The effect of a firm’s zombie status on its innovation in research and development (R&D)-active small and medium-sized enterprises (SMEs): focusing on the moderating effect of zombie congestion

The previous research on zombie firms has primarily focused on four areas: (1) the adverse effects of zombie firms on their industries, (2) the negative impact of zombie firms on the performance of non-zombie firms, (3) the negative effect of government intervention on zombie firms, and (4) the recovery of zombie firms.

First, prior research has demonstrated that zombie firms negatively affect their respective industries. Caballero et al. (2008) defined zombie firms as entities that received subsidized credit from banks. They found that industries with a higher concentration of such entities experienced lower levels of capital investment, job creation, and productivity in Japan. McGowan et al. (2018) identified zombies based on the interest-coverage ratio and confirmed that their presence in an industry had a detrimental effect on productivity in OECD countries. This impact manifests in two ways: zombie firms themselves have low productivity levels, and the market congestion caused by such firms can crowd out non-zombie investments and hinder market expansion. Geng et al. (2021) found that zombie firms hinder industrial upgrading in China by 0.85% per 1% increase in zombie firm assets, owing to resource mismatch and innovation suppression. European Commission (2018) also found that zombie firms hinder the growth of other companies, particularly young ones in Europe.

Second, additional studies have shown that zombie firms negatively affect the financialization and innovation of non-zombie firms. Wu & Pan (2023) studied Chinese listed firms and found that zombie firms increase the financialization of non-zombie firms, exacerbating financing constraints. This effect is more pronounced among non-state enterprises, regions with low financial development, and before 2015 supply-side reforms. Qiao et al. (2022) argued that the prevalence of zombie firms reduces the patent output of healthy firms and jeopardizes innovation. Zombie firms distort the credit resources of healthy firms, especially those that are state-owned or in high-tech industries, thereby reducing innovation output. Yu et al. (2023) also confirmed that zombie firms significantly inhibit normal enterprises, especially those with high innovation intensity, through competition distortion and financing constraints.

Third, there have been studies examining the negative impact of government intervention on zombie firms. One study by Qiao & Fei (2022) analyzed data from Chinese industrial firms to investigate the effects of government subsidies on enterprise operating efficiency. The study found that while subsidies generally improve operating efficiency, moral hazard reduces their effectiveness in boosting efficiency and profits in zombie firms. Similarly, Chang et al. (2021) revealed that government intervention could increase the risk of zombie formation, which in turn can cause economic harm. They recommend that market forces should play a greater role in supply-side structural reforms. Another study by Liu et al. (2019) found that Chinese zombie firms that received government subsidies had lower capacity utilization and that subsidies distorted investment behavior, leading to poor performance. The study also showed that the adverse effect of government subsidies on the ability of zombie firms to utilize their capacity is particularly significant in the case of state-owned zombie firms, those with low government intervention, and those with inadequate financial reporting. The findings suggest that subsidies and fiscal policies specific to zombie firms should be used carefully.

Last, there is research focusing on the potential for recovery of zombie firms despite the prevailing negative perspective on them. According to Goto & Wilbur (2019), Japan’s zombie firms are financially weak firms sustained by discounted interest rates and evergreen lending, and many exist SME. Although these firms can hinder industry efficiency, many of them can revive, making elimination inappropriate. Nurmi et al. (2020) support this argument, suggesting that zombie firms in Finland are not always truly distressed and can recover. While subsidies increase the likelihood of firms surviving, they do not necessarily aid in the recovery of zombie firms. Baek et al. (2021) also conducted research on zombie firms in Korea. They found that if the zombie firms are young, then government R&D funding and the presence of a research department within the firms can have a positive impact on their revival as normal firms.

The existence of zombie firms has been found to have several negative effects on their respective industries, the financialization and innovation of non-zombie firms, and government intervention. In Korea, policymakers have responded to this issue by attempting to limit funding for R&D activities for zombie firms due to their detrimental effects on the macroeconomy. However, it remains unclear whether the negative perceptions of zombie firms can be applied equally to those who engage in R&D activities.

In order to gain a deeper understanding of the relationship between zombie firms and innovation, it is important to consider the possibility of heterogeneity between zombie firms with R&D activities and those without. In Korea, a quite small proportion¹ of firms engage in R&D activity, suggesting that there may be significant differences in behavioral patterns between zombie firms with R&D activities and those without. Moreover, the stylized fact on zombie firms has largely been based on studies targeting relatively large companies, such as listed firms (Caballero et al., 2008; Liu et al., 2019; McGowan et al., 2018; Schaozhen et al., 2019), and as such, the behavior of zombies in SMEs may differ.

