Financial Constraints to Exporting: Experimental Evidence from Rwanda

1) We will use administrative data from the Rwandan Revenue Authority (RRA) including customs data, VAT tax data, corporate income tax data, and other datasets to estimate a model predicting likely adoption of the EGF based on firm characteristics. This analysis will be accessed on an on-site secure RRA data facility. The model will be used to identify ~5,000 firms with the highest propensity score (excluding existing recipients).

We will then conduct a listing exercise with these ~5,000 firms. This exercise will be conducted mostly on the phone, with an in-person follow-up for (potentially a subset) those unreachable via phone. Firms will be given a short survey to collect additional contact information and gauge general interest in expanding access to credit and exporting. This will be used to screen out firms who (1) are hard to reach/unwilling to be surveyed; (2) do not express interest in increasing access to credit or (3) do not express interest in exporting in the future. We expect this to result in a reduced target sample of 2,000 firms.

2) We will randomize these 2,000 firms into 500 treatment and 1,500 control. As detailed below in the Power Calculation section, we have chosen to have more controls than treated firms in this experiment due to the high cost of treatment relative to data collection, especially for outcomes available in the administrative data. In these cases, a greater number of controls is optimal for power given a fixed project budget (Duflo et al, 2006). We will use the administrative data to stratify our randomization by firm size, sector, and prior export history. This administrative data from the pre-experiment period will also be used to test for balance among treatment and control groups, as well as to serve as baseline controls in our analysis.

We will then conduct a door-to-door campaign in conjunction with the BRD for treatment firms. This campaign will be comprised of two components:

(i) An information campaign, in which the firm will be given information about the EGF, including details about the loan facility, the application procedure, and the typical uses of the loan among existing borrowers.

(ii) Application support, in which the BRD will send a loan officer to the firm’s establishment multiple weeks after the first visit (giving the firm time to compile the documents) to assist in assembling and filling out the application paperwork, which includes a business plan; financial reports (profit and loss statements, balance sheets and cashflow statement); and bank statements (or mobile money statements) documenting the financial conditions of the firm.

The BRD will then conduct its loan review as under its standard operating procedure for all applicants. The goal is that the door-to-door campaign serves as "randomized encouragement" for firms to apply, such that we can use treatment as an instrument for taking out an EGF loan. We will then measure the impact of taking out an EGF loan on a variety of firm and supply chain outcomes (see below for more details on outcomes).

2024-08-01

2025-03-01

Using a combination of administrative data (customs, VAT, Corporate income tax, firm-level employees information from PAYE) and firm surveys we will measure the impact of the Export Growth Fund (EGF) over several key outcomes:

1) Total firm credit access.
2) Firm revenue, employment and export performance

1) Total firm credit access: we will measure total borrowing from firm-level surveys on the loan portfolio. The objective is to analyze whether firms substitute away from commercial loans at market rates and whether they increase their total borrowing.

2) Firm revenue, employment and export performance: we will measure these outcomes in administrative data, using CIT, PAYE and customs data, respectively. Export performance will be separated into the extensive margin (any exporting) and several intensive margin measures (total exports, number of export destinations, number of products exported).

1) Mechanisms affecting firms' performance outcomes: quality upgrading, certification, export promotion, capacity expansion, etc.

(2) Impact on industry -level outcomes and spillover effects on non-recipients, including upstream firms

(1) Mechanisms affecting firms' performance outcomes: this will be measured a using a combination of firm surveys and administrative data. This will include measures of quality upgrading (measured using self-reported quality measures), number and type of certifications obtained (obtained from the Rwandan Standards Board (RSB) and self- reported in firm surveys), total investment and expenses in training and research (as reported in administrative data), export promotion activities (as reported in firm surveys) and capacity expansion (as reported by self-assessed total capacity in firm-level surveys).

(2) Impact on aggregate outcomes and spillover effects on non-recipients, including upstream firms: For aggregate outcomes, we will analyze how industry-level variables such as total exports, total domestic sales and total number of employees (constructed from administrative data) are affected (using, for instance, the share of treated firms across sector/regions, conditional on the share in our study sample). We will also estimate spillover effects on upstream suppliers, using pre-existing firm-to-firm linkage data in the VAT data.

