how would you go by testing to prove/disprove their predictive power once the computer plots the line correctly?
Excellent question, Eusdaiki. Well, if your computer program knows what trendlines and support/resistance lines are (assuming we can clearly define them in computer terms) it will be able to test all kinds of classical chart patterns that supposedly make money in the long run, starting with the ones popularized by Edwards and Magee. For instance, we could test the breakout theory : We could tell the computer to buy any breakout of a resistance line and short any breakdown of a support line and see if we could indeed make any money that way (using all types of stops and profit targets). The breakout itself could be clearly defined as follows, for example : 2 consecutive candles completely above a resistance line (for long positions) or below a support line (for short positions). This would open a whole new world of trading research and possibilities!
I am actually coming more and more to the view that a random walk model of the market is the best description of how prices move through time. This would mean that pricing models that give future value are the most "true" model, and that models which tell you when to buy and when to sell are BS. In fact all of TA would be BS in a random walk, and of course your trendlines would be BS. Although there may be a bit of a self-fulfilling prophecy aspect to support/resistance levels. I still like the intellectual pursuit of TA, it's just that I've never made any net profit using it. It's quite a natural thing for people to look for something that explicitly tells them to buy now or sell now. It could be chasing a feather in the wind though.
models that give future value? you mean like fundamental models based on the dividends and other corporate events?
the problem with this approach is that you are testing 2 things at once. 1. the predictive value of the signal or pattern 2. the effectiveness of the rules which you use to trade as a consequence of the pattern. separating the 2 effects can be tricky.
No, because a chart pattern by itself is meaningless, you need the pattern AND the trigger to initiate the position. For example waiting for a head and shoulders formation to complete is only the first step, now you need a penetration of the neckline (the trigger) to initiate a long or short position. The trigger here is of course the tricky part, as you can design/backtest hundreds of them. And if you add one or more filters (each with its own numerical value), for confirmation purposes, now we are talking about thousands of possible scenarios. That's why we need a freaking computer to test the alleged predictive power of these chart patterns, assuming we can code the (rather subjective) trading rules