Skip to content

EXAMPLE.md

Latest commit

 

History

History
254 lines (165 loc) · 7.33 KB

EXAMPLE.md

File metadata and controls

254 lines (165 loc) · 7.33 KB
Raw

End-to-end research loop

Audience: User

A compact command recipe for the full mtdata research loop: discover methods → fetch data → add context → forecast → size risk → validate. Prefer this when you already know the CLI; if you want narrative “why each step,” start with SAMPLE-TRADE.md.

Related: Setup · CLI · Sample trade · Forecasting

Assumptions

  • MetaTrader 5 is running and logged in
  • You call tools via mtdata-cli <command> ...
  • Swap EURUSD / H1 for your market

Tip: add --json when you want structured output for scripts or agents.

1) Discover what’s available

Forecast methods + availability (optional libraries show up here):

mtdata-cli forecast_list_methods --json

Forecast model spaces:

mtdata-cli forecast_list_library_models native --json
mtdata-cli forecast_list_library_models statsforecast --json
mtdata-cli forecast_list_library_models sktime --json
mtdata-cli forecast_list_library_models pretrained --json

Indicators:

mtdata-cli indicators_list --limit 50
mtdata-cli indicators_describe rsi --json

2) Inspect symbol and fetch candles

mtdata-cli symbols_list --limit 20
mtdata-cli symbols_describe EURUSD --json
mtdata-cli data_fetch_candles EURUSD --timeframe H1 --limit 300 --json

Optional liquidity snapshot. This requires MTDATA_ENABLE_MARKET_DEPTH_FETCH=1 and broker support for Level 2/DOM data:

mtdata-cli market_depth_fetch EURUSD --spread true --json

--spread is a Boolean output switch. It computes spread prices in symbol price units and, where metadata permits, also reports points, pips, and basis points from the DOM or fallback bid/ask quote.

3) Prepare data: indicators + denoise

Compute a few indicators:

mtdata-cli data_fetch_candles EURUSD --timeframe H1 --limit 1500 --indicators "ema(20),ema(50),rsi(14),macd(12,26,9)" --json

Denoise before indicators (smoother inputs):

mtdata-cli data_fetch_candles EURUSD --timeframe H1 --limit 1500 --indicators "rsi(14),ema(50)" --denoise ema --denoise-params "columns=close,when=pre_ti,alpha=0.2,keep_original=true" --json

Denoise after indicators (smoother signals):

mtdata-cli data_fetch_candles EURUSD --timeframe H1 --limit 1500 --indicators "rsi(14)" --denoise ema --denoise-params "columns=rsi_14,when=post_ti,alpha=0.3,keep_original=true" --json

4) Baseline price forecasts

Simple baseline (Theta):

mtdata-cli forecast_generate EURUSD --timeframe H1 --horizon 12 --library native --method theta --json

Pattern-based “analog” forecast (nearest-neighbor style):

mtdata-cli forecast_generate EURUSD --timeframe H1 --horizon 12 --library native --method analog --params "window_size=64 search_depth=5000 top_k=20 scale=zscore" --json

Optional: foundation model (if available):

mtdata-cli forecast_generate EURUSD --timeframe H1 --horizon 24 --library pretrained --method chronos2 --params "context_length=512" --json

Optional: Monte Carlo simulation forecast (range of outcomes):

mtdata-cli forecast_generate EURUSD --timeframe H1 --horizon 12 --library native --method mc_gbm --params "n_sims=3000 seed=7" --json

5) Volatility forecast (risk sizing)

EWMA (fast default):

mtdata-cli forecast_volatility_estimate EURUSD --timeframe H1 --horizon 12 --method ewma --params "lambda_=0.94" --json

HAR-RV (uses intraday realized volatility):

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

6) Barrier analytics (TP/SL planning)

Barrier probabilities (TP/SL odds within the horizon):

mtdata-cli forecast_barrier_prob EURUSD --timeframe H1 --horizon 12 --method hmm_mc --barrier '{"kind":"tp_sl","unit":"pct","take_profit":0.5,"stop_loss":0.3}' --params "n_sims=5000 seed=7" --json

Barrier optimization (search a TP/SL grid):

mtdata-cli forecast_barrier_optimize EURUSD --timeframe H1 --horizon 12 --method hmm_mc --mode pct --grid-style volatility --search-profile medium --params "n_sims=5000 seed=7" --json

Interpretation shortcuts:

  • edge ≈ P(TP first) − P(SL first)
  • prob_resolve ≈ 1 − P(no hit)
  • Wider stops/targets may look “worse” on edge but can be better after accounting for time-to-hit and payoff ratio.

7) Regime detection (gatekeeper)

Change-points (BOCPD):

mtdata-cli regime_detect EURUSD --timeframe H1 --fetch-limit 1500 --method bocpd --threshold 0.6 --lookback 300 --json

Regime labels (HMM):

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

8) Quick backtest (sanity check)

Run a rolling-origin backtest to compare a few methods:

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

9) Pattern detection (optional context)

Candlestick patterns:

mtdata-cli patterns_detect EURUSD --timeframe H1 --mode candlestick --lookback 500 --robust-only true --json

Classic chart patterns:

mtdata-cli patterns_detect EURUSD --timeframe H1 --mode classic --lookback 800 --json

Harmonic Fibonacci-ratio patterns:

mtdata-cli patterns_detect EURUSD --timeframe H1 --mode harmonic --lookback 800 --json

Bill Williams fractal price-structure levels:

mtdata-cli patterns_detect EURUSD --timeframe H1 --mode fractal --lookback 300 --json

Volume-profile structure levels from bounded tick data:

mtdata-cli volume_profile_levels EURUSD --start "2026-01-07" --end "2026-01-08" --source auto --price-source mid --bucket-points 10 --json

Equivalent H1 lookback form:

mtdata-cli volume_profile_levels EURUSD --timeframe H1 --lookback 168 --source auto --price-source mid --bucket-points 10 --json

Combine the two by opting fractal detection into volume-profile confluence:

mtdata-cli patterns_detect EURUSD --timeframe H1 --mode fractal --lookback 300 --config '{"volume_profile":true,"volume_profile_tolerance_points":25}' --json

volume_profile_levels returns POC, VAH, and VAL. The defaults admit up to one day and 50,000 raw ticks; source=auto falls back to an explicitly labeled M1 bar approximation for larger windows. Broker FX volume is usually tick activity, not exchange-traded volume.

10) Combine signals into a decision (simple recipe)

A pragmatic decision checklist:

  • Forecast direction: do multiple methods agree on the sign of the next horizon bars?
  • Volatility: is projected volatility within your risk budget?
  • Regime: are you trading in a regime your strategy is designed for?
  • Barriers: does the TP/SL pair have acceptable edge / prob_resolve?
  • Patterns: do detections support or contradict the idea (optional)?

Appendix: Troubleshooting

See TROUBLESHOOTING.md.

See Also