For years I’ve had the idea that there’s something to channels, an edge that could be exploited. In my early days as a trader I manually tried it a couple of times. Mixed results, mostly negative. Then later, as I still saw those channels holding up when I just looked at the charts with my own eye, I tried with truly automatic strategies. Alas, the backtests never held up. Now, with the help of Claude, it was time to get this out of the way once and for all. Claude produced two papers (not peer reviewed) that definitively answers the question: are regression channels special? Unfortunately, the answer is no. The paper explores 28 futures markets at various time frames and finds that regression channels are no better, maybe even worse, than randomly drawn lines. You can read and download the papers price-channels.pdf and regression-channels.pdf
The problem centers around two observations. The first is that if you have a sequence of top – bottom – top, then you can draw a line between the two tops, and a line parallel to that through the bottom. What I regularly see is that the price bounces of the lower line, a bit like here:
That’s completely disproved. The reason why I see it, and probably many others, is that we always look after the fact: hey, it worked here!
Another pattern that I think I see, but really isn’t there is the “cleanliness” of a channel. In my mind, if I see a real clean channel it goes on forever. Until it doesn’t of course, but I think it goes on longer than you would expect. Again, this is not true. Below is a picture of what I think is a clean channel: it is long, contains many bars and they repeatedly bounce against the bands.
You’d think the next bars will be contained in the channel as well. Not true. The channels are just as good as random lines.