough with IBs systems to give a definitive answer. Clearly, if you have a non-margin account, you're not going to go short. I am not aware if IB permits you to lock yourself to a specified set of securities.
  
  • I can't speak for IB, but it's your brokerage account - they'll do what they usually do, including regular statements. For Quantopian, you'll get the reports you see in the live algo dashboard. Reconciliation happens continuously.
    
  • It's all automated. We may put in a manual review method in the future.
    
  • We continue to test language to describe what Quantopian does, and what Quantopian does continues to evolve. Algorithmic, systematic, quantitative, automated, trading, investing. . . there is a long list of relevant terms. Quantopian does support trading frequencies of dozens of trades per day, and it supports much lower rates too. Quantopian certainly can be used to manage a retirement portfolio. I hope we see a lot of that. Vanguard is an interesting company to compare us to. Vanguard's leading products (in my mind) are passive investments. I think you'd want to compare us to their actively managed products to make a better comparison.
    
  • If you are looking to open an IB account, we have someone we can refer you to.
    
  • When something is new and unexplored, it can be daunting. I encourage you to take a few more steps if you want to learn more. Do paper trading on Quantopian. Open an IB account, get an IB paper trading account, and start trading that. We won't charge you for it during the beta, and you'll be able to see exactly how it works. I think that will go a long way towards settling your understanding of how it works. There is no way to replace the experience of doing it yourself.
    

Thanks Dan,

I'll stick with tinkering around on Quantopian for awhile, until I see some photos of all of the customers' yachts.

Grant

I can't tell if you're kidding or not!

• Paper trading counts as tinkering, yes? There's no money on the line. Give it a try.
  
• If you start seeing pictures of yachts, that's when you should get suspicious. We're not a get-rich-quick scheme. We're a platform for algorithmic investing.
  

Sorry Dan,

Got carried away with the hyperbole. Basically, I need to continue moving up the learning curve before getting into an Interactive Brokers account for paper/live trading. Also, since both Python and algorithm writing are new to me, I'm slow. We'll see if I can sort out another algorithm to launch into your paper trading application.

By the way, I very much enjoy the ethos of Quantopian, so keep up the good work!

Grant

Thanks Grant. You've been a great early power user, and you've helped us a ton. I'm glad you're here to help us.

Dan

Peter, thanks for the info.

I have another question regarding market data. What source do you use for market data? Is it a source external to my running TWS instance or do you use the market data that is enabled for my IB account (i.e. depending on whether or not my account subscribes to some US equity data this feed could be delayed 15 mins)?

Thanks

Hello David,

I'm still in the process of applying for an IB account. Dan can elaborate but Quantopian prices are provided via Nanex. See: https://www.quantopian.com/posts/new-features-paper-trading-live-trading

P.

Anony,

Another angle on this is how much annual revenue does Quantopian need to generate? Let's say they need $10M/yr. to make it worthwhile and expect 10,000 users. So, back-of-the-envelope yields $10^7/10^4 per user = $1K per user annually.

By the way, I'm not sure I'd categorize 10-50 orders per month as investing. The guidance I've seen is that quarterly re-balancing is more prudent, since investing has a 5-year minimum time horizon, as a rule-of-thumb. My view so far is that Quantopian with a linked brokerage account may be more trouble than it is worth for investing.

Grant

Thanks Anony,

The 1%-2% max per year as a total of working capital is what I'd thought. Regarding the academic world, I've yet to see any interest. One hang-up, I think, is that there is restricted access to the data (and the exact details of how the data are gathered and reduced are missing, as well). So, the necessary analyses for explaining the underlying fundamentals can't be carried out. And there is no way to run batches, parallel code, download and plot results, etc.--not very attractive if one is writing a book/paper/thesis. Besides, my sense is that major research universities already have access to the kind of data used by Quantopian (e.g. via http://www.crsp.com/ or similar).

Grant

Getting back to IB and Quantopian, why can't a user access IB's tick-by-tick data, rather than minute data? I'm already paying IB for the data, and not to use it means that I'm giving up a lot of alpha in execution, more than just the half-spread (on average).

Hi Abraham,

We wanted an identical execution environment at every step of the platform. This way it is easier to build your strategy, backtest it, and the deploy it to the live market using the same foundation - so that you know exactly how it should perform. The reason we don't provide our backtest in tick data is because of our license agreement with our data vendor, we will support minutely bars as the highest frequency.

