Are Green Bonds Associated with Shareholder Value? Market Reaction and Firm-Valuation Evidence from Thailand

Abstract

Whether green-bond issuance is associated with shareholder value remains unclear in small, concentrated emerging markets, where signaling and information-asymmetry channels from developed-market studies may not operate the same way. This study examines short-term stock-market reactions and medium-term firm valuation associated with green-bond issuance among the full population of 20 eligible Thai listed issuers: 34 analytical issuance events from May 2019 to May 2026, and an annual panel of 20 firms observed from 2015 to 2025, comprising 220 repeated firm-year observations and 187 complete cases in the baseline regression. Event-study models find no statistically detectable average abnormal return around the issue date, robust across benchmark models and event windows; two-way fixed-effect panel regressions, likewise, find no statistically detectable average association with Tobin’s Q, with a confidence interval wide enough to admit economically meaningful effects in either direction. The matching and selection analyses remain inconclusive. The IV model has a weak first stage and is therefore reported only as a diagnostic. Because the design identifies within-firm variation among issuers, rather than a comparison with matched non-issuers, the results should be read as associations, rather than causal effects. The study provides census-based evidence that, in Thailand’s small and concentrated green-bond market, a green label may not yet carry detectable average value relevance for shareholders.

1. Introduction

The global sustainable-debt market has expanded substantially, reaching USD8.1 trillion in cumulative GSS+ debt by the end of 2025 (Climate Bonds Initiative, 2026), alongside continued development of regulatory infrastructure in ASEAN and Thailand (ACMF, 2018; SEC Thailand, 2018; Bank of Thailand, 2022, 2023; ThaiBMA, 2026b). Whether this growth translates into a shareholder-value effect for issuing firms, however, is a distinct empirical question from whether the market itself has grown. Existing theory predicts that green-bond issuance may affect firm value through signaling, reduced information asymmetry, reputational benefits, and a potentially lower cost of capital, but whether these channels generate measurable valuation effects in a smaller, less mature market like Thailand’s remains unclear, and is the central question this study addresses.

This paper offers, to the authors’ knowledge, one of the first Thai firm-level studies to jointly examine green-bond issue-date returns and medium-term firm value using the full population of eligible listed issuers, rather than a sample. The empirical approach combines an event study of short-term market reaction with a fixed-effects panel regression of medium-term firm value, supplemented by matching and selection specifications that probe, without fully resolving, the familiar concern that firms already predisposed toward higher valuations may simply be the ones that choose to issue; an instrumental-variable specification is reported separately, in Appendix A, as a diagnostic, rather than as part of the main evidence base. Full methodological detail is given in Section 3.

Consistent with ASEAN-level evidence from Aini et al. (2023), who similarly find no significant stock-market reaction to green-bond issuance, this study finds no detectable average association between green-bond issuance and either cumulative abnormal stock returns around the issue date or Tobin’s Q in the years that follow, under the primary benchmark specifications. This finding contrasts with more positive evidence reported for some developed markets (e.g., Flammer, 2021). The matching and Heckman sensitivity analyses remain inconclusive because of residual covariate imbalance, outcome non-comparability, and limitations of the exclusion restriction; they are, therefore, not used to corroborate the baseline estimates. A separate, exploratory instrumental-variable diagnostic in Appendix A is uninformative because the proposed instrument has a weak first stage. In a market as concentrated and institutionally less mature as Thailand’s, green-bond issuance alone may not yet constitute a value-relevant signal for investors.

The study contributes to the literature in several respects. First, and most centrally, it tests a boundary condition of signaling theory that the largely developed-market literature has not directly addressed: whether the signaling and information-asymmetry channels through which green bonds are theorized to affect shareholder value extend to Thailand’s small, concentrated market. The results are compatible with a weaker value relevance of a green label in such settings, although the underlying mechanisms are not tested directly here. Second, it provides firm-level evidence using the full population of eligible Thai listed issuers, rather than a sample, avoiding sampling variation from selecting only a subset of eligible Thai issuers, while not overcoming the limited statistical power or external generalizability inherent to this small population. Third, it jointly examines both short-term equity-market reaction and medium-term firm valuation within a single study, rather than treating these as separate research questions. Fourth, it reports this null finding transparently, alongside confidence intervals and small-sample considerations, rather than treating the absence of statistical significance as proof that no economically meaningful effect exists.

The remainder of the paper proceeds as follows. Section 2 reviews the relevant literature and develops the study’s hypotheses and research question. Section 3 describes the data and methodology. Section 4 presents results. Section 5 discusses the findings. Section 6 concludes the paper, and includes limitations.

2. Literature Review

2.1. Green Bonds and Firm Value: Theoretical Channels

Three theoretical channels connect green-bond issuance to firm value: signaling, information asymmetry, and stakeholder legitimacy. The signaling logic is that green-bond issuance can operate as a costly and partly verifiable signal, particularly when it is accompanied by external review, credible use-of-proceeds commitments, and transparent post-issuance reporting. Y. Liu et al. (2026) find a marked rise in environmental disclosure quality following green-bond issuance in China, stronger when the bond carries third-party certification, funds pollution abatement, or represents a first issuance—evidence of credible signaling, rather than relabeling. A study of UK issuers reaches a similar conclusion by a different route: Arhinful et al. (2025) find that green-bond issuance, alongside environmental expenditure and emission-reduction policy, is positively associated with the firm’s subsequent reputation.

A second, related channel runs through information asymmetry: disclosure and third-party review can narrow the information gap with investors and lower the cost of capital. Agliardi and Agliardi (2021) model the green-bond premium as jointly determined by cash-flow and project-effectiveness uncertainty, helping explain why the empirical greenium literature is so mixed. Hyun et al. (2023) find that frequent green-bond issuers obtain more favorable bond pricing than infrequent issuers, consistent with repeated issuance serving as an information channel that reduces information asymmetry, and Muhammad et al. (2026), using a global propensity-score-matched difference-in-differences design, report lower overall capital costs and market risk among green-bond issuers relative to matched non-issuers, findings consistent with an informational channel.

Stakeholder legitimacy, the third channel, is conditional rather than automatic. Petreski et al. (2025), studying Swedish real-estate issuers, find that repeated green-bond issuance lowers the cost of capital, but a first or one-time issuance raises it. Legitimacy appears to be earned through a track record, not granted on the first attempt. Gaur et al. (2026) emphasize that narrative quality, transparency, and certification practices jointly shape how investors assess the credibility of green-bond disclosures.

These three channels imply green-bond issuance need not raise firm value uniformly. These studies suggest that any valuation effect may depend on certification, disclosure quality, and issuance history, rather than arising uniformly across issuers. A review of 59 studies published between 2019 and 2025 describes the evidence as fragmented, and concentrated in developed markets (Rodrigues Loiola et al., 2025). Wu et al. (2025) report associations between selected green-bond features and firms’ environmental and ESG disclosure performance, though these relationships are generally weak and not uniform across performance measures. Whether these channels operate similarly in smaller and less mature sustainable-finance markets remains less well established.

2.2. Empirical Evidence from Developed Markets

Flammer (2021) provides influential international corporate evidence that green-bond announcements are associated with positive stock-market reactions, particularly for first-time and independently certified issuances, alongside improvements in post-issuance environmental performance. Earlier international evidence also pointed in the same direction: Glavas (2020), using corporate green-bond announcements across 18 countries, documents a positive stock-price response to green-bond issuance, with the response becoming stronger after the 2015 Paris Agreement. At a more aggregate level, Li et al. (2024) find that Chinese cities’ green-bond issuances generate a positive operating-performance spillover onto local, especially environmentally-oriented, firms.

