Journal Issue
Share
Article

Kingdom of Swaziland: Selected Issues

Author(s):
International Monetary Fund. African Dept.
Published Date:
December 2015
Share
  • ShareShare
Show Summary Details

Macro-Financial Risks in Swaziland’s Banking Sector Associated with the Sacu Revenue Fall1

The fall in the revenues from the Southern African Customs Union (SACU) could have serious macroeconomic consequences. In addition to adverse impacts on fiscal and external balances, financial sector would also be affected. This section focuses on the likely impacts of prospective SACU revenue shock on the financial sector. Specifically, staff analysis demonstrates that the SACU revenue fall would increase sovereign spread and reduce private credit growth, while the government itself may face liquidity problem and run arrears (as experienced in the 2010–11 fiscal crisis). These developments would also affect borrowers’ payment capacity and eventually worsen bank’s asset quality, possibly resulting in further negative feedback to the economy.

A. Background

1. Swaziland’s banking system currently functions well with sufficient capital buffer (Table 7 of the Article IV staff report), though it is vulnerable to exogenous shocks.2 Capital ratios of commercial banks are above the regulatory minimum, and their profitability is high. The banking sector, however, is highly exposed to exogenous shocks, reflecting the economy’s vulnerability (e.g., a small open economy and heavy reliance on the SACU revenues). Furthermore, the high level of concentration on the balance sheets and the almost unitary business model of three of the banks pose additional structural weaknesses which would augment the negative impact of various shocks (in particular, the high volatility in the SACU revenues).

2. During the 2010–11 fiscal crisis (following the SACU revenue fall), nonperforming loans (NPL) increased, while commercial bank credit declined (Figure 1). The rise in NPLs started immediately with the outbreak of the 2010–11 fiscal crisis, while the loan-to-deposit ratio dropped and stayed low due to a decline in private sector credit growth during the crisis period. In 2012, the NPL increased sharply owing to economic contraction which led to defaults of private borrowers in the government-related sectors (e.g., construction) and the small and medium-sized enterprises (IMF, 2013). The change in NPLs exhibits positive correlation with the sovereign spread throughout the period, which remained high during the period when the sovereign risk premium was above 50 bps (in the bottom right chart in Figure 1).

Figure 1.Banking System Performance

3. The degree of structural vulnerability is high compared with other SACU countries (Text Figure 1). Although the NPL ratio declined to 6.4 percent in 2015Q2, the average NPL ratio in recent years was the highest among the SACU countries, and was one of the highest after the 2010–11 crisis among the Sub-Saharan African countries. This exhibits high sensitivity of asset quality in Swaziland’s banking sector to shocks. Bank-level data also show that the deterioration in asset quality since the 2010–11 fiscal crisis was wide-spread across all four banks. Some banks have much higher NPLs than others for their high exposures to construction and other small businesses which experienced financial distress during the crisis. These exposures are not only large but also volatile, which has caused high volatility in NPLs at both country- and individual bank-levels.

Text Figure 1.Level and Disparity in Non-Performing Loans

4. In view of the expected fall in the SACU revenues in coming years, this note explores the impact (stress) of the prospective fiscal shock on the stability of the banking system and its feedback effect on the real economy. The stress testing exercise is carried out in the following three stages:

  • (i) Exploring the macro-financial links from stress event (SACU revenue shock) to macroeconomic variables using a vector auto-regression (VAR) model,

  • (ii) Mapping the macroeconomic shocks to banks’ asset quality (i.e., NPL ratio) using cross-country panel regression, accounting for the feedback effect of banks’ credit risk on the real economy, and

  • (iii) Stress testing based on the calibrated NPL shock and the stress-testing model developed in the recent MCM TA (IMF, 2015), based on the standard macro stress-testing framework (Cihák, 2007).

B. Exploring Macro-Financial Linkages

5. The dynamic impact of the SACU shock is mapped into macroeconomic variables using the VAR model. The SACU shock would directly affect the real economy by worsening fiscal position, possibly crowding out private sector credit and increasing sovereign risk premium (and thus real interest rate), and could indirectly affect the economy through the deterioration of asset quality in the banking sector. Shocks to the banking sector may also feedback to the economic growth through bank’s deleveraging which reduces private investment. The following reduced form VAR is used:

where yt is a set of endogenous variables with one quarter lag. The quarterly series of following endogenous variables are included: (i) the growth in electricity consumption (in millions kilowatt hours, as proxy for real GDP) and in credit to private sector (quarter-on-quarter) and (ii) the de-trended sovereign spread (Swazi 3-month T-bill rate minus South African T-bill rate) and the current gross official reserves at the Central Bank of Swaziland which remove their historical trend components. 34 The deviation of actual SACU disbursements from the trend, computed by applying HP filter to the actual and projected SACU disbursement in the baseline framework, is included as an exogenous variable zt (see Figure 2).