Best,
Alisa

Alisa's point about backtesting and live trading is dead-on. One of the key design concepts for Quantopian is to have backtesting and live trading be as close to the same as possible. That means that you trade with the same data as you backtest.

There are a couple other tick-level reasons for the way we did it: 1) We're going to support other brokers than IB, and we want to keep the brokerage connection as generic as possible and 2) tick-level infrastructure is a lot more expensive. All in all, there are number of reasons for sticking with the minute-level data.

I want to be clear about one thing: in live trading, you are not waiting a minute to submit your order to IB. When handle_data() runs and an order is created, that order is in IB's hands for execution very quickly - a second or two. The algorithm will wait for the next minute (the next handle_data()) before it shows you the order execution - but that's a display question, not an execution question. You are not giving up a minute's worth of alpha in the execution of orders.

Imagine your strategy entails pairs trading of SPY vs QQQ and the minute data shows the quote in SPY as 179.90-179.91 800x1000, but when your strategy order fires it is actually 180.07-180.08 100x1000. So you ending up paying 180.08 on SPY, when you really wanted to pay 179.91.

Trading on real-time data is so fundamental to most strategies, that it blows my mind that someone could argue for using minute data when real-time data is available and the trader is actually paying for it. So it's not just a display question, but an execution question of utmost importance. How much slippage are you willing to live with? If it's my money, the answer is "very little."

Hello Abraham,

I'm not sure that we're talking about the same thing yet.

I want to take your example and extend it. For clarification, we're talking about trading with an IB-integrated account, not the backtester - if we're talking about a backtesting situation, it's slightly different; for this purpose I'm talking about either IB paper trading or IB real-money trading. Also, I want to be clear, Quantopian is only working on as-traded data, not quote data.

Imagine it's 10:00:00 Eastern. A trade crosses the wire, SPY trade at 179.90, and arrives at the Quantopian server. Over the next 60 seconds, dozens of trades arrive, priced at 179.90 or 179.91. At 10:00:59, just before it we get to 10:01:00, a trade comes in at 179.91. Quantopian wraps all those trades into a bar - Open:179.90; High:179.91, Low:179.90; Close:179.91; Volume: 8000. (Only the close is relevant in this example, but including the rest for completeness. That bar is passed to your algorithm, and handle_data() runs, and in this example the algorithm determines it should place an order to buy 100 shares of SPY. Quantopian's servers talk to IB's servers and get a confirmation that that the order is placed - in most cases. that process takes a second or two. At some point in the future, IB tells Quantopian that the order has been filled. For a stock like SPY, in most cases that is very quick, under a second. In the meantime, a new bar has started, and at 10:01:59, the bar is completed and passed to your algorithm, along with the information that the order has been filled. handle_data() runs again, looks at the new price, looks at the filled order, and decides what to do next.

So all that said, does that clarify for you how live trading works?

If you are looking for second-to-second trading decisions, and that level of slippage matters to your strategy, then indeed this is not the correct platform for that strategy. On the other hand, Quantopian is a good platform if your strategy supports action within a few seconds of learning the prices.

Hello Anony,

Would there be an example we could run on Quantopian to illustrate your claim "I'm convinced Quantopian is geared toward multi-day trades"? In other words, do you know of a proven intra-day algorithm that worked on another system, that could be coded into Python? I say "worked" since my assumption is that any algorithm that is currently working would not be one you would want to share. Basically, the idea would be to highlight, via an example, your point. If you are correct, then the known algorithm should do poorly on Quantopian.

Grant

So Dan, what do you do with low-cap stocks that had no trades in the last 10 minutes or 3 hours, but the quote was often revised. Say ANCB which might trade 42,000 shares one day and only 200 on another day. It's quoted wide, but the midpoint of the spread is a better indicator than the last trade which occurred 3 hours ago. Also, the other person willing to give up 20 ticks per share for a mean-reverting trade would come out a loser more often than not. I'm quite dumbfounded by the assertion that you can give up so much alpha to reduce data costs and to create a uniform environment. When HFTs play the game in microseconds (and in the future nanoseconds) how can anyone expect to play in minutely insane minutes. I guess I need to sleep on it.

Abraham,

Take a chill pill. Even if you trade direct with IB, you'd get eaten alive by the HFT shops if you are truly trading a a High Frequency strategy requiring such rapid execution. IB samples prices three or four times a second which seems fast but is very slow for HF traders. Speed is always relative.

First of all, I have no affiliation with Quantopian and am not trying to defend them. However, unless you are a professional working for a HF hedge fund or an investment bank, you are unlikely to find a platform on which to trade with such low latency.

