This is the methods section for my larger research paper, The Effect of Credit Bureau arrival on Access to Finance and Lending in Sub-Saharan Africa. The corresponding blog entry presents the main findings, but I figured not everyone would want to dive into the methods of why and how I did what I did. This is for everyone that is interested. I’ve left the references and footnotes in because I cannot link to many of these sources.
To see the results page, click here.
If you have any questions, comments, or criticisms, please feel free to contact me. One of the reasons I’m posting this here is so I can continue learning as well.
Methods
I used firm-level data from World Bank’s Enterprise Surveys to examine the effect of credit bureaus on overall lending in a country and access to finance.
This data analysis draws largely on papers from Triki and Gajigo (2014) and Peria and Singh (2014).[1] The main dependent variable captures a firm’s perception of access to finance as an obstacle (See Table 1 for Summary Statistics). I use the Enterprise Survey question, How Much of An Obstacle: Access to Finance, which scores responses on a scale of 0 to 4. A score of “0” represents “No Obstacle” and “4” represents a “Severe Obstacle”. I also use the dependent variables such as “Establishment has a line of Credit or a Loan” and “% of Working Capital Borrowed from a Bank” to try and capture objective measures of financial access.
The data set includes countries with at least two years of survey data. I only use firms that appeared in both surveys in order to measure a change in access to finance. There are both advantages and disadvantages to this approach. The panel data reduces the total countries we can include, and the firms are not random. Regardless of how the firms were chosen for the initial survey, appearing on two surveys means that the firm survived for at least three years. This means that all of the firms used in the dataset were relatively successful, at least compared with ones that did not exist by the second survey. This biases the dataset in favor of firms that can access finance more easily, as older firms generally face fewer obstacles.[2] However, even though the surveys most likely bias in the direction of more efficient firms that should struggle less with access to finance, it can still tell us something about the lending environment. In having firms that all survived at least three years from the previous survey, these businesses are more alike and better for comparative purposes. This way we will not merely compare the efficiency and competency of businesses, but actually capture the impact of a credit bureau on the lending environment.
This panel study uses a differences-in-differences approach and focuses on a grouping of countries from Sub-Saharan Africa that share some common features in the first year of the survey. While it is impossible to create a perfect match among countries, these are all on the same continent and many share a region, and some even share languages (See Figure 1). While there still are some issues regarding the common trends assumption that is vital to the differences-in-differences approach, these countries share more in common than similar studies that use sets of 63 countries across continents.
I created two groups of countries, a treatment and a control (See Figure 1). The firms in the treatment group are in countries that received a credit bureau between surveys. The control group includes firms in countries that did not gain a credit bureau in the time period between surveys. There are ten countries in the control group and six in the treatment group. The first survey year is the Pre-Treatment period, and the second survey year is the Post-Treatment period, after the establishment of the credit bureau. This method allows us to compare if credit bureau arrival had any causal effect on access to finance.
There are also three sample decisions I made that are worth noting here. First, I run models that include Kenya in the Treatment Group and ones that do not. Some studies include Kenya as receiving a credit bureau between its survey years of 2007 and 2013, but Doing Business Credit Bureau coverage data shows an active bureau in the country in 2007. Their first licensed bureau did not begin operations until after 2009. Since there is a discrepancy, the main model excludes Kenya, but a separate one factors it in.
Second, I run models that include Malawi in the Treatment Group and ones that do not. Some studies identify Malawi as having a credit bureau, but having coverage less than .1% of the population. Malawi introduced a Credit Bureau Act in 2010, but there were significant issues in getting commercial banks to adhere to the legal requirements. Since Doing Business reports do not include Malawi as having a credit bureau during this period, I include Malawi in some models, but exclude it from the main one.
Finally, I include Tanzania in the control group even though they received licensing for a bureau in 2013. Since their survey was reported in 2013, the businesses surveyed would have responded prior to the credit bureau’s initial operations.
Figure 1: Control and Treatment Groups
| Control | Treatment |
| Angola 2006 & 2010 (183)
Burkina Faso 2006 & 2009 (88) Cabo Verde 2006 & 2009 (53) Cameroon 2006 & 2009 (75) DRC 2010 & 2013 (92) Ethiopia 2011 & 2015 (372) Mali 2007 & 2010 (152) Niger 2005 & 2009 (71) Senegal 2007 & 2014 (238) Tanzania 2006 & 2013 (115) |
Ghana 2007 & 2013 (31)
Kenya 2007 & 2013 (151) Malawi 2009 & 2014 (87) Rwanda 2006 & 2011 (70) Uganda 2006 & 2013 (209) Zambia 2007 & 2013 (152) |
Firms in Parenthesis
I condition this study on a few control variables. I include Firm Age as older firms face fewer challenges in accessing finance.[3] I also include Manager Experience and a variable, Audit, measuring whether a firm has its financial statements externally audited. These variables measure both efficiency and competency of a firm, as well as transparency. Both could make financial access easier.[4] Larger companies also face fewer obstacles accessing finance, so I use the Employees variable as a measure of the number of full-time employees. The last firm-specific control variable is Foreign Ownership, as firms owned by private foreign individuals or companies generally report access to finance as a lower obstacle.[5]
In addition to the firm-specific control variables, I also include a binary variable indicating whether the country has a Credit Registry. Even though this study mainly focuses on the presence of a credit bureau, it must still take account of other credit-sharing institutions in the country that influence the ability of a firm to access credit.
