On this page
The strategy autopilot described on this page is paused on real accounts for now. Your strategies keep running on the demo account in Replay and in backtests.
Where to find it
- Where
- Chart → Indicators panel → group « Quant models » → « Hurst MR (signals) »
- Default
- Off. Threshold 0.58, Hurst window 50 bars, Z lookback 19 bars, Entry K 2.75σ, timeout 120 bars
- Orders
- None. The overlay draws signals, levels and a status panel; it has no path to a broker
What it does
On every closed bar the overlay computes a Hurst exponent by the rescaled-range method on the log returns of the last « Hurst window » bars (12 log-spaced lags between 4 and min(window/2, 50), non-overlapping chunks, log-log regression). It also computes a z-score of the close against the mean and standard deviation of the « Z lookback » bars strictly before the current bar; the current bar is excluded from the statistics.
The regime is mean-reverting when H is below the Hurst threshold. In that regime a signal fires when |Z| ≥ K: Z > 0 marks a hypothetical short toward the mean, Z < 0 a hypothetical long. The suggested stop is max(SL min, SL × σ) capped at SL cap; the target is fair value ± TP overshoot × σ. A state machine follows the hypothetical trade bar by bar: stop first, then target, then timeout, then the MR/Trend trail, which moves the stop to fair value once price crosses it while H exceeds the trail threshold. After an exit, two bars of cooldown pass before a new signal.
The forming bar never triggers a decision; it only feeds the live values (H, Z, bands) shown in the status panel. This is the on-bar-close semantics of the original NinjaTrader strategy the overlay ports.
Settings
| Setting | Default | What it changes |
|---|---|---|
| Hurst threshold | 0.58 (0.30–0.70) | H below this value = mean-reverting regime |
| Hurst window | 50 bars (20–200) | Rolling window of the R/S estimate |
| Z lookback | 19 bars (10–100) | Rolling mean / std window of the z-score, current bar excluded |
| Entry K (σ) | 2.75 (1.0–5.0) | |Z| needed to mark a signal; also the width of the ±Kσ bands |
| SL × σ | 0.65 (0.1–3.0) | Stop distance as a multiple of σ |
| SL min / SL cap | 5 pts / 20 pts | Floor and ceiling of the stop distance in price points |
| TP overshoot | 0.15 σ (0–2) | Target beyond fair value; 0 = the mean itself |
| MR/Trend trail | On | Moves the stop to fair value when price crosses it in a trending regime |
| Trail H thresh | 0.51 (0.40–0.60) | H above this value counts as trending for the trail and its exits |
| Timeout | 120 bars (15–300) | Mark-to-market exit after N bars |
| ±Kσ bands / Status panel / Trade levels | On / On / On | Display toggles for the bands, the live panel and the entry, stop and target lines |
How to use it
The defaults are the « champion v9 » configuration validated on 1-minute bars; changing the timeframe changes what 50 and 19 bars mean, so re-read the windows when you switch. Use the status panel to see H, Z, the current regime and the stop that would apply if a signal fired now.
Bands and levels are drawn as polylines cut at session gaps: an interval more than three times the median bar step is treated as a gap, so a Friday-to-Sunday break does not produce a line across the chart.
Limits and pitfalls
- Signals are hypothetical. Nothing is routed; the sizing, daily-loss and forced-flat logic of the original strategy are deliberately not ported.
- The Hurst estimate is single-scale and unstable on short windows; it does not establish an edge.
- Stop and target fills are simulated intrabar on high/low, which is optimistic compared with real fills.
Related pages
In the same section
- Imbalances
- Indicator guide cards
- History loading
- Indicators panel
- Hedging pressure
- Advanced averages and trend
- Heatmap profile window
- Buy/sell flow
This page in other languages
Frequently asked questions
- Can Hurst MR place trades automatically?
- No. The overlay is described in the panel as « signals only, never orders ». Automated execution exists only through the separately armed autopilot, which is never persisted across sessions.
- Why is the current bar excluded from the z-score statistics?
- To match the backtest the strategy was validated with: the mean and standard deviation use the lookback bars before the bar being evaluated, so the bar cannot influence its own reference.
- What are the exit reasons shown on the chart?
- sl, tp, timeout, trail_z3 (opposite Z ≥ 3.0 while trailing) and trail_hz (trending regime with opposite Z ≥ 2.5, or the trailed stop hit).