2.1 Theoretical underpinnings

The resource-based view posits that firms with access to unique and valuable resources are better positioned to achieve competitive advantages, enhance productivity, and attain higher survival rates (Barney, Reference Barney1991; Wernerfelt, Reference Wernerfelt1984). This potential can be realized when enterprises have access to appropriate resources, typically obtained through investment. In modern economies, investment projects are often financed through external loans provided by financial institutions. Consequently, firms lacking easy access to external credit are likely to lag behind their competitors. Brush and Cooper (Reference Brush and Cooper2012) attribute the performance disparities between MOEs and WOEs primarily to the latter’s limited access to financial resources. In the absence of sufficient financial capital, WOEs fail to invest adequately in new assets, which are the main conduits of technological progress ultimately restricting their potential to grow. In a developing economy setting, these constraints are often more pronounced, as financial markets are typically less developed, and formal lending institutions exhibit discriminatory practices against female entrepreneurs. WOEs in developing countries struggle to secure external financing due to a combination of limited collateral, higher perceived credit risk, and gender biases (Muravyev et al. Reference Muravyev, Talavera and Schäfer2009). As a result, WOEs often rely on informal sources of finance, which tend to be insufficient for larger, productivity-enhancing investments.

In the context of gender disparities and economic performance, early developments in human capital theory (HCT) (Becker, Reference Becker1962) suggest that differences in education, skills, and experience between male and female entrepreneurs may lead to variations in firm productivity and performance. Differences in human capital are proven important sources of gender inequalities in a developing country setting (Alibhai et al. Reference Alibhai, Buehren, Papineni and Pierotti2017; Essers et al. Reference Essers, Megersa and Sanfilippo2021). Historically, gender biases in acquiring educational and professional skills have disadvantaged women entrepreneurs to build sufficient levels of human capital, potentially limiting the growth potential of WOEs. However, more recent evidence challenges this long-held view of women’s educational disadvantage. Terjesen and Singh (Reference Terjesen and Singh2008) indicate that women often bring unique perspectives and skills, which can enhance decision-making and boost firm performance when placed in leadership positions. Nonetheless, the propositions of HCT alone are insufficient to fully explain the gender-performance gap, suggesting that a more comprehensive approach is needed to address other structural barriers (Islam et al. Reference Islam, Gaddis, Palacios López and Amin2020).

Institutional theory (IT) posits that the gender-performance gap in entrepreneurship is largely influenced by broader social, economic, and institutional constraints. IT underscores the role of formal and informal institutions, including legal frameworks and cultural norms that shape business practices, often causing additional barriers for WOEs (North, Reference North1987). From an IT perspective, discriminatory practices and unfavorable gender norms, particularly in developing countries with entrenched patriarchal systems, lead to unequal treatment of women in financial markets, limiting their access to networks, credit, and business opportunities. In many developing countries, formal institutions such as property laws, inheritance rights, and legal systems may be biased against women, impeding their ability to offer collateral for loans or register businesses in their own names. This institutional bias often perpetuates women’s exclusion from formal financial systems and reinforces gender inequalities in firm performance (Brush et al. Reference Brush, De Bruin and Welter2009).

The adverse effects of institutionalized gender barriers for women entrepreneurs are also documented in Elam and Terjesen (Reference Elam and Terjesen2010). In developing countries, there are societal expectations around women’s roles in the household and the labor market that can lead to time and mobility constraints, which also limit the ability of female entrepreneurs to engage in business networking or expand their firms. Overall, informal institutions’ limitations prevent WOEs from benefiting from economies of scale and accessing larger markets.

Finally, social capital theory (SCT, hereafter) provides valuable insights into the gender-performance nexus by highlighting the importance of networks in achieving business success. Social capital, in this context, refers to the resources embedded within social networks, including trust, reciprocity, and shared norms that facilitate cooperation among individuals and organizations (Coleman, Reference Coleman1988). For WOEs, social capital manifests in various forms, such as access to business groups, industry networks, and relationships with key stakeholders, all of which are essential for acquiring financial resources, market information, and business opportunities (Marlow and Patton, Reference Marlow and Patton2005). However, because many business and financial networks are traditionally male-dominated, WOEs are likely to struggle to tap into these valuable social resources. The lack of social capital impedes the growth of WOEs by restricting access to finance, mentorship, and market information (María Viedma Marti, Reference María Viedma Marti2004; Glaeser et al. Reference Glaeser, Laibson and Sacerdote2002; Nahapiet and Ghoshal, Reference Nahapiet and Ghoshal1998). According to SCT, social capital plays a pivotal role in establishing trust, particularly for small and medium-sized enterprises, which are the typical size of WOEs in developing countries (Deller et al. Reference Deller, Conroy and Markeson2018).

