SmartQuant Discussion

suppose I'm a complete newbie, no broker account, no market experience
but programing experience and a simple idea for a trading strategy,
easy to implement.

How much time and money does it take me to get my strategy programmed
and executed on a real market -say- forex for -say- 1 month ?


all inclusive, creating an account, buying a computer + software,
creating internet connection , installing, transferring money


your estimates, please

This is a big question with a lot of different variables.

With Interactive Brokers you can open an account for 5k, and 25k if you want margin.
Registration with IB after the paper work takes about a week.
Cash transfer takes an additional week.
http://www.InteractiveBrokers.com

If the system generates a lot of cash you should incorporate for tax purposes.
Incorporation takes about 4-6 weeks

As for hardware it depends on the system you want to trade.
If its simple (just a single stock) then just use the PC that you have right now.
If you need more horsepower I'm a big fan of second hand Dell 1955 servers.
They are getting cheaper and they come with a blade enclosure for easy management and job deployment.
Great for monte carlo, machine learning, and training neural networks.
http://www.stikc.com/Catalog/PowerEdge-1955

Internet Connection.
If you auto trade all day long you will definitely need a quick internet connection as well as a backup internet connection. 100-150/month
Don't forget APS (power backup). 250.
(At this point you could also consider finding a hosted solution close to a fiber end point - then it becomes a number crunching game)

Software
OpenQuant - check the web site
IB's TraderWorkstation - free from their website.
If you need to do some math I would reccomend R or Matlab.
R is free, Matlab is 5k.

Programming ?
If you can do it yourself then just take the time and learn C# and the OpenQuant API.
If not, there are people in the forums that do it for $. (See rent a coder)
Offshore is around 20/hr, home grown could be much more (50-100/hr).
Depending on complexity you could be looking at a week, month, or longer of work.

I would suggest getting this book.
It will answer a lot of questions for you.

http://search.barnesandnoble.com/Quantitative-Trading/Ernie-Chan/e/9780470284889

Just a word of advice.
Don't invests large sums until you can make sure the idea works.
If you don't have OQ yet you can always use excel and do your calculations there to see if your idea works.

CS

isn't there a company, a service provider, or even a private investor, who can do all this
for me, while I just provide the strategy in form of a computer program and the money ?

And I pay the fees and a to-be-negotiated percentage of the predicted/assumed/hoped wins,
while possible losses are paid by me entirely from the provided money=margin ?

Or would that be illegal, do I _have_ to read and understand these dozends of pages
in the terms and conditions of the brokers, trading-software-providers, barnesandnoble books

maybe it's legally possible then -and easier- in other countries ?

You can do it legally it shouldn't be a problem.
But the legal documentation required might be hefty.
You would definitely want a legal contract between you and the hosting company.

What if:
they didn't back test?
they missed a test case?
the strategy didn't work?
the strat you gave them works but the one they coded didn't?
there was a system system outage?
worse...
there was a system outage with open positions and the market tanked,
ahhh the sweet sound of liability bells ringing! =D

(just a little added humour there!)

Also it would also probably depend on the strategy.

Intraday strategies can generate a lot of capital quickly (or lose it twice as quickly) and they are usually quite difficult to get perfect. (anything less than perfect is almost always a money loser)

Usually intraday strats need to be adaptive meaning you need to train, tune, optimize them at the end of every day.
Better... they are tuned intraday while trading to take in to account the changes in volatility, frequency, cycle length, distance to/from the mean.
(that's what a server farm is good for... training and tuning your system intraday)

Trading daily / weekly / monthly trading is a little less intensive than intraday but still requires some work.
Usually because its (somewhat) easier to detect a stationary pattern of some kind in daily data.
But you need to make sure the software is up and running throughout the day, every day, without fail.

They might also be able to host it and maintain it for you as well.
But as with all service based contracts this would most likely come at some cost.
Fixed price 100/day 1000/month?
Per trade? 2$ 5$ ?
Worse yet... Per share? ugh!
(no idea... just throwing out conversation numbers)
After fee's does the strategy still make money?

For strategy development you should be able provide a trading strategy to a developer or development firm (local or remote) in the form of a paper document, working system that you want to automate, previous trading results, pen on napkin, and the they should be able to build a system from that information. After development comes the task of back testing, estimating slippage, estimating missed trades (not getting a position filled), and looking for something missing. You should also verify that the results are statistically significant. (was this just pure luck or was there some sort of quantifiable significance about the results.)

Then... verify the results, paper trade for a couple days and then go live! =D
Bingo, bango, bongo!

One last item to chew on is portfolio optimization and risk.
Do you keep the same position size for every trade or do you pyramid / martingale.
inversely proportional to volatility, (to mitigate risk)?
proportionally to their probability of prediction?
(The be sure to optimize the thresholds on these variables as well)

I hope this was helpful!

=D
CS

> What if:
> they didn't back test?
> they missed a test case?
> the strategy didn't work?
> the strat you gave them works but the one they coded didn't?
> there was a system system outage?
> worse...
> there was a system outage with open positions and the market tanked,
> ahhh the sweet sound of liability bells ringing! =D
>
> (just a little added humour there!)
>
> Also it would also probably depend on the strategy.

I would just stop.
I would start with small amounts to test it and then increase only if it works well.
When they make an error, the chance is still almost 50%-50% that it produces win or loss.

