Making Financial Pattern Discovery Rigorous with Statistical Methods
Summary
The document discusses how financial pattern discovery can extend beyond traditional technical analysis. It distinguishes loosely defined chart patterns and indicator rules from approaches that state hypotheses mathematically and evaluate them with statistical methods. Examples include testing a parameterized trading system with cross-validation and analyzing asset returns rather than only price levels.
It also points to machine learning and models such as hidden Markov models as ways to study patterns or changing market regimes, including regimes associated with macroeconomic or policy events. The central lesson is methodological: technical or fundamental ideas can be investigated more rigorously when their definitions, tests, and predictive performance are made explicit. The responses offer conceptual distinctions rather than a particular fitted model, empirical dataset, or evidence that any proposed pattern predicts returns reliably; careful validation remains necessary.
Key ideas
- Rigor depends on clearly defined methods and tests, not only on the type of market data used.
- Quantitative methods can formalize trading ideas and assess their predictive performance.
- Cross-validation can test whether a parameterized strategy generalizes beyond its design data.
- Returns, regime models, and machine-learning methods offer avenues beyond price-chart rules.
- The discussion does not establish that any specific pattern or model is profitable.
Tags
Full text
# Data mining of financial time series and pattern discovery - beyond technical analysis? # Data mining of financial time series and pattern discovery - beyond technical analysis? There are lot of works about pattern discovery in financial time series - but all of them (known to me) are connected with technical analysis and that is field (as said by many) that is not rigorous science. So - are there pattern discovery approaches that goes beyond technical analysis and that can be called science - e.g. ones that try to discover parameters of hidden markov model or that otherwise try to model the different financial regimes as responses to the macroeconomic events (e.g. country's position in the economic cycle) or as responses to the other events, e.g. by changing tax legislation, by monetary policy changes (interest rate etc.), by governments' stimulus efforts, etc.? ## Answer by vonjd (score 4) https://quant.stackexchange.com/a/31890 I think it is important to understand the following: Technical analysis is not rigorous not because of the used data per se but because of the used methodology! Classical technical analysis uses ill-defined graphical patterns and has no good quantitative methodology for testing the accuracy of the results. Many of the same problems often haunt classical fundamental analysis which uses different metrics so the big difference to quantitative finance - as I understand it - is a mathematical systematization of those concepts and using the scientific method to find out what works and what doesn't. On top of that with quantitative finance und especially machine learning (which is the term used nowadays) you could use the full construction kit of algorithms that can learn from and make predictions on data. But even classical technical analysis can be made rigorous, see e.g. my answer here: https://quant.stackexchange.com/a/8258/12 ## Answer by Donny Lee (score 3) https://quant.stackexchange.com/a/31004 Here is my understanding of when technical analysis and mathematical techniques differ. Technical analysis predominantly utilizes three things: - Price levels - Indicators on price levels - Trading using rules based on 1 and 2. Anything outside of 1, 2 and 3 is beyond technical analysis. Naturally, your analysis can involved all 3, but once you use a technique outside those three, you are slowly inching towards statistical and mathematical techniques. Examples follow. The typical technical analysis backtester is, pick an asset, pick an interval, define an indicator on OHLC bars in that interval and then buy when indicator > value and sell when indicator < value. Suppose, now you have a trading systems defined by parameter set. And you perform cross validation to test its predictive power. While you have used technical analysis to devise the trading rules, cross validation is a statistical technique. Suppose now that you've decided to perform you analysis. But instead of looking at price levels, you look at the returns of an asset instead of price levels. To me, using returns of an asset is more in the domain of mathematical analysis than technical analysis. Finally, there is a host of other trading strategies that originates from the return series. I can convincingly say that these are NOT technical analysis.
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.