ESG Performance, Firm Value, and Market Risk in Maritime and Fishery Firms: Panel Evidence from Korea

Abstract

Environmental, social, and governance (ESG) ratings aggregate heterogeneous corporate activities whose valuation and risk associations may vary by industry and dimension. This study examines a balanced panel of 556 non-financial KOSPI-listed firms in Korea from 2012 to 2025 (7784 firm-year observations), partitioned into 24 maritime and fishery firms (336 observations) and 532 non-focal comparison firms (7448 observations). Firm value is measured by Tobin’s Q and market risk by the annual standard deviation of daily stock returns. Separate firm and year fixed-effects regressions describe within-group associations, while pooled sector-by-ESG interaction models formally test slope differences. In the focal sample, the overall ESG, social, and governance scores are negatively associated with firm value, whereas in the joint risk model, the environmental pillar is positively associated with volatility and the social pillar is negatively associated. Formal interaction tests indicate a significant sector difference for the environmental-value association and for the environmental and social risk associations; joint tests reject the null that the E-S-G sector-interaction terms are jointly zero for both outcomes (p = 0.037 for value and p < 0.001 for risk). The sector differences are more stable for market risk than for firm value: environmental and social risk interactions remain statistically detectable under richer and more flexible pooled specifications, whereas the joint value interaction weakens when nuisance slopes are allowed to vary by sector. Leave-one-firm-out and leave-one-subsector-out diagnostics preserve the main environmentally positive and socially negative risk signs, although statistical strength varies with subsector composition. Wild Cluster Bootstrap checks further support the environmental and social risk patterns, while lagged and post-2022 analyses narrow the temporal and valuation claims. Detailed-component results are treated as hypothesis-generating and evaluated using false-discovery-rate adjustment. The results support a differentiated, sector-sensitive interpretation of ESG ratings but remain associational rather than causal.

1. Introduction

Environmental, social, and governance (ESG) considerations have become central to corporate sustainability, capital allocation, and risk assessment. This shift is particularly consequential in the maritime economy, where firms operate under simultaneous pressure from decarbonization requirements, environmental regulation, occupational safety concerns, global supply-chain expectations, and heightened investor scrutiny. Recent reviews of maritime ESG research document a rapid movement from an environment-dominant agenda toward a broader E-S-G framework, while also emphasizing fragmented disclosure practices and the lack of industry-specific evaluation frameworks [1,2]. Evidence from listed maritime firms likewise suggests that ESG integration is not uniform: firms combine sustainability and financial strategies in different ways, and the social dimension has become more strategically salient since 2020 [3].

The financial implications of ESG performance remain contested. One strand of research argues that stronger ESG performance can improve resource efficiency, reputation, stakeholder relations, information quality, and access to capital, thereby supporting market value and resilience [4,5,6]. A second strand emphasizes costs, adjustment burdens, managerial discretion, and measurement heterogeneity. From this perspective, ESG expenditure may not immediately translate into a valuation premium, and the relationship can depend on the dimension being measured, the outcome variable, the industry, and the time horizon. Recent international evidence on firm value and risk confirms that ESG controversies and ESG performance can have different direct and indirect associations with market valuation and volatility [6].

The same ambiguity is evident in corporate risk. ESG disclosure can reduce information asymmetry and constrain excessive risk-taking, and several studies report lower market-based risk in firms with stronger ESG practices [7,8]. However, disaggregated evidence shows that environmental, social, and governance dimensions may be associated with risk in different directions [9]. Consequently, an aggregate ESG score may be informative for broad benchmarking but insufficient for identifying which sustainability activities are linked to valuation or risk outcomes.

This issue is especially relevant for maritime and fishery firms. The sector is capital-intensive, regulation-sensitive, globally exposed, and strongly connected to natural resources, labor conditions, safety, local communities, and supply chains. Korean evidence already points to sector-specific patterns. Shipping firms have been found to differ from the broader KOSPI market in both ESG levels and valuation relationships [10,11]. Research on Korean logistics firms links ESG management to firm value and default-related outcomes [12], while recent shipping studies show that environmental and social investments can have different associations with liquidity, economic value added, and financing costs [13,14]. These findings caution against applying a single cross-industry ESG interpretation to maritime and fishery firms.

Three issues motivate the present analysis. First, much of the sector-specific Korean literature focuses on shipping or logistics rather than a broader maritime and fisheries portfolio that also includes fisheries and shipbuilding-related firms. Second, value and risk are often examined separately, even though a given ESG dimension may have different associations with valuation and uncertainty. Third, aggregate ESG ratings can obscure offsetting patterns among E, S, G, and their subcomponents. Recent maritime ESG reviews similarly call for more industry-sensitive and multidimensional empirical evidence [1,2].

This study examines these issues using a balanced panel of KOSPI-listed non-financial firms from 2012 to 2025. The full balanced panel contains 556 firms and 7784 firm-year observations. For descriptive and separate fixed-effects analyses, it is partitioned into a non-focal comparison sample of 532 firms and 7448 observations and a focal maritime and fisheries sample of 24 firms and 336 observations, comprising shipping and logistics, fisheries, and shipbuilding-related companies. Pooled sector-interaction models use the full 556-firm panel to test slope differences formally. The analysis considers firm value, measured by Tobin’s Q, and market risk, measured by annual stock-return volatility. Overall ESG performance, the E, S, and G dimensions, and eight detailed ESG components are examined. Firm and year fixed effects form the main empirical specification, with firm-clustered standard errors and Wild Cluster Bootstrap inference for the smaller focal sample.

The paper addresses two research questions. RQ1 asks whether ESG–value and ESG–risk associations differ between the focal maritime and fishery firms and the remaining non-financial KOSPI firms. The analysis addresses this question directly using pooled firm and year fixed-effects models with maritime sector–ESG interaction terms and formal joint tests of the E, S, and G interactions. RQ2 asks which ESG dimensions and detailed components are associated with firm value and market risk within the focal sector. The questions remain non-directional because the literature, both general and maritime-specific, reports positive, negative, and insignificant relationships.

The study contributes in four ways. Relative to prior Korean shipping and logistics studies that primarily document ESG–finance relationships within shipping-related samples [10,11,12,13,14], the incremental contribution here is not a new sign estimate for ESG per se but a formal test of whether ESG slopes differ across sectors and outcomes. First, the study uses mutually exclusive focal and non-focal samples for descriptive regressions and pooled sector-interaction models to test E-S-G slope differences directly. Second, it examines firm value and market-based risk within the same design; the most stable sector-specific evidence is concentrated in the environmental and social associations with market risk rather than in a generic ESG–value relationship. Third, it extends the empirical setting beyond shipping and logistics to a broader sector-focused portfolio that includes fisheries and shipbuilding-related firms, while explicitly evaluating whether the results depend on subsector composition. Fourth, it stress-tests the focal-sample evidence using Wild Cluster Bootstrap p-values, leave-one-firm-out and leave-one-subsector-out diagnostics, lagged ESG specifications, richer time-varying controls, sensitivity to the 2022 KCGS rating revision, flexible sector-specific nuisance specifications, common-threshold Winsorization checks, and false-discovery-rate adjustment for detailed-component tests. The resulting contribution is therefore a bounded, sector-materiality assessment of where ESG–value and ESG–risk relationships differ and where they do not, while retaining an explicitly associational interpretation.

