QQ 201 Intermediate
Testing a strategy
Course description
A backtest that makes money has at least five explanations other than a real edge. This course takes them one at a time: information from the future, tuning to noise, the cost of searching, testing on data the strategy has already seen, and costs that were left out.
Each lesson describes the failure, then a test that catches it. The last course in the Academy shows how QubitQuant Terminal implements those tests.
What you will be able to do
- Say what a backtest does and does not establish
- Detect look-ahead bias with a truncation test and a delay test
- Compute how a search inflates the chance of a false positive
- Design a walk-forward test and keep a holdout honest
- Judge a strategy's edge against its costs
Syllabus
- Lesson 1What a backtest is A simulation of trades you did not make. Useful for ruling ideas out, weak for ruling them in.
- Lesson 2Look-ahead bias The backtest used something it could not have known yet. It is the most common way a result is wrong.
- Lesson 3Overfitting and multiple testing Try enough variations and one will look brilliant by luck. How to count what the search cost you.
- Lesson 4Walk-forward and holdouts The only fair test is on data the strategy has never seen. Two ways to arrange that.
- Lesson 5Costs Spread, commission, slippage and financing, which are small on each trade and together often larger than the edge.
- AssessmentProblem set Six problems on false positives, costs, look-ahead, walk-forward and holdouts, with worked solutions.