Research has shown that SMEs may have a higher potential for recovery from zombie status. For instance, Goto & Wilbur (2019) examined zombie firms in SMEs and found that these firms can revive over time, indicating that uniformly promoting their elimination from the economy may be inappropriate. Similarly, Nurmi et al. (2020) reported that two-thirds of firms with zombie status in Finland became healthy again between 1999 and 2017. They identified cyclical factors, rather than a secular trend, as the main cause of the increase of zombie firms in Finland over 15 years.

Several studies have investigated the relationship between a firm’s financial condition and its decision to invest in R&D. Some have found a weak or little impact of financial constraints on a firm’s R&D investment and innovation output. For instance, Alfranseder & Dzhamalova (2014) reported that financially constrained firms invested more in R&D during the 1998 financial crisis than financially non-constrained ones. Similarly, Lahr & Mina (2013) found that financial constraints had little effect on a firm’s R&D investment decision. Moreover, Bontempi (2016) and Alfranseder & Dzhamalova (2014) proposed a plausible mechanism underlying a firm’s R&D investment behavior. Specifically, a firm’s decision to invest in R&D is likely to be influenced more by uncertainty about future market prospects than by its own financial condition. In other words, external demand-side shocks may have a more significant impact on a firm’s R&D investment decision than supply-side shocks within the firm. In the context of Korea, the unique characteristics of the country’s strong R&D policies may contribute to this uncertainty, significantly impacting firms’ own R&D investments. Traditionally, Korea encouraged private R&D investment through government-led initiatives (Kim et al., 2014; Yoon, 2014), primarily using government R&D budgets and policy measures. Many SMEs, lacking sufficient internal R&D budgets, tend to heavily rely on substantial government R&D funding for their technological development. Additionally, firms often view the future technology areas identified by the government as more promising, making them more inclined to invest in these areas.

Other research has shown that firms tend to invest persistently in R&D, a behavior known as R&D smoothing (Brown & Petersen, 2011; Himmelberg & Petersen, 1994). Firms recognize that it takes a relatively long time to translate R&D investment into economic profit. Stopping or reducing R&D investment would entail an additional adjustment cost in restructuring the R&D human resource or facility (Bernstein & Nadiri, 1989; Hall, 2002).

Upon considering all factors mentioned above, including the negative impact of zombie congestion on industry performance, the heterogeneity of zombie firms, the higher potential for recovery among SME, the weak relationship between a firm’s financial condition and its decision to invest in R&D and the tendency of firms to invest persistently in R&D, this study proposes that R&D-active firms are unlikely to decrease their R&D investment despite financial difficulties. Instead, they are more likely to continue investing in R&D to escape the zombie status and secure new competitiveness. Consequently, this study will hypothesize the following:

With regards to the relationship between a firm’s zombie status and its project output, Kam & Jung (2018) discovered that there was no significant difference in the probability of technological success between zombie and non-zombie firms. Their findings were based on a sample of Korean SMEs (7,757) that undertook government-sponsored R&D projects supported by the Ministry of SMEs and Start-ups in Korea. In light of this, this study proposes the following hypotheses:

Our firm-level data include 1,461 SMEs that received more than one government grant before the end of 2015 from the KEIT, the largest funding agency for industrial R&D. The firms participated as a principal research organization for their government-sponsored R&D projects. Each firm’s financial information was also acquired through Korean Enterprise Data, a credit reporting company. The data covers financial information from 2005 through 2015 for each firm, which spans 11 years. Considering that a small portion of SMEs perform R&D activity and government funding agencies choose R&D beneficiary firms on a competitive basis (Ma et al., 2022), our data represents well R&D-active SMEs and is suitable for our research objective to find innovation behavior in R&D-active zombie firms.

In Fig. 1, we observe a considerable gap between firms with and without a zombie status. Over the entire period, firms with a zombie status have shown lower R&D investment. Remarkably, firms with a zombie status seem to have suffered more from the global financial crisis. In 2009, their R&D investments showed a considerable decline, particularly in firms with a zombie status. All these, however, need cautious interpretation because these are simple results that do not control for each firm’s dynamics and heterogeneity. Instead, this study will provide a more reliable estimation through the dynamic panel data analysis considering the within-firm effect (i.e., the effect of a firm’s zombie status on its R&D investment occurring within the firm over the years).