2. Literature Review and Analytical Framework

2.1. ESG Performance and Firm Valuation

Firm value reflects investors’ assessments of expected future cash flows, growth opportunities, intangible assets, and risk. ESG performance can influence this assessment through several channels. Environmental efficiency may reduce waste, regulatory exposure, and resource costs; social performance may strengthen employee, customer, supplier, and community relationships; and governance quality may improve monitoring and reduce agency problems. Empirical evidence has therefore often documented a positive relationship between selected ESG attributes and market valuation. Derwall et al. [4], for example, report an eco-efficiency premium, while Black et al. [5] show that stronger corporate governance is associated with higher Tobin’s Q in Korean public firms. Korean studies likewise report value relevance and dimension-specific relationships across ESG, social-responsibility, and governance measures [15,16,17,18].

Positive valuation is not guaranteed, however. ESG programs require expenditures, organizational change, monitoring systems, new capital equipment, and compliance resources. Markets may discount these costs when benefits are uncertain, deferred, or difficult to verify. In addition, aggregate ESG ratings combine heterogeneous activities that can carry different cost and benefit profiles. Korean evidence shows that the link between ESG and financial performance can vary by dimension and performance measure [17]. This heterogeneity provides a reason to distinguish the overall ESG score from E, S, and G rather than interpreting ESG as a single treatment.

The sectoral setting intensifies this ambiguity. In shipping, Kim and Hong [11] find that ESG activities are positively associated with firm value in the broader KOSPI sample but negatively associated with value among shipping firms. Kim and Lim [13] similarly report that environmental and social investments can reduce the economic value added for shipping firms relative to non-shipping firms. These results are consistent with the possibility that investors treat sustainability commitments differently in industries with large transition costs, volatile freight or commodity cycles, and high capital requirements. The key empirical question is therefore not whether ESG is universally “good” or “bad” for value, but whether valuation differs by ESG dimension and industry.

2.2. ESG Performance and Corporate Risk

Corporate risk is another important dimension of sustainable performance. ESG activities can be associated with lower risk when they improve disclosure quality, stakeholder trust, operational discipline, governance, or preparedness for environmental and social shocks. Menla Ali et al. [7] show that stronger ESG disclosure is associated with lower corporate risk-taking based on both accounting- and market-based measures. Du and Nik Azman [8] also report that stronger ESG performance is associated with lower risk-taking among listed Chinese firms. International panel evidence further suggests that high ESG scores can soften the risk consequences of ESG controversies [6], while socially responsible investment research also underscores the importance of considering return and risk jointly [19].

Yet disaggregation again matters. Sun et al. [9] find that ESG components have different relationships with corporate risk-taking, with environmental performance inhibiting risk-taking while social and governance performance can move in the opposite direction in their setting. Such results do not necessarily contradict the risk-management view of ESG; rather, they show that different ESG activities can be associated with different investment, operating, and financing choices. Market-based volatility may also capture transition uncertainty: firms with high exposure to environmental constraints may intensify ESG activity precisely because they face higher underlying transition risks.

For maritime and fishery firms, risk can arise from freight-market volatility, commodity prices, environmental incidents, vessel and plant safety, labor conditions, regulation, supply disruptions, and capital-market financing. Shipping-specific evidence indicates that ESG activity can reduce the cost of debt [14], but the literature has not established that every ESG pillar consistently reduces stock-market volatility. This is important because a sustainability initiative may be costly from a near-term valuation perspective, yet still improve stakeholder stability or risk control, while another initiative may coincide with heightened transition exposure.

2.3. Sector Context and Research Questions

Maritime ESG research has historically concentrated on environmental issues, especially emissions, energy efficiency, and decarbonization. Recent reviews find growing attention to labor, welfare, board structure, disclosure, and stakeholder governance, but they also identify fragmentation between sustainability and corporate-governance research [1,2]. This fragmentation is consequential because shipping, fisheries, logistics, and shipbuilding firms face overlapping environmental and social obligations that cannot be represented well by a single composite rating.

The Korean setting provides an informative test bed. Listed maritime and fishery firms are exposed to global trade and regulatory conditions but are evaluated within the same domestic capital market and ESG-rating framework as other listed companies. This allows the focal sector sample to be compared with a mutually exclusive non-focal group while holding the national institutional setting constant. The focal sample is also heterogeneous enough to capture multiple parts of the blue economy: shipping and logistics firms are exposed to transport and trade cycles; fishery firms are connected to biological resources, food safety, and processing; and shipbuilding-related firms face technology, energy-transition, and capital-investment pressures.

The empirical design therefore does not impose a single expected sign on ESG coefficients or assume that shipping and logistics, fisheries, and shipbuilding-related firms are economically identical. The combined focal group is treated as a sector-focused portfolio whose members share substantial exposure to capital intensity, environmental transition and regulation, labor and safety conditions, natural-resource or supply-chain constraints, and global market conditions, while differing in the materiality of those exposures. This motivates both a pooled sector test and explicit sensitivity checks for subsector composition. RQ1 formally tests whether the ESG–value and ESG–risk slopes differ between focal and non-focal firms through sector-interaction terms. RQ2 examines E, S, G, and detailed components within the focal group. Contrasting signs across outcomes are interpreted as empirical associations rather than evidence of a causal trade-off.

The analytical framework combines sector materiality with stakeholder, agency, signaling, and legitimacy perspectives. From a stakeholder perspective, social performance can be economically relevant in maritime and fisheries activities because labor safety, supplier continuity, food safety, and community relations directly affect operations and stakeholder support [20]. Agency theory implies that governance arrangements can influence monitoring, managerial discretion, financing, and valuation, although the direction of observed rating-value associations need not be uniformly positive [21]. Signaling and legitimacy perspectives suggest that ESG disclosure and visible sustainability activity can communicate quality or responsiveness to external expectations, but they may also intensify during periods of greater underlying exposure, so observed associations can reflect both corporate responses and investor interpretation [22,23]. Finally, the materiality perspective predicts that financially relevant ESG issues differ across industries; environmental transition exposure is therefore expected to be especially salient in capital-intensive and regulation-sensitive maritime activities, while the importance of social and governance dimensions can vary across shipping, fisheries, and shipbuilding-related firms [24]. These complementary perspectives motivate disaggregated ESG tests and formal sector interactions rather than a uniform expected sign.

3. Materials and Methods

3.1. Data and Sample Construction

The empirical analysis uses firms listed on the Korea Exchange KOSPI market over 2012–2025. ESG information was obtained from the Korea Institute of Corporate Governance and Sustainability (KCGS), and financial-statement and stock-market variables were obtained from FnGuide DataGuide. Financial firms and firm-year observations missing required ESG or financial variables were excluded. Firms were then required to have continuous observations over the 14-year study period. The resulting full balanced panel comprises 556 non-financial firms and 7784 firm-year observations. For the separate group analyses, the panel is partitioned into 532 non-focal comparison firms (7448 observations) and 24 focal maritime and fishery firms (336 observations); the two groups are mutually exclusive.

