Market Prediction: Efficiency and Anomalies

Module: Fintech AI | Difficulty: Advanced

Efficient Market Hypothesis

• Weak: Prices reflect all past information
• Semi-strong: Prices reflect all public information
• Strong: Prices reflect all information

Anomaly Detection

Predictability

| Horizon | IC | Source | |---------|-----|--------| | Intraday | 0.1-0.2 | Microstructure | | Daily | 0.02-0.05 | Momentum | | Weekly | 0.01-0.03 | Mean reversion |

import numpy as np

class MarketAnomalyDetector:
    def __init__(self, lookback=252):
        self.lookback = lookback
    def detect(self, returns):
        mu = np.mean(returns[-self.lookback:])
        sigma = np.std(returns[-self.lookback:])
        z_score = (returns[-1] - mu) / sigma
        return {
            'z_score': z_score,
            'anomaly': abs(z_score) > 3,
            'direction': 'up' if z_score > 0 else 'down'
        }
    def regime_detection(self, returns, n_regimes=2):
        from hmmlearn import hmm
        model = hmm.GaussianHMM(n_components=n_regimes)
        model.fit(returns.reshape(-1, 1))
        return model.predict(returns.reshape(-1, 1))

Research Insight: Markets are predictably inefficient. Momentum and mean reversion are persistent anomalies. Machine learning can exploit these anomalies, but transaction costs and market impact limit practical profitability. The key is finding signals that decay slowly and are robust to regime changes.