IBrokers Presentation

Algorithmic trading with R

Kyle [email protected]

Finance department, Anderson School of Management, UC Los Angeles

September 28, 2010

Kyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 1 / 28

mailto:[email protected]


Before we start

Identifying by name certain brokerages, assets, programminglanguages, etc. is a necessity for clarity and concreteness. Pleasedo not mistake this for an endorsement of any of them. I derive nopersonal gain from their mention, and, to the extent that I simplydescribing them, I do not recommend them.Please understand that I can take absolutely no responsible foranything that anyone does subsequent to viewing this talk.What I am discussing is staking real money on ones ability todevelop and implement a strategy superior to many very smartpeople with the same goal.



Outline

1 Some BackgroundAlgorithmic trading at a high levelIndividual investing

2 Algorithmic trading in RDetailsImplementation

3 Further



What is algorithmic trading?

To the best of my knowledge, there is not a well-formed definitionof algorithmic trading, so absent any strong guidance I woulddefine algorithmic trading as a trading strategy that executestrades without human oversight.Statistical models feature prominently in algorithmic tradingstrategies.



Who is trading algorithmically?

Big banks: Bank of America, Citigroup

Sell side: Goldman Sachs, Credit Suisse

Buy side investment funds (better execution): Fidelity,BlackRock

Hedge funds: Famous players include RenaissanceTechnologies, Citadel, and AQR

Market makers: Getco, Madison Tyler Holdings, SusquehannaInternational Group

Small investors: Me, and potentially you!



How they are trading algorithmically

As a rule1:Lots of highly-paid smart people working on sophisticated secretalgorithms to be implemented by tons of developers in low levellanguages on high-tech hardware located proximate to theexchange’s servers.

Which begs the question . . . How should I be tradingalgorithmically?

1This is almost entirely anecdotal: gleaned from popular reporting, aminimal amount of academic research, the boasting of friends, as well as thepuffery of firms and headhunters trying to get me to interview.Kyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 6 / 28



As a rule1:Lots of highly-paid smart people working on sophisticated secretalgorithms to be implemented by tons of developers in low levellanguages on high-tech hardware located proximate to theexchange’s servers.Which begs the question . . .

How should I be tradingalgorithmically?




As a rule1:Lots of highly-paid smart people working on sophisticated secretalgorithms to be implemented by tons of developers in low levellanguages on high-tech hardware located proximate to theexchange’s servers.Which begs the question . . . How should I be tradingalgorithmically?



What we will not be discussing tonight

Living in Los Angeles one is already effective excluded from ultrahigh frequency trading.



What we will be discussing tonight

So we rely on our ability to take favorable action on the futureevolution of prices.

A difficult exerciseBut one at which we have a shot at success




So we rely on our ability to take favorable action on the futureevolution of prices.A difficult exercise

But one at which we have a shot at success




So we rely on our ability to take favorable action on the futureevolution of prices.A difficult exerciseBut one at which we have a shot at success



And how do we do this?

By building and implementing profitable forecasting models. Theformer is up to you2, the latter is what I will discuss tonight.Specifically, I will guide you through the steps necessary toautomatically trade assets using R and the IBrokers libraryavailable on CRAN.

2for those unsure where to begin, “pairs trading” and “market making” arenatural strategies to read aboutKyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 9 / 28


But first, a bit of background on investing

How do you and I trade? Through a brokerage:

In person or on the phone

Online (slow): Scottrade, E-Trade

Online (fast): Ninjatrader, thinkorswim, or . . .

















Interactive Brokers

A well-regarded online “deep discount” broker. Some quick detailsrelated to a trading strategy3

At least 18 years of age, not binding in light of:

Substantial experience trading desired asset classes measuredin years (honor system)

At least $25,000 in capital to trade actively (United Stateslaw)

Must pay at least $30 / month in commissions, or a small(≈ $10) fee is assessed for access to market data

3not exhaustive, read the (large quantities of) fine printKyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 11 / 28


Trading platform

To have a completely automated trading system, what would weneed? At a high level:

Data access and management

A framework for analyzing and implementing trade strategies

Order handling and execution

Platforms abound: a surfeit of proprietary software, along withsome moderate open source work, generally driven by Python orC++. As I will show presently, an Interactive Brokers account +IBrokers package handles the data access and order handling. Revidently excels at the rest.



