Gender and bank lending after the global financial crisis: are women entrepreneurs safer bets?

Abstract

Using gender as a theoretical framework, we analyse the dynamics of bank lending to small- and medium-sized enterprises (SME) in the aftermath of the 2008 global financial crisis. Using six waves of the SME Finance Monitor survey, we apply a formal Oaxaca–Blinder decomposition to test whether gender impacts upon the supply and demand for debt finance by women. Reflecting established evidence, we found women had a lower demand for bank loans; contradicting accepted wisdom however, we found that women who did apply were more likely to be successful. We argue that feminised risk aversion might inform more conservative applications during a period of financial uncertainty which may be beneficial for women in terms of gaining loans. However, we also uncover more subtle evidence suggesting that bank decisions may differ for women who may be unfairly treated in terms of collateral but regarded more positively when holding large cash balances.

Appendix. Details of the Blinder–Oaxaca decomposition

In the context of this study, let Y_F and Y_M denote the outcome variables for models estimating the probabilities of demanding loans or loan application being granted for male and female owned ventures, respectively; and X a vector of predictors for Y. We are interested in how much of the mean outcome difference,

$$ \varDelta =E\left({Y}_M\right)-E\left({Y}_F\right) $$

(1)

where E(Y) is the predicted outcome variable from a linear probit model, is accounted for by group differences in the predictors X.

Further let β be the coefficient estimates from the model, it can be shown that following the standard assumptions of linear regressions, the mean outcome difference Δ can be written as:

$$ \varDelta =E\left({Y}_M\right)-E\left({Y}_F\right)=E{\left({X}_M\right)}^{\hbox{'}}{\beta}_M-E{\left({X}_F\right)}^{\hbox{'}}{\beta}_F $$

(2)

Now, if we introduce a ‘non-discriminatory’ coefficient vector β* embedded in the coefficient estimates β, that accounts for the part of mean outcome probability differences explained by the predictors, Eq. (2) can be rearranged as:

$$ \varDelta ={\left[E\left({X}_M\right)-E\left({X}_F\right)\right]}^{\hbox{'}}{\beta}^{\ast }+\left[E{\left({X}_M\right)}^{\hbox{'}}\left({\beta}_M-{\beta}^{\ast}\right)+E{\left({X}_F\right)}^{\hbox{'}}\left({\beta}^{\ast }-{\beta}_F\right)\right] $$

(3)

Equation (3) can be seen as a twofold decomposition where the first component,

$$ Q={\left[E\left({X}_M\right)-E\left({X}_F\right)\right]}^{\hbox{'}}{\beta}^{\ast } $$

(4)

is defined as the ‘quantity effect’ referring to the part of the outcome differential explained by group differences in X. In turn, the second component,

$$ U=E{\left({X}_M\right)}^{\hbox{'}}\left({\beta}_M-{\beta}^{\ast}\right)+E{\left({X}_F\right)}^{\hbox{'}}\left({\beta}^{\ast }-{\beta}_F\right) $$

(5)

can be seen as the unexplained part usually attributed to discrimination,⁶ as it depicts the differences in the sensitivities to predictors for different groups. The non-discriminatory coefficient vector β* is suggested in previous studies either to be equal to the coefficient estimates of one of the groups (e.g. male) assuming that discrimination is only directed towards the other group (e.g. female), or a weighted average of group coefficient estimates β_M and β_F, or the coefficients from a pooled regression. In this study, we define β* as the coefficients from pooled probit regressions on loan demand and supply.⁷