System / Setup Design process

L

liquidtox

Hi all,

Just thought I'd throw out a new topic for discussion. I realize a lot of people are looking for details of working setups, but personally I think the thought process in developing working systems is much more relevant. As we know most setups / patterns only work for so long and only for certain market conditions. So I believe it would be more useful to discuss how a trader goes about finding the workable setups...

Does one:

a) watch the markets for a couple of years and hope a pattern dawns on them ? and hope it still works for time to come?

b) use backtesting software and sim and sim until you come up with a optimized system that may work for a bit longer?

c) cruise the web sites / chat rooms and piece together styles from all over then watch them realtime for a while until you see it working...

Any insight would be most appreciated, whether it be into mechanical, discretionary or hybrid system design. Perhaps the more experienced traders on this board would be willing to shed some light onto this topic...
 
liquidtox,

I went looking for this basic topical area tonight, and found your thread title sounded closest to the question/discussion of interest to me. But, boy, do I feel like a minority! It's been over a year since anyone has commented here.

Anyhow, if you're still around, while I'm undoubtedly a novice compared to many in the ET community, I'm finding in my own systems design work that it helps to start by identifying a single basic belief I have about the market, and then looking for a way to express that in completely measurable terms.

My own beliefs about the market could generally be categorized as statistical and behavioral, so if I were to try and describe my research and modelling, I guess this would be about right: I generally look to exploit what appear to be overreactions and/or ill-timed commitments of other traders, often in the area of relative price extremes. This normally results in my first pondering pattern failures. I want to make sure there's a compelling story that can be told about the psychology of decision-making there. Then I look for ways to mathematically express the story so that I can measure it.

But, like you, I'm interested to hear how other systems designers would characterize their starting points. Thanks for getting the discussion started. Here's to hoping it continues.
 
Quote from NorskTrader:

liquidtox,

... I'm finding in my own systems design work that it helps to start by identifying a single basic belief I have about the market, and then looking for a way to express that in completely measurable terms.

...
More...

I think this is a great way to build your system. You need to start with one solid piece to build on.

What I am trying to do is find *MY* way to view the market. Most of the folks here on ET will say you don't need any indicators, just price, S/R, and trendlines. I am finding some indicators that are quite useful for viewing the market and anticipating what it is going to do. An indicator is just an alternate way to view the market, a different lens. Just because it is derived doesn't mean it is less useful than just the price action. What I am looking for is an indicator or two that distill the price action into something that gives me a good indication of the coming trend or lack of trend.

To tie this back to your point, my single basic belief about the market is that it moves in cycles. So what I am looking for is an indicator (or a couple) that helps me see these cycles and gets me in and out at the right points in the cycle. I think a key for finding good indicators is to work backwards. Find the trends/cycles for the day and see what the indicators were doing as they developed. It seems most people start with the indicator and work from there.

Sorry if this is hard to follow. I shouldn't post from work. :)
 
Biomech,

No need to apologize for your post. What you said makes sense to me.

While I'm not sure how many traders are hard-core anti-indicator types, I like what you've said about indicators providing another lens. For me the essential test of an indicator is its explanatory power over a wide array of conditions and over a large population of trade instances. If Indicator A, all by its lonesome, consistently segments a new population of trades into a preponderance of winners on one side and a preponderance of losers on the other, it gets my attention. Especially if it does so at roughly the same demarcation point(s). And if my own research is any guide, there are some indicators that do this again and again and again.

As for cycles, I also agree. For me the research question is then, "which recurrent market principle's cycle do I suspect is most relevant here?" We all probably have our favorites there, too.

Anyhow, thanks for helping revive this conversation.
 
This picture is a random walk process image extracted from famous Berstein's book.

<IMG SRC="http://www.elitetrader.com/vb/attachment.php?s=&postid=207812">

You can calculate indicators as you like even with random walk process. For example a moving average and you will still have the illusion that it give you signal although anybody knows that it cannot be on a random process it is by construction of indicator itself that you have an artificially correlated serie. So the problem is how do you distinguish random from not random before talking of any indicator. If not so you can just do pure Money Management.


Quote from NorskTrader:

Biomech,

No need to apologize for your post. What you said makes sense to me.

