hmmm, if im not mistaken this software creates code for metastock or tradestation and then you use it there...
Thanks Ron will experiment with the AND method and see if I can get better results.
Alex, thanks for the feedback. Do you look for a certain number of signals on the same bar such as number >= 3 for example, or wait a few bars for a signal in the same direction?
The general idea is based on using a number of patterns to indicate market direction only and then some other patterns to establish positions. This always increases profit potential. When using AND the first condition can can be shifted a few bars backwards to relate to a past signal while the second condition refers to a current signal.
I think you should now have a general idea of how this might work. You must also try to understand why your method is very simple and does not take advantage of the tremendous power of APS to find patterns.
I'm currently investigating the possibility of using a genetic algorithm with the operators AND, OR to automatically design systems for me using as input the APS patterns.
Ron
futz_trader
Newbie
- Messages
- 2
- Likes
- 0
Ron,
Regarding "AND"ing patterns, are you essentially talking about "Split" patterns (from Harris' books) or something more general that allows bar spacing between patterns?
I'll understand if you don't want to get more specific but from reading this thread a few times both you and Alex seem to be experienced trading pattern systems. Comments from you guys are appreciated.
Thanks,
futz
I guess "split patterns" are just one kind that could involve ANDing. I think Ron was very generous indeed to even mention ANDing and I wouldn't expect him doing so, neither I would ask him going into more detail about his methods.
Here's is an example: you have a tool that calculates moving averages of all kinds. But there a zillions ways of using MAs. Some are too obvious and naive but others are very sophisticated and often successful.
In a similar way, APS can find many profitable patterns for you. Using the patterns in a profitable way depends on the specific strategy that employs them.
I guess confusion arises from the fact that each pattern code is generated by APS in the form of a complete strategy. I take that to be useful only for the purpose of backtesting the pattern codes in another platform rather than using them to trade as is.
Alex
futz_trader
Newbie
- Messages
- 2
- Likes
- 0
Alex,
Thanks for the response and your comments on specific/complete strategy makes sense.
After re-reading the thread, there are a few things I'll add to my list of pattern-related stuff to look in to. I'm just getting started with mechanical systems development. As a frustrated discretionary trader, any tidbits are useful to a beginner like me and in that regard this thread has been useful.
Thanks,
futz
Good luck to you futz. I recommend reading the online articles by Michael Harris, especially the one about the Profitability Rule:
Tradingpatterns.com Michael Harris article
Ron
I am thinking of buying APS but I would like to know whether I need to also buy another program to go with it. That is, can it be used as just a stand alone program? I do not have any trading programs at all. At present, I do not trade. If I do need another program to effectively use APS, can I get a suitable free one? I have very limited financial resources. So if a relatively simple but free program can do the job I don't mind if it does not have all of "the bells and whistles" of the commercial one.
My second question concerns data. If I purchase APS would I be able to get free data from somewhere (maybe a broker, for example) to go with it? I am thinking of just focusing on trading FX, using daily data for a few currency pairs. As I understand that FX is traded around the clock (i.e. 24 hrs/day), is it true that different data suppliers make different (arbitrary?) open and close prices for the same data stream? That is, if my thinking is correct, the starting time of each 24-hour cycle may vary from data supplier to data supplier. Would that have any impact on the working of APS?
Once I do get a supply of FX historical data would it be feasible for me to manually update it every day? Does APS have that capability or would I have to go into a WordPad (for example) file to manually update the data there.
My second question concerns data. If I purchase APS would I be able to get free data from somewhere (maybe a broker, for example) to go with it? I am thinking of just focusing on trading FX, using daily data for a few currency pairs. As I understand that FX is traded around the clock (i.e. 24 hrs/day), is it true that different data suppliers make different (arbitrary?) open and close prices for the same data stream? That is, if my thinking is correct, the starting time of each 24-hour cycle may vary from data supplier to data supplier. Would that have any impact on the working of APS?
Once I do get a supply of FX historical data would it be feasible for me to manually update it every day? Does APS have that capability or would I have to go into a WordPad (for example) file to manually update the data there.
