QuoteOriginally posted by: Traden4AlphaI, personally, don't restrict my attention to the first moment of price movements. And I agree that the emphasis on first moments is one of the deep failings of TA. QF's tools for the second moment provide great value, although QF has its own failing where it uses distributions that that empirically invalid.What distributions you're talking about? There are different models and different distributions, some of them (gaussian) further, some of them (Levy-syle) closer to reality. It seems to me you're captive of some 40-years old basic research like Black-Scholes QuoteNo they aren't "words, words, words", they are the proper decision-theoretic perspective on claims made by TA (versus price prediction or correlation). TA does necessarily NOT claim that if f(Price)>80, then E(dPrice/dt) <0. Instead the claim is more of the form that if f(Price)<80 and you sell short at that point and you cover immediately if dPrice(t)/dt>0 and hold the short position as long as dPrice(t)/dt <0 or some other indicator provides a buy signal, then the total price drop (and total profit form shorting) will be large. The expectation for profit can be very different from the expectation for price change depending on the exit tactics used by the trader (QF provides some nice tools for thinking about stochastic systems and branching games with exit conditions). Moreover the expectation of profit for a roundtrip trading strategy based on a given indicator can be very different from the correlation structure between the indicator and future price movement. My point is that correlation between an indicator and prices is neither necessary nor sufficient for profits.Aaargh I understand your concern about correlation, ok, now could you PLEASE specify some strategy, some statement, conditional on something in the past probably, but well-defined, so we could test it? Otherwise we're just wasting time speaking about the maturity of pink elephants.QuoteThe fact that GARCH does a better job of simulating price data is intriguing to me for what it says about risk vs. return. Isn't there a bit of a paradox to persistence of volatility and non-persistence of excess rates of return? Why not consistently buy persistently volatile security to earn a higher rates of return? And if a forecast of high volatility isn't a forecast of high return, then why not avoid all high volatility securities?Very interesting remark, never thought about this. Is volatility persistent on the big timeframes (i.e. daily/weekly)? I thought it's persistent on the small timeframes like M1 - M15, so there isn't a lot of forecasting power there?QuoteThen why do so many traders still use TA? Why are TA tools built into trading platforms, data services, and Bloomberg? We could just scoff at traders' use of TA as just a case of financial astrology, but that would miss a great opportunity to learn something deeper about the markets. If a sufficient fraction of volume is driven by TA-influenced trading, then what will that do to prices?Well, it's not a bad proof, that someone uses it ergo it's good. But we don't know whether someone uses it. Yes, some trader can say: I use TA & TA is useful. But they won't share his secrets with you therefore they don't make any claims. So their theory couldn't be verified. Moreover, I don't know any sucessful trader using TA (although I don't know many succesful trader at all). Surely you can use moving average and say that you are using TA - but that would be a bit unfair. Likewise I can use moving average and say that I use regression and square-least error thus I use QA.

QuoteOriginally posted by: amiralievAaargh I understand your concern about correlation, ok, now could you PLEASE specify some strategy, some statement, conditional on something in the past probably, but well-defined, so we could test it? Otherwise we're just wasting time speaking about the maturity of pink elephants.A statement to test: prices are subject to support & resistance (S&R). S&R are the most fundamental explananda of TA; trends can be seen as a sequence of S&R. Note that they are more or less symmetrical. So we can test the following thesis:Hypothesis: Existance of Resistance.Defintion of Resistance (1st try): if a price p has a level A which is reached at time t the probablility q: p*<p is higher for all t* > t, with t* < some t#. If p*>p then the probability u: p*' > MA_x(p') is higher than 50%.In words: if there is Resistance at A, the price is more likely to reverse than not. If the price crosses A, the momentum is bigger than usual (the moving average momentum of last x days).There is evidence that S&R are in fact psychological phenomena, because people tend to think of levels in analogy to a hurdle. People tend to think in terms of discrete prive movements.

Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical ImplementationLO, MAMAYSKY, WANGThe presence of clearly identified support and resistance levels, coupledwith a one-third retracement parameter when prices lie between them,suggests the presence of strong buying and selling opportunities in thenear-term.with this one:The magnitudes and decay pattern of the first twelve autocorrelationsand the statistical significance of the Box-Pierce Q-statistic suggest thepresence of a high-frequency predictable component in stock returns.Despite the fact that both statements have the same meaning—that pastprices contain information for predicting future returns—most readers findone statement plausible and the other puzzling or, worse, offensive.