More recent studies using larger samples report less consistent results. Pietrzyk et al. (2025), examining 1122 European issuances between 2013 and 2024, find no significant average stock-price effect, and any effects that do exist are weakly negative at best and concentrated among larger, uncertified issuances. The Nordic evidence is starker: Aleknevičienė and Vilutytė (2024) report a significant negative reaction, partly attributed to the COVID-19/war-era rotation from equities into bonds. Across a broader set of 3618 ESG bond issuances (2021–2023), Ordonez-Borrallo et al. (2026) find a negative average stock-market response, particularly when the use-of-proceeds mandate is relatively inflexible.

The related literature looks not at stock prices, but at the green-bond “greenium”, the yield discount investors accept relative to a conventional bond. The results are similarly mixed, though on balance they suggest the market does price environmental credentials into fixed-income securities (Zerbib, 2019; Baker et al., 2018).

Two systematic reviews confirm this heterogeneity is not an artifact of any one study: Negi et al. (2025), synthesizing 73 studies (2007–2024), and Rodrigues Loiola et al. (2025), synthesizing 59 studies (2019–2025), both find the evidence fragmented and overwhelmingly drawn from Europe, the United States, and China—a gap that motivates testing whether these dynamics hold in a market as different as Thailand’s.

2.3. Empirical Evidence from Asia and ASEAN

Evidence from outside the developed world is considerably thinner, though a growing cluster of studies now speaks to different pieces of the puzzle. The work closest to the present study—on how green-bond issuance affects firm-level stock market outcomes—comes from Aini et al. (2023), who examine issuers across five ASEAN countries using content analysis and an SEM-PLS approach. They find issuance raises investor attention roughly in proportion to the amount issued, but has no statistically significant effect on stock market reaction, directly or through that attention channel. This finding provides a relevant ASEAN comparison for the present study, although differences in sample construction and methodology limit direct comparability. The most direct Thai evidence to date comes from Wanganusorn et al. (2026), though at the bond rather than the firm level—using 62 ThaiBMA-registered green bonds issued between 2019 and 2025, they document a positive relationship between bond return and risk, but find issuance volume unrelated to either, leaving the firm-value question specifically unanswered. Chowiendo et al. (2025) take a different approach altogether, examining carbon emissions, rather than firm value, in a panel of ASEAN-listed companies from 2019 to 2023, and find green-bond issuance associated with lower emissions, with firm size itself also negatively associated with emissions as a separate effect.

A separate strand of ASEAN research examines macroeconomic and environmental outcomes, rather than firm value; because these studies do not test the shareholder-value question this paper addresses, they are summarized only briefly here, as background context. At the country level, green-bond and green-finance activity has been linked to higher renewable energy consumption and lower emissions in some studies (Dong et al., 2023; Ahmed et al., 2022), though with effects that emerge only in the long run in others (Agustin et al., 2025), and effects on broader economic growth that are mixed and conditional on countries’ income levels and stages of financial development in others still (Phung Thanh, 2022; Kong et al., 2026). Qualitative and review evidence similarly describes ASEAN green finance markets as constrained by financing gaps, limited investor awareness, and immature infrastructure (Le & Vivian, 2025; Lin et al., 2026). Taken together, this literature is best read as conditional, rather than uniformly positive or negative: reported effects depend on the outcome measured, the time horizon, and the specific market studied, which is consistent with the firm-level heterogeneity discussed above, rather than contradicting it.

Mixed or muted effects of green finance are therefore not unusual in ASEAN and Asian settings, although the outcomes examined across studies differ substantially: no significant stock-market reaction is found at the cross-country firm level (Aini et al., 2023); at the macroeconomic level, green-bond-market growth shows mixed and conditional links to broader economic performance, depending on income level and financial development (Phung Thanh, 2022; Kong et al., 2026), a related but distinct outcome from firm value; and qualitative accounts describe persistent market immaturity (Le & Vivian, 2025). Recent Thai evidence has also begun to examine firm-level stock-market responses. Sukcharoensin and Apinyakul (2026), using 34 ESG-bond announcement and issuance cases among SET-listed firms, document a significantly positive stock-price reaction over selected post-announcement windows. Their study focuses on short-term ESG-bond market reactions, whereas the present study restricts attention to green bonds and jointly examines issue-date returns and medium-term firm value, while addressing selection and endogeneity concerns. Wanganusorn et al. (2026) address the Thai market only at the bond level; this study aims to close that gap directly.

2.4. Why Results Might Differ in Thailand

Several structural features of the Thai setting could weaken or eliminate the value effects documented elsewhere. Thailand’s relatively young and concentrated green-ond market may limit the informational value of issuance announcements. Ramos Farroñán et al. (2025), synthesizing 50 studies on emerging-market green finance, find effectiveness conditional on regulatory oversight and liquid secondary markets, conditions a smaller, younger market like Thailand’s may not yet meet. Antar (2026) discusses the potential for tokenization to improve the efficiency, transparency, and liquidity of blue-bond markets, although these applications remain prospective. Tuncel and Yaman (2026) find that the link between financial development and environmental outcomes is asymmetric in mature G7 economies but symmetric in emerging E7 economies, evidence that the level of market development itself shapes how these signals get transmitted.

The sample’s concentration in Energy and Utilities may also matter: valuations of regulated and state-linked firms may be influenced by administrative and sector-specific frameworks, potentially reducing the relative salience of discretionary ESG signals. Evidence on how state ownership moderates green-bond effects is itself mixed. L. Liu et al. (2025) find the tax-aggressiveness-reducing governance benefit stronger among non-state-owned Chinese issuers; Chang et al. (2025) find the innovation-stimulating benefit stronger among state-owned ones. These studies do not point in a single direction, but they suggest that ownership structure may condition the effects associated with green-bond issuance. Whether a similar mechanism operates in Thailand remains an open empirical question.

Thailand’s experience may also simply reflect the developmental stage that green-finance markets outside the largest economies tend to share. Liang and George (2026) find that green finance reduces CO₂ emissions by only around 0.12% per 1% increase across the BRICS economies, a modest effect consistent with a gradual financing channel. Broader reviews of emerging sustainable finance instruments point to a mix of persistent constraints, including regulatory fragmentation, limited investor awareness, and implementation gaps (Dahiya et al., 2026; Srivastava et al., 2026; Mishra et al., 2026), and Kumar and Sikka (2025) document comparable hurdles in India, another emerging-market setting facing green-finance implementation challenges. These institutional differences may weaken the valuation effects documented in more mature markets, a claim this study does not test directly.

2.5. Hypotheses and Research Question

Empirical evidence on the direction of these effects is mixed across markets (Section 2.2 and Section 2.3): some studies find positive announcement returns and valuation effects, others find null or even negative results, and the balance of evidence appears more consistently positive in larger, more liquid developed markets than in smaller or less mature ones. H1 and H2 are, nonetheless, stated in the positive direction predicted by signaling theory (Section 2.1)—the theoretical mechanism most commonly invoked to justify a positive effect—rather than as a prediction that this direction will necessarily hold empirically in the Thai setting; Section 2.4 sets out reasons the theoretical prediction might attenuate or fail to materialize here specifically.

H1. 

Green-bond issuance is associated with positive cumulative abnormal return (CAR) around the issue date.

H2. 

The presence of an outstanding green bond is positively associated with firm valuation, measured by Tobin’s Q, in a firm-year panel.

