Miles'o: I think correlation coefficient is used as a normalized measure of the variation between 2 data sets (I think it's covariance divided by standard deviation), so I think if data set #2 were a standard normal distribution, that would give us what we're looking for. Now that I think about it though, I'm not sure it would tell us anything we couldn't already see: The corr-coeff would be near 1 when the data hugged the MA within the observation window, and would move toward zero probably exactly in sync with the price plot moving outside a 1 standard deviation band and beyond, and probably goes toward -1 at turning points, until the moving average catches up with the turn. My guess would be the longer the window size during a consolidation, the more closely the price distribution approaches a standard normal distribution. One of the key things you're supposed to look for with BB's is a narrowing down to a neck, and then an expansion. But, this is generally obvious from the price plot. In fact, you can usually sketch what the Bollinger bands would look like before you plot them. Any time there's a rise, a pause, then a rise again, the bands expand during the rise, contract during the pause, and reverse and expand again when the trend resumes. If the window size is less than the duration of the pause, the bands will run parallel during the pause (because the data within the window is approximately normally distributed, flat, with small variance), then expand. If the pause is shorter than the window, the data in the window will still have some slope to it (not zero mean) and more variation when the price breaks out again, so the bands will curve down (but never quite run parallel to the price), and then immediately curve up when the breakout occurs (more parabolic shaped). You can sculpt these to look almost any way you want by playing with the parameters, which is why I think that if a person is going to use these, they should look at 2-3 timeframes, like 8,34, and 144 days, or better yet, have the 3 time frames selected automatically based on some cycle measurement of the stock in question. That way you'll always be consistent. Doesn't mean it will work better just because you're consistent, but you can't figure out how to adjust it to get better performance unless it's used consistently.
dh |
