http://stockornot.i2pi.com/?s=1564875868.gif http://blog.i2pi.com/2007/03/28/stock-or-not/ any comments from the TA contingent? thanks! good luck, surf
Mandelbrot shows the same idea in fractal discussion. The issue is does trend following or anything else have any predictive value? That is, tell me today what the dow will close at tomorrow. Trends are historical only. http://www.elitetrader.com/vb/showthread.php?s=&postid=1313553#post1313553
I read this in the comments: "Perhaps my success is attributable to my fall-back plan of picking the asset with the lower variability." That seems to ring true, I'm up over 60% with that method.
These so-called experts also missed big. Actually they are shooting the wrong target. In the testing website you always have to choose betwen the two graphes each time, even though you are not sure most of the times. This is not what pro traders do in reality. You don't have to be in market all time. You will only trade when you find a high probability, or high expectation set up. In the analog of the testing website, you should be allowed to pass if you are not sure which one is true market data. If the rules are changed this way I best the performance will improve dramatically.
The Misbehavior of Markets by Bernard Mandelbrot. pages 118-119. He constructs some fractal charts which look very real. I am sure some one is making money off this. I had a guy sending me emails for a while trading off this idea, I don't know what he was doing but almost all his calls, like 8 out of 10 were right, and hit the profit target. He was over in Britain.
Garbage! Even though the dude apparently used a GARCH process, I was able to consistently exceed 60% on numerous occasions (each with 50+ trials). Lucky? Maybe. But even to the naked eye the flawed price generation process makes easier to pick - and I didn't even pass on the ones that looked close - bet i can get a better score if I did. And this is from someone from the Quant contigent! In addition surf, you appear to be missing the point here? The test isn't about finding trends or price prediction, it is about whether a human can discern the differences between actual and artificial time series. What I'd like to see is whether people with serious screen time performed better than noobs, as the current test does not differentiate between experience levels. Almost like asking anyone at random whether they can tell the difference between an original artwork and a fake. Try that with people having some knowledge in the field, and then you will not be getting the usual monkeys yelling how statistics lie, where in fact it is the testing methodology is biased thereby skewing the results.