Research and Performance Testing · CFDs Trading Guide

A rule can fit old data beautifully and fail immediately afterward. Independent testing helps separate a repeatable idea from a lucky selection.

Count the ideas you tried

Testing many indicators, thresholds and markets increases the chance of finding an attractive historical result by luck. The final chart hides the rejected alternatives. Keep a research log of all versions and explain why each change was made. A simple rule with an economic rationale is easier to assess than a collection of exceptions designed around past losses.

Protect unseen data

Use an earlier period to develop the rule and a later period to evaluate it after the design is fixed. Preserve chronological order. Once results from the later period influence a change, that period is no longer an independent test. Reserve a further untouched interval or use a carefully specified rolling evaluation process.

Look for stability rather than one perfect setting

If a 19 day threshold performs very well but 18 and 20 days fail badly, investigate why. A broad region of reasonable results is more credible than a narrow optimum, although neither guarantees future success. Test different volatility regimes, markets and cost assumptions without quietly selecting only the combinations that work.

Move cautiously from research to observation

A demonstration account can reveal workflow and signal timing problems, but its fills may not match live execution. Compare the model with subsequent observed quotes and costs before drawing conclusions. A failed independent test is valuable evidence. It can justify rejecting the idea rather than searching for another parameter that restores the old headline return.

Try it yourself

You examine a reserved test period, change the exit rule to improve its result, then report that same period as unseen validation. What is wrong with this process?

Show the worked answer

The reserved period influenced the model, so it became development data. A genuinely independent later sample is needed to evaluate the revised rule.

Further reading

Educational material, not a recommendation to trade. Numerical examples are hypothetical. CFDs involve leverage and substantial risk of loss. Availability, costs and protections depend on the contract, legal entity, jurisdiction and client classification.

Continue learning