SETESG Selection-Bias Audit

One hundred and twenty-one of 641 SET Mainboard firms are in SETESG — the voluntary ESG index that selects governance leaders. Zero of those 121 firms is currently Kill-Switch-flagged.

That is the cleanest selection-bias finding in this Note. The hard floor against currently-distressed firms works. Of the 40 universe-wide Kill Switch firms, none have been admitted to the index. That is the easy story.

The complicated story is everything above the floor.

Fifty-three percent of SETESG firms are Chameleon — almost identical to the Large-cap segment that dominates the index. Eighty-two percent of Large-cap firms are in SETESG; 6% of Small-cap firms are. The voluntary participation infrastructure of the index — the disclosure cost, the internal sustainability-team cost, the ratings-engagement cost — is itself a cap filter, and most of what looks like governance-quality screening in the SETESG cohort is, on inspection, the inherited archetype distribution of the Large-cap segment that bears the cost.

This Note documents the selection bias. Note 1 of this series footnoted that this question — whether SETESG is genuinely screening for governance leaders or whether it is admitting a different cohort altogether — was the subject of a later piece. Note 6 is that piece, and the answer is more nuanced than either of the two endpoints the question implies.

SETESG: what the index is, and what it is not

The SET Sustainability Index family has a multi-stage history. SET launched THSI — the Thailand Sustainability Investment Index — in 2015 as a voluntary recognition list for Thai listed companies meeting SET's sustainability assessment criteria. An annual SET ESG Ratings system was introduced in subsequent years, with tier letters from B through AAA determining inclusion. The index was rebranded as SETESG in 2023, with broader scope and more granular methodology integrated with the rating tier.^[1]

Participation is voluntary. Firms apply to SET Sustainability Assessment, complete a self-disclosure questionnaire, and receive a rating. The listing rule does not mandate ESG assessment. Firms can be removed from the index if they fail subsequent annual review.

Within the framework's scored universe of 641 SET Mainboard firms, 121 are in SETESG at the production refresh date — a 19% inclusion rate. The architecture of the signal is structurally different from Korea's KCGS rating system, which assigns a public letter rating to essentially every KOSPI firm: in Korea, the signal is the position on the rating gradient; in Thailand, inclusion in the index is itself the signal.^[2] Thailand has no mandatory ESG-disclosure floor; Korea's universal-coverage rating effectively imposes one through transparency.

That structural difference matters for Note 6's central question. When inclusion is itself the signal, the participation cost becomes the dominant predictor of who gets selected.

The baseline reading

SETESG cohort statistics against the non-SETESG universe:

Metric	SETESG (n=121)	Non-SETESG (n=520)	Universe (n=641)
T mean	73.55	67.11	68.33
B mean	70.56	73.31	72.79
R mean	73.36	73.35	73.35
Composite g mean	72.58	71.47	71.68
Celestial %	29.75%	14.62%	17.47%
Hidden Gem %	13.22%	39.42%	34.48%
Chameleon %	52.89%	35.38%	38.69%
Poison Apple %	4.13%	2.88%	3.12%
KS %	0.00%	7.69%	6.24%
Mean cap (THB B)	94.4	14.0	—
Median cap (THB B)	26.0	1.7	—

Four findings sit on the surface.

SETESG selects on transparency. SETESG firms post a T-axis mean of 73.55 against the non-SETESG cohort's 67.11 — a 6.4-point overweight. That is the largest single-axis gap between the two cohorts.

SETESG does not select on Balance of Power. B-axis mean in SETESG is 70.56; in the non-SETESG cohort, 73.31. The voluntary index runs marginally weaker than the non-SETESG universe on the axis where Thailand's structural governance problem actually lives. The reading is not that SETESG firms are governance-weak — the gap is small — but that the index does not screen on the dimension that matters most for minority-shareholder protection.

Conflict-of-Interest Risk is essentially identical between cohorts. R-axis mean is 73.36 in SETESG, 73.35 outside it. The voluntary index admits firms across the same R-axis distribution as the universe baseline.

Zero Kill Switch firms in SETESG. The single cleanest selection-bias signal. None of the 40 universe-wide Kill Switch firms is in the index. The hard floor screen works, and it works completely.

