Methodology: Gradient Signal Scoring
Swiss-Quant uses a proprietary gradient scoring engine instead of traditional binary thresholds. This page explains how our signals are calculated.
Why Gradient Scoring?
Traditional screeners use binary rules: RSI < 30 = "oversold" = buy. RSI 31 = nothing. This creates false signals at arbitrary boundaries. Our gradient approach assigns continuous scores from -1.0 (strong sell) to +1.0 (strong buy), eliminating cliff effects.
The 8 Technical Indicators
1. RSI (Relative Strength Index)
Measures momentum on a 0-100 scale. Gradient: RSI 10 → +1.0 (strong buy), RSI 30 → +0.2, RSI 50 → 0 (neutral), RSI 70 → -0.2, RSI 90 → -1.0 (strong sell). Smooth piecewise interpolation between anchor points.
2. MACD (Moving Average Convergence Divergence)
Measures trend momentum via the MACD histogram. Gradient: histogram / |signal line| ratio, clamped to [-1, +1]. A large histogram produces a strong signal; a tiny histogram approaches zero.
3. Stochastic Oscillator
Measures the closing price relative to the high-low range over 14 periods. Same gradient shape as RSI with adjusted boundaries optimized for mean-reversion signals.
4. ADX (Average Directional Index)
Measures trend strength (not direction). Used as an amplifier rather than a directional signal: ADX < 15 (choppy market, multiplier ×0.7), 15-25 (ramp to ×1.0), 25-60 (ramp to ×1.4), 60+ (capped at ×1.4).
5. CCI (Commodity Channel Index)
Measures price deviation from its statistical mean. Gradient: -200 → +1.0, -100 → +0.3, 0 → neutral, +100 → -0.3, +200 → -1.0.
6. EMA Cross (12/26)
Measures short-term vs long-term trend alignment. Gradient uses a hyperbolic tangent: tanh(spread_pct × 0.5) where spread = (EMA12 - EMA26) / EMA26 × 100. This naturally saturates at extreme values.
7. Bollinger Bands
Measures price position within the 20-period volatility envelope. At the lower band → +1.0 (buy), middle band → 0 (neutral), upper band → -1.0 (sell). Linear interpolation between bands.
8. Volume Confirmation
Validates signal strength using volume relative to the 20-day average. Volume ≥ 1.0x average confirms the signal at full weight. Below 0.5x average reduces signal weight, filtering out low-conviction moves.
Composite Score Calculation
Individual indicator scores are combined using a weighted sum:
The raw weighted sum is then multiplied by the ADX amplifier (0.7 to 1.4) and the volume confirmation factor, producing a final composite score.
Signal Quality Tiers
| Tier | Requirements |
|---|---|
| HIGH | ≥5/7 indicators agree + volume ≥ 0.7x avg + ADX ≥ 20 + |score| ≥ 0.3 |
| CONFIRMED | ≥4/7 agree + (volume OK or ADX ≥ 25) + |score| ≥ 0.2 |
| SPECULATIVE | |score| ≥ 0.15 but missing confirmations |
| WEAK | |score| < 0.15 |
Action Logic
BUY: score > 0.35 (strong) or score > 0.15 + volume confirmed.
SHORT: score < -0.35 (strong) or score < -0.15 + volume confirmed + ADX ≥ 20.
HOLD: Everything else -- avoids false signals from weak or unconfirmed scores.
AI Forecast Models
Our equity, crypto, forex, and commodity forecasts use an XGBoost + LightGBM + Ridge stacking ensemble with walk-forward validation (5 splits, 7-day purge gap). Only predictions meeting quality gates -- directional accuracy > 50% and confidence > 50% -- are published.
Feature engineering includes technical indicators, macro data from FRED, sentiment analysis via Alpha Vantage, and asset-specific signals such as on-chain data for crypto and CFTC positioning for commodities. Feature selection uses mutual information regression to retain the top 30 features from approximately 80 candidates.