Reducing asymmetric cost behaviors: Evidence from digital innovation

The paper examines how firm-level digital innovation affects asymmetric cost behavior (cost stickiness), a phenomenon where costs increase more with rising sales than they decrease with falling sales. Amid the pandemic-driven economic downturn and accelerated digital economy, the study distinguishes digital innovation—combining digital technologies with products, processes, and business models—from broader digital transformation. Prior work has focused largely on digitization levels or narrow technologies (e.g., robots) and often lacks robust mechanism analysis. The research question is whether and how digital innovation mitigates cost stickiness, improving cost efficiency and firm performance. The study argues that digital innovation could reduce cost stickiness via enhanced internal controls, improved resource-adjustment efficiency, and tempered managerial over-optimism, thereby underscoring its strategic importance beyond mere digitization.

The review situates digital innovation within broader corporate digitalization research, which links digitization to gains in productivity, innovation capacity, financial performance, resilience, value creation, financing access, and environmental outcomes. Digital innovation is framed along three dimensions: product/service innovation (leveraging AI, IoT, blockchain to enhance offerings), operational process innovation (automation, robotics, BPM to boost efficiency and flexibility), and business model innovation (new revenue models and customer engagement through data analytics and AI). The cost stickiness literature identifies managerial decisions as central, with key drivers including resource-adjustment frictions, managerial expectations (over-optimism), and agency issues. Governance mechanisms (board structure, ownership, investor scrutiny) and incentives influence cost asymmetry. Building on these, the paper proposes that digital innovation reduces cost stickiness through three channels: strengthening internal controls (data-driven risk assessment, standardized procedures, audit quality), improving resource-adjustment efficiency (ERP-enabled transparency, faster asset and labor adjustments), and mitigating managerial over-optimism (data-informed decisions, enhanced managerial digital competencies).

Data: Panel of Chinese A-share listed firms (Shanghai/Shenzhen) from 2007–2022 drawn from CSMAR and Wind. Exclusions: financial firms, delisted firms, *ST/ST firms, dual A&B shares, IPO year or earlier, negative total assets/equity, missing data, and anti-sticky observations (positive STICK). Final sample: 10,822 firm-year observations.

Cost stickiness measurement: Following Weiss (2010), the study computes STICK as the log-difference in the slope of the cost function between the latest quarters of sales increases vs decreases within year t, using operating costs (COST) and sales (SALE). Negative STICK indicates stickiness; the analysis retains only negative STICK and uses Abs_STICK = |STICK| as the dependent variable (higher values denote greater stickiness). Robustness includes an ABJ (Anderson, Banker, Janakiraman, 2003) model on the full sample (positive and negative STICK) to test asymmetric cost behavior and interaction effects with digital innovation.

Digital innovation (DI): Firm-level digital innovation is proxied by the logarithm of (1 + count) of firm patent applications in key digital technologies. Patents are identified via text analysis (abstracts, descriptions, claims) and matched to categories from the China National Intellectual Property Administration’s “Key Digital Technology Patent Classification System (2023).” A robustness alternative uses an expanded classification-based count (DIGITAL_r).

Controls: Firm size (log assets, SIZE), age (AGE), asset tangibility (TANG), gross profit margin (GrossProfit), liquidity (LIQUID), two-year consecutive sales decline dummy (IncomeDD), sales growth (GROWTH), independent director share (INDIR), CEO-chair duality (DUALITY), institutional ownership (INST), and operating expense ratio (OER). Continuous variables are winsorized at 5%/95%.

Baseline model: Abs_STICK_it = β0 + β1 DI_it + X_it + year FE (θ_t) + industry FE (η_i) + ε_it, with firm-level cluster-robust standard errors.

Endogeneity strategies: (1) Instrumental variables: city-year average DI of other firms (excluding firm i) as IV, leveraging common local digital infrastructure. Two-stage least squares with standard diagnostics (Kleibergen-Paap Wald rk F and Cragg-Donald F). (2) Propensity score matching (PSM): treated = DI above 75th percentile; 1:1, 1:2, 1:3 nearest-neighbor matching; balance tests confirm covariate balance. (3) Placebo tests: random assignment of DI repeated 1000 times to test spurious correlations.

Robustness: Alternative DI measure (DIGITAL_r); alternative fixed-effects structures (industry-year, city-year); ABJ cost stickiness specification with interactions (IncomeD × SaleR × DI) and controls to confirm that DI attenuates stickiness.

Mechanism (mediation) tests: Three-channel mediation using equations:

• CONTROL_it = α0 + α1 DI_it + … (internal control quality via log DiBo Internal Control Index).
• TURNOVER_it = α0 + α1 DI_it + … (resource-adjustment efficiency via total asset turnover).
• STREAK_it (managers’ forecast success streak per Hilary et al., 2016) and an alternative CEO relative compensation (Compen, per Hayward & Hambrick, 1997) as proxies for managerial over-optimism. Mediation is inferred if DI significantly affects the mediator and DI’s coefficient magnitude on Abs_STICK declines when the mediator is included.

Heterogeneity analyses: Cross-sections by firm size (Big dummy above industry median sales), life cycle (Growth dummy per Dickinson, 2011 cash-flow patterns), and regional digital environment: digital governance (did1; Information Benefit Pilot Policy cities post-2014), digital taxation (Golden Tax Project III; did2), and IP protection level (did3). Model (2) is re-estimated within each subgroup.

