Hurst Exponent: Identify Whether to Trend-Follow or Mean-Revert

The Hurst Exponent does not tell you when to buy or sell. It tells you which type of strategy is mathematically valid in the current market regime — trending, mean-reverting, or random — before you commit to any indicator at all.

Section 1: Core Mechanics

The Hurst Exponent was developed by hydrologist Harold Edwin Hurst in 1951 while studying the Nile River's flood cycles. Benoit Mandelbrot introduced it to financial markets in the 1960s. It measures the long-range dependency of a time series — specifically whether the series has memory (persistence) or anti-memory (anti-persistence).

Three regimes:

• H > 0.5: Persistent (trending) — past trends tend to continue. Momentum strategies are valid.
• H = 0.5: Random walk — no predictable structure. No strategy has mathematical edge.
• H < 0.5: Anti-persistent (mean-reverting) — price movements tend to reverse. Mean reversion strategies are valid.

Method 1: Rescaled Range (R/S) Analysis

The classical calculation developed by Hurst himself.

Step 1: Divide the price series of length N into sub-periods of length n.

Step 2: For each sub-period, compute the range R (max minus min of cumulative deviations from mean) and the standard deviation S.

Step 3: Compute the ratio R/S for each sub-period.

Step 4: Average R/S across all sub-periods of length n.

Step 5: Repeat for multiple values of n (e.g., n = 8, 16, 32, 64, 128).

Step 6: Plot log(R/S) against log(n). The slope of the regression line is the Hurst Exponent H.

In log space:

Where H is the slope and C is a constant. Estimate H via linear regression of log(R/S) on log(n).

Method 2: Detrended Fluctuation Analysis (DFA)

DFA is more robust for financial time series because it explicitly removes polynomial trends before measuring fluctuation — reducing the risk of overestimating H in trending series.

Key DFA formula for fluctuation F at scale n:

Where is the cumulative sum of the return series and is the local polynomial fit within each window of size n. Plot log(F(n)) vs log(n) — slope = H.

Inputs

• Price series: Close prices (log returns for DFA; raw or log prices for R/S)
• Length: 500 bars minimum; 1,000+ bars preferred for R/S accuracy
• Sub-period sizes (R/S): Standard sequence: 8, 16, 32, 64, 128 (powers of 2)
• Rolling window: 252 bars (1 trading year) for a rolling H estimate

Parameters

Output Range and Meaning

H is always between 0 and 1:

• 0.55 to 1.0: Persistent — trend-following strategies (MA crossovers, momentum, breakouts)
• 0.45 to 0.55: Near-random — no reliable directional edge; reduce position sizes
• 0.0 to 0.45: Anti-persistent — mean reversion strategies (Z-Score, DPO, Bollinger Bands, ConnorsRSI)

Typical equity indices on daily data fall between 0.50 and 0.60 — slightly persistent overall. Individual stocks vary significantly.

Visual Behavior

Hurst does not plot as an intraday line overlay. It is computed as a single scalar value for a time window, then updated as the window rolls forward. On a chart, it appears as a slowly-moving indicator line between 0 and 1 with horizontal reference lines at 0.45 and 0.55.

Section 2: Interpretation & Signals

H Value Trading Map

Regime Switching Application

The most powerful application is rolling Hurst as a strategy selector:

1. Calculate H on a rolling 252-day window, updated daily
2. When H rises above 0.55: activate trend-following indicators (SMA crossovers, MACD, ADX)
3. When H falls below 0.45: activate mean reversion indicators (Z-Score, DPO, ConnorsRSI)
4. When H is between 0.45 and 0.55: reduce position sizes to 25-50% of normal — no clear edge

Divergence

Hurst does not produce price divergence. Instead, watch for H regime transitions: H crossing from above 0.55 to below 0.55 signals a shift from persistent to random/anti-persistent. This is not a buy/sell signal — it is a strategy-switch signal. Position size reduction at the transition is prudent.