The cap statistics show why most of the rest of the cohort patterns emerge. SETESG firms have mean market capitalization of THB 94.4 billion against THB 14.0 billion outside — a 6.7-multiple gap. Median cap runs THB 26.0 billion against THB 1.7 billion — a 15.3-multiple gap. SETESG is not, in distribution-shape terms, a representative sample of SET Mainboard. It is the Large-cap segment with a few mid-and-small-cap participants.

Cap is the dominant predictor

Cap band	Universe n	In SETESG	Inclusion rate
Large (≥ THB 100B)	28	23	82.1%
Mid (10–100B)	113	69	61.1%
Small (< 10B)	486	29	6.0%

The inclusion rate ratio between Large-cap and Small-cap is 13.7-to-one. Eighty-two percent of Large-cap firms participate in SETESG. Six percent of Small-cap firms do.

The mechanism is participation cost. SETESG inclusion requires sustained investment: disclosure infrastructure, internal sustainability-team capacity, ratings-engagement bandwidth. Large-cap firms have budgets for all three; Small-cap firms generally do not. Small-cap firms self-select out of the index by inability or unwillingness to bear the disclosure cost — not necessarily by governance quality.

This cap-driven selection bias is the dominant driver of every other SETESG observation. The 12.3-percentage-point Celestial over-representation in SETESG is partly the inheritance of the Large-cap archetype distribution, which itself runs Celestial-and-Chameleon-heavy. The 14.2-percentage-point Chameleon over-representation is the same inheritance running through a different archetype slot. The 21.3-percentage-point Hidden Gem under-representation is also cap-correlated, because Hidden Gems are 90.5% small-cap (per Note 1), and small-cap firms are essentially absent from SETESG.

The index, once decomposed, looks much less like a governance-quality screen and much more like a Large-cap-segment-plus-aspirants index where the inclusion threshold is paying for the disclosure stack.

Cluster overlap: the family-cluster amplifier

Cluster	Cluster n	In SETESG	Inclusion rate
BDMS	2	2	100.0%
Sri Trang	2	2	100.0%
PTT (state-cohort)	56	41	73.2%
Osotspa	5	3	60.0%
Royal-adjacent cohort	6	3	50.0%
Thai Union	5	2	40.0%
Mitr Phol	6	2	33.3%
TCC / ThaiBev	6	2	33.3%
Land & Houses	6	2	33.3%
Central	12	3	25.0%
CP Group	10	2	20.0%
Boonrawd	17	3	17.6%
Bangkok Bank	36	6	16.7%
KBank / Lamsam	14	2	14.3%
King Power	8	1	12.5%
Saha Pathanapibul	14	0	0.0%
Thai Summit	8	0	0.0%
Minor	1	0	0.0%

Aggregate: clustered cohort SETESG inclusion rate is 35.5% (76 of 214); non-clustered cohort inclusion rate is 10.5% (45 of 427). Family-cluster membership raises SETESG inclusion probability by 3.4× the non-clustered baseline.

The PTT state-cohort posts the highest inclusion rate among large clusters at 73.2%. State-enterprise hybrids carry mandatory-equivalent ESG-disclosure expectations through the State Enterprise Policy Committee framework, even though SETESG itself is voluntary.^[3] The state-cohort effectively operates with a soft mandate that family-controlled clusters do not face — and the inclusion-rate gap reflects that.

The royal-adjacent institutional shareholding cohort has 3 of 6 firms in SETESG (50.0%), reported here at the cohort-aggregate level consistent with the framework's PUBLIC_GUARDRAILS protocol on aggregate institutional-shareholding cohorts.

Three clusters post zero SETESG inclusion: Saha Pathanapibul (14 firms), Thai Summit (8 firms), and Minor International (1 firm). Saha's absence is the most notable: a 14-firm consumer-products cluster with B-axis at 76.6 (comfortably above universe baseline) and zero KS firms — yet zero SETESG participation. The absence reflects a voluntary-disclosure-engagement gap, not governance distress. Some clusters have built ESG-disclosure infrastructure; others have not. The gap is operational, not structural.