RQ3. Do selected issuer and bond characteristics help explain heterogeneity in issue-date CAR? Rather than a single directional hypothesis, this is treated as an exploratory research question, addressed through parsimonious models linking specific characteristics to the theoretical channels set out in Section 2.1: firm size and leverage (information asymmetry and cost-of-capital channel), credit rating and third-party certification (signaling and stakeholder-legitimacy channels), and industry (Energy and Utilities concentration, discussed in Section 2.4).

3. Materials and Methods

The analysis comprises three complementary components, each mapped to a specific hypothesis, research question, or identification concern: the event study (Section 3.2) tests H1 and, through the cross-sectional specifications in Section 3.2, RQ3; the firm-value panel regression (Section 3.3) tests H2; and the selection and endogeneity sensitivity checks (Section 3.4) address the concern that green-bond issuance is not randomly assigned, which could otherwise bias the test of H2. Estimating equations are given alongside each specification, below; this section also flags points where the implementation departs from a textbook specification, together with the reason for each departure.

3.1. Population and Sample

The population is all companies listed on the Stock Exchange of Thailand (SET) that have issued a bond classified as a “Green Bond” under ThaiBMA’s classification, identified from the ThaiBMA bond registry. Given the small size of the Thai green-bond market, the study uses census sampling, the full population of qualifying issuers, rather than a random sample. Inclusion criteria are (i) the issuer is listed on SET as a public company in its own right, rather than an unlisted subsidiary of a listed parent; (ii) the issued debt instrument is classified as a “Green Bond” under ThaiBMA’s classification; and (iii) the firm has continuous daily stock price data covering both the estimation window and the event window. Exclusion criteria are: (i) issuers not listed directly on SET (for example, foreign subsidiaries issuing offshore, or unlisted subsidiaries of a listed parent); and (ii) firm-days on which the stock was suspended from trading within the event window. Potentially confounding corporate disclosures were systematically screened within ±14 calendar days of each of the 34 event dates, covering M&A announcements, tender offers, disposals, and major shareholding changes. Rather than being excluded automatically, events with higher-risk overlapping disclosures were retained in the primary sample and separately excluded in a sensitivity analysis (Section 4.1). The final sample comprises 20 SET-listed green-bond issuers, observed repeatedly over 2015–2025 to yield 220 firm-year observations for the panel analysis; the effective cross-sectional sample for inference purposes remains 20 firms, since firm-years for the same issuer are not statistically independent, and the baseline regression (Section 3.3) uses 187 complete firm-year observations after listwise deletion. Within-firm identification is aided by the fact that 19 of the 20 sample firms exhibit an observed transition from no green bond outstanding to a green bond outstanding within the 2015–2025 panel window; the exception, Sansiri PCL, has no green bond outstanding during the panel period because its only issuance falls just after the panel’s final year (Supplementary Table S1). Most of these 19 firms experience only a single 0→1 adoption transition, rather than repeated switching, so this count indicates that within-firm variation exists but does not by itself address the possibility that the timing of adoption is correlated with time-varying firm characteristics (Section 3.4). For the event study, an initial 63 ThaiBMA-registered issuances collapse, after grouping same-day multi-tranche issuances, to 35 distinct issuance events; one of these, Event #10 (Bangkok Mass Transit System PCL), does not meet the direct-listing inclusion criterion above because the issuer is an unlisted subsidiary of BTS Group Holdings PCL and its market reaction can only be proxied via the parent company’s returns. This event is therefore excluded from all event-study specifications, leaving an analytical sample of 34 events.

3.2. Event Study

Event Day 0 is the ThaiBMA-recorded issue date, used as a proxy for the public disclosure date because the exact SET announcement date could not be independently verified for every one of the 35 events (Supplementary Table S1 documents this limitation event-by-event); this is a documented limitation, discussed further in Section 6, and means that results reported below are more precisely described as issue-date market reactions than as announcement-period reactions in the strict sense. Supplementary Table S1 lists, for each of the 35 identified events, the issuer, event date, bond type, first-versus-repeat issuance status, bond rating and certification, and confounding-disclosure screening outcome. Abnormal returns are estimated using three benchmark models over a 110-trading-day estimation window (Brown & Warner, 1980, 1985) ending 11 trading days before the event: the Market Model; a Fama–French three-factor model (Fama & French, 1993) using daily Thai market (already in excess of the risk-free rate), SMB, and HML factors from Thailand’s Factor Library (Thailand Factor Library, 2026; methodology in Charoenwong et al., 2021), with a daily Thai risk-free rate (ThaiBMA government bond yield curve, 1-month Treasury bill; ThaiBMA, 2026a) used separately to construct excess stock returns on the left-hand side; and a market-adjusted model. The annualized 1-month Treasury bill yield was converted to a daily effective rate using a 365-day convention, R_f,t = (1 + y_t/100)^(1/365) − 1, and merged with stock and factor returns by exact calendar date; trading days without a corresponding published yield quotation were excluded from the excess-return calculation for that date, rather than imputed. The primary event window is [−1, +1] trading days (MacKinlay, 1997), chosen to minimize contamination from unrelated news while still capturing the issue day and immediate market response; [−3, +3], [−5, +5], and [−10, +10] are reported as robustness windows. Repeat issuances are treated as separate event observations because each issuance discloses new, issuance-specific information (amount, terms, and use-of-proceeds allocation) at a distinct point in time, even though a firm’s overall green-bond commitment is already known to the market after its first issuance; this is a standard event-study convention, but means event observations from the same firm are not fully independent. Because 14 of the 34 analytical events are repeat issuances by firms that appear more than once in the sample, results are also reported (a) collapsed to one average CAR per firm, and (b) restricted to first-issuance events only. The first-issuance-only cut is the theoretically sharper test of the signaling channel discussed in Section 2.1, since a first issuance plausibly conveys more new information to investors than a subsequent one, and is treated as a key robustness check, rather than a secondary one. Significance is assessed with a parametric cross-sectional t-test, the Patell (1976) standardized-residual test, and the non-parametric Corrado (1989) rank test.

Formally, letting i index firms and t index trading days:

R_i,t = α_i + β_iR_m,t + ε_i,t

(1)

$$A{R}_{i,t}=R_{i,t}-\left({\widehat{\alpha }}_i+{\widehat{\beta }}_i{R}_{m,t}\right)$$

(2)

$$\mathit{CA}{R}_i\left(t_1,t_2\right)={\displaystyle \sum _{t=t_1}^{t_2}}A{R}_{i,t} ; \mathit{CAAR}\left(t_1,t_2\right)={\displaystyle \frac{1}{N}}\sum _{i}CA{R}_i\left(t_1,t_2\right)$$

(3)

R_i,t − R_f,t = α_i + β_MKTMKT_t + β_SMBSMB_t + β_HMLHML_t + ε_i,t

(4)

3.3. Firm-Value Panel Regression

Tobin’s Q is a market-based valuation measure, approximated here as (market capitalization + total liabilities)/total assets, and is used as a proxy for medium-term firm valuation, rather than a direct measure of shareholder value or a statistical test in itself; it captures a different dimension of market performance from the short-term issue-date abnormal returns examined in Section 3.2, and the two should not be treated as interchangeable operationalizations of a single underlying construct.