QuoteOriginally posted by: yabbadabbaQuoteOriginally posted by: Traden4AlphaI think that TA and QF (and science) bring some powerful (but flawed) tools to the table. Adding in insights from agent-based computational economics and complex adaptive systems should provide a better way to look at the markets.Is there any research in this direction? Do you think it is an easy way from the explanation to the decisions? What about machine learning? Machine learning as the science of pattern recognition could add some insights.Machine learning (and artificial intelligence) does provide some useful tools for both theory (mathematical representations of pattern-finding and decision making) and practice (algorithmic structures for sense-deliberate-respond cycles embedded in a real-time environment). I noticed that 9 Wilmott job postings in the last 9 months asked for machine learning expertise.I'm not sure whether machine learning falls in the QA or the TA camp. Some of machine learning research (mostly pattern recognition and iterated game theory) seems like the kind of hard-core math and statistics used in QA. In that regard, machine learning can fall into the same trap as QA (many examples of machine learning take place in closed toy-worlds designed for their analytic tractability). But other work (often in robotics and embedded agents) uses less rigorous ad hoc heuristics more common in TA. One could also argue that machine learning subsumes both in being an attempt to reduce the scientific method to algorithms. As such, machine-learning-as-science uses both QA and TA concepts as atoms of hypotheses or implementation methods in testing those hypotheses.

I think, that most of the research analysing support and resistance looks at them from the stop-order structure point of view. I have seen some calculations done, comparing price changes near integer numbers vs. far from integers. The results suggested, that price behaviour is statistically different in the surroundings of integer price levels. I do not remember any details, though.

QuoteOriginally posted by: KackToodlesQuoteOriginally posted by: torontosimpleguy But science is a tool to understand world around us. by your definition, then, wall street and QF does no science. People in QF and wall street don't care about "understanding" the world. They care about making money. Whatever makes their stock picks go up. They don't understand it, but if they make money, they are happy. You will NEVER see a wall streeter exclaim, "I understand it and don't care if we didn't make any money!" That's a good point. There are several issues involved. First I want to drop the name V.Niederhoffer who proclaims a scientific trading methodology. The problems you are ignoring when you only care about your P&L are a) the dimension of riskb) the dimension of continuityc) the dimension of control and transparency (typically the managed capital is borrowed, thus the principial-agent problem has a big impact).The object of any financial activity is to maximize profits while adhering to the contraints a,b,c.

Quote A statement to test: prices are subject to support & resistance (S&R). S&R are the most fundamental explananda of TA; trends can be seen as a sequence of S&R. Note that they are more or less symmetrical. So we can test the following thesis:Hypothesis: Existance of Resistance.Defintion of Resistance (1st try): if a price p has a level A which is reached at time t the probablility q: p*<p is higher for all t* > t, with t* < some t#. If p*>p then the probability u: p*' > MA_x(p') is higher than 50%.In words: if there is Resistance at A, the price is more likely to reverse than not. If the price crosses A, the momentum is bigger than usual (the moving average momentum of last x days).There is evidence that S&R are in fact psychological phenomena, because people tend to think of levels in analogy to a hurdle. People tend to think in terms of discrete prive movements. Very good, now we have something to talk about. So now we should conduct some kind of experiment to prove this. As I see you are accept the probability paradigm (as you say "probability" ). As we have limited data we will have some estimate of probability you mentioned. So what I have to test you hypothesis is around 20 years of EURUSD (earlier some index of euro area currencies vs dollar) daily, here (probably you suggest some other source, it doesn't matter). As you clearly don't accept that markets are brownian motion or exp-bm or something else we have two options1) you suggest a model to calculate a treshhold (i.e. 55% or 60%) from which we will accept the hypothesis2) you give me this level ad hoc and I agree with it or disagree if you take something too narrow At least we discuss this.Then we take MetaTrader write the algorithm and see the mean percentage of guesses. Agreed? By the way, as you understand the percentage of guesses doesn't matter - in fact even in perfect market the prediction "prices will go 10 points up , not 90 points down" will be right in 90% cases. So probably you should modify your statement so that it includs some kind of strategy, buying and selling, and we will prove that it generates return . Or?