About IBrokers

Implements high level wrappers to the Interactive BrokersJava API in R

Still in active development (last release: early September2010)

Some necessary logic not fully supported

A few short vignettes, minimal activity on the lists, and somepresentations given by Jeffrey Ryan (the package’s author),but still sparsely documented



How IBrokers4 works at a high level

1 Get some data from the exchanges

2 Take some action based on that data, goto ↑

More concretely, “some data” might be a price at which traderstands ready to buy or an update to the total volume for the day.“some action” might involve incorporating this new realization intoa model and then taking action on the output of this model.Happens dozens of times a second.

4and event processing in generalKyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 14 / 28


Here are the necessary steps to get going with R + IBrokers



Sign in

Sign in via GUI: TWS or IB Gateway. After an aggravating phonecall to technical support I have ascertained that they do notsupport automated login (e.g. via command line argument), sopossibly roll your own.



Enable connections



Instantiate connection object

In R, create a connection object with the twsConnect command

1 l i b r a r y ( I B r o k e r s )2 tws <− twsConnect ( )



Start looking at some data

The commands5

1 t i c k e r <− ”SPY”2 c on t r a c t <− twsEqu i t y ( t i c k e r , ”SMART” , ”ISLAND” )3 reqMktData ( tws , c o n t r a c t )

will dump to the console all of the data coming in from InteractiveBrokers, based upon the standard methods given in the defaulteWrapper argument.

<20100519 06:29:01.340431>,id=1,pos=0,operation=insert,side=bid,price=29.67,size=1<20100519 06:29:01.379101>,id=1,pos=1,operation=insert,side=bid,price=29.26,size=4<20100519 06:29:01.381317>,id=1,pos=2,operation=insert,side=bid,price=29.17,size=34<20100519 06:29:01.383101>,id=1,pos=3,operation=insert,side=bid,price=27.68,size=1<20100519 06:29:01.384798>,id=1,pos=0,operation=insert,side=ask,price=30.08,size=1<20100519 06:29:01.386646>,id=1,pos=1,operation=insert,side=ask,price=30.09,size=2<20100519 06:29:01.390194>,id=1,pos=2,operation=insert,side=ask,price=30.95,size=34<20100519 06:29:01.392034>,id=1,pos=3,operation=insert,side=ask,price=32.27,size=1

5“SPY” is the ticker for a leading S&P 500 index exchange traded fundKyle Matoba (UCLA) Algorithmic trading with R September 28, 2010 19 / 28


An eWrapper object is composed of an environment in which tostore data, accessor functions, and methods executed wheneverdata of a particular type is received.

> head(eWrapper(), 5)

$.Data

<environment: 0x102632430>

$get.Data

function (x)

get(x, .Data)

<environment: 0x100d4e4e8>

$assign.Data

function (x, value)

assign(x, value, .Data)


$remove.Data

function (x)

remove(x, .Data)


$tickPrice

function (curMsg, msg, timestamp, file, ...)

{

symbols <- get.Data("symbols")

e_tick_price(NULL, msg, timestamp, file, symbols, ...)

}




A natural next step is to tweak some of the eWrapper methods; for

example, I began by storing incoming data in a circular matrix (within my

custom eWrapper()-like function):

1 eW$ t i c k P r i c e <− f u n c t i o n ( curMsg , msg , t imestamp , f i l e , . . . ) {2 t i ckType = msg [ 3 ]3 msg <− as . numeric (msg )4 data <− eW$get . Data ( ” data ” )5 i ndex <− eW$get . Data ( ” i ndex ” )6 data [ index , ] <− NA78 f i l e <− u n l i s t ( f i l e )9 i f ( t i ckType == . twsTickType$BID){