The baseline specification regresses Tobin’s Q on a green-bond-outstanding indicator and firm-level controls, with firm and year fixed-effects. The controls address the most commonly documented correlates of both green-bond issuance and Tobin’s Q in the corporate finance literature: firm size (larger firms have greater access to capital markets, including green-labeled debt, and differ systematically in valuation); leverage (capital-structure decisions correlate with both financing choices and valuation); return on assets (profitability affects both the capacity to issue debt on favorable terms and market valuation); and sales growth (growth firms may be more likely to fund expansion with new debt instruments, including green bonds, and are typically valued differently by the market). These variables are not intended as a complete model of selection into green-bond issuance, and Section 3.4 addresses selection more directly. Firm age is examined only in an alternative industry-and-year fixed-effects specification (Appendix B, Table A3), because it is mechanically collinear with year fixed-effects once firm fixed-effects are included (an age-period-cohort identification problem). With only 20 firm clusters, conventional cluster-robust standard errors (Petersen, 2009) can understate estimation uncertainty; the green-bond coefficient is therefore additionally reported with CR2 standard errors and Satterthwaite degrees of freedom, and with a wild (Rademacher) cluster bootstrap-t p-value following Cameron et al. (2008), alongside the conventional cluster-robust result. Model specification tests reported include a Wald test of coefficient equality between fixed-effects and pooled-OLS estimates, the Breusch–Pagan Lagrange multiplier test, variance inflation factors, the Wooldridge test for serial correlation, and a Modified Wald test for groupwise heteroskedasticity (Appendix B, Table A3).

Q_i,t = α_i + γ_t + β₁GreenBond_i,t + β₂X_i,t + u_i,t

(5)

3.4. Selection and Endogeneity Sensitivity Checks

Green-bond issuance is not randomly assigned: firms already predisposed toward higher valuations, or operating in sectors with easier access to green-labeled debt, may be the ones that choose to issue. Firms may also be more likely to issue when they are already entering capital-intensive investment cycles, restructuring their capital base, facing heightened regulatory or investor pressure to demonstrate environmental commitments, or able to access unusually favorable financing conditions—any of which could independently affect Tobin’s Q. Because the baseline panel identifies within-firm variation among issuers, without a matched non-issuer control group, it cannot fully separate the effect of green-bond issuance from these correlated forces. Two complementary specifications examine the sensitivity of the panel result to this concern. Propensity-score matching (Rosenbaum & Rubin, 1983; Austin, 2011) combined with difference-in-differences (PSM-DiD) matches treated firms to non-issuing candidates within industry on size, ROA, and leverage; a panel DiD is estimated for the Energy and Utilities sub-sample (the only group with multi-year matched-control data), complemented by a matched-pairs comparison for the remaining six industries. Matching with replacement was necessary because the candidate pool was smaller than the number of treated Energy and Utilities firms. A Heckman two-stage model estimates a firm-level probit selection equation using an industry-level peer-issuance-rate exclusion restriction, with the resulting inverse Mills ratio included in an industry-and-year (rather than firm) fixed-effects outcome equation (Heckman, 1979). Neither specification is presented as identifying a causal effect: matched-sample standardized covariate bias remains above conventional thresholds for several covariates even after matching, and the Heckman exclusion restriction (industry peer-issuance rate) could plausibly affect firm value through channels other than selection, such as sector-wide investor sentiment. Both are therefore reported as sensitivity checks on the baseline panel result, not as endogeneity corrections in the causal-inference sense.

$$Y_{i,t}={\gamma }_i+{\gamma }_t+\delta \left(\mathit{Treate}{d}_i\times \mathit{Pos}{t}_t\right)+{\gamma }_2{X}_{i,t}+u_{i,t} ; \mathit{ATT}=\widehat{\delta }$$

(6)

Pr(GreenBond_i,t = 1) = Φ(δ₀ + δ₁PeerIssuanceRate_industry,t + δ₂X_i,t)

(7)

Q_i,t = α + β GreenBond_i,t + β₂X_i,t + λ IMR_i,t + u_i,t

(8)

An exploratory instrumental-variables (2SLS) specification, instrumenting issuance with an industry-year peer-issuance rate, appears only in Appendix A: the first-stage F-statistic falls well below conventional weak-instrument thresholds (Staiger & Stock, 1997; Stock & Yogo, 2005), and the resulting estimate is not interpreted substantively.

3.5. Robustness Checks

Robustness checks include alternative event windows and benchmark models (Section 3.2); a placebo/falsification test (500 iterations, each drawing one randomly assigned fake date per firm-event, at least 150 trading days from every real event date for that firm, seed = 20260805); panel diagnostic tests including tests for serial correlation and groupwise heteroskedasticity (Appendix B, Table A3); an alternative industry-and-year fixed-effects specification including firm age (Appendix B, Table A4); lagged and led green-bond indicators (Appendix B, Table A5); return on assets as an alternative outcome, and winsorization at the 1st and 99th percentiles (Appendix B, Table A6). All analyses were performed in R version 4.5.2 (R Core Team, 2025).

3.6. Data

Firm financial data were hand-collected from audited financial statements, Form 56-1/One Report filings, and SET Corporate Actions capital-movement records for all 20 sample firms across the 2015–2025 panel window, cross-verified against multiple primary sources. Daily stock price/volume data and daily Thai market and factor returns were obtained from SET/SETSMART-sourced series. Bond credit ratings and certification standards were obtained from a ThaiBMA green-bond registry extract. Daily Thai SMB and HML factor returns were obtained from Thailand’s Factor Library (Thailand Factor Library, 2026), supported by the Thailand Capital Market Development Fund, the Capital Market Research Institute, and SETSMART Enterprise of the Stock Exchange of Thailand; the underlying methodology is described in Charoenwong et al. (2021). Table 1 reports descriptive statistics for the firm-year panel.

Table 1. Descriptive Statistics (Firm-Year Panel, N = 220 firm-years, 20 firms, 2015–2025).

Variable	N	Mean	SD	Min	Median	Max
Tobin’s Q	199	1.416	0.960	0.621	1.134	10.257
Total Assets (THB mm)	208	134,676	122,106	4963	94,791	773,810
Firm Size (ln Assets)	208	11.391	0.973	8.510	11.459	13.559
ROA	208	0.038	0.049	−0.084	0.033	0.462
Leverage	208	0.581	0.151	0.122	0.586	0.937
Sales Growth (YoY)	189	0.204	1.299	−0.510	0.041	17.052
Firm Age (years)	199	15.38	12.20	0	12	47

Note: Sales Growth is heavily right-skewed by a small number of genuine, explainable business events (notably a post-IPO consolidation jump); winsorized results at the 1st/99th percentile are reported alongside the baseline model, as a robustness check (Appendix B, Table A6).

Of the 220 firm-year observations, 61 (28%) are green-bond-outstanding firm-years. Thirteen of the 20 sample firms (65%) are classified in the Energy and Utilities sector, reflecting the genuine composition of Thailand’s green-bond issuer base; this concentration is addressed in the issue-date heterogeneity analysis (Section 4.1) and the Discussion (Section 5).

This study relies exclusively on secondary company and financial-market data, including publicly available company filings and ThaiBMA records, as well as licensed SETSMART data; it does not involve human participants, surveys, or personally identifiable information.

3.7. Use of Generative AI Tools

Claude Sonnet 5 Medium (Anthropic, San Francisco, CA, USA) and GPT-5.5 Instant (OpenAI, San Francisco, CA, USA) were used to assist with language editing, manuscript organization, and the drafting of selected data-analysis code. All code, statistical outputs, source interpretations, and manuscript text were independently checked and validated by the authors, who take full responsibility for the accuracy and integrity of the work.

4. Results

4.1. Event Study

Table 2 reports CAAR(−1, +1), the primary window, under three benchmark models. Under both the Market Model and the full three-factor Fama–French specification, CAAR is small, negative-to-positive across specifications, and statistically indistinguishable from zero. A placebo test drawing, in each of 500 iterations, one random fake date per firm-event (at least 150 trading days from every real event date for that firm, not only the event being replaced, seed = 20260805), produces an empirical distribution of placebo CAARs centered near zero (mean −0.0002, SD 0.0060, 95% empirical range [−0.0125, 0.0118]); the real primary-window CAAR of 0.0008 falls well within this range (empirical p = 0.878), consistent with the benchmark-model results not reflecting a mechanical artifact of the estimation procedure.

