 Posted 2/13/2007 17:09:17
|
|
|
|
| Can RE backtest a portfolio of systems on a portfolio of stocks?
|
|
|
Posted 2/13/2007 17:18:38
|
|
|
|
| At this point it can back test a portfolio of stocks. If you want to run more than one system, you'll have to run more than one instance of RE. jlfvkRES (2/13/2007)
Can RE backtest a portfolio of systems on a portfolio of stocks?
|
|
|
Posted 2/13/2007 18:56:02
|
|
|
|
| I should add that you could run multiple trading strategies under the same RightEdge system. You would have to manage both strategies in the system code. If your strategies are simple, they might just generate signals to buy and sell. So you would your strategies and then write the main RightEdge system to check the signals the strategies are generating and take action. If you wanted to allocate a certain percentage of your capital to each strategy, for example, you could write code that would manage that. If you want to do something more complicated, it may be more complicated to write, but it should be possible. You have the full power of C# or Visual Basic, so you can code it pretty much any way you want. Thanks,
Daniel
|
|
|
Posted 2/14/2007 00:38:39
|
|
|
|
dplaisted (2/13/2007)
You have the full power of C# or Visual Basic, so you can code it pretty much any way you want.
Let me jump in here.
Of course "evrything is possible" with a modern programming language. On the other hand the whole reason of being for software packes like RE is simplification:
I write some lines for my trading system and it runs in a specialized environment.
In this context it becomes highly preferable if there can be several "simultaneous" systems - not specially coded into one source code, but simly some checkmarks before the names of existing, simple systems.
This has always been one of the major weaknesses of WL and the top feature request for WL over years.
www.finantic.de
Our Trading System at C2: Topaz
|
|
|
Posted 2/14/2007 07:38:17
|
|
|
|
| Agreed. I think the point was really to offer a work around given the current situation. I will say that we really haven't received many requests for this type of feature, but we're definitely open to the idea. While I have you both (and others feel free to chime in), I'd like some ideas/opinions/suggestions on how multiple systems would make the most sense to you. With the current environment you load a "project", much like any development environment. This project contains files that build a single program. In other words, in Visual Studio, you cannot tell the environment to run multiple programs. So I'm very curious how that would work. Would there be another "mode" so to speak that would allow you to select multiple assemblies? How does synchronization work (or does it work?) between multiple systems. Another thing that always helps programmers is understanding the practical application. What situation would someone want to do this? It seems like one would always want to monitor each system individually. DrKoch (2/14/2007)
dplaisted (2/13/2007)
You have the full power of C# or Visual Basic, so you can code it pretty much any way you want.Let me jump in here. Of course "evrything is possible" with a modern programming language. On the other hand the whole reason of being for software packes like RE is simplification: I write some lines for my trading system and it runs in a specialized environment. In this context it becomes highly preferable if there can be several "simultaneous" systems - not specially coded into one source code, but simly some checkmarks before the names of existing, simple systems. This has always been one of the major weaknesses of WL and the top feature request for WL over years.
|
|
|
Posted 2/14/2007 08:28:16
|
|
|
|
There is a simple answer:
Look at this very practical situation:
I developed say three systems. Each works reasonably well in backtest. To "diversify" risk and invested money the obvious idea is to trade all three systems in parallel. Before I do so, I want to backtest this "combined approach".
In terms of an expected user interface it looks like this:
I have a "list" with my three systems, (and possibly some more). I set a checkmark on my three systems and assign some "position sizing hint". We use to say soemthing like 2.5% risk for system A, 1.5% Risk for system B and so on.
After I run this "Basket" I can see all the usual results (performance report, list of trades, equity curve,...) for the combined system.
On the implementation side you need two distinct phases in system execution: Trading Signals and Position sizing.
First you produce all trading signals for all systems. Then you run through all signals in date/time order and do the position sizing based on available equity and so on.
We implemented this for WL using external ASCII files for trade signals. It works pretty good but is a bit complicated to run because its a combination of several scripts... cumbersome...
www.finantic.de
Our Trading System at C2: Topaz
|
|
|
Posted 2/14/2007 08:37:19
|
|
|
|
| I'll digest the rest of the post, but one point I need clarified first, is this only applicable to backtesting or would you run this situation live as well?
|
|
|
Posted 2/14/2007 09:07:03
|
|
|
|
billb (2/14/2007)
....or would you run this situation live as well?
Guess what?
Well, serious, I am more in backtesting. The "real trading" runs on quite another class of system.
But I think "they" will expect this to work live as well...
After all, given this event driven design it should not be too hard to make this work in RE.
www.finantic.de
Our Trading System at C2: Topaz
|
|
|
Posted 2/15/2007 18:57:36
|
|
|
|
| I noted the need for multiple systems in my posts before, but I figured this can be added later on. The ability to test multiple systems with different weight assignments is something nice to have and some of the neural-networked based systems like NeuroShell provide for that. However, here I wanted to add some more notes to the posts on this tread. For multiple systems to run in real-time, there are more issues involved. Do the systems “talk” to each other synchronously, asynchronously, or both? Are there real-time requirements on the “cross-talk” (and this does not mean just being fast as the term real-time is loosely used, but specific time requirements). These will involve memory model access and design as well. Will there be multiple systems within an execution and/or outside that need to “talk”. For cross application talk, there are many middle-wares that can be used, but an API probably needs to be added to RE that can be extended to implement the interfaces.
|
|
|
Posted 2/16/2007 05:55:24
|
|
|
|
| I would have resource pools (funds and constraints), possibly shared (point of intersection), and multiple concurrent strategies. Bar frequency becomes a conundrum: is it an attribute of a strategy, of a symbol or of a symbol within a strategy? It would seem collecting from a provider would have to be done at the lowess common denominator. I contemplate three distinct stores per symbol: intraday, daily and refreshed from the internet on reference, like charts (this avoids all long term store maintenance questions at the local level). - Pete
-Pete
(See also rightedge-ats Yahoo group  )
|
|
|
|