Studying the impact of profitability, bankruptcy risk, and pandemic on narrative tone in annual reports in an emerging market in the East

The narrative content of annual reports is increasingly central to corporate reporting, providing essential, price-sensitive, and sustainability-related information beyond traditional accounting measures. Following major accounting scandals, interest in narrative disclosures intensified, prompting research into how narrative tone relates to firm performance and whether it serves communication or impression management purposes. Prior work shows narrative tone may predict profitability and bankruptcy risk and correlates with firm performance. This study examines whether narrative tone in Vietnamese listed non-financial firms’ annual reports is influenced by profitability, bankruptcy risk, and the COVID-19 pandemic. It asks: (1) Is narrative tone influenced by profitability, bankruptcy risk, and the pandemic? (2) Do profitable and unprofitable firms differ in tone? (3) Do financially distressed/bankruptcy-risk firms differ in tone versus non-distressed? (4) Does tone differ before, during, and after COVID-19? Contributions: (i) investigates an Eastern emerging market (Vietnam) with distinct cultural and linguistic characteristics relative to Western settings; (ii) constructs a Vietnamese finance/accounting sentiment lexicon by translating and adapting Loughran–McDonald and augmenting with locally used terms; (iii) analyzes tone across groups (profitable vs. unprofitable; bankruptcy-risk vs. non-risk; pre-, during, and post-COVID-19). Using a large panel (468 firms; 4,212 reports; ~250k pages), the study initially finds that higher profitability coincides with more positive and net tone and less negative tone. Unprofitable firms show more negative tone, while differences in positive tone between profitable vs. unprofitable firms are limited. Financially distressed firms use more negative and less net tone. During COVID-19, firms used more positive and negative tones; post-COVID-19, positive tone rose and negative tone fell compared to the outbreak period. Findings inform impression management, agency, and signaling theories, underscoring the need for investors to assess narrative tone critically.

Theoretical basis integrates impression management, agency, and signaling theories to explain how and why narrative tone conveys value-relevant information. Impression management posits managers strategically present information to shape stakeholders’ perceptions, potentially accentuating positives and downplaying negatives. Agency theory suggests managers may disclose more favorable narratives to bolster investor confidence and mitigate perceived agency conflicts, sometimes engaging in earnings/tone management. Signaling theory explains the use of positive/negative language as signals to reduce information asymmetry and differentiate firm quality. Literature review highlights the “narrative turn” in accounting, expanding use of rhetorical and narratological tools. Textual analysis methods span dictionary-based approaches and machine/deep learning. Prior research documents links between tone and profitability (e.g., Li 2010; Aly et al. 2018; Kang et al. 2018), and between tone and bankruptcy risk (Lopatta et al. 2017; Lohmann and Ohliger 2020). Studies during crises (e.g., COVID-19) show firms vary emotional tones, often using more negative affect to convey concerns. Hypotheses developed include: H1 (profitability positively relates to positive and net tone, negatively to negative tone); H2 (tone differs between profitable vs. unprofitable firms); H3 (tone differs between distressed/bankruptcy-risk vs. non-distressed firms); H4 (tone differs before, during, and after COVID-19); H5 (bankruptcy risk and COVID-19 moderate the profitability–tone relationship).

