Average True Range - GV

Average True Range (ATR) is based on volatility like Bollinger Bands (BB), which finds the simple mean and adds and subtracts the standard deviation from the mean for a band.

Chaikin's Volatility and Vertical Horizontal Filter (VHF) have curves similar to ATR for a given range. While the former two cannot be straightaway used as stoploss bands, both BB and ATR can be, as they are real-time range measurements as against indicators of volatility.

A few of us may be acquainted with the details of ATR, but for those who are not, here it is: ATR is the mean/simple average of the maximum of the following range alternatives, for a period of time:

(i) range between the high and lows of the day;
(ii) range between yesterday's close and today's high (mostly in a rising mkt) or yesterday's close and today's low (mostly in a falling market).

To convert this in to a band, one needs to add or subtract it with a choice of either the close or the extremes. One would choose the close if he wants to play safe and is concerned that he might lose out from the gains; or one would choose from extremes if he wants to make sure that the stoploss does not him out of a trend.

Again one can play with the 'averaging number' - 14 bars or 20 bars - as well as the multiplication factor for the range - whether 1 ATR or 2 ATR (just like the number of standard deviations chosen in BB).

Effectively this throws up three variables for designing. No set of variables work for all markets or stocks. They could work for one range of historical-volatility (HV) patterns exhibited by the market, or stocks, during the testing period. Even if HV changes for the same underlying, performance deviates significantly from the anticipated. This could be a huge amount, since the stoploss could put you out of the trend, with no re-entry in sight. Hence it is likely that on some occasions you would have lost a sizeable part of the trend while taking in all expected whipsaws.

Now comes the worst. I am not well acquainted with foreign markets, for example the US where most designers come from. It is thus logical that they have tested primarily their own markets. Specifically, the US market does not throw up gaps like Indian market does. So, if the US market is as advanced as claimed, then they should be less manipulated than ours, and therefore a smoother market, in terms of absence of gaps.

The big problem with gaps for trading systems is that they simply massacre your volatility-based stoplosses. Imagine a gap of 100 points. The range the expands to 100 for a specific bar and hence the simple average of this range, goes up roughly by a factor of 5 (if you took the standard 20-bar as the range).

Now multiply it with the factor of 2 or 2.5, it becomes very large. If you are subtracting it from the extremes (hoping to avoid losing a trend, which you could if you subtract from close or mean, then the range - and hence the stoploss - really expands indeed. Due to simple averaging used, the adverse effect stays for 19 more bars during which the trend could change and you would have lost out on a good part of the gain. This was common in 2008.

Alternatively, after a gap, the price action could remain narrow for over an hour or two, which is not uncommon and towards the end of which your stoploss would have come to very close to the price action. Then the price action may enlarge into a (even) a slightly a larger trading range triggering your stoploss.

The problem is because of the narrow range after a steep gap-up/gap-down and simple averaging, the discussed enlargement of the trading range would not be large enough to cause a moving average convergence (Maco), but would have put you out of market - that is no Maco if you had adopted larger parameters for your MAs (and smaller parameters will by themselves whipsaw any number of times and throw up so many trades that you would have become the loverboy of your broker and the villain of your dealer).