Selected Issues


Selected Issues

Analyzing the Evolution of Credit and Non-Performing Loans Based on Credit Registry Data1

A. Background

1. The banking system of São Tomé and Príncipe is characterized by persistent high levels of non-performing loans (NPLs). The NPL ratio of the banking system tripled between 2013 and end-June 2017 to 32 percent, peaking at about 36 percent in 2016. This level of NPL is significantly higher than in peer countries, such as Caribbean island states or small states in general (Pacific islands excluded), and also exceeds that of neighboring Angola. The share of the number of loans that are non-performing, the “unweighted” NPL ratio, however, is substantially lower, at less than 10 percent of GDP. This indicates that the high NPL ratio, which is measured in value terms, is driven by the poor performance of a limited number of large loans.

Figure 1.
Figure 1.

São Tomé and Príncipe: NPL Ratio, 2013-17

Citation: IMF Staff Country Reports 2018, 322; 10.5089/9781484385005.002.A004

Sources: BCSTP and IMF staff calculations

2. As part of a comprehensive strategy to address the high NPL ratio, a central credit registry (CRC) has been created. The BCSTP made the CRC operational in March 2011, collecting information on the characteristics and status of loan exposures on a monthly basis. The coverage of the credit registry has increased from about 50 percent in the initial stages to 96 percent recently.2 Besides providing a range of descriptive statistics, micro data can help identify trends within banks’ credit portfolios, spanning both performing and non-performing loans, and derive indicators to be used for financial stability analysis and banking supervision.

Figure 2.
Figure 2.

Selected Countries: NPL Ratio in 2017

Citation: IMF Staff Country Reports 2018, 322; 10.5089/9781484385005.002.A004

Source: World Bank

3. This paper analyzes the evolution of credit and credit quality in São Tomé and Príncipe, using micro data from the CRC. In the following sections, the paper first presents the main characteristics of banks’ loan portfolio and, in particular, non-performing loans, then it conducts a vintage analysis of default rates and loan transition matrices that give indications of how loan quality evolved during the sample period. It concludes with recommendations on measures for improving the CRC and for using the rich data to guide bank supervision and support the implementation of authorities’ NPL reduction strategy.3

B. The Credit Registry and Credit Market Characteristics

4. The analysis uses a multi-year data sample. It covers March 2011 to June 2017, contains 469,823 observations from up to 12 banks (including banks that have meanwhile been closed or taken over), and are related to 29,238 loans and 13,039 borrowers (around seven percent of the total population or 14 percent of the adult population). Banks used to report data with a certain lag which will be reduced significantly through a new online reporting system. During this time period, the number of loans covered in the sample grew steadily while the share of performing loans declined. In addition, the rate of loan origination varied over time (Figure 3).

5. The credit registry collects a wide range of borrower information, as reported by banks. In addition to borrowers’ personal information, the CRC contains information on the bank name, date of loan origination, maturity date, loan amount at origination, carrying amount, currency, interest rate, status of the loan in given month (category of loan classification), type of loan rate (fixed vs. variable rate), type of borrower (household vs. firm), the sector/occupation of the borrower, and the district of the borrower. The CRC also provides an indirect measure of borrower income, expressed by the share of disposable income in total debt. Unfortunately, currently no information on collateral value is provided.

6. The CRC data show that the credit market in São Tomé and Príncipe is highly concentrated. On the demand side, the largest firms (top 5 percent) received 54 percent of the total corporate credit, while the largest individual borrowers received 29 percent of the retail credit. On the supply side, the largest bank extended almost half of the loans in the sample and the other two largest banks around 15 percent each. Geographically, the loans are concentrated in the district of the capital (59.4 percent).

Figure 3.
Figure 3.

São Tomé and Príncipe: Characteristics of Bank Credit

Citation: IMF Staff Country Reports 2018, 322; 10.5089/9781484385005.002.A004

Sources: BCSTP and IMF staff calculations

7. High loan concentration makes banks’ balance sheets vulnerable to the financial health of large borrowers. At one small bank the largest borrower accounts for 40 percent of the loan portfolio, while at three others that share amounts to 20 percent of total loans. Even at the largest banks the largest 10 borrowers, representing less than 1 percent of the total numbers of borrowers, account for over 30 percent of the loan volume. The default of a single large borrower is thus likely to have a substantial impact on individual banks.

