To prevent overfitting, SQX splits historical data into two segments:
| Pillar | Purpose | Key Techniques | |--------|---------|----------------| | | Clean, aligned, survivorship-free datasets | Point-in-time databases, anomaly detection, corporate actions adjustment | | Signal Generation | Predict future returns | Linear models (PCR, Ridge), tree-based (GBRT), neural nets, NLP from filings | | Portfolio Construction | Combine signals into positions | Mean-variance, risk parity, machine learning optimization, constraints | | Risk Management | Limit drawdowns & volatility | VaR, CVaR, factor risk models, stop-loss rules, regime detection | | Execution | Minimize market impact & delay | VWAP, TWAP, adaptive algorithms, liquidity-aware slicing | | Backtesting | Validate real-world viability | Walk-forward, cross-validation, monte carlo with transaction costs | strategy quant x
class QuantX: def __init__(self, capital, lookback=60): self.capital = capital self.lookback = lookback def regime(self, df): aroon_up = (df['high'].rolling(25).apply(lambda x: x.argmax()) / 25) * 100 if aroon_up.iloc[-1] > 70: return 'trend' elif aroon_up.iloc[-1] < 30: return 'revert' else: return 'neutral' To prevent overfitting, SQX splits historical data into