2.2 Mechanisms of the gender-productivity nexus and hypotheses formulation

The previous discussion outlines the theoretical frameworks that explore women’s role in leadership positions and provide explanations about the inferior corporate performance of WOEs relative to MOEs. This section overviews some of the empirical findings of this voluminous literature and formulates the testable hypotheses of the present study.

Bardasi et al. (Reference Bardasi, Sabarwal and Terrell2011) found that WOEs in developing countries tend to be smaller, less productive, and more likely to operate in the informal sector. Similarly, Sabarwal & Terrell (Reference Sabarwal and Terrell2008), focusing on South Asian and Eastern European firms, discovered that WOEs generally exhibit lower sales and employ fewer workers. These disparities are frequently attributed to the disadvantaged treatment WOEs experience, particularly in securing access to finance—an essential conduit for investing in fixed capital assets and innovation. This is also evident in firms across sub-Saharan Africa (Aterido et al. Reference Aterido, Beck and Iacovone2013), and Central Asia (Muravyev et al. Reference Muravyev, Talavera and Schäfer2009), where credit constraints limit WOEs’ ability to expand operations and enhance productivity.

Another stand of literature investigates the resilience and adaptability of WOEs, particularly within a highly competitive business environment. Coleman and Robb (Reference Coleman and Robb2016), as well as Manolova et al. (Reference Manolova, Brush, Edelman and Shaver2012), suggest that, despite structural disadvantages, WOEs exhibit significant resilience by capitalizing on innovation, flexibility, and niche market strategies to maintain their competitive edge. Marlow and McAdam (Reference Marlow and McAdam2013) contend that the observed lower productivity of women entrepreneurs is not a consequence of within WOEs’ internal management deficiencies and (or) misallocation of the entrepreneurial capital, but rather the result of substantial constraints in accessing financial resources and business networks. Fairlie and Robb (Reference Fairlie and Robb2009) found that the lower performance of WOEs in terms of profits, employment, and sales are primarily attributed to disparities in initial capital, industry selection, and differences in business experience, rather than any inherited gender-oriented limitations in business capabilities. When controlling for variables such as firm size, industry, and geographical location, empirical studies reveal that the performance gap between WOEs and MOEs diminishes substantially. This suggests that, when external barriers are mitigated, WOEs can perform as well as MOEs. Footnote ³ Morazzoni and Sy (Reference Morazzoni and Sy2022) show that after guaranteeing equal access to credit to male and female entrepreneurs, this improves aggregate talent in the economy and decreases the misallocation of total capital by 12%.

The key takeaway from the existing evidence is that WOEs consistently demonstrate inferior performance, irrespective of other factors or conditions ("unconditional"), suggesting that this underperformance is more likely driven by external barriers rather than intrinsic gender-related deficiencies. Building on this discussion, we propose the first hypothesis of the paper:

H1: WOEs are expected to have unconditionally lower productivity compared to MOEs.

Among the most significant obstacles preventing firms from reaching their full potential is restricted access to finance. Firms require credit to operate their businesses and invest in new capital assets. Nonetheless, there are gender asymmetries in obtaining finance that are not resulted from worse risk profiles or lower profitability of WOEs. The gender-based heterogeneous treatment for accessing credit exists even in developed economies like the United States. Morazzoni and Sy (Reference Morazzoni and Sy2022) found that US female entrepreneurs are 10% more likely to have their loan applications rejected compared to their male counterparts. In the absence of other societal barriers, this financial penalty for WOEs suggests the existence of gender selection bias, where only more productive WOEs can overcome financial barriers and either launch a new business or expand an existing one.Footnote ⁴ In developing countries, access to finance is a common challenge for businesses, but gender disparities worsen the situation for women entrepreneurs. Cultural norms often lead to discriminatory lending practices, stricter collateral requirements, and smaller loan amounts for WOEs (Aterido et al. Reference Aterido, Beck and Iacovone2013; Muravyev et al. Reference Muravyev, Talavera and Schäfer2009). Overall, the disadvantaged treatment of WOEs in financial markets perpetuats a gender gap in firm productivity. Based on this discussion, we propose the second hypothesis of the paper as follows:

H2: The productivity gap between firms whose bigger obstacle is access to finance is more pronounced among WOEs.

Financial frictions that disadvantage WOEs are expected to be more pronounced in sectors characterized by greater financial dependence, typically capital-intensive industries. This may also explain why WOEs are more frequently found in the informal sector,Footnote ⁵ which tends to have a lesser need for external financing (Islam and Amin, Reference Islam and Amin2023). These financial constraints limit WOEs’ ability to operate in formal sectors such as manufacturing, ICT, and finance, confining them to industries with smaller-scale production. As a result, WOEs tend to have low capitalization, preventing them from investing in productivity-enhancing activities, which in turn contributes to a productivity gap compared to MOEs. The third hypothesis of the paper is formulated as:

H3: As the financial dependence of the industry increases, WOEs are more likely to suffer a productivity penalty relative to MOEs.

