[Bollinger Bands]------------------------------------------
Bollinger bands, a.k.a. trading bands, are said to be an indicator of volatility. They are plotted as an envelope of plus and minus two standard deviations of some price measure, offset from the mean of this price measure. Bollinger's recommended period for the average is 20.
The Misuse? of Bollinger Bands:
Most implementations of Bollinger bands use a 20-day average of the CLOSING price. Bollinger recommends using either the "typical price" (H+L+C)/3 or the "weighted close" (H+L+C+C)/4 as the daily price measure, rather than the close. This measure is used for both the moving average and band calculations. His examples use the close just to make the overall description easier to understand. This, however, is only a minor consideration since cases could be found where one price measure outperforms the other and vise-versa. Additionally, it is commonly believed that one should use the crossing of the bands (by the price plot) as buy/sell signals. In fact, Bollinger himself says: "Trading bands do not give absolute buy and sell signals; they provide a framework for relating the price to other indicators." He recommends "On Balance Volume", "RSI," and "Money Flow" as the indicators to defer the trading band's signal to.
Bollinger Band Observations:
If price is moving sideways, with very little change among the daily weighted closes for the 20-period observation window, clearly the variance (and therefore the standard deviation since it is the square root of variance) is small. Thus the bands contract toward the mean. We don't need Bollinger bands to tell us this--it's obvious from the chart that the day-to-day price progress is small. "Small" or "tight" aren't quantified relative to standard deviations, so I'd be willing to bet that the trading bands add nothing to our analysis of price consolidation that we couldn't already see from the chart because it's relative either way.
If prices are range-bound (bouncing back and forth between horizontal or diagonal trendlines), and that range tightens up, so do the bands. When price breaks out from a consolidation region and ramps up, the price will tend to ride the upper band. But of course!, the new prices during the breakout are statistical outliers relative to the tight, flat, consolidation region (which still makes up most of the observation window's data). Naturally this new data is going to ride the plus-2-standard-deviation band, or even go well outside it for strong breakouts. Stated another way, this breakout is surely going to be in the 90+ percentile of the dataset, as will the confirmation day, and the next confirmation day, and so on. A breakout trend hugging the upper band isn't a surprise. Same with a breakdown trend, except it hugs the lower band.
Also, it's expected that most of the data is going to fall within the bands--that's what 2 standard deviations means! The fact that they tend to bounce from the upper band to the lower is also of questionable significance. The bands are self-adjusting to fit the data, because they are based on the data. And the definition of the standard deviation wont let today's data go outside the 2-standard deviation band unless it is way out of line from the norm (which would have been clear just from looking at the price chart). Since price movements in general don't stray too far from the "normal" movements except when strongly trending, and the intermediate-term price tends toward its average value, it makes sense that a move to one band is many times followed by a counter move to the other band, such that the net over the period of these two extremes is the mean.
Which gets me to my point. I think they have a valid use, but they are being used incorrectly as a volatility indicator. They don't indicate volatility as much as they indicate trend, and the thrust of the trend. As outlined earlier, when the price is moving sideways with little price progress, the bands are narrow ("trendless"). When the price breaks out into a trend, the bands expand. When the thrust of the move begins to decline (i.e. the day to day net progress in price is shrinking), the bands contract. The bands always overshoot reversal tops (v-tops), and are far too wide at that point to be of any use as an exit indicator.
The fact that a contraction in the bands tends to be followed by a breakout is no new revelation either. The tightening of the price range towards the end of a basing region is part of the constructive setup conditions necessary for a strong breakout. This all can be read from the chart. I think when people look at Bollinger Bands without really thinking about them, they are left with the impression that the have an uncanny ability to follow the price swings. But they should! They're sculpted from the price and price swings.
In short, all I think they do is confirm that prices tend toward the intermediate-term mean except when prices are strongly trending. When prices are strongly trending up, most of the price activity is above the moving average, so the data is skewed toward the top half-band. When prices are strongly trending down, most of the price activity is below the moving average, so the data is skewed toward the lower half band. The more exponential the rise (or decline), the wider the bands get (i.e indicating strength of trend).
Some day soon, I'll give explanations and links to charts which use the VLC I talked about in previous posts, and some other statistical band indicators that I think are an improvement on Bollinger bands-especially using Bollinger bands in isolation, as many people do. These new indicators give a complete view by characterizing volume, price progress day-to-day, and price progress intraday as separate indicators. When using multiple indicators, they must each be derived from different pieces of the only data we have available: O-H-L-C-V, and their changes with time. Otherwise, they will of course confirm the signaling indicator since they are all just derivatives of the same data (i.e. if all the confirming indicators are derived from closing price only). The next step after this is to make them adaptive, or self-adjusting (but this is a way's down the road).This is too long already!
dh |