Table 2. CAAR(−1, +1) Under Three Benchmark Models.

Model	N	CAAR	t-Statistic	p-Value
Market Model	34	0.0008	0.136	0.892
Fama–French 3-Factor (excess returns)	34	−0.0002	−0.040	0.968
Market-Adjusted	34	−0.0034	−0.603	0.551

Note: Estimation window [−120, −11] trading days. N = 34 across all three models reflects the exclusion of Event #10 (Bangkok Mass Transit System), whose issuer is not listed directly on SET in its own right and whose market reaction can only be proxied via the parent company, BTS Group Holdings; this does not meet the direct-listing inclusion criterion set out in Section 3.1 and is therefore excluded from all specifications, not only those requiring an estimation window. The Fama–French specification uses excess returns constructed from a daily Thai risk-free rate (ThaiBMA 1-month Treasury bill yield) and Thai daily market excess-return, SMB, and HML factors.

Table 3 reports CAAR under the [−1, +1] primary window and the [−3, +3], [−5, +5], and [−10, +10] robustness windows, under the Market Model.

Table 3. CAAR Across Event Windows (Market Model).

Window	N	CAAR	t-Statistic	p-Value
[−1, +1] (primary)	34	0.0008	0.136	0.892
[−3, +3]	34	0.0034	0.522	0.605
[−5, +5]	34	0.0045	0.483	0.632
[−10, +10]	34	−0.0055	−0.442	0.661

Note: no window, from the narrowest to the widest tested, yields a statistically significant CAAR; the primary [−1, +1] window CAAR of 0.0008 has a 95% confidence interval of [−0.0112, 0.0128]. The Patell (1976) and Corrado (1989) test statistics, computed alongside the cross-sectional t-test for all four windows, confirm the same conclusion (Appendix B, Table A7).

Because 14 of the 34 analytical events are repeat issuances, Table 3 findings are supplemented by two additional cuts. Collapsed to one average CAR per firm (N = 20), the mean [−1, +1] CAR is −0.0013 (t = −0.277, 95% CI [−0.0113, 0.0087]); restricted to first-issuance events only (N = 20), the mean [−1, +1] CAR is −0.0049 (t = −0.605, 95% CI [−0.0216, 0.0119]). Neither estimate is statistically significant. Given the small samples (N = 20) and the width of the confidence intervals, these analyses should be interpreted as failing to detect an average issue-date reaction, rather than as establishing that pooling repeat issuances is immaterial or that economically meaningful effects are absent. Taken together, these results do not support H1: issue-date cumulative abnormal returns are not statistically distinguishable from zero under any benchmark model, event window, or issuer-pooling choice tested.

RQ3 asks whether selected issuer and bond characteristics help explain heterogeneity in issue-date CAR. Rather than a single richly parameterized model, three parsimonious specifications are estimated, each linking a small set of characteristics to the theoretical channels in Section 2.1, with firm-clustered standard errors throughout.

Table 4 reports two coefficients provide weakly suggestive evidence at the 10% level, but neither is statistically significant at the conventional 5% level. The negative coefficient on the numeric rating scale suggests that higher-rated issues may receive more favorable issue-date reactions; however, the estimate is significant only at the 10% level, and does not establish either a credit-quality or certification mechanism, particularly since certification itself is not statistically significant in the same model. Energy and Utilities issuers show higher CAR than other industries, controlling for size and leverage. Both results are reported as suggestive, rather than confirmatory, given the small sample and the number of specifications examined. In answer to RQ3, credit rating and industry membership each show weakly suggestive associations with issue-date CAR at the 10% level, while size, leverage, and certification show none; none of these characteristics is significant at the conventional 5% level. Given the small number of events and the number of characteristics examined, these associations are also subject to a multiple-testing concern that this analysis does not formally correct for; they are, accordingly, presented as preliminary and exploratory, rather than as definitive evidence of heterogeneity.

Table 4. RQ3: Issue-Date CAR and Issuer/Bond Characteristics.

Model/Variable	Coefficient	Cluster SE	p-Value
Model 1: Size + Leverage
Size	0.0037	0.0060	0.544
Leverage	0.0342	0.0768	0.662
Model 2: Rating + Certification
Rating (1 = AAA…16 = B-)	−0.0028	0.0015	0.076
Certification (CBS = 1)	0.0203	0.0196	0.314
Model 3: Energy & Utilities + controls
Energy & Utilities dummy	0.0188	0.0105	0.090
Size	0.0056	0.0058	0.344
Leverage	0.0369	0.0679	0.594

Note: dependent variable is CAR(−1, +1). N = 34 (Model 1, 3); N = 33 for Model 2, reflecting the 34-event analytical sample less Event #8 (B.Grimm Power, issued 6 July 2021), which is absent from the ThaiBMA ESG bond registry used to construct the rating and certification variables, a documented data-availability gap, rather than an exclusion choice.

4.2. Firm-Value Panel Regression

Table 5 reports the baseline two-way fixed-effects regression of Tobin’s Q on the green-bond-outstanding indicator and firm-level controls.

Table 5. Panel Regression of Tobin’s Q on Green-Bond Issuance (Firm and Year Fixed-Effects).

Variable	Coefficient	Std. Error	p-Value
GreenBond (outstanding)	0.066	0.091	0.478
Firm Size (ln Assets)	0.010	0.027	0.716
ROA	−0.535	0.573	0.359
Leverage	0.029	0.151	0.849
Sales Growth	0.007	0.136	0.960

Note: N = 187 complete cases; number of firms = 20; firm fixed-effects = yes; year fixed-effects = yes; within R-squared = 0.650; conventional firm-clustered standard errors (20 clusters).

With only 20 firm clusters, conventional cluster-robust inference can be unreliable. Table 6 reports the GreenBond coefficient under CR2 standard errors with Satterthwaite degrees of freedom, and under a wild (Rademacher) cluster bootstrap-t procedure (999 replications).

Table 6. GreenBond Coefficient: Small-Cluster-Robust Inference.

Statistic	Value
Coefficient	0.066
Conventional cluster-robust (CR1) SE	0.091
CR1 95% CI	[−0.112, 0.243]
CR2 SE (Satterthwaite df = 6.1)	0.083
CR2 95% CI	[−0.138, 0.269]
Conventional cluster-robust p-value	0.478
Wild cluster bootstrap-t p-value (B = 999)	0.498

Note: the wild-cluster-bootstrap p-value is close to the conventional cluster-robust p-value, indicating that the small number of firm clusters does not materially change inference for the GreenBond coefficient.

Taken together, the panel results do not support H2: the GreenBond coefficient is not statistically significant under either conventional or small-cluster-robust inference, and the CR2 95% confidence interval [−0.138, 0.269] is wide enough to admit economically meaningful positive or negative effects, so this result should be read as an absence of a detectable average association, rather than evidence that no effect exists. Thirteen of the 20 sample firms are Energy and Utilities issuers; because valuations in this sector may be shaped by regulation, capital intensity, and ownership structure, rather than discretionary market signaling, this concentration is a further reason the panel estimate is best interpreted as specific to this issuer population, not a general effect of green-bond issuance on firm valuation.

