Connors RSI (CRSI)

Larry Connors' three-component composite RSI. Averages a fast RSI on close, an RSI on the up/down streak run-length, and a percent rank of the 1-period return — all individually bounded in [0, 100], so the composite is too. Designed for short-term mean-reversion: Connors' rule-of-thumb is CRSI < 5 is oversold and CRSI > 95 is overbought, both rare conditions on the recommended (3, 2, 100) configuration.

Quick reference

Item	Value
Family	Momentum Oscillators
Input type	f64 (close)
Output type	f64
Output range	[0, 100]
Default parameters	period_rsi = 3, period_streak = 2, period_rank = 100 (ConnorsRsi::classic())
Warmup period	max(period_rsi + 1, period_streak + 2, period_rank + 1) — 101 for classic
Interpretation	< 5 oversold, > 95 overbought; cross of 50 is the mid-bias flip

Formula

Three components, each bounded in [0, 100]:

1. close_rsi_t    = RSI(close, period_rsi)_t

2. streak_t       = +1, +2, ...  for runs of consecutive up closes
                    -1, -2, ...  for runs of consecutive down closes
                    0            on an unchanged close
   streak_rsi_t   = RSI(streak, period_streak)_t

3. roc1_t         = (close_t − close_{t−1}) / close_{t−1}
   percent_rank_t = 100 * count(r in last period_rank rocs where r < roc1_t)
                    / period_rank

CRSI_t = (close_rsi_t + streak_rsi_t + percent_rank_t) / 3

The streak counter resets to 0 on an unchanged close. The ROC window is skipped on bars where prev_close == 0.0 (return undefined). See crates/wickra-core/src/indicators/connors_rsi.rs:89-135.

Parameters

Name	Type	Default	Constraint	Description
period_rsi	usize	3	> 0	Period of the close-based RSI component.
period_streak	usize	2	> 0	Period of the streak-based RSI component.
period_rank	usize	100	> 0	Lookback for the 1-period return percent rank.

ConnorsRsi::new returns Error::PeriodZero if any of the three periods is zero. ConnorsRsi::classic() returns the (3, 2, 100) factory (connors_rsi.rs:74-77).

Inputs / Outputs

Indicator<Input = f64, Output = f64>. Python: ConnorsRSI.batch(prices) returns a 1-D np.ndarray with NaN in the warmup prefix. Node: ConnorsRSI.batch(prices) returns Array<number> (NaN in warmup slots); update(value) returns number | null.

Warmup

warmup_period() == max(period_rsi + 1, period_streak + 2, period_rank + 1).

• The close-RSI needs period_rsi + 1 prices to seed.
• The streak-RSI needs period_streak + 1 streak values, each of which itself requires a prior price — so period_streak + 2 prices.
• The percent-rank branch needs period_rank + 1 prices (period_rank 1-period returns).

For classic (3, 2, 100) parameters the rank branch dominates at 101. Pinned by accessors_and_metadata (connors_rsi.rs:181-187) and warmup_emits_first_value_at_warmup_period (connors_rsi.rs:195-206).

Edge cases

• Unchanged close. Streak resets to 0; pinned by streak_resets_to_zero_on_unchanged_close (connors_rsi.rs:241-254).
• Non-finite input. NaN / ±∞ are dropped — the indicator returns its previous value without advancing state. Pinned by ignores_non_finite_input.
• prev_close == 0.0. The 1-period return is undefined; the ROC window is left unchanged on that bar. Pinned by zero_prev_skips_roc_update.
• Output bounded [0, 100]. Each component is in [0, 100], so the average must be too. Pinned by output_is_bounded on a 300-bar sine wave.
• Pure uptrend. All three components are pushed toward 100; the composite stays well above 50 (pure_uptrend_saturates_high asserts > 60 on a monotonic 1..=200 ramp with classic params).
• Reset. reset() clears both inner RSIs, the streak counter, the previous price, the ROC window, and the cached output.

Examples

Rust

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

fn main() {
    let prices: Vec<f64> = (1..=200)
        .map(|i| 100.0 + (f64::from(i) * 0.2).sin() * 5.0 + f64::from(i) * 0.1)
        .collect();
    let mut crsi = ConnorsRsi::classic();
    let out = crsi.batch(&prices);
    let v = out[120].unwrap();
    println!("row 120 CRSI = {v:.4}");
}

Python

import numpy as np
import wickra as ta

prices = 100 + np.sin(np.linspace(0, 30, 200)) * 5 + np.arange(200) * 0.1
crsi = ta.ConnorsRSI(3, 2, 100)
out = crsi.batch(prices)
print('warmup:', crsi.warmup_period())  # 101
print('row 120:', out[120])

Node

const wickra = require('wickra');

const crsi = new wickra.ConnorsRSI(3, 2, 100);
const prices = Array.from({ length: 200 },
  (_, i) => 100 + Math.sin(i * 0.2) * 5 + i * 0.1);
const out = crsi.batch(prices);
console.log('row 120:', out[120]);

Streaming

use wickra::{ConnorsRsi, Indicator};

let mut crsi = ConnorsRsi::classic();
let price_stream: Vec<f64> = Vec::new(); // your live price feed
for bar in price_stream {
    if let Some(v) = crsi.update(bar) {
        if v < 5.0 { /* extreme oversold — Connors fade-the-dip setup */ }
        if v > 95.0 { /* extreme overbought */ }
    }
}

Interpretation

CRSI is a short-term mean-reversion oscillator. Connors' canonical use:

• CRSI < 5. Enter long with a price-confirmation trigger (e.g. close above the prior bar's high), exit when CRSI > 50 or on a fixed N-day timeout.
• CRSI > 95. Symmetric short setup, though much less common in bull-biased equity tape.

Because each component is itself a mean-reversion-flavoured measure, CRSI extremes are rarer than RSI extremes. A standard 2-period RSI hits < 5 reasonably often; CRSI < 5 requires all three components to be near zero simultaneously, which is a much stronger condition.

Common pitfalls

• Treating CRSI like RSI. Same scale, very different distribution. CRSI 30 is not "oversold" — it's roughly average. Calibrate thresholds against the historical CRSI distribution of the instrument you're trading.
• Insufficient warmup. The default period_rank = 100 means the first 101 bars return no value. Short backtests on small data windows will look entirely like warmup.
• Streak-reset surprise. A single unchanged close resets the streak to 0, which sharply drops the streak-RSI component. On thinly-traded instruments with frequent flat closes this can generate spurious CRSI swings.

References

• Larry Connors & Cesar Alvarez, Short Term Trading Strategies That Work (2009) — the original introduction of CRSI as a refinement over RSI(2).
• Connors Research, An Introduction to ConnorsRSI (2012, white paper) — formal definition.

See also

• Rsi — the underlying building block.
• LaguerreRsi — Ehlers' alternative smoothing-based RSI variant.
• StochRsi — another RSI composite that rescales RSI through a stochastic.
• Indicators-Overview — full taxonomy.