Design: Panel study of Vietnamese listed non-financial firms on HOSE and HNX from 2015–2023, covering pre-, during-, and post-COVID-19 periods and the disclosure regime under Circular 155/2015/TT-BTC. Sample: 468 non-financial firms; 4,212 annual reports; ~250,013 pages. Data processing: Annual reports (PDF/Word/scans) converted to text using OCR and Python. Narrative tone measured by frequency of positive and negative words using a Vietnamese finance/accounting sentiment lexicon built by translating Loughran–McDonald (2011) and augmenting terms after reviewing ~200 Vietnamese annual reports. Dependent variables: POSTONE (positive word frequency), NEGTONE (negative word frequency), NETTONE (POSTONE−NEGTONE). Main model (1): TONE_it = β0 + β1 ROA_it + β2 PLOS_it + β3 DIST_it + β4 COVID_it + Σγ Controls_it + ε_it, where TONE ∈ {POSTONE, NEGTONE, NETTONE}. Key independents: ROA (return on assets); PLOS (loss dummy: 1 if ROA<0); DIST (Altman Z''-score based categories: 0 safe>5.85; 1 grey 4.15–5.85; 2 distress<4.15); COVID coded as 1 pre (2015–2019), 0 during (2020–2021), 2 post (2022–2023). Controls: SIZE (ln assets), AGE (ln years listed), AUD (Big4=1), RET, SRET (volatility), BTM, LIQ, LEV, FGRO, ACC, LENG (ln non-numeric word count). Moderation models: (2) add ROA×DIST dummies to test bankruptcy-risk moderation (H5a); (3) add ROA×COVID dummies to test COVID moderation (H5b). Estimation strategy: OLS, REM, FEM tested; Hausman suggests FEM for POSTONE and NETTONE, REM for NEGTONE. Diagnostic tests (White heteroscedasticity, Wooldridge autocorrelation) indicate heteroscedasticity/autocorrelation (p=0.000), so FGLS is used for efficient inference. Robustness: paired bootstrap (1000 iterations) on FGLS; dynamic panel with SGMM including up to three lags of TONE (model 4) to address endogeneity and dynamics: TONE_it = β0 + β1 TONE_i,t−1 + β2 TONE_i,t−2 + β3 TONE_i,t−3 + β4 ROA_it + β5 PLOS_it + β6 DIST_it + β7 COVID_it + Σγ Controls_it + ε_it. Descriptives: Mean POSTONE 3.46% (min 0.13%, max 7.63%); mean NEGTONE 1.22% (min 0, max 2.85%); mean NETTONE 2.24% (min −0.28%, max 6.86%). VIF avg 1.59 (max 2.67), indicating no multicollinearity. COVID-period observations ≈22.22% of total.

- Profitability and tone (FGLS): - POSTONE: ROA positively related (β≈0.0066, p<0.01). - NEGTONE: ROA negatively related (β≈−0.0032, p<0.01). - NETTONE: ROA positively related (β≈0.0099, p<0.01). - Supports H1a, H1b, H1c. - Profitable vs. unprofitable (PLOS): - NEGTONE higher for loss firms (β≈+0.0007, p<0.01); NETTONE lower (β≈−0.0010, p<0.01); POSTONE difference not significant. Supports H2 for negative and net tone. - Bankruptcy risk (DIST): - Distress (DIST=2) associated with higher NEGTONE (β≈+0.0004 to +0.0005, p<0.01) and lower NETTONE (β≈−0.0010, p<0.01); POSTONE shows no significant difference. Supports H3. - COVID-19 periods (vs. during-pandemic baseline 2020–2021): - Pre-COVID (2015–2019): lower POSTONE (β≈−0.0005, p<0.05), lower NEGTONE (β≈−0.0011, p<0.01), higher NETTONE (β≈+0.0004, p<0.10). - Post-COVID (2022–2023): higher POSTONE (β≈+0.0005, p<0.05), lower NEGTONE (β≈−0.0002, p<0.05–0.10), higher NETTONE (β≈+0.0007, p<0.01). - During the outbreak, firms used more positive and negative tones than pre-COVID; after stabilization, positive increased and negative decreased relative to the outbreak. Supports H4. - Moderation (H5): - Bankruptcy risk × ROA: Significant only for NEGTONE at high risk (DIST=2×ROA β≈−0.0032, p<0.05): when bankruptcy risk is high, the negative association between ROA and NEGTONE strengthens (more negative tone as profitability declines). H5a effectively shows moderation for negative tone; authors note broader H5a rejected for other tones. - COVID × ROA: Interactions insignificant for all tones; H5b rejected. - Controls (FGLS): - SIZE: +POSTONE, −NEGTONE, +NETTONE (all p<0.01). - BTM: −POSTONE, +NEGTONE, −NETTONE (significant where reported). - LIQ: −POSTONE and −NETTONE (p≤0.01); NEGTONE ns. - LEV: Negative with POSTONE, NEGTONE, and NETTONE (p<0.01). - FGRO: −POSTONE and −NETTONE (p<0.01); NEGTONE ns. - AUD (Big4): +POSTONE and +NETTONE (p<0.01); NEGTONE ns. - LENG: −POSTONE and −NETTONE; +NEGTONE (all p<0.01). - RET: weak negative relation to NEGTONE (p<0.10) only. - SRET: −POSTONE (p<0.05), −NETTONE (p<0.01); NEGTONE ns. - Robustness: - Bootstrap (1000 iterations) yields small SEs; effects remain significant and stable. - SGMM: Lagged tones (t−1, t−2) significantly predict current tone; ROA effects persist (positive on POSTONE/NETTONE, negative on NEGTONE). COVID effects weaken in significance; PLOS and DIST lose significance in SGMM, indicating tone persistence is strong.