4.3. Selection and Endogeneity Sensitivity Checks

Propensity-score matching was implemented separately for the Energy and Utilities sub-sample (panel DiD, 13 treated firms) and, using matched pairs, for the remaining six industries. The Energy and Utilities panel DiD ATT, using ROA as the outcome (matched-control data for Tobin’s Q were not available for this sub-sample), is 0.016 (p = 0.109) and is therefore not directly comparable in magnitude with the Tobin’s Q baseline estimate in Table 5; a matched-pairs estimate for one of the six remaining industries is significant (p = 0.041) but does not survive a leave-one-out sensitivity check, and is therefore interpreted as fragile, rather than confirmatory. Standardized covariate bias after matching remains above conventional thresholds (10%) for several covariates, so these estimates are reported as sensitivity checks on the baseline panel, result rather than as evidence of a causal effect.

The Heckman two-stage model’s first-stage selection equation, using an industry-level peer-issuance rate as the exclusion restriction, is a statistically significant predictor of issuance status. The resulting inverse Mills ratio is not significant in the outcome equation (p = 0.646), and the green-bond coefficient changes little between corrected and uncorrected specifications; this provides no evidence that the modeled selection mechanism materially changes the baseline estimate, though it does not rule out selection operating through channels the model does not capture, including the exclusion restriction’s own plausibility, as discussed in Section 3.4.

An exploratory instrumental-variables specification is reported in Appendix A. Its first-stage F-statistic falls well below conventional weak-instrument thresholds, and the resulting 2SLS estimate is not interpreted substantively.

5. Discussion

The primary event-study and panel estimates are each statistically indistinguishable from zero, but they concern different outcomes and are not treated as mutually confirming evidence. The matching and Heckman analyses remain inconclusive because of residual covariate imbalance, outcome non-comparability, and limitations of the exclusion restriction; they are therefore not used to corroborate the baseline results. The available data do not distinguish among the following explanations. They are discussed below as possible interpretations of the results. Placed alongside findings from other developing markets, these considerations provide context for why the Thai estimates may differ from results reported in some developed markets.

In terms of the study’s hypotheses, neither H1 (positive issue-date abnormal returns) nor H2 (a positive association between green-bond issuance and Tobin’s Q) are supported: neither result is statistically distinguishable from zero under the primary specifications, and both confidence intervals remain wide enough to admit economically meaningful effects in either direction. RQ3 finds weakly suggestive, non-conventional-significance associations between issue-date CAR and both credit rating and Energy and Utilities industry membership. Signaling theory (Section 2.1) predicts that a green-bond announcement conveys information about a firm’s environmental commitment and, indirectly, its financial and governance quality; the absence of a detectable market or valuation reaction is consistent with that signal not yet being priced distinctly from other available information in a market of this size and concentration, rather than necessarily reflecting a Thailand-specific institutional peculiarity. Prior developed-market evidence of a positive reaction (e.g., Flammer, 2021) and mixed ASEAN evidence (e.g., Aini et al., 2023) both fit within this same signaling framework: the channel is theoretically available, but its empirical strength appears to depend on market conditions that this study cannot fully disentangle from country-specific characteristics with a single-country census.

One possible interpretation concerns market maturity. The Thai green-bond market remains relatively young and concentrated next to the markets studied by Flammer (2021) and Glavas (2020), and the absence of a detectable issue-date reaction accords with the limited market depth discussed in Section 2.4, though the available data cannot isolate this channel from others. Related studies link green-finance effectiveness to regulatory oversight and secondary-market liquidity (Ramos Farroñán et al., 2025) and show that relationships between financial development and environmental outcomes differ across mature and emerging economies (Tuncel & Yaman, 2026).

The sample’s concentration in regulated Energy and Utilities firms, 13 of the 20 sample firms, may also matter. Regulatory and ownership characteristics in this sector could affect how investors interpret green-bond issuance and may therefore moderate the signaling, cost-of-capital, and stakeholder-legitimacy channels set out in Section 2.1. The RQ3 finding of a weakly suggestive Energy and Utilities coefficient on issue-date CAR (Section 4.1, Table 4) fits this interpretation, though at a 10% threshold and across a single specification, it should not be read as confirming it. State-linked ownership itself seems to shape how green-bond effects transmit, though not in one direction: L. Liu et al. (2025) find the governance benefit stronger among non-state-owned Chinese issuers; Chang et al. (2025) find the innovation-stimulating benefit stronger among state-owned ones. Ownership and administrative context appear to matter, even if the net effect cannot be signed ahead of time with the present data.

The negative coefficient on the numeric rating scale in Table 4 is consistent with higher-rated issues receiving more favorable issue-date reactions. However, the estimate is significant only at the 10% level, and does not establish a certification or signaling mechanism; certification itself, tested separately in the same model, is not statistically significant. This result is offered as a direction for future research, rather than a confirmed channel.

A similar null result is reported by Aini et al. (2023), who find no significant stock-market response in a five-country ASEAN sample, and by Wanganusorn et al. (2026), who find Thai green-bond issuance volume unrelated to bond return or risk. The estimates here diverge, however, from the positive post-announcement effects reported by Sukcharoensin and Apinyakul (2026) for Thai ESG bonds; differences in bond coverage, event definitions, and benchmark models may partly explain the contrasting results.

Evidence based on macroeconomic and environmental outcomes is less directly comparable. At the macroeconomic level, Phung Thanh (2022) and Kong et al. (2026) report no uniform direct growth effect; the estimated relationships vary across income groups and levels of financial development. Studies reporting effects on renewable energy deployment and emissions reduction (Dong et al., 2023; Ahmed et al., 2022; Chowiendo et al., 2025), with effects surfacing only in the long run (Agustin et al., 2025), and qualitative accounts of financing gaps and immature infrastructure (Le & Vivian, 2025; Lin et al., 2026; Dahiya et al., 2026; Kumar & Sikka, 2025; Srivastava et al., 2026; Mishra et al., 2026), can provide institutional context, but they do not establish whether green-bond issuance affects issuer valuation. Even in the U.S. and European settings where a positive firm-value effect was first documented, the picture is less uniformly positive than the earliest studies suggested (Pietrzyk et al., 2025; Aleknevičienė & Vilutytė, 2024; Ordonez-Borrallo et al., 2026).

Wanganusorn et al. (2026) document Thai green-bond market behavior at the bond level, reporting a positive relationship between bond return and risk, but no relationship between issuance volume and either. The present study does not provide evidence that these bond-level relationships are accompanied by detectable issuer-level valuation effects; however, the two studies examine different outcomes, and are not a direct transmission test. The Thai estimates resemble those from selected studies reporting muted market or valuation responses in ASEAN, although direct comparisons are limited by differences in outcomes and research designs; the present study extends this line of evidence to Thailand specifically at the firm level, while examining the sensitivity of the result to selection and endogeneity concerns.

For issuers, the results do not provide evidence that the green-bond label alone generates an immediate equity-market premium in the current Thai market; if a valuation benefit exists, it is not detectable, on average, in this sample, though the suggestive rating association means credit quality may still be worth emphasizing in investor communications around an issuance. For investors, the absence of a detectable average reaction does not imply that green-bond issuance is uninformative in every individual case, only that no systematic average pattern is detectable across this population; investors evaluating a specific issuance may still find firm- or issue-specific characteristics, such as rating or industry, informative, though the present evidence for these is only weakly suggestive. For regulators and market infrastructure bodies, the findings are silent on whether green bonds improve underlying project financing or environmental outcomes, since this study examines shareholder value, rather than project-level effects; to the extent market depth and investor recognition condition the value relevance of green labels, as Section 5 discusses, continued efforts to deepen secondary-market liquidity and standardize certification practices remain relevant, independent of this study’s null finding on shareholder value specifically.