The focal sample is identified using the sector flag and subsector labels contained in the final analysis dataset. Operationally, a firm is classified as focal only when the pre-existing sector flag assigns it to one of three pre-defined subsectors—shipping and logistics, fisheries, or shipbuilding-related activities; all other listed firms form the non-focal comparison group. This classification is fixed before estimation, and no firm is added, removed, or reclassified on the basis of the regression results. The focal sample contains 24 firms and 336 firm-year observations: 11 shipping and logistics firms, 8 fishery firms, and 5 shipbuilding-related firms. The non-focal comparison sample contains the remaining 532 firms and 7448 firm-year observations. Each firm in both groups is observed in every year from 2012 to 2025. Appendix A reports the firm name, security code, KSIC section observed in the dataset, and focal subsector for all 24 focal firms. The three focal subsectors are not assumed to be homogeneous: shipping and logistics firms are primarily exposed to transport, trade, and freight conditions; fishery firms to biological resources, food safety, processing, and commodity conditions; and shipbuilding-related firms to capital investment, technology, and energy-transition demand. The combined focal sample is therefore interpreted as a sector-focused portfolio, and leave-one-subsector-out analyses assess whether any single group dominates the results. The available analytical file retains the operational sector flag and subsector labels but does not preserve a finer provenance record for how those labels were originally constructed; accordingly, the paper reports the complete roster and observed industry coding for replication and does not present the flag as an official industrial taxonomy. The balanced design limits changes in sample composition over time but can create survivor or coverage selection because firms with incomplete histories are excluded. A genuine pre-balance unbalanced-panel survivor-bias re-estimation cannot be reconstructed from the available analysis files because they contain complete 2012–2025 histories rather than firm-specific coverage gaps. We therefore treat survivor selection as an explicit limitation rather than presenting an auxiliary balanced sample as a substitute for an unbalanced-panel test.

Table 1. Sample composition.

Sample/Sector	Firms	Years	Firm-Year Obs.	Share Within Focal Sample
Full non-financial listed-firm panel	556	2012–2025	7784	—
Non-focal comparison sample	532	2012–2025	7448	—
Maritime and fisheries sample	24	2012–2025	336	100.0%
Shipping and logistics	11	2012–2025	154	45.8%
Fisheries	8	2012–2025	112	33.3%
Shipbuilding-related	5	2012–2025	70	20.8%

Notes: The full balanced panel is partitioned into mutually exclusive non-focal and focal samples. Sector counts follow the final analysis dataset; subsector shares are calculated within the focal sample.

summarizes the sample composition.

3.2. Variables

The first dependent variable is firm value (Value), measured by Tobin’s Q. The Tobin’s Q variable used in the analytical dataset follows the DataGuide-based definition documented in Korean capital-market research: (market capitalization + total liabilities)/total assets [25]. Debt therefore enters through total liabilities, and no separate preferred-share term is included in the source-variable definition used in the analytical data. The second dependent variable is market risk (Risk), measured as the standard deviation of each firm’s daily stock returns within a calendar year; a larger value denotes greater annual stock-return volatility. Because Tobin’s Q and stock-return volatility capture market-based outcomes, supplementary analyses also examine ROA as an accounting-performance outcome and financial leverage (total liabilities/total assets) as a balance-sheet risk-related outcome.

The main explanatory variable is the continuous overall ESG field, named esg in the analytical workbook. This field is retained from the KCGS-sourced ESG data used to construct the analysis file and is used directly in estimation; the authors did not create it by numerically converting public KCGS letter grades or by reconstructing an overall score from the E, S, and G pillars. Public KCGS ratings are reported on the S, A+, A, B+, B, C, and D scale [26,27], whereas the ESG field observed in the estimation sample ranges from 0.00 to 6.00 in 0.05-point increments, with larger values denoting stronger ESG performance under the retained source coding. The stated 0.00–6.00 range is the observed range in the non-financial estimation sample used throughout this study. The analytical workbook does not retain the corresponding public letter-grade field, the year-specific grade thresholds, or additional metadata needed to reconstruct a defensible grade-to-score crosswalk. No ex post conversion is therefore imposed. Accordingly, the sample mean of 2.5918 is interpreted only as the mean of the continuous ESG field, not as the numerical equivalent of a particular public KCGS letter grade. The environmental (E), social (S), and governance (G) pillar indicators and their detailed components are expressed as proportions. Environmental components are environmental management (EB), environmental performance (EP), and environmental stakeholder communication (ES); social components are internal (S_INT) and external (S_EXT) social performance; governance components are shareholder rights (GS), board (GB), and audit (GA). The continuous overall score and the proportional pillar scores are therefore interpreted on their own scales. To assess comparability over time, Supplementary Table S5 reports the annual score distributions and year-to-year increases, decreases, and unchanged observations. The sharp 2021–2022 shift is treated explicitly because KCGS revised its assessment model in 2022 [26]. Linear treatment is evaluated against both a quadratic specification and a within-year quintile specification.

Control variables capture size, leverage, operating cash flow, R&D intensity, sales growth, ownership concentration, foreign ownership, and listing age. For the sample-specific regressions, continuous financial controls are Winsorized at the 1st and 99th percentiles within the corresponding estimation sample. For the pooled sector-interaction models, Winsorization thresholds are calculated from the pooled 556-firm sample. ESG indicators are not Winsorized. Sales growth is expressed as a decimal rate. The controls follow established research on valuation, leverage, growth, innovation, ownership, foreign investment, and firm age [28,29,30,31,32,33,34,35,36].

Table 2. Variable definitions.

Variable	Definition	Role
Value	Source-variable definition: (Market capitalization + total liabilities)/total assets; market-based firm-value proxy	Dependent variable
Risk	Annual standard deviation of daily stock returns	Dependent variable
ESG	Continuous ESG field retained from KCGS-sourced ESG data in the analytical workbook (observed range: 0.00–6.00; higher values denote stronger ESG performance); no author-created letter-grade conversion	Main explanatory variable
E	Environmental pillar score	ESG dimension
S	Social pillar score	ESG dimension
G	Governance pillar score	ESG dimension
EB/EP/ES	Environmental management/performance/stakeholder communication	Detailed E components
S_INT/S_EXT	Internal/external social performance	Detailed S components
GS/GB/GA	Shareholder rights/board/audit	Detailed G components
Size	Natural log of total assets	Control
Lev	Total liabilities/total assets	Control
Cfo	Operating cash flow/total assets	Control
R&D	R&D expenditure/total assets	Control
Sg	Annual sales growth rate	Control
Own	Largest shareholder and related-party ownership share	Control
For	Foreign investor ownership share	Control
Age	Natural log of listing age + 1	Control

Notes: KCGS: Korea Institute of Corporate Governance and Sustainability. E, S, G and detailed-component scores are proportional measures; the overall ESG indicator remains on its source scale. For sample-specific regressions, continuous financial controls are Winsorized at the 1st and 99th percentiles within the corresponding estimation sample; pooled interaction models use pooled-sample thresholds.

summarizes the variable definitions and roles.

3.3. Empirical Specification

The empirical strategy prioritizes within-firm variation. Pooled OLS models were examined as a diagnostic benchmark, whereas the principal results use firm fixed effects and year fixed effects. Firm fixed effects absorb time-invariant firm characteristics that may correlate with ESG, and year effects absorb common annual shocks. Variance inflation factors were examined and did not indicate a material multicollinearity problem. A Hausman specification test favored the fixed-effects approach over a random-effects alternative [37].

