Basic Anchored Oscillators
To take a brief but related digression from the Adaptive Time Machine, I would like to discuss the importance of anchored versus rolling oscillators. Almost all conventional indicators are rolling–that is for example an rsi2 only looks at the current position of the market over the last two days without regard to what has happened before that. In relation to the different runs based entries and exits, it is not so obvious that a strategy such as entering on two days down and exiting on two days up could actually last for two weeks or more. This is because an up day that fails to follow through can occur numerous times prior to exit. Furthermore, the entry stipulation of a down run of two days can also occur under similar circumstances. A down run could also occur after two very large down days, or perhaps two small down days. Clearly not all up or down runs are made equal.
In fact I would argue that one of the main reasons that traders sometimes “disagree” with what an indicator is saying is precisely for this same reason: often the current level fails to take into account what appears to be common sense to the trader. What we really want to do is to create an indicator “with a memory” and also use different methods of normalization in order to avoid such situations. The Runs Oscillator https://cssanalytics.wordpress.com/2010/07/26/runs-oscillator/ addressed the issue of normalizing run length, but being a rolling indicator, it fails to address some of the problems in the examples above. Rolling indicators make life difficult for adaptive systems as well because we typically utilize some performance selection criteria in order to identify the most favorable situations. If the comparable situation is not like past situations, then we may do better or worse than what was observed in the past. In a sense, this contributes to the noise involved with classification.
A basic anchored indicator will have both a trigger point as well as an exit point. Once a situation is triggered, we measure the distance traveled in terms of both time and price until an exit occurs. A simple example pertains to daily mean reversion as defined in this recent post by MarketSci: http://marketsci.wordpress.com/2010/08/01/the-new-state-of-short-term-mean-reversion-july-2010/ as buying on down days and shorting on up days. In this case for longs, the trigger would be one down day and the exit would be one up day. Once a trade was initiated (on one down day) we would measure the distance traveled since entry in % terms or atr, and also the days that have elapsed- in this case a down run- until an up day occurs (the exit). This method of combined analysis that looks at magnitude (price) and stretch (time) underlies the DVI – which is also a rolling indicator. This is what makes the DVI so useful, but also- excepting the intentional smoothing- somewhat ineffective at identifying turning points. In this case, a complex anchored runs oscillator would consider also the distance travelled either during the trigger condition, following the trigger condition, the exit condition or all three. There is a lot in that last sentence to think about, in a subsequent post we will provide an example.