I believe 1 min bars is a very reasonable compromise, especially given data licensing issues. As Dan says, the trades only take a second or two to execute after the close of the bar. If you need to trade with higher granularity, you can use more complicated orders using stops, limits, OCAs or combination thereof that would hit intrabar if the price triggers.

That said, when you request data from IB you can specify Bid, Ask or Trade data. I assume the price bars are mid, but it would be nice to access the same level of pricing granularity.

In general, I don't like responding to anonymous posters, but I will this time. A particular hedge fund may have several desks each deploying strategies, some of which look like (or are) HFT strategies, while others deal with illiquid stocks. The requirements for each strategy may differ, some absolutely requiring gpu or fpga, and co-lo and true-real-time, others less restrictive, but none using aggregated minute data. Heck, if you are going to use minute data, why not just use free 20 minute delayed data? As for C/C++ vs Java vs Python, it will depend on the type of strategy. Some strategies require C/C++ with user controlled memory management, others are relaxed enough to use Java's ephemeral garbage collection, and so on and so forth. But using stale data, whether it's one minute or 20 minutes, just doesn't make sense. And to Colin Alexander, take your own advice about chill pills and learn more about market infrastructure and trading. For the type of strategies which a retail investor can deploy IB provides a decent environment comparable to many institutional products from EMS/OMS providers and brokerage firms. So it would really make sense to have Quantopian properly interface with IB and for customers who already pay for IB's data, to use that data. Quantopian should have handlers for different data feeds, rather than hard coding to one.

I thought the java garbage collector problems were largely solved if one moves to the Azul JVM? I was planning on doing some Scala on that assumption...

Simon, you might be right. I'm not a Java programmer, although I took a Udacity course on Java. Check with Java gurus.

Abraham,

Sorry to have offended you, but I just meant that professionals can and will build there own systems interacting directly with there brokers, whereas Quantopian appears more geared towards retail/semi-pro. I agree that it would be nice to use data that you already pay for from IB, but I can imagine challenges that would present Quantopian with potential licensing issues and with system development where each system is fed data from that user's IB account.

Hi Grant,

We read every comment posted by our users, and we carefully consider all of them. You really don't need to keep repeating the same question over and over again.

Thanks,

Jonathan Kamens

Sorry...I removed the post. I agree that you do a great job of listening...no need for me to use a bull horn. --Grant

But what is the response to Grant's question for those of us who have not seen it before, and have not seen your answer before?

Hello Abraham,

I asked a similar question here https://www.quantopian.com/posts/help-needed-with-batch-transform to Grant a while ago about Quantopian publishing a change log and I was directed to zipline on Github.

My experience - and I suspect Grant's - is that in my I.T. career that testing, version control, change management and communication to users are absolutely critical. I would expect this from a system that potentially puts my money at risk.

P.

Grant asked:

Any discussion within Quantopian and with Interactive Brokers of setting up a true test instance, that is walled-off from your production instance? Eventually, it seems, you're gonna need one, so that you don't put users real money at risk with upgrades. Within the IT world, this is the best practice, as I see it.

I do agree that we need to build more test mechanisms into our system. We'll get there in the long run. While we build out our platform there will continue to be sharp edges that our customers need to be aware of. For instance, we still have a number of single points of failure in our platform. As we mature, we will become more robust and one by one remove those failure points. Rome was not built in a day, nor was a fully redundant trading platform.

As a side - I don't think you need any special IB test instance. IB paper trading is exactly what you need. If you can run your algo against IB paper, then you can do it in real money.

Here's the way we are managing the testing problem for now. First, we make sure we "fail safe." When there is a problem, the algorithm shuts down and does nothing. That prevents trades from being placed that are unexpected. Linked to "fail safe" is "prompt notification." If your algo stops working, then we need to let you know promptly so that you can manually unwind any positions that you think are too risky. For some strategies, that's not a good set of operational parameters. For others, it works just fine. If you are running an algo that rebalances weekly, you don't mind if you get an outage here or there. If you are going in and out of the market every minute, then you need our platform to mature more before you invest with us.

Over time we'll get more robust and we'll start being able to guarantee uptimes. Guaranteed uptimes will certainly include more robust testing methods.

I hope that answered the question.

