Quote from intradaybill:
Your conclusion does not follow from your premises.
To get to the core of this, as jcl, another member, said in another thread, series autocorrelation is empirically measured to last for up to 4 bars. I say maybe up to 10 bars. Then it vanishes. It is then unrealistic to expect patterns formed over long periods in daily or weekly or even monthly charts to convey any information about the future.
TA is a very broad subject. Some aspects of TA may work well, like support and resistance, price patterns, some algos using price, volume and open interest, even trendlines when properly drawn and used for very short term forecasts. Chart patterns - things like triangles, double bottoms, head and shoulders, rectangles, etc. formed over many bars, weeks or even months cannot convey any information because the autocorrelations at the start of these patterns change continuously throughout their formation and information is lost. Thus, these patterns are random in the best case. According to this view, any books that still mention such formations as viable signals are in the best case pranks. What they do is showing the formations that worked and not the numerous ones that did not. This is selection bias. Both wrong and even unethical.