Best Application

Hurst is not a standalone trading signal. Use it as a meta-indicator — a filter that decides which other indicators are valid in the current regime. Before applying any indicator from this course, calculate the rolling H on daily data and select the indicator family that matches the current H reading.

Chart Setup — Hurst Regime Identification

Section 3: Pass vs. Live — Real-Time Reliability

Hurst Exponent is a slowly-evolving indicator. On daily data with a 252-bar window, it updates once per day and changes slowly. This means it identifies regimes that have already been established — not emerging ones. It is best used weekly to assess whether the current regime justifies the strategy family you are using.

Do not use Hurst on intraday data. The computational requirements (500+ bars) mean intraday Hurst calculations are based on hours or minutes of data — statistically unreliable and practically useless for signal generation.

Section 4: Practical Use Cases

Real-world application: A systematic fund using rolling Hurst on SPX daily data 2010-2023 switched between a 3-SMA trend system (when H > 0.55) and a Z-Score(20) mean reversion system (when H < 0.45) with neutral positioning (H 0.45-0.55). Backtested Sharpe ratio was 1.4 vs 0.9 for trend-only or mean-reversion-only approaches. The regime-aware switching avoided the worst drawdowns of both systems by staying out of their worst regime.

Section 5: Pseudo Code

INPUT: close_prices[], method="RS", window=252, min_series=500

# METHOD 1: Rescaled Range (R/S) Analysis
PROCESS (R/S):
  Step 1: Compute log returns: returns[i] = log(close[i] / close[i-1])

  Step 2: Define sub-period sizes: n_values = [8, 16, 32, 64, 128]
           (all n must be <= len(returns) / 2)

  Step 3: For each n in n_values:
            rs_values = []
            for each non-overlapping sub-window of size n in returns:
                mean_r = mean(sub_window)
                deviations = cumulative_sum(sub_window - mean_r)
                R = max(deviations) - min(deviations)
                S = std(sub_window)
                if S > 0:
                    rs_values.append(R / S)
            avg_rs[n] = mean(rs_values)

  Step 4: Fit log(avg_rs) ~ H * log(n_values) via least-squares regression
           H = slope of the regression line

# METHOD 2: Detrended Fluctuation Analysis (DFA)
PROCESS (DFA):
  Step 1: Compute cumulative sum of log returns: Y[k] = sum(returns[0:k])
  Step 2: For each scale n in [8, 16, 32, 64, 128]:
            F_values = []
            for each segment of size n in Y:
                fit a polynomial (degree 1) to segment
                residuals = segment - polynomial_fit
                F_values.append(sqrt(mean(residuals^2)))
            F[n] = mean(F_values)
  Step 3: H = slope of log(F) vs log(n_values) via linear regression

OUTPUT:
  H — single float between 0 and 1
  interpretation: "TREND" if H > 0.55, "RANDOM" if 0.45 <= H <= 0.55, "MEAN_REVERT" if H < 0.45

# ROLLING HURST:
  For each day i starting at i = window:
    hurst[i] = calculate_hurst(close_prices[i - window : i], method)

EDGE CASES:
  - len(close_prices) < 500: raise warning "insufficient data for reliable Hurst estimate"
  - All returns identical (flat price): S = 0 in R/S — skip that sub-window
  - H estimate outside [0, 1]: data quality issue — check for price gaps or errors
  - Method="RS" with series < 128 bars: n_values cannot reach 128 — reduce to available powers of 2

Section 6: Parameters & Optimization

Standard Lookback Conventions

Parameter Impact

Market-Specific H Characteristics

• S&P 500 daily: H approximately 0.53-0.58 in bull markets — slight persistence
• Gold (XAU/USD) daily: H approximately 0.50-0.56 — near-random to slightly persistent
• Bitcoin daily: H approximately 0.57-0.68 in trending phases — strongly persistent when trending
• Forex (EUR/USD) daily: H approximately 0.48-0.54 — near-random, occasional mean reversion
• Individual large-cap stocks: High variance (0.40-0.70) depending on sector and news cycle

Section 7: Synergies & Conflicts

Section 8: Common Mistakes

Section 9: Cheat Sheet