Two intuitive upgrade families, both killed by the same unusual control: a random-veto baseline.
Idea 1: add a fourth factor to the score
Our stock score is a three-factor composite. We tested adding a profitability factor, a buyback factor, and
both. Each addition lowered returns — and, crucially, all of them scored below a null where the
fourth factor is random noise (the random-fourth-factor baseline averaged 20.0%/yr with a best of 22.2%,
versus 24.0% for the untouched score). When adding a real, sensible factor does worse than adding noise,
the composite is at a local maximum: any dilution hurts.
Idea 2: veto the scary candidates
Exclusion filters are the most natural risk idea there is: skip the heavy share issuers, the loss-makers,
the high-volatility names, the aggressive asset-growers. We tested each veto against a baseline that vetoes
the same number of names at random. Every real filter scored below the random veto on the
large-cap list, and all were outright negative on the mid/small-cap list.
share of the alpha. Filtering for comfort sells the edge. “Risky” and “mispriced” overlap — that
overlap is largely what the score is finding.
Verdict: no filters, no fourth factor. Risk management stays where it measurably works: position
sizing, the trend gates, and portfolio caps — not candidate vetoes.