The previous hypothesis also serves as a test of the average returns to capital between genders in firm ownership and management. If financial dependence increases and women face harsher conditions in obtaining external credit, one might expect the lower productivity of WOEs to indicate that women entrepreneurs lack sufficient entrepreneurial capacity to use limited financial resources efficiently and stimulate productivity. On the other hand, if the average returns to capital increase, it suggests that WOEs are equally capable of utilizing financial resources in a way that effectively enhances firm productivity.

The lower productivity of WOEs is also linked to their reduced capacity to innovate and adopt new technology. Restricted access to finance often forces WOEs in both developed (Ferrando and Ruggieri, Reference Ferrando and Ruggieri2018) and developing countries (Cirera and Muzi, Reference Cirera and Muzi2020) to abandon innovation projects. Davis et al. (Reference Davis, Haltiwanger and Schuh2019) and Haltiwanger et al. (Reference Haltiwanger, Jarmin and Miranda2013) indicate that firms engaging in innovation typically experience higher sales growth and profitability. Innovation is also expected to have a positive impact on productivity (Hall, Reference Hall2011). A key finding in the literature is that product innovation significantly enhances revenue productivity, with evidence from European firms showing that the introduction of new products increases productivity by 2.7% (Nathan and Rosso, Reference Nathan and Rosso2022). However, these positive effects are not uniform across all firms. They are more pronounced in the services sector and, to a lesser extent, in manufacturing, with firm size also playing a crucial role. Small-sized firms often lack the financial capacity to invest heavily in innovation. This pattern is particularly relevant to WOEs in developing countries, as Cirera and Sabetti (Reference Cirera and Sabetti2019) found that these firms are less likely to introduce new products or processes, contributing to an innovation gap that limits their growth potential (Cirera and Muzi, Reference Cirera and Muzi2020).

Additionally, the managerial style in women-led firms, often characterized by greater risk aversion, poses further challenges to innovation. Risk aversion can deter innovation, which is inherently uncertain and involves unpredictable returns. WOEs may prioritize stability and employee welfare over high-risk, growth-oriented strategies, leading to lower dynamic efficiency and long-term performance. While this approach may provide greater security for employees, it constrains the firm’s ability to pursue rapid, innovation-driven growth. Therefore, the fourth hypothesis of the paper is written as:

H4: Limited access to finance constrains innovation in WOEs, leading to lower productivity compared to MOEs.

4.1 The production function model

We consider the following Cobb–Douglas production function:

(1) \begin{equation} y_{ict}=\alpha _{0} +\mathbf {z}^{\prime}_{ict}\alpha _{l}+\mathbf {x}^{\prime}_{ict}\alpha _{k}+\omega _{it}+u_{ict} \end{equation}

where $y_{ict}$ is the natural logarithm of value-added of firm $i$ , in country $c$ , in year $t$ , $\mathbf {z}^{\prime}_{it}$ is a $1\times J$ vector of variable production inputs, such as labor, and $\mathbf {x}^{\prime}_{it}$ is a $1\times K$ vector of state variables, such as capital, all in logarithmic form. The industry index $j$ is omitted for the ease of the exposition. Parameter $\omega _{ict}$ measures productivity, which is observed for the firm manager but remains unobserved for the econometrician. Finally, $u_{ict}$ represents random noise that captures measurement errors. Parameters $\omega _{ict}$ and $u_{ict}$ differ in that the former is a state variable, impacting firm i’s decision rules, while the latter has no impact on $i$ ’s decisions. Before estimating (1), we need to state that the fundamental issue is the endogeneity bias emerging between the evolution of $\omega _{ict}$ and the selection of inputs $z_{ict}$ . The presence of endogeneity bias implies that the $l_{ict}$ is not chosen independently from $\omega _{ict}$ , which makes OLS to generate an upward biased coefficient for $l_{ict}$ and other variable inputs.

To mitigate endogeneity between $l_{ict}$ and $\omega _{ict}$ in (1) from selection bias, we first assume a benchmark specification in which $\omega _{ict}$ follows an AR(1) process:

(2) \begin{equation} \omega _{ict}=E[\omega _{ict}|\omega _{ict-1}] +\xi _{ict} \end{equation}

