The Maths Every Day Trader Should Know
What would convince us a trading idea works?
Turn a plausible trading idea into a fair comparison with declared rules, available-at-the-time information, costs, and fresh evidence.
By ContextChart Research · 5 min
Published
Choose before the next results arrive
Try it yourself
Every candidate stays in the comparison. A better past fit has not been assumed to carry forward.
- Candidate B · cumulative net R
Read the plotted values
| Trade | Candidate B · cumulative net R |
|---|---|
| 0 | 0 |
| 1 | 0.95 |
| 2 | 1.9 |
| 3 | 2.85 |
| 4 | 3.8 |
| 5 | 2.75 |
| 6 | 3.7 |
| 7 | 4.65 |
| 8 | 5.6 |
Three deliberately constructed candidate records, each with eight past and eight later ±1R trades. This exercise illustrates selection and unseen evidence; it is not a statistical test or a sample of real strategies.
The idea
An attractive explanation is the beginning of a trading question. A test asks what observation would separate it from a convincing story.
Start by defining the decision. What information would have been available? What action follows? At what horizon should the prediction be judged? How will entry, exit, cost, and failure be recorded?
Then choose a relevant comparison. A price level can be compared with nearby shifted levels. A forecast can be compared with a simple prior. A selection rule can be compared with the candidates it declined. The comparison should preserve enough of the original situation to isolate what the proposed idea adds.
Write those rules before examining the outcomes used to evaluate them. Otherwise it is easy to keep changing a definition until the sample appears to agree.
Keep fresh evidence for a later check. If the rule changes after that check, acknowledge the new version. The earlier results informed its design; they cannot all remain independent confirmation of it.
A useful test gives an idea a fair chance to fail and keeps the evidence that disagrees with it.
How this experiment works
Compare the impression created by a selected successful result with the evidence from the complete experiment. Choose a candidate, change the cost, then reveal how the selected past result compares with the fresh eight-trade sequence.
The exercise is about how much context a result needs. It does not supply a universal pass mark. A real test needs a decision-specific baseline, realistic costs, and data that represent the path the application will encounter.
How we use this
Our research process begins with a trader's question, defines its unit and timescale, identifies look-alikes, and tests whether a measure distinguishes them. Results are examined per instrument before pooling and checked across sample halves where appropriate.
That process has killed plausible ideas and corrected earlier conclusions. The runner audit removed a counting error. The strength work separated a useful description of currency movement from unsupported direction signals. The time-aware range work checked that changing future bars left earlier estimates unchanged.
These are concrete improvements in the reliability of the measurement. They do not make every remaining hypothesis true. Reproducible calculations and forward evidence still have different jobs.
Show the maths
An advantage is always relative to a defined alternative:
Measured lift = outcome under the proposed rule − outcome under the comparison
The outcome might be net R, forecast error, calibration, or a measurement-consistency score. Choose it to match the question. A gain on one measure does not automatically establish a gain on another.
Repeatedly choosing the best of many noisy results creates selection bias. Recording the attempted variants and testing a frozen rule on fresh observations helps expose that problem. Dependence and changing conditions still need attention.
What this does not establish
A passing historical comparison cannot guarantee future profits. It can justify a more specific next experiment. Keep data scope, costs, revisions, and unknowns beside the result so the next reader can judge how much weight it deserves.