Struct HurstExponent 

pub struct HurstExponent { /* private fields */ }

Hurst Exponent of the last period values, estimated by rescaled-range (R/S) analysis.

The classic Hurst-Mandelbrot estimator forms log-log pairs of (n, R(n)/S(n)) for several window lengths n and reports the slope of the least-squares fit. Wickra uses a streaming-friendly variant that partitions the trailing window into chunks of equal size, computes (R/S) for each chunk length, and fits a log-log line to the resulting points:

for each chunk size m ∈ {n/2, n/3, …, n/chunks}:
    mean_m   = (1/m) · Σ x_i               over the chunk
    dev_m_i  = (Σ_{j ≤ i} (x_j − mean_m))  // cumulative deviation
    R_m      = max(dev_m) − min(dev_m)
    S_m      = population_stddev(chunk)
    pair     = (log m, log(R_m / S_m))
H = slope of OLS line through the (log m, log(R/S)) points

The interpretation is unchanged from the textbook:

• H ≈ 0.5 → random walk; recent moves carry no information about future direction (the efficient-markets baseline).
• H > 0.5 → persistent / trending; up moves are likelier to be followed by more up moves.
• H < 0.5 → anti-persistent / mean-reverting; up moves tend to reverse.

Use it as a regime filter: trend-following strategies prefer H > 0.55; mean-reversion prefers H < 0.45. The output is clamped to [0, 1] to absorb degenerate fits on very small windows.

period must be at least 2 · chunks so every chunk has at least two points (otherwise its stddev is zero). A perfectly flat window has all R/S = 0 and the indicator returns 0.5 (random-walk baseline) to avoid divide-by-zero / log-zero failures.

Each update is O(period); the window is stored in a deque and the chunked R/S computation runs once per emission, not per input.

Example

use wickra_core::{HurstExponent, Indicator};

let mut indicator = HurstExponent::new(100, 4).unwrap();
let mut last = None;
for i in 0..200 {
    last = indicator.update(f64::from(i));
}
assert!(last.is_some());

Implementations

impl HurstExponent

pub fn new(period: usize, chunks: usize) -> Result<Self>

Construct a new Hurst Exponent over a window of period inputs, fitted across chunks log-log points.

chunks controls the number of R/S pairs that go into the slope fit; the typical value is 4 (the original Hurst paper used 5 — 9 points; smaller windows constrain the choice).

Errors

Returns Error::InvalidPeriod if chunks < 2 or period < 2 · chunks.

pub const fn period(&self) -> usize

Configured window period.

pub const fn chunks(&self) -> usize

Configured chunk count.

Trait Implementations

impl Clone for HurstExponent

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for HurstExponent

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Indicator for HurstExponent

type Input = f64

Type of one input data point (typically f64 for a price, or Candle / Tick).

type Output = f64

Type of one output value.

fn update(&mut self, value: f64) -> Option<f64>

Feed one new data point into the indicator and return the freshly computed output, or None if there is no value for this input.

fn reset(&mut self)

Reset all internal state, leaving the indicator equivalent to a freshly constructed instance with the same parameters.

fn warmup_period(&self) -> usize

Number of inputs required before the first non-None output can be produced.

fn is_ready(&self) -> bool

Whether the indicator has emitted at least one value since the last reset.

fn name(&self) -> &'static str

Stable, human-readable indicator name. Used by chaining and diagnostics.

fn batch_nan_into(&mut self, inputs: &[Self::Input], out: &mut [f64])

where Self::Input: Copy, Self::Output: Into<f64>,

Run the indicator over inputs, writing one output per input into the caller-owned out buffer (NaN where update returns None). It exists for every indicator whose output converts to an f64; the batch pipelines use it for the scalar f64 -> f64 ones.

fn batch_fast_into(&mut self, inputs: &[Self::Input], out: &mut [f64])

where Self::Input: Copy, Self::Output: Into<f64>,

Opt-in fast batch: like batch_nan_into, but an indicator with a vectorised kernel may reassociate its arithmetic to run it in SIMD lanes. Each value then agrees with the exact batch to within the tolerance the indicator documents (a few units in the last place), not bit for bit; warmup positions, NaN placement and the output length are identical. The kernels are deterministic: the same input produces the same bits on every platform, with or without SIMD hardware.