where $\xi _{ict}$ is an innovation term in the evolution of productivity capturing factors that cannot be explained in the persistence nature of $\omega _{ict}$ .Footnote ¹² To obtain consistent estimates of (1) for parameters $\alpha _{l}$ and $\alpha _{k}$ , we use the semi-parametric approach of Ackerberg, Caves, and Frazer (Reference Ackerberg, Caves and Frazer2015) (hereafter ACF). ACF is implemented in two stages. The first stage of ACF estimates the following specification: $\theta _{ict}= \alpha _{0}+\alpha _{l}l_{ict}+\alpha _{k}k_{ict}+m^{-1}(l_{ict}, k_{ict}, m_{ict})$ , where $m^{-1}$ is the inverse function of intermediate material inputs. The functional form of $\theta _{ict}$ in the first-stage estimation of ACF is a polynomial equation and seeks to net out random noise and (or) measurement errors related to $\epsilon _{ict}$ . The logic of this approach is that materials $m_{ict}$ can be used as a proxy variable for unobserved productivity shocks. The only condition needed is that the demand for materials is monotonically increasing in $\omega _{ict}$ , the function $m$ can be inverted. Coefficients for $l_{ict}$ and $k_{ict}$ are derived in the second stage of ACF based on the following orthogonality conditions: in $E[k_{ict},\xi _{ict}]=0$ , $E[l_{ict-1},\xi _{ict}]=0$ . Once we recover coefficients for $\alpha _{0}$ , $\alpha _{l}$ , and $\alpha _{k}$ , then productivity is measured as $\hat {\omega }_{ict}=y_{ict}-\hat \alpha _{0}-\hat \alpha _{l}l_{ict}-\hat \alpha _{k}k_{ict}$ . The results of estimating (1) following ACF are shown in Table 5 and their summary statistics in Table 6.

Table 5.

Production function estimates

Note: Bootstrap standard errors in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ . A third degree of polynomial approximation for the first stage is used. This method uses a Generalized Method of Moments (GMM) routine for the second stage. The Wald test refers to the constant returns to scale hypothesis: $\beta + \gamma = 1$ .

Table 6.

Summary statistics for $\hat {\omega }_{ict}$ ( $TFP_{ict}$ )

Note: This table presents the summary statistics of the estimated productivity using ACF approach.

5.1 Econometric specification

The econometric analysis starts by exploring the effects of gender on TFP through access to finance. To implement this task, we first specify the following model:

(3) \begin{equation} TFP_{ict}=\beta _{0}+\beta _{1}Gender_{ict}+\beta _{2}C_{ict}+\beta _{3}( Gender_{ict}\times C_{ict})+ \mathbf {M}^{\prime}_{ict}\mathbf {B}_{4}+\lambda _{c}+\gamma _{j}+\mu _{t}+\epsilon _{ict} \end{equation}

where $TFP_{ict}$ is the empirical counterpart of $\hat {\omega }_{ict}$ ; $Gender_{ict}$ is the main variable of interest, and we use interchangeably $WOEs_{ict}$ and $WMEs_{ict}$ ; $C_{ict}$ includes $FinOB_{ict}$ whether firm $i$ faces financial barriers; $FinOBI_{ict}$ measures the intensity of financial barriers; and $innovation_{ict}$ refers to whether firm $i$ innovates. Again, these channels are used alternatively and represent the channels through which $Gender_{ict}$ affects $TFP_{ict}$ within our framework. We also include a vector of firm-specific characteristics, $\mathbf {M}_{ict}$ . Parameter $\beta _{1}$ measures the gender premium (or penalty) in $TFP_{ict}$ , while $\beta _{2}$ , $\beta _{3}$ and $\mathbf {B}_{4}$ are other parameters of interest to be estimated. Finally, specification (3) includes country( $\lambda _{c}$ ), industry( $\gamma _{j})$ and year ( $\mu _{t}$ ) fixed effects to capture unobserved country and industry idiosyncrasies and time-variant macroeconomic effects.

5.2 Measuring the gender penalty

We start with a parsimonious specification without considering any channel. Results are shown in Table 7. The coefficients of $WOE_{ict}$ and $WME_{ict}$ are statistically negative indicating a productivity penalty for female-led enterprises regardless of whether we consider the gender effect through ownership (i.e., WOEs) or management (i.e., WMEs). The size of the magnitude indicates that $WOE_{ict}$ tends to be 5.3% less productive than $MOEs_{ict}$ , while the size of the penalty is even bigger equal to 10% when we measure gender as $WMEs_{ict}$ in column (3). When we measure the size of the gender penalty on TFP conditional to other firm characteristics in columns (2) and (4), the coefficients of $WOE_{ict}$ and $WME_{ict}$ remain negative with very small changes in the size of the effect. Overall, Table 7 indicates that WOEs are less productive than MOEs and this is true regardless of whether we account for other firm characteristics. Evidence from Table 7 supports H1.Footnote ¹³

Table 7.