8. System wide, the higher amount of bank credit goes to enterprises and are in local currency. While loans to individuals account for 97 percent in terms of number of loans, loans to firms represent 71 percent of the total loan amount. Almost all loans are denominated in local currency (New Dobras, STD—95.6 percent), followed by EUR (2.6 percent), and USD (1.8 percent). Only 5 percent of the borrowers in the sample have access to foreign-currency credit. For individuals, the median domestic currency loan value is STD 35,000 (about USD 1,750), with 10 percent of the loans being above STD 130,000 (about USD 6,500). These figures are large given that São Tomé and Príncipe’s GDP per capita is only USD 1,772 (end-2017). For firms, that median loan value is STD 500,000 (about USD 25,000), with 10 percent of the loans exceeding STD 4.9 million (about USD 245,000). The median loan value for USD and EUR loans are much larger at about USD15,000 and EUR 30,000, respectively.

9. There is considerable variation in interest rates charged and maturities. Interest rates may be as high as 54 percent and maturities range from 1 day to almost 37 years. The average interest rate for STD loans is much higher than for foreign currency loans (20 vs 11 percent). In addition, the maturity of STD loans is shorter, on average 2.7 years, compared with the average of 5.0 years for foreign currency loans. This difference probably reflects unobservable differences in the creditworthiness of borrowers obtaining loans in local currency and in foreign currency. About 80 percent of the loans have a fixed rate.

10. Loan characteristics also vary across the four largest lenders. In terms of loan size, the largest and third largest bank extend loans which are on average twice as large as the loan amount of the other two large banks. On the other hand, the two largest banks charge an average interest rate which is slightly above the sample average of 20 percent, while the third largest bank charges a rate that is significantly lower at 6 percent. The largest bank supplies 72 percent of the foreign currency loans.

11. The majority of the borrowers in the country have a single banking relationship and one loan. Most of the borrowers have loans only from one bank (80 percent), a few have credit with two banks (16 percent), and the remaining 4 percent with three or more banks. Around half of the borrowers have only one loan in the dataset, 25 percent have two loans and 14 percent have three loans. Another 1.5 percent of borrowers have ten or more loans. The four largest borrowers with over 30 loans in the sample borrow exclusively from one bank, and all of them are enterprises.

C. The Evolution of Non-Performing Loans and Default Rates

12. Loan classifications in São Tomé and Príncipe are in line with international practices. According to the regulation (BCSTP, 2007), a loan exposure is classified in one of five categories, principally according to the criterion of days past-due but also taking into account the financial condition of the borrower and the status of collateral. Category VI is reserved for written-off loans. The standard definition of a NPL loan is that the borrower is late in making its scheduled payment by 90 or more days. Hence, categories III to V are considered non-performing.

Table 1.

BCSTP’s Loan Classification System

article image

13. The CRC data permit extracting descriptive statistics on NPLs. On average during the sample period, seven percent of the loans became non-performing and a 20 percent had a delayed payment of at least 30 days. Those loans that become non-performing during their lifetime do so after 15 months on average. Around 11 percent of the borrowers in the sample have had at least one loan classified as non-performing. However, the probability of a firm having a non-performing loan is much higher than the probability of an individual having a non-performing loan: 39 percent of the firms in the sample had at least one non-performing loan compared to 10 percent among individual borrowers.

14. As of end-June 2017, non-performing loans represented 32 percent of the total loan amount and are concentrated among a small number of borrowers as well as banks. The top 10 delinquent borrowers (about 1 percent of the borrowers with NPLs) account for 43 percent of the total NPL amount, with the top 3 borrowers representing 22 percent. Among the top 10 borrowers with NPLs, nine are firms, mainly from the wholesale trading sector. One active bank holds 32 percent of the loans classified as a loss (unweighted number), while a liquidated bank holds another 40 percent.

15. NPLs have significantly different characteristics than performing loans (Table 2). NPLs have a higher interest rate, a longer maturity, are more likely to be denominated in local currency, are larger and tend to have a fixed rate. Although these differences in characteristics between NPL and non-NPL loans are statistically significant, their magnitude is fairly small, with the exception of the loan amount and having a fixed rate. At the same time, these differences can also reflect unobservable borrowers’ creditworthness.

Table 2.

