OmniTrader Upgrade Forums
Trade Plans
Two "Exponential" Moving Average Crossover
113 replies, 12129 viewings
|
OmniTrader Upgrade Forums
Trade Plans
Two "Exponential" Moving Average Crossover
|
Last Activity 1/2/2025 4:50 PM 113 replies, 12129 viewings |
|
| Printer friendly version |
Jim Dean Sage   Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA  User Profile |
| ||
| SteveL Veteran   Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
Perhaps a minor bug? | ||
| SteveL Veteran Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
My opinion is that the a strategy that allows reversing signals is preferrable. Jim suggested a comparison between the Rev2in1 option vs. the two SB's and a Vote block change (see attachment aSJLstrDoubleDualEMA-2in1sysBx2.ots http://www.omnitrader.com/omnitrader/support/OT2004/forum/get-attachment.asp?attachmentid=2844) in post #21060). For the past 12 months, there are 2 differences in the Dow30 between these two strategies. In both cases, it happened when the stock was relatively directionless. In both strategies, you will see (below) that both long and short signals are occurring every few days. In the 2 SB version (2in1sysBx2), the signals show as reinforcing signals, whereas in the 1 SB version (2in1sysB-Rev) the signal changes. I don't recognize any benefit in one vs. the other. As you can see in the charts below, they give opposite signals in these two cases where there are differences. The 2 SB version (2in1sysBx2) turns out to produce a better result in these two cases (over the past 12 months). But in a 10 year PortSim, the 1 SB version (2in1sysB-Rev) produces slightly better results. [Edited by SteveL on 7/10/2010 11:31 PM]  Attached file : 2010-07-10 GE 2010-02-26 signal comparison 2in1sysBx2 vs 2in1sysB-Rev.png (38KB - 770 downloads) Attached file : 2010-07-10 WMT 2010-02-10 signal comparison 2in1sysBx2 vs 2in1sysB-Rev.png (38KB - 776 downloads) | ||
This accout has been deleted |
"Use reversing signal" is important for a relatively rapidly swinging market and it should be implemented only as a last resort in my opinion. The key is to create a system that is reliable to your spec without having OT do 'optimizing' and 'reversing signal' for you. We don't want variables because then our thinking will be variable... as in we are trying to compensate for that with 'Oh, if OT said this and we thought opposite by looking at the chart, then...." This is where we can get lost because of conflicting signals. This is okay when we manually enter our trades and make final decisions but if we implement OT to use Integrated Brokerage entry... then it may not be our liking. In all these years I have been doing technical analysis, the structure for system must be simple and close enough to what you need. I just never forage into strategy programming. When we see our chart, we know what to do instinctively, yet sometime we betray our knowledge and at time it is exhausting to screen all potential stocks. That is why I have decided... uh... 15 years later... that I must implement strategy programming to make my workload easier. | ||
| SteveL Veteran Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
Both the 2 SystemBlock strategy, and the 1 SystemBlock strategy Using Reversing Signals are very close to the same thing. In both strategies, reversing signals are getting through faster than in the original/starting strategy. In both of these modified strategies, the reversing signal is not inhibited by the earlier signal being active for N-Bar=20 and enables the possibility of recognizing more quickly a change in direction. I wouldn't characterize that choice as an optimization. (I think of optimization as the process of letting OT run through a set of parms looking for that set which produces the best result for the input. That's not what we're doing at the moment, although I did stray into that territory by comparing PortSim results. But in my defense, I did that to see whether the strategies were equivalent over a longer period than 12 months.) The DoubleDualEMA crossover system is still firing the signal. The question is whether OT should or should not allow a reversal in the stock to be recognized ASAP, or hidden for up to 20 days. In all cases, once a trade is entered, it is not exited until the N-Bar=10 exit triggers. So, all we are currently trying to do is develop a a good entry signal. But bottom line, if you prefer to stick with the more basic strategy as a starting point, without reversing signals, I'm fine with that as this is a learning exercise, and this started as your thread. :^) Steve | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
