Our experimental design is motivated by two common challenges faced by researchers when analyzing the effect of financial education on financial policies: i) the endogenous decision to appoint a financial expert CEO / to obtain financial education; and ii) limited availability of data.
The literature on the effects of managerial human capital on firm policies has mostly relied on cross-sectional analysis, which renders causal inference very challenging as endogenous matching between firms and managers biases the estimates (Guenzel and Malmendier (2020)). Since Bertrand and Schoar (2003), most studies have used panel regressions to estimate potential CEO effects using within-firm variation due to CEOs switching firms. However, Custodio and Metzger (2014) and Fee et al. (2013), for instance, cast doubt on this methodology for identifying managerial effects on policy choices. They argued that CEO turnover events are endogenous, and managerial "style changes" are anticipated by corporate boards at the time of the CEO selection decision. While firm-fixed effects absorb firm heterogeneity that is time invariant, it cannot be ruled out that firm time-varying characteristics, unobserved by the econometrician, such as some strategic decisions, drive both financial policies and the characteristics of the appointed CEOs. In the context of financial expertise, Custodio and Metzger (2014) showed that firms run by managers with past work experience in finance have better access to external financing and allocate their firms’ financial resources more efficiently. However, this study also shows that financial expert CEOs are more likely to be appointed by older firms, which suggests an endogenous matching.
To identify a treatment effect of financial expertise on firm policies, one would need to randomize financial expertise across firms. One way of doing so could be an actual random allocation of CEOs to firms, which would take care of endogenous matching. However, this experiment is not feasible in practice. Moreover, a random allocation of CEOs to firms does not deal with the concern that there are unobservable characteristics of CEOs that correlate with financial expertise. For instance, CEOs with financial expertise might be of higher (or lower) ability or talent.
To overcome endogeneity concerns we propose randomizing financial education of top managers while maintaining the match between CEOs and firms. To be specific, we treat managers with financial education by offering free MBA-style lectures on corporate finance and risk management to top managers. Such a randomized controlled trial (RCT) can be used to identify a treatment effect of finance education on financial policies. The second challenge for our study is the availability of data. First, most companies in Mozambique are private, and access to financial statements is limited. Moreover, some outcomes, such as the use of specific valuation techniques or risk management instruments, are difficult to measure in those statements.
In order to address both concerns, endogeneity and data availability, we implemented the intervention in a staggered way, i.e., we ultimately taught both, the treatment and the control group. By treating both groups, we provide incentives to firms to share their financial statements with us, as well as to participate in face-to-face surveys, allowing us to collect data on nonstandard outcomes. The first cohort – the treatment group – received the treatment in May 2017, while the second cohort – the control group – received the same treatment in November 2018/April 2019.
The staggered nature of the intervention also helps to address the concern that the formation of expectations could bias our estimates (Chemla and Hennessy (2019)) because despite the greater uncertainty for the control group, which is treated later, both the treatment and control groups expect to be treated. Last, it reduces ethical concerns of providing a permanent advantage to one of the groups.
To address the concern of endogenous selection into our treatment, we conducted the randomization among the firms that applied to the program. We also stratified the randomization by industry to ensure that the same industries were represented in both groups. As noted by Sutton (2014), a sample stratified by industry provides a "fair and complete picture of the country’s industrial capabilities". Because there were subsidiaries of business groups in our sample (i.e., companies belonging to the same group that were managed by one or more participating managers) we made sure the these companies were part of the same group to minimize contamination concerns.