Figure 2.Stress Test Calibration—Macroeconomic Assumptions

6. The VAR estimates and the impulse response functions show that a SACU revenue fall would adversely affect real economic growth by increasing sovereign risk premium (and thus interest rate as well as market confidence), which results in lower credit extension to the private sector. The VAR regression in Box 1 and the impulse response function in Figure 3 show that growth is sensitive to a reduction in private credits (with a lag of one quarter), while a fall in SACU revenues (below its trend) would significantly reduce the international reserves, leading to a hike in sovereign spread (above its trend) and a reduction in private credit growth. A large negative coefficient of international reserve on the hike of sovereign spread suggests high elasticity of Swaziland’s spread in response to the SACU revenue shock. The impulse response of sovereign spread to international reserve shock demonstrates that the risk premium would be peaked with 4 quarter lag after the SACU revenue shock, and the risk premium remain elevated over the medium-term. This would have immediate impact on growth and credit intermediation through direct channels (by worsening market confidence which could potentially trigger short-term outflow of capital from Swaziland banking sector, affecting the liquidity situation) as well as indirect channels (by increasing real interest rate).

Figure 3.VAR Analysis — Impulse Response Functions (without Feedback Effect)

Source: IMF staff calculation

7. Based on the VAR estimates, output growth could be as low as -2 percent in 2016–17. Based on the estimates, the evolutions of macroeconomic variables for 2015-2018 are explored under two difference scenarios. The benchmark scenario assumes that the economy will experience the SACU shock as same magnitude as projected in the staff report (Scenario 1). An alternative scenario (Scenario 2) puts prolonged shock in SACU revenue which persistently stays lower than the historical trend in SACU disbursement. The charts in Figure 2 highlight a large and immediate spike in sovereign spread and sharp slowdown in growth due to sluggish credit intermediation, resulting in annual growth of electricity consumption to about -2 percent during the shock period.5

C. Mapping the Macroeconomic Shocks to Bank’s Asset Quality

8. The sensitivity of loan quality to the change in macroeconomic shock is assessed using a cross-country satellite panel model. This regression estimates the transmission of macroeconomic shocks to the financial sector. Given the insufficient length of time series data for Swaziland, the fixed effect estimation based on the panel data of the CMA countries (in our case, Lesotho, Swaziland, and Namibia) is used:6

where ΔNPLit stands for the change in aggregate NPL ratios for country i in year t, and ΔMACROi,t includes the changes in sovereign spread and SACU revenues.7 A lag of private sector credit growth variable is also added. To capture an immediate and sharp rise in NPL ratio in Swaziland (see the bottom chart in Figure 1), a crisis dummy (Crisisi,t), which indicates the period when the NPL ratio grows by more than 30 percent, is included. The sample covers the period of 2009-2014 in quarterly frequency. To account for country-specific and period-specific factors, country dummies (μi) and quarter dummies (λt) are included. α1 capture the average effect of each macroeconomic variables on the growth in NPLs across three countries.

9. The estimates of the cross-country and bank-level satellite models suggest that the NPL ratios would rise following the SACU revenue fall due to higher sovereign risk premium, slower growth, and the likelihood for government arrears for suppliers (Box 2). In the cross-country satellite panel regression, the coefficients had the expected signs overall. A spike in the country’s spread by 100 bps beyond the historical trend has immediate impact in higher NPL ratio by 10.8 percent. Private sector credit growth also appears to increase the NPL ratio. Based on the previous crisis episode, this result may reflect an increase in unsecured lending and/or lumpy credit extension to specific borrowers (including construction and small and medium-sized enterprises), although the effect is not statistically significant. We also run the satellite regression using the NPL data for four individual banks in Swaziland. This regression could also include the quarterly electricity consumption growth as one of the explanatory variables (see Box Table 3). The result using bank-level data also confirms that higher sovereign risk premium (which is associated with government’s liquidity problem and an event of government domestic arrears during the fiscal crisis) and stagnant real economy activities (which is captured by the electricity consumption growth) would contribute to higher NPL ratios as observed in the past for Swaziland.