The baseline fixed-effects specification for the overall ESG indicator is:

$$Y_{it}=\beta {ESG}_{it}+{\gamma }^{\prime X_{it}}+{\alpha }_i+{\lambda }_t+{\epsilon }_{it}$$

(1)

where $Y_{it}$ denotes either Tobin’s Q or annual stock-return volatility, $X_{it}$ is the vector of firm-level controls, ${\alpha }_i$ represents firm fixed effects, ${\lambda }_t$ represents year fixed effects, and ${\epsilon }_{it}$ is the idiosyncratic error term. The ESG pillars are also estimated separately and jointly. The joint specification is:

$$Y_{it}={\beta }_E{E}_{it}+{\beta }_S{S}_{it}+{\beta }_G{G}_{it}+{\gamma }^{\prime X_{it}}+{\alpha }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

An analogous specification replaces E, S, and G with the eight detailed ESG components. Standard errors are clustered at the firm level to account for within-firm serial correlation and heteroskedasticity [38]. Because the focal sample contains only 24 firm clusters, bootstrap p-values are additionally obtained from a firm-level Wild Cluster Bootstrap-t procedure using 9999 Rademacher draws. The detailed-component regressions evaluate multiple related coefficients and are treated as hypothesis-generating rather than confirmatory. Benjamini–Hochberg false-discovery-rate (FDR) q-values are therefore reported for the eight detailed components, and isolated 10% findings are not treated as strong evidence. All statistical analyses were conducted in Stata 18.0 (StataCorp LLC, College Station, TX, USA).

To test RQ1 formally, the pooled fixed-effects model is augmented with sector interactions: Y_it = β_E E_it + β_S S_it + β_G G_it + δ_E(Marine_i × E_it) + δ_S(Marine_i × S_it) + δ_G(Marine_i × G_it) + γ′X_it + α_i + λ_t + ε_it. The time-invariant Marine indicator is absorbed by firm fixed effects. The interaction coefficients directly test whether the within-firm ESG slopes differ between the focal firms and the remaining KOSPI firms; a joint Wald test evaluates whether all three E, S, and G sector-interaction coefficients are zero.

Sensitivity analyses include one-year-lagged ESG specifications; leave-one-firm-out and leave-one-subsector-out diagnostics; post-2022 KCGS interaction and split-period models; exclusion of zero-coded ESG observations; additional available controls for profitability (ROA) and financial distress (an indicator for negative ROA); nonlinear quadratic and within-year quintile treatments of the overall ESG score; broad-industry-by-year fixed effects; a more flexible pooled interaction model that allows the control-variable slopes and year effects to differ by focal status; subgroup regressions using common pooled-sample Winsorization thresholds; and alternative outcomes using ROA and financial leverage. The supplied analytical file does not contain current-ratio or investment variables, so those covariates are not included. Year fixed effects absorb economy-wide annual shocks, including common oil-price, exchange-rate, and macroeconomic movements, while the industry-by-year and fully interacted nuisance specifications relax progressively more of the common-shock and common-slope structures. Firm-specific freight exposure, firm-specific stock returns, and other unavailable time-varying shocks cannot be fully controlled. These sensitivity checks assess robustness and temporal ordering; they do not establish causality.

The separate non-focal and focal regressions remain useful for describing within-group patterns, but sector differences are evaluated using the formal interaction models above rather than comparing coefficients or statistical significance across the two separate samples. All estimates remain associational: fixed effects do not remove time-varying omitted variables, reverse causality, or selection into ESG activity. Accordingly, the Results and Discussion Sections (Section 4 and Section 5) use terms such as ‘associated with’ and ‘related to’ and distinguish contemporaneous evidence from the one-year-lagged sensitivity results.

3.4. Descriptive Statistics

Table 3. Descriptive statistics: non-focal comparison and maritime and fisheries samples.

Variable	Non-Focal Mean	Non-Focal SD	Maritime Mean	Maritime SD
Value	1.0998	0.8862	0.9641	0.3529
Risk	0.0256	0.0110	0.0258	0.0097
ESG	2.5783	1.3399	2.8915	1.2982
E	0.3147	0.2397	0.3712	0.2438
S	0.2908	0.2193	0.3534	0.2061
G	0.2770	0.1250	0.2951	0.1270
Size	20.3761	1.6388	20.9629	1.6672
Lev	0.4515	0.2025	0.5594	0.1862
Cfo	0.0471	0.0635	0.0422	0.0655
R&D	0.0075	0.0148	0.0007	0.0016
Sg	0.0495	0.2013	0.0594	0.2127
Own	0.4381	0.1603	0.4180	0.1687
For	0.1008	0.1270	0.0882	0.0948
Age	3.2227	0.5530	3.2536	0.5781

Notes: N = 7448 for the non-focal comparison sample and N = 336 for the focal sample. Continuous controls reflect 1%/99% Winsorization within each descriptive sample. Sales growth is shown as a decimal rate.

compares descriptive statistics for the non-focal comparison and focal samples. The maritime and fishery firms have a lower average Tobin’s Q (0.9641 versus 1.0998) and a similar average stock-return volatility (0.0258 versus 0.0256). Their mean overall ESG indicator and E, S, and G pillar scores are higher than those of the non-focal comparison firms. They are also larger and more leveraged on average, while their R&D intensity is lower. These unconditional differences describe the two mutually exclusive samples; they are not interpreted as ESG effects.

4. Results

4.1. ESG Performance and Firm Value

Table 4. Firm fixed-effects estimates for firm value (Tobin’s Q).

Variable	(1) ESG	(2) E	(3) S	(4) G	(5) E+S+G
Panel A. Non-focal comparison firms
ESG	−0.0381 ** (−2.04)
E		−0.2614 *** (−3.12)			−0.2625 *** (−3.01)
S			−0.1852 (−1.43)		0.0510 (0.35)
G				−0.2851 (−1.60)	−0.1248 (−0.80)
Controls	Yes	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Within R2	0.0805	0.0814	0.0792	0.0792	0.0816
Observations	7448	7448	7448	7448	7448
Firms	532	532	532	532	532
Panel B. Maritime and fishery firms
ESG	−0.0653 ** (−2.33)
E		−0.3131 (−1.68)			−0.1286 (−0.75)
S			−0.4164 *** (−2.90)		−0.2318 ** (−2.17)
G				−0.6967 ** (−2.76)	−0.5269 ** (−2.10)
Controls	Yes	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Within R2	0.4003	0.3886	0.3996	0.4068	0.4201
Observations	336	336	336	336	336
Firms	24	24	24	24	24

Notes: Firm-clustered t-statistics in parentheses. All models include the controls in Table 2, firm fixed effects, and year fixed effects. Continuous financial controls are Winsorized within each estimation sample. ** p < 0.05, *** p < 0.01. Panels A and B are mutually exclusive; cross-panel coefficient comparisons are descriptive, and formal sector differences are tested in the pooled interaction models reported below.

reports the firm fixed-effects estimates for Tobin’s Q. In the non-focal comparison sample (Panel A), the coefficient on overall ESG is negative (beta = −0.0381, p < 0.05). The environmental pillar is also negative when entered separately (beta = −0.2614, p < 0.01) and remains negative in the joint E-S-G model (beta = −0.2625, p < 0.01). Social and governance performance are not statistically significant in either the separate or joint specifications. The non-focal comparison sample therefore does not display a uniform positive valuation association for higher ESG scores.

In the focal sample (Panel B), overall ESG is negatively associated with firm value (beta = −0.0653, p < 0.05). The environmental pillar is not statistically significant. Social performance is negative both when estimated separately (beta = −0.4164, p < 0.01) and in the joint model (beta = −0.2318, p < 0.05). Governance is also negative in the separate model (beta = −0.6967, p < 0.05) and remains negative in the joint model (beta = −0.5269, p < 0.05). These coefficients describe the focal sample; their differences from Panel A are not formal tests of coefficient equality.

