Time-Series Analysis — 20 Years of Google Stock

1. Core Principle: Price is Noise, Returns are Signal

Raw price levels are non-stationary. To build a robust model, prices must be transformed into returns — a stationary stochastic process centered around a mean.

2. ETL Pipeline

• Ingest — raw OHLCV CSV with strict temporal ordering
• Transform — rolling features: volatility, moving averages, log returns
• Visualise — structural market regimes (Bull vs Bear)
• Quantify — Sharpe ratio, Beta, Max Drawdown, volatility clustering

3. Max Drawdown Implementation

def calculate_max_drawdown(price_series: pd.Series):
    """Quantifies worst-case capital destruction from peak to trough."""
    rolling_peak = price_series.cummax()
    drawdown = (price_series - rolling_peak) / rolling_peak
    return drawdown, drawdown.min()

4. Regime Detection

def detect_regimes(returns: pd.Series, window: int = 252) -> pd.Series:
    rolling_vol = returns.rolling(window).std() * np.sqrt(252)
    regime = pd.cut(rolling_vol, bins=3, labels=['Low Vol', 'Mid Vol', 'High Vol'])
    return regime

5. Key Findings

• Fat-tail distribution: crashes occur more frequently than normal curve predicts
• Volatility clustering: calm periods systematically followed by chaotic periods
• Max drawdown reveals worst-case investor psychology (2008, 2022 corrections clearly visible)
• Sharpe ratio confirms strong risk-adjusted performance over 20-year horizon