Saying Is Believing: How Do Digital Disclosure Anomalies Affect Corporate Innovation Efficiency

Abstract

Digital transformation has become a new driving force for corporate development, widely embraced by managers. However, in the actual digital transformation process, some companies’ performances are characterized by “more talk, less action”. To measure the phenomenon, this paper innovatively constructs the Digital Disclosure Anomaly (DDA) indicator using Chinese A-share listed companies’ data from 2013 to 2021 and examines the impact of DDA on corporate innovation efficiency by Saying is Believing theory. We find that DDAs show a long-term negative impact on corporate innovation efficiency. Our mechanism analysis suggests that DDAs reduce innovation efficiency through resource misallocation, which is driven by two mechanisms: the manager squandering effect and the inefficient investment effect. Our results suggest that DDAs are harmful for corporate innovation efficiency by the Saying is Believing (SIB) effect and corporate DDA behavior is detrimental to sustainability in digital economy development.

1. Introduction

Digital transformation has become a critical factor of global economic development, playing a key role in enhancing enterprise innovation, optimizing resource allocation, and improving economic efficiency [1,2,3]. In China, digital transformation supports the national digital economy strategy by integrating advanced technologies such as data, algorithms, and computing power into various business scenarios. This integration fosters high-quality digital economic development and sustainable growth [4,5,6].

However, digital transformation does not always deliver its intended benefits. Recent reports and market analyses reveal instances of underperformance. For example, “Liao Wang” Weekly News pointed out that the performance of enterprises in the digital transformation arena is uneven in 2022. It presented that some enterprises have more plans, but fewer practices “https://lw.news.cn/2022-02/14/c_1310468941.htm (accessed on 4 May 2024)”. These anomalies not only distort market expectations but also introduce inefficiencies into corporate innovation and investment practices.

A major challenge in addressing these anomalies lies in the lack of quantitative methods to measure and evaluate their presence and impact. Current assessments of digital transformation rely heavily on qualitative analyses. These methods often fail to detect subtle but systemic issues, such as inflated digital disclosures or inefficient investment patterns. The absence of robust quantitative metrics limits the ability of stakeholders to recognize and mitigate these challenges effectively.

Developing quantitative measures for digital transformation anomalies is both necessary and meaningful. First, such measures help investors evaluate corporate performance, reducing information asymmetry and protecting their interests. Second, such measures can guide regulatory bodies and policymakers in fostering a more transparent and equitable capital market, which is essential for sustainable and high-quality economic development. This study addresses the gap by quantifying DDAs and examining their implications for corporate innovation efficiency, thereby contributing to the sustainability and high-quality growth of the digital economy.

2. Literature Review

Many scholars have explored the factors influencing digital transformation. Externally, a favorable business environment will accelerate digital transformation by attracting government subsidies [7], and the uncertainty of climate policy will inhibit corporate digital transformation by increasing information asymmetry [8]. Other influencing factors include local competition [9], digital infrastructure [10], and the clustering level of digital services [11]. Internally, factors include technical founders [12] and the IT experience of managers [13].

The implementation of digital transformations by enterprises has a comprehensive self-promotion effect. These effects include improving corporate ESG performance [14], lifting supply chain management levels [15], reducing carbon emission intensity [16], narrowing the gap between management and ordinary employees [17], green innovation [18], etc. Research on the role of digital transformation in corporate innovation has attracted the attention of many scholars, because innovation is the core driving force for enterprise development [19,20]. The findings indicate that firm digitalization can promote firm innovation. Furthermore, higher digital industrialization levels in a region enhance the marginal innovation efficiency of nearby firms through a “competitive effect” [21].

The current measurement method for the digital transformation level of listed companies mainly relies on word frequency calculations of related “words” in the annual reports of listed companies [22], and some also include digital personnel or digital-related input and output data for improvement. In fact, the digital word frequency calculation in the annual report can be considered as “what they said” while the digital personnel or digital-related inputs can be considered as “how they did” [23]. These two behaviors are not completely consistent in reality. Text analysis is crucial for identifying inconsistencies, especially within the context of DDAs. Analyzing textual data from annual reports helps detect gaps between corporate statements and actual behavior [24].

Many previous studies have focused on the phenomenon of “inconsistency between words and deeds” in information disclosure. Their topics include “management tone manipulation”, where the management tone deviates from corporate fundamentals [25]. Advanced natural language processing (NLP) techniques are used to capture the identification of patterns in management tone, contextual sentiment, and even hidden biases in corporate communication [26]. Scholars believe that financing needs [27], earnings management [28], and market pressure [29] are vital reasons behind tone anomalies. Other studies have examined inconsistencies in disclosures of corporate social responsibility [30,31], green development [32], and investment efficiency [27], finding that such disclosure strategies are structurally disordered and are rejected by investors and regulators.

