Theoretical Arguments for and Against Chart Pattern Prediction
Summary
The document considers whether patterns found in historical prices can forecast future returns, contrasting pattern recognition with trend-following and mean-reversion. One view invokes weak-form market efficiency: tests are described as supporting the idea that past prices do not reliably predict US stock returns over horizons longer than a few minutes. Apparent chart-based profits may instead reflect hidden risks, including the possibility that a convergence trade moves further against an investor before it reverts.
Other replies argue that a statistically meaningful pattern can represent structure rather than noise and support a model for forecasting. They also point out that this reasoning rests on induction: past regularities do not prove future repetition. The document mentions research that tests returns conditional on predefined chart patterns and reports a difference, but gives no methodology or detailed results. Its replies are opinions and brief references rather than a consensus or a practical validation framework; the existence and persistence of predictive patterns remain unresolved.
Key ideas
- Weak-form market efficiency challenges the use of past prices to predict returns.
- Apparent technical profits may involve risks that are not visible in chart patterns.
- A pattern that separates structure from noise can serve as a model for forecasting.
- Inferring future behavior from historical regularities relies on induction and cannot be proved by past success alone.
- Research cited in the discussion examines returns conditional on predefined chart patterns, but details are not provided.
Tags
Full text
# Is there any theoretical basis for pattern-recognition strategies? # Is there any theoretical basis for pattern-recognition strategies? Mean-reversion and trend-following strategies have some kind of a theory behind them that explains why they might work, if implemented well. Pattern-recognition, on the other hand, seems like nothing more than data mining and overfitting. Could patterns possibly have any predictive value? In other words, is there any theoretical reason why a pattern observed in historical data would be repeated in the future, other than random chance or self-fulfilling prophecy where the pattern "works" because enough traders use it? ## Answer by Richard Herron (score 13) https://quant.stackexchange.com/a/87 Weak form market efficiency says that you can't predict prices based on past prices. Or that technical analysis doesn't work. I think that the tests of weak form market efficiency are pretty conclusive and show that the US stock market is weak-form efficient; at least on a a timeline longer than a few minutes. That's not to say that markets are "efficient". The tests of semi-strong form efficiency (i.e., can't predict prices from all public info) are still debated a little, but I think most would say that markets are not semi-strong form efficient. You can do fundamental analysis have a chance a determining winners and losers. And markets are definitely not strong form efficient (i.e., can't predict prices from all info, public and private). So does technical analysis work? I don't think so. Some may be earning abnormal returns, but they're likely taking risks that aren't obvious from the charts. Or in the context of mean reversion, yes, most of the time things revert to the mean, but they may not revert to the mean within your tolerance for pain. I think the best light read on the mean reversion topic is "When Genius Failed". Their convergence trades on off-the-run and on-the-run Treasuries were "right", but they went further away from the mean before converging after insolvency. ## Answer by vonjd (score 8) https://quant.stackexchange.com/a/83 General answer to a very general question: If you find a significant pattern which distinguishes between structure and noise you understand something about that system. You have a model about it so you can extrapolate and forecast. On that basis you can use this model to make money. In that sense mean-reversion and trend-following are also "only" strategies that use a model derived from the data (or where ever from). Take evolution as a proxy: The living organism also have a model about their environment (which is partly stochastic, too). The successful organisms use that to survive and breed. As an aside: In terms of a philosophical basis you can say that it is the belief that the past has meaning for the future (inductive argument) - but this is only a belief which cannot be justified in itself (saying that it worked well in the past is itself inductive and therefore we have a catch-22 here) ## Answer by Zarbouzou (score 4) https://quant.stackexchange.com/a/127 To my point of view there are never any reason that a pattern or anything else would repeat in the futur. Actualy I don't see any difference between pattern recognition and mean reversion/Trend Following in term of theoretical proof. One can read Pr Andrew Lo : "Foundations of Technical Analysis". It tries to give a theoretical background to TA by studying the enpirical distribution of stock returns conditionned or not on the presence of predefined chart pattern. His result is that there is a diference that justifies the use of TA. ## Answer by Samir musa (score 1) https://quant.stackexchange.com/a/46147 I do not believe market is random and neither do many members of this forum. Simply by that notion you believe that, market follow a systematic and cyclical approach. The success of any investor is to identify those cycles and take advantage of them. Systematic, cyclical both imply the existence of pattern.However, we all believe the existence of theme or context which is the greatest cause of confusion to any trader. Is the basis of all of our problems.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.