São Tomé and Príncipe: Average Characteristics of NPL and Performing Loans

article image

16. Having had a NPL in the past does not seem to reduce the chances of getting a new loan in the future. In fact, 45 percent of the borrowers who had a delay in loan repayment of more than 90 days are able to contract a new loan afterwards. Of these new loans, 41 percent is with the same bank where the delay occurred and 59 percent are with a different bank.

17. Borrowers having relationships with more than one bank are more likely to have a NPL. As a matter of fact, 10 percent of the loans from multiple-bank borrowers were NPL, compared to 6 percent of the loans from single-bank borrowers. On the other hand, having more than one loan does not seem to influence the probability that a borrower will have a NPL loan. This suggests that some borrowers explored banks’ incomplete information about their credit histories.

18. Default rates of new loans increased during 2012-15 (Figure 4). Vintage analysis shows that the incidence of newly-granted loans turning non-performing within the first 12 months rose from 0.2 percent for the 2012 vintage to 3.5 percent in the 2015 vintage.4 If weighting this rate by the number of months that a loan stayed in default during that 12-month period, the default rate drops significantly, as on average a loan is only around 4 months in default within the first year before performing again.5 This weighted default rate has also trended up, but more moderately. The BCSTP reported to the mission that the 12-month default rate picked up toward the end of 2017 (i.e. outside the data sample), reaching 6.7 percent in December 2017. It is natural that lifetime default rates (depicted on the right axis) are much higher, having peaked at around 10 percent in 2014, but have been trending lower in recent years because loans are still relatively new to default.

Figure 4.
Figure 4.

São Tomé and Príncipe: Default Rates per Origin Year, 2012-15

(Percent of loans originated in a given year)

Citation: IMF Staff Country Reports 2018, 322; 10.5089/9781484385005.002.A004

Sources: BCSTP and IMF staff

19. A more comprehensive vintage analysis illustrates that default rates at different time horizons have been rising throughout the sample period (Table 3). It is natural that the default rate rises as a loan ages. The vintage table below shows that loans issued in 2015 showed a 1.7 percent delinquency rate after only one month, with the rate increasing gradually to 5.8 percent after 18 months.6 However, as with the 12-month default rates, default rates at other time horizons also increased during 2012-15, before receding somewhat in 2016.7 For example, the 3-month default rate increased from 0.4 percent in the 2012 vintage to 1.7 percent in the 2015 vintage.

Table 3.

São Tomé and Príncipe: Comprehensive Vintage Analysis, 2012-16

(In percent)

article image

20. To understand in more detail the dynamics of loan developments, transition matrices are contructed to show the migration within the loan classification system. Table 4 presents the monthly transition matrix of loan classifications, averaged over the entire 2011-17 sample period. The entries in the matrix are sample frequencies of transitions from one category to another which occurred in a given period (Feng et al., 2008). It can also be construed as the probability that a loan will change its classification category next month or remain unchanged, given its current classification. On average, two percent of “regular” (category I) loans were re-classified as “under supervision” (watchlist loans with payment delayed by 30-89 days). Once a loan is classified as “under supervision,” there is still a high chance of 42 percent that it will turn “regular” again in the next month. However, this probability decreases considerably to at most 5 percent once a loan is classified as non-performing (“below standard”, or worse). Once a loan is classified as “below standard”, there is actually a higher probability (9 percent) that the classification of this loan will worsen to “doubtful” (category IV) or “loss” (category V) in the next month.

Table 4.

São Tomé and Príncipe: Transition Matrix

(Average of Monthly Transitions, 2011-17 Period, in percent)

article image

21. A comparison of transition matrices at different years reveals recent trends in loan classifications. Comparing the average monthly transitions for 2015, 2016, and the first half of 2017 (Tables 5 to 7), three trends become apparent. First, there is an increased transition rate for category II watchlist loans back to category I (54.8 percent versus 40.0 percent). Second, once loans are classified in the “early” NPL categories III and IV, a higher share now migrates to the loss category V (a combined rate of 17.9 percent compared to 11.5 percent in 2015); this is reportedly explained by more stringent loan classification, particularly in the context of on-site inspections. Third, the unsound practice of ad-hoc re-classifcation of loss loans as regular category I loans—perhaps following loan restructuring—has been diminished in 2017 compared to previous years (0.8 percent versus 5.7 percent in 2016). Similarly, there are no more cases of written-off loans being put back into category I as was the case in previous years (particularly in 2015 with a transition rate of 7.6 percent).

