Hi everyone Just getting started in systems writing. I would appreciate any comments from experienced systems developers. I am working on EOD data. 1. I am finding that I _can_ develop potentially tradable non-discretionary systems; the problem is, the good ones tend to go off very infrequently. I am wondering - do those who trade these systems exclusively have a large number of different systems that each go off rarely? Of course the type of system you are shooting for matters - I am speaking here of what should probably be called 'swing' trading systems (although in my development work I have found that a swing trade is simply one which you are not stopped out of!). Part of my problem may be that I have this idea in my head that I should be able to devise a system that wins a large percentage of the time. I therefore work each system to the point that it produces a higher percentage of wins but triggers infrequently. 2. More importantly - I have found that the best stuff I have would be hard to trade, psychologically speaking - that is, it is very hard to devise a 'swing' type system that reliably goes off in every dip during a nice clean uptrend; it is easier to develop a system that goes off 1/2 way through up legs. What kind of an answer am I looking for here? Not a pat on the back - just want to hear from others who have been down this road and know what I'm talking about. Are your good systems 'hard' to trade in that they give signals in counter-intuitive spots? (Hmmm... what's counter-intuitive to one person might be perfectly reasonable to another, right?) I am using Metastock right now but the system tester in v 8.0 is not great; I am considering WLD but I would have to learn Pascal for that... any comments about the best system development program for non-programmers? Some of the posts here are obviously written by individuals with heavy programming and high math chops. That's not me - I am hoping that a good system can be devised by someone like me who substitutes hard work for high concept.
If I have understood you correctly then..... 1) i would always opt for "trade less often but better" but be careful not to over optimise it. 2) by it's nature a system would have neither a psychological or intuitive element, a system can of course perform well in "testing" but be hard/impractical to trade from a "real world" point of view but I don't think that was your question. anyway just my 2p worth - good luck.
Answer to you first question. It all depends on you. What you feel comfortable with. When I first started developing systems...I developed systems that produced a high number of trades with a 60 to 70% edge. They made good money...but my personality couldn't take the beatings that the 30 to 40% losers gave me. So, I added some extra filters to give less trades and bump up the winning edge to something more comfortable. The end result? Less trades and less money. The hardest part about this side is the less trades. There's so much time between trades that you get ancy...feel like you're never going to get a trade again. Anyway, the way I solved that is develop more systems. It will take a while...but eventually you develop enough systems that give you plenty of trades. And you get a higher winning edge with each system. Ahhhh, there's the problem. When you combine all those high winning edge systems together and trade them as one...your total system winning edge drops down and you are almost right back where you started. For me, the difference now is that I'm more comfortable trading those systems that I've developed...I can weather those losses cause I know the winners will come. But, it is definitely an evolving process. Developing a system is easy. Trading it will take years before you feel truly comfortable with it. One more thing on developing your systems...when you begin tweaking it to reduce the number of trades and boost your winning edge...do not forget that you still have to produce enough trades to prove your system is valid. I ran into that one as well. I develop on WLD...and it works great for me. The best part about this software is the scanning. You can develop your systems, add them to the MultiScan utility, and hit go and it does all the work running your systems and triggering your trades. Great if you have numerous systems to run each night. And yes, many times my systems produce trades that are counterintuitive. But, that's why I develop and trade systems...cause my trading intuition sucks. Good luck!
I agree with almost everything you said and for the most part this has been my experience as well. One thing about WLD that IMHO kills it as a serious system testing platform is that it will not generate orders in the "middle" of a bar. This is a gross oversite for the high-frequency-data, high-frequency-number-of-trades trader. Also, I believe it has some strange rules in a system about looking at the Bid/Ask. Otherwise, the program is terrific. nitro
>One thing about WLD that IMHO kills it as a serious >system testing platform is that it will not generate >orders in the "middle" of a bar. I am a WLD user and perhaps I'm misunderstanding you Nitro when you use the term "generate orders". If you are saying that during a testing run orders can only be placed at the OHLC prices of a bar then I will have to disagree with you. Orders can be placed anywhere within confines of HL. I'm sure I'm just misunderstanding you so I'll let you elaborate if you will. Thanks JB
JB, Yes, you can put the order in, _but_ I believe (unless my prowess with WLD is not up to par) that if the bar when it is completely formed does not actually generate the order, then WLD internally will be flat, even though intrabar you would have taken a position. The key is realtime signals for _limit_ and _stop_ orders. For market orders or EOD systems, this is not an issue, but for realtime, tick by tick traders, it is worthless. It is probably _the_ #1 reason that Tradestation people will not move to WLD. nitro
Hey Nitro, thanks for the response. I now think the confusion between us has nothing to do with any misunderstanding you have of WL, but perhaps rather is caused by your use of the term "testing". You said: (emphasis mine) >One thing about WLD that IMHO kills it as a serious >system testing platform is that it will not generate >orders in the "middle" of a bar. I think from your second post you are actually referring to a (current) limitation in WL use as a real time trading platform rather than back testing. In back testing bar data one can't dig any deeper than the bar resolution allows so you can only deal with a bar at it's end. One can however make careful assumptions within that bar after it's end and "generate orders" within the bar. For live trading however with the proper data feed trade one can (and often must) make decisions from the data stream mid candle. I'm now pretty sure this is what you are referring to and yes, this has been a WL weakness. >It is probably _the_ #1 reason that Tradestation >people will not move to WLD. I hear (but as I'm not associated with WL in any way please do not put any faith in my statement) that in just a few weeks the next build of WL will allow mid-bar trades during live trading. Perhaps someone from WL could confirm or deny this. JB
MarketDog, Excellent post. Could you elaborate on: "Ahhhh, there's the problem. When you combine all those high winning edge systems together and trade them as one...your total system winning edge drops down and you are almost right back where you started." Is there a statistical principle involved here? If you're running five systems with a win ratio of 70%, then their combined stats would be 70%. Same for any other ratios. What am I missing? Thanks Wayne
JB, Yes, you read me correct. I think I misused the term testing in this case, because to me, I almost always test forward on realtime data, not back However, even in the case of backtesting on say 30 second bars, I believe the problem is the same? (not sure here.) In tradestation, you can tell it to use tick resolution so that even though it is _displaying_ 30 second bars, _internally_ it is using every tick to make it's calculations. This more accurately simulates what is happening in realtime when you go live. Tradestation used to have the "intrabar tick logic" that it used to use to guess what happened intrabar. That was also worthless. A "serious" system testing tool has to be able to deal with the granularity that the data is naturally coming at you from the feed/exchange. It has to be able to access _every_field_ available to the datafeed that you are using. Exciting news if that is true about WLD! :eek: nitro