Hi Jack,
APS works like this: you put in data to the program, it search for patterns for just that stock and then it creates a code that can be used with tradestation. Therefore, to use this program you will need TradeStation or one of the other programs that it works with and a data supplier, i guess you could use Yahoo Finance as a data provider if you are not going to day trade.
///Victor
APS works like this: you put in data to the program, it search for patterns for just that stock and then it creates a code that can be used with tradestation. Therefore, to use this program you will need TradeStation or one of the other programs that it works with and a data supplier, i guess you could use Yahoo Finance as a data provider if you are not going to day trade.
///Victor
Hi Jack, you can use MetaTrader or NinjaTrader data for free. I use Metatrader data:
APS Automatic Pattern Search - Metatrader
I use the free data with APS and then I program the patterns in my platform's native code.
Most data vendors use the same time for the open and close.
You need to update manually only if you are using the scan option or system tracking option to generate your signals. In that case, you need to subscribe to an EOD data service that automatically updates your data files. I do that for my stock trading I use an EOD service that automatically updates over 1,000 stock data files in the format needed by the program to scan and track signals.
Alex
Hi Alex,
I also looked at ASP but havent been sure if it is right for me. I trade FX with metatrader, do you know if the patterns it produce works for the FX market too?
Victor--
Try to think in terms of high probability trading strategies using patterns instead of individual high profitability patterns. I had APS do many searches for high profitability FX patterns in 15, 30 and 60 minute bars. I got many patterns, maybe close to one hundred or more for a single currency pair in all three timeframes. The problem is identifying the ones with the highest probability of staying profitable into the future. You have to develop such identification process yourself. For that you must employ Statistical Analysis and I think I have said enough...
Ron
trader3cnd
Member
- Messages
- 73
- Likes
- 3
Victor--
Try to think in terms of high probability trading strategies using patterns instead of individual high profitability patterns. I had APS do many searches for high profitability FX patterns in 15, 30 and 60 minute bars. I got many patterns, maybe close to one hundred or more for a single currency pair in all three timeframes. The problem is identifying the ones with the highest probability of staying profitable into the future. You have to develop such identification process yourself. For that you must employ Statistical Analysis and I think I have said enough...
Ron
Does the program pay off then? Are you willing to share your strategies for a fee? Pls send pm.
I have been thinking that my confidence in APS could be increased if APS could be put to the test, using random data.
What I am getting at is this: run a large data set of random data through APS and see whether APS finds any tradeable patterns. I thought about this because of my past experience in playing roulette. Roulette players often imagine that there are patterns in the outcomes of a roulette wheel. Some say there is a “dealer’s signature”, other say that the roulette wheel is slightly imperfect or “biased” while others say that they follow statistical patterns. We, as humans, are always looking for patterns in everything. It seems to be “natural” for us to do that. What we need to be careful of is to not imagine that there are patterns where there are none. A random sequence can be very deceptive, in that regard.
It could be argued that, in the case of the price of a security, it is mostly random but that it does exhibit some predictability from time to time. Presumably, that predictability is what APS uncovers. I think that the only way for us to be strongly confident of APS is to conduct a series of experiments using random data. After all, one’s confidence is crucial when trading for real.
The random data to be used in an APS experiment could be generated by just scrambling some existing historical HLOC data. Each HLOC set could be retained, unchanged and each HLOC set could be assigned to a random place within a new series. If, for example, you had five years of historical data for a particular security, you could randomly assign each daily (or hourly) HLOC to another day (hour), to make a new, five-year data set. I do not know how the scrambling process could be achieved in order for the new data set to be truly random. However, I feel confident that one or more of the readers here would have the necessary skill to do that, provided that they feel that the effort would be worthwhile.
After a large random data set is created (the larger the better), APS could be used to test the data for “tradeable” patterns. Hopefully, APS would not find any patterns. That would be the ideal outcome - one which would greatly increase one’s confidence in APS. The experiment could be done several times with the same original data, but scrambling it again, again and again. If APS did find “good” tradeable patterns in the scrambled data, what would that mean? We could compare those “good” patterns with the ones that APS found in the original, unscrambled, historical data. I will not speculate any more about the probable results. But I think that the experiment is worth doing. What do you think?
