MacdFix

MACD Fix (MACDFIX) — the classic MACD with the fast and slow EMAs fixed at 12 and 26, leaving only the signal period configurable. Following TA-Lib, the two EMAs smooth with the fixed constants 0.15 / 0.075 rather than 2/13 / 2/27.

Quick reference

Field	Value
Family	Trend & Directional
Input type	f64 (single close)
Output type	MacdOutput { macd, signal, histogram }
Output range	unbounded; histogram = macd − signal
Default parameters	signal is required (fast/slow fixed at 12/26)
Warmup period	same as MacdIndicator::new(12, 26, signal) (26 + signal − 1, 34 for signal = 9)
Interpretation	TA-Lib's fixed-constant 12/26 MACD; only the signal smoothing is tunable.

Formula

fast      = EMA(close; seed SMA(12), α = 0.15)
slow      = EMA(close; seed SMA(26), α = 0.075)
macd      = fast − slow
signal    = EMA(macd, signal_period)          // ordinary α = 2 / (signal_period + 1)
histogram = macd − signal

This is TA-Lib's MACDFIX: the 12- and 26-bar EMAs use Gerald Appel's rounded smoothing constants 0.15 and 0.075 instead of the period-derived 2/13 ≈ 0.1538 and 2/27 ≈ 0.0741 (both still seeded with their simple means). MacdFix is therefore not identical to MacdIndicator::new(12, 26, signal) — the values differ slightly, though the warmup and output shape are the same. The signal line is an ordinary signal-period EMA of the MACD line. The output is the usual MacdOutput triple. See crates/wickra-core/src/indicators/macd_fix.rs.

TA-Lib parity. MacdFix(9) uses the same constants and emits its first row on the same bar (33) as TA-Lib MACDFIX(9), and is identical to it (to 1e-9) from bar 164 on in the TA-Lib reference suite. The early rows differ only by the EMA seed: Wickra seeds each EMA with the SMA of its own first window (the fast EMA from bars 0–11), while TA-Lib delays the fast EMA's seed so it ends on the slow EMA's first bar; the gap decays by a factor 0.85 per bar.

Parameters

Name	Type	Default	Valid range	Description	Source
signal	usize	none	>= 1	Signal-line EMA period (the classic value is 9). signal = 0 errors with Error::PeriodZero.	macd_fix.rs:45

The fast (12) and slow (26) EMA periods are fixed and not configurable; use MacdIndicator if you need to change them.

Inputs / Outputs

From crates/wickra-core/src/indicators/macd_fix.rs:

use wickra::{Indicator, MacdFix};
use wickra::MacdOutput;
// MacdFix: Input = f64, Output = MacdOutput
const _: fn(&mut MacdFix, f64) -> Option<MacdOutput> = <MacdFix as Indicator>::update;

In Python update returns a (macd, signal, histogram) tuple (or None during warmup) and batch returns an (n, 3) Matrix. In Node update returns a { macd, signal, histogram } object and batch a flat Array<number> of length n · 3.

Beside the generic batch, MacdFix carries MACD's fused flat batches in Rust: batch_macd / batch_macd_into (exact, [macd, signal, histogram] per row, NaN warmup rows, bit-identical to streaming) and batch_macd_fast / batch_macd_fast_into (the opt-in SIMD kernel, agreeing to within a few ULP). The _into forms panic unless out.len() == inputs.len() * 3. Every binding exposes them, exactly as for MACD:

Language	Exact batch	Fast batch
Python	MACDFIX.batch → (n, 3) Matrix	MACDFIX.batch_fast → (n, 3) Matrix
Node	batch → flat Array<number>; batchInto(prices, out)	batchFast → flat Float64Array; batchFastInto(prices, out)
WASM	batch → flat Float64Array; batchInto(prices, out)	batchFast → flat Float64Array; batchFastInto(prices, out)
C ABI	wickra_macd_fix_batch	wickra_macd_fix_batch_fast
C# / Go	Batch	BatchFast
Java	batch / batchInto	batchFast / batchFastInto
R	batch()	batch_fast()

The Node and WASM flat layouts are [macd0, signal0, histogram0, macd1, ...]; the Into forms need an output of length 3 · n. See Streaming vs Batch for the fast-batch contract.

Warmup

MacdFix::new(signal).warmup_period() equals MacdIndicator::new(12, 26, signal).warmup_period() — the slow EMA must seed and then the signal EMA must seed on top of it. The unit test accessors_report_config pins this equality directly.