Table 5.

São Tomé and Príncipe: Transition Matrix

(Average of Monthly Transitions, January-December 2015, in percent)

article image
Table 6.

São Tomé and Príncipe: Transition Matrix

(Average of Monthly Transitions, January-December 2016, in percent)

article image
Table 7.

São Tomé and Príncipe: Transition Matrix

(Average of Monthly Transitions, January-July 2017, in percent)

article image

22. Expanding the analysis further to include the transition matrix based on loan value shows that small loans are more likely to become temporarily non-performing. Contrary to the previous matrices that count the number of loans maintaining or changing their status, this variant sums up the carrying value of the loans in each cell and then expresses the transitions in percentage terms as before. Transition rates will be relatively higher when it is mostly large loans that migrate to other categories and vice versa. Comparing the two matrix variants for the first half of 2017 (Table 8) illustrates that the transition rates for category II and III loans returning to category I (regular) are much lower when basing the transitions on carrying amounts (see Table 6). The implication is that many small loans—probably mostly consumer loans—become temporarily past-due before debtors can clear their arrears. As the mission learned from banks, the delay is in some cases caused by late salary payments to the borrowers.

Table 8.

São Tomé and Príncipe: Transition Matrix, Amount-Weighted

(Average of Monthly Transitions, January-June 2017, in percent)

article image

23. The low transition rate from category V to VI is consistent with the fact that banks do not have the practice of writing off loans. During the sample period, only one financial institution—a bank being liquidated—wrote off loans from its balance sheet. On average, loans are labeled as a “loss” (over 365-day payment delay) but are still not written off for 9 months (median: 5 months). Ten percent of the category V loans are labeled as a loss for at least 22 months, with 80 percent of these loans concentrated in one bank. This bank has in its portfolio a loan labeled as a loss for over five years.

24. The data also suggest that loan classification appears imperfect at individual banks, based on the criterion of number of days past due. A necessary condition for a well-functioning banking supervision is that loans are classified correctly in a timely manner. If loans are classified correctly, the changes in loan categories are stepwise, i.e. that a loan from category I (regular) will move to category II (30-89 days) first and not directly to category IV (180-359 days), although that may be sensible in rare cases when loan repayment suddenly becomes doubtful.8 A correct reclassification in accordance with the criterion of days past due occurred in 86 percent of the cases; for the rest a loan worsened by more than one category within a month, which is inconsistent with the classification schedule that downgrades loans with intervals of longer than 30 days. Other problematic cases include loans that were downgraded only after more than two months even though they should have been downgraded within a shorter interval.9 One in every five loans that have a category worse than “regular” seems to be misclassified by staying longer than appropriate in a given category. Among the four largest banks, the loan classification that would appear correct solely based on the days past-due criterion ranges from 20 to 96 percent.

D. Conclusions and Recommendations

25. This paper analyzes credit developments in the banking system based on loan information from São Tomé and Príncipe’s credit registry. The main conclusions from the analysis are that (i) the high credit concentration gives rise to vulnerabilities as banks depend on the performance of a few clients; (ii) mainly enterprises are responsible for the excessive levels of NPLs, while households’ contribution is small; and (iii) banks are reluctant to write off NPLs, which may be due to protracted or partial recovery of underlying collateral. Vintage analysis shows that default rates increased through 2015 before receding somewhat. Finally, transition matrices calculated for the entire sample illustrate that once a loan became non-performing, it tends to deteriorate further. Nevertheless, matrices for recent years show that more loans are recovered during early past-due periods; the high rate of transition from past-due to the loss category also suggests that loan classification practices are becoming more consistent.

26. The existence of a comprehensive credit registry is a major accomplishment for a small fragile state, and going forward, it can facilitate financial stability analysis. The scope of the information gathered in the credit registry is unusually wide for a small developing country. Indeed, it provides a wealth of data, encompassing many important characteristics of the bank-borrower relationship. In addition, issues with data quality appear relatively minor. The credit registry data can be used consistently for banking supervision purposes. In particular, it can be used to track the evolution of the loan quality for the entire system or individual banks and portfolios. In this, the objective is to spot emerging loan deterioration early on —such as a rising transition rate from category I to II—in order to take remedial measures before there is pronounced rise in NPLs. Also, the computed 12-month default rates can potentially be used to derive an estimate of expected loan loss, provided that information on loan recovery rates is gathered as well.