10. The feedback effect between the real economy and banking sector is also incorporated into the VAR regression which confirms the strong macro-financial linkages in Swaziland (Figure 4). While lower growth significantly increases NPLs mainly by weakening borrowers’ solvency to repay debt, the deterioration of bank’s asset quality could also affect the real economy through credit supply channel (see Klein (2013)). We use the same VAR specification as in Eq. (1) with the change in aggregate NPL ratio as an additional endogenous variable to capture this feedback effect. The impulse response function in Figure 4 confirms that an increase in NPLs leads to a temporary reduction in private credit growth and contraction in economic growth. The contraction in the private economy and deterioration in fiscal situation would worsen the payment capacity of borrowers in government-related sector (e.g., construction) and small and medium-sized enterprises, partly due to the government’s arrears to suppliers and vendors as observed after the 2010–11 fiscal crisis period, which then contribute to higher NPL ratio.

Figure 4.VAR Analysis — Impulse Response Functions (with Feedback Effect)

Source: IMF staff calculation.

Note: The impulse resonse is estimated based on the VAR model as specified in Eq. (1) with the change in NPL ratio and crisis dummy additionally controlled (both variables are defined in Eq. (2).

11. Finally, the evolution of NPL ratios is predicted based on the estimates in the VAR model as specified in Eq. (1) (which incorporated the feedback effect). On the basis of future SACU revenue streams as assumed in scenario 1 and 2, the evolution of NPL ratio is projected as shown in the last chart in Figure 2. When prolonged SACU revenue shock is assumed, the NPL ratio is projected to rise rapidly when the SACU crisis materializes in 2016 which would be peaked in the same year as similarly found by Hardy and Schmieder (2013) from past crisis episodes. Based on the calibrated credit shock, the next section examines the adverse impact of the NPL shock on the capital adequacy ratio (CAR) when the NPL ratio is peaked in each scenario.

D. Stress Testing Exercise

12. A stress test considers the impacts of the materialization of credit risk combined with interest rate risk and concentration risks on the capital adequacy ratio under three scenarios. Table 1 provides a detailed description of each scenario. Besides the credit risk estimated by the VAR model, we incorporate interest rate risk and concentration risk in alternative scenarios to conduct stress tests. The hurdle rate used in the stress test is summarized as follows:

  • Scenario 1: Only a credit shock (an increase in NPLs) as calibrated from the benchmark model (assuming SACU disbursements projected in the staff report) is applied.

  • Scenario 2: Larger credit shock due to more prolonged SACU shock is applied. Moreover, the interest rate shock (an increase in interest rates by 10 percentage points as assumed in IMF (2015)) is added as an increase in real interest rates could potentially affect the bank’s net interest income due to the re-pricing of interest earning assets and interest bearing liabilities.

  • Scenario 3: A larger credit shock is applied for vulnerable banks (whose NPLs spiked up much larger than others) where the level of NPL shock is set at their historical maximum during the sample period. Moreover, the loss-given-default for the largest exposure is set at 100 percent only in scenario 3. For the three South African subsidiary banks, the counterparty risk of largest exposure is assumed to be partly taken by South African parent banks via off-balance risk sharing agreements. The analysis assumes that the losses for subsidiaries will be capped at the 50 percent of the regulatory capital in the case of a default of largest exposure. 8

Table 1.Kingdom of Swaziland Stress-Testing Scenario
Scenario 1. Decline in SACU revenues
Expected impact on the economyExpected impact on banksExpected indirect impact
Lower SACU revenues worsen fiscal position and external balance, that causes:
  • - Reduced fiscal spending

  • - Increase in government’s debt, arrears, and contingent liabilities

  • - Higher risk premium

  • - Subdued economic activity

Higher credit risk (an increase in NPLs)The reduction in lending to the private sector which could trigger further deceleration in economic activities and credit intermediation
Scenario 2: Prolonged decline in SACU revenues + Increase in Real Interest Rates
Expected impact on the economyExpected impact on banksExpected indirect impact
Prolonged decline in SACU revenues worsen fiscal and external positions, that causes:
  • - Reduced fiscal spending

  • - Increase in government’s debt, arrears, and contingent liabilities

  • - Higher risk premium Subdued economic activity

Higher credit risk (an increase in NPLs; effect stronger than scenario 1)

