It works but it is getting harder and harder because more people use software and computers to analyze prices and as a result anomalies that give rise to profitable patterns and trading rules are eliminated at higher rates than before. I found many answers in the recent book by M. Harris "Fooled by Technical Analysis" In the book Harris talks about what works and what does not. He makes some interesting arguments against the constant use of backtesting and warns about the dangers of data-mining bias. Format is web based but worth reading it.
Rather than searching for "anomalies", suggest looking for "correlations". They are still there... haven't really changed in the last 20 years or more. Anyone who is "fooled by TA", doesn't know how to do it correctly.
But correlations do not indicate direction and are non-stationary unless you are a very long-term trader.
(1) I don't know what you mean by "non-stationary". (2) Correlations absolutely DO* indicate direction short term. * meaning, with "tradable and risk/reward-favorable probability". Nothing of course is absolute... which is why trading with stops is essential
Point of view (A): I define technical trading specifically as mumbo jumbo methods that consist of placing a chicken bone on my tablet over the top of a price graph and seeing if there is any price visible. This method did not work when I backtested it for 3 days in 2004, for the S&P 500 (and for robustness I used 3 different chicken bones). Therefore technical trading does not work. Point of view (B): Anyone who looks at a chart is a technical trader. Anyone who looks at a price when they trade is trading techically (try putting a post it note on your screen on the part of your brokers website where the bid/offer comes up). Anyone who knows what a price is, is technical trading. Since some traders make money, and since all traders do at least some of these things, technical trading works. To save time I will copy and paste this reply to all TA threads from now on. GAT
If TA = anything based on past prices then yes, it absolutely and undeniably works. Renaissance Tech is collecting huge amounts of past data, they wouldn't do so if it wouldn't work. Google their track record. There is no crystal ball, there are only probabilities.
Great points and I've witnessed it myself as well. Most great edges have limited liquidity and when too many jump into those, we know what happens.
As a day trader, I use charts because I am not smart enough nor do I have the time to figure out the meaning of life every time I take a position.