A Python tool for calculating the Hurst Exponent of financial time series — a statistical measure that reveals whether a market is trending, random, or mean-reverting. Built for systematic traders, options sellers, and anyone who wants to understand the structural character of a market before placing a trade.
Example output for USO, including price history, R/S, DFA, and rolling 63d/126d/252d Hurst series.
The Hurst Exponent (H) is a single number between 0 and 1 that describes the memory of a time series:
| H Value | Regime | What It Means |
|---|---|---|
| H < 0.45 | Mean-Reverting | Prices tend to revert after moves. Ideal for short volatility strategies. |
| 0.45 – 0.55 | Random Walk | No persistent structure. Neutral environment. |
| H > 0.55 | Trending | Prices tend to continue in the same direction. Favors trend following. |
Unlike a moving average — which tells you where a trend is going — the Hurst Exponent tells you whether trending behavior exists in the first place. It is a regime detector, not a directional signal.
pip install yfinance numpy pandas matplotlib scipyPython 3.8 or higher is required.
# Basic usage — 2-year default window
python src/hurst.py --ticker SPY
# Specify a period
python src/hurst.py --ticker USO --period 5y
# Specify exact date range
python src/hurst.py --ticker MA --start 2022-01-01 --end 2024-12-31
# Use only one calculation method
python src/hurst.py --ticker QQQ --period 3y --method dfa
# Run without chart output
python src/hurst.py --ticker GLD --period 2y --no-plot
# Disable rolling Hurst (enabled by default)
python src/hurst.py --ticker IWM --no-rolling| Argument | Default | Description |
|---|---|---|
--ticker |
(required) | Any ticker symbol supported by Yahoo Finance (e.g. SPY, USO, AAPL, GLD) |
--period |
2y |
Lookback period: 1mo 3mo 6mo 1y 2y 5y 10y ytd max |
--start |
— | Optional start date in YYYY-MM-DD format; used with --end |
--end |
— | Optional end date in YYYY-MM-DD format; used with --start |
--method |
both |
Calculation method: rs, dfa, or both |
--no-rolling |
off | Disable rolling Hurst calculation (enabled by default) |
--no-plot |
off | Suppress chart output, print results to terminal only |
--min-window |
10 |
Minimum segment size for the regression (advanced) |
The script calculates the Hurst Exponent using two independent methods and compares the results.
The classical method introduced by hydrologist Harold Hurst in 1951, later applied to financial markets by Benoit Mandelbrot.
Steps:
- Divide the log-return series into segments of varying length n
- For each segment, calculate the Range R (max cumulative deviation minus min) and the Standard Deviation S
- Compute the ratio R/S for each segment length
- Plot log(R/S) against log(n)
- The slope of the regression line is H
If the slope is greater than 0.5, the series has long-term memory — returns are positively autocorrelated and trends persist. If the slope is below 0.5, returns are negatively autocorrelated and the series mean-reverts.
A more modern and robust approach, better suited to non-stationary financial series.
Steps:
- Integrate the return series (cumulative sum of deviations from the mean)
- Divide into segments of length n
- For each segment, fit and remove a linear trend (DFA-1)
- Calculate the root-mean-square fluctuation F(n) of the residuals
- Plot log(F(n)) against log(n)
- The slope is H
DFA is less susceptible to short-term autocorrelation artifacts and tends to give more conservative estimates than R/S, which can overestimate H for short series.
In addition to the full-period calculation, the script computes rolling Hurst Exponents for 252-day, 126-day, and 63-day windows, each stepping forward every 21 days. This gives a long-, medium-, and shorter-term view of regime change, which is critical for understanding when a market changes character, not just what its average behavior looks like.
- Terminal report with H values, R², standard errors, and a plain-language interpretation
- 4-panel chart saved as
hurst_<TICKER>.png:- Price history
- R/S log-log regression plot
- DFA log-log regression plot
- Rolling Hurst over time for 252d, 126d, and 63d windows with trending/mean-reverting zones highlighted
The Hurst Exponent answers one of the most important pre-trade questions: is this underlying likely to stay within a range, or is it likely to trend away from my short strikes?
- H < 0.48 → favorable environment for short strangles, iron condors, and cash-secured puts. The underlying oscillates and tends to revert, which means collected premium has a higher probability of expiring worthless.
- H > 0.55 → elevated risk for short volatility structures. The market has momentum — one side of your strangle can go deep in-the-money and stay there.
A practical rule: check the 63-day rolling H before opening a new position. If H is above 0.52, consider widening your strikes, reducing position size, or waiting for the regime to normalize.
H > 0.55 is statistical confirmation that trend-following signals (moving average crossovers, breakouts, momentum) have structural support. The market is not in a random walk — persistence is measurable. This provides a principled basis for staying in winning trades longer and trusting the trend.
Use Hurst as a meta-filter before applying other signals:
IF Rolling H (63d) > 0.55:
Activate trend-following rules
Widen or avoid short-vol positions
IF Rolling H (63d) < 0.45:
Activate mean-reversion rules (short strangles, range trading)
Ignore trend signals — they will generate whipsaws
The rolling chart is particularly valuable. A sudden shift in H from below 0.5 to above 0.6 — as seen in USO during the 2026 Hormuz crisis — is an early warning that the market character has changed. Combined with IV Rank or VIX, it can help you recognize when a low-volatility range environment is transitioning into a trending regime before the damage is done.
The Hurst Exponent is a powerful concept, but it has real limitations in practice:
Statistical noise in short windows. For windows below 100 data points, the H estimate carries a standard error of ±0.10 to ±0.15. A 21-day rolling H reading of 0.60 could easily be 0.45 in reality. Use 63- or 126-day windows for more stable readings.
Estimation bias in R/S. The classical R/S method is known to produce upward-biased estimates for finite series — meaning it tends to report H > 0.5 even for purely random data. DFA corrects for much of this bias. When the two methods disagree significantly, trust DFA more.
Not a standalone trading signal. Using H alone to time entries and exits will likely underperform a simple moving average. Its value is as a regime classifier that improves the quality of other signals, not as a trigger itself.
Markets change. A market with H = 0.48 over five years can exhibit H = 0.80 for three months during a crisis. The rolling calculation exists precisely to capture this, but no static measure fully anticipates regime shifts.
The Hurst Exponent originates from Harold Edwin Hurst's 1951 study of the long-term storage capacity of the Nile River. Benoit Mandelbrot recognized its applicability to financial markets in the 1960s and 1970s, arguing that price series exhibit long-range dependence and fractal self-similarity — properties incompatible with the Efficient Market Hypothesis.
This is the core argument of Richard Brennan's The Fractals of Finance (2026): markets are not random machines but complex adaptive systems with structural memory. The Hurst Exponent is one of the clearest empirical signatures of that memory. An H consistently above 0.5 across dozens of futures markets and decades of data is not a coincidence — it is evidence that trends are a structural feature of how markets function, not a temporary anomaly.
MIT License. Free to use, modify, and distribute.