The relationship between corporate innovation and digital transformation has been widely studied. At the corporate level, Xue [33] suggests that digital transformation promotes green technology innovation. At the regional level, green innovation is considered critical for improving digital infrastructure across China [18]. Previous studies mainly explore how digital transformation influences innovation and the methods used to analyze this impact.

From the review of the literature above, it is not difficult to find that current research on corporate digital transformation primarily focuses on the influencing factors and their effects as well as the effect between digital transformation itself and innovation. However, there is a lack of enough attention being paid to the phenomenon of “inflated” digital information disclosure, namely, digital disclosure anomalies (which will be called DDAs in this paper), in the market. The following discussion will construct a measurement method for DDA, and based on the “Saying is Believing” theory, analyze the impact of listed companies’ DDA on corporate innovation efficiency and the methodology.

Therefore, the contributions of this paper include (1) developing a new way to measure the difference between what companies say about their digital transformation and what they actually do; (2) using the “Saying is Believing” theory to research the abnormal phenomenon of corporations “talking more than doing” and providing an explanation; (3) studying the impact of DDAs on corporate innovation efficiency through two mechanisms.

3. Theoretical Analysis

Symbol management theory [34] proposes that communicators may adjust the information they convey according to the preferences of their audience. Moreover, during the communication process, the communicator’s own views and memories are also influenced. While providing biased information, communicators will begin to believe what they are conveying, even if the information-related events did not occur [35]. This is the “Saying is Believing” (SIB) effect. Influential factors like communication goals, feedback and the relationship of the communicator will induce the SIB effect [36]. The SIB effect has been confirmed in psychology [37] and marketing [38]. Psychological research has found that when people are in a positive emotional state, they tend to make optimistic decisions [39], which may lead to managers’ overconfidence. Managers will be overconfident when they believe the digital programs have been accomplished by the SIB effect, and they will overestimate the returns of their investment projects and the costs of external financing [40], which may lead to resource distortions. The flow chart demonstrating how DDAs influence corporate innovation efficiency is presented in Figure 1 below.

The stock market is essentially an “information market”. Listed companies disclose corporate operating information through annual reports and other channels. Investors predict company profitability based on the collection and analysis of this information [41]. According to the SIB effect, public company DDA is a two-step process. First, corporations generate exaggerated information. Digital transformation is a key indicator of modern enterprise competitiveness, and the digital information disclosed by the company will affect investors’ perceptions and the company’s stock price. Companies will send positive signals to the market by emphasizing their digital achievements, trying to highlight their future growth potential and industry leadership.

Additionally, management may be driven by personal interests, cater to investors’ expectations and maybe over-promoting digital achievements to stabilize stock prices. Because of these motivations, listed companies may “hollow reputation without basis” into digital information disclosure, which creates DDA information. Second, corporations disclose overvalued digital-related information. Through information disclosure methods such as annual reports, listed companies disclose whitewashed digital information after many internal meetings about digital transformation disclosure. According to the SIB effect, managers not only believe the information they disclose [42], but they also become overconfident. This overconfidence will lead them to overestimate the return of other programs [43,44].

Because of SIB effect, overconfident managers think they have achieved what they promised in their annual reports. Additionally, they overestimate the returns of other programs and move innovation resources away from the digital program they mentioned based on their beliefs. The blind spot of this decision is based on overly positive information, which may lead to resource misallocation and results in the inability of enterprises, ultimately inhibiting the innovation efficiency of enterprises and affecting their sustainable development [45]. Thus, we propose our first hypothesis:

H1. 

DDA has a long-term negative effect on corporate innovation efficiency.

As shown in Figure 1, the SIB effect contributes to resource misallocation, primarily through two channels: the executive squandering effect and the inefficient investment effect.

First, the SIB effect may intensify the executive squandering effect. On the one hand, due to the influence of the SIB effect, corporate managers believe that digital transformation projects have already been completed, leading to the premature release of digital resources and the common overconfidence among management. This exacerbates managers’ rent-seeking behavior, where resources are diverted toward unproductive activities. Such diversion manifests as the emergence of opportunistic behaviors among executives, increased management expenses, and a reduction in innovation efficiency [46,47]. On the other hand, the SIB effect may mislead companies to overestimate executive performance, resulting in inflated manger compensation that depletes company resources. This leads to a misallocation of scarce resources. Thus, we propose our second hypothesis:

H2. 

DDA triggers resource misallocation through the executive squandering effect, thereby reducing innovation efficiency.

Second, the SIB effect may also lead to inefficient investment practices. As corporate managers believe that digital transformation projects have already been completed, by the SIB effect, managers will undoubtedly be overconfident. Meanwhile, management overconfidence may lead to bad performance of the corporations. The released resources that were originally allocated to optimal digital transformation projects may remain idle or be diverted to suboptimal projects, exacerbating the resource curse and contributing to underinvestment [46]. Thus, we propose our third hypothesis:

H3. 

