|
| Autor |
Nachricht |
SwingManT
Anmeldedatum: 17.08.2005
Beiträge: 1700
Wohnort: Frankfurt am Main
|
FOREX Predicting Price Movement (Devisenpaare)
Verfasst am: 30.01.2006, 16:33 |
|
|
Link:
http://www.strategybuilderfx.com/forums/showthread.php?t=11530
Download-Link:
http://www.strategybuilderfx.com/forums/attachment.php?attachmentid=7624
when to trade
http://www.strategybuilderfx.com/forums/attachment.php?attachmentid=7670
Andere reports des Autors
http://www.fxengines.com/reports/
Zitat:
It seems like some additional explanation is needed, so here are a few points:
- Our goal here was not to create a trading system. It was to study data in situations that frequently occur during trades and try to understand what is most likely to happen in those situations.
- The methodology is contained in the report, but it is somewhat technical. Basically we took a closing price for an interval and looked at it over the following 24 periods. We were looking for breakouts, since most successful trades (even those which do not use breakouts as a signal) enjoy some form of price breakout. We decided to look for breakouts of 10, 15, 20, 25, and 50 pips. We did not care if it was a long or short trade. We watched the price and if it hit any of the trigger values above (or some, all of them), we called that a trade. From that point on, we studied the trade to see how far it would go within the 24 periods. We looked for target pips of 20, 30, 40, 50, 60, 70, 80, 90, and 100+ pips. If the target was hit before the stop of -20, the probability of hitting that target for THAT TRADE ONLY was 100%. If it did not hit the target, the probability for that trade was 0%. To obtain the probabilities shown we combined all of the observations.
- These results will be difficult to achieve in live testing, especially without automation. There are many reasons for this: no spread is included, there is a limited data set, the data is checked every interval, not every tic, etc., etc. The fact that it is difficult to achieve these exact results does not make the data any less valuable.
- The results do not specify anything about what the net performance of a system using this approach would be. For example, the stop threshold is -20 but the actual point at which the stop is hit could be more than that. You can see the difference in the 5 min vs. other intervals.
- The intervals serve only one purpose: the help us to measure a total span of time. We had to use some metric, and intervals are familiar to most traders. All they do is give us different total time spans for the 24 periods. For example, if we look at the 5 minute interval 24 periods equals 120 minutes. 24 periods on the 30 minute chart is 12 hours. By looking at the different intervals we were able to observe what happened to price over varying time spans, which is valuable.
- This report was created using advanced modeling techniques. If you have a background in econometric analysis you would probably be able to duplicate it, although our data would be different from your data so the results would be different.
- You can use the results of this report in three main ways, though there may be others. First, you can use breakout signals for market entry. Second, you can hone your stops to take advantage of expected price moves. Lastly, you can create systems that use limit exits to capture profits along the lines shown in the report. We cannot vouch for any of these specifically, but they all seem to be reasonable ideas for utilizing this data.
We have a bunch of things coming out that will speak to specific trading systems, but we have always wanted to do this report just to satisfy a curiosity. The data is good, and is valuable. It's not a system in itself, so anyone looking for that will be disappointed. But really for anyone who trades we believe there is an inherent value here, because it's information. Forex is a market that seems to thrive in a vacuum, and when we can we will share tidbits of data that fill that up a bit more.
We also meant for the report to be somewhat thin on opinion, to let the report's data speak for itself. I have said about all I can about how we did it above, but I would be happy to answer any questions.
- Scott
Zuletzt bearbeitet von SwingManT am 30.01.2006, 16:42, insgesamt einmal bearbeitet |
|
| |
|
 |
SwingManT
Anmeldedatum: 17.08.2005
Beiträge: 1700
Wohnort: Frankfurt am Main
|
Verfasst am: 30.01.2006, 16:46 |
|
|
Zitat 2:
Well this is an interesting turn of events... sure, I'll play ball.
Before I respond, I would like to point out to anyone reading this that I have answered every question asked of me honestly, directly and to the best of my ability, and will continue to do so. I'm not sure why anyone would think I'd be shy about chiming in. I made this report available without asking for a thing, and despite my many warnings that this was an advanced report and that it was not meant to drive explicit trading strategies, I have received these questions which obviously are directed without any knowledge of those warnings. If you don't have interest in the report, don't want to make the effort to dig into it, or don't agree with it, fine - pitch it.
