Numinor Hidden Sector Risk: Finding the Industry Risk That Shenwan-Neutral A-Share Portfolios Still Carry

A-share managers remove industry risk by balancing long and short positions within each Shenwan industry, and Shenwan puts every company in exactly one industry. The businesses underneath are more mixed: about a fifth of listed revenue sits outside the company's own Shenwan level-1 label (19.4% with ChinaScope's own product-to-industry map). A portfolio balanced on labels can therefore carry industry exposure its manager believes is gone.

We measure that exposure with ChinaScope SAM's product-segment data. Each month from 2016 we build 204 long/short portfolios, all exactly neutral on Shenwan level 1, estimate every factor on one random half of the market and hold the portfolios in the other, and record how much of each portfolio's daily return each classification explains.

The finding. On top of Shenwan level 3, SAM uncovers nearly as much hidden industry risk again as level 3 itself does: 18.7 against 20.0 bps a year in the random portfolios, significant at t = 13.2 and positive in every year from 2016 to 2025. It holds on a much stronger benchmark (14 styles and 10 statistical factors: 15.3 against 17.1 bps) and when SAM is delayed 60 days after each filing. It comes from what SAM knows about each company: shuffling product mixes among companies in the same Shenwan level-3 industry removes 65% of it.

What it is for. Adding SAM to a Shenwan level-3 neutralisation lowers the portfolios' realised volatility in 9 of 10 years, while adding the same exposures shuffled between companies raises it. Removing the risk, though, costs signal: exact neutrality gives up more signal than it saves risk, and in concentrated books of 100 to 300 names a light penalty on SAM exposure lowers realised volatility in most years but still lowers the information ratio. SAM's value is in seeing, reporting and weighing this industry risk in the risk model, not in forcing it out of a portfolio.

---

Headline findings

Finding	Value	Basis
Hidden industry risk SAM uncovers beyond Shenwan level 3	18.7 bps a year	200 random L1-neutral portfolios
... compared with what Shenwan level 3 uncovers beyond level 1	20.0 bps a year	same
SAM's incremental explanatory power, all portfolios	+0.65pp adjusted R² (t 13.2)	2,039 portfolio-years
Years positive	10 of 10	2016-2025
Later sub-period, 2023-2025	+0.71pp (t 7.4)	date fixed before the study
Stronger model (14 styles + 10 statistical factors)	SAM 15.3 vs Shenwan L3 17.1 bps (t 10.0)	2017-2025
SAM delayed 30 / 60 days after filing	+0.57pp / +0.65pp	against +0.65pp
Share lost when product mixes are shuffled	83% across the market; 65% within Shenwan level 3	100 draws each
Realised volatility, Shenwan L3 + SAM neutral vs L3 alone	0.64% lower, 9 of 10 years; shuffled SAM 0.30% higher	random portfolios
Concentrated books (100-300 names), light SAM penalty	volatility -0.6% to -1.3% (lower in 7-8 of 9 years); information ratio -5.5% to -19.6%	random signals
Rebuilt in the Numinor Matrix on its own data	SAM +0.65pp (t 17.0), 11 of 11 years positive	package 0.3.2

Read with these limits

• A risk result. It measures industry co-movement a label-neutral portfolio still carries, not a return signal.
• Small in size. The hidden risk is under 1% of a diversified portfolio's variance; the value of removing it is a scenario of about USD 0.18m to 0.27m a year per USD 1bn, depending on the risk model, and rests on assumptions the paper tests and qualifies.
• Shenwan before mid-2021. Most Shenwan labels before July 2021 are its 2021 classification dated back (70% at level 3), which gives Shenwan the benefit of hindsight in those years.

Build it yourself

Everything rebuilds in the Numinor Matrix from five SAM fields (USD 2,250 a year each at list), with the helper package numinor_hidden_sector_risk_sandbox_helpers that produced the paper's results. A buyer's agent can rerun the test, the shuffles and the concentrated books before taking anything home.

Numinor Systems · ChinaScope · October 2026. Contact: tyl@numinor.io