DDA induces underinvestment through inefficient investment effects, resulting in resource misallocation and finally reduced innovation efficiency.

4. Materials and Methods

4.1. Data and Sample Selection

This paper selects data about listed Chinese companies from 2013 to 2021, resulting in 21,816 samples after excluding ST companies and financial companies. Digital word frequency is counted with the RESSET text tool from annual reports, where the digital words used for counting word frequency are collected from CSMAR. Digital transformation input data, digital transformation output data, and control variable data are collected from the CSMAR. We also cross validate control variable data from Wind for data robustness. Moreover, CNRDS provides the innovation data. Due to the high proportion of missed digital transformation data in 2022 and 2023 from CSMAR, this paper deletes the samples in 2022 and 2023 in further analyses.

4.2. Variable Selection and Definition

Dependent variable: Corporate innovation efficiency is the dependent variable of this study. Innovation efficiency is calculated using the number of patent applications per unit of R&D investment.

Independent variable: digital disclosure anomalies (DDAs) is the independent variable of this study, which is a binary variable (0 or 1). When DDA equals 1, it means that the listed company’s digital information disclosure level in a given year is much higher than its actual digital transformation efforts. To measure the DDA of companies, we refer the model construction approach suggested by Huang X. [48] and Lin Wanfa [49]. We calculate DDA by obtaining the residuals from a regression equation that relates digital word frequency to digital actions (Equation (1)).

$$\begin{gathered} \ln \_frequenc{y}_{i,t}={\beta }_0+{\displaystyle \sum }_j{\beta }_i^{1}DigOutpu{t}_{i,t,j}+{\displaystyle \sum }_j{\beta }_i^{2}DigInves{t}_{i,t,j}+{\displaystyle \sum }_j{\beta }_i^{1}DigOutpu{t}_{i,t+1,j} \\ +{\displaystyle \sum }_j{\beta }_i^{2}DigInves{t}_{i,t+1,j}+{\beta }_{1}control{s}_{i,t}+{\beta }_{2}control{s}_{i,t+1}+{\mu }_{i,t}+{\lambda }_{p,t}+{\epsilon }_{i,t} \end{gathered}$$

(1)

where, $\ln \_frequenc{y}_{i,t}$, meaning “what they said”, represents the log of the frequency of words about digital project-related terms in annual reports. $DigInves{t}_{i,t,j}$ and $DigOutpu{t}_{i,t,j}$ represent “how they did”. $DigInves{t}_{i,t,j}$ represents enterprises’ digital investments, including MDADigJob (whether the enterprise has CDO/CTO/CIO or not), Nstinnnbasesaccumulated (whether the enterprise has technological innovation bases), and Innovqual (log of the number of digital innovation recognition projects). $DigOutpu{t}_{i,t,j}$ represents digital output, including Iswonaward (whether the enterprise won the national digital prize or not), DigInnStan (whether the enterprise meets the digital innovation standard), Paperjour (the number of digital innovation papers), and Diginvpatautnum (the number of digital invention patent authorizations). Detailed explanations are shown in

Table 1. Variables and Explanations.

Variable Type	Variable Name	Explanations
Dependent Variables	Inveff	Innovation efficiency (number of patents per R&D investment)
Independent Variables	DDA	Level of DDA (dummy variable, valued at 1 if residual is above the 90th percentile, otherwise 0)
Variables used to construct explanatory variable models	Ln_frequency	Log of frequency of the words related to digital projects in companies’ annual reports
	MDADigJob	Number of management digital roles (number of CIO, CTO, and CDO)
	Nstinnnbasesaccumulated	Construction of scientific and technological innovation bases, valued at 1 if the company has an innovation base, otherwise 0
	Innovqual	Log of the number of digital innovation recognition projects + 1
	iswonaward	Whether the company won the national digital prize or not
	DigInnStan	Digital innovation standard work, valued at 1 if the company meets the digital innovation standard, otherwise 0
	Paperjour	Number of digital innovation papers
	Diginvpatautnum	Number of digital invention patent authorizations
Control Variables	Ret	Annual monthly returns for individual stocks
	Std_Ret	The standard deviation of annual monthly returns for individual stocks
	Cr	Current ratio
	Btm	Market value ratio
	Ln_priope	Log of main business income
	Size	Total assets of the company
	Lev	Asset–liability ratio
	Roa	Return on total assets
	Std_Roa	Standard deviation of return on total assets
	Delta_Roa	Difference in return on total assets
	F_Roa	Return on total assets for the next period

.