Now, on to the actual report and its veracity. Let me first reply to a few specific points:
- The data is not "optimized" or "modeled" in any way. The data was supplied to us by a major FX dealer, and is the actual historical tick database of their prior rates. The only thing we did to the data was build it into intervals, from the tics.
- Regarding our "no representation..." warning: The data didn't originate with our own dealing activity, so we don't make any claims about it, other than the above. That warning is also in place because we didn't want the kind of brief look at the report that seems to have driven these challenges to be the basis for anyone's trading activity. Again, this is not a trading system - we never claimed it was a money machine. Rather, we said, and maintain, that the results of this report help you to have a better sense of what to expect in the market. I will give an example of that momentarily.
Alderlug - You say: "Either probabilities are accurate or bogus. Period. It's really that simple." I would like to see the results of your analysis which refute our own. Please share with us your data source, methodology, and qualifications for building an advanced statistical model of this kind. I'd like to know more about the specific parts of this report you challenge, and I would like the opportunity to refute each, one at a time.
Garyh - I would like to see a summary of the MT expert you created (as long as you acknowledge the shortcomings of MT that may render those exerts invalid) and the actual test results of the trades you claim to show that the report's data are inaccurate.
Amarnath - Thanks for pointing out that it is good information (it is) that's not well suited for newbies (it's not). As for the sharing part, this little inquisition makes me wonder what possessed me.
Coastie - Thanks for being one who has clearly *read* the report. Many have read the report and take it as we intended. Many have not, or may not understand it, and have misinterpreted it badly.
For my own response to these comments, it's tough to say more than the lengthy post I made above, and the comments that preceded the data in the report itself. The idea for this report was pretty simple. Every successful trade over 10 pips has some kind of price breakout. So we thought that no matter what type of entry a trader used, it might be useful to have some historical sense of what prices have done in similar situations. Of course we could have done a report that says: "when a 30 minute MACD cross occurs, what happens to price?" That would have meant that either we did just a couple of those specific analyses, or that we did far more than we could have practically done. So we chose the breakout idea as the trigger to enter these "trades".
|
|
| |
|
|
SwingManT
Anmeldedatum: 17.08.2005
Beiträge: 1700
Wohnort: Frankfurt am Main
|
Verfasst am: 30.01.2006, 16:49 |
|
|
Zitat 3:
So, on to the trades themselves. The way the analysis works is this: We monitored price for the particular interval until it had moved a certain number of pips in either direction - "the trigger". Let's say the trigger was 10, which is the first group of results. We said: "When the price moves 10 in either direction, mark it, and watch if for the next 24 periods." Of course 24 5-minute periods is much shorter than 24 4-hour periods, so this really was just a way of watching the price longer. Once we marked and watched the next 24 periods, we looked for one of three events: (1) the target pips of 10, 20, 30, etc. (2) the target stop of 20 was hit OR EXCEEDED, or (3) the 24 periods expired with neither a price target nor a stop being hit. If the target value was hit, we recorded that as "true". If not, it was false. In the end, we summed up all of these values and calculated the probabilities of the targets being hit that are shown in the report. Keep in mind that the stop could have been exceeded. We only checked to see if 20 was breached or not. It could have been -21 or -210. We didn't record that, because we didn't build the report to trade, as least not explicitly.
Now if you tried to take these numbers and trade them, all kinds of things could happen. Your success % could be nearly identical to what we show in the report, but the system (we call them engines) could be either a loser, a modest gainer, or a big gainer. You can have a success rate of 90%, but if your losses exceed the gains, it's a losing system. Here's a good example of that from my own tests:
(this is a back test summary from FX Engines)
"This engine had a net performance of -432 pips with an 84% success rate. Had all trades been exited at the high, the engine would have been worth -99 pips. However, an average of 1.99 pips per trade was "given back" from the high to the point where you exited. The maximum drawdown for this engine was 390 pips, with 2 losing trades during the worst losing streak and 23 winners during the best winning streak."
Interestingly, the entry signal for this engine was a breakout of 25 pips above the high for the last 24 periods on the 2 hour chart. The limit exit on this engine test was 30 pips. If you look at the report for GBPUSD (that was the pair), you will see we quote a probability of 96.3% for a target of 30 pips when using a 25 pip trigger on the 2 hour chart. Of course, this engine's test had an 84% success rate and lost money. The difference between the 96% and 84% is attributable to many factors, the most important of which are that the data I ran this test on is more recent than the data from the report, and the method of entry is different from the method of entry used in the report. But it's pretty close.