The gender penalty in firm TFP: 30 developing countries

Note: This table shows the estimates of the gender ( $WOE_{ict}$ and $WME_{ict}$ penalty in TFP. The latter is estimated using the ACF approach. $\mathbf {M}_{ict}$ include $age_{ict}$ , $human_{ict}$ , $innovation_{ict}$ , and $onwership_{ict}$ . See Tables A1 and B1 in the Appendix for the description and coefficients of these variables. Clustered standard errors at the industry level are shown in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

5.3 Gender and finance

Next, we examine whether the productivity gap is more pronounced among WOEs facing financial obstacles. To do this, we construct a variable using data from the WBES, which asks firms to identify the most significant obstacle affecting their operations.Footnote ¹⁴ Accordingly, we construct a dummy variable, $FinOB_{ict}$ , which takes the value of 1 if firm $i$ identifies access to finance as its biggest obstacle, and 0 otherwise. We also create an alternative variable, $FinOBI_{ict}$ , to capture the intensity of the financial constraint, based on the survey question: “To what extent is access to finance an obstacle to the current operations of this establishment” Responses to this question are ranged from 1 to 5 starting from “no obstacle” to “very severe obstacle.”Footnote ¹⁵ As the value of $FinOBI_{ict}$ increases, the perceived severity of the access to finance obstacle also increases indicating difficulty in accessing external finance. Results from $FinOB_{ict}$ and $FinOBI_{ict}$ , are presented in Tables 8 and 9, respectively.

Table 8.

TFP, gender, and financial obstacles: 30 developing countries

Note: This table shows the estimates of obstacles to accessing finance interacting with gender variables ( $WOE_{ict}$ and $WME_{ict}$ ) on TFP. The latter is estimated using the ACF approach. $\mathbf {M}_{ict}$ includes $age_{ict}$ , $human_{ict}$ , $innovation_{ict}$ , and $ownership_{ict}$ . See Table A1 and B2 in the Appendix for the description and coefficients of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

Table 9.

TFP, gender, and intensity of financial obstacles: 30 developing countries

Note: This table shows the estimates of obstacles to accessing finance using an intensity measure interacting with gender variables ( $WOE_{ict}$ and $WME_{ict}$ ) on TFP. The latter is estimated using the ACF approach. $\mathbf {M}_{ict}$ includes $age_{ict}$ , $human_{ict}$ , $innovation_{ict}$ , and $ownership_{ict}$ . See Table A1 and B3 in the Appendix for the description and coefficients of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

$WOE_{ict}$ and $WME_{ict}$ encounter a productivity penalty that ranges from 2% to 10.4% in Tables 8 and 9. While the presence of financial obstacles $FinOB_{ict}$ in Table 8 does not affect productivity, nonetheless, the intensity of these obstacles, $FinOBI_{ict}$ , in Table 9 leads firms to improve productivity. This suggests that financial pressure induces efficiency gains in the use of available capital. The coefficient of $FinOBI_{ict}$ remains statistically significant even after including $\mathbf {X}_{ict}$ in columns (3) and (6). The coefficients of interaction terms, $WOE_{ict}\times FinOB_{ict}$ and $WOE_{ict}\times FinOBI_{ict}$ are negative in both tables indicating that access-to-finance barriers reduces disproportionately more the productivity of $WOE_{ict}$ and $WME_{ict}$ . Overall, evidence from our baseline specifications in Tables 8 and 9 support hypothesis H2, showing that financial obstacles disproportionately exacerbates productivity challenges in women-led firms, with the most pronounced impact on WOEs.

Next, we investigate whether increased financial dependence in an industry exacerbates the financial challenges faced by WOEs, leading to a further productivity penalty for WOEs compared to MOEs. Following Rajan and Zingales (Reference Rajan and Zingales1998) and Acharya and Xu (Reference Acharya and Xu2017), we first construct a firm-level measure of external financial dependence. This measure captures firm $i$ ’s reliance on external financing, calculated as $firm\_dep_{ict} = \frac {(100 - k5a)}{100}$ . This is the fraction of fixed assets that are not funded from internal cash flows. We use the WBES variable k5a, which reports the percentage of fixed assets financed by internal funds or retained earnings. Based on this calculation, a value of zero indicates no external dependence, as all fixed assets are funded internally. Next, we aggregate this firm-level measure to create an industry-level external financial dependence variable. For each industry-year combination, we calculate the median external dependence of firms within the same Standard Industrial Classification (SIC) code, and then we calculate the time-series median of external financial dependence for each industry across the years in the dataset. Finally, we rank industries by their time-series median external financial dependence in Table 10. We assign percentile ranks across all industries with a higher rank representing greater dependence on external financing.Footnote ¹⁶

Table 10.