The control variables consider the management level and performance of the enterprises, and the explanation of control variables are shown in Table 1. ${\mathsf{\beta }}_{\mathrm{i}}^1$ and ${\mathsf{\beta }}_{\mathrm{i}}^2$ represent variable coefficients of $\mathrm{t}$ period and $\mathrm{t}+1$ period, respectively; $\mathrm{i}$ represents company, $\mathrm{t}$ represents year, $\mathrm{j}$ represents the indicators of digital investment or digital output and $\mathrm{p}$ represents province. ${\mathsf{\mu }}_{\mathrm{i},\mathrm{t}}$ is the industry-year fixed effect, ${\mathsf{\lambda }}_{\mathrm{p},\mathrm{t}}$ is the province-year fixed effect. The model also handles clustered robust standard errors for companies. Considering that digital programs will last a long time and the stock market responds quickly, the regression only controls t + 1 variables. ${\mathsf{\mu }}_{\mathrm{i},\mathrm{t}}{\mathrm{and}\mathsf{\lambda }}_{\mathrm{p},\mathrm{t}}$ represent year-industry fixed effects and year-province fixed effects.

Most importantly, ${\mathsf{\epsilon }}_{\mathrm{i},\mathrm{t}}$, the residual of the regression, represents the inconsistency between “what they said” and “how they did”. Considering corporations may exaggerate their digital transformation programs, this paper marks the top 10% of residual values as 1, representing DDAs; the rest are marked as 0. In other words, if the difference between a company’s words and deeds is in the top 10%, these companies have DDAs.

After determining the DDA indicators for companies, the following models are used to regress corporate innovation efficiency or innovation levels on DDAs:

$$\mathrm{I}nvef{f}_{i,t}={\beta }_0+{\beta }_{1}DD{A}_{i,t}+{\beta }_{2}control{s}_{i,t}+{\mu }_{i,t}+{\lambda }_{p,t}+{\epsilon }_{i,t}$$

(2)

Control variables: Based on the research of Lin Wanfa [34], our control variables contain different dimensions of corporate performance, including corporate operational performance, management performance, and stock performance. Table 1 presents the control variables used in the model for constructing the DDA indicator and control variables used in the baseline regression. It also presents the dependent variables and independent variable in this paper.

5. Empirical Analysis

5.1. Descriptive Statistics

Table 2. Descriptive Statistics.

Variable Type	Variable Name	Sample Size	Mean	Std. Dev	Min.	Max.
Dependent variable	inveff	21,816	0.170	0.650	−10.770	26.570
Independent variable	DDA	21,816	0.100	0.300	0.000	1.000
	Administrative Expense Ratio	21,816	0.060	0.100	−0.110	7.280
Mechanism Analysis Variables	Managers Compensation	21,816	0.000	1.000	−0.940	21.280
	inefficient investment	21,816	0.040	0.090	0.000	7.250
Control Variables	ret	21,816	0.170	0.340	−0.560	1.800
	std_ret	21,816	0.140	0.050	0.050	0.720
	cr	21,816	2.720	1.030	0.620	20.320
	btm	21,816	0.610	0.150	0.170	1.240
	lnpriope	21,816	19.120	0.670	13.240	22.770
	lnsize	21,816	22.210	0.700	20.670	25.760
	lev	21,816	0.400	0.090	0.150	0.850
	roa	21,816	0.050	0.020	−0.090	0.180
	Delta_roa	21,816	−0.010	0.020	−0.130	0.140
	F_roa	21,816	0.040	0.020	−0.090	0.180

presents the descriptive statistics of the main variables. The mean value of the independent variable, DDA, is 0.1 after data processing. For innovation efficiency (inveff), the median is 0.17, with a standard deviation of 0.65. The minimum value is 0, and the maximum value is 26.5, indicating significant variation among enterprises.

5.2. Baseline Regression

Considering the persistence of the SIB effect, the baseline regression includes corporate innovation efficiency for the current period as well as the subsequent two periods, as presented in Equations (2), (3), and (4), respectively. The first, second, and third columns of

Table 3. Regression Results for Innovation Efficiency.

Variables	Innovation Efficiency
	Fixed	F1.Inveff	F2.Inveff
DDA	−0.0344 ***	−0.0291 **	−0.0352 **
	(0.0107)	(0.0118)	(0.0138)
std_ret	0.153 *	0.0948	0.1723
	(0.0864)	(0.0995)	(0.1351)
ret	−0.0104	−0.0107	−0.0241 *
	(0.00982)	(0.0115)	(0.0146)
std_roa	−0.123	−0.0865	−0.0117
	(0.169)	(0.211)	(0.3096)
lnsize	−0.0321	−0.0243	−0.1121 *
	(0.0349)	(0.0447)	(0.0587)
lev	−0.0183	−0.0266	−0.3957
	(0.189)	(0.231)	(0.3033)
cr	0.0102	0.0125	0.00251
	(0.00660)	(0.00986)	(0.0133)
btm	0.107 **	0.0766	0.1077
	(0.0474)	(0.0589)	(0.0664)
roa	−0.0885	−0.0468	−0.5147
	(0.336)	(0.385)	(0.4624)
Delta_roa	−0.166	−0.260	0.0078
	(0.240)	(0.273)	(0.3142)
F_roa	−0.00921	0.0229	−0.1513
	(0.203)	(0.226)	(0.2632)
lnpriope	0.00696	−0.00104	−0.0020
	(0.0137)	(0.0164)	(0.0215)
Industry × Year fixed effects	Yes	Yes	Yes
Provincial × Year fixed effect	Yes	Yes	Yes
Individual effect	Yes	Yes	Yes
Constant	0.660	0.668	2.830 **
	(0.681)	(0.884)	(1.1823)
N	21,294	17,977	15,001
R2	0.798	0.814	0.829

