Momentum-Based Learning Strategies for Technical Analysis: Implications for the Efficient Market Hypothesis
Technical analysis seeks to find patterns in past stock prices that are profitable to trade on in the future. Allen and Karjalainen’s (1999) genetic algorithm finds technical analysis trade rules while avoiding the pitfall of data-snooping. However, they acknowledge that their approach leaves room for a specially- designed algorithm. Using their results with past profitable trade-rules in the literature, particularly those of Hsu and Kuan (2005), I design a learning strategy algorithm that generates simple technical analysis rules. With it, I demonstrate eight time deciles of technical analysis profit beyond realistic transaction costs over the last twenty years in the component DJIA stocks over the 5-minute time frequency, and nine to ten time deciles in the ETF DIA (a DJIA tracking ETF) over the 5-minute time frequency and the MA parameter, depending on the summary statistic. These results are largely robust to slippage. As a whole, the results are on the margin of violating the Efficient Market Hypothesis (EMH) but ultimately confirm a weak form of the EMH.
Svogun, Daniel Philip, "Momentum-Based Learning Strategies for Technical Analysis: Implications for the Efficient Market Hypothesis" (2020). ETD Collection for Fordham University. AAI27956713.