Adventures in Technical Analysis, McClellan Oscillator Edition

When I dumped on technical analysis last month, one of the more unexpected results was a long email thread with a chap called Jeff Drake, on the subject of something called the McClellan Oscillator. He's a big fan, while I was far from convinced. So we worked out a deal, pitching Jeff's oscillator against my random results.

Jeff would let me know when the Oscillator next hit 60 (don't ask), and compare his own three-day return from shorting the market to the three-day return I would have made shorting the market on a random date: the first of that month. He'd even spot me 50 points.

Last Thursday I got an email from Jeff, saying that the Oscillator gave a sell signal:

You sold the market on your random day of July 1 and closed 3 days later down 190 points (hmm....you're pretty good at this) and then we'll add 50 points for a total gain of 240 points. I'm selling the market today at 11,632 and we'll see where we stand at the close next Monday.

Where did the Dow close Monday? 11,131: a drop of 500 points. Clearly, Jeff did a lot better with his carefully-timed technical analysis than I did throwing a dart at the calendar. And he's quite evangelical about it, too:

I hope you finally realize that I've given you quite possibly the most financially valuable gift you'll ever receive in your lifetime. Use it wisely and with strong discipline and you will easily and consistently beat the best annual return on investment that most long-term investors like Buffett and Lynch have ever gotten in their entire careers (and that's with your portfolio sitting in cash 90% of the time).

I should note that this wasn't a one-off thing: according to Jeff, the Oscillator has given buy/sell signals so far this year on February 1, March 3, March 10, June 12, June 24, July 15, and July 24. In most cases, buying/selling the broad stock market and then unwinding the trade after three days would have been very profitable; the June cases, not so much.

After all this, I decided that at the very least I should bother to find out what this Oscillator actually is. Investopedia came to the rescue, kinda:

To calculate subtract a 39 day EMA (of advancing issues - declining issues) from a 19 day EMA (of advancing issues - declining issues).

An EMA is an "exponentially weighted moving average," which means that more recent days are weighted higher than older days.

The main thing which strikes me here is how arbitrary all the numbers are. Why 39 days and 19 days? Why are the key levels on the oscillator +60 and -60? Why is the amount of time to hold the trade 3 days? These aren't the kind of things which can be worked out ex ante: they're used because they're the numbers which have worked in the past. And, for all I know, they'll work in the future as well. On the other hand, all trades work until they don't, and nothing works forever.

The reason I'll never trade on the McClellan Oscillator is simple: I don't trade. (I'm a fan of the dictum that you should never enter any position you wouldn't be happy holding for three years.) But, even if I did trade, I would never trade anything I don't understand. And the McClellan Oscillator, with its emphasis on the difference between two pretty similar moving averages, is not something I can come close to understanding.

Could it be that there are some genuine reasons related to market psychology why the McClellan Oscillator has worked in the past? Well, possibly. But psychology is hardly an exact science either. And the idea that you can start with some kind of psychological tenets and end up with the difference between a 39-day moving average and a 19-day moving average - that's surely impossible.

So I'm glad that Jeff is making lots of money trading his oscillator. But even he admits that he stayed short in June when the oscillator gave a buy signal, "because I follow a LOT of technical charts and I suspected the bounce would reverse quickly because the other charts were still very bearish".

Me, I'm much happier not following any technical charts. It's much less confusing that way.