Skip to content

Doji

Patterns candlestick indecision reversal

A single-candle pattern in which the open and close are virtually equal, producing a very small body with upper and lower wicks. It signals momentary equilibrium between buyers and sellers and potential exhaustion of the prior directional move.

Visual Example

Doji candlestick showing a very small body (open nearly identical to close) with wicks extending in both directions. Annotation highlights the near-zero body size relative to the full range.

Synthetic ideal engineered so |open − close| ≪ (high − low). Matches the exact condition used by TA-Lib CDLDOJI / QuantWave CDLDOJI (and .ta.cdl_doji()). Generated 2026-05-31 IST via docs/gen_candle_previews.py.

Description

The Doji forms when buying and selling pressure are in near-perfect balance at a given price level. In the context of a strong uptrend or downtrend it frequently marks a point of indecision and the first sign that momentum may be fading. It does not, by itself, constitute a buy or sell signal; it is best interpreted as a warning that the prevailing control (bulls or bears) has temporarily loosened.

Practitioners incorporate Doji into confluence setups: after a confirmed Market Structure bias flip, alongside volume contraction/expansion, or as a sparse but high-information binary feature in ML pipelines for regime classification. Variants (Gravestone, Dragonfly, Long-Legged) add directional flavor to the indecision message.

Formula / Specification

Recognition Rules (exact implementation in QuantWave / TA-Lib CDLDOJI):

  1. For the current bar compute body length = |close − open|.
  2. Compute full range = high − low.
  3. The bar qualifies as a Doji when body length is negligible relative to range (TA-Lib reference implementation uses a small fractional tolerance derived from the series; the wrapper returns a non-zero signal when the condition holds).
  4. Output: typically +100 / −100 / 0 following TA-Lib candle pattern convention (context-dependent sign in some higher-level consumers).
  5. The streaming implementation (Next<(f64,f64,f64,f64)>) buffers the four price series and delegates the final test to the underlying talib_rs function. No additional state or smoothing is maintained across calls.

The pattern is stateless beyond the single bar under test.

Parameters

Parameter Default Description
(none) — Pattern recognition only; no tunable parameters. The CDLDOJI struct maintains internal OHLC history solely to satisfy the talib_rs API contract.

Usage Examples

Streaming (Rust)

use quantwave_core::indicators::CDLDOJI;
use quantwave_core::traits::Next;

let mut doji = CDLDOJI::new();
for (open, high, low, close) in &ohlcv_bars {
    let signal = doji.next((*open, *high, *low, *close));
    if signal != 0.0 {
        // Doji detected — combine with bias, volume, or structure
    }
}

Streaming (Python)

from quantwave import CDLDOJI

doji = CDLDOJI()
for o, h, l, c in ohlcv_bars:
    signal = doji.next((o, h, l, c))
    if signal != 0:
        ...

Polars Batch (Python — primary research / feature surface)

import polars as pl
import quantwave as qw

df = (
    pl.read_csv("ohlcv.csv")
    .lazy()
    .with_columns([
        pl.col("open").ta.cdl_doji("open", "high", "low", "close").alias("doji_signal"),
    ])
    .collect()
)
# Column contains +100 (bullish context), -100 (bearish), or 0

All surfaces are bit-identical (enforced by the universal Next<T> trait and proptests).

Edge Cases & Limitations

  • First bar of any series always returns 0 (insufficient history for the pattern test).
  • In a powerful trend a Doji is frequently a continuation warning rather than reversal; always require trend or structure context.
  • Near-zero volatility bars can trigger spurious Doji classifications — filter with ATR or range thresholds.
  • Literature tolerance for "virtually equal" varies (5–15 % of range); QuantWave follows the TA-Lib reference exactly.
  • High false-positive rate in choppy, low-volume markets without additional confirmation (volume spike, next-bar close direction, higher-timeframe bias).
  • Best paired with Market Structure or Ehlers regime tools rather than used in isolation.
  • No look-ahead bias.

Boundary Behavior

Condition Behavior
Warm-up Pattern functions emit 0 (no pattern) until enough bars exist.
period > len Short series returns all zeros (no pattern detected).
NaN inputs Bars with NaN OHLC are treated as no pattern (0).
Invalid params N/A for most candlestick patterns.
Empty data Empty input returns an empty integer series.

Sources & References

Primary Source: TA-Lib CDLDOJI specification and reference implementation (ta-lib.org), wrapped via talib_rs and quantwave-core/src/indicators/pattern.rs (the talib_cdl! macro).

Visual: Generated 2026-05-31 IST via docs/gen_candle_previews.py using synthetic data that satisfies the exact recognition rules used by the library.

Additional Context: Nison, Japanese Candlestick Charting Techniques (1991) for market-psychology interpretation (used only for descriptive depth; no duplicated boilerplate). MQL5 Price Action articles for practical confluence with structure detectors.

Implementation Provenance: Universal Next<T> contract and batch/streaming parity documented in quantwave-core/src/traits.rs and Agents.md.