Open methodology

No black box. Every hypothesis is inspectable.

DirtyWater measures observed technical structure and risk on a 0–100 scale. Thresholds describe factor state; agreement reflects consistency, sample depth, continuity and data quality.

State thresholds: 40 / 60
Factors: 5
Recalculation: Every dataset

Composite engine

Five factor families. Declared weights.

Directional evidence stays separate from risk quality, so a strong trend cannot erase volatility, tail-loss or data-integrity warnings.

01Weight 29%

Trend structure

Protected trend output

Direction and strength from EMA, SMA, MACD and ADX.

02Weight 24%

Momentum

Protected momentum output

Agreement across RSI, Stochastic and MACD acceleration.

03Weight 13%

Volume evidence

Protected volume output

Price versus VWAP and normalized OBV participation.

04Weight 14%

Volatility quality

Protected volatility output

ATR and annualized dispersion penalize unstable samples.

05Weight 20%

Risk efficiency

Protected risk output

Drawdown, CVaR and Sharpe describe path and tail behavior.

Calculation set

Technical, path and tail risk in one context.

The terminal calculates SMA, EMA, Bollinger, RSI, MACD, Stochastic, ATR, ADX, OBV, VWAP, trend slope/R², efficiency, annualized return/volatility, volatility regime, drawdown, VaR/CVaR 95% and 99%, Sharpe, Sortino, Calmar, Ulcer index, Omega, gain-to-pain, skew, kurtosis, autocorrelation and data-quality diagnostics.

Open authenticated terminal

Protected model layer

Advanced calculations are executed inside DirtyWater systems and are not published as formulas.

Auditable outputs

The site shows source, horizon, confidence, risk and limits without exposing proprietary logic.

Legal boundary

Research output only: no personal advice, no trade instruction, no promise of future price.

Model registry · 3.0

Separate models. Explicit purpose.

Only predictive models that pass a chronological holdout enter the separate ML consensus. Pricing, portfolio and execution models remain diagnostics and never manufacture a directional vote.

01DW 3.0

Predictive ML

HMM · Random Forest · SVM · LSTM

Point-in-time features, chronological train/holdout split, accuracy and Brier gating.

02DW 3.0

Derivatives

Black–Scholes · CRR · Crank–Nicolson · Heston · SABR

Reference pricing, complete Greeks and stochastic-volatility diagnostics with declared assumptions.

03DW 3.0

Portfolio

ERC · Markowitz · Black–Litterman · PCA · Engle–Granger

Covariance-aware allocation references, factor concentration and long-run relationship tests.

04DW 3.0

Risk & execution

GARCH · VaR/ES · Stress · TWAP/VWAP/POV · Almgren–Chriss · Kelly

Tail risk, execution schedules, duration, convexity and bounded sizing diagnostics.

Declared limitations

What the model cannot know.

Regime breaks

Historical relationships can fail abruptly; thresholds cannot prevent gaps.

Data quality

Corporate actions, illiquidity, missing volume and source errors can distort results.

Model risk

Parameters can overfit. Outputs describe the selected sample and never claim certainty.