Abstract
<title>Abstract</title> <p> Technology-transfer and innovation-support systems often use sector taxonomies to reduce the number of firms requiring costly, detailed assessment. This study argues that a taxonomy can be descriptively valid yet selection-invalid when the population used to construct it differs from the population it is used to screen. It develops the concept of target-population validity and tests it with a proprietary Korean bank panel. Sector screens are fixed with 2012–2015 information and evaluated on later outcomes for 116,193 young SMEs in 54 industries. Three near-capacity-matched rules are compared: the OECD whole-industry R&D taxonomy, an R&D screen recalibrated on the SME population, and an SME multi-signal screen adding registered-patent and recognised-intangible indicators. The main downstream criterion combines 2016–2019 sales growth of at least 10% with a BBB-or-better rating. After controlling for age, size, prior growth, profitability, leverage, liquidity, debt, legal form, prior technological activity, macro-sector, and timing, the SME R&D and multi-signal screens are associated with 2.23 and 2.39 percentage-point higher joint-success probabilities; the OECD screen’s adjusted difference is 0.85 points. Industry-clustered, wild-bootstrap, logit, attrition-weighted, placebo, and threshold analyses support the pattern. Recalibrating the same R&D construct on the target SME population explains nearly all practical improvement; adding patent and intangible signals contributes little at the margin. A dual-route architecture raises recall, but only when the value of avoiding false negatives exceeds added evaluation costs. The findings distinguish descriptive, target-population, and selection validity in upstream technology-transfer screening. <bold>JEL Classification:</bold> C52; L25; L26; O31; O32; O38 </p>