Summary of "Introduction to Quantitative Trading"
Introduction to Quantitative Trading
Quantitative Trading Overview
Quantitative trading involves creating a statistical edge and executing it to generate risk-adjusted returns. It consists of two core components:
- Model: Generates the statistical edge, often using machine learning techniques.
- Strategy: Executes the model’s predictions in the market.
Execution quality is as important as the model itself; poor execution can eliminate any edge.
Models in Quantitative Trading
Types of Models
- Regression models: Predict real-valued outputs such as future price, price delta, or returns. Returns are unitless, allowing cross-asset comparison.
- Classification models: Predict categories (e.g., price up/down) with associated probabilities.
Features and Model Complexity
- Features are inputs to models. Univariate models (using a single feature) are preferred for simplicity and robustness to avoid overfitting.
- Simpler models (e.g., linear regression) are favored over complex ones (e.g., neural networks) due to the risk of overfitting and better interpretability, following the principle of Occam’s Razor.
- Models typically use time series inputs, predicting future values based on past values (lags). The simplest case uses the last known value as a feature (AR1 model).
Core Skills for Quantitative Trading
- Econometrics: Foundation for time series analysis, including concepts like autocorrelation, non-stationarity, and co-integration.
- Machine Learning: Scalable and flexible, with no strong assumptions, ideal for large or high-frequency data.
- Programming: Essential for latency-sensitive execution and building reliable, scalable systems.
- Mathematics: Key areas include linear algebra, multivariable calculus, probability, and statistics.
Statistical Edge & Performance Metrics
- Expected Value (EV): The average net profit and loss (P&L) per trade after costs. Focus on EV rather than win rate.
Example: A high win rate does not guarantee profitability if losses are large. Even a small positive EV (1-2 cents) compounded over many trades can lead to strong returns.
-
Risk-Adjusted Returns: Measured by the Sharpe Ratio, calculated as expected returns minus the risk-free rate divided by the standard deviation of returns.
- For intraday strategies, the risk-free rate is often ignored due to short holding periods.
- A higher Sharpe ratio indicates a smoother equity curve and safer leverage usage.
-
Returns Calculation:
- Simple returns are asymmetric (e.g., +20% up, -6.6% down).
- Log returns are symmetric and time-additive, making them preferable for modeling.
Market Microstructure & Order Book Fundamentals
- Order Book: Displays supply (asks) and demand (bids) at various price levels.
- Spread: The difference between the best bid and ask prices; wider spreads occur in less liquid markets and represent a hidden cost for liquidity takers.
- Mid Price: The average of the best bid and ask prices; preferred over the last trade price to avoid bid-ask bounce noise.
- Market Orders: Remove liquidity (taking), filled at the best opposite side price, and can cause slippage if order size exceeds available quantity.
- Limit Orders: Add liquidity (making), earn rebates or lower fees, but may not always fill.
Quantitative Trading Strategies
Two Broad Types
-
Market Making
- Adds liquidity via limit orders.
- Profits from capturing the spread.
- Key parameters:
- Spread: Profit margin.
- Bias/Skew: Adjusts bid/ask to increase fill probability aligned with model predictions.
- Trade-off between spread size (profit vs. fill likelihood) and adverse selection risk.
-
Market Taking
- Removes liquidity via market orders.
- Pays higher fees, better suited for longer time horizons.
- Key decisions:
- Timing: When to trade.
- Time-based: Trades executed at fixed intervals (e.g., hourly, daily).
- Predicate-based: Trades triggered only when model predictions cross thresholds (e.g., predicted return > 0.01).
- Sizing: How much to trade.
- Constant: Fixed trade size regardless of prediction strength.
- Piecewise linear: Scale trade size linearly with prediction strength, bounded by a maximum size (e.g., using a hard tangent function).
- Nonlinear: Smooth scaling using functions like tanh, allowing finer sensitivity to prediction strength.
- Timing: When to trade.
Modeling Trading Behaviors with Linear Regression (AR1 Model)
- Mean Reversion: Negative weight in the linear model; price tends to revert to the mean.
- Momentum: Positive weight; price tends to continue in the same direction.
- The linear regression model has two parameters (weight and bias), offering:
- High interpretability.
- Good generalization.
- Very fast computation (3-5 CPU cycles per prediction).
Parameter Optimization Methods
-
Closed Form Solution (Ordinary Least Squares):
- Provides an exact analytical solution.
- Efficient for small datasets but not scalable to large or high-frequency data.
-
Gradient Descent:
- Iterative approach improving parameters stepwise.
- Suitable for large datasets.
- Learning rate controls step size; choice of loss function is flexible and critical.
- Can handle non-convex loss functions with local minima.
Risk Management & Leverage
- Strategies with low Sharpe ratios limit leverage due to the risk of large drawdowns.
- High Sharpe ratio strategies enable safer leverage and more stable returns.
- Conservative trade sizing is important to avoid blowups.
Explicit Recommendations & Cautions
- Focus on expected value rather than win rate.
- Prefer log returns over simple returns for modeling.
- Start with simple univariate models before exploring complex ones.
- Backtest any stop-loss or take-profit rules to ensure they improve expected value.
- Use constant trade sizing as a baseline before exploring scaling methods.
- Be cautious with leverage, especially with low Sharpe ratio strategies.
- Aggressively bias quotes in market making to increase fill rates if the edge is small.
- Avoid overfitting by limiting model complexity.
- Reliability and scalability of execution systems are as important as raw speed.
- Backtest all strategy parameters thoroughly.
- Understand trade-offs in spread size and fill probability in market making.
Assets, Instruments, and Sectors Mentioned
- No specific tickers or companies mentioned.
- Instruments include:
- Stocks (implied).
- Order book data.
- Limit orders (market making).
- Market orders (market taking).
- Time series data (price, returns).
- High-frequency data (order book ticks).
Methodology / Framework Summary
- Modeling: Build a predictive model (linear regression AR1) using time series features (last value).
- Optimization: Use closed form or gradient descent to find model parameters.
- Edge Quantification: Calculate expected value (EV) and Sharpe ratio.
- Strategy Execution: Convert model predictions into trades via:
- Market making (limit orders with spread and bias).
- Market taking (market orders with timing and sizing).
- Trade Sizing: Choose constant, piecewise linear, or nonlinear sizing based on prediction strength.
- Backtesting: Validate all assumptions, parameters, and filters.
- Risk Management: Monitor Sharpe ratio, control leverage, and manage drawdowns.
Disclosures / Disclaimer
- The video is educational and introductory.
- No explicit financial advice is given.
- Emphasizes backtesting before applying any strategy or parameter.
- Real-world complexities like market impact and latency optimization are acknowledged but out of scope.
Presenter / Source
- Single presenter (unnamed in subtitles).
- Presents as an experienced quant trader/educator sharing foundational knowledge on quantitative trading.
End of Summary
Category
Finance
