RTTADocumentation Built by Goblin Reactor ↗

DirectionalChangeDetector

Incremental, causal technical analysis documentation

Summary

DirectionalChangeDetector samples intrinsic time via directional-change (DC) events: a move of relative size \(\theta\) from the last extremum defines a DC; the path beyond that event is reported as overshoot. Outputs include event flag, overshoot, current extremum, and trend mode.

Update API

import rtta

ind = rtta.DirectionalChangeDetector(threshold=0.01)  # 1% relative
result = ind.update(price)
# result.event ∈ {-1, 0, +1},
# result.overshoot, result.extremum, result.direction

threshold is a relative fraction (\(0.01 = 1\%\)). direction / mode is \(+1\) in an uptrend (seeking a downturn), \(-1\) in a downtrend (seeking an upturn), \(0\) until the first DC.

Theory Of Operation

Directional-change methods (Glattfelder, Dupuis, Olsen and related intrinsic-time work) replace calendar sampling with event sampling: time advances when price has moved by a fixed relative amount from a local extreme. After a DC up, the algorithm tracks the new high as extremum until a \(\theta\) drop; after a DC down, it tracks lows until a \(\theta\) rise. Overshoot measures how far price has continued beyond the last DC price in the current mode.

Recurrence

Let \(\theta>0\) denote threshold. On the first price \(p_0\): extremum \(E = p_0\), last DC price \(p^{\mathrm{dc}} = p_0\), mode \(m=0\), event \(0\).

Bootstrap (\(m=0\)): retain both the highest and lowest prices seen since initialization, \(B^H\) and \(B^L\):

\[\begin{aligned} p_t \ge B^L(1+\theta) &\Rightarrow m\leftarrow +1,\ \mathrm{event}\leftarrow +1,\ p^{\mathrm{dc}},E\leftarrow p_t,\\ p_t \le B^H(1-\theta) &\Rightarrow m\leftarrow -1,\ \mathrm{event}\leftarrow -1,\ p^{\mathrm{dc}},E\leftarrow p_t,\\ \text{else} &\Rightarrow B^H\leftarrow\max(B^H,p_t),\quad B^L\leftarrow\min(B^L,p_t). \end{aligned}\]

Keeping both bootstrap extrema allows several sub-threshold moves to add up to the first event. Replacing one anchor with every incoming price would detect only a one-tick move of size \(\theta\).

Uptrend (\(m=+1\)): \(\theta\)0. If \(\theta\)1, set \(\theta\)2, \(\theta\)3, \(\theta\)4.

Downtrend (\(\theta\)5): \(\theta\)6. If \(\theta\)7, set \(\theta\)8, \(\theta\)9, \(0.01 = 1\%\)0.

Overshoot (when \(0.01 = 1\%\)1):

\[\mathrm{overshoot}_t = \begin{cases} (p_t - p^{\mathrm{dc}})/p^{\mathrm{dc}}, & m=+1,\\ (p^{\mathrm{dc}} - p_t)/p^{\mathrm{dc}}, & m=-1,\\ 0, & m=0. \end{cases}\]

Outputs: \(0.01 = 1\%\)2, \(0.01 = 1\%\)3.

Implementation Notes

The recurrence is implemented in src/rtta/indicator.cpp in class DirectionalChangeDetector. Result type is DirectionalChangeResult.

Reference