QuoteOriginally posted by: amiralievQuote A statement to test: prices are subject to support & resistance (S&R). S&R are the most fundamental explananda of TA; trends can be seen as a sequence of S&R. Note that they are more or less symmetrical. So we can test the following thesis:Hypothesis: Existance of Resistance.Defintion of Resistance (1st try): if a price p has a level A which is reached at time t the probablility q: p*<p is higher for all t* > t, with t* < some t#. If p*>p then the probability u: p*' > MA_x(p') is higher than 50%.In words: if there is Resistance at A, the price is more likely to reverse than not. If the price crosses A, the momentum is bigger than usual (the moving average momentum of last x days).There is evidence that S&R are in fact psychological phenomena, because people tend to think of levels in analogy to a hurdle. People tend to think in terms of discrete prive movements. Very good, now we have something to talk about. So now we should conduct some kind of experiment to prove this. As I see you are accept the probability paradigm (as you say "probability" ). As we have limited data we will have some estimate of probability you mentioned. So what I have to test you hypothesis is around 20 years of EURUSD (earlier some index of euro area currencies vs dollar) daily, here (probably you suggest some other source, it doesn't matter). As you clearly don't accept that markets are brownian motion or exp-bm or something else we have two options1) you suggest a model to calculate a treshhold (i.e. 55% or 60%) from which we will accept the hypothesis2) you give me this level ad hoc and I agree with it or disagree if you take something too narrow At least we discuss this.Then we take MetaTrader write the algorithm and see the mean percentage of guesses. Agreed? By the way, as you understand the percentage of guesses doesn't matter - in fact even in perfect market the prediction "prices will go 10 points up , not 90 points down" will be right in 90% cases. So probably you should modify your statement so that it includs some kind of strategy, buying and selling, and we will prove that it generates return . Or?On this subject, I suggest Osler (2000) - Support for Resistance, and Osler (2001) - Currency Orders and Exchange-Rate Dynamics.

Hi amiraliev, thanks for your suggestions. I agree with the procedure. My problem with a better defintion is the concept of timeframes. Every R-level is valid for a certain time on a certain time scale. One would have to formally define time frames. Perhaps this has been done - I don't know. Every price series has a fractal nature, but there are timescales where changes are discrete (ticks).Another question is how many R-levels we should expect to find in a TA series versus a RW series. Let's first improve the defintion and perhaps get some technical traders to review it, before we move on to a test.(Minor change: a further implication is that if the R-level is rejected then the momentum is higher than usual. Therefore the probablity u is higher than usual for every p*.)

C'mon you guys. It's the one that makes the most money. Therefore QF > TA.QF is more accurate at forecasting more of the time.Except for them Balck Swans. And then we're most certainly fooled by randomness .

QuoteOriginally posted by: TraderJoeC'mon you guys. It's the one that makes the most money. Therefore QF > TA. . If what you say is true, why does wall street have more highly paid analysts doing TA than QF? is wall street throwing their money away stupidly?

My guess is that TA-using/understanding traders are easier to find

QuoteOriginally posted by: yabbadabbaHi amiraliev, thanks for your suggestions. I agree with the procedure. My problem with a better defintion is the concept of timeframes. Every R-level is valid for a certain time on a certain time scale. One would have to formally define time frames. Perhaps this has been done - I don't know. Every price series has a fractal nature, but there are timescales where changes are discrete (ticks).Another question is how many R-levels we should expect to find in a TA series versus a RW series. Let's first improve the defintion and perhaps get some technical traders to review it, before we move on to a test.(Minor change: a further implication is that if the R-level is rejected then the momentum is higher than usual. Therefore the probablity u is higher than usual for every p*.)1) Yes, it's ok with the timeframes, they are predefined. I suggest daily, so we will have ~ 20 years of data.2) Probably my English betrays me, but I don't grasp some sides of the definition. Probably you can write (in english) the entry and exit rules (i.e. buy when the price hits A and bounces back to A+15 in the next 2 days then buy).3) Concerning the Osler article I hadn't enough time but generally he manages to grasp some important points like the fact that "indicative" (i.e. Reuters) quotes differ dramatically from "real" (those you could work with, interbank). Moreover he uses bootstrap and I like this too. The general remark I would like to do preliminary is that the statement (like "there is a support at 1.23") could be exact and useless at the same time, because even martingales can have support & resistance. Nevertheless, even the "weak" fact like "support exists" (not "support exists and you can profit from this") is quite interesting and deserves to be explained by some scientific research (not ad hoc TA arguments). I don't believe personally that it is just psychological effect.

QuoteOriginally posted by: KackToodlesQuoteOriginally posted by: TraderJoeC'mon you guys. It's the one that makes the most money. Therefore QF > TA. . If what you say is true, why does wall street have more highly paid analysts doing TA than QF? is wall street throwing their money away stupidly?Prove it! Where's your stats, Professor???