I took the following passage from Ernie Chan´s book:
The notion of scaling into a position with a mean-reverting strategy is familiar
to many traders. (Another name for it is averaging-in.) As the price (of an
asset, a spread, or a portfolio) deviates further and further from its mean,
the potential profit to be reaped from an eventual reversal is also increasing;
thus, it makes sense to increase the capital invested. This is exactly what
our linear mean-reversal strategy does. Note also that this type of scaling-in
strategies also scale out gradually: We do not have to wait until the price
reverts to its mean before taking profits. The advantage of being able to exit
whenever the price reverts by a small increment is that even if the price
series is not really stationary and therefore never really reverts to its mean,
we can still be profitable by constantly realizing small profits. An added benefit
is that if you are trading large sizes, scaling-in and -out will reduce the
market impact of the entry and exit trades. If we want to implement scaling in
using Bollinger bands, we can just have multiple entries and exits: for
example, entryZscore = 1, 2, 3, …, N and exitZscore = 0, 1, 2, …, N − 1. Of
course, N is another parameter to be optimized using a training data set.
