SAM Product Momentum — A Product-Spillover Alpha Signal for Chinese A-Shares Under Two Residualization Regimes

We document a cross-sectional alpha signal for Chinese A-share equities constructed from ChinaScope's SAM product-segment revenue panel. For each focal stock at each rebalance date, the signal captures the recent movement of its product environment — the market-cap-weighted average return of the pure-proxy basket of peers that derive more than 50% of their revenue from the same SAM Level-2 product — weighted by the focal's own revenue mix across SAM products. The methodology is evaluated under two parallel residualization regimes: a raw-return construction (Construction R) that keeps the signal as a deterministic function of stock returns and revenue mix, and a source-residualized construction (Construction S) that residualizes daily returns against the 22-factor base before signal construction.

Evaluated against a standard 22-factor (12 Barra-style + 10 supplementary) institutional risk model, the signal produces positive OOS multi-offset orthogonal ICIR: +0.352 (Construction R) / +0.307 (Construction S), with all 20 OOS rebalance-grid alignments positive and positive annual mean orth-ICIR in each of the 8 calendar years studied. Under an industry-included variant adding 31 SYWG L1 controls, the Construction R OOS multi-offset ICIR rises to +0.435. A comprehensive robustness battery confirms the signal's standalone-alpha character: orth-ICIR moves broadly upward as the evaluation base is replaced with random noise (factor quality shuffle to 100% noise: +0.794 / +0.859), falls monotonically as the base widens from 5 to 22 factors (K=5 → +0.72 / +0.84; K=22 → +0.35 / +0.31, R / S), and strengthens at longer forward horizons — peaking at 60 days before overlap adjustment, with the de-overlapped 60-day ICIR remaining positive at approximately +0.27 / +0.28.

Findings are based on a 10-year sample (2016–2026) using the live ChinaScope production data feed: company filings, segment revenues, and factor values are point-in-time with a publish_date + 30 days availability buffer modelling realistic vendor-delivery and buyer-ingestion lag. Numinor and ChinaScope are inviting selected institutional A-share quant funds to evaluate the SAM Product Momentum signal on their own factor stack under NDA.

---

Headline empirical findings

Finding	Construction R	Construction S	Reference
OOS multi-offset orth-22 ICIR (canonical)	+0.352	+0.307	§4.1 / §5.1
OOS multi-offset orth-22+industry ICIR	+0.435	+0.301	§4.5 / §5.5
% positive offsets, OOS, orth-22	100% (20/20)	100% (20/20)	§4.2 / §5.2
OOS per-offset ICIR range	[+0.097, +0.589]	[+0.089, +0.440]	§4.2 / §5.2
Annual positive rate (mean orth-22 ICIR)	8 / 8 years	8 / 8 years	§4.4 / §5.4
Weakest year (mean orth-22)	2022: +0.039	2022: +0.113	§4.4 / §5.4
Strongest year (mean orth-22)	2021: +0.792	2023: +0.710	§4.4 / §5.4
Standalone-alpha signature: 100%-noise ICIR	+0.794	+0.859	§6.1 / §7.1
Base count scaling: K=5 → K=22 ICIR	+0.716 → +0.352	+0.841 → +0.307	§6.3 / §7.3
Multi-horizon 60-day OOS ICIR (raw / de-overlapped)	+0.462 / +0.27	+0.483 / +0.28	§6.4 / §7.4
Annualized portfolio turnover	4.76×	4.41×	§6.5 / §7.5
Pre-orthogonalized delivery available	Yes (ICIR invariant)	By construction	§6.6

All numbers are gross of transaction costs. Net IR transferability requires the buyer's own portfolio-construction, cost, liquidity, borrow, and impact models — see Scope and limitations below.

---

What SAM Product Momentum is

SAM (Sector Analysis & Mapping) is ChinaScope's product taxonomy and segment-revenue panel for the Chinese A-share market. Approximately 8,000 product nodes are organized across a nine-level hierarchy; the methodology in this paper operates at Level 2 — 462 product sub-categories fine enough to be economically distinct (e.g., passenger-vehicle tires vs commercial-vehicle tires) and coarse enough to maintain meaningful peer density. Listed companies disclose segment revenue in their semi-annual and annual filings; ChinaScope's Data Automation System aggregates these into a per-stock per-period revenue mix across SAM products.

For each Level-2 product at each rebalance date, we identify the pure-proxy basket — listed companies deriving more than 50% of their revenue from that product (with at least 5 peers per basket to maintain diversification). At Level 2, approximately 90% of A-share company-report observations have a single Level-2 product exceeding the 50% threshold — most A-share listed companies are concentrated enough at L2 to qualify as a pure proxy for some product.

The signal for focal stock $i$ at trading date $t$ has two components:

• biz_mom — the focal's expected return given its product environment. For each SAM product $p$ in the focal's revenue mix, we compute the market-cap-weighted average return of the pure-proxy basket for $p$ (leave-one-out: the focal is excluded from its own basket). The focal's biz_mom is then the revenue-mix-weighted average of these basket returns, minus the focal's own return — capturing the spread between where the focal should be and where it actually is. Aggregated over a 20-day rolling window.
• biz_resvol — the 20-day residual standard deviation from regressing the focal stock's return series on its implied product-mix return series. Captures dispersion between the focal and its product environment, a complementary view of decoupling.

The cross-sectional z-score combination of the two components is the composite SAM Product Momentum signal. The pipeline is evaluated against the buyer's 22-factor risk model at the cross-section per date to produce the orthogonal incremental ICIR (the headline empirical metric).

