I think the reasoning for the use of ATR (which was introduced by traders after some years of using Donchian Channels) was to smooth out the variation in position sizes. This would make sense when you consider this strategy was developed for commodity futures and currencies which have prices that can cycle up and down for many years or decades without much of a dramatic major trend up or down like equities (which always have a very long upward trend). The trader using ATR to calculate position sizes trading EUR/USD would be using very similar position sizes year after year after year (over the last decade EUR/USD has largely traded less than 7% away from 1.3000).
Suppose, for example that you got a long entry signal in EUR/USD at 1.1100. At the time, the price range of the channel you are trading is 500 pips. Suppose you have a $5000 account and want to risk $50 (1% of equity) on the trade so you put on 1000 units (1000 units fixes a risk of .10 per pip and .10*500=$50).
Suppose you are filled long at 1.1100 but the price goes only to 1.1105 and then turns back south and stops you out. This quick whipsaw gives a full loss of $50. The price then continues south another 200 pips below your stop loss level and then bounces back up.
At that point, you would place your new long entry order just over 1.1105, but now the price range is 700 pips tall and you want to risk 1% of your new balance which is $4950. To risk $49.50 on this new 700 pip range you put on 707 units (707 units fixes risk of .0707 per pip and .0707*700=$49.49). Now you are entering at almost the same level with a 30% smaller position size.
So then suppose this second entry goes on to a profit. Now you have profited 30% less on an entry just 5 pips above your initial entry at 1.1100 simply because of that whipsaw.
An initial stop based on ATR will smooth that out, the size of the second trade would have been much more similar to the size of the first. The position size on the first would have been a bit smaller and the position size on the second may have been a bit bigger. So the use of an ATR stop is not only good to keep you from dropping your position size way down in a situation like that above, but also it is good to prevent you from putting on a position size that is too big when a price range gets really narrow. In fact, I have mentioned this many times, Jerry Parker uses a minimum ATR to prevent positions from going too big. The purpose of that is to protect the trader against slippage.
ATR stops will allow position sizes to vary but not to as great a degree as pure Donchian Channel stops. Generally traders who use Donchian Channels and ATR set their initial stop loss based on ATR. Then, once the channel stop becomes profitable, they use the channel stop to get out with a profit. Others use ATR for position size calculation and the Donchian Channel stop for an exit whether it is profitable or not.
To put this particular subject into perspective, it should not go unnoticed that the impact of these nuances is lessened considerably by system diversification (and also by market diversification). Suppose under the circumstances as outlined above you risked only 0.333% on each trade and simultaneously ran three systems on EUR/USD. The position that endured the whipsaw above would have been only a third the size and a second position using a longer channel would have had an even smaller size with a wider stop and a third position using an even longer channel would have had an even smaller size with an even wider stop. Suppose then that only the first system endured the whipsaw while the other two went on to profits. This would severely dampen the effect of the variation of position size caused by the use of pure Donchian Channel stops. In fact, it would dampen that effect more so than the use of an ATR stop. As far as I understand there are many successful trend following traders that use initial stops and position sizes based on ATR and use system diversification.
As for me, for now, I am going to focus more on system diversification and run pure Donchian Channel stops.
-Adrian
So my short answer is: system diversification.
