mt-data-mcp/docs/SAMPLE-TRADE-ADVANCED.md

Advanced forecast-to-trade playbook

Audience: User

Builds on the basic sample trade (one barrier pair via forecast_barrier_prob) with regime filters, conformal intervals, HAR-RV volatility, Monte Carlo barrier grids, and disciplined risk/execution gates. Modular by design: run each block, inspect the output, and only continue when your own thresholds (calibrated via backtests) say so.

Not financial advice. Use a demo account for any execution steps. See TRADING_SAFETY.md.

Dense terms used below: BOCPD · HMM · HAR-RV · Conformal · Kelly · VaR · Edge

Related: Sample trade · CLI · Barriers · Regimes · Glossary

Assumptions

Item	Value
Symbol / timeframe	EURUSD / H1
Horizon	12 bars (~half day)
Interface	mtdata-cli <tool> ... --json

---

0) Safety and hygiene

• Skip high-impact news windows (for example ±60 minutes around CPI, NFP, FOMC).
• Enforce a daily loss cap (for example 1–2× average daily VaR) and a per-trade risk cap (for example 0.25–1.0% of equity).
• Consider a spread/liquidity filter (for example spread under 1.5× median) before entry.

---

1) Regime & Break Detection (Gatekeeper)

Detect structural breaks and label regimes so you avoid trading through hostile phases.

1.1 BOCPD change‑points (returns)

mtdata-cli regime_detect EURUSD --timeframe H1 --fetch-limit 1500 --method bocpd --threshold 0.6 --lookback 24 --json

• Gate: if transition_summary.max_transition_probability >= 0.6 → stand down or reduce size; retrain/recalibrate models.

1.2 HMM‑lite regimes (returns)

mtdata-cli regime_detect EURUSD --timeframe H1 --fetch-limit 1500 --method hmm --params "n_states=3" --lookback 300 --json

• Derive a simple regime tag: {trend‑lowvol, trend‑highvol, range} from state and state_probabilities.
• Gate: trade only when regime in {trend‑lowvol, trend‑midvol}; reduce risk in range/highvol.

Optional: MS‑AR(1) (statsmodels)

mtdata-cli regime_detect EURUSD --timeframe H1 --fetch-limit 1500 --method ms_ar --params "n_states=2 order=1" --json

---

2) Realized Volatility & Risk Budget (HAR‑RV)

Estimate daily realized variance from intraday returns, then map to H1.

mtdata-cli forecast_volatility_estimate EURUSD --timeframe H1 --horizon 12 --method har_rv --params "rv_timeframe=M5,days=150,window_w=5,window_m=22" --json

• Extract volatility_per_bar (per-bar sigma) and volatility_horizon (k-bar sigma).
• Risk budget: set per‑trade risk ≤ min(0.7× daily VaR, fixed cap). Use σ to set realistic TP/SL and lot size.

---

3) Denoise + Quick Technical Context

Pull data with light denoising and a few TIs for situational awareness (no heavy feature stacks in this flow).

mtdata-cli data_fetch_candles EURUSD --timeframe H1 --limit 300 --indicators "ema(20),ema(50),rsi(14),macd(12,26,9)" --denoise ema --denoise-params "columns=close,when=pre_ti,alpha=0.2,keep_original=true" --json

• Context: price vs EMA(20/50), RSI near extremes, MACD momentum slope.
• Gate: prefer longs if price > EMA(20)>EMA(50) and regime=trend.

---

4) Forecast with Valid Intervals (Conformal)

Calibrate per‑step residual quantiles via rolling backtest; then get point + conformal bands.

mtdata-cli forecast_conformal_intervals EURUSD --timeframe H1 --method fourier_ols --horizon 12 --steps 25 --spacing 12 --ci-alpha 0.1 --json

• Use lower_price/upper_price (conformal), not model CIs, for entry gating and sizing.
• Gate: only take longs if conformal.lower_price[h-1] > pivot and point forecast trend=up.

---

5) Barrier Analytics (MC + Closed‑Form)

5.1 Optimize TP/SL grid with HMM MC paths

mtdata-cli forecast_barrier_optimize EURUSD --timeframe H1 --horizon 12 --method hmm_mc --mode pct --grid-style volatility --params "n_sims=5000 seed=7 refine=true refine_radius=0.35 tp_min=0.25 tp_max=1.5 tp_steps=7 sl_min=0.25 sl_max=2.5 sl_steps=9 return_grid=false" --top-k 5 --json

• Choose a combo by objective (edge/kelly/ev/ev_cond/ev_per_bar/prob_resolve/profit_factor/min_loss_prob/utility) subject to constraints:
  • Use min_prob_win, max_prob_no_hit, and max_median_time (bars) to enforce hit-rate and timing limits.

