Fashion firms show no robust digital-sustainability financial synergy

A time-lagged observational study of 73 publicly listed fashion companies found no robust evidence that digital capability, sustainable capability, or their combination was associated with stronger financial performance. The analysis identified no statistically reliable synergy between the two agendas, while an apparent advantage for fast-fashion firms in ordinary least-squares estimates disappeared when the researchers applied bootstrap resampling and heteroskedasticity-consistent standard errors. The result challenges the assumption that digital and environmental transformations automatically reinforce one another, although the modest sample could not reliably detect small interaction effects.

The research focused on global apparel manufacturing, apparel retail, and luxury-goods companies whose financial information was publicly available. Corporate disclosures from fiscal year 2024 were used to measure capabilities, while operating performance and market valuation were assessed with trailing 12-month data extending into the 2025 and 2026 fiscal periods. This one-year separation was intended to establish temporal order and reduce simultaneity bias: disclosed strategic action would precede the financial outcomes observed in the analysis.

It cannot, however, eliminate omitted-variable bias or demonstrate causation, so the findings are reported as directional associations rather than proof that the capabilities caused changes in performance. The authors grounded their model in dynamic capabilities and institutional theory. Digital capabilities were treated as “sensing” mechanisms that can improve visibility into demand, inventory, and supply chains, potentially helping firms detect changing conditions and redirect resources.

Sustainable capabilities were framed as “transforming” mechanisms, encompassing practices such as circular business models, supply-chain transparency, and decarbonization programs that reorganize how physical operations work. This framework captures a plausible technical link between the two: digital systems can provide the information needed to identify waste, trace materials, and coordinate environmental interventions. The study nevertheless asked whether that theoretical connection appears in financial results when digital sensing and sustainable transformation are deployed independently and together.

Measuring genuine capability rather than promotional language was a central technical challenge. The researchers built a computer-assisted text-analysis pipeline using Python’s spaCy library and syntactic dependency parsing. Rather than simply counting whether terms such as “digital” or “sustainability” appeared, the algorithm identified the operational verb governing each capability keyword.

Sixteen distinct verbs were assigned scores on a 0–5 ordinal scale, with passive expressions such as “exploring” receiving less weight than active language such as “deployed.” The procedure also nullified statements containing negation or cancellation terms, including words such as “not,” “abandon,” “delay,” and “failed,” and normalized each firm’s score by the total number of words analyzed to limit distortion from unusually long reports. A reliability check compared the algorithm with manual coding for a random 20% subsample, approximately 15 firms. The resulting Cohen’s kappa exceeded 0.85, indicating high agreement, although the primary researcher both developed the capability dictionary and performed the manual comparison, limiting the degree of independence in the validation.

The final company classifications were also generated algorithmically. K-means clustering divided the firms according to gross margin and inventory turnover into 42 low-margin, high-velocity companies labelled Fast Fashion and 31 high-margin, low-velocity companies labelled Luxury. The two-cluster solution had the highest silhouette score among the configurations tested, 0.398, but this indicated only moderate separation.

Post-hoc tests found that the fast-fashion group had faster inventory turnover and lower gross margins than the luxury group, with both differences significant at p less than 0.001. Ordinary least-squares regression included controls for firm size, firm age, geographic scope, and the number of report tokens analyzed, with continuous variables Winsorized at the fifth and 95th percentiles. The authors entered control variables first, followed by mean-centered digital and sustainable capabilities, their interaction, and segment-specific moderation terms.

Centering reduced the maximum variance inflation factor from 8.30 to 4.51. Because the financial data showed unequal residual variance, the researchers also used HC3 robust standard errors, while a 5,000-iteration bootstrap tested whether estimates remained stable without relying on normal-distribution assumptions. For return on assets, the full model explained 36.8% of the observed variation, but this fit statistic did not translate into support for the proposed capability effects.

Digital sensing had no statistically significant unconditional association with operating performance: its coefficient in the block without interaction terms was 2.507, with p equal to 0.200. Sustainable transformation likewise had no significant sample-wide association, with a coefficient of 0.620 and p equal to 0.131. Their interaction term was negligible, with a coefficient of 0.506, p equal to 0.723, and an increase in explained variance of less than 0.001.

Adding digital and sustainable capabilities to the control model raised R squared from 0.227 to 0.282, but that block was not conventionally significant, with F(2, 65) equal to 2.48 and p equal to 0.092. The apparent fast-fashion advantage for digital capability in the initial parametric model was also unstable. The interaction’s bootstrap 95% confidence interval ran from −1.459 to 17.004 and included zero.

The HC3 standard error of 5.539 was 66% larger than the conventional OLS estimate of 3.338, and the corresponding p-value rose from 0.010 to 0.111. The moderation term for sustainable capability was not significant under conventional estimation either, with a coefficient of −0.918 and p equal to 0.434; its bootstrapped confidence interval ranged from −3.026 to 1.283. The two segment-moderation terms were significant when entered together, adding 0.086 to explained variance, with F(2, 62) equal to 4.21 and p equal to 0.019, but neither digital nor sustainability moderation was robust when evaluated separately.

That joint signal was therefore presented as a direction for larger studies rather than proof of a particular conditional effect. The market-valuation model produced an even less conclusive result. Its overall fit was not statistically significant, with F(10, 62) equal to 0.849, p equal to 0.584, and R squared of 0.075.

Because the model could not distinguish explanatory signal from random variation, the authors did not interpret its individual coefficients for digital capability, sustainable capability, synergy, or market segment. The authors proposed that the absence of a detectable valuation premium might reflect a “legitimacy gap,” in which firms adopt sustainability structures to satisfy external expectations without receiving an immediate market reward. The text algorithm was specifically designed to distinguish substantive strategic action from empty rhetoric, but the analysis did not establish investor motives or determine how markets price the capabilities.

The legitimacy-gap account therefore remained a conceptual proposition for subsequent research rather than a proven mechanism. The study’s limits are important to the apparently clean conclusion that “twin transformation” is not automatically profitable. With 73 firms, the design had limited power to detect small effects or subtle interactions; its reported minimum detectable effect was f squared of 0.111, leaving scope for Type M magnitude errors.

The sample excluded private companies, small enterprises, broadline department stores, and other sectors, so generalization beyond listed apparel and luxury businesses is uncertain. Important controls such as prior-year performance and capital structure were unavailable, and disclosure-based measurement could not capture capabilities that firms chose not to reveal. The segment labels also depended on two selected clustering variables, while alternative indicators could produce different assignments.

Longitudinal research with larger samples, direct measures of organizational integration, and more complete financial controls would be needed to determine whether synergies emerge slowly, operate differently within particular business models, or remain absent in publicly traded fashion firms.

Subject of Research: Digital and sustainable capabilities and financial performance among publicly listed fashion companies

Article Title: The Twin Transformation illusion: testing the limits of digital and sustainability synergy in fashion

Article References: Aitnazarova, A., Haj Ammar, O., & Remsei, S. (2026). The Twin Transformation illusion: testing the limits of digital and sustainability synergy in fashion. Frontiers in Sustainability, 7, 1977391. https://doi.org/10.3389/frsus.2026.1977391

Image Credits: AI Generated

DOI: 10.3389/frsus.2026.1977391

Keywords: fashion industry, digital transformation, corporate sustainability, firm performance, digital-sustainability synergy, return on assets, market valuation, syntactic dependency parsing, dynamic capabilities, publicly listed companies, Twin, Transformation