Note: The standard error is shown in brackets, and *, **, and *** represent significance levels of 10%, 5%, and 1%, respectively.

respectively present the regression outcomes for corporate innovation efficiency under the conditions of with fixed effects (Equation (2)), for the next period (Equation (3)), and for the next two period (Equation (4)). As shown in Table 3, the results show that DDAs have a significantly negative effect on corporate innovation efficiency and the effect remains for two periods. Due to the existence of the SIB effect, management will invest more resources to the innovative project which they believe is the most optimal one (but actually is not). Thus, DDA results in a decrease in corporate innovation efficiency, supporting H1.

$$Invef{f}_{i,t+1}={\beta }_0+{\beta }_{1}DD{A}_{i,t}+{\beta }_{2}control{s}_{i,t}+{\mu }_{i,t}+{\lambda }_{p,t}+{\epsilon }_{i,t}$$

(3)

$$Invef{f}_{i,t+2}={\beta }_0+{\beta }_{1}DD{A}_{i,t}+{\beta }_{2}control{s}_{i,t}+{\mu }_{i,t}+{\lambda }_{p,t}+{\epsilon }_{i,t}$$

(4)

5.3. Robustness Test

To ensure the robustness of the regression, we use the machine learning regression method instead of the econometric regression method to re-measure the deviation of “What they said” and “How they did” and then generate new DDAs.

Considering that our data encounter the challenges of high-dimensionality and overfitting, the SVR model based on grid-search and cross-validation is chosen to be the machine learning method. On the one hand, the SVR machine learning model shows better performance dealing with high dimension data by using kernel function. On the other hand, through parameter optimization and cross-validation, SVR can avoid overfitting and perform better. It ensures the robustness of model and the reliability of the residuals.

We use the variables in Table 1 to generate a new DDA and use the new DDA in the baseline regression. For companies above the 90th percentile, the DDA is set to 1, and for those below the 90th percentile, the DDA is set to 0. As shown in

Table 4. Robustness Test for Innovation Efficiency.

Variables	Innovation Efficiency
	Now	F1.Inveff	F2.Inveff
	(1)	(2)	(3)
DDA	−0.0341 ***	−0.0289 **	−0.0344 **
	(0.0107)	(0.0118)	(0.0138)
Control variables	Yes	Yes	Yes
Industry × Year fixed effects	Yes	Yes	Yes
Provincial × Year fixed effects	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes
N	21,294	17,977	15,001
R2	0.784	0.814	0.828

The standard error is shown in brackets, and **, and *** represent significance levels of 5% and 1%, respectively.

, the first, second, and third columns of Table 4, respectively, present the regression outcomes of innovation efficiency for the current period (Equation (2)), the next period (Equation (3)), and the next two periods (Equation (4)). As shown in Table 4, the results remain significant.

5.4. Endogeneity Analysis

This study may have endogeneity issues. When corporate innovation capacity is unsatisfactory, they will also exaggerate their digital projects in the annual report, which will cause a reverse causality issue. To ensure the reliability of the research conclusions, this paper will use the two-stage instrumental variable method, which is shown in Column 1 and Column 2 in

Table 5. Endogeneity Tests of Innovation Efficiency.

	Innovation Efficiency
	2SLS (DDA)	2SLS (Inveff)	Heckman (DDA)	Heckman (Inveff)
First-Stage (IV)	6.8368 ***		4.9744 ***
	(16.73)		(20.24)
Second-Stage (DDA)		−0.0992 ***		−0.0345 ***
		(−6.63)		(−2.60)
AR Test	114.2700 *** [0.0000]
Wald test	83.2600 *** [0.0000]
IMR			−0.0507 (−1.41)
N	21,054	21,054	21,054	21,054

The standard error is shown in brackets, and *** represent significance levels of 1%.

.

Considering the “bellwether effect”, enterprises are influenced by the performance of neighboring leading enterprises, eventually leading to DDAs. Meanwhile, corporate innovation efficiency is not related to the leading corporate DDAs. Thus, this paper selected the average DDA level of the top 20 enterprises with the largest business scale in this province as our instrumental variable. As shown in Table 5, with the AR Test and Wald Test, the instrumental variable used in the model is not a weak instrumental variable. The results of the Second Stage in column 2 show that the coefficient for innovation efficiency is significantly negative, which solves the reverse causality issue.