We are not yet really trying to improve the strategy, but rather are seeking to investigate alternative means of modeling the same thing. It's a discovery process. I think that using the simple model rev2in1 for now is the best for moving ahead. We have very usefully identified several means of using the tools - that's why I'm here. Ready to move on? | ||
This accout has been deleted |
| ||
| SteveL Veteran Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
| ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
| ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
First, I owe Steve a (belated) response to the strategy he posted a while back, that separated the processing into one System block that created only Long signals, and another SysB that did only Short signals. These are connected to a Vote block that takes simple votes of the results, with the same 5-day lookback that we've been using. Both of Steve's SysB's are otherwise identical ... they have the "2in1" approach (both slow and fast MV2 active in one block), and both use the Number of Bars = 20 cutoff, and NO reversal checkbox. There was one small discrepancy between that strat and our others ... it used a Market On Open exit. I've modified it to be MOC, and attached it here. The DJ-30/12-mo eval showed only ONE discrepancy between this strat and the Rev2in1, in GE at the end of Feb09. This period of time is CLEARLY a tight consolidation, where no one would be likely to enter a "real" trade based on these systems. A filter could be built to reject trades like this ... but that's a jackrabbit that we'll not chase. As Steve pointed out, using the "Double" approach (two sep sys blocks) offers more versatility for optimization (either automated or via Strat Wiz) ... and it's common "wisdom" that short and long trades act differently, so this is a great technique to remember ... WHEN one gets around to tweaking parameters (and we ain't quite there, yet :~) So, save this one for later. Thanks, Steve! [Edited by Jim Dean on 7/12/2010 7:51 AM] Attached file : aJDgyDblDualEMA-SvsL.ots (780KB - 441 downloads) | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
Reminder: the Confirm strategy put ONE Of the MV2-C systems in the System Block, and added a Confirm block after the Vote block. The Vote and Orders block were unchanged. The ConB had Confirmation Tolerance settings of 5 bars "before signal" and 0 bars "after signal", with Any number of signals acceptable. We never discussed why it was 5/0 vs 0/5 vs 5/5 vs 0/0 or something else. The Help manual seems a bit ambiguous here: Confirmation Tolerance—Specifies the number of bars that OmniTrader will look back in order to accept a signal to be used for confirmation purposes. This refers to the signals generated by the Confirmation block as well as any signals passed to the block from a Secondary Signal Line (see Primary and Secondary Signal Lines). The question is, when the input form says "Before|After Signal", is it referring to the CONFIRMATION-system signal (before or after it), or to the VOTED signal that "feeds" the ConfB? An experiment will answer that question. Create three strats - identical except for these two CB inputs. One has 5/5 input, another has 5/0 input, and the third has 0/5 input. None has Reversal ... all use N=20, like the original strat did. Now, find a case where the 5-0 and the 0-5 versions give DIFFERENT results ... the 5-5 should agree with one and not the other ... and our original DoubleDualEMA strat should (ideally) agree with one of them as well. CAREFULLY examine the votelines of the four cases, and see if you can figure out whether the "Signal" that the Before/After is refering to is the VOTED signal, or the Confirm-Block signal. Hint#1: not a trick - you can know for sure Hint#2: check out XOM on 3/18/10 (too easy, drill sergeant!) Hint#3: check out IBM on 5/21/10 (for the flip side of the coin) Hint#4: check out KO on 3/26/10 (for a "two-fer") To make sure we're on the same sheet of music, here are the four strategies that I'm speaking of ... [Edited by Jim Dean on 7/12/2010 9:49 AM] Attached file : aJDgyDoubleDualEMA.ots (744KB - 435 downloads) Attached file : aJDgyDualEMAconf-0-5.ots (772KB - 426 downloads) Attached file : aJDgyDualEMAconf-5-0.ots (772KB - 418 downloads) Attached file : aJDgyDualEMAconf-5-5.ots (772KB - 451 downloads) | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
