Backtesting is the backbone of quantitative research. It’s where ideas become data-driven strategies and where many good ideas are filtered out. Interviewers will test your understanding of not just how to run a backtest, but also how to diagnose bias, evaluate performance, and ensure robustness.
🧠 1. What is backtesting and why is it important?
Definition:
Backtesting is the process of applying a trading strategy to historical data to simulate how it would have performed.
Purpose:
- Validate whether a signal or model has predictive power
- Estimate returns, risk, and performance stability
- Diagnose overfitting, assumptions, and hidden biases
A backtest is only useful if it accurately reflects realistic trading conditions — otherwise, it’s just curve-fitting to historical noise.
📉 2. What is slippage and why does it matter?
Definition:
Slippage is the difference between the expected execution price and the actual execution price.
Sources of slippage:
- Bid–ask spread
- Market impact
- Latency and order queue positioning
- Rapid price moves after a signal triggers
Modeling slippage:
Simplified:
\(\text{Slippage Cost} = \text{Trade Size} \times (\text{Market Price} - \text{Execution Price})\)
In more advanced setups:
- Kyle’s lambda model for impact
- Almgren–Chriss optimal execution framework
- Volume-based impact models
Interview Tip: Always emphasize that ignoring slippage leads to overly optimistic backtest results.
🔁 3. What is turnover and how is it computed?
Definition:
Turnover measures how frequently a portfolio’s positions change, representing trading intensity (and thus cost).
Formula: \(\text{Turnover}_t = \frac{1}{2} \sum_i |w_{i,t} - w_{i,t-1}|\)
Interpretation:
- High turnover → more trading, more transaction costs
- Low turnover → more stable portfolio, lower cost
Use cases:
- Constraint in portfolio optimization
- Metric for execution cost modeling
- Stability diagnostic for factor strategies
🧬 4. What is alpha decay?
Definition:
Alpha decay measures how quickly a signal loses predictive power after it is discovered or after it is generated.
Types:
- Post-discovery decay: Alpha degrades as more capital crowds the signal.
- Signal timing decay: Alpha weakens as execution is delayed.
Empirical Example:
A signal with strong 1-day predictive return may decay to zero by day 3.
Quantifying alpha decay:
Compute IC or return spread over different holding periods:
\(\text{Decay}(k) = IC_{t+k}\)
Strategies with slow decay → easier to execute, less sensitive to slippage.
Fast-decay signals → require low latency and aggressive execution.
📊 5. What are the key performance metrics in backtesting?
| Metric | Meaning | Purpose |
|---|---|---|
| CAGR | Annualized return | Long-term growth |
| Volatility | Std. dev. of returns | Measures risk |
| Sharpe Ratio | Excess return per unit risk | Risk-adjusted performance |
| Sortino Ratio | Downside-only version of Sharpe | Asymmetric payoff strategies |
| Max Drawdown | Worst peak-to-trough drop | Tail risk & resilience |
| Calmar Ratio | CAGR / Drawdown | Drawdown-aware performance |
| Hit Rate | % of positive-return trades | Stability |
| Win/Loss Ratio | Avg win ÷ avg loss | Trade distribution |
⚙️ 8. What is realistic transaction cost modeling?
Good backtests include:
- Bid–ask spread cost
- Market impact proportional to trade size
- Variable commissions (based on region/asset)
- Short borrow fees
- Slippage models based on volatility & volume
A common simplified model: \(\text{Cost} = a \cdot |w_t - w_{t-1}| + b \cdot (|w_t - w_{t-1}|)^2\) where (a) models linear costs and (b) models market impact.
🚀 9. What makes a backtest “too good to be true”?
Red flags:
- Extremely high Sharpe (>3 in equities)
- Low drawdown with high return (unrealistic)
- Zero or near-zero turnover
- No parameter sensitivity
- Smooth PnL curve with no volatility
- Gains persisting across all regimes without explanation
Interviewers may ask you to critique a “perfect” backtest. Mention:
“It likely suffers from overfitting, unrealistic assumptions, or hidden data leakage.”