6. Conclusions

This study contributes to the green-bond literature by testing whether the valuation implications documented in larger developed markets are also observed in Thailand’s small, concentrated market. Neither H1 (a positive issue-date market reaction) nor H2 (a positive association between green-bond issuance and firm valuation) are supported: under the primary event-study benchmarks, the panel specification, and small-cluster-robust inference for the panel result, neither a statistically detectable positive issue-date reaction nor a statistically detectable positive association with Tobin’s Q is observed. Although isolated alternative specifications yield significant estimates, these results are model-sensitive or fail sensitivity checks, leaving no evidence of a robust valuation premium. The results are consistent with, but do not establish, the possibility that green-bond issuance has limited average value relevance in Thailand’s current market setting, at this stage of development of the country’s sustainable finance market and given the concentration of issuers in regulated Energy and Utilities firms. As Section 5 discusses, the absence of a robust positive valuation effect is consistent with a broader Asian and ASEAN literature reporting limited or conditional effects of green finance. Much of that evidence, however, concerns macroeconomic or environmental outcomes, a different question from firm value, so the present results cannot establish whether the Thai finding reflects a broader developing-market pattern. Future research, ideally revisited as the Thai green-bond market matures, could usefully extend this design to sustainability-linked bonds and to a comparative cross-country ASEAN panel that tests directly whether market liquidity, state ownership, and financial development jointly explain the cross-market heterogeneity discussed in this study.

Several limitations should inform interpretation. First, Event Day 0 is operationalized as the ThaiBMA-recorded issue date, rather than an independently verified SET public-disclosure date. If the true disclosure date differs from the issue date for some events, the event-study results are better described as issue-date market reactions than as announcement-period reactions in the strict sense. If the true disclosure date falls outside the selected event window, the estimated abnormal return may be attenuated or otherwise distorted; however, the direction and magnitude of any timing bias cannot be determined from the available data. A specific version of this concern is that information may reach the market before the issue date—for example, through pre-issuance disclosures, roadshows, or underwriter communications—in which case, the [−1, +1] window centered on the issue date could miss the informative event entirely, rather than merely mismeasure it. This earlier-information-leakage scenario is a plausible contributing explanation for the study’s null findings that the present data cannot rule out. Second, the census-based sample (20 firms and 34 analytical events) is a deliberate design choice, given the small size of the Thai green-bond market, but it limits statistical power, particularly for the selection-sensitivity sub-sample analyses in Section 4.3. Propensity-score matching did not achieve full covariate balance by conventional standards, given genuine common-support constraints. The instrumental-variables estimate, reported in Appendix A, is undermined by a weak first stage concentrated in a single industry sub-sample, and is not interpreted substantively. Bond rating and certification data were unavailable for one of the 34 analytical events (Event #8, absent from the ThaiBMA ESG bond registry used to construct these variables); the RQ3 analysis in Section 4.1 is therefore based on 33 of 34 events. The systematic confounding-event screening described in Section 3.1 flagged two higher-risk events (Events 7 and 26); excluding them left the Section 4.1 conclusion unchanged (N = 32, CAAR = 0.0006, t = 0.105, p = 0.917, versus N = 34, CAAR = 0.0008, t = 0.136, p = 0.892 for the full sample), but the underlying corporate actions themselves (an acquisition and a tender offer) were not separately modeled, and remain a residual source of noise in those two event-level CAR estimates specifically. With only 20 firm clusters in the panel regression, conventional cluster-robust inference could understate estimation uncertainty; this is addressed directly in Section 4.2 using CR2 standard errors and a wild cluster bootstrap-t procedure, both of which are close to the conventional result and so do not suggest materially different conclusions. Finally, the comparative discussion in Section 5, situating this study’s null finding alongside other Asian and developing-market evidence, draws on studies that differ in methodology, sample period, and outcome variable (several examine macroeconomic or environmental outcomes instead of firm value or stock returns); this comparison is therefore best read as suggestive of a regional pattern, rather than as a formal cross-country meta-analysis.

Appendix A.1. Exploratory Instrumental-Variable Diagnostic

An instrumental-variables (2SLS) specification instrument green-bond issuance with an industry-year peer-issuance rate. The first-stage F-statistic is 0.77, well below the conventional rule-of-thumb value of 10 used in the weak-instrument literature (Staiger & Stock, 1997; Stock & Yogo, 2005). The proposed instrument does not provide sufficiently strong identifying variation, and the resulting 2SLS estimate is therefore not interpreted substantively; it is reported here only for transparency and completeness, and should not be read as a robustness check that supports or undermines the paper’s main conclusions.

Table A1. Exploratory IV/2SLS Estimate (Not Interpreted Substantively).

Statistic	Value
First-stage F-statistic	0.77
2SLS coefficient (GreenBond)	−2.854
2SLS p-value	0.559

Table A1. Exploratory IV/2SLS Estimate (Not Interpreted Substantively).

Statistic	Value
First-stage F-statistic	0.77
2SLS coefficient (GreenBond)	−2.854
2SLS p-value	0.559

Appendix A.2. Variable Definitions

Table A2. Variable Definitions.

Variable	Definition	Source
Tobin’s Q	Approximate Tobin’s Q: (Market Cap + Total Liabilities)/Total Assets	SET/SETSMART prices; audited financials
GreenBond	1 if firm has a green bond outstanding in year t, 0 otherwise	ThaiBMA bond registry
Size	Natural log of total assets (THB mm)	Audited financial statements
Leverage	Total Liabilities/Total Assets	Audited financial statements
ROA	Net income attributable to owners/Total Assets	Audited financial statements
Sales Growth	Year-over-year % change in total revenue	Audited financial statements
Firm Age	Current year minus year of SET listing/IPO	SET factsheets
Rating	Bond-issue credit rating, converted to a 1–16 numeric notch scale (1 = AAA)	ThaiBMA ESG bond registry
Certification	1 if bond standards include Climate Bonds Standard (CBS), 0 otherwise	ThaiBMA ESG bond registry
Industry	SET official Sector classification (8 categories)	SET listed company master list

Note: The Tobin’s Q proxy follows the approximation of Chung and Pruitt (1994).

Table A2. Variable Definitions.

Variable	Definition	Source
Tobin’s Q	Approximate Tobin’s Q: (Market Cap + Total Liabilities)/Total Assets	SET/SETSMART prices; audited financials
GreenBond	1 if firm has a green bond outstanding in year t, 0 otherwise	ThaiBMA bond registry
Size	Natural log of total assets (THB mm)	Audited financial statements
Leverage	Total Liabilities/Total Assets	Audited financial statements
ROA	Net income attributable to owners/Total Assets	Audited financial statements
Sales Growth	Year-over-year % change in total revenue	Audited financial statements
Firm Age	Current year minus year of SET listing/IPO	SET factsheets
Rating	Bond-issue credit rating, converted to a 1–16 numeric notch scale (1 = AAA)	ThaiBMA ESG bond registry
Certification	1 if bond standards include Climate Bonds Standard (CBS), 0 otherwise	ThaiBMA ESG bond registry
Industry	SET official Sector classification (8 categories)	SET listed company master list

Note: The Tobin’s Q proxy follows the approximation of Chung and Pruitt (1994).

Appendix B.1. Panel Diagnostic Tests

Table A3. Panel Diagnostic Tests.