And ... again ... if any of you "lurkers" out there want to jump in, feel free ... but ONE proviso - please READ THE PRIOR POSTS, all the way from the beginning, and absorb them, FIRST. I want to keep the "flow" going forward if possible. Thanks! Steve - please check your Private Mail inbox. Moving further, working on the Confirm Block ... using the "5/0" input per our original case .. You may have wondered, when I originally set up the Confirm Strategy, what rationale I used to split up the two variations of the MV2-C system. That is, the "fast" 5+20 version ended up in the System Block, and the "slow" 10+40 version ended up in the Confirm Block. Why not the other way around? Does it matter? Hmm - sounds like another experiment is in order!! Set up two strat's, both using N=20 (not reversal), and both using 5/0 for the Confirm Block input. One strat has 5+20 in SysB and 10+40 in ConB, and the other has 10+40 in SysB / 5+20 in ConB. I've made it easy for you ... the those strat's are (the first two) attached. Do the DJ30+12mo walkthru comparison of these, versus the venenerable DoubleDualEMA (avail for download in post below). Do you find any differences? Which of the ConfirmB strategies emulates the DDE strategy the best? Hint: INTC. Check out the expanded votelines and provide an explanation as to why one approach is better than another. THEN "for extra credit" ;~> ... do the SAME THING for those two variations of the Confirm strategy, USING REVERSAL logic ... compare these to the Rev2in1 strategy. I've attached those three strat's as well, to speed things up for you ... I presume that by now you know how to make the mod's yourself. [Edited by Jim Dean on 7/12/2010 12:48 PM] Attached file : aJDgyDualEMA-FSconfirm.ots (772KB - 427 downloads) Attached file : aJDgyDualEMA-SFconfirm.ots (772KB - 417 downloads) Attached file : aJDgyDualEMArev-2in1.ots (744KB - 391 downloads) Attached file : aJDgyDualEMArev-ConfFS.ots (772KB - 434 downloads) Attached file : aJDgyDualEMArev-ConfSF.ots (772KB - 454 downloads) | ||
This accout has been deleted |
The after signal of 5 is similar to setting Vote Box lookback of 10 days for the CB and SB to confirm each other. With the lookback of 5 bars it only allows the twos ystem 5 bars (days) to confirm each other to fire a strategy vote. With the confirm box, since there is no way to get the confirm system to confirm with the System Box for the Vote Box to fire a strategy vote signal. Therefore, the confirm box has both the SB and VB functionalties together to allow just as much flexibilities as separate system box and votebox. | ||
This accout has been deleted |
However, on other stocks, it went from some long and short signal into absolutely no signals at all for entire year. FS and SF strategy makes no difference in the strategy vote line. It doesn't matter if you set the fast system first or the slow. They both fired same signals. The 5-0 strategy is the same as DDE strategy. | ||
This accout has been deleted |
Now to current chart. there is a slight difference between the FS and SF voting signal... only on some but not all. The rev2in1 and FS strategy is same exact voting signals. | ||
This accout has been deleted |
As for the variation in strategy votes on FS and SF strategies. The SF is somehow a bit better. For medium term trades, having a long term signal in bull run is better before the short term system fires a long position. I am speculating that for short term traders, this may yield too few trades. Then again, the focus of this lesson is to teach us how to implement a strategy... learning what each function does... This is not the lesson to show all the secret holy grail for everlasting profit. | ||
| SteveL Veteran Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