For economic interpretation on a common scale, a 0.10-point increase in the social and governance pillar scores in the joint focal model corresponds to approximately −0.023 and −0.053 in Tobin’s Q, respectively. Relative to the focal-sample mean Tobin’s Q of 0.9641, these magnitudes are about 2.4% and 5.5%. These calculations describe contemporaneous associations and should not be interpreted as causal valuation effects.

4.2. ESG Performance and Market Risk

Table 5. Firm fixed-effects estimates for market risk (annual stock-return volatility).

Variable	(1) ESG	(2) E	(3) S	(4) G	(5) E+S+G
Panel A. Non-focal comparison firms
ESG	−0.0002 (−0.98)
E		−0.0010 (−1.05)			−0.0021 * (−1.85)
S			0.0005 (0.45)		0.0026 * (1.95)
G				−0.0016 (−0.96)	−0.0019 (−1.04)
Controls	Yes	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Within R2	0.1495	0.1495	0.1493	0.1495	0.1502
Observations	7448	7448	7448	7448	7448
Firms	532	532	532	532	532
Panel B. Maritime and fishery firms
ESG	0.0001 (0.08)
E		0.0063 (1.53)			0.0119 *** (2.86)
S			−0.0063 * (−1.79)		−0.0095 ** (−2.26)
G				−0.0098 (−1.18)	−0.0091 (−1.10)
Controls	Yes	Yes	Yes	Yes	Yes
Firm FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Within R2	0.2742	0.2814	0.2811	0.2817	0.3052
Observations	336	336	336	336	336
Firms	24	24	24	24	24

Notes: Firm-clustered t-statistics in parentheses. All models include the controls in Table 2, firm fixed effects, and year fixed effects. Continuous financial controls are Winsorized within each estimation sample. * p < 0.10, ** p < 0.05, *** p < 0.01. Panels A and B are mutually exclusive; cross-panel coefficient comparisons are descriptive, and formal sector differences are tested in Table 6.

reports fixed-effects estimates for annual stock-return volatility. In the non-focal comparison sample, overall ESG and the separately entered E, S, and G pillars are not statistically significant after firm and year fixed effects are included. In the joint E-S-G specification, however, E is weakly negative (beta = −0.0021, p < 0.10) and S is weakly positive (beta = 0.0026, p < 0.10), while G remains statistically insignificant. These 10% findings are marginal and are not treated as strong evidence; formal sector differences are evaluated in the pooled interaction models in Table 6.

The focal sample displays more differentiated estimates. Overall ESG is not significantly associated with market risk, and neither the separately entered environmental pillar nor the governance pillar is significant. The social coefficient is negative when entered separately at the 10% level and remains negative in the joint model (beta = −0.0095, p < 0.05). In contrast, the environmental coefficient is positive and statistically significant in the joint model (beta = 0.0119, p < 0.01). Governance remains statistically insignificant. The joint E-S-G specification therefore yields opposite environmental and social signs to the marginal joint estimates in the non-focal comparison sample, but the cross-sector contrast is interpreted through the formal interaction tests rather than by comparing separate-regression significance. These associations should not be read as causal effects.

Using the same 0.10-point change for comparability, the joint-model environmental coefficient corresponds to an increase of about 0.00119 in annual return volatility, while the social coefficient corresponds to a decrease of about 0.00095. Relative to the focal-sample mean risk measure of 0.0258, these magnitudes are approximately +4.6% and −3.7%, respectively.

4.3. Formal Tests of Sector Differences

Table 6 reports the pooled fixed-effects interaction models using all 556 firms to formally test RQ1. For firm value, the Marine × E interaction is positive and statistically significant (beta = 0.4991, 95% CI [0.107, 0.892], p = 0.013), indicating that the environmental-value slope differs between focal and non-focal firms. Marine × S and Marine × G are not individually significant. A joint Wald test rejects the null that the three E-S-G interaction terms are jointly zero at the 5% level (F = 2.85, p = 0.037). The environmental slope within the focal group itself is positive but not statistically significant in the interaction model, so the result is evidence of a sector difference rather than evidence that environmental performance raises focal-firm value.

The sector contrast is clearer for market risk. Marine × E is positive (beta = 0.01363, 95% CI [0.00595, 0.02131], p < 0.001), whereas Marine × S is negative (beta = −0.01078, 95% CI [−0.01875, −0.00280], p = 0.008); Marine × G is not statistically significant. The joint test strongly rejects the null that the E-S-G interactions are jointly zero (F = 7.06, p < 0.001). The implied focal-group slopes are positive for E (0.01163, p = 0.002) and negative for S (−0.00824, p = 0.034). Thus, sector distinctiveness is dimension-specific and is most clearly supported for the environmental and social associations with market risk.

Table 6. Formal sector-interaction tests.

Measure	Value: Interaction Coef (p)	Value: Focal Slope Coef (p)	Risk: Interaction Coef (p)	Risk: Focal Slope Coef (p)
ESG	−0.0416 (p = 0.263)	−0.0796 (p = 0.026)	0.00057 (p = 0.422)	0.00039 (p = 0.581)
E	0.4991 (p = 0.013)	0.2335 (p = 0.198)	0.01363 (p < 0.001)	0.01163 (p = 0.002)
S	−0.2469 (p = 0.153)	−0.2014 (p = 0.098)	−0.01078 (p = 0.008)	−0.00824 (p = 0.034)
G	−0.4577 (p = 0.139)	−0.6043 (p = 0.045)	−0.00432 (p = 0.416)	−0.00635 (p = 0.223)

Notes: Coefficients are from pooled firm and year fixed-effects models with the controls in Table 2, pooled-sample Winsorization thresholds, and firm-clustered standard errors. The sector-interaction coefficient denotes the difference between the focal and non-focal slopes. The focal slope equals the non-focal slope plus the interaction term. Joint Wald tests of H0: Marine × E = Marine × S = Marine × G = 0 yield p = 0.0368 for firm value and p = 0.00012 for market risk.

Table 6. Formal sector-interaction tests.

Measure	Value: Interaction Coef (p)	Value: Focal Slope Coef (p)	Risk: Interaction Coef (p)	Risk: Focal Slope Coef (p)
ESG	−0.0416 (p = 0.263)	−0.0796 (p = 0.026)	0.00057 (p = 0.422)	0.00039 (p = 0.581)
E	0.4991 (p = 0.013)	0.2335 (p = 0.198)	0.01363 (p < 0.001)	0.01163 (p = 0.002)
S	−0.2469 (p = 0.153)	−0.2014 (p = 0.098)	−0.01078 (p = 0.008)	−0.00824 (p = 0.034)
G	−0.4577 (p = 0.139)	−0.6043 (p = 0.045)	−0.00432 (p = 0.416)	−0.00635 (p = 0.223)

Notes: Coefficients are from pooled firm and year fixed-effects models with the controls in Table 2, pooled-sample Winsorization thresholds, and firm-clustered standard errors. The sector-interaction coefficient denotes the difference between the focal and non-focal slopes. The focal slope equals the non-focal slope plus the interaction term. Joint Wald tests of H0: Marine × E = Marine × S = Marine × G = 0 yield p = 0.0368 for firm value and p = 0.00012 for market risk.

4.4. Detailed ESG Components in Maritime and Fishery Firms

Table 7. Detailed ESG components: maritime and fisheries fixed-effects estimates.