Meanwhile, it is the enterprise itself that decides to launch digital transformation or not, which may cause sample selection bias. Thus, we employ the Heckman test to ensure the result, which is shown in Columns 3 and 4 in Table 5. In the Heckman test, the first stage will create a new variable, Inverse Mill’s Ratio (IMR), to identify sample selection bias. The coefficient of IMR indicates that the model does not actually suffer from sample selection bias.

5.5. Heterogeneity Analysis

Considering that supervision will influence corporate disclosures in annual reports, this paper also analyzes the impact of DDA on innovation efficiency under conditions of external supervision or internal supervision.

For external supervision, this paper classifies companies into state-owned and non-state-owned enterprises, where state-owned enterprises are associated with stronger external supervision by the State-Owned Assets Supervision and Administration Commission [50]. The results in

Table 6. Heterogeneity Analysis of Ownership.

Variables	Innovation Efficiency
	Non- State	Non- State	Non- State	State- Owned	State- Owned	State- Owned
	Now	F1.Inveff	F2.Inveff	Now	F1.Inveff	F2.Inveff
DDA	−0.0270 **	−0.0205 *	−0.0366 ***	−0.0209	−0.00682	0.0106
	(0.0107)	(0.0114)	−0.0366 ***	(0.0250)	(0.0278)	(0.0316)
std_ret	0.154 *	0.0915	−0.0259	0.141	0.103	0.558 **
	(0.0933)	(0.0995)	(0.139)	(0.178)	(0.197)	(0.249)
ret	−0.0161	−0.0103	−0.00787	−0.0102	−0.0244	−0.0648 **
	(0.0105)	(0.0116)	(0.0149)	(0.0208)	(0.0235)	(0.0287)
std_roa	−0.000246	−0.145	−0.134	−0.0722	0.0467	0.0744
	(0.195)	(0.231)	(0.344)	(0.321)	(0.371)	(0.536)
lnsize	0.0173	−0.0495	−0.0330	0.0653	−0.0373	−0.418 ***
	(0.0395)	(0.0456)	(0.0607)	(0.0669)	(0.0827)	(0.102)
lev	−0.319	−0.140	−0.338	0.714 **	0.834 *	0.215
	(0.219)	(0.248)	(0.337)	(0.354)	(0.427)	(0.533)
cr	−0.00779	0.00753	−0.00634	0.0215	0.0443 **	0.0415
	(0.00617)	(0.00876)	(0.0126)	(0.0140)	(0.0211)	(0.0277)
btm	0.0427	0.0729	0.0248	0.149	0.0708	0.225 *
	(0.0524)	(0.0605)	(0.0696)	(0.0946)	(0.114)	(0.123)
roa	0.768 **	0.349	0.456	−0.804	−1.128	−2.365 ***
	(0.380)	(0.406)	(0.512)	(0.646)	(0.715)	(0.803)
delta_roa	−0.256	−0.0266	−0.108	−0.440	−0.486	0.244
	(0.275)	(0.290)	(0.341)	(0.448)	(0.498)	(0.544)
f_roa	−0.0670	−0.0706	−0.0779	0.378	0.0997	−0.292
	(0.221)	(0.230)	(0.273)	(0.409)	(0.436)	(0.498)
lnpriope	−0.00810	−0.00792	−0.00804	0.0323	0.0376	0.0453
	(0.0165)	(0.0186)	(0.0260)	(0.0244)	(0.0279)	(0.0342)
Industry × Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Provincial × Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
N	14,266	11,840	9702	6924	6037	5206
R2	0.853	0.863	0.872	0.776	0.801	0.826

The standard error is shown in brackets, and *, **, and *** represent significance levels of 10%, 5%, and 1%, respectively.

show that the coefficient for non-state-owned enterprises is significantly negative, while that for state-owned corporations is not. This suggests that state-owned enterprises, with stricter external supervision, are less affected by the SIB effect. In contrast, non-state-owned enterprises are more prone to resource misallocation due to weaker external oversight.

For internal supervision, the analysis uses the controlling shareholder’s shareholding proportion. A higher shareholding ratio is associated with weaker internal supervision, as the controlling shareholder possesses complete authority in decision-making [51]. Corporations are grouped into two categories (0 or 1) based on the median shareholding ratio of the largest shareholder. Companies with a shareholding ratio above the 50th percentile are marked as 1, while those below the 50th percentile are marked as 0. The results in

Table 7. Heterogeneity Analysis of Shareholding Proportion of The Controlling Shareholder.