Part of the problem as I see it is that there's a high "hurdle" to the business model. The expected value of a catastrophic loss of capital due to bugs and upgrade risk is larger than any reasonable beta-testing rebate. In that sense, an algo-platform-as-a-service doesn't lend itself to the "build fast, break things" startup model, since rational user uptake will be nil until the financial and operational risk control systems are in place. In a way, the risk control IS the business opportunity, since it's what everyone hates to build, but is also exactly what is so vital to avoid terminal drawdowns. On the operational side, fully versioned rollback-able environments, instance boot-up unit tests and algo regression tests, connectivity guarantees and failover plans (what happens when EC2 goes down?). On the financial risk side, position limits, risk limits, monte-carlo simulation for VAR limits across multiple algos, etc etc.

None of this is the sexy code, nobody (that I know) likes doing it, it's extremely important, and therefore it is what people might pay to outsource. Finding profitable alpha models is very hard, but to be honest, Quantopian's platform is not good for that anyway. Once one or more is found, given the data, coding one up for python in Quantopian vs C# in QuantConnect vs TradeStation vs MATLAB vs C++ vs ..., I mean, it takes time, but not too much time. There aren't any big barriers to entry or exit there. The pain points which can be solved for added value and vendor lockin potential are further up the stack.

My 2c!

I guess my next question is whether Quantopian, its parent, or its subsidiary, is a registered Broker-Dealer or not.

Thanks Dan,

I'm not trying to create fear-uncertainty-doubt, but rather to highlight a standard approach to beating down the risk. With regard to "I hope that answered the question" you have not directly, but that is o.k. Reading between the lines, obviously you are aware of the walled-off test instance approach and chose not to implement it, but might in the future.

As these things go, sometimes there are spin-off insights, such as Simon's. The over-arching problem from a user standpoint is "What will I get for the risk?" You definitely have a "build fast, break things" approach, but you really need to move toward something like highly engineered, highly reliable, proven firmware. Personally, I concluded that there is more downside risk than upside benefit to being an early-adopter of Quantopian-Interactive Broker real-money trading (aside from my total lack of trading experience and probably five other valid reasons).

Grant

So to answer my own question, at this time Quantopian is NOT a registered Broker Dealer (BD) and does not need to comply with (but should seriously consider compliance with, nonetheless) SEC Rule 15c3-5 (17 CFR Part 240)

https://www.sec.gov/rules/final/2010/34-63241.pdf

On the flip side, not being a BD severely limits Quantopian's ability to charge on a transactional basis.

I agree with Anony's sentiment regarding eating your own dog food (hey, if I can't figure out algorithmic trading, at least I can learn some slang). Perhaps this is already happening, but not publicly, since I did see somewhere a comment by Jess Stauth that the algorithm posted by Dan above has been running live on Interactive Brokers (but perhaps only paper trading).

In some sense, Quantopian wants to have its cake and eat it too, by getting lots of free testing help and community involvement, but keep other aspects of its efforts and business plan private. One of the challenges for users is that we don't have context. Presumably, as a business, you need to do certain things by a certain time under certain budget constraints--we have no idea how many days/weeks/months/years you have to achieve milestones to thrive as a business. What do you need to do and when?

ICYMI check out our latest blog post (http://blog.quantopian.com/real-money/) for a more in depth update on the real money pilot, including the first month of performance for the real money algo we introduced at the top of this thread. Here is that algo's performance on a real money account updated through market close today:

I'm curious what Quantopian can do for the student market? Granted, backtesting things is great and even just having access to this data is a huge privilege, but when it comes to transitioning from the "laboratory" to the actual market, students are often inhibited by exceedingly low account balances. Would Quantopian ever be interested in doing anything with student discounts or the like? I would hate to see this forum become too focused on its heavy hitters and lose the educational value that I have gleaned from it. I don't necessarily have any short-term fears of that happening, but I would be interested in seeing what could be done to help students and other people with limited capital. I know IB has some minimum account balances and the like, and I understand the limits that puts on Quantopian, but how could this challenge be addressed? Other brokers maybe?

I apologize if this question seems outside the scope of this thread, but it seems relevent and doesn't seem to have been discussed here yet.

Thanks Jess,

I too had to look up ICYMI (and "sentencronym" ain't a real word, Anony). Does the Quantopian dashboard include all of the user costs associated with the trading? Also, when Quantopian starts charging, will the algorithm cost be rolled into the dashboard performance metrics/chart? Presumably, the (monthly?) charge would be taken out of the cash balance associated with the algorithm, so it could be tracked on a per algorithm basis. Or would accounting for costs need to be done separately from the algorithm dashboard?

