Stochastic Methods Workflow

The Stochastic Methods tab is the model-driven cousin of Mean Reversion. Instead of tuning a threshold on an RSI, you fit a continuous-time model to the price series — Ornstein–Uhlenbeck, Kalman filter, Hurst, HMM — and trade when the model says the current state is extreme relative to its equilibrium. It is the tab we recommend for users who've backtested indicator rules and want a more statistically disciplined alternative.

What lives in this tab

One module, Stochastic Calculus Builder (SC001STCB), that hosts several methods:

• OU Z-Score (single-asset) — Fits an Ornstein–Uhlenbeck process to log-price. You trade when the z-score relative to the OU mean exceeds a threshold and exit when it reverts. The fit gives you a half-life — the expected time to revert — which doubles as a sanity check on whether OU is the right model for the series.
• OU Pairs / Kalman Pairs — Same idea applied to a hedged pair. OU uses a static hedge ratio; Kalman updates the hedge ratio dynamically. Pairs Scanner on the same page produces candidate pairs ranked by correlation, cointegration, OU half-life and β stability.
• MR Scanner — Ranks tickers by Hurst → OU half-life → ADF → KPSS to surface candidates that are statistically mean-reverting before you commit to a backtest.
• HMM regimes — Hidden Markov Model to classify market states; usable as a filter on top of the OU/Kalman trade signal.
• Grid Search + Walk-Forward — The two validation tools that matter most. Grid Search sweeps the parameter plane and shows a robustness surface; Walk-Forward runs multiple sequential IS/OOS windows and produces a verdict score, composite OOS equity and decay ratio.

Canonical workflow

1. Find a candidate. Single-asset users: run MR Scanner against your index — shortlist tickers that pass Hurst and have an OU half-life roughly in the range you want to trade (e.g. 10–60 bars). Pairs users: run Pairs Scanner on a basket and shortlist pairs with high correlation, cointegration, reasonable half-life and stable hedge ratio.
2. Configure the method. Pick OU Z-Score or Kalman Pairs, set the entry/exit z-thresholds, pick a lookback window for the parameter estimates, set direction, stops.
3. Run a single backtest to confirm the setup produces a reasonable trade count and equity curve.
4. Grid Search. Sweep two parameters (e.g. entry z-threshold and exit z-threshold, or half-life window and z-threshold). The output is a robustness surface. A robust strategy looks like a wide plateau on this surface — if the best point is a narrow spike surrounded by bad neighbours, you've found an overfit.
5. Walk-Forward. Click the → Walk-Forward button (now opens in a new tab so the backtest stays put). It imports your grid-search robust zone, locks the swept parameters to zone centres, and then runs multi-fold walk-forward by sweeping different parameters. This is the anti-overfitting validation — you're testing that the robust-zone choice holds up on pure OOS data across multiple windows.
6. Read the verdict. Walk-Forward returns a verdict score (0–100), composite OOS equity curve, per-fold parameter stability and a decay ratio (how much IS performance survives to OOS). A verdict score in the 60s+ on a composite OOS curve that looks like the IS curve is what you want.
7. Save the config. It lands in the Execution ribbon. Forward-test in Stochastic Paper Trading (PT005STPT).

Reading the outputs

• OU half-life — If the estimated half-life is either very short (1–2 bars) or longer than your holding horizon, OU is probably the wrong model for this series.
• Grid Search robustness surface — Look for a wide plateau, not a peak. The reported "robust zone" is the rectangle of parameters where behaviour is consistent.
• Walk-Forward verdict — The composite score reflects OOS-only performance across folds. Pay attention to the decay ratio too — a ratio close to 1.0 means IS performance translates to OOS; a ratio much below 0.5 is the smoke signal for overfitting.
• Per-fold parameter stability — If the optimal parameters jump around wildly between folds, the strategy is fitting noise even if the composite curve looks OK.

Common pitfalls

Deeper reading

• Trading Strategies → Mean Reversion Overview, Hurst Exponent, Half-Life, Variance Ratio — the math underneath MR Scanner and OU.
• Trading Strategies → Market Regime Detection for HMM and regime-aware filtering.