Quantitative Trading
Definition
Quantitative trading is a systematic, automated approach to trading: strategies are expressed as explicit rules, validated on historical data, and executed by software rather than discretion. Its appeal is scalability and low time cost — “trading millions from home” — but only if the strategy is consistently profitable. Overriding the system manually usually makes it worse. Notes here follow Ernest Chan’s Quantitative Trading.
Core Ideas
Sourcing strategies
Ideas are not scarce — the work is filtering. Sources: academic finance sites (SSRN, NBER, Quantpedia), retail-focused blogs/podcasts, trader forums (Elite Trader), and Twitter. Screen a candidate strategy before deep backtesting:
- Does it outperform a benchmark and have a high enough Sharpe Ratio?
- Small enough drawdown and short enough drawdown duration?
- Does it suffer survivorship bias, or lose steam in recent years?
- Does it have a “niche” that shields it from institutional competition?
- Does it fit your constraints — time, programming skill, capital, income vs long-term-gain goal?
Academic strategies are often too complex, out of date, dependent on expensive data, or only work on illiquid small-caps.
Backtesting and its biases
A realistic backtest is the heart of the discipline. Key hazards:
- Survivorship bias — data missing delisted stocks inflates results (Yahoo has it; CRSP/Sharadar are survivorship-bias-free).
- Look-ahead bias — using information not available at decision time.
- Data-snooping bias — too many parameters or too many tweaks over-fit history; defend with large samples, out-of-sample testing, and sensitivity analysis.
- Transaction costs — slippage, commissions, and market impact when order size is large relative to volume/market cap.
Platforms: QuantConnect (LEAN engine, realistic fills/slippage/costs, backtest→live with no code change), Blueshift (QuantInsti, free minute data). Performance is measured by annualized Sharpe ratio and maximum drawdown.
Business structure — retail vs proprietary
| Retail brokerage | Proprietary firm | |
|---|---|---|
| To open | None | Pass NASD Series 7 |
| Leverage | ~2x overnight / 4x intraday (Reg T) | Firm’s discretion, 20x+ |
| Liability | Unlimited (unless via LLC/S-corp) | Limited to deposit |
| Guidance | None | May mentor (sometimes for a fee) |
| Trade-secret risk | Low | Managers can piggyback |
The essentials regardless of choice: low commissions, variety of instruments, deep liquidity, and — most importantly — an API for real-time data and order transmission. A non-US resident can incorporate a US LLC to cap liability.
Automated trading (ATS)
An automated trading system generates and transmits orders from your strategy. Advantages: faithful adherence to the backtest, ability to run multiple strategies at once, and fast execution (essential for high-frequency). Tools: Backtrader (Python), IBrokers (R), Trading Toolbox (MATLAB), Blueshift (low-code). Paper trading first surfaces software bugs, hidden look-ahead/data-snooping bias, operational issues, and realistic transaction costs.
Risk management
The dominant cause of blowups is overleveraging a good model, not a bad model. Controls:
- Size positions with the Kelly Criterion — and typically use half-Kelly for safety.
- Stop-losses help in momentum regimes but are harmful in mean-reverting regimes; distinguish news-driven (momentum) moves from liquidity-event (mean-reverting) moves.
- Guard against model risk, software risk, and natural-disaster risk.
- Psychological discipline: do not succumb to despair or greed; start small and scale up gradually. Loss aversion means losses feel 2–2.5× the pain of equivalent gains.
- Trade for expectancy (期望), not for hit rate. A high win probability is not the objective; positive expected value per trade is. A strategy that wins 30% of the time with a 5:1 payoff beats one that wins 70% of the time with a 1:3 payoff. Optimising for “being right often” is the same trap Measurement Dysfunction describes — a proxy metric standing in for the goal.
Strategy families and frequency
Every quant strategy exploits either mean reversion or momentum — see Mean Reversion and Momentum for the statistical machinery (stationarity, cointegration, Ornstein-Uhlenbeck half-life, CPO regime optimization). Two structural results from Chan (ch. 7–8) worth carrying:
- High-frequency → high Sharpe. By the law of large numbers, more independent bets shrink the percent deviation from mean return, so HFT strategies carry the highest Sharpe and long-term compounded growth — at the cost of being hard to backtest and technology-reliant.
- Leverage low beta. A highly-leveraged low-beta portfolio should out-compound an unleveraged high-beta one, since low-beta stocks empirically deliver higher Sharpe.
Relationships
- Kelly Criterion — optimal leverage and capital allocation across strategies
- Mean Reversion and Momentum — the two regimes every strategy exploits
- Sharpe Ratio — the risk-adjusted performance screen
- Portfolio Risk Management — drawdown, beta neutrality, volatility targeting
- Machine Learning — signal generation and price prediction inputs
- Trading & Finance — parent topic