Grant

Hello Jess,

Looking at the screenshot above I can see that $802.14 is (more or less) 2.70% of the $30,0000 starting capital. But $1,525.60 + $28,618.98 = $30,144.58 which is quite a distance from $30,802.14. What am I missing?

P.

Hi Peter,

I just went back and checked, looks like the starting capital base when the algo was deployed was actually $29,342.44. Then the logic in the algo aims to allocate out 95% of the portfolio value, leaving 5% as a cach buffer. I think that all works out, but if you find it doesn't let me know!

Best wishes,
Jess

Hello Jess,

Thank you. So the return is 2.73%? Sorry...

P.

Hi Peter,

might be there were a few trades/commissions that hadn't cleared yet, I can get it within $10 from the numbers in that snapshot as follows:
$29,342.44 * 2.7% return = 30134.69


long exposure + cash = 30144.58

I just checked the current numbers and can get it within a dollar or two - so it looks to be accounting for things the way I expect.

Hi Jess,

What is the accounting for commissions? The anticipated Quantopian monthly algo charge?

Grant

Jessica, Peter's observation is spot on. The return should be measured on the portfolio value including the cash buffer, not just the amount invested. Thus it would be approximately 2.7% * 95% = 2.57%.

Or at least it should be clear what the base unit is on which the return is measured against.

Hello Jess,

I'm happy with your numbers but I think 'Return' should be displayed to one decimal place as '2.7%' is almost certainly correct but '2.70%' is very likely not correct.

P.

Made some changes to allow for variable weighting of the portfolio to be rebalanced. I chose the weights arbitrarily, so I'd suggest researching a bit and assigning your own weighting. Changing context.useEqualWeights = True will return the behavior back to Dan's original implementation.

Hello All,
How long does it take for paper trading beta application to be approved? I submitted my form 2-3 days ago but have not heard back yet.

Hi Dev,

Thanks for submitting the form and you are now all set to begin live trading! I have sent you an email to follow-up.

Best,
Alisa

Not sure if this is a good place to ask this question.
In the example live trading algo, I am wondering if anyone is able to clarify why context.weights is defined like this:

context.weights = 0.99/len(context.secs)  

Instead of like this:
context.weights = 1/len(context.secs)

Additionally, I am having trouble understanding when this if statement would ever not be true. Context.rebalance_date will always equal none (or whatever else you define it as), so what is the purpose of including the first part of the if statement?
if context.rebalance_date == None or exchange_time > context.rebalance_date + datetime.timedelta(days=context.Rebalance_Days):

On the weights, my guess is that it has to do with rounding in Python and the desire to keep the sum less than or equal to 1, rather than a tad over 1.

Perhaps also to leave enough money to cover fees?

Tim- I don't think that is the case as there is a separate function handling commissions in line 32:

set_commission(commission.PerTrade(cost=0.0))  

Abraham- Even with 1 instead of .99 the sum will be less than or equal to 1 as 1/9=.1 repeating. And I don't think there would be any negative implication if the sum was slightly greater than 1 instead of less than 1- the amount ordered via the order_target_percent function would just differ very slightly.

Good point - I didn't realize that function was only being used for backtests at first glance. It doesn't sound like Quantopian has any sort of special agreement with IB on commissions so it seems like a bit of a losing proposition to just rebalance with 1 or 2 share trades @ $1 a pop. I guess it is intended to be more of a demonstration that live trading is possible.

Hi guys,

I am running this algo in a real money account - but part of the purpose of running it is to get burn in with seeing orders placed and filled - so that's why I'm asking it to rebalance every day, though that generates much more trading than you'd need to track the target portfolio (we don't have an volume discount with IB at this point, so there is indeed a $1/trade minimum fee). In order to rebalance less often you can modify the rebalance parameter to something like:

context.Rebalance_Days = 21

Like I used in the simple asset allocation example I referred to further down in the thread.

Also I left a 1% cash buffer in the account just to be conservative and know that I'd have a buffer in case there was some rounding issue that came up - it hasn't been an issue however and that's probably not necessary in this case.

That makes sense, better safe than sorry. I'm having trouble finding your example of modifying the rebalance parameter, do you think that you could include a link?

Sure thing Colin - this example was posted earlier above and maybe in a separate post as well (but might have been edited, so here it is again)

See the section below, you can edit the context.rebalance_days variable and your algo should only trade every N days based on that.