Edge cases

• Fixed smoothing constants. MacdFix::new(s)'s MACD line equals an SMA-seeded α = 0.15 EMA minus an SMA-seeded α = 0.075 EMA, and differs from MacdIndicator::new(12, 26, s) on the same series. The unit test uses_the_fixed_smoothing_constants pins both.
• Flat batch. batch_macd matches streaming bit for bit (flat_batch_matches_streaming).
• Zero signal. MacdFix::new(0) returns Err(Error::PeriodZero). The unit test rejects_zero_signal pins this.
• Constant series. Once warmed, a flat price gives macd = 0, signal = 0, histogram = 0 (both EMAs converge to the constant). See the example below.
• Reset. m.reset() clears all three EMA states. The unit test reset_clears_state pins this.

Examples

Rust

use wickra::{BatchExt, Indicator, MacdFix};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut m = MacdFix::new(9)?;
    let out = m.batch(&[100.0; 60]);
    println!("{:?}", out.last().unwrap());
    Ok(())
}

Output:

Some(MacdOutput { macd: 0.0, signal: 0.0, histogram: 0.0 })

On a constant series both the 12- and 26-period EMAs converge to 100, so macd = 0, the signal EMA of zero is 0, and histogram = 0. On real data the three lines diverge; the histogram = macd − signal identity always holds.

Python

import numpy as np
import wickra as ta

m = ta.MACDFIX(9)
out = m.batch(np.full(60, 100.0))   # (60, 3) Matrix
print(out[-1])   # last row: macd, signal, histogram

Output:

array('d', [0.0, 0.0, 0.0])

The fast batch has the same shape and agrees with batch to within a few units in the last place:

import math
import wickra as ta

x = [100 + 5 * math.sin(0.3 * i) for i in range(80)]
exact = ta.MACDFIX(9).batch(x)
fast = ta.MACDFIX(9).batch_fast(x)
print([round(v, 6) for v in exact[-1]])
print([round(v, 6) for v in fast[-1]])

Output:

[-1.039229, -0.18508, -0.854149]
[-1.039229, -0.18508, -0.854149]

Node

const ta = require('wickra');
const m = new ta.MACDFIX(9);
let last = null;
for (let i = 0; i < 60; i++) last = m.update(100);
console.log(last);

Output:

{ macd: 0, signal: 0, histogram: 0 }

The flat batches write [macd, signal, histogram] per row:

const ta = require('wickra');
const x = Float64Array.from({ length: 80 }, (_, i) => 100 + 5 * Math.sin(0.3 * i));
const out = new Float64Array(x.length * 3);
new ta.MACDFIX(9).batchInto(x, out);              // exact, into a reused buffer
console.log(Array.from(out.slice(-3), (v) => +v.toFixed(6)));
const fast = new ta.MACDFIX(9).batchFast(x);      // Float64Array, NaN during warmup
console.log(Array.from(fast.slice(-3), (v) => +v.toFixed(6)));

Output:

[ -1.039229, -0.18508, -0.854149 ]
[ -1.039229, -0.18508, -0.854149 ]

Interpretation

MacdFix is TA-Lib's fixed-constant MACD(12, 26, 9) momentum oscillator. Read it the standard way: macd crossing its signal line is the primary trade trigger, the histogram (their difference) shows momentum building or fading, and macd crossing zero marks the fast/slow EMA crossover. Use MacdFix when you need to reproduce TA-Lib's MACDFIX; use MacdIndicator when you need non-standard fast/slow lengths.

Common pitfalls

• Expecting to change fast/slow. They are fixed at 12/26 by design; reach for MacdIndicator to vary them.
• Treating it as an alias of MACD(12, 26, 9). The 0.15 / 0.075 constants make the values differ slightly from MacdIndicator::new(12, 26, 9); on 100 + 5·sin(0.3·i) for 80 bars the last MACD value is -1.0392 for MacdFix versus -1.1066 for MacdIndicator.
• Comparing histograms across instruments. MACD is in price units, so its scale depends on the instrument; normalise (e.g. by ATR) before comparing.

References

Gerald Appel's MACD (1979); the fixed-12/26 packaging and the 0.15 / 0.075 smoothing constants match TA-Lib's MACDFIX.

See also

• Indicator-MacdIndicator — the fully configurable MACD.
• Indicator-MacdExt — MACD with selectable moving-average types.
• Indicators-Overview — the full taxonomy.