Findings address the research questions by showing profitability strongly shapes narrative tone: more profitable firms emphasize positive language and reduce negative language, aligning with agency and signaling theories that managers communicate success to reduce information asymmetry and build investor confidence. Differences between profitable and unprofitable firms manifest primarily in negative and net tones, suggesting managers of loss-making firms do not meaningfully increase positive tone but do acknowledge difficulties through more negative wording. Financial distress and bankruptcy risk are reflected in more negative and less net tone, indicating limited use of impression management under stress and a tendency to convey conditions transparently. The moderation by bankruptcy risk strengthens the inverse profitability–negative tone link, reinforcing that adverse performance under high risk elicits more candid negative narratives. COVID-19’s direct effects on tone show firms used both more positive and more negative language during the outbreak (communicating resilience and challenges). Post-pandemic stabilization shifted tone towards more positivity and less negativity, consistent with signaling recovery. However, COVID-19 did not moderate the profitability–tone relationship, implying profitability remains the primary driver of tone regardless of crisis context. Control variables reveal that larger, Big4-audited firms tend to present more positive and net tone, while higher leverage, higher BTM, longer reports, and higher volatility relate to more cautious or negative-leaning narratives. Overall, results suggest narrative tone in this Eastern emerging market conveys informational content about profitability and financial condition more than it serves as opportunistic impression management during macroeconomic stress.

This study shows that profitability increases positive and net tone while reducing negative tone in Vietnamese firms’ annual reports. Loss-making firms convey more negative and less net tone than profitable firms. Financial distress elevates negative tone and reduces net tone. During COVID-19, firms used more positive and negative tone than pre-pandemic; in the post-pandemic period, positive tone rose and negative tone fell relative to the outbreak. Bankruptcy risk strengthens the inverse link between profitability and negative tone, while COVID-19 does not moderate profitability’s impact on tone. These results support impression management, agency, and signaling theories, indicating that while managers signal strong performance, they are less likely to conceal difficulties during adverse conditions. Implications: Investors and stakeholders can use narrative tone as a signal of profitability and financial health but should remain vigilant to potential strategic tone usage. Policymakers and regulators may encourage transparent narrative practices to reduce information asymmetry. Future research: Extend samples beyond Vietnam to other emerging markets, examine predictive power of tone for future profitability, and analyze broader narrative features (readability, rhetoric, storytelling). Incorporate social media disclosures and advanced NLP/deep learning methods for richer textual insights.

- External validity: Sample limited to Vietnamese listed non-financial firms; results may not generalize to other countries or sectors. - Scope of narrative analysis: Focus on tone only; other narrative attributes (readability, rhetoric, storytelling) not analyzed. - Dictionary approach: Vietnamese sentiment lexicon derived from translation and manual augmentation may omit context-specific nuances; alternative NLP methods could yield different insights. - Dynamic/endogeneity controls: Although SGMM and lags are included, residual endogeneity or omitted variables may remain. - COVID coding: Period classification (pre/during/post) is coarse and may not capture firm-specific pandemic impacts or timing heterogeneity.