Backtests and robots: what simulation demonstrates

The essentials

A backtest applies rules to historical data using execution and cost assumptions. It describes an experiment, not future profit. Optimizing many parameters on one period may fit noise instead of a stable relationship.

Understand the mechanism, not just the definition

A backtest evaluates rules under particular data and assumptions, not continued validity or correct handling of every live situation. Repeated parameter tuning can fit noise. Retain samples not used for parameter selection and examine costs, data quality and sensitivity.

Example and verification

Keep version, data, period, spread, commission and execution model. Reserve a period unused during optimization and vary costs. Record errors rather than only the best-looking curve.

Worked case, step by step

Two variants produce 100 gross. Ten trades at cost 2 leave 80, while one hundred trades at the same cost leave −100. Omitting commissions or using unrealistic spread can favor the most cost-sensitive variant. Evaluate adverse periods and equity declines, not only ending balance.

What to verify when applying this

Before automation, check permissions, maximum size, stop controls and disconnection behavior. Copying and signals add dependencies; provider results do not guarantee yours. A VPS keeps processes running but does not eliminate strategy or execution risk.

Practice: verify it in your environment

Learn by observing the process and keeping its data. Complete this practice in demo or using documentation, without sending real orders or adding capital for the exercise. Record differences between the example and your account specifications.

1. Keep version, parameters, data, period and costs for each test.
2. Test out of sample and with harsher costs without using real money for the exercise.

Test your understanding

Try the question before revealing the answer. Justify the conclusion and its assumptions rather than memorizing a number.