Test	Statistic	df	p-Value
Breusch–Pagan LM (pooled OLS vs. panel structure)	χ2 = 202.43	1	<0.001
Wald test of coefficient equality (FE vs. pooled OLS) †	χ2 = 19.88	5	0.001
Wooldridge test for serial correlation	t = 2.33	185	0.021
Modified Wald test for groupwise heteroskedasticity	χ2 = 296.58	20	<0.001
VIF (maximum, across regressors)	1.50 (Leverage)	–	–

† The Wald test compares the fixed-effects and pooled-OLS coefficient vectors directly (base R implementation); it is distinct from the standard Hausman test, which compares fixed-effects against a random-effects GLS estimator and is not reported here. All VIF values are below 2, indicating no material multicollinearity. The Breusch–Pagan and Modified Wald results support the use of a panel structure with heteroskedasticity-robust inference, consistent with the clustered and CR2 standard errors reported in Section 4.2. The Wooldridge test (based on an AR(1) auxiliary regression of first-differenced residuals, following Wooldridge, 2010) rejects the null of no serial correlation at the 5% level, suggesting some within-firm serial correlation; the CR2 standard errors and wild cluster bootstrap-t in Table 6 are robust to this.

Table A3. Panel Diagnostic Tests.

Test	Statistic	df	p-Value
Breusch–Pagan LM (pooled OLS vs. panel structure)	χ2 = 202.43	1	<0.001
Wald test of coefficient equality (FE vs. pooled OLS) †	χ2 = 19.88	5	0.001
Wooldridge test for serial correlation	t = 2.33	185	0.021
Modified Wald test for groupwise heteroskedasticity	χ2 = 296.58	20	<0.001
VIF (maximum, across regressors)	1.50 (Leverage)	–	–

† The Wald test compares the fixed-effects and pooled-OLS coefficient vectors directly (base R implementation); it is distinct from the standard Hausman test, which compares fixed-effects against a random-effects GLS estimator and is not reported here. All VIF values are below 2, indicating no material multicollinearity. The Breusch–Pagan and Modified Wald results support the use of a panel structure with heteroskedasticity-robust inference, consistent with the clustered and CR2 standard errors reported in Section 4.2. The Wooldridge test (based on an AR(1) auxiliary regression of first-differenced residuals, following Wooldridge, 2010) rejects the null of no serial correlation at the 5% level, suggesting some within-firm serial correlation; the CR2 standard errors and wild cluster bootstrap-t in Table 6 are robust to this.

Appendix B.2. Alternative Panel Specification: Industry and Year Fixed-Effects, Including Firm Age

Table A4. Panel Regression of Tobin’s Q with Industry and Year Fixed-Effects, Including Firm Age.

Variable	Coefficient	Cluster SE	p-Value
GreenBond (outstanding)	0.217	0.266	0.425
Firm Size (ln Assets)	0.104	0.133	0.443
ROA	3.552	2.309	0.141
Leverage	1.037	1.165	0.390
Sales Growth	0.445	0.029	<0.001
Firm Age	−0.023	0.010	0.029

Note: N = 187. Industry and year fixed-effects (firm age is separately identified because it is not collinear with industry fixed-effects, unlike with firm fixed-effects). GreenBond remains statistically insignificant in this alternative specification, consistent with the firm-and-year-FE baseline in Table 5.

Table A4. Panel Regression of Tobin’s Q with Industry and Year Fixed-Effects, Including Firm Age.

Variable	Coefficient	Cluster SE	p-Value
GreenBond (outstanding)	0.217	0.266	0.425
Firm Size (ln Assets)	0.104	0.133	0.443
ROA	3.552	2.309	0.141
Leverage	1.037	1.165	0.390
Sales Growth	0.445	0.029	<0.001
Firm Age	−0.023	0.010	0.029

Note: N = 187. Industry and year fixed-effects (firm age is separately identified because it is not collinear with industry fixed-effects, unlike with firm fixed-effects). GreenBond remains statistically insignificant in this alternative specification, consistent with the firm-and-year-FE baseline in Table 5.

Appendix B.3. Lagged and Led GreenBond Indicators

Table A5. Panel Regression of Tobin’s Q on Lagged and Led GreenBond Indicators.

Specification	N	Coefficient	Cluster SE	p-Value
GreenBond (t − 1)	187	0.054	0.095	0.575
GreenBond (t − 2)	174	−0.004	0.111	0.975
GreenBond (t + 1, led)	167	0.009	0.094	0.924

Note: each row is a separate two-way fixed-effects regression, otherwise identical to the Table 5 baseline, replacing contemporaneous GreenBond with the indicated lag or lead. None of the lagged, led, or contemporaneous (Table 5) specifications is statistically significant, and the led indicator in particular shows no evidence of a pre-existing trend, though with only 20 firms this diagnostic has low power, and a modest pre-trend could go undetected.

Table A5. Panel Regression of Tobin’s Q on Lagged and Led GreenBond Indicators.

Specification	N	Coefficient	Cluster SE	p-Value
GreenBond (t − 1)	187	0.054	0.095	0.575
GreenBond (t − 2)	174	−0.004	0.111	0.975
GreenBond (t + 1, led)	167	0.009	0.094	0.924

Note: each row is a separate two-way fixed-effects regression, otherwise identical to the Table 5 baseline, replacing contemporaneous GreenBond with the indicated lag or lead. None of the lagged, led, or contemporaneous (Table 5) specifications is statistically significant, and the led indicator in particular shows no evidence of a pre-existing trend, though with only 20 firms this diagnostic has low power, and a modest pre-trend could go undetected.

Appendix B.4. Alternative Standard Errors and Outcome Variable

Table A6. GreenBond Coefficient: Alternative Standard Errors, Outcome Variable, and Winsorization.

Specification	N	Coefficient	SE	p-Value
Baseline (Table 5, for reference)	187	0.066	0.091 (CR1)	0.478
ROA as alternative outcome	189	−0.024	0.024 (CR1)	0.339
Winsorized (1st/99th pct)	187	0.028	0.080 (CR1)	0.735

Note: GreenBond remains statistically insignificant under the alternative ROA outcome and winsorization, consistent with the baseline result in Table 5 and Table 6.

Table A6. GreenBond Coefficient: Alternative Standard Errors, Outcome Variable, and Winsorization.

Specification	N	Coefficient	SE	p-Value
Baseline (Table 5, for reference)	187	0.066	0.091 (CR1)	0.478
ROA as alternative outcome	189	−0.024	0.024 (CR1)	0.339
Winsorized (1st/99th pct)	187	0.028	0.080 (CR1)	0.735

Note: GreenBond remains statistically insignificant under the alternative ROA outcome and winsorization, consistent with the baseline result in Table 5 and Table 6.

Appendix B.5. Patell and Corrado Test Results

Section 3.2 specifies the Patell (1976) standardized-residual test and the Corrado (1989) rank test alongside the cross-sectional t-test reported in Table 3. Results for all three tests, across all four event windows, are reported here.

Table A7. CAAR Significance Tests Across Event Windows (Market Model).

Window	CAAR	Cross-Sectional p	Patell p	Corrado p
[−1, +1] (primary)	0.0008	0.892	0.647	0.802
[−3, +3]	0.0034	0.605	0.393	0.881
[−5, +5]	0.0045	0.632	0.389	0.665
[−10, +10]	−0.0055	0.661	0.957	0.335

Note: N = 34 for all tests. Cross-sectional p-values are as reported in Table 3. None of the three test statistics is significant at conventional levels in any window.

Table A7. CAAR Significance Tests Across Event Windows (Market Model).

Window	CAAR	Cross-Sectional p	Patell p	Corrado p
[−1, +1] (primary)	0.0008	0.892	0.647	0.802
[−3, +3]	0.0034	0.605	0.393	0.881
[−5, +5]	0.0045	0.632	0.389	0.665
[−10, +10]	−0.0055	0.661	0.957	0.335

Note: N = 34 for all tests. Cross-sectional p-values are as reported in Table 3. None of the three test statistics is significant at conventional levels in any window.