# Initialize context variables the define rebalance logic:  
context.rebalance_date = None  
context.next_rebalance_Date = None  
context.rebalance_days = 28  
context.rebalance_window_start = 10  
context.rebalance_window_stop  = 15

This is a different strategy that makes allocations across a diversified portfolio of asset classes (US and developed ex-US stocks, bonds, real estate, and commodities). We initialize the last and next rebalance date variables as 'None', and also set a trading window of 10am to 3pm (the strategy will only try to rebalance in that window and once a rebalance has occurred the last and next rebalance date variables are updated).

Hope this is useful and please let me know if you have additional questions!
Best, Jess

Thanks for the response. I realized where I was getting confused- had trouble finding where context.rebalance_date was set in my algorithm because ctrl+F wasn't finding anything beyond the viewable code. It would be pretty cool if there was version control on algorithms- do you guys have any plans to implement this in the future?

I have a few questions regarding live trading this algo:

• I created another IB account which is hooked up to my original live trading account and moved a few thousand bucks in there to test with. Both accounts are accessed using the same username. I added the new live test account to Quantopian (but did not add my main account to Q). This should work yes? ie. Q wont accidentally hook into my main account (hopefully)?

• Say that Q did hook into my main account and say that this account has $100k in stock/options, $30k in cash, and since it's a portfolio margin account there is $500k buying power. If this algo went live on that account how would it trade? would it use the cash only or would it go after the entire margin?

Thanks!

I created another IB account which is hooked up to my original live trading account and moved a few thousand bucks in there to test with.

That seems like a prudent strategy. ;-)

Both accounts are accessed using the same username. I added the new live test account to Quantopian (but did not add my main account to Q). This should work yes? ie. Q wont accidentally hook into my main account (hopefully)?

That's correct. If you don't give Quantopian the account number of your main account, it can't access it.

Say that Q did hook into my main account and say that this account has $100k in stock/options, $30k in cash, and since it's a portfolio margin account there is $500k buying power. If this algo went live on that account how would it trade? would it use the cash only or would it go after the entire margin?

Quantopian doesn't enforce any limits; you would be able to trade whatever IB lets you trade.

Thanks for the reply, That clears things up. So am I correct in assuming that this line

context.weights = 0.99/len(context.secs)

determines how much of the total account is available to this algo? .99/9 = .11 so each position will be valued at 11% of available capital? So if that .99 is changed to .33 then positions will be sized to 3.6%?

@Beau, yes that's correct

@Alisa, Thanks for your response. If I were to start live trading with one IB account and decide few months later to switch my IB account ( say move from individual account to IRA account) will I still be able to keep my one free lifetime algorithm?

Hi Dev,

The one free lifetime algorithm is a "slot" that is good for any single real money algorithm, covered up to $100,000, on your Quantopian account. In the future, when we begin charging for the live trading service, you will not have to pay to live trade one algorithm. This can be any algorithm. You can start it, stop it, change it, swap it out etc. You will have one slot for a live trading algorithm with no fees. You will receive this promotion if you deploy a live trading algorithm by March 31, 2014.

Thanks @Alisa.
I have deployed live trading algorithm. Looking forward to working with your platform. Thanks and this is awesome!

Running a cloned algo in my IB simulator account.

It's 9:41am but status shows MARKET CLOSED.

Click on logs: Waiting for logs...

Live Algorithm Settings
Status: Broker Login Needed
Started Trading: 3/21/2014 9:28:30 AM
Capital base: Unknown
Source Algorithm: Cloned from "Paper Trading With Interactive Brokers - Open Beta Launch" (Go to algorithm )
IB Account: Abe @ IB simulator (...)

Also, how do I log out of my IB account from the Quantopian interface? Like when I want to access Trader Workstation directly to reset some account parameters, or to monitor what's going on?

Best,
Abe

Hi Abe,

If you log into your live trading dashboard before market open (ie 9:29AM) it will say MARKET CLOSED. Once the market is open, you will need to refresh your browser to see the updated status and positions.

Regarding your question about logging into IB and Quantopian simultaneously, see my answer in this thread. The short answer is that IB only allows one connection per account. If you have multiple connections (running a Quantopian algorithm and logging into TWS/Webtrader) IB will terminate one connection. To prevent this you can setup a second login (for free!) for real money accounts.

For more information, take a look at our live trading FAQ.

Best,
Alisa

Hi Alisa,

It seems the browser display real-time stats (or near real-time). Why would the "market closed" area not be able to self-refresh?

Best,
Abe

I agree, the browser should refresh upon market open, but it's a feature that we haven't built yet. We're working hard to get the infrastructure up and running for live trading. This is one of the enhancements left on our to-do list :)