NVDR overlap: the counter-intuitive finding

NDR bucket	n	In SETESG	Inclusion rate
NDR ≤ 5%	257	24	9.3%
5–15%	152	73	48.0%
15–30%	32	21	65.6%
30–70%	5	0	0.0%
> 70%	1	0	0.0%

The inclusion rate is non-monotonic. SETESG inclusion peaks in the 15–30% NDR bucket at 65.6% — the structural minority-disenfranchisement cohort surfaced in Note 2. Of the 38 firms with NVDR Dilution Ratios above 15% in the SETESG cross-tabulation production cut, 21 are in SETESG (55.3%). The Note 2 figure of 39 firms reflects a slightly different refresh window; the structural finding is unchanged across either count.

The reading is counter to the naive assumption that high-NVDR firms would face exclusion from a sustainability index for the structural minority-voice problem they create. The opposite happens. High-NVDR firms attract foreign-investor attention precisely because of the FOL-bypass utility of the NVDR mechanism, and foreign-investor attention drives ESG disclosure investment as part of the cross-border-capital-attraction stack. The voluntary index responds to foreign-capital-driven incentives. It does not respond to the structural voting-rights asymmetry the NVDR mechanism creates.

The exceptions are at the extreme tail. Zero of the five firms in the 30–70% NDR bucket are in SETESG, and the single firm at NDR > 70% — F&D, the active KS-04 trigger documented in Note 2 — is not in SETESG.^[4] The voluntary index successfully screens out the structural-vacuum firms at the extreme end. It admits firms in the 15–30% bucket where the asymmetry is real but has not yet crossed into Kill Switch territory.

Note 2 and Note 6 surface complementary readings of the same structural feature. NVDR voting separation is a B-axis structural feature that the framework's NDR indicator captures monotonically. SETESG inclusion is an ESG-disclosure-quality signal that responds to foreign-investor demand. The two signals are not aligned. A firm with substantial NVDR exposure can be a SETESG member because foreign investors want it disclosed, even though the same NVDR exposure is what makes the firm's voting-rights architecture structurally asymmetric.

What the index actually does

The Note's central question — whether SETESG selects for governance leaders, admits problem firms, or sits in some middle position — has a more nuanced answer than the binary alternatives suggest.

SETESG screens hard against current Kill Switch firms. Zero of the 40 universe-wide KS firms are in the index. The hard floor works completely. Above that floor, SETESG operates as a cap-driven voluntary-participation filter rather than a governance-quality filter. The 82.1% Large-cap inclusion against 6.0% Small-cap inclusion confirms cap as the dominant predictor of SETESG presence; the +12.3-percentage-point Celestial over-representation and the +14.2-percentage-point Chameleon over-representation are both inherited from the Large-cap archetype distribution, which itself runs Celestial-and-Chameleon-heavy at 78.5% combined within the Large-cap segment.

Within the Large-cap SETESG cohort, the index does not strongly discriminate between Celestial firms and Chameleon firms. SETESG admits Chameleons at high rates because Chameleons are common in the Large-cap segment that bears the disclosure cost. The above-floor discrimination is weak.

Five SETESG firms (4.1% of the 121-firm cohort) are classified Poison Apple by the framework — transparent on the surface, structurally weak on Balance-of-Power or Conflict-of-Interest Risk underneath. SETESG admits these firms because the index reads the transparency side and not the structural-weakness side. The framework reads both. The two systems disagree on these five firms — and where they disagree, the question is which system is reading the structure correctly.

The aggregate count of those five firms is publishable. The framework's reading of which five firms is not. Investors and ESG-rating systems that treat SETESG inclusion as a uniform governance-quality signal are taking the position that disclosure-side transparency is a sufficient screen. The framework's position is that it is not — and the five Poison Apple firms in SETESG are the operational demonstration.

Korea, Thailand, and the floor-versus-gradient distinction

Market	Voluntary ESG architecture	Universe coverage	Inclusion mechanism
Korea	KCGS ESG rating (D through A+)	KOSPI essentially universal	Annual public letter rating per firm
Japan	FTSE Blossom Japan, JPX ESG, MSCI ESG Japan (multiple parallel voluntary indices)	Voluntary, multiple architectures	Various
Thailand	SETESG (binary in-or-out)	121 / 641 = 19%	Voluntary annual application
Singapore	SGX Sustainability Reporting + voluntary indices on top	Mandatory floor since 2017, voluntary indices above	Hybrid mandatory floor + voluntary

The structural divergence between Korea and Thailand is the floor-versus-gradient distinction. Korea's KCGS rating effectively functions as a mandatory-coverage system through its universality — every KOSPI firm receives a public letter rating, and the signal is the position on the gradient rather than the fact of inclusion. Thailand's SETESG is opt-in, and the cap-driven participation cost becomes the dominant filter on who gets in.

