Gold trading basics · Lesson 83 of 100

Gold Performance Attribution: Separate Within-Condition Results from the Opportunity Mix

Calculate weighted subgroup results, distinguish a changing sample mix from method improvement and avoid hindsight regime labels.

Weights Change The Average: instructional diagram for lesson 83 — CYGURU GOLD · Original lesson schematic. Examples are illustrative.

An overall average combines outcomes and their weights

A strategy’s net average can change because its result within each environment changes, because it trades a different mixture of environments or because both occur together. A single average hides those mechanisms. Define condition labels that are observable before the decision, such as a specified session or a documented event flag. A label based on a later realized trend can explain history but cannot automatically serve as a live filter. Record the rule assigning each observation and preserve unclassified cases instead of forcing every trade into the category that makes the interpretation most persuasive.

Calculate a mixture change with stable subgroup results

Imagine two predefined conditions A and B. In the first period there are 10 A trades averaging +0.30R and 30 B trades averaging −0.10R. Weighted contribution is +3R−3R=0R across 40 trades, an overall mean of zero. In a second period there are 30 A and 10 B trades with exactly the same subgroup means. Total becomes +9R−1R=+8R, averaging +0.20R. Overall performance improved even though neither subgroup’s conditional mean improved. The change came entirely from opportunity weights under this invented example, not from a more skillful execution or a newly discovered edge.

Use an explicit reference weighting for comparisons

One diagnostic is to apply the same reference weights to both periods’ subgroup means. With 50% A and 50% B, the example standardized mean is 0.5×0.30+0.5×(−0.10)=+0.10R in each period. This is an analytical comparison, not the return actually earned by the unequal-count samples. State the chosen weights and why they are useful. Equal weighting, previous-period weighting and exposure-time weighting answer different questions. If risk per trade varies, a trade-count-weighted mean R can differ from dollar contribution or account return. Keep the accounting unit explicit.

Separate sample mix from subgroup results
Quantity	Period 1	Period 2
A count / mean R	10 / +0.30	30 / +0.30
B count / mean R	30 / −0.10	10 / −0.10
Total contribution R	0	+8
Overall mean R	0	+0.20
50/50 standardized mean R	+0.10	+0.10

Fixed subgroup means; standardized means are analytical comparisons, not actual account returns.

Check whether the subgroup evidence supports the story

Small cells can produce unstable conditional averages, and repeated overlapping signals can make counts overstate independent information. A subgroup defined after noticing a successful episode can merely restate that episode’s outcome. Report counts, dispersion, dates and costs within each group, along with the number of alternative groupings examined. A shift in the mix might reflect a genuine change in market conditions, a changed trading schedule or a data-collection problem. Do not declare a new favorable regime from one coefficient or a few trades. Investigate mechanisms before deciding whether the observation justifies a rule change.

Separate descriptive attribution from a prospective selection rule

Knowing after the period that condition A performed better does not establish that a trader could identify and exploit that difference in advance. A proposed filter needs a pre-entry definition, implementation timing, exposure limits and a separate evaluation. Removing B trades can also alter diversification, concentration and the amount of capital left idle. Compare the full process with a baseline rather than preserving only conditional win rates. Record how a filter behaves when the condition is missing or ambiguous. An attribution table can identify a research question while leaving the future usefulness of that question unresolved.

Practice task: reconcile weighted and standardized results

Reproduce the zero first-period mean, +0.20R second-period mean and +0.10R equal-weight standardized mean in both periods. Explain why the standardized figure is not a realized account return. Add a third unclassified category and describe how silently dropping it could alter the conclusion. Your report should state the classification time, weighting method, costs and sample counts. Finish with a prospective test plan for any new condition filter. The central discipline is to distinguish a changed mixture of opportunities from improved within-condition performance before advertising that the strategy has become better.

Gold Performance Attribution: Separate Within-Condition Results from the Opportunity Mix

An overall average combines outcomes and their weights

Calculate a mixture change with stable subgroup results

Use an explicit reference weighting for comparisons

Check whether the subgroup evidence supports the story

Separate descriptive attribution from a prospective selection rule

Practice task: reconcile weighted and standardized results

Can you apply the idea?

Make this lesson part of your process.

Gold Trading Mastery

Continue the beginner course

Lesson sources and further study

Core course references

Connected reading