Fig. 1. Time trend of major variables. R&D, research and development.

Download Original Figure

The mean of a firm’s zombie status is 0.11 in Table 2, implying that 11% of all firm-year observations fall into the category of zombie status. This is consistent with the findings of the Bank of Korea (2014), which reported that zombie firms accounted for about 10% to 15% of all firms. In this study, the firm’s zombie status is a dummy variable defined as one if its annual interest-coverage ratio remains below one consecutively over the past three years (i.e., persistent interest-coverage ratio <1) and otherwise 0. Fig. 1 shows that the annual percentage of a firm’s zombie status has risen rapidly since 2012. In 2015, the percentage reached 15.9%, which ignited deep concerns about zombie firms in Korea.

To identify industries with relatively lower zombie shares, we examined eight technological fields, including Machine, Electronics, Information Technology, Chemical, Biomedical, Energy, Knowledge Service, and Ceramics, using our data set. Then, we compared the zombie shares of each technological field to the average zombie share across all fields. Based on this comparison, industries that demonstrated a below-average zombie share were categorized as possessing a low zombie share. Thus, among the eight technological fields, the Machine, Information Technology, and Knowledge fields were identified as industries with lower zombie shares.⁴

It is also important to mention that the percentage of missing data in the R/C variable is up to about 70%.⁵ This matches findings by Hall & Oriani (2006), who report between 65% and 88% missing values for the R/C variable. Missing R&D data is known to come from one of two reasons: first, a firm’s information service provider could not collect and supply the firm’s financial information, and second, R&D information is originally blank on the firm’s financial statements by intentional choice (Koh & Reeb, 2015).

Either case poses potential selection bias in the estimation. For this reason, we explored the randomness of the omission in our data according to a firm’s zombie status or sales. First, the missing percentage of R&D data in zombie firms is not much different from that of non-zombie firms.⁶ Second, the nullity correlation between R/C and the sales variable is found to be less than 0.2, which also confirms that the missingness of R&D data does not depend on the sales volume. Hence, we presume that the missingness of R&D data would be quite random, which will not severely cause sample selection bias in our estimation.

To test the relationship between a firm’s zombie status and project output, this study collected data on government-sponsored R&D projects from KEIT. The project-level data includes 1,066 projects that were implemented by private firms, not by other types of research organizations (e.g., universities, public research institutes, etc.), and were fully completed before the end of 2015. Each firm’s financial information (2012–2014) is also merged.

This study uses two dependent variables for technological and commercial success: patent application and commercialization. All these are dichotomous indicators because the dataset does not contain detailed information on the number of patent applications. In Table 3, the mean of patent applications (Patent) is 0.65, which indicates that 65% of projects in our data reported that they applied for a patent produced from a government-sponsored R&D project. The mean of commercialization (Comm) is 0.56, lower than the mean of patent applications.

In Table 3, 16% of all principal firms in the projects belong to zombie status. The mean government fund is $1.514M, with a standard deviation of 1.467. The average project’s total duration is 3.06 years, with a standard deviation of 1.54. Within each project, there are an average of 3.21 participating organizations besides the principal firm, and the average TRL upon the project completion is 6.7 (i.e., somewhere between the prototype stage and production stage).

In Fig. 2, we observe the simple mean difference in the two dependent variables according to a firm’s zombie status. A firm with a zombie status produces higher patent applications and lower commercialization than one without a zombie status. Contrary to the ordinary notion of zombie firms, firms with a zombie status in government-sponsored R&D projects are more aggressive in securing technological competitiveness. We might conjecture that a firm probably became a zombie status not because of losing competitiveness but because of persistently innovative investment. In the following section, we will provide more reliable results through our model that controls for other variables affecting patent and commercialization.

Fig. 2. The mean of patent (1/0) and commercialization (1/0) according to a firm’s zombie status.

Download Original Figure

The strongest correlation among the exploratory variables is 0.49, and it lies between the government fund and project duration indicators. This suggests that there is no strong multicollinearity in the exploratory variables. A detailed description of the variables is shown in Table 4 below.

A two-step system GMM methodology is applied to test our hypotheses 1a and 1b. Both estimations (specifications 1 and 2) in Table 5 meet the requirement for validating the estimation results. The Hansen (1982)J test does not reject the validity of instrumental variables (i.e., the validity of over-identifying restriction) at the 5% statistical significance level. AR test also confirms no second-order autocorrelation at the 5% statistical significance level.