Jack
What I am getting at is this: run a large data set of random data through APS and see whether APS finds any tradeable patterns. I thought about this because of my past experience in playing roulette. Roulette players often imagine that there are patterns in the outcomes of a roulette wheel. Some say there is a “dealer’s signature”, other say that the roulette wheel is slightly imperfect or “biased” while others say that they follow statistical patterns. We, as humans, are always looking for patterns in everything. It seems to be “natural” for us to do that. What we need to be careful of is to not imagine that there are patterns where there are none. A random sequence can be very deceptive, in that regard.
It could be argued that, in the case of the price of a security, it is mostly random but that it does exhibit some predictability from time to time. Presumably, that predictability is what APS uncovers. I think that the only way for us to be strongly confident of APS is to conduct a series of experiments using random data. After all, one’s confidence is crucial when trading for real.
The random data to be used in an APS experiment could be generated by just scrambling some existing historical HLOC data. Each HLOC set could be retained, unchanged and each HLOC set could be assigned to a random place within a new series. If, for example, you had five years of historical data for a particular security, you could randomly assign each daily (or hourly) HLOC to another day (hour), to make a new, five-year data set. I do not know how the scrambling process could be achieved in order for the new data set to be truly random. However, I feel confident that one or more of the readers here would have the necessary skill to do that, provided that they feel that the effort would be worthwhile.
After a large random data set is created (the larger the better), APS could be used to test the data for “tradeable” patterns. Hopefully, APS would not find any patterns. That would be the ideal outcome - one which would greatly increase one’s confidence in APS. The experiment could be done several times with the same original data, but scrambling it again, again and again. If APS did find “good” tradeable patterns in the scrambled data, what would that mean? We could compare those “good” patterns with the ones that APS found in the original, unscrambled, historical data. I will not speculate any more about the probable results. But I think that the experiment is worth doing. What do you think?
Jack
intradaybill
Well-known member
- Messages
- 358
- Likes
- 14
Could you explain why you think this is a good test and provide a bit of theoretical justification?
AFAIK, one can find all kinds of patterns in random data. Suppose APS finds patterns in random data. Do you conclude that:
(1) The market data is also random if APS finds patterns there.
(2) The patterns in market data will not stay profitable in the future
or if APS does not find patterns in random data do you conclude that:
(3) The market data is not random if patterns are found
(4) The patterns in market data are not random and will stay profitable
I think none of the above conclusions can be reached.
The only test that makes sense is actual trading and real P/L.
Bill
Bill, I will answer your points, one by one.
Could you explain why you think this is a good test and provide a bit of theoretical justification? If we were to run random data through APS I would hope that APS would not come up with any tradeable patterns. I think that this would be especially true for a very large sample. Previously, I used an illustration of roulette. The roulette wheel can be regarded as a random number generator. If that asumption is true, a roulette wheel is not predictable, not even using APS. I am not suggesting that the securities that we trade are wholly random. Perhaps they are mostly random but sometimes not random.
AFAIK, one can find all kinds of patterns in random data. Suppose APS finds patterns in random data. Do you conclude that: Bill, I do not understand your abbreviation "AFAIK". You will have to explain that to me.
(1) The market data is also random if APS finds patterns there. Not at all! The fact that APS finds patterns in any data (whether random or not) would not prove anything to me. Anyway, in your statement in the previous paragraph you said that "suppose that APS finds patterns in random data" and then you went on in this point to ask whether "the market data is also random if APS find patterns there". You already asked me to assume that the data is random. Therefore, whether APS found any patterns or not would be irrelevant to deciding whether the data is random. It already is random, according to you!
(2) The patterns in market data will not stay profitable in the future Once again, you have already asked me to assume that the data in question is random. Therefore, I would conclude that if APS did find some patterns in random data, the patterns would not last and therefore would not be profitable into the future. I would love APS to be able to find profitable patterns in the outcomes of roulette spins, for example. If so, I would take a (small) computer with me (with APS loaded in it) to the casino and clean up!
or if APS does not find patterns in random data do you conclude that:
(3) The market data is not random if patterns are found Once again, you question is nonsensical. You asked me to assume that the data is random and then you ask me whether I would think, if APS does not find some patterns in random data, that that would prove that the data is not random! Of course not.
