Look-ahead bias, survivorship, over-fitting and unrealistic costs — the errors that make bad systems look brilliant.
Key takeaways
- Include realistic costs and slippage or the test is fiction.
- Reserve out-of-sample data before you start optimising.
- Prefer fewer parameters and a longer sample over a perfect curve.
The purpose of a backtest is not to find the best-performing rule set. It is to estimate whether a plausible idea has an edge that survives contact with costs and variance. Optimising until the equity curve looks beautiful achieves the opposite.
Trap one is look-ahead bias: using the day's close to decide a trade taken during that day, or using an adjusted price series that already reflects future corporate actions. It is easy to do accidentally and it inflates results dramatically.
Trap two is survivorship. Testing today's index constituents over ten years quietly excludes every company that failed, which makes almost any long-only rule look good.
Trap three is over-fitting. Every added parameter buys a better historical curve and reduces the probability of live performance. Hold out at least a quarter of your data and never touch it during development.
Trap four is cost realism. Include brokerage, exchange fees, GST, STT, stamp duty and a slippage assumption that reflects the instrument's actual spread. Strategies with a small per-trade edge usually die here — better to learn it in the test than with capital.
Related guides
Systems
How to evaluate whether a strategy is actually working
Sample size, expectancy, adherence and drawdown — four metrics that answer the question P&L cannot.
Systems
Win rate versus risk-reward: the trade-off nobody escapes
High-accuracy small-winner systems and low-accuracy trend systems both work. Choosing between them is about temperament.
Systems
Market regimes: why one strategy cannot work all year
Trending, ranging and volatile-choppy conditions reward different behaviour. How to identify the regime you are in.
Systems
Rule-based and algorithmic trading in India: where to start
You do not need code to be systematic. A written rule set, then a spreadsheet, then automation — in that order.
Your edge deserves
real capital
Start today from ₹1,499, trade Equity, F&O, Currency or MCX your way, and get paid in rupees on exactly the terms you were shown on day one.
Trade · Prove · Get Funded — evaluation fee refunded with your first INR payout.