Variables	Innovation Efficiency
	<50%	<50%	<50%	>50%	>50%	>50%
	Now	F1.Inveff	F2.Inveff	Now	F1.Inveff	F2.Inveff
DDA	−0.00772	−0.00912	−0.0174	−0.0396 **	−0.0350 *	−0.0415 *
	(0.0118)	(0.0124)	(0.0130)	(0.0177)	(0.0193)	(0.0243)
std_ret	0.123	0.101	0.0248	0.195	0.0592	0.491 **
	(0.101)	(0.106)	(0.133)	(0.136)	(0.149)	(0.212)
ret	−0.00929	−0.00684	−0.00601	−0.0169	−0.0207	−0.0493 **
	(0.0113)	(0.0121)	(0.0141)	(0.0161)	(0.0181)	(0.0239)
std_roa	−0.274	−0.206	−0.111	0.0853	0.0299	0.0961
	(0.231)	(0.262)	(0.350)	(0.254)	(0.296)	(0.446)
lnsize	0.0225	−0.0293	−0.0503	0.0378	−0.0402	−0.259 ***
	(0.0431)	(0.0494)	(0.0604)	(0.0537)	(0.0637)	(0.0862)
lev	−0.372	−0.456 *	−0.461	0.426	0.784 **	−0.00442
	(0.227)	(0.253)	(0.303)	(0.298)	(0.348)	(0.493)
cr	−0.00145	−0.000123	−0.00319	0.0000891	0.0480 ***	0.0276
	(0.00655)	(0.00907)	(0.0114)	(0.0105)	(0.0161)	(0.0257)
btm	0.116 **	0.146 **	0.127 *	0.0368	−0.0339	0.0968
	(0.0569)	(0.0636)	(0.0653)	(0.0746)	(0.0897)	(0.108)
roa	0.150	−0.157	−0.248	0.392	−0.623	−1.275 *
	(0.419)	(0.441)	(0.499)	(0.513)	(0.552)	(0.691)
delta_roa	0.101	0.261	0.387	−0.635 *	−0.418	0.00627
	(0.305)	(0.320)	(0.342)	(0.357)	(0.379)	(0.455)
f_roa	−0.189	−0.327	−0.363	0.343	0.0758	−0.149
	(0.245)	(0.252)	(0.266)	(0.316)	(0.328)	(0.414)
lnpriope	−0.0164	−0.0153	−0.0189	0.0159	0.0399 *	0.0480
	(0.0178)	(0.0193)	(0.0238)	(0.0204)	(0.0229)	(0.0311)
control variable	Yes	Yes	Yes	Yes	Yes	Yes
Industry × Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Provincial × Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
N	10,587	9203	7934	10,578	8591	6869
R2	0.831	0.861	0.885	0.734	0.784	0.782

The standard error is shown in brackets, and *, **, and *** represent significance levels of 10%, 5%, and 1%, respectively.

indicate that, for companies with a high shareholding ratio of the controlling shareholder, the negative effect of DDAs remains significant and persists for two periods, exacerbating resource misallocation due to weak internal supervision. In contrast, for companies with a low shareholding ratio of the controlling shareholder, the negative effect of DDA on innovation efficiency is not significant.

6. Further Analysis

This section examines the executive squandering effect and inefficient investment effect caused by DDA and their implications for corporate innovation efficiency. First, to test managerial rent-seeking behavior and compensation misallocation, this study uses the administrative expense ratio and standardized executive compensation levels as dependent variables. Additionally, to evaluate the inefficient investment effect of management, investment efficiency is measured using the absolute value of residuals derived from the Richardson [52] investment model. For negative residuals, larger absolute values indicate stronger underinvestment. The absolute values of residuals from negative residual samples are used as dependent variables.

As shown in

Table 8. Mechanism Analysis of the Manager Squandering Effect and Inefficient Investment Effect.

Variables	Managers Squandering Effect		Inefficient Investment Effect
	Administrative Expense Ratio	Manager Compensation	Underinvestment
DDA	0.00799 ***	0.0537 **	0.00422 ***
	(0.00223)	(0.0218)	(0.00130)
std_ret	0.115 ***	−0.278	0.152 ***
	(0.0241)	(0.236)	(0.0144)
ret	−0.00905 ***	0.0979 ***	−0.00895 ***
	(0.00269)	(0.0263)	(0.00155)
std_roa	−0.0628 **	0.463	−0.0189
	(0.0304)	(0.298)	(0.0184)
lnsize	−0.0126 *	0.553 ***	0.0172 ***
	(0.00671)	(0.0658)	(0.00413)
Lev	−0.131 ***	−1.746 ***	0.00422
	(0.0369)	(0.361)	(0.0222)
cr	−0.00116	0.0279 **	0.000157
	(0.00135)	(0.0133)	(0.000758)
btm	−0.00706	0.129	−0.0557 ***
	(0.0115)	(0.113)	(0.00681)
roa	0.331 ***	−0.344	0.150 ***
	(0.0715)	(0.700)	(0.0426)
delta_roa	−0.186 ***	−0.482	−0.0709 **
	(0.0556)	(0.544)	(0.0336)
f_roa	0.0528	−2.185 ***	0.0797 ***
	(0.0497)	(0.487)	(0.0296)
lnpriope	−0.0193 ***	0.277 ***	−0.0138 ***
	(0.00336)	(0.0329)	(0.00203)
Industry × Year fixed effect	Yes	Yes	Yes
Provincial × Year fixed effect	Yes	Yes	Yes
N	21,816	21,816	13,685
R2	0.324	0.094	0.358

The standard error is shown in brackets, and *, **, and *** represent significance levels of 10%, 5%, and 1%, respectively.