(4) The patterns in market data are not random and will stay profitable I do not even understand this point of yours. You already stated that I should assume that the data is random so why would I conclude that the data is not random on the basis that APS did not find any patterns? You added "and will stay profitable". How will anything in this situation be profitable if APS did not find any patterns, as you stated above? To me, it would be excellent if APS does not find any patterns (or very few weak ones, at least) in random data. After all, random data by definition means that the outcomes are not predictable. They happen purely by chance. There is no guiding hand of god, fate, luck, fortune or karma to determine the outcome. As they say for roulette, "the wheel does not have a memory".
I think none of the above conclusions can be reached.I agree with you!
The only test that makes sense is actual trading and real P/L. I do not agree with this point, for the reason already stated above.
Bill
Could you explain why you think this is a good test and provide a bit of theoretical justification? If we were to run random data through APS I would hope that APS would not come up with any tradeable patterns. I think that this would be especially true for a very large sample. Previously, I used an illustration of roulette. The roulette wheel can be regarded as a random number generator. If that asumption is true, a roulette wheel is not predictable, not even using APS. I am not suggesting that the securities that we trade are wholly random. Perhaps they are mostly random but sometimes not random.
AFAIK, one can find all kinds of patterns in random data. Suppose APS finds patterns in random data. Do you conclude that: Bill, I do not understand your abbreviation "AFAIK". You will have to explain that to me.
(1) The market data is also random if APS finds patterns there. Not at all! The fact that APS finds patterns in any data (whether random or not) would not prove anything to me. Anyway, in your statement in the previous paragraph you said that "suppose that APS finds patterns in random data" and then you went on in this point to ask whether "the market data is also random if APS find patterns there". You already asked me to assume that the data is random. Therefore, whether APS found any patterns or not would be irrelevant to deciding whether the data is random. It already is random, according to you!
(2) The patterns in market data will not stay profitable in the future Once again, you have already asked me to assume that the data in question is random. Therefore, I would conclude that if APS did find some patterns in random data, the patterns would not last and therefore would not be profitable into the future. I would love APS to be able to find profitable patterns in the outcomes of roulette spins, for example. If so, I would take a (small) computer with me (with APS loaded in it) to the casino and clean up!
or if APS does not find patterns in random data do you conclude that:
(3) The market data is not random if patterns are found Once again, you question is nonsensical. You asked me to assume that the data is random and then you ask me whether I would think, if APS does not find some patterns in random data, that that would prove that the data is not random! Of course not.
(4) The patterns in market data are not random and will stay profitable I do not even understand this point of yours. You already stated that I should assume that the data is random so why would I conclude that the data is not random on the basis that APS did not find any patterns? You added "and will stay profitable". How will anything in this situation be profitable if APS did not find any patterns, as you stated above? To me, it would be excellent if APS does not find any patterns (or very few weak ones, at least) in random data. After all, random data by definition means that the outcomes are not predictable. They happen purely by chance. There is no guiding hand of god, fate, luck, fortune or karma to determine the outcome. As they say for roulette, "the wheel does not have a memory".
I think none of the above conclusions can be reached.I agree with you!
The only test that makes sense is actual trading and real P/L. I do not agree with this point, for the reason already stated above.
Bill
A roulette wheel is a completely deterministic system and can be fully predicted as was recently done by a team of Eastern Europeans using a super computer and used their system to break several casinos in England.
All dynamical systems that conform to Newton's laws are deterministic. It is highly questionable that truly random number generators exist. This means that patterns will repeat and can be predicted if the seed (initial conditions) of the number generator is known and its dynamics.
Now, to the problems you are discussing, APS is not trying to predict the next outcome. I think you need to re-consider your views about such and other mechanical trading systems. APS provides probabilities based on historical performance. This is something completely different.
In that respect, a system designed for a particular market that performs well in that market may completely fail when using random data. That proves the data is random and nothing else. Performance using random data and Monte Carlo simulation should not be confused. These are very different things.
Alex
Last edited:
True