TFP, gender, and industry-wide external dependence

Note: This table shows the estimates of external dependence interacting with gender variables ( $WOE_{ict}$ and $WME_{ict}$ ) on TFP. The latter is estimated using the ACF approach. $\mathbf {M}_{ict}$ include $age_{ict}$ , $human_{ict}$ , $innovation_{ict}$ , $onwership_{ict}$ , and $digit_{ict}$ . See Table A1 and B4 in the Appendix for description and coefficients of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

The coefficient of $External\_dependence_{j}$ is consistently positive in all specifications of Table 10 implying that in industries with heavy needs for external finance, TFP tends to be higher. A scenario compatible with the proposition of finance-led growth as external funds offer the required credit for productivity-enhancing investment (Ferrando and Ruggieri, Reference Ferrando and Ruggieri2018). The productivity penalty remains in Table 10 and becomes slightly bigger for $WOE_{ict}$ at the order of nearly 11% in column (6) compared to findings in Tables 8 and 9. However, the coefficient of the interaction term $WOE_{ijct}\times External\_dependence_{j}$ suggests a mitigating effect. Specifically, as industries become more dependent on external financing, the negative productivity impact associated with women managers is diminished. In other words, the relative scarcity of external finance that WOEs face leads to more efficient resource use, or equivalently, to a higher average return on capital, which positively impacts $TFP_{ict}$ . We interpret the higher average revenue product of capital as an evidence that women entrepreneurs do not exhibit any intrinsic deficiency in allocating capital efficiently compared to their male counterparts. Thus, the disproportionately negative effect on WOEs’ TFP due to limited access to capital, as seen in the tables, stems from discriminatory lending practices that restrict the ability of women entrepreneurs to secure the necessary funds for productivity and business growth. These results contribute to the misallocation literature (Hsieh and Klenow, Reference Hsieh and Klenow2009), demonstrating that stricter requirements for accessing finance in women-led enterprises lead to a narrower entrepreneurial pool, which can ultimately reduce aggregate productivity. Overall, our findings in Table 10 don’t support hypothesis H3.

In summary, the results of this section highlight that while both WOEs and WMEs face significant productivity challenges due to discriminatory lending practices that limit access to external credit, the stereotype that women are less capable of managing businesses is unfounded. In fact, in our sample, women are found to perform better than men in industries with greater reliance on external financing. This suggests that women can efficiently run businesses and stimulate productivity, even when faced with severe financial constraints. We find no evidence that women self-select into sectors with low financial dependence, which might otherwise offer a comparative advantage over MOEs. Our evidence points to a gender asymmetry in access to finance that negatively impacts the productivity of women-led firms like other recent female entrepreneurship literature suggests (Morazzoni and Sy, Reference Morazzoni and Sy2022).

5.4 Gender and innovation

In this section, we investigate the sources of $TFP_{ict}$ differentials between WOEs and MOEs, focusing on the innovation channel and its interaction with the firm’s gender status in driving $TFP_{ict}$ . Specifically, we examine how WOEs may face unique challenges in adopting product and process innovations due to constraints in accessing finance (Cirera and Muzi, Reference Cirera and Muzi2020; Ferrando and Ruggieri, Reference Ferrando and Ruggieri2018). Our approach analyzes whether innovation mediates the gender productivity gap, identifying if industries with a stronger reliance on innovation exhibit a more pronounced gender disparity in TFP.

Before showing results from estimating equation (3), we conduct preliminary tests to examine whether a more risk-averse behavior, combined with the anticipation that women will face discrimination in accessing finance, influences their decision to innovate. To capture structural trends at the industry level and avoid underlying selectivity bias at the firm level, we investigate the relationship between the presence of female leadership and innovation at the industry-year level. We regress the percentage of firms that innovate in each industry-year combination on the proportion of $WOE_{jt}$ and $WME_{jt}$ in industry $j$ at year $t$ . The results, presented in Table 11, show that industries with higher proportions of $WOE_{jt}$ and $WME_{jt}$ tend to have lower levels of innovation. Column (1) suggests that a 10% increase in $WOE_{jt}$ in the industry decreases industry innovation by 2.25%; the effect is higher if we consider an increase in the share of $WME_{jt}$ (i.e., 4.8%). This evidence aligns with the notion that female-led firms tend to innovate less. Next, we investigate whether this effect is linked to financial obstacles and whether this negative tendency further contributes to adverse productivity outcomes.

Table 11.

Innovation and female leadership: industry-level evidence

Note: This table shows coefficients from regressing the % of innovating firms in $j$ on the % of $WOE_{jt}$ and $WME_{jt}$ . Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

Turning to results in Table 12 on the effects of gender on TFP through the channels of financial obstacles and innovation, we observe that, as expected, innovation improves TFP. The coefficients range from 0.084 to 0.121 across columns (1) to (8), indicating that innovative firms can be up to 12.1% more productive than non-innovative ones. Regarding the gender effect, the productivity penalty for WOEs remains between 5% and 6%.

Table 12.

TFP, gender, financial obstacles and innovation: 30 developing countries

Note: This table shows the estimates of $innovation$ interacting with gender variables ( $WOE_{ict}$ and $WME_{ict}$ ) on TFP. The latter is estimated using the ACF approach. $\mathbf {M}_{ict}$ include $age_{ict}$ , $human_{ict}$ , and $onwership_{ict}$ . See Tables A1 and B5 in the Appendix for the description and coefficients of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

Table 13.