The variable of the most interest in this study is a firm’s zombie status. The coefficient of zombie status (ϕ₁) is revealed to be significantly positive in both specifications (specifications 1 and 2), which might indicate that a firm’s zombie status has a positive association with its R&D investment. The coefficient (ϕ₁) is estimated to be 5,158.9 in specification 1, which means that a firm with a zombie status would invest $5,158.9 more than it would invest if it were a non-zombie status. This study finds evidence to support our hypothesis 1a that the zombie status of SMEs will not reduce their R&D investment.

This study also found a significantly positive coefficient (ϕ₂) in support of hypothesis Hypothesis 1b. It indicates that the effect of a firm’s zombie status on its R&D investment is more pronounced in industries with a lower share of zombie firms. Specifically, firms with zombie status in industries with lower congestion are likely to invest more in R&D activities than those with higher congestion. This highlights the potential impact of zombie congestion on a firm’s investment behavior. This finding is quite similar to previous research by McGowan et al. (2018), which suggests that the market congestion caused by zombie firms can crowd out non-zombie investments and hinder market expansion. This study also suggests that even firms with zombie status are influenced by the level of zombie congestion in their respective industries.

The coefficient (λ₁) of the lagged dependent variable is estimated to be negative, which is consistent with the findings of Cincera et al. (2016). It is natural that the speed of R&D stock accumulation (R/C value) follows a decreasing trend. The value (R/C) in a firm goes up only in cases where the firm makes R&D investments that are more than 1.4 times⁷ that of the previous period. Regarding the change in log sales (△y_i,t, △y_i,t–1), all coefficients (λ₂ and λ₃) are revealed to be significantly positive. Given that the variables are indicative of either future expectations or investment opportunities, this study can interpret that a firm’s positive outlook and opportunity would speed up its R&D stock accumulation.

The error-correction term (λ₄) is found to be significantly negative, and this term captures the speed of adjustment from short-run dynamics into long-run equilibrium. The magnitude of the coefficients is estimated between –1 and 0, which shows that the model is converging to long-run equilibrium and the estimation is stable. The coefficient is estimated as –0.199 in specification 1, implying that 19.9% of disequilibrium between short-run and long-run equilibrium will be corrected before the next period.

The coefficient (λ₅) of the log sales (y_i,t–2) is found to be significantly negative. This term captures returns to scale. The negative coefficient can be interpreted as that firms, on average, exhibit decreasing returns to sales. Holding all other variables constant, the firm with a larger scale of sales would experience less increase in its R&D stock. It suggests that a firm with more sales would expect less return on its R&D stock increase. Thus, the firm will invest less in R&D than other firms with lower sales. This finding is also consistent with Kim(2018), which found diminishing returns to R&D investment in Korean manufacturing SMEs.

As a robust check of GMM estimation, the study implemented a fixed effect and pooled ordinary least squares (OLS) regression results. Table 5 shows that the coefficient of zombie status (ϕ) is still found to be significantly positive in all specifications. In addition, the coefficient (λ₁) of the lagged dependent variable in GMM estimation lies between the lower-bound estimate of fixed-effect regression and the upper-bound estimate of pooled OLS regression.⁸ This confirms the validity of our GMM estimation results. We also tested other specifications by adopting the revised indicator⁹ of a firm’s zombie status. It reveals an insignificant effect on a firm’s R&D investment, which confirms that the finding still accepts our hypothesis. Furthermore, we tried to control the effect of the global financial crisis¹⁰ in our model. This does not make any difference in the effect of a firm’s zombie status on its R&D investment.

As an additional test for the validity of our results, we implemented GMM estimation, as shown in Table 6, for three subgroups: firms in industries with a lower proportion of zombie firms, firms in industries with a higher proportion of zombie firms, and firms that have experienced zombie status ever before. Consistent with the moderating effect of industry zombie share on R&D investment, we found that zombie status in an industry with a lower zombie share is strongly associated with R&D investment. The coefficient of zombie status (ϕ₁) is 156,423.6 in the subgroup of the industry with a lower zombie share, while the coefficient is only 5,328.1 in the industry with a higher zombie share. Furthermore, when we focused only on firms that have previously experienced zombie status more than once, we obtained quite interesting results that differed from those of the overall sample. The coefficients () turn into all insignificant, indicating that neither the speed of R&D stock accumulation nor the change in sales had an impact on the firm’s R&D investment. We conjecture that a different behavioral mechanism may exist for zombie firms that have engaged in R&D activity persistently. These firms may be more likely to actively pursue R&D investment to overcome their zombie status rather than relying on previous patterns of R&D investment or sales performance.