10 cat ( paste ( t imestamp , msg [ 5 ] , msg [ 4 ] , ”” , ”” , ”” ,11 ”” , ”” , sep = ” , ” ) , ”\n” , f i l e = f i l e , append = TRUE)12 data [ index , 1 : 2 ] <− msg [ 5 : 4 ]13 data [ index , nc ] <− as . numeric ( Sys . t ime ( ) )14 }15 e l s e i f ( t i ckType == . twsTickType$ASK) {16 cat ( paste ( t imestamp , ”” , ”” , msg [ 4 ] , msg [ 5 ] , ”” ,17 ”” , ”” , sep = ” , ” ) , ”\n” , f i l e = f i l e , append = TRUE)18 data [ index , 3 : 4 ] <− msg [ 4 : 5 ]19 data [ index , nc ] <− as . numeric ( Sys . t ime ( ) )20 }21 e l s e i f ( t i ckType == . twsTickType$LAST) {22 cat ( paste ( t imestamp , ”” , ”” , ”” , ”” , msg [ 4 ] , ”” ,23 ”” , sep = ” , ” ) , ”\n” , f i l e = f i l e , append = TRUE)24 data [ index , 5 ] <− msg [ 4 ]25 data [ index , nc ] <− as . numeric ( Sys . t ime ( ) )26 }27 i ndex <− ( i ndex %% h i s t l eng th ) + 128 eW$ a s s i g n . Data ( ” i ndex ” , i ndex )29 eW$ a s s i g n . Data ( ” data ” , data )30 }



Implement trade logic!

Either here, or by modifying the callback that functions likereqMktData take as an argument. The callback is literally anunending loop that listens for data, then passes it to theappropriate method

> deparse(twsCALLBACK)[35:47]

[1] " while (TRUE) {"

[2] " socketSelect(list(con), FALSE, NULL)"

[3] " curMsg <- .Internal(readBin(con, \"character\", 1L, "

[4] " NA_integer_, TRUE, FALSE))"

[5] " if (!is.null(timestamp)) {"

[6] " processMsg(curMsg, con, eWrapper, format(Sys.time(), "

[7] " timestamp), file, ...)"

[8] " }"

[9] " else {"

[10] " processMsg(curMsg, con, eWrapper, timestamp, "

[11] " file, ...)"

[12] " }"

[13] " }"



For example...

If one stores recent data as I have shown, then a basic strategymight be to periodically refit a time series model and use it toforecast the direction of prices a bit in the future.

Shameless plug: I am wrapping up a paper with Gareth James andPeter Radchenko on an adaptive lasso-like regularization forexactly this sort of problem. Email me for the code, or it will be onCRAN shortly.



Placing orders?

Not production-ready, but exported and works. Look at the sourceof placeOrder().



Some things to look at next

1 .twsIncomingMSG for a mapping of the low level messagetypes to the methods that process them

2 The event handlers IBrokers:::e order status, etc.

3 Recording data with eWrapper.MktData.CSV()



IBrokers resources

The Interactive Brokers API guidehttp://www.interactivebrokers.com/php/apiUsersGuide/apiguide.htm

The vignettes on CRANhttp://cran.r-project.org/web/packages/IBrokers/vignettes/

Some assorted talkswww.rinfinance.com/agenda/2010/JeffRyan_Tutorial.pdf

https://www.rmetrics.org/files/Meielisalp2009/Presentations/Ryan.pdf

Some scant banter on the listsRSiteSearch("IBrokers")


http://www.interactivebrokers.com/php/apiUsersGuide/apiguide.htm

http://cran.r-project.org/web/packages/IBrokers/vignettes/

www.rinfinance.com/agenda/2010/JeffRyan_Tutorial.pdf

https://www.rmetrics.org/files/Meielisalp2009/Presentations/Ryan.pdf


That’s it. Any questions?



Thanks for your time

Further questions, feedback, anecdotes, large cash payments, wantto talk more about this? Whatever it is, contact me at:[email protected]

This presentation will also be available at:http://personal.anderson.ucla.edu/kyle.matoba.2014/



http://personal.anderson.ucla.edu/kyle.matoba.2014/


About the code in this presentation

Everything was run in the environment:

> sessionInfo()

R version 2.11.1 (2010-05-31)

x86_64-apple-darwin9.8.0

locale:

[1] en_US.UTF-8/en_US.UTF-8/C/C/en_US.UTF-8/en_US.UTF-8

attached base packages:

[1] stats graphics grDevices utils datasets methods base

other attached packages:

[1] IBrokers_0.2-8 xts_0.7-1 zoo_1.6-4

loaded via a namespace (and not attached):

[1] grid_2.11.1 lattice_0.18-8

IBrokers Presentation

Documents

sophisticated

low level

algorithmic

trading algorithmically

kyle matoba

academic research

minimal amount

phone online