Second-stage results from 2SRI regression

Note: This table shows the second stage of 2SRI estimates in which $WOE_{ict}$ , $WME_{ict}$ , $FinOB_{ict}$ , and their interactions are considered endogenous. The instrument used is the inverse of $IFD_{jt}\times WBL_{jt}$ . In the first stage, a probit model for each endogenous binary variable is employed and includes the instrument, $M_{ict}$ and industry fixed effects. $\mathbf {M}_{ict}$ includes $age_{ict}$ , $human_{ict}$ , and $onwership_{ict}$ . See Table A1 in the Appendix for the description of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

Table 14.

Second-stage results from 2SLS regression

Note: This table shows the second stage of 2SLS estimates in which $WOE_{ict}$ , $WME_{ict}$ , $FinOBI_{ict}$ , and their interactions are considered endogenous. The instrument used is the inverse of $IFD_{jt}\times WBL_{jt}$ . In the first stage, a probit model for each endogenous binary variable is employed using the residuals as a regressor in the second stage. For non-binary endogenous variables, predicted values (i.e., $\hat {Y}_{FinOBI_{ict}}$ , $\hat {Y}_{WOE_{ict}\times FinOBI_{ict}}$ ) and $\hat {Y}_{WME_{ict}\times FinOBI_{ict}}$ from the first stage are used as regressors in the second stage. Both models include the instrument, $\mathbf {M_{ict}}$ and industry fixed. $\mathbf {M}_{ict}$ include $age_{ict}$ , $human_{ict}$ , and $onwership_{ict}$ . See Table A1 in the Appendix for the description of these variables. Clustered standard errors at the industry level are in parentheses. $^{*}$ $p\lt 0.10$ , $^{**}$ $p\lt 0.05$ , $^{***}$ $p\lt 0.01$ .

Moreover, the interaction effect of $WOE_{ict} \times innovation_{ict}$ becomes significantly negative in column (6) when additional firm controls are included. This suggests that the productivity gains from innovation are smaller for WOEs. Specifically, innovation-led TFP is reduced by 2.5% for WOEs compared to MOEs. The overall effect of $innovation_{ict}$ on $TFP_{ict}$ in column (6), holding other factors constant, shows that $innovation_{ict}$ increases $TFP_{ict}$ by approximately 6.6% for WOEs, compared to 9.1% for MOEs. In other words, the TFP differential from innovation for WOEs, calculated as $TFP_{ict, innovation=1}$ - $TFP_{ict, innovation=0}$ , is 0.066.

In column (7), we also investigate whether financial obstacles are the underlying reason innovation does not mitigate the productivity gap between WOEs and MOEs. To examine this, we include interaction terms between $FinOBI_{ict}$ , $WOEs_{ict}$ , and $innovation_{ict}$ . The interaction term, $FinOBI_{ict} \times innovation_{ict}$ , shows a negative effect (−0.013), confirming that financial barriers hinder the full absorption of innovation-related productivity gains. Similarly, the presence of financial obstacles disproportionately affects $TFP_{ict}$ in WOEs (−0.018), a finding also evident in Table 9. The triple interaction term ( $WOE_{ict} \times FinOBI_{ict} \times innovation_{ict}$ ) in column (7), Table 12 suggests that there is no clear additional productivity impact from the combination of these three factors beyond their individual or two-way interaction effects.

The conditional effects from Table 12 indicate that the $TFP_{ict}$ of a $WOE_{ict}$ that innovates but faces financial obstacles is equal to 7.59, which is the same as the $TFP_{ict}$ of a non-innovative MOE with no financial obstacles. Overall, the findings in Table 12 confirm that innovative firms can achieve substantial TFP gains, averaging up to 12%. However, consistent with H4, these benefits are significantly diminished for WOEs due to their greater challenges in accessing credit. While credit obstacles alone are associated with a slight increase in TFP, likely due to greater efficiency in the use of resources, they clearly limit the productivity gains that come from innovation.

5.5 Robustness analysis

The main takeaway from the previous analysis is that financial obstacles disproportionately affect TFP of WOEs and WMEs. However, this effect may reflect discriminatory lending practices by financial institutions, which perceive lending to WOEs as a riskier decision. Similarly, women entrepreneurs may anticipate such discrimination, leading them to face less favorable lending criteria or even opt out of seeking external finance altogether. Despite our findings in Table 10, which show that women perform as well as, if not better than, men in industries characterized by external financial dependence, there may still be industry selection bias. This bias suggests that women entrepreneurs are more likely to self-select into specific industries or sectors that have lower average productivity or lower growth potential compared to industries dominated by male entrepreneurs. These unobserved underlying forces imply that the performance of WOEs might be affected by selectivity bias, without necessarily indicating that WOEs are inherently less productive than MOEs. To ensure that our previous results are robust to unobserved selectivity bias, we implement an IV estimation following the approach proposed by Rajan and Zingales (Reference Rajan and Zingales1998).Footnote ¹⁷ We construct an instrument using the interaction term between the inverse of a country- and time-variant Financial Development Index and the inverse of a Gender Equality Index. The relevance of this instrument stems from the idea that countries with higher levels of financial development typically offer better access to external financing, which can, in turn, influence firms’ ability to invest and innovate. Additionally, the gender equality measure reflects the extent to which WOEs can access the financial market system without discrimination. This instrument proves relevant as it encapsulates structural and economy-wide characteristics that affect the availability of finance and ensure equal access for entrepreneurs, irrespective of gender. Essentially, we assume that the interaction between country- or industry-level characteristics affects firm performance solely through its impact on access to resources, not through other unobserved factors. Thus, our instrument is considered exogenous, based on the premise that the historical and institutional factors driving gender norms and financial development are unrelated to specific firm productivity shocks. This satisfies the exclusion restriction, thereby isolating the causal effect of gender and financial constraint variables on $TFP_{ict}$ .

