Evidence-Based Technical Analysis: Applying the Scientific Method and Statistical Inference to Trading Signals

by David R. Aronson

Quick Summary

Aronson challenges traditional technical analysis by applying the scientific method and rigorous statistical inference to evaluate trading signals. The book covers epistemological foundations, cognitive biases that corrupt subjective analysis, hypothesis testing, data-mining bias, and presents a case study testing over 6,400 trading rules on the S&P 500 using bootstrap methods to assess statistical significance. It argues that only objective, testable TA rules validated against data-mining bias qualify as legitimate knowledge.

Detailed Summary

David Aronson's "Evidence-Based Technical Analysis" is a landmark work that subjects traditional technical analysis (TA) to the rigorous standards of scientific inquiry. Written by an adjunct professor at Baruch College's Zicklin School of Business who also has extensive industry experience, the book argues that TA must evolve from a "faith-based folk art" into a rigorous observational science.

Part I lays extensive methodological, psychological, philosophical, and statistical foundations across seven chapters. Chapter 1 distinguishes objective TA rules (which can be precisely defined and tested) from subjective TA methods (which rely on human interpretation and cannot be rigorously evaluated). Chapter 2 is a devastating 70-page critique of subjective technical analysis, drawing on cognitive psychology research to show how heuristic biases -- representativeness, availability, anchoring, confirmation bias, hindsight bias, and the illusion of validity -- systematically corrupt human pattern recognition in financial data. Aronson demonstrates that the human visual system is wired to perceive patterns even in random data, making subjective chart reading inherently unreliable.

Chapter 3 introduces the scientific method as the only rational framework for extracting knowledge from market data, distinguishing between deductive and inductive reasoning, emphasizing falsifiability as the criterion for meaningful claims, and explaining the role of parsimony and the discernible-difference test. Chapter 4 covers fundamental statistical concepts: probability distributions, sampling theory, the central limit theorem, and standard error. Chapter 5 presents hypothesis testing and confidence intervals in detail, including Type I and Type II errors, p-values, and statistical power.

Chapter 6 is the book's most critical contribution: a thorough treatment of data-mining bias. When hundreds or thousands of trading rules are tested on the same dataset, some will appear profitable by chance alone. Aronson explains White's Reality Check bootstrap method for testing the best rule discovered via data mining against the null hypothesis that no rule has genuine predictive power, and presents Tim Masters' Monte Carlo permutation alternative (published here for the first time). The chapter covers Bonferroni correction, stepwise testing, and the relationship between the number of rules tested and the expected magnitude of the best spurious result.

Chapter 7 surveys theories of why prices might exhibit non-random behavior, including behavioral finance, limits to arbitrage, adaptive markets hypothesis, and information cascades, providing the theoretical motivation for believing that some TA rules might work.

Part II presents a massive empirical case study testing 6,402 binary trading rules on the S&P 500 index from 1980 to 2005. Chapter 8 describes the rule universe (moving average crossovers, channel breakouts, filter rules, and momentum rules with various parameter combinations), the detrending methodology, and the statistical testing procedure. Chapter 9 presents the results: after correcting for data-mining bias using the bootstrap Reality Check, none of the 6,402 rules showed statistically significant predictive power at conventional significance levels.

This is one of the most intellectually rigorous books in the trading literature. Its conclusion -- that the most commonly used objective TA rules have no statistically verifiable predictive power on the S&P 500 after correcting for data mining -- is sobering but essential knowledge for any serious trader.