Leave-one-out cross-validation
The test that measures what is left of an indicator once one of its inputs is removed.
The idea in one sentence
For each input that composes an indicator, we remove that input, recompute, and compare. A truly robust composite produces a coherent reading even when amputated of one of its ingredients. A composite that collapses at the slightest subtraction is not a composite, it is a single-variable disguise.
Why this method
A composite indicator can be silently dominated by a single one of its inputs. The apparent weighting can suggest a balanced split, while the result actually tracks one underlying series at 95%. Three problematic consequences follow : the reading becomes fragile when that source is degraded, the noise of that source dominates the whole suite, and the reading advertised as multi-factor is no longer such.
How Trinity applies it
Every Trinity composite is tested by subtracting each of its inputs in turn. The correlation between the complete version and each amputated version is measured. If a single subtraction collapses the correlation to the point where the reading changes interpretive zone, the indicator is rejected or recalibrated. The composites that survive this test are those whose value genuinely emerges from the synthesis.
What it does not guarantee
Leave-one-out validation guarantees that a composite is not dominated by a single input at the moment of the test. It does not guarantee that this independence will hold across all future regimes. It is a necessary but not sufficient filter, to be combined with the other pillars.
Going further
Leave-one-out cross-validation is one of the four Trinity validation discipline pillars, alongside Walk-Forward validation, Popper falsification and perturbation.
Institutional disclaimer
Trinity Insights is an educational and analytical tool. Content does not constitute investment advice. Trinity Insights is not a Crypto-Asset Service Provider (CASP) registered under MiCA Regulation (EU) 2023/1114, nor a Financial Investment Advisor (CIF), nor a PSAN. See the full disclaimer.