I'm not quite sure how to express the result, but I'll give it a try. ConfirmBlock (5-0, 0-5, 5-5) experiment results demonstrate that the CB "expands" the SystemBlock results forward and backward. That is it opens up the SystemBlock signal window. Example: In our test, we have a VB 5 bar lookback. So, once the SB fires, the opportunity window is open for 5 bars from the initial signal for the next 5 bars. The CB 5-0 (5 back, 0 forward) adds 5 bars to the front of that window, and 0 to the end of that window. So, with CB 5-0, IF the CB fired 1 to 5 periods before the SB (or anytime during the VB 5 bar lookback), it would pass the voted signal through. Or alternatively, if we had a CB 0-5, it is similar (but not the same) as setting the VB to 10 bars, extending the SB signal window by an additional 5 bars. | ||
| SteveL Veteran Posts: 262 Joined: 8/19/2005 Location: Boulder, CO User Profile |
You asked "Which of the ConfirmB strategies emulates the DDE strategy the best?" The FS strategies emulate the rev-2in1 the best. Which is better? Hmmm. Let's just say that with my current view of the market, I would expect the FS strategy to be better because it fits my personal view of the world which is to look for a signal, and then wait for it to be confirmed, rather than have a "confirmed signal", waiting for the "signal" to arrive. That is to say, I prefer the horse before the cart, not the other way around. (But I'm willing to change my market-view given sufficient good evidence that the cart before the horse makes more sense.) So, bottom line, a "fast" signal confirmed by a "slower" confirmation block makes more sense to me than the other way around. | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
First, in my various "hints", I indicated which symbols showed DIFFERENCES between the strat's being compared. I trust that y'all saw those differences, and realize that the FS and SF strats DO produce different vote lines, as do the 5/0 and 0/5. Adding Reversals to the picture - same conclusion. For each of those variants, there ARE differences. Stan expressed the 5/0 thing in a useful way ... it's sort of like "extending the Vote Block" lookback. However, OT requires a Vote Block, so we cannot remove it :~) ... also, let me add a caution that the two are ONLY equivalent in CERTAIN (very simple) strategies ... you can have much more complex configurations for the VB or for the CB ... multiple "rows" can feed either or both of them, etc yadda yadda. So ... think of those two lookback inputs (VB and CB) separately. I'm not quite sure if all the goals of this "exercise" were met ... my purpose here, besides just showing some interesting stuff about Confirm Block options, was to try to "train" you to be slow and careful and very deliberate about how you check out stuff like this. I realize this is not "natural" for many/most people, but it's a valuable trait for trading. The better you understand your tools, the fewer mistakes you will make (known or unknown), and the MORE CONFIDENCE you will have in them. Psychology is a key ingredient to successful trading. So, I'm attaching a snapshot that I put together, to help illustrate what is going on when the CB inputs are changed from 5/0 to 0/5 to 5/5. It shows three symbols that had DIFFERENT responses to the inputs, and expands the votelines to show what's going on. I had asked: The question is, when the input form says "Before|After Signal", is it referring to the CONFIRMATION-system signal (before or after it), or to the VOTED signal that "feeds" the ConfB? Maybe the question, much like the Help manual, was unclear :~( However, I believe that it is very helpful to realize the "Signal" being referred to in those labels is the VOTED SIGNAL ... ie, "Bars After Voted Signal" or "Bars Before Voted Signal" would offer clearer understandings. But, maybe that's just me 8~> *I* think the graphics speak for themselves ... but if this raises some additonal questions, please ask them! That's why we're going through this process. Thanks for hanging in there! [Edited by Jim Dean on 7/13/2010 5:15 AM] Attached file : Confirm Before vs After Signal.png (192KB - 705 downloads) | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