Component	Value: Pillar Model	Value: Joint Model	Risk: Pillar Model	Risk: Joint Model
EB: Environmental management	−0.3521 ** (−2.64)	−0.2554 ** (−2.17)	0.0006 (0.18)	0.0037 (1.25)
EP: Environmental performance	0.3872 *** (2.84)	0.5542 *** (3.30)	−0.0001 (−0.02)	0.0045 (1.00)
ES: Stakeholder communication	−0.1627 (−1.41)	−0.1198 (−1.07)	0.0060 * (2.02)	0.0081 *** (3.13)
S_INT: Internal social	−0.2913 * (−1.84)	−0.2947 (−1.68)	−0.0002 (−0.04)	−0.0065 (−1.14)
S_EXT: External social	−0.1897 (−1.65)	−0.0900 (−0.75)	−0.0063 (−1.26)	−0.0080* (−1.95)
GS: Shareholder rights	−0.3432 (−1.71)	−0.3674 * (−1.74)	−0.0082 (−1.26)	−0.0066 (−1.38)
GB: Board	−0.1993 (−0.84)	−0.0408 (−0.19)	−0.0040 (−0.39)	−0.0086 (−0.85)
GA: Audit	−0.3350 * (−1.72)	−0.3221 (−1.67)	−0.0045 (−1.12)	−0.0045 (−1.20)

Notes: N = 336; 24 firms. ‘Pillar model’ estimates components within their E, S, or G block; ‘joint model’ includes all eight detailed ESG components. All models include controls, firm fixed effects, and year fixed effects. * p < 0.10, ** p < 0.05, *** p < 0.01 report conventional firm-clustered inference. Benjamini–Hochberg FDR correction is additionally applied across the eight joint-component tests for each outcome; EP for value (q = 0.015) and ES for risk (q = 0.020) remain significant at q < 0.05.

reports detailed-component estimates for the focal sample. These regressions are explicitly treated as hypothesis-generating because they consider eight related coefficients. For firm value, environmental management (EB) is negative and environmental performance (EP) is positive in the joint model. After Benjamini–Hochberg correction across the eight joint-component tests, only EP remains significant at a 5% false-discovery rate (q = 0.015); EB does not (q = 0.116). This narrows the component-level evidence relative to conventional unadjusted p-values.

For market risk, environmental stakeholder communication (ES) is positively associated with stock-return volatility in the joint model. This result remains significant after false-discovery-rate adjustment (q = 0.020). The S_EXT (external social performance) coefficient is negative and conventionally significant in the joint model but does not survive the FDR correction (q = 0.186). The remaining component coefficients also have q-values above 0.05. Component-level results other than EP for value and ES for risk are therefore interpreted cautiously.

4.5. Small-Cluster and Influence Sensitivity Checks

Because conventional clustered inference can be sensitive when the number of clusters is small, the focal-sample coefficients were re-evaluated using a firm-level Wild Cluster Bootstrap-t with 9999 Rademacher draws.

Table 8. Wild Cluster Bootstrap p-values: maritime and fishery firms.

ESG Variable	Firm Value Wild p-Value	Market Risk Wild p-Value
Overall ESG	0.049	0.931
E (joint E-S-G model)	0.521	0.017
S (joint E-S-G model)	0.060	0.039
G (joint E-S-G model)	0.093	0.311

Notes: Entries are two-sided Wild Cluster Bootstrap-t p-values for the focal-sample specifications corresponding to Table 4 and Table 5. p-values are computed using restricted residuals and 9999 firm-level Rademacher draws. The procedure is a small-cluster sensitivity check and does not substitute for a larger cross-sectional sample.

reports bootstrap p-values for the corresponding focal-sample coefficients in Table 4 and Table 5. For firm value, the overall ESG coefficient remains statistically detectable (wild p = 0.049), while the joint social and governance coefficients weaken to p = 0.060 and p = 0.093, respectively. For market risk, the positive environmental coefficient (wild p = 0.017) and negative social coefficient (wild p = 0.039) remain statistically detectable; governance remains insignificant (wild p = 0.311). These results place the strongest small-cluster support on the environmental and social risk associations.

Influence diagnostics further assess the small focal sample. In 24 leave-one-firm-out repetitions, the overall ESG–value coefficient remains negative in all 24 runs and is significant at 5% in 22. The environmental risk coefficient is positive and significant at 5% in all 24 runs, while the social risk coefficient is negative in all 24 runs, significant at 10% in all 24 and at 5% in 19. Leave-one-subsector-out diagnostics provide a stricter composition check. In the joint risk model, the environmental coefficient remains positive and the social coefficient remains negative after excluding each of the three subsectors in turn, although statistical significance varies: excluding shipping and logistics gives E = 0.0109 (p = 0.116) and S = −0.0070 (p = 0.130); excluding fisheries gives E = 0.0142 (p = 0.020) and S = −0.0116 (p = 0.027); excluding shipbuilding-related firms gives E = 0.0105 (p = 0.071) and S = −0.0075 (p = 0.297). The signs are therefore compositionally stable, but the statistical strength is not, which supports treating the 24-firm focal sample as heterogeneous rather than homogeneous. Full leave-one-subsector-out results are reported in Appendix B.

4.6. Additional Robustness and Sensitivity Analyses

One-year-lagged specifications provide a stricter temporal ordering. In the focal sample, lagged overall ESG remains negatively associated with firm value (beta = −0.0590, p = 0.034). By contrast, the contemporaneous environmentally positive and socially negative risk coefficients do not remain statistically significant when the ESG pillars are lagged for one year; lagged governance is negative for risk (beta = −0.0178, p = 0.019). The lagged sector-interaction terms are also not statistically significant. These results indicate that the contemporaneous risk pattern is not fully stable to temporal reordering and should not be interpreted causally.

An available-data control-augmentation check adds ROA and a negative-ROA distress indicator. In the focal sample, the main qualitative pattern is preserved: overall ESG remains negatively associated with value, social performance remains negative in the joint value model, governance remains only marginally negative, and the joint risk model retains a positive environmental coefficient and a negative social coefficient. In a pooled sector-interaction model augmented with the same ROA and distress controls, the environmentally positive and socially negative risk interactions remain statistically significant, while the valuation evidence remains comparatively less stable. Because the current-ratio and investment series are not present in the analytical file, the study does not claim to control directly for those channels. In a further pooled interaction specification with broad-industry-by-year fixed effects, the environmental and social risk interactions remain significant (p = 0.004 and p = 0.009; joint p < 0.001), whereas the corresponding value interactions weaken (joint p = 0.138). Thus, the strongest sector-specific evidence is again concentrated in the market-risk outcome.

The baseline pooled interaction model constrains the non-ESG control slopes and year effects to be common across focal and non-focal firms. To relax that assumption, an additional specification interacts focal status with all eight financial controls and with the year indicators. In this more flexible model, the three value interactions are no longer jointly significant (F = 1.40, p = 0.241), whereas the market-risk interactions remain jointly significant (F = 5.19, p = 0.0015). For risk, Marine × E remains positive (beta = 0.01384, p < 0.001) and Marine × S remains negative (beta = −0.01193, p = 0.005). This additional test reinforces the interpretation that the sector-specific market-risk pattern is more stable than the value pattern.

