Anchored RSI

A cumulative Relative Strength Index whose averaging begins at a runtime-chosen anchor bar instead of over a fixed Wilder period.

Quick reference

Field	Value
Family	Momentum Oscillators
Input type	f64 (single close)
Output type	f64
Output range	0.0 ..= 100.0
Default parameters	none — re-anchor at runtime with set_anchor()
Warmup period	2 (needs one price change)
Interpretation	RSI of the entire move since the anchor, not a rolling window.

Formula

For every bar since the current anchor:

gain_i = max(close_i - close_{i-1}, 0)
loss_i = max(close_{i-1} - close_i, 0)
RSI_t  = 100 - 100 / (1 + Σ gain_i / Σ loss_i)

The relative strength is the ratio of cumulative gains to cumulative losses since the anchor; the per-bar count cancels, so this equals avg_gain / avg_loss over the anchored window. Where Rsi applies Wilder's period-length smoothing (α = 1/period), Anchored RSI gives equal weight to every move since the anchor — so it answers "what is the RSI of the whole leg that started at my anchor?". See crates/wickra-core/src/indicators/anchored_rsi.rs.

Saturation follows the standard convention: a window with no losses yet (and at least one gain) reads 100, no gains yet reads 0, and a perfectly flat window reads the neutral 50 — identical to the convention used by Rsi.

Parameters

None. The averaging window is opened at construction (the first close is the anchor) and re-opened at runtime:

Method	Description
set_anchor()	Marks a re-anchor; the next update clears the running sums and previous close before folding in its own bar, starting a fresh anchored window. Mirrors AnchoredVwap's "click to anchor" workflow.

Inputs / Outputs

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

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

Python streams as float | None, batches a close column to a 1-D numpy.ndarray (NaN for warmup). Node streams as number | null, batches as Array<number> with NaN placeholders.

Warmup

AnchoredRsi::new().warmup_period() == 2. The first close only seeds the previous-close reference and returns None; the first value follows on the second close (one price change is enough). After a set_anchor(), the first bar of the new window likewise re-seeds and returns None. The unit test first_bar_seeds_and_returns_none pins this contract.

Edge cases

• Pure uptrend → 100. Monotonically rising closes have zero cumulative loss, so RSI saturates at 100. The unit test pure_uptrend_saturates_at_100 pins this.
• Pure downtrend → 0. Monotonically falling closes give zero cumulative gain → 0. Pinned by pure_downtrend_saturates_at_0.
• Flat window → 50. No movement at all reads the neutral 50. Pinned by flat_window_reads_50.
• Re-anchor. set_anchor() discards the old window; a downtrend that read 0 followed by a re-anchored uptrend reads a fresh 100. Pinned by set_anchor_clears_old_window.
• Non-finite inputs. NaN / ±inf are ignored: they do not advance the window and the last value is returned unchanged. Pinned by ignores_non_finite_input and non_finite_before_any_bar_returns_none.
• Reset. reset() clears the window and previous close; the next close re-seeds. Pinned by reset_clears_state.

Examples

Rust

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

fn main() {
    // prices 10 -> 11 (+1) -> 9 (-2) -> 12 (+3)
    let mut rsi = AnchoredRsi::new();
    let out: Vec<Option<f64>> = rsi.batch(&[10.0, 11.0, 9.0, 12.0]);
    println!("{:?}", out);
}

Output:

[None, Some(100.0), Some(33.333333333333336), Some(66.66666666666667)]

After bar 2 the cumulative gain/loss is 1 / 2 → 100 − 100/1.5 = 33.33; after bar 3 it is 4 / 2 → 100 − 100/3 = 66.67. This matches the cumulative_reference_values unit test.

Python

import wickra as ta

rsi = ta.AnchoredRSI()
for x in [10.0, 11.0, 9.0, 12.0]:
    print(x, '->', rsi.update(x))
print('value:', rsi.value)
rsi.set_anchor()  # next update opens a fresh anchored window

Output:

10.0 -> None
11.0 -> 100.0
9.0 -> 33.333333333333336
12.0 -> 66.66666666666667
value: 66.66666666666667

Node

const ta = require('wickra');
const rsi = new ta.AnchoredRSI();
for (const x of [10, 11, 9, 12]) {
  console.log(x, '->', rsi.update(x));
}
rsi.setAnchor();

Output:

10 -> null
11 -> 100
9 -> 33.333333333333336
12 -> 66.66666666666667

Streaming with a re-anchor

import wickra as ta

rsi = ta.AnchoredRSI()
for x in [20.0, 19.0, 18.0, 17.0]:  # downtrend -> 0
    rsi.update(x)
assert rsi.value == 0.0
rsi.set_anchor()
assert rsi.update(50.0) is None     # re-seeds
assert rsi.update(51.0) == 100.0    # fresh up-leg

Interpretation

Anchor the index at a structurally meaningful bar — a swing low, an earnings gap, a session open — and Anchored RSI reports the momentum of the whole leg since that point, with no period to tune. Because it is cumulative, it is steadier than a rolling Rsi deep into a trend (one more bar barely moves a long-running average), which makes divergence against the anchored leg easy to read. Use the rolling Rsi when you want a fixed-lookback oscillator; use Anchored RSI when the relevant question is "how strong is this move from this point".

Common pitfalls

• Expecting Wilder values. Anchored RSI is not Rsi with a reset — it uses simple cumulative averages, not Wilder's α = 1/period smoothing, and it has no period. Numbers will not match an Rsi(14) line.
• Forgetting the re-seed after set_anchor(). Because RSI needs a price change, the first bar after an anchor returns None; the first value of the new leg appears on the second bar, not the first (unlike AnchoredVwap, which emits immediately).

References

The Relative Strength Index is due to J. Welles Wilder Jr., New Concepts in Technical Trading Systems (1978). The "anchored" variant applies the same runtime-anchor idea popularised for VWAP by Brian Shannon to the RSI ratio.

See also

• Indicator-Rsi — the fixed-period Wilder RSI.
• Indicator-AnchoredVwap — the volume-weighted anchored sibling.
• Indicators-Overview — the full taxonomy.