This study applied logistic regression to test the relationship between a firm’s zombie status and its technological and commercial output. Table 7 shows the results of zombie status on technological and commercial output, respectively.

The zombie status (β₁) of SMEs is revealed to be a significant positive predictor of patent applications in specification 1. Specifically, if an SME with a zombie status implements a government-sponsored R&D project, the likelihood of producing a patent application is 1.55 times higher than that of a non-zombie firm. We confirm that an SME with the zombie status taking R&D activity persistently is more likely to achieve technological success. However, specification 3 shows that the effect of zombie status on the commercialization probability is insignificant. Both findings support our hypothesis 2a that the zombie status of SMEs will not have a negative association with their innovation output from government-sponsored R&D projects. Nevertheless, it should be noted that an SME with zombie status persistently engaging in R&D activity might face a limitation in achieving commercial success from the technology, as suggested by the negative coefficient (β₁), even though it is not statistically significant.

With regards to the moderation effect, the coefficient (β₂) is revealed as insignificant in all specifications (2 and 4). While the moderating effect of industry zombie share on the zombie firm’s R&D investment is evident, the moderating effect disappears in the relationship between a zombie status and innovation output from government R&D funding. Regardless of whether firms belong to industries with a lower or higher zombie share, there may be no difference in the performance made by firms to produce innovation output from government R&D funding. Therefore, we reject hypothesis 2b, saying that the effect of a firm’s zombie status on its innovation output will be stronger in industries with a lower share of zombie firms.

The estimated effect of the magnitude of government funding (β₃) on the probability of producing a patent application from an R&D project is significantly positive, but on the probability of fostering commercialization, it is close to 0. These results are consistent with the findings of Park (2014), which were drawn using 1,929 industrial government-sponsored R&D projects (2006–2010) in Korea. In contrast, the duration of the R&D project (β₄) is a strong and significant predictor of commercialization but has no effect on patent applications.

The number of partners involved in each R&D project (β₅) is found to influence both patent applications and commercialization significantly positively. Collaboration seems to matter more than any other factor in producing technological and commercial output. These results are consistent with previous findings in Park (2014), and Ma & Dwyer (2020) despite different measures of outcomes.

In addition, we need to mention the reverse causality of technological output and a firm’s lack of finance. For example, suppose a firm evaluates that the technology becomes more promising during its development. In that case, it invests more in technological development (i.e., higher technological output), and accordingly, the firm might face a more serious lack of finance (i.e., zombie status). This would lead to biased estimation. However, we presume that severe reverse causality is not likely to happen in our estimation because there is some time lag between a firm’s zombie status and its project output. As explained in Table 4, the zombie indicator captures a firm’s financial condition from 2012 to 2014, while the firm’s project output is measured on a cumulative basis by the end of 2015. Nevertheless, this study did not strictly control reverse causality using instrumental variables, which is the limitation of this study.

As an additional test, we first implement the same logistic regression for each subgroup of lower and higher zombie congestion industries, respectively. The lower zombie congestion industry includes the Machine, Information Technology, and Knowledge fields, among all eight technological fields. From the estimation results in Table 8, we found that the positive association between a firm’s zombie status and the probability of producing a patent application still holds true, particularly in the lower zombie congestion industry, while there is no relationship in the higher zombie congestion industry. Regarding commercialization, there seems to be no association between a firm’s zombie status and the probability of commercialization in both groups. All these results still do not reject our hypothesis 2a.

Second, we implement the same logistic regression for each subgroup of zombies and non-zombies, respectively. In Table 9, the effect of government funds on the probability of patent application is found to be significant in both zombie firm and non-zombie firm groups. However, the magnitude of the effect is much stronger¹¹ in the zombie firm group than in the non-zombie firm group. It confirms that zombie firms better utilize government funds to produce technological output. Regarding commercialization, the effect of government funds is revealed to be insignificant. Remarkably, in all estimations, the zombie firms do not benefit from other project partners, unlike the non-zombie firms. It might also suggest that zombie firms lack networking capabilities to acquire external technological knowledge and business opportunities.

Table 10 summarizes the test results of all hypotheses. Hypotheses 1a, 1b and 2a are accepted, and on the contrary, hypothesis 2b is rejected.