> Intraday strategies can generate a lot of capital quickly (or lose it twice as quickly) and

the amount is specified by the order

> they are usually quite difficult to get perfect. (anything less than perfect is almost always a money loser)
>
> Usually intraday strats need to be adaptive meaning you need to train, tune, optimize them
> at the end of every day.

this is a separate process, separated from the mere ordering.
It can be tested with small depots, or just virtual data

> Better... they are tuned intraday while trading to take in to account the changes in volatility,
> frequency, cycle length, distance to/from the mean.
> (that's what a server farm is good for... training and tuning your system intraday)

the strategy will be adapted due to experience from actual orders.
But again, that's another problem.

> Trading daily / weekly / monthly trading is a little less intensive than intraday but still requires some work.

same program, same work in principle. Start with daily,weekly and then maybe proceed to
hourly,minutely,millisecondly when it work well.

> Usually because its (somewhat) easier to detect a stationary pattern of some kind in daily data.
> But you need to make sure the software is up and running throughout the day, every day, without fail.

yes, extra-connection. Buy the whole system from the broker.
Include your executable to regularly read quotes and account data, analyze them
and submit orders.

> They might also be able to host it and maintain it for you as well.

good !
I just submit the money and my program

> But as with all service based contracts this would most likely come at some cost.
> Fixed price 100/day 1000/month?
> Per trade? 2$ 5$ ?
> Worse yet... Per share? ugh!
> (no idea... just throwing out conversation numbers)
> After fee's does the strategy still make money?

fees will go down. There are no broker-costs once everything is automated.
It just makes sense, so it will come earlier of later IMO.
(in countries that legally support it.)

> For strategy development you should be able provide a trading strategy to a developer or development
> firm (local or remote) in the form of a paper document, working system that you want to automate,
> previous trading results, pen on napkin, and the they should be able to build a system from
> that information. After development comes the task of back testing, estimating slippage,
> estimating missed trades (not getting a position filled), and looking for something missing.
> You should also verify that the results are statistically significant. (was this just pure luck
> or was there some sort of quantifiable significance about the results.)

that's the task of the system developer, programmer. The broker would just
provide the compatible platform.

> Then... verify the results, paper trade for a couple days and then go live! =D
> Bingo, bango, bongo!

or just do simulated,hypothetical,no-money trading or small amount trading to test it.
The broker could support this as well - lower liability,lower fees.

> One last item to chew on is portfolio optimization and risk.
> Do you keep the same position size for every trade or do you pyramid / martingale.
> inversely proportional to volatility, (to mitigate risk)?

the strategy will calculate it. Basically optimize the expectation value,
decreasing the performance-volatility can be done by increasing the number
of transactions with same amount of invested money

> proportionally to their probability of prediction?
> (The be sure to optimize the thresholds on these variables as well)
>
> I hope this was helpful!

where can I find more about it ?
It should exist already ... so important !

>I would just stop.
>I would start with small amounts to test it and then increase only if it works well.
>When they make an error, the chance is still almost 50%-50% that it produces win or loss.

Where I was going with this was that it is usually the speed at which the system can trade that can make a win or loss. So you're saying that you would put a timer on the system and only allow it to trade once every minute, or once every five minutes?
This is possible.

>> Intraday strategies can generate a lot of capital quickly (or lose it twice as quickly)
>the amount is specified by the order

Its not really the order size but the frequency at which the orders are executed.

>> Trading daily / weekly / monthly trading is a little less intensive than intraday but still requires some work.
> same program, same work in principle. Start with daily,weekly and then maybe proceed to

> hourly,minutely,millisecondly when it work well.

Yes that plan would work. (Just turn the juice up on the system!)

>> They might also be able to host it and maintain it for you as well.
>good !
>I just submit the money and my program

>> But as with all service based contracts this would most likely come at some cost.
>> Fixed price 100/day 1000/month?
>> Per trade? 2$ 5$ ?
>> Worse yet... Per share? ugh!
>> (no idea... just throwing out conversation numbers)
>> After fee's does the strategy still make money?

>fees will go down. There are no broker-costs once everything is automated.
>It just makes sense, so it will come earlier of later IMO.
>(in countries that legally support it.)

There should always be a broker fees.
OpenQuant only connects you to a broker like Interactive Brokers for automating your trading system.
OpenQuant still needs to talk to a broker to execute the orders.

And the fee's that I mentioned above are in addition to the broker fees.
These fees would be the monthly maintenance charges from the hosting company.

You mentioned "legally" a couple of times.... I think I see where you're going with this.
So do you mean legally support it because the hosting company could fall under OSC regulation because they are providing a financial gateway?
Hmmmm... interesting...

Just to give you an answer... Sorry... I don't know of anyone offering retail OpenQuant hosting.

QuantHouse (will host SmartQuant/OpenQuant solutions for you at the exchange) but from what I've read they only deal with institutional traders and I can't even begin to imagine what the hosting fees are like.

Taking a slight step back...

Really, its not too difficult to download and run OpenQuant.
After watching the videos and playing around with it you should be able to get the hang of running things fairly quickly. As for strategy development you would need to get the hang of C# which could take some time.
But are a large number of C# tutorials out there.

For maintenance...
Scripts can be setup to automate the nightly download of data run optimization scripts.
You'll probably still want to review the historical back tests and keep an eye on things.

If you don't want to do this yourself you could always start up a company and hire some software contractors to do it for you and get them to maintain it. ( This might also get around any OSC concerns. )

At this point the discussion starts to sound a bit like that Barenaked Ladies song...
http://en.wikipedia.org/wiki/If_I_Had_$1000000 =D

Anything is possible with a strategy that can consistently generate winners.

=D

CS