Higher interest rate risk (through changes in spread between interest earning assets and interest bearing liabilities)
An increase in real interest rate and the sharper reduction in lending to the private sector which could trigger further deceleration in economic activities and credit intermediation (effect stronger than in scenario 1)
Scenario 3: Prolonged decline in SACU revenues + Concentration risk (a default of major borrowers)
Expected impact on the economyExpected impact on banksExpected indirect impact
Prolonged decline in SACU revenues worsen fiscal and external positions, that causes:
  • - Reduced fiscal spending

  • - Increase in government’s debt, arrears, and contingent liabilities

  • - Higher risk premium

  • - Subdued economic activity

  • - Cut in subsidy to SOEs

Higher credit risk (an increase in NPLs; effect stronger than scenario 1 and 2 due to a default of large exposures)

Higher concentration risk (a default of the single largest exposure or other major borrowers)
An increase in real interest rate and the sharper reduction in lending to the private sector which could trigger further deceleration in economic activities and credit intermediation (effect much stronger than in scenario 1 and 2)

13. The stress test is performed with a balance sheet approach based on a modified Čihák framework to examine the CAR relative to the regulatory requirement under the Basel I. The bank-level balance sheet and income statement data as of end-December 2014 are used for this assessment. Currently, all commercial banks in Swaziland need to comply with the Basel I capital requirement which requires banks to hold at least 4 percent Tier 1 capital and 8 percent total regulatory capital compared with the risk weighted asset (RWA). As the case of most Sub-Saharan African countries, the CARs of all four commercial banks in Swaziland comfortably exceed the regulatory minimum at end-December 2014.

14. The results show that the SACU shock could depress the asset quality, especially when the NPL shock is combined with concentration risk, though the shocks would be generally absorbed by the current large capital buffer. The stress-testing results for the entire banking system under three different scenarios (summarized in Text Table 1) show the reduction in the initial capital ratios due to three shocks (NPL shock, interest rate shock, and default of largest exposure). Under scenario 1, the deterioration in loan quality caused by the SACU shock (under the baseline) alone is estimated to create losses of 1.6 percent of the Tier 1 regulatory capital, while the capital would stay well above the regulatory requirement.

Text Table 1.Stress Tests: Shocks to Bank’s Capital (In percent)
Entire banking system
Initial Tier 1 capital-to-RWA ratio19.1%
Substractions due to shocks:
1.Aggregate NPL shock
Scenario 1−1.6%
Scenario 2−3.3%
Scenario 3−4.4%
2.P/L effect of interest rate shock (scenario 2 and 3)0.2%
3.Default of largest exposure (Scenario 3)−7.1%
Tier 1 capital-to-RWAratio (under stress)
Scenario 117.4%
Scenario 215.9%
Scenario 37.6%
Source: IMF staff calculation.Note: The NPL shock assumes a jump in NPLs from the current level to the peak NPL level calibrated from the VAR model as shown in Figure 1. Scenario 1 and 2 apply a uniform rise in NPLs for all banks, while scenario 3 puts larger credit shock for vulnerable banks which experienced larger spikes in NPLs in the past.
Source: IMF staff calculation.Note: The NPL shock assumes a jump in NPLs from the current level to the peak NPL level calibrated from the VAR model as shown in Figure 1. Scenario 1 and 2 apply a uniform rise in NPLs for all banks, while scenario 3 puts larger credit shock for vulnerable banks which experienced larger spikes in NPLs in the past.

15. The capital adequacy ratios will decline further if losses due to the default of the largest exposures are added. Under scenario 2, the prolonged SACU shock would trigger a sharper rise in NPLs and create losses of 3.3 percent of regulatory capital. The effect on the net-interest margin appears to be small but positive as the changes in interest rates are passed into both risk-earning assets and risk-bearing liabilities, leaving the net interest margin close to unchanged with marginally positive impact on the capital adequacy ratio. With the severe shock under scenario 3, the credit shock would create slightly larger losses in capital, driven by a sharper rise in NPLs for vulnerable banks. The concentration risk is estimated to be sizable, demonstrating the losses of 7.1 percent of regulatory capital should the largest exposures experience defaults due to economic contraction during the SACU crisis. This extreme shock scenario would significantly depress the capital adequacy ratio down to 7.6 percent.