Here are two more examples:
Example 1: Entry signal is a 60 minute breakout of 15 over the high of the last period, limit exit of 30. The probability in the report for GBP 60 MIN 15 trigger is 85.7%. This test yielded 84%, and made money.
"This engine had a net performance of 1298 pips with an 84% success rate. Had all trades been exited at the high, the engine would have been worth 2521 pips. However, an average of 1.97 pips per trade was "given back" from the high to the point where you exited. The maximum drawdown for this engine was 454 pips, with 3 losing trades during the worst losing streak and 46 winners during the best winning streak."
EXAMPLE 2: Entry signal is a 60 minute breakout of 25 over the high of the last period, limit exit of 30. The probability in the report for GBP 60 MIN 25 trigger is 95.2%. This test yielded 84%, and made money.
"This engine had a net performance of 941 pips with an 84% success rate. Had all trades been exited at the high, the engine would have been worth 1832 pips. However, an average of 2.02 pips per trade was "given back" from the high to the point where you exited. The maximum drawdown for this engine was 452 pips, with 3 losing trades during the worst losing streak and 25 winners during the best winning streak."
All three examples given here are from 2003-2004 (24 months). They are not "well-chosen" examples - just tests I had done before. I don't even think I was thinking about the report when I made them. I am sure there are examples where the report's data are not as close, but there are good reasons for that too. The data and methodology - and the results - of the report, are nonetheless rock solid. I reiterate - the report is not about specific examples. It's about patterns and learning more about price action in forex markets.
|
|
| |
|
|
SwingManT
Anmeldedatum: 17.08.2005
Beiträge: 1700
Wohnort: Frankfurt am Main
|
Verfasst am: 30.01.2006, 16:49 |
|
|
Zitat 4:
Of course I am not going to provide an example for every data point in the report. If you have a specific test that does not agree with this report and you want me to respond to it, fire away, but please only do so with supporting info about the nature of your test, the data source, dates, etc. If you want to try shooting the report down I will gladly take all comers, but at least give me a chance to reply completely. It would also be nice if you gave me the benefit of the doubt, as my words and actions have given no one reason to question my integrity or honesty.
p.s. - Note that one of the examples yields 1298 pips over two years. Big deal? Well, it makes money 84 out of 100 times, and it's completely automated. With an engine like that I would gladly trade 10 standard contracts with $100K in capital. 1298 x 10 = 12,980 pips over two years. I will take that, and it seems to be a benefit of the report, if you are looking for one.
Zitat 5:
fxengines
Didn't mean to offend you. The information you have posted has been very helpful to me. Was just posting information that I tried to develop a system using the information you provided and it failed to produce usable results. I'm not doubting the integrity of the intent of your information, I'm just stating that the methods of creating that information may not have been compiled in a fashion that would allow it be accurately used in the way we think we can. In other words, we may be reading more into the information than you are presenting. Thank you for providing more details regarding your information. As far as my personal testing goes, I can't find the expert I ended up with (not unusual since I'm closely approaching a 1000 test files). I will continue to look for it and post it when I find it. (or maybe I'll just write another one)
Gary
Zitat 6:
Conceptcar - it seems you are missing a few key points here, and I am really, really reluctant to get into them into any detail here and extend this forum beyond where it is now. It seems each iteration causes more confusion!
There is only one place in the report where 95.6% is mentioned, at the DAY interval for a 10 pip trigger with a 20 pip target. That says: When the DAY chart moves 10 pips, it has a 95.6% chance of hitting +20 (moving ANOTHER 10 pips) before it meets or breaks the -20 stop. That sounds completely reasonable to me, and quite expected. I don't know where you get the 80-200 range you are talking about, and I can point out many examples on a day chart where price "runs away" from a particular point (i.e. a stop) and never really retraces. Additionally, it's a key conceptual point to remember that we are checking only every DAY in this example, not each tic.
If you, or anyone else, would like to discuss this further please contact me directly at fxengines@hotmail.com. If any of those discussions add to this forum I will copy them here.
Scott
Im letzten Beitag auf StrategyBuilderFX werden ein paar Unstimmigkeiten erwähnt. Mal abwarten was der Autor zu sagen hat.
Zuletzt bearbeitet von SwingManT am 30.01.2006, 16:55, insgesamt einmal bearbeitet |
|
| |
|
|
|
|
Registrieren
Login