In light of the findings, the credit registry and loan classification practices may be improved further:

The credit registry can be further improved by:

  • making available more information on the borrowers, such as income, wealth, and gender;

  • collecting information on the collateral underlying a loan and its estimated value, which could be used for macroprudential purposes; and

  • classifying borrowers/loans according to these principal economic sectors (e.g. agriculture, manufacturing, construction trade, services etc.). The credit registry already includes information about the sector of the borrower, but the classification is so granular and not standardized, making it difficult to conduct an analysis.

The regulation and supervision can be enhanced by:

  • obliging banks to write-off non-performing loans from their balance sheets after a certain period (e.g., after two years in category V).

  • reclassifying restructured loans as performing only after certain period of compliant debt service (e.g., 12 months),10 instead of immediately.

  • continuing to strive for a timely and correct classification of loans so that such action is fully inconsistent with the BCSTP’s loan classification system.


  • Banco Central de São Tomé and Príncipe (2007): “Normas de Aplicação Permanente sobre Identificação de Classificação de Activos e Provisões”, NAP 34/2007.

    • Search Google Scholar
    • Export Citation
  • European Central Bank (2016): “Draft guidance to banks on non-performing loans”, September 2016.

  • Feng, D., C. Gourieroux and J. Jasiak (2008): “The ordered qualitative model for credit rating transitions”, Journal of Empirical Finance, Vol. 15(1): 111130.

    • Search Google Scholar
    • Export Citation
  • Hanson, S. and T. Schuermann (2004): “Estimating Probabilities of Default”, Federal Reserve Bank of New York Staff Report No. 190.


Prepared by Luiza Antoun de Almeida and Torsten Wezel.


The CRC covered 269 loans in 2011, 6,787 in 2013, and 10,864 in 2017.


As noted in the staff report, the authorities have adopted a NPL reduction strategy plan, but its implementation has been hampered by an inefficient judiciary system that makes recovering collateral difficult. While the government has sought to establishing arbitration centers to facilitate out-of-court settlements, these centers are not yet operational for lack of start-up funding. Meanwhile, the continued high credit risk has stalled new credit to the economy.


It is possible that the timing of NPL recognition and opportunistic behavior on the part of borrowers at a failing bank affected the default rate in 2015.


The unweighted variant of the 12-month default rate is called “cohort approach” which takes only the observed proportions form the beginning of the period to the end into account, whereas the weighted variant (“duration approach”) recognizes the time spent in the starting state to obtain the migration intensity (Hanson and Schuermann, 2004). Another variant of the latter takes account of multiple intra-year classification changes (e.g. from category I to III, and then from category III to II), but this more intricate weighted default rate is not implemented in this paper.


For 2012, a default rate of 5.7 percent after 48 months is compatible with the lifetime default rate of 7.1 percent shown in figure 4. When calculating the default rate after 66 months for the 2012 vintage (the highest span possible in the case that a loan was extended in January 2012 and observed until June 2017), the same default rate of 7.1 percent is obtained.


The maximum time horizon in 2015 and 2016 is less than the full two years and one year, respectively (limited to 18 and 6 months) because the data sample ends in June 2017.


Obviously, there may be other triggers for re-classifying a loan by more than would be appropriate based on the “days past-due” criterion that make a borrower unlikely to pay such as the personal situation (e.g. sudden loss of income) or non-compliance with the requirement to provide updated financial information to the bank.


For example, while it is possible for a loan to stay longer than two months in category II (30-89 days) because the borrower keeps paying but with a one-month lag, such a scenario is unlikely.


The guidance by the European Central Bank on the treatment of restructured loans that are to be re-classified from non-performing to performing status requires, inter alia, the completion of a “cure period” of one year and that the debtor’s behavior demonstrates that financial difficulties no longer exist, specifically that there is no past-due amount and that the borrower has a demonstrated an ability to comply with the post-forbearance conditions (European Central Bank, 2016).

Democratic Republic of São Tomé and Príncipe: Selected Issues
Author: International Monetary Fund. African Dept.