Korea's published research has documented that KCGS A+ and A rated firms still experience governance events at non-trivial rates — the rating gradient does not perfectly predict actual outcomes above the bottom tier. Thailand's SETESG shows the same pattern through a different architecture: the hard floor screen against current distress works (zero KS firms in SETESG, mirroring Korea's near-zero KS rate at A+ tier), but the discrimination above the floor is weak (52.9% Chameleon in SETESG, comparable to weak-signal-above-floor patterns in Korea published research).^[5]

Both ESG-rating systems, despite different architectures, function similarly: hard floor on currently-distressed firms, weak signal above the floor. The architectural difference does not change the structural finding.

Foreign investors are not the audience

A counter-intuitive footnote that closes the cohort reading. SETESG firms (117 of 121 with available foreign-ownership data) post mean foreign ownership of 32.6% — below the non-SETESG cohort's 38.4% (516 of 520 with data). The voluntary index is marketed partly toward foreign-investor screening, but foreign capital on SET Mainboard is not concentrated in the SETESG cohort. It is concentrated outside it.

The mechanism has three drivers. Foreign-parent-controlled firms (the 20-firm owner-type-P cohort: DELTA, BBGI, BCPG, BGRIM, and others) typically do not participate in SETESG — the foreign parent's holding already establishes foreign-investor presence, so the voluntary-disclosure signal is redundant. The PTT state-cohort, which has 73% SETESG inclusion, has only moderate foreign ownership through NVDR-mediated channels. And foreign investors who want Thai equity exposure use direct holdings or NVDRs as their access mechanism, not SETESG inclusion as a screen.^[6]

The voluntary index does not function as a foreign-investor-acquisition tool in the Thai context. It functions as a domestic-institutional and Thai-retail signal. That is itself a structurally interesting finding — and one that the cross-market comparison in Note 7 footnoted at high level but did not quantify until now.

What the selection-bias reading shows

Three structural takeaways.

First, SETESG's hard floor screen against current distress is real and complete. Zero of the 40 universe-wide Kill Switch firms are in the index. The floor mechanism functions exactly as designed.

Second, above the floor, SETESG is cap-driven rather than governance-driven. The 13.7-multiple Large-cap-to-Small-cap inclusion gap is the dominant predictor of presence in the index. Most of what looks like governance-quality screening in the SETESG cohort is the inherited archetype distribution of the Large-cap segment that pays the participation cost. The +12.3-point Celestial over-representation and the +14.2-point Chameleon over-representation are both cap-inheritance, not governance discrimination.

Third, the framework and SETESG agree at the floor and disagree above it. They agree completely on Kill Switch screening — neither admits a current KS firm. They disagree on five Poison Apple firms that SETESG admits and the framework reads as transparent-but-structurally-weak. The disagreement is the operational test: when a voluntary disclosure-quality index meets a structural-prediction framework on the same universe, the floor is shared and the gradient is not.

Voluntary ESG indices, in Thailand and in Korea, do floor screening well. They do not discriminate above the floor. The framework's contribution is to read the gradient that the indices do not — and the Note's reading of SETESG suggests that investors who treat voluntary-index inclusion as a governance-quality signal above the floor are reading a partial signal as a complete one.

The hard floor works. The gradient is the question. The framework reads the gradient.

SETESG inclusion rate by cap band. Large-cap dominates at 82.1%; small-cap barely registers at 6%. Participation cost is the cap filter.

---

The Apex G-Score framework currently covers 641 SET Mainboard listed companies as of the April 2026 production snapshot. Underlying data: FY2025 cross-section. Scoring under TBR v2.0 weights: Transparency 0.30, Balance of Power 0.30, Conflict-of-Interest Risk 0.40.