There ARE differences. Let me emphasize that ... even though maybe in our limited DJ30/12mo comparison not many showed up, that's a very small sample. And, when the expanded votelines are engaged, we can use them to see what's going on behind the scenes. And that is the point of the exercise. We are NOT trying to figure out which gives "better" signals ... but rather WHY they give DIFFERENT signals. This question of "WHY" is so VERY important to learn to use!! I teach a lot of people about a wide variety of things, and I've noticed HOW HARD it is to get people to ASK questions, these days. Let me encourage you to cultivate the art of asking questions. It requires you to examine things closely, and actually think about them. It encourages intelligent creativity. It helps a trader continually improve their track record. OK ... end of sermon (for a while ;~) The first snap shows the (single) example from the DJ30/12mo search that showed FS different than SF, for non-reversing systems. Look it over carefully ... make sure you understand WHY, from the detail on the voteline, FS fired but SF did not fire. The three following snaps show examples that use the Reversing flag ... this produces more signals, so more differences show up. Note the comments in red on the right side of the chart, in each case. [Edited by Jim Dean on 7/13/2010 6:10 AM] Attached file : Confirm FastSlow vs SlowFast.png (242KB - 647 downloads) | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
First, the Reversing option (or presumably the N=1 option we spoke about a long while back) simply provides MORE signals ... and the N=20 "lockout effect" might be useful for keeping our charts & examples "cleaner" (see the first chart vs the other three, in the last post). I suggest we move ahead with the investigations using N=20 at first, but adding Reversing to check for additional patterns and to confirm our initial conclusions. My leaning would be to use the Rev or N=1 approach for our "final" strat template ... since there is no logical rationale (that I can think of) for creating an arbitrary N=20 "signal lockout" zone. Second, the 5/0 vs 0/5 seems to be obvious. Our tests were run with the FS combination, and as such, the 5/0 clearly was a better "match" to the DoubleDualEMA strat that is our "control" for the testing. Let's not forget that, long ago and far away in the original post that started this thread, the request was for a strategy that used two instances of MV2-C in conjunction with one another. 5/0 (with FS) matches that. If you were curious (I was), you might have tried out 0/5 with SF, to see how it performed vs the DDE. We did not show the charts ... but I skipped over it since it was a poorer fit. Following from that ... Third, we saw that FS matched DDE, but SF did not. So, FS (5x20 in Sys, 10x40 in Conf) is the way to go. More importantly, it MAKES SENSE! The Fast pair will fire FIRST ... and our Vote Block "extends" that firing impact a ways further ... and our Confirm Block "reaches back" to find it. Again ... on some tests I did separately ... reversing the order, and changing the CB settings to 0/5 (or even 0/10 with VB=0) did not perform acceptably). So, hopefully these conclusions MAKE SENSE to you now, in relation to HOW the strategy-logic is actually functioning. I hope that these "lessons" will give you a better "gut feel" for the purpose and mechanisms of these blocks we're working with. | ||
| Jim Dean Sage Posts: 3022 Joined: 9/21/2006 Location: L'ville, GA User Profile |
We will construct another variant of the DDE now, using a Filter Block INSTEAD of a Confirm Block. We'll put one of the MV2 pairs into the FB and leave the other in the SB, and see how that works. It WILL be different! Strictly speaking, by doing this, we are moving outside the bounds of the original post. However, it is useful to check it out ... oft-times our original ideas need to be massaged a bit, even at their foundational levels ... or, at least, we should investigate alternatives to see if the foundation is actually a sound one. It's worthwhile to read the Help sections about Filter (and Confirm) blocks. The primary distinction between them is that Confirm focuses on a POINT-EVENT in time ... the moment that the confirm-system fires ... but the Filter looks at a STATE/CONDITION that might last for a long time. Since the original request was for two systems that fire at POINTS in time, I did not discuss the Filter Block. However, it's worth a look ... maybe the CONCEPT that originally was requested could be better handled with a Filter plus one System ... OR, maybe the DoubleDualEMA pair would be improved by a Filter, somehow. We'll wait on that second idea till later, though :~) The "state" that the Filter will be testing is of course the relationship of the two EMA's to one another. If the faster of the two EMA's is on top, it's bullish ... and bearish if it's on the bottom. This is commonly referred to as "STACKING" ... what order are the EMA's "stacked up" on the