Improve AMA indicator parity with Cython

[nicolad]

Detailed comparison

Concern	Python / Cython (source of truth)	Rust (before)	Rust (after this PR)
Parameter validation	✔ Positive ints, slow > fast	✘	✔ identical asserts
count increments	✔	✘	✔ now increments + tested
prior_value reset	✔	✘ (left stale)	✔ cleared in reset
Comments accurate	n/a	✘ (misleading)	✔ fixed
Test coverage	full	partial	full parity

---

flowchart TD
    A[call<br>AdaptiveMovingAverage::new] --> B{Validate arguments}
    B -->|any zero| P1[[panic]]
    B -->|slow ≤ fast| P2[[panic]]
    B --> C[Construct fields<br>+ alpha_fast/slow<br>+ EfficiencyRatio::new]
    C --> D[[Ready instance]]

Loading

sequenceDiagram
    participant Caller
    participant AMA as AdaptiveMovingAverage
    participant ER  as EfficiencyRatio

    Caller ->> AMA: update_raw(v)
    AMA    ->> AMA: count += 1
    AMA    ->> AMA: first input?\n(no → skip init branch)
    AMA    ->> ER : update_raw(v)
    AMA    ->> AMA: sc = (ER.value · α_diff + α_slow)^2
    AMA    ->> AMA: value = prior + sc · (v - prior)
    AMA    ->> AMA: prior = value_old
    AMA    ->> AMA: initialized = ER.initialized()

Loading

stateDiagram-v2
    direction LR
    [*] --> Idle
    Idle   -->|first input| Primed
    Primed -->|ER window filled| Ready
    Ready  -->|reset()| Idle

Loading

---

Related to #2507

---

[nicolad]

Ten practical ways to put AdaptiveMovingAverage (AMA) to work

#	Use-case	Why AMA adds value	How you’d wire it in
1	Dynamic trend-following strategy	AMA hugs price when markets are quiet and backs off in noisy bursts, reducing whipsaws vs. fixed-length EMAs.	Feed live QuoteTicks; go long when price crosses above AMA, flat/short on cross-down.
2	Adaptive trade exit filter	Instead of a static stop, trail the AMA a few ticks behind the position—lag expands in choppy phases, avoiding premature exits.	Recompute AMA on every trade tick; update stop-loss to AMA(value) – k·ATR.
3	Signal confirmation layer	Pair with fast oscillators (RSI, Stoch). Trade only when oscillator crosses and price is also on the AMA’s “right” side.	Require RSI > 70 and price > AMA for shorts, the reverse for longs.
4	Volatility-weighted position sizing	AMA’s α adapts to Efficiency Ratio; its smoothing constant is a real-time proxy for noise. Larger dispersion ⇒ smaller size.	Compute sc = (ER·α_diff + α_slow)^2; scale order size by 1 / sc.
5	Regime detection (trend vs. range)	Flat AMA (low slope) + low ER → ranging; steep AMA + high ER → trending.	Monitor slope = (AMA_t – AMA_{t-1}) / price; switch strategies per regime.
6	Event-driven alpha tests	Measure how quickly AMA “snaps” after news releases—fast convergence implies strong consensus.	Timestamp news headline; record ticks until `
7	Portfolio rotation filter	On daily bars across a watch-list, rank symbols by distance above/below their AMA; rotate into top momentum names.	Batch-update AMA per symbol in a nightly job; sort by (close/AMA – 1).
8	Latency-robust back-tester	Because AMA embeds its own noise filter, it’s forgiving of sparse historical trade data—fewer look-back artifacts than HMA/TEMA.	Drive update_raw() with OHLC Bars instead of tick data.
9	Risk clock for market-making	Widen spreads when AMA widens (indicating volatility) and tighten when AMA compresses.	Compute σ of (price – AMA) over sliding window; map to spread tier.
10	Embedded metric in on-chain oracle	AMA’s low-noise property lends itself to DeFi price oracles that must resist manipulation spikes but stay reactive.	Run AMA in a Rust WASM module; publish AMA.value to the chain each block.

---

Implementation pointers

• Period selection – Common combos: (ER=10, fast=2, slow=30) intraday; (ER=20, fast=2, slow=20) daily. Tune with walk-forward.
• Feed choice – Pass PriceType::Last for mid-quotes; PriceType::Bid/Ask for aggressive market-making logic.
• Thread safety – AdaptiveMovingAverage is #[repr(C)] but not Send + Sync; wrap in Mutex for multi-threaded engines.
• Zero-state re-use – Call reset() between back-test runs to avoid cross-contamination of statistics.

[faysou]

Thanks, I like your practical ways table, I suppose it's possible to get similar tables from LLMs for other indicators, it's useful.

[cjdsellers]

Great work @nicolad!

Many thanks for the contribution 🙏, the chart and table are useful.

[cjdsellers]

👌