QuantVise
HomeSymbolsStrategiesPredictionsData
Login

QuantVise

ML-enhanced trading strategies · Cost-aware evaluation

Systematic strategy development with rigorous backtesting, real-cost accounting, and cross-market validation to separate genuine alpha from overfitting.

ML-Enhanced Strategies

Machine learning models trained on macro data, volatility, and cross-market signals to predict regime changes.

Rigorous Backtesting

Repeatedly testing and evaluating all strategies across 22 indices to find opportunities and avoid disasters.

Cost-Aware Evaluation

Every result accounts for spreads, swap costs, and capital efficiency — showing effective return relative to actual capital needed.

Portfolio Management
Coming Soon

Platform for supervising and running strategies for clients, strategies as diversified portfolio.

Regime Detection

ML models identify bull, bear, and sideways regimes to rotate exposure — reducing drawdowns in crashes and staying invested during rallies.

Macro-Driven Signals

Models ingest VIX, yield curve, credit spreads, fed funds rate, and market breadth — not just price action.

Champion vs Benchmark

#StrategySymbolCapital EfficiencyTradesSharpePFSQNR²Worst YearBest Year
How We Measure Performance

Effective Return

Effective Return measures return per unit of capital actually deployed, not just per total investment. A strategy that earns +50% but only needs 1x capital scores much higher than one earning +50% that requires 6x capital in a bad year.

effective_return = net_pnl / (max_capital_x * avg_margin) * 100

The TOTAL Effective Return is the average of per-year values.

Capital Efficiency shows what percentage of allocated capital is actually used on average. A strategy scoring 91% never needed more than ~10% above its average margin — extremely efficient. A strategy at 16% needed over 6x average margin during drawdowns, meaning most capital sits idle waiting for worst-case scenarios.

capital_efficiency = (1 / max_capital_x) * 100

Higher is better. The champion achieves 91% vs BuyAndHold at 16%.

  • Walk-forward training — Models are trained on historical data and generate predictions only on unseen future periods, never on data they were trained on.
  • Cross-market validation — The same models are tested across 22 global indices to confirm signals generalize beyond a single market.
  • Cost-aware evaluation — All results include real trading costs: spreads, overnight swap fees, and commissions. No hypothetical frictionless returns.
  • Multi-year robustness — Strategies must perform across different market regimes — bull, bear, and sideways — not just one favorable period.
  • Capital efficiency filter — High-capital-x strategies are penalized in the ranking, preventing leveraged overfitting where a strategy looks good only because it uses excessive capital.