I also condition for financial sector development by including the variable Commercial Bank Branches (per 100,000 adults) to account for the role bank presence has in access to finance. This is an attempt to recognize the problem identified by Triki and Gajigo, in that even though Private Credit Bureaus are associated with more access to finance, that could be the result of a country’s strong financial infrastructure.[6] Unfortunately, this will not solve the problem entirely, as studies also show that foreign banks are more likely to enter a country if a private credit bureau is already present.[7] Since screening borrowers costs money, and foreign banks will not automatically possess data on the new clientele base, credit bureaus reduce information costs and make it easier for foreign banks to gain information about new customers.[8] While the presence of a credit bureau or credit registry may be the product of increased development of the financial sector, the bureau could also influence the development of the financial sector. Still, a higher rate of bank branches in a country is important to include because it captures financial infrastructure better than other variables.
Finally, I include macro or country-level variables to use as a robustness check. Since Triki and Gajigo’s study resulted in the presence of a bureau losing its significance when they controlled for macroeconomic factors, I include Domestic Credit to Private Sector (% of GDP) to measure overall lending, GDP Per Capita, GDP Per Capita Growth, and Inflation. These variables capture the overall lending environment that could greatly influence the ability for firms to access finance.
For purposes of robustness and transparency, I ran the same regression without controls, with firm level controls, and with firm and macro level controls. There is a danger that control variables create new and different samples. Running the analysis under multiple conditions should tell us if that is indeed occurring.
Summary Statistics
Table 1 Firm-Level Summary Statistics
| Mean | Std. Dev. | Min. | Max. | ||
| How Much Of An Obstacle: Access To Finance | 2.10 | 1.41 | 0.00 | 4.00 | |
| % Of Working Capital Borrowed From Banks | 8.23 | 19.53 | 0.00 | 100.00 | |
| Establishment has A Line Of Credit Or Loan From A Financial Institution? | 0.23 | 0.42 | 0.00 | 1.00 | |
| Treatment Group | 0.35 | 0.48 | 0.00 | 1.00 | |
| Post Treatment | 0.49 | 0.50 | 0.00 | 1.00 | |
| Firm Age | 23.01 | 121.86 | 1.00 | 2019.00 | |
| How Many Years Of Experience Working In This Sector Does The Top Manager Have? | 14.67 | 9.61 | 1.00 | 55.00 | |
| Num. Permanent, Full-Time Employees At End Of Last Fiscal Year | 62.23 | 267.33 | 1.00 | 8000.00 | |
| Financial Statements Checked & Certified By External Auditor In Last Fiscal Yr? | 0.46 | 0.50 | 0.00 | 1.00 | |
| % owned by Private Foreign Individuals, Companies Or Organizations | 11.86 | 30.05 | 0.00 | 100.00 | |
| Commercial Banks per 100,00 Adults | 3.72 | 3.62 | 0.62 | 28.18 | |
| Domestic Credit to the Private Sector (% of GDP) | 16.85 | 9.33 | 3.92 | 57.96 | |
| GDP Per Capita | 1121.30 | 890.62 | 311.25 | 3886.48 | |
| GDP Per Capita Growth | 3.80 | 4.24 | -4.43 | 16.64 | |
| Inflation | 7.71 | 6.6! | -0.77 | 22.39 | |
| Observations | 2724.00 |
Source: World Development Indicators, Enterprise Surveys (http://www.enterprisesurveys.org), The World Bank.
References
[1] Thouraya Triki and Ousman Gajigo, “Credit Bureaus and Registries and Access to Finance: New Evidence from 42 African Countries,” Journal of African Development 16, no. 2 (2014); Maria Soledad Martinez Peria and Sandeep Singh, The Impact of Credit Information Sharing Reforms on Firm Financing? (2014).
[2] Thorsten Beck et al., “The Determinants of Financing Obstacles,” Journal of International Money and Finance 25, no. 6 (2006): 933, http://dx.doi.org/10.1016/j.jimonfin.2006.07.005.
[3] Ibid.
[4] Peria and Singh, 13.
[5] Beck et al., 933.
[6] Triki and Gajigo, 5.
[7]Hsiangping Tsai, Yuanchen Chang, and Pei-Hsin Hsiao, “What Drives Foreign Expansion of the Top 100 Multinational Banks? The Role of the Credit Reporting System,” Journal of Banking and Finance 35, no. 3 (2011): 589, http://dx.doi.org/10.1016/j.jbankfin.2010.05.016.
[8] Ibid., 597.
Enterprise Surveys (http://www.enterprisesurveys.org), The World Bank.
Due North 