While I'm not sure how many traders are hard-core anti-indicator types, I like what you've said about indicators providing another lens. For me the essential test of an indicator is its explanatory power over a wide array of conditions and over a large population of trade instances. If Indicator A, all by its lonesome, consistently segments a new population of trades into a preponderance of winners on one side and a preponderance of losers on the other, it gets my attention. Especially if it does so at roughly the same demarcation point(s). And if my own research is any guide, there are some indicators that do this again and again and again.

As for cycles, I also agree. For me the research question is then, "which recurrent market principle's cycle do I suspect is most relevant here?" We all probably have our favorites there, too.

Anyhow, thanks for helping revive this conversation.
More...
 

Attachments

harrytrader,

Your point about the danger of false inferences from spurious correlations -- including merely random patterns that give the illusion of systematic correlation -- is an excellent one. The danger is real. In fact, in a recent paper I presented, I began on page 1 with just such a "reality check," showing one such "correlation" that was the product of nothing more than the random number generator in my Excel spreadsheet.

That's why I worded my last post as I did. As a researcher, while I'm sufficiently acquainted with the problems inherent in Karl Popper's hypothesis falsification principle, my bias is to begin as a skeptic. But when I see a certain indicator repeatedly tell essentially the same story, and to have roughly the same segmenting power in context after context, my confidence in it rises.

Here's a worthwhile exercise, which is designed to acquaint the observer with pictures of random patterns of correlation. And, of course, it is also designed to give one some basis for recognizing when something may _not_ be so random.

Take a random number generator and spit out a bunch of "trade results" that roughly mimic the range of results you might see in your favorite trading setup. Freeze the values. Then run the random number generator again and create an ordinal series with which you can sort them. Go ahead and sort them, and plot the "trade results" in that sequence. Keep the random number generator active, and have it keep generating new sequences, which you then again sort.

What you'll see is many cases in which portions of the sequence on the graph seem to be laden with winners, and others with losers. But you'll also see many cases in which it looks...well...random.

The trick is to then compare these displays with displays of the same sort built from correlations between trade performance and your favorite indicator applied in context after context.

My hunch is that if you've got a good indicator, you'll see a pattern that is too consistent across many trade environments to leave you comfortable with the null hypothesis of randomness there, too.

Which brings me to traderkay. I'd be happy to, but I don't want to spoil your own joy of discovery.

:)
 
Quote from NorskTrader:

harrytrader,

Your point about the danger of false inferences from spurious correlations -- including merely random patterns that give the illusion of systematic correlation -- is an excellent one. The danger is real. In fact, in a recent paper I presented, I began on page 1 with just such a "reality check," showing one such "correlation" that was the product of nothing more than the random number generator in my Excel spreadsheet.

That's why I worded my last post as I did. As a researcher, while I'm sufficiently acquainted with the problems inherent in Karl Popper's hypothesis falsification principle, my bias is to begin as a skeptic. But when I see a certain indicator repeatedly tell essentially the same story, and to have roughly the same segmenting power in context after context, my confidence in it rises.

Here's a worthwhile exercise, which is designed to acquaint the observer with pictures of random patterns of correlation. And, of course, it is also designed to give one some basis for recognizing when something may _not_ be so random.

Take a random number generator and spit out a bunch of "trade results" that roughly mimic the range of results you might see in your favorite trading setup. Freeze the values. Then run the random number generator again and create an ordinal series with which you can sort them. Go ahead and sort them, and plot the "trade results" in that sequence. Keep the random number generator active, and have it keep generating new sequences, which you then again sort.

What you'll see is many cases in which portions of the sequence on the graph seem to be laden with winners, and others with losers. But you'll also see many cases in which it looks...well...random.

The trick is to then compare these displays with displays of the same sort built from correlations between trade performance and your favorite indicator applied in context after context.

My hunch is that if you've got a good indicator, you'll see a pattern that is too consistent across many trade environments to leave you comfortable with the null hypothesis of randomness there, too.

Which brings me to traderkay. I'd be happy to, but I don't want to spoil your own joy of discovery.

:)
More...

That's funny, acrary was just discussing a very similar method for testing the validity of system design... and how he goes about finding an edge

you can find it here:
http://elitetrader.com/vb/showthread.php?s=&threadid=14001&perpage=6&pagenumber=7
 
You must log in or register to reply here.
Back
Top