A separate preprocessing sensitivity check applies the pooled 556-firm 1%/99% Winsorization thresholds to both subgroup regressions instead of recalculating thresholds within each subgroup. The non-focal coefficients are effectively unchanged. In the focal joint risk model, the environmental coefficient remains positive (beta = 0.01179, p = 0.009) and the social coefficient remains negative (beta = −0.00930, p = 0.039). By contrast, the focal social coefficient in the joint value model moves from p = 0.041 under sample-specific thresholds to p = 0.052 under common thresholds. The core risk findings are therefore not an artifact of subgroup-specific Winsorization, whereas some valuation inference remains sensitive at conventional significance cutoffs.

The 2022 KCGS rating-model revision matters particularly for valuation. In the full sample, the continuous overall ESG score mean increases from 1.916 in 2012 to 3.178 in 2021 and then falls to 1.989 in 2022; 474 of 556 firms show a year-to-year score decrease in 2022. This distributional break is consistent with the documented KCGS model revision [26]. In the focal sample, the overall ESG × Post2022 differential is positive but only marginally significant (beta = 0.0591, t = 1.90), while the E × Post2022 differential is positive and significant (beta = 1.0019, t = 3.09), indicating that some valuation slopes change around the revision. Post-2022 focal estimates are based on only 96 observations and are therefore imprecise. Supplementary Table S5 reports the annual score distribution and transition counts so that the scale break is visible rather than treated as a purely technical sensitivity check.

Split-period focal regressions provide a complementary sensitivity check around the 2022 rating-model revision (Supplementary Table S6). The overall ESG–value coefficient remains negative but is statistically insignificant in both 2012–2021 (beta = −0.0342, p = 0.227) and 2022–2025 (beta = −0.0606, p = 0.166). In the joint pillar value model, the social coefficient changes from 0.0257 (p = 0.842) before 2022 to −0.6892 (p = 0.033) after 2022, while the governance coefficient is marginally negative before 2022 (−0.4716, p = 0.057) but insignificant afterward. For market risk, the social coefficient is negative before 2022 (−0.01145, p = 0.040); in 2022–2025, the environmental coefficient is positive at the 10% level (0.02490, p = 0.071) and the social coefficient remains negative but imprecise (−0.01663, p = 0.115). Because the post-2022 focal period contains only 96 observations, these split-period estimates are treated as descriptive sensitivity evidence rather than separate regime-specific conclusions.

Nonlinear and categorical checks further limit overinterpretation of the overall score. A quadratic specification indicates possible nonlinearity for focal-sample risk. When the continuous overall score is replaced by within-year quintiles, the four quintile indicators are jointly significant for focal-sample risk (p = 0.034) but not for focal-sample value (p = 0.401), reinforcing the conclusion that a single linear coefficient does not summarize every aspect of the score distribution. Alternative-outcome analyses also show that the focal ESG results do not reproduce at conventional levels for ROA, while financial leverage shows no significant association with overall ESG in the focal sample (p = 0.333); the environmental pillar is only marginally negative for leverage (p = 0.073). These checks narrow, rather than broaden, the claims made from Tobin’s Q and stock-return volatility.

Taken together, the robustness checks place the strongest evidentiary weight on the environmentally positive and socially negative market-risk pattern. Those signs are not driven by a single firm, they persist across leave-one-subsector-out samples, and they survive the available ROA/distress control augmentation, common-threshold Winsorization, industry-by-year fixed effects, and a fully interacted nuisance specification. Their statistical strength nevertheless varies with subsector composition and weakens when ESG is lagged, so the evidence remains contemporaneous and associational. The valuation results are less stable across timing, rating-regime, nuisance-slope, and preprocessing choices. A direct unbalanced-panel survivor-bias check cannot be implemented because the available analytical data contain only firms with complete 2012–2025 histories; this limitation is retained explicitly.

5. Discussion

5.1. Contrasting Value and Risk Patterns

The estimates show that ESG dimensions do not have a uniform relationship with value and market risk, and the formal interaction tests refine the sector interpretation. In the non-focal comparison sample, overall ESG and the environmental pillar are negatively associated with firm value, while social and governance coefficients are not significant in the joint value model. For market risk, the separate non-focal pillar models are insignificant; the joint model produces only marginal 10% evidence of a negative environmental and positive social association. Within the focal group, overall ESG, social performance, and governance performance are negatively associated with value, whereas the joint risk model yields a positive environmental coefficient and a negative social coefficient. The pooled interaction models show selected slope differences across the two groups, but the robustness hierarchy is asymmetric: the joint value interaction weakens when control slopes and year effects are allowed to differ by sector, whereas the environmental and social risk interactions remain statistically detectable. The evidence therefore supports a stronger sector-specific interpretation for market risk than for firm value and does not imply that every ESG dimension is sector-specific.

The negative value coefficient on overall ESG in the focal sample is consistent with earlier Korean shipping evidence reporting a negative ESG–value relationship for shipping firms [11]. The one-year-lagged overall ESG coefficient remains negative and statistically significant, which reduces concern that the baseline result is purely contemporaneous. At the same time, the 2022 KCGS interaction results show that valuation slopes are sensitive to the rating regime. Capital-intensive maritime and fisheries businesses may incur substantial expenditures for vessels, equipment, facilities, compliance, safety, and transition technologies, but the present design cannot identify whether such costs explain the estimated coefficients.

The social pillar has a negative value coefficient and a negative risk coefficient in the contemporaneous focal models. The negative social risk coefficient remains statistically detectable under the Wild Cluster Bootstrap and retains its sign in every leave-one-firm-out run, which makes it one of the more stable focal findings. However, lagged social performance is not significantly associated with risk, so the evidence does not establish a temporal or causal risk-reduction effect. Direct measures of safety, labor, supplier, customer, and community outcomes are needed before attributing the pattern to a specific operating channel.

The environmental results require similar caution. The environmental pillar is not significantly associated with focal-sample value in the separate focal model, yet the formal sector interaction for value is significant, indicating that the environmental-value slope differs from that of non-focal firms. For risk, the positive environmental coefficient remains statistically detectable under the Wild Cluster Bootstrap, survives the available ROA/distress-control specification, and remains positive and significant in all 24 leave-one-firm-out runs. Nevertheless, lagged environmental performance is not significant for risk, and the valuation relation changes around the 2022 KCGS revision. These patterns are consistent with sector-specific transition exposure but do not identify its source.

Governance performance is negatively associated with value in the conventional focal fixed-effects model but is not significantly associated with risk. The value coefficient weakens under the Wild Cluster Bootstrap (wild p = 0.093), so the evidence for governance is less robust to small-cluster inference than the baseline clustered result suggests. The formal Marine × G interaction is also not statistically significant for either outcome. Governance should therefore be interpreted as a focal-sample association rather than as a demonstrated sector difference.

5.2. Implications for ESG Strategy

For managers, the results caution against treating ESG primarily as a composite-score target. A change in the overall rating can reflect underlying E, S, and G movements that have different contemporaneous associations with market value and volatility. The formal interaction results further indicate that some of these associations differ from those observed among other KOSPI firms. Maritime and fisheries companies may therefore benefit from monitoring environmental outcomes, workforce and supply-chain indicators, governance quality, financing conditions, and transition exposure separately. The empirical results do not establish that changing any one ESG pillar will mechanically change value or risk.

The social results also suggest that immediate valuation and market-risk indicators need not move together. A social initiative that does not coincide with a higher contemporaneous Tobin’s Q may still be relevant to risk management, but this proposition requires direct evidence on safety, labor, supplier, customer, and community outcomes. The present results should therefore be used to formulate more specific tests rather than to infer an insurance effect from the ESG score itself. This direction is consistent with the growing attention on social sustainability in maritime ESG research [1,3].