Further analyses: Two-stage framework assessing benefits of reducing stickiness. First stage estimates predicted Abs_STICK via Model (2). Second stage regresses risk and performance measures on Abs_STICK: risk via industry-adjusted ROA volatility (RISK1, std. dev.; RISK2, range) over three years; profitability via net profit margin (PERF1) and ROA (PERF2).

Superiority of digital innovation vs general digital transformation: Splitting the sample by DI activity (DI_dummy2 = 1 if firm has any digital innovation) and testing whether digital transformation measures—keyword frequency in MD&A (DT) and digital intangibles share (DA)—relate to Abs_STICK within innovating vs non-innovating firms using Abs_STICK = β0 + β1 DT + controls.

• Baseline: Digital innovation significantly reduces cost stickiness. In the fully specified model (Table 2, Col. 4), DI’s coefficient is −0.0233 (t = −2.905), indicating that a one-rank increase in DI associates with a 0.0233 decrease in Abs_STICK.
• Endogeneity (IV): The city-year average DI (excluding the firm) is a strong instrument (first-stage IV coefficient positive and significant at 1%; Cragg-Donald F = 97.256; Kleibergen-Paap rk Wald F = 38.368). Second stage retains a negative, significant DI effect on Abs_STICK (Table 3), corroborating causality.
• PSM: Across 1:1, 1:2, and 1:3 matches, DI remains negative and significant (e.g., −0.0178*, −0.0197**, −0.0210**; Table 4), confirming robustness to selection.
• Placebo: Randomized DI assignments yield coefficients centered near zero and insignificant (Figure 3), supporting non-spurious findings.
• Additional robustness: Alternative DI measure (DIGITAL_r) and alternative fixed-effects structures keep DI significantly negative (Table 5). The ABJ model on the full sample shows DI reduces stickiness (IncomeD × SaleR × DI coefficient −0.0206**, Table 6), while base effects confirm stickiness (SaleR positive; IncomeD × SaleR negative).
• Mechanisms: • Internal control: DI increases internal control quality (α1 > 0, 0.0258***), and DI negatively affects Abs_STICK (β1 < 0). Including CONTROL reduces the absolute DI effect (|δ1| < |β1|), indicating mediation (Table 7). • Resource-adjustment efficiency: DI raises turnover (α1 = 0.0333***). DI’s negative effect on Abs_STICK attenuates after adding TURNOVER (|δ1| < |β1|), supporting mediation (Table 8). • Managerial over-optimism: DI reduces STREAK (−0.0241***), and DI’s negative impact on Abs_STICK persists with attenuation when including STREAK; results hold using CEO relative compensation (Compen) as an alternative proxy (Table 9).
• Heterogeneity: • Firm size: Effect concentrated in larger firms (Big = 1: DI = −0.0278***; Big = 0: not significant; Table 10). • Life cycle: Stronger in non-growth firms (Growth = 0: DI = −0.0332***; Growth = 1: not significant; Table 11). • Region: Stronger where digital governance is advanced (did1 = 1: significant negative DI; did1 = 0: negative but not significant). Similar patterns for digital taxation system maturity (did2) and strong IP protection (did3): significant negative effects in high-development/protection regions, insignificant elsewhere (Table 12).
• Benefits of reducing stickiness: In two-stage analyses, higher Abs_STICK is associated with higher ROA volatility (risk) and lower profitability, while reducing stickiness aligns with improved financial outcomes (Table 13; signs and significance reported across risk and profitability specifications).
• Superiority of digital innovation: Digital transformation (DT, DA) significantly reduces cost stickiness only among firms that also engage in digital innovation (DI_dummy2 = 1: β1 < 0 and significant at 10%; DI_dummy2 = 0: β1 not significant; Table 14), emphasizing the primacy of innovation over mere digitization.

The study demonstrates that firms’ digital innovation materially weakens asymmetric cost behavior by enhancing internal governance, improving the agility and efficiency of resource reallocation, and tempering managerial over-optimism. These mechanisms link digital innovation to more flexible cost structures, directly addressing inefficiencies inherent in cost stickiness. The effects are especially salient in larger and later-stage firms—where coordination, resource allocation, and managerial discretion are more complex—and in regions with mature digital governance, taxation, and IP protection, which provide supportive institutional environments. Collectively, the results establish digital innovation as a key lever in cost management strategy, offering firms resilience and strategic advantage by enabling faster, data-driven cost adjustments aligned with demand fluctuations. The findings also position digital innovation as a critical component within broader digital transformation programs, with innovation-led efforts yielding superior cost-behavior improvements compared to non-innovative digitization alone.

Using patent-based measures of key digital technology innovation for Chinese listed firms (2007–2022), the paper finds that digital innovation significantly reduces cost stickiness. The results remain robust to IV, PSM, placebo, alternative measures, and the ABJ model. Mechanism tests indicate that improved internal control quality, enhanced resource-adjustment efficiency, and reduced managerial over-optimism mediate the effect. Heterogeneity analyses show stronger effects in large firms, non-growth-stage firms, and regions with advanced digital governance, taxation, and IP protection. Further analyses suggest that lowering cost stickiness reduces risk and enhances profitability, and that digital transformation’s impact on cost behavior is meaningful primarily when accompanied by digital innovation. Policy prescriptions emphasize prioritizing digital innovation (through R&D and tax incentives and funding), tailoring support to large and mature firms, upgrading regional digital infrastructures and legal frameworks, embedding digital innovation into risk and financial management, and investing in workforce digital skills development.