The Export Growth Fund (EGF) was first launched in 2017, strategically conceived by the Ministry of Trade and Industry (MINICOM) and implemented by the Rwandan Development Bank (BRD). Since then, more than 120 loans have been issued to approximately 80 firms, totaling $21 million USD in lent funds. However, this falls far below the 200-500 beneficiaries and $50 million USD per year in loans targeted by MINICOM (MINICOM National Export Strategy, 2015). The slow launch has been linked to low awareness of the facility. Qualitative exploratory work conducted by the PIs, including interviews with non-recipients, confirms limited awareness of the facility among exporters and large firms (although interest is high upon being informed).

Given their plans to significantly expand access in the coming years, the BRD and MINICOM have agreed to partner with the PIs to conduct a randomized door-to-door campaign to target firms. First, we will use rich administrative data, to which we have been given access by the Rwandan Revenue Authority (RRA), to identify target firms.

Then, with a random subset of these firms, we will conduct a door-to-door campaign in conjunction with the BRD. This campaign will be comprised of two components: the first is an information campaign, in which the firm will be given information about the EGF, including details about the loan facility, the application procedure, and the typical uses of the loan among existing borrowers. The second component is application support, in which the BRD will send a loan officer to the firm’s establishment multiple weeks after the first visit (giving the firm time to compile the documents) to assist in assembling and filling out the application paperwork, which includes (i) a business plan; (ii) financial reports (profit and loss statements, balance sheets and cashflow statement); and (iii) bank statements (or mobile money statements) documenting the financial conditions of the firm.

The BRD and MINICOM have agreed to randomize this door-to-door outreach among target firms. We will therefore use randomized receipt of the door-to-door campaign as an instrument for EGF loan take-up. Though we do not expect perfect compliance (see the below section on Power Calculations for take-up assumptions), we do expect this to serve as a strong first-stage for EGF adoption, given that (i) firms are pre-screened as likely adopters in the administrative data; (ii) firms have expressed a high degree of interest for information + application support, and (iii) the BRD is eager to issue a large number of loans to this target group of firms.

Not available

Randomization will be done by the PIs on a computer at a secure facility of the Rwanda Revenue Authority

Firms

No

2000 Firms

2000 Firms

1500 Firms Control, 500 Firms Treatment

We conduct power calculations using administrative data from the RRA, along with the results from a retrospective difference-in-difference analysis of a previous non-experimental round of the EGF program. The average revenue of the 2000 most similar firms to EGF recipients according to our PSM analysis is 280 million Rwandan Franc (RWF) ($232,000 USD). The standard deviation is 233 million RWF ($192,000 USD). Since we will be using administrative data to evaluate the impact of the intervention, we will have access to several post-intervention measurements. For the purpose of this power calculation, we analyze the impact with 3 follow-up measurements and 1 baseline. The within-firm correlation of revenues across periods in the administrative data is 0.81. As explained above, we will use a larger control set than our treatment, taking advantage of the almost null cost of additional control units using administrative data. We use a treatment-to-control ratio of 0.25, with 1500 firms in the control and 500 in the treatment. To be conservative, we assume a take-up rate of 25%, below the average of 30% from 5 microcredit studies in Banerjee et al. 2015. However, there are two factors that we expect to drive a strong take-up rate in our case: first, interest rates under EGF are highly subsidized and significantly below commercial bank rates. Secondly, we will target a group of likely adopters (those that share similar characteristics to prior EGF recipients), randomizing within this group of firms. Our simulations suggest that with 500 treated firms, we will be able to detect an increase in revenues of 0.26 standard deviations. Importantly, this is less than half of what we find in our difference-in-difference analysis, where we find an average increase of 0.6 standard deviation 2 years after receiving an EGF loan. In fact, with 500 treated firms, we would be able to detect an impact like that we find in the DiD retrospective analysis with a take-up rate as low as 11%.