5.2 TP/SL odds for the chosen combo

mtdata-cli forecast_barrier_prob EURUSD --timeframe H1 --horizon 12 --method hmm_mc --barrier '{"kind":"tp_sl","unit":"pct","take_profit":0.4,"stop_loss":0.8}' --params "n_sims=5000 seed=7" --json

5.3 Closed‑form GBM sanity check (fast)

mtdata-cli forecast_barrier_prob EURUSD --timeframe H1 --horizon 12 --method closed_form --direction long --barrier '{"kind":"single_price","level":1.1795}' --json

• Flag discrepancies (e.g., MC>>GBM) to reduce size or re‑check calibration.

---

6) Labeling & Threshold Calibration (Optional but Recommended)

Use triple‑barrier labels offline for signal evaluation and meta‑models.

mtdata-cli labels_triple_barrier EURUSD --timeframe H1 --lookback 2000 --horizon 12 --barrier '{"unit":"pct","take_profit":0.4,"stop_loss":0.8}' --label-on high_low --detail full --json

--lookback controls the labeled history; --limit only caps the compact or standard sample, while --detail full returns the complete labeled series.

• Compute in‑sample precision/recall for your entry rules; adjust thresholds (edge, cp_prob, RSI, EMA alignment) to reach desired trade quality.

---

7) Position Sizing & Execution

• Position size (conservative Kelly/VaR):
  • Kelly_cap = 0.25 × Kelly (from optimizer) or ≤ 1.0% equity, whichever is smaller.
  • VaR sizing: risk_to_SL ≤ risk_budget, where risk_to_SL uses spread‑adjusted SL and conformal lower.
• Stops & targets:
  • TP/SL from optimizer; time stop at horizon if neither is hit.
  • Optional partial TP at 0.5×TP; move stop to breakeven.
• Costs: subtract spread/commission from TP; inflate SL by typical slippage.
• If you convert a plan into a live order, start with trade_place --dry-run true and review the CLI trade controls (--require-sl-tp, --magic, --expiration) before removing the preview. The unprotected-position close fail-safe is always enabled.

---

8) Backtest & Walk‑Forward Checks

1. Rolling backtest for chosen forecast method(s)

mtdata-cli forecast_backtest_run EURUSD --timeframe H1 --horizon 12 --steps 50 --spacing 5 --methods "theta fourier_ols" --json

2. Stress‑test entry thresholds

• Sweep edge minima (0.05→0.15), cp_prob caps, regime sets, and confirm Sharpe/win rate stability.

---

9) Trade Plans (Examples)

Plan A – Breakout with pullback filter

• Gates: regime in {trend}, cp_prob<0.6, price>EMA(20)>EMA(50), conformal.lower>pivot.
• Targets: TP=0.40%, SL=0.80% (from optimizer); time stop at 12 bars.
• Size: min(VaR_budget, 0.25×Kelly) with HAR‑RV sigma.

Plan B – Mean‑reversion in range regime (reduced size)

• Gates: regime=range & low cp_prob; RSI>70 near R1 or <30 near S1.
• Smaller size; tighter SL and 0.20–0.30% TP; MC must show positive edge.

---

10) Monitoring & Drift Handling

• Update BOCPD and HMM twice per session; stand down on cp spikes.
• Refresh HAR‑RV daily (intraday M5 RV aggregation).
• Re‑calibrate conformal residuals weekly; re‑grid barrier optimizer monthly.
• Track realized vs. forecast errors; trigger re‑training on degradation.

---

TL;DR – Advanced Flow

1. Filter: regime & cp_prob gates.
2. Calibrate risk with HAR‑RV; set budget.
3. Denoise + quick TI context (EMA/RSI/MACD).
4. Forecast with conformal intervals; gate by bands.
5. Optimize TP/SL via MC; sanity‑check with GBM closed‑form.
6. (Optional) Label history with triple‑barrier; tune thresholds.
7. Execute with VaR/Kelly‑capped sizing, time stop, and costs.
8. Walk‑forward checks; adjust thresholds and monitoring cadence.

---

See Also

• SAMPLE-TRADE.md — Basic workflow (start here if new)
• FORECAST.md — Forecasting methods guide
• BARRIER_FUNCTIONS.md — Barrier analytics reference
• forecast/REGIMES.md — Regime detection details
• forecast/VOLATILITY.md — Volatility estimation methods
• forecast/UNCERTAINTY.md — Conformal intervals