5.3. Implications for Investors, Lenders, and Policy

For investors and lenders, the results illustrate the information lost when an aggregate ESG score is treated as a sufficient sustainability signal. In the focal sample, pillar-level coefficients differ across value and risk outcomes, and the detailed components do not move uniformly. Sector-specific materiality screens and disaggregated indicators may therefore complement aggregate ratings, particularly where environmental transition obligations, labor conditions, and capital requirements differ markedly across firms.

For policymakers, the findings support examining industry sensitivity in ESG evaluation rather than assuming uniform cross-industry relationships. The formal interaction results are strongest for environmental and social associations with market risk, while the 2022 rating-revision analysis also shows that valuation results can depend on the evaluation regime. This favors transparent, dimension-specific assessment and links between ratings and physical outcomes such as emissions, energy use, accidents, labor outcomes, and resource sustainability. The regressions do not estimate the effects of any specific subsidy, regulation, or disclosure mandate, so policy effectiveness requires designs that directly identify those interventions.

The distinction between inputs and outcomes is also important for evaluation. Environmental management systems, reporting activity, and transition expenditure are inputs or organizational responses, whereas emissions, energy intensity, accidents, labor outcomes, and resource sustainability are closer to realized operating outcomes. Future policy and financing assessments should link ESG ratings to such physical indicators before concluding that a change in a rating represents a corresponding change in environmental or social performance.

5.4. Limitations and Future Research

Several limitations bound the interpretation. First, the focal sample contains only 24 listed firms. Wild Cluster Bootstrap inference and leave-one-firm-out diagnostics reduce concern that the principal risk results are artifacts of conventional small-cluster inference or a single influential firm, but they cannot add cross-sectional information. Second, the balanced-panel requirement improves comparability over time but can generate coverage or survivor selection by excluding firms that delist, merge, experience financial distress, or lack continuous ESG coverage. The available analytical data contain complete 2012–2025 histories for the estimation firms, so a genuine pre-balance unbalanced-panel survivor-bias test cannot be reconstructed. The zero-coded-exclusion analysis produces an unbalanced estimation sample for a different purpose and is not presented as a solution to survivor selection. Third, listed firms are generally larger and more formalized than many small- and medium-sized businesses in fisheries, processing, marine services, and related supply chains, limiting external validity.

Fourth, the analysis is associational. Firm fixed effects control for time-invariant unobserved heterogeneity, but time-varying omitted variables and reverse causality remain possible. One-year-lagged specifications improve temporal ordering but do not establish causality, and some contemporaneous risk results weaken when ESG is lagged. Additional controls using the available ROA and negative-ROA distress indicator, together with industry-by-year fixed effects, address part of the omitted-variable concern. However, the supplied analytical file does not contain current-ratio or investment variables, and firm-specific freight exposure, stock-return shocks, financing distress, oil-price exposure, exchange-rate exposure, and other unobserved time-varying factors remain. Fifth, the main risk measure is annual stock-return volatility. Supplementary financial-leverage estimates provide one balance-sheet risk-related comparison, but the available file does not contain the daily return series or credit-market data needed to construct idiosyncratic volatility, downside volatility, default probability, credit spreads, or cost of debt. The exploratory ROA analysis also indicates that market-based findings should not automatically be generalized to accounting performance.

Sixth, the focal category combines shipping and logistics, fisheries, and shipbuilding-related firms. Leave-one-subsector-out diagnostics preserve the environmentally positive and socially negative risk signs but show that statistical strength varies with the composition of the focal portfolio, so the combined sample should not be interpreted as fully homogeneous. The analytical file preserves the operational focal flag and subsector labels but not a more granular provenance record for the original label construction; Appendix A therefore reports the complete roster and observed KSIC section and the study treats the flag as an operational research classification rather than an official industrial taxonomy. Seventh, the KCGS overall ESG indicator is retained on its source scale and modeled linearly in the baseline, while the pillar and component indicators are continuous proportions. The 2022 KCGS revision materially affects some valuation slopes, and the quadratic robustness check indicates possible nonlinearity for focal-sample risk. Eighth, the detailed-component regressions involve multiple related tests; applying the Benjamini–Hochberg correction leaves only EP for value and ES for risk significant at a 5% false-discovery rate.

Finally, the analysis reports formal sector-interaction tests and multiple robustness checks, but the remaining limitations should be kept explicit rather than treated as resolved. In particular, a genuine pre-balance unbalanced panel and some alternative market-risk measures cannot be recovered from the available data. The analytical workbook retains the continuous ESG field used in estimation but not the corresponding public KCGS letter-grade field, year-specific grade thresholds, or additional metadata needed to reconstruct a defensible grade-to-score mapping. The study therefore does not impose an unsupported conversion. Instead, the paper documents the observed 0.00–6.00 range of the ESG field, its direction, annual distributions, score transitions, the 2022 break, nonlinear and within-year categorical robustness, and the DataGuide-based source-variable definition for Tobin’s Q. Larger sector samples, richer provider metadata, and direct physical sustainability indicators would further clarify economic materiality across maritime subsectors.

6. Conclusions

This study examines associations between ESG performance, firm value, and market risk in a balanced panel of 556 KOSPI-listed non-financial firms observed from 2012 to 2025. The panel is partitioned into 24 maritime and fishery firms and 532 non-focal comparison firms for separate group analyses, while pooled interaction models use the full sample. The design combines firm and year fixed effects, disaggregated ESG indicators, formal sector-interaction tests, Wild Cluster Bootstrap inference, influence diagnostics, lagged specifications, and sensitivity checks for the 2022 KCGS rating revision and multiple testing.

Three empirical patterns summarize the results. First, the focal sample does not show a uniform positive valuation association with ESG: overall ESG and the social and governance pillars are negatively associated with Tobin’s Q, while the environmental pillar is not significant in the separate focal regression. In the non-focal comparison sample, overall ESG and the environmental pillar are negatively associated with value, and the baseline formal interaction test identifies an environmental-value slope difference. However, the joint value-sector test weakens in a fully interacted nuisance specification, so the valuation evidence is specification-sensitive. Second, the joint focal risk model yields a positive environmental coefficient and a negative social coefficient. The corresponding non-focal joint coefficients are only marginal at the 10% level and have the opposite signs, while both environmental and social sector interactions are statistically significant. These risk interactions remain statistically detectable after adding the available ROA/distress controls, under industry-by-year fixed effects, with common Winsorization thresholds, and in a fully interacted nuisance specification. The focal risk signs also persist in every leave-one-subsector-out sample, although significance varies, and the lagged models are weaker. Third, detailed-component evidence is substantially narrower after false-discovery-rate adjustment: environmental performance remains significant for value and environmental stakeholder communication for risk.

The principal implication is methodological and interpretive: aggregate ESG ratings can conceal different associations across dimensions, outcomes, sectors, and rating regimes. The most stable sector-specific evidence in this study concerns the environmental and social associations with market risk, while the valuation evidence is more sensitive to specification and timing. The interaction tests therefore support selected sector differences rather than a blanket maritime effect, and the sensitivity analyses show where the evidence is stable and where it is not. Managers, investors, lenders, and policymakers should avoid assuming that a higher composite rating necessarily implies a contemporaneous valuation premium or lower market volatility. The estimates remain associational and do not identify the causal effects or the channels behind the coefficients.