Data for the index of financial development ( $IFD_{ct}$ ) are collected from the International Monetary Fund (IMF) database.Footnote ¹⁸ $IFD_{ct}$ measures a country’s relative ranking in terms of depth, access, and efficiency of financial institutions and markets in the country. The higher the value of $IFD_{ct}$ , the more efficient the country’s financial system. We also extract data from the World Bank’s Women, Business, and the Law ( $WBL_{ct}$ ) index, which measures how laws and regulations in country $c$ affect women’s economic opportunities on a scale from 0 to 100, where 100 means equal legal rights for men and women.Footnote ¹⁹ To facilitate interpretation and align with our endogenous variables, $FinOB_{ict}$ and $FinOBI_{ict}$ , which interact with gender variables $WME_{ict}$ and $WOE_{ict}$ , we reverse the interpretation of the indices by subtracting each value from the maximum in our sample.

In the IV estimations, Tables 13 and 14, we instrument $WOE_{ict}$ , $FinOB_{ict}$ , $FinOBI_{ict}$ , and the interaction terms $FinOB_{ict}\times WOE_{ict}$ and $FinOBI_{ict}\times WME_{ict}$ applying two different approaches depending on the nature of the endogenous variables. For the binary endogenous variable $FinOB_{ict}$ , the respective interaction terms $\times WOE_{ict}$ , and gender variables $WOE_{ict}$ ,Footnote ²⁰ we use the two-stage residual inclusion (2SRI) approach as outlined by Terza et al. (Reference Terza, Basu and Rathouz2008). In this method, the first stage specifies the following probit model, $Pr(Y_{ict}=1|Z_{ct})=\Phi [\gamma _{0}+\gamma _{1}Z_{ct}+\gamma _{j}+\phi _{ict}]$ , where $Y_{ict}$ includes each time the endogenous variable, $WOE_{ict}$ , $FinO_{ict}$ , and $WOE_{ict} \times FinO_{ict}$ . Variable $Z_{ct}$ is our instrument defined as the interaction of $IFD_{ct}\times WBL_{ct}$ . The first-stage probit model includes industry fixed effects, $\gamma _{j}$ , to capture industry-variant idiosyncrasies as well as the respective control variables $M_{ict}$ . The second-stage estimation includes as regressors the first-stage residuals of the endogenous binary variables.Footnote ²¹ For the non-binary endogenous variables, $FinOBI_{ict}$ and the respective interaction term $FinOBI_{ict}\times WOE_{ict}$ , we implement the standard two-stage least squares (2SLS) approach where in the second stage, we include the predicted values from the first-stage regressions.

Estimates from the 2SRI model in Table 13 confirm the productivity penalty for $WOE_{ict}$ , which is found to be approximately 6.5%. Additionally, $WOE_{ict}$ experience a further productivity reduction of 5.1% compared to $MOE_{ict}$ when access to finance is an obstacle. Column (2) shows that the productivity penalty for $WME_{ict}$ remains around 10%, but there is no additional adverse effect from financial obstacles. The 2SLS estimates in Table 14 present a similar pattern. The productivity of $WOE_{ict}$ and $WME_{ict}$ consistently remains lower than that of $MOE_{ict}$ . Moreover, the productivity gap for women-led firms widens as financial obstacles become more severe. Since $TFP_{ict}$ is in logs, the impact on productivity is exponential, meaning the negative effect grows significantly as obstacles intensify. For example, as the intensity of obstacles increases from 4 to 5, the productivity of $WME_{ict}$ decreases by approximately 275.9% compared to male-owned firms, while for $WOE_{ict}$ , the reduction is 78%. Overall, Tables 13 and 14 demonstrate that unobserved selection bias does not affect the results regarding the persistently lower productivity of $WOE_{ict}$ and $WME_{ict}$ . On the contrary, after controlling for this bias, the productivity gap between $WOE_{ict}$ and $MOE_{ict}$ becomes even larger when financial obstacles are present and severe.