---

Mechanism

Product-segment revenue disclosure is underused information in Chinese A-share quant. Companies sharing a primary Level-2 product face common product-environment shocks — demand swings, regulatory shifts, input-cost moves — that propagate to their stock prices with measurable lag. The empirical results are consistent with focal stocks partially lagging their pure-proxy product peers over 1–4 week horizons. The economic mechanism is limited-attention spillover: market participants process individual-stock idiosyncrasies more quickly than they process product-environment context, leaving systematic cross-sectional information unpriced for the rebalance horizons institutional quants trade.

The signal's content is concentrated in microstructure / risk-factor space rather than fundamental space (factor family drop analysis, §6.2): volatility, beta, and liquidity factor families absorb most of the signal's content under orthogonal evaluation. The standalone-alpha character is confirmed by two independent diagnostics — orth-ICIR moves broadly upward as the evaluation base is replaced with random noise (§6.1) and falls monotonically as the base widens from 5 to 22 factors (§6.3) — patterns that identify the signal as carrying its own predictive content rather than being a derived linear combination of the factor base.

The signal strengthens at longer forward horizons, peaking at 60-day forward returns (raw OOS ICIR +0.46 / +0.48 under R / S; +0.27 / +0.28 after standard √3 overlap correction). This horizon dependence is consistent with a product-spillover propagation mechanism that operates over weeks rather than days — supporting an economic-information interpretation rather than a short-horizon trading-noise interpretation.

---

Scope and limitations

This is diagnostic evidence, not a production-validated backtest. Five points are relevant for institutional readers:

• The paper is a signal paper, not a deployment paper. Orth-ICIR is a cross-sectional rank-correlation statistic, not a realized portfolio Sharpe. Turnover and cost-drag figures (~4.4–4.8× annualized, ~95–285 bps drag at 10–30 bps execution) support implementation-cost modeling but do not establish post-cost profitability. A separate deployment-focused study would address portfolio construction, capacity, transaction-cost-aware optimization, and benchmark-specific results.
• Effective IS truncation. The 22-factor base requires d_at_yoy (year-over-year asset-turnover change), which is unavailable in the current Tushare-sourced fundamental feed for fiscal periods before 2018. The effective in-sample period for 22-factor orth-eval is therefore 2019-04 to 2022-12 (~3.7 years), with OOS 2023-01 to 2026-04 (~3.3 years). A subsequent revision will rebuild the F2 panel from ChinaScope's broader fundamentals feed to recover the intended 2017–2018 IS window.
• OOS regime composition is favorable in some respects. The IS/OOS split places the weak 2022 year in IS and the strong 2023 year in OOS. However, final results show OOS performance broadly comparable to IS rather than mechanically higher: Construction R is +0.352 OOS vs +0.356 IS, while Construction S is +0.307 OOS vs +0.303 IS. The year-by-year breakdown is reported in §4.4 / §5.4 so readers can map performance to regime context directly.
• PIT buffer sensitivity not yet measured. The methodology canonical is publish_date + 30 days. Sensitivity to longer buffers (45 or 60 days) is a natural follow-up robustness test; we have not yet run it.
• Industry-classification convention. The +0.435 / +0.301 industry-included results use SYWG L1 (31 industries), the de facto standard for Chinese A-share institutional research. Buyers using a different industry classification (CITIC, GICS, proprietary) can substitute that table by matching the shipped schema; the qualitative pattern is expected to be preserved but the precise magnitude is buyer-dependent.

---

What the trial includes

We are inviting a small number of institutional A-share quant funds to evaluate the SAM Product Momentum signal in their own environment under NDA. Each trial includes:

• The full ChinaScope SAM data package. Full SAM taxonomy (~8,000 nodes) and Level-2 product mapping, per-stock per-period segment-revenue records covering ~5,500 active A-share companies across the 2016–2026 research window, reporting-period filing calendar, and SYWG L1 industry-classification history (for the industry-included orth-eval variant). Delivered as clean parquet + CSV files with full bilingual data dictionary and source-schema mapping for buyers wiring up their own production pipelines.
• Reference Python implementation. The complete SAM Product Momentum pipeline as a proper Python package (numinor-sam-pm): pure-proxy basket construction, leave-one-out canonical aggregation, biz_mom and biz_resvol computation, source residualization, composite signal building, and orthogonal-ICIR / multi-offset / year-by-year / robustness-battery evaluation primitives. Drop the buyer's own 22-factor base in; reproduce the analysis under their universe, factor stack, and constraints.
• Four notebooks covering the full whitepaper:
  • 01_quickstart.ipynb — synthetic toy demo of the full pipeline.
  • 02_replication.ipynb — reproduces §4–§5 headlines including the industry-included variant; targets Whitepaper v2.2 headline ICIR to three-decimal agreement (±0.001 tolerance) under the frozen reference data package.
  • 03_custom_application.ipynb — buyer-customization template for arbitrary factor-base counts (N = 12, 22, 50, …).
  • 04_robustness.ipynb — reproduces the full §6 / §7 battery (factor quality shuffle, family drop, base count scaling, multi-horizon, turnover, pre-orthogonalized delivery equivalence).
• Methodology consultation. Direct technical engagement with Numinor's research team on integration, robustness testing against the buyer's specific factors, and production deployment considerations.

---

Next steps

• Read the full technical whitepaper (Numinor SAM PM Whitepaper v2.2, 54 pages) for the complete methodology, the parallel §6 / §7 robustness batteries across both constructions, and the full supporting figure set.
• Request a data trial through your existing ChinaScope or Numinor relationship, or contact us directly at the address below.
• For technical methodology or replication questions, reach out to Numinor Systems directly.

---

Numinor · ChinaScope · May 2026 Contact: tyl@numinor.io