, DDAs significantly increase the administrative expense ratio and executive compensation, indicating that DDA fosters managerial rent-seeking behavior and inflated executive compensation. When corporations mistakenly believe that they had already done what they said in the annual report, the managers’ compensation will be wrongly increased by the inflated performance of digital programs, and the rent-seeking behavior will expand among management because of idle resources. These findings suggest that DDAs induce the manager squandering effect, wherein limited and valuable corporate resources are misallocated to satisfy managerial interests rather than organizational goals. The misallocation not only compromises resource utilization but also reduces innovation efficiency by redirecting resources away from value-generating activities, thereby supporting H2.

Moreover, DDAs significantly exacerbate underinvestment, as digital transformation projects may be mistakenly perceived as complete. This misperception leaves company resources idle or diverted to suboptimal digital activities, leading to resource misallocation. Furthermore, resource misallocation reduces the budget and productive investments allocated to the organization’s innovation activities, affecting corporate innovation efficiency. Specifically, resource allocation may lead to underinvestment in innovation projects that should be the key factor of a company’s sustainable growth and competitiveness [53]. Therefore, by exacerbating underinvestment of these critical innovation projects, DDAs indirectly decrease corporations’ innovation efficiency. Such misallocation diminishes the organization’s innovation activities budget and productive investments, ultimately impairing corporations’ innovation efficiency and supporting H3.

These results emphasize the need for companies undergoing digital transformation to implement robust monitoring mechanisms and strategic resource allocation frameworks. By addressing managerial rent-seeking behavior and ensuring accurate assessment of digital transformation progress, corporations may avoid resource squandering and underinvestment. Doing so may optimize resource use and enhance innovation efficiency, which is critical for sustaining a competitive advantage in a digitally-driven market environment.

7. Conclusions

This paper uses data from 2013 to 2021 to measure DDAs and analyzes the relationship between DDAs and innovation efficiency. It finds that, due to the SIB theory, companies tend to think they have accomplished what they disclosed in their annual reports. This “exaggeration” result leads to overconfident management. It also results in resource misallocation through two main channels, the manager squandering effect and the inefficient investment effect, which ultimately impairs corporate innovation efficiency. The manager squandering effect occurs when DDAs lead to a premature release of digital transformation resources, fostering rent-seeking behaviors among overconfident executives. The inefficient investment effect occurs when DDA fosters underinvestment as overconfident managers mistakenly perceive digital projects as complete by SIB theory, leaving resources idle or allocating them to inefficient projects. In our heterogeneity analysis, DDA shows a negative effect on non-stated-owned enterprises and low shareholding proportion of the controlling shareholder enterprises. Thus, DDA is detrimental to corporate innovation development and will lay a ticking time bomb that undermines innovation capabilities and sustainable development.

Based on our research, we have four suggestions for different parts of the financial market to meet the sustainable development of digital market. For investors, we suggest improving their ability to identify and analyze corporate disclosures, especially those with a cross-period nature. For market supervisors, we suggest adopting a “memorial supervision” method, which means continuously tracking and recording the disclosure of corporate digital project information over the years, to ensure that companies do not yield DDAs and to persistently safeguard the market’s sustainability. For corporations, we suggest establishing robust internal governance structures to monitor and evaluate digital transformation progress. This structure should include an independent audit department responsible for reviewing the consistency between digital transformation plans and actual investments. Such measures can help mitigate the risks of DDAs and reduce the unexpected influence of the largest shareholder. For analysts, paying more attention on the inconsistency between “what they said” and “how they did” will provide deeper insights into the corporate performance and figure out uncover potential risks related to digital transformation.

Future studies could explore deeper into refining the measurement of DDAs, considering the evolving nature of digital transformation and disclosure practices. Additionally, exploring the relationship between DDAs and broader organizational behaviors, such as corporate governance or cultural factors, could yield valuable insights. At the meantime, this study has certain limitations. First, the generalizability of findings may be constrained by focusing only on listed companies, potentially overlooking private firms or SMEs where DDA might manifest differently. Second, while this paper identifies key mechanisms such as the manager squandering effect and inefficient investment effect, there may be other underlying mechanisms through which DDA impacts corporate innovation efficiency that remain unexplored. Third, due to significant data omissions related to innovation and digital transformations from 2022 to 2024, this study excludes these years from the analysis, making the data less timely. Addressing these gaps could significantly enrich the understanding of DDA and its broader implications.