chart. The Guppy strategies have this (as well as crossovers and other things) at their core, fwiw. SO ... you should be able to do this in your sleep now ... let's once again use the DDE as our basis, and create TWO new strat's that use FB's ... the first strat will have the 5 & 20 (fast) EMA's in the Filter, and the second will have the 10 & 40 (slow) EMA's in the Filter. But, this time, BEFORE we do the experiments, let's stop and THINK about which approach will "likely" work better ... but let's not get "married" to our presumptions (KISS "methods" without experimentation to back them up are often a KISS ... of DEATH ;~) The filter is checking a "state" based on the stacking of two EMA's. AS LONG AS that state is true, then any signals that hit the filter are allowed to pass through it. Keep in mind that the signals that come INTO the filter are NOT a "state" ... they are points in time. Yes, the Vote Block lookback can be used to "stretch" that point out (our 5-bar setting is reasonable for those param's, IMO) ... but if the filter state is OFTEN DELAYED in its appearance till AFTER the system-signal fires, then the signal gets lost. Maybe we want that ... maybe not. Let's assume, without being sure, that we DON'T want to "lose" any signals from the System block. That would mean that it would be better for our Filter "state" to "open up" BEFORE the SysB signal occurs. So, that means the "fast" 5+20 EMA pair should be in the Filter, and the "slow" 10+40 EMA pair should be in the System. Our "prediction" on the basis of this reasoning is that the strat set up with a "fast" filter will be closer to the DDE strat than one which has a filter using the slow 10+40 EMA's. This prediction, btw, is often known as a "straw man". We set it up to scare away the scavengers ... but also to use as target practice. Sometimes, it falls apart upon examination. Other times, we actually decide to dress it up with some coverals and a hat and maybe even a corncob pipe. Here are the two strat's for you to compare ... activate them with the DDE and find out which is closest in how the Vote line appears, for the DJ30/12mo test-bed. Hint: this is a no-brainer. Enjoy it while you can! To help you along ... please don't shortcut taking some time to review how a Filter Block is set up. The snapshot highlights some things that many folks overlook ... how to Rename, and most especially how to set up adjustable Parameters. Please ask any questions that come to mind! [Edited by Jim Dean on 7/13/2010 7:39 AM] Attached file : aJDgyDualEMA-FiltFast.ots (744KB - 461 downloads) Attached file : aJDgyDualEMA-FiltSlow.ots (744KB - 404 downloads) Attached file : Filter Block in Strategy.png (93KB - 668 downloads) | ||
This accout has been deleted |
1. System in the System Box, the First priority that has higher ranking than the system in Confirmation Box. 2. The system used in Confirmation Box is now the 'confirming' system that would only fire appropriate signal based on the signal from System Box. That is a big difference. Whereas two in one or double system box before the voting box, they have "equal" power. Therefore, in your term "not implied" of priority. | ||
This accout has been deleted |
IBM: No signal were generated because 0-5 specification eliminated any 'before' signal confirmation. KO: In this case the 5 "after" signal specification fired off a confirmation signal. The 'reinforced" signal on conrimation system is also counted for the confirmation voting process using the "before" signal specification. | ||
This accout has been deleted |
Back to the point... Jim, how come you set the Lookback in the VB to 0? I was trying to understand why the signal won't fire when they concur with each other. Now it makes sense to me about the way the filter block woeks. The lookback period is still required to get a long/short strategy vote signal, unless it is perfectly lined up with each other. I also noticed for the filterFast, you set the FB to 5/20 and SB to 10/40. For the filterSlow, you reversed it rather than keeping the SB the same in both cases. This also makes it a bit harder to correlate the fast and slow. I think I blew a few brain cells trying to decipher your lessons... but all is good because those are 'blockade" brain cells. LOL. |
|
|
|
Administrator
Forum Moderator |
Registered User
Unregistered User |
 E-Mail this thread to a friend |
Toggle e-mail notification |