16. The stress testing result demonstrates that the banking sector as a whole could still be compliant with the Basel I’s capital adequacy requirement but with large variability across individual banks. Under scenarios 2 and 3, the maximum impact of underlying shocks on the capital adequacy ratio is estimated to be significantly larger than the average impact. Although the capital adequacy ratio for the entire banking system could still be compliant with the Base I requirement, it will likely affect its financing intermediation function. Under the most extreme case, the capital adequacy ratio could potentially be negative. This points the importance of strong bank supervision and regulation.

Text Figure 2.Stress Tests: Distribution of Shocks across Individual Banks (Percent)

Source: IMF staff calculation

Note: Sample covers four individual banks in Swaziland.

E. Conclusion

17. The macro-financial analysis reveals that the prospective SACU revenue shock could trigger a rise in NPL ratios through the emergence of the government arrears, higher real interest rate (with higher risk premium), and sluggish private credit growth, which could put banking system stability at risk. The VAR regression reveals that a rise in risk premium, caused by lower SACU revenues, could lead to higher real interest rate and subdued economic activities, which also trigger a rise in NPL ratios as shown by the satellite regressions. These results suggest that, not only to ensure fiscal and external sustainability, but also to safeguard the financial sector stability, fiscal consolidation would be critical to mitigate the adverse impacts of the SACU revenue shock. Strong policy response with fiscal consolidation would help contain a rise in risk premium and mitigate the crowding-out of private credits, while allowing the government to avoid possible cash management problem.

18. While the banking system currently shows high capital ratios, the SACU revenue shock could worsen the capital adequacy of the whole Swazi banking sector close to the regulatory limit, particularly if the SACU shock should be accompanied by increases in interest rate and default of large borrowers. Banks seem to be sufficiently capitalized to withstand a SACU revenue fall alone, but with more adverse scenarios (with interest rate and concentration risks), borrowers’ payment capacity could be eroded, and the banks’ asset quality would worsen. Therefore, to address the expected SACU revenue shock, in addition to a strong policy response with fiscal consolidation, it is essential to strengthen supervision for financial sector, as recommended by the IMF (2015), and also to mitigate the risks to banks stemming from the high concentration risk.

Box 1.VAR Estimate on the Impact of SACU Shock on the Economy

The following table shows estimates of the VAR regression as specified in Eq. (1). We run the model using macroeconomic series for Swaziland (2009Q3-2014Q4) to analyze how the exogenous shock of the SACU revenue fall will endogenously affect growth, credit, real interest rate, sovereign spread, and gross international reserves (GOR). The preferred specification includes growth rate of electricity consumption (Δelectricity; proxy for quarterly GDP growth due to data limitation), private sector credit growth (Δprivate credit), the detrended sovereign spread (above South African T-bill rate; Spread^) and gross international reserve (GOR^) which remove the respective historical trends as endogenous variables, all are lagged by one quarter. It also includes the detrended SACU disbursements which removes its historical trend (SACU^) as an exogenous variable.

The result below shows significant positive correlation between SACU disbursement and the gross international reserve, and significant negative correlation between GOR and sovereign spread. A large negative coefficient of international reserve on the hike of sovereign spread suggests high elasticity of Swaziland’s spread in response to the SACU revenue shock. This VAR estimate implies that a fall in SACU (below its trend) would significantly reduce the international reserves, leading to a hike in sovereign spread (above its trend).

Box Table 1.Macroeconomic Impacts of SACU Shock
ΔelectricityΔprivate creditSpread^GOR^
Endogenous variable
L.Δelectricity−0.521718−0.019287−0.8408−0.078058
(0.19551)(0.18743)(0.82942)(0.16792)
[-2.66855][-0.10290][-1.01373][-0.46486]
L.Δprivate credit0.405596−0.0721090.6141510.539687
(0.26616)(0.25516)(1.12915)(0.22860)
[ 1.52390][-0.28260][ 0.54391][ 2.36086]
Spread^0.0429−0.0283260.6306480.054652
(0.03544)(0.03397)(0.15033)(0.03044)
[ 1.21066][-0.83381][ 4.19503][ 1.79570]
L.GOR^0.132894−0.047377−0.9977180.831425
(0.09583)(0.09187)(0.40654)(0.08230)
[ 1.38682][-0.51571][-2.45419][ 10.1019]
Constant−0.0079910.031063−0.018394−0.018291
(0.01504)(0.01442)(0.06379)(0.01291)
[-0.53146][-0.28835][-1.41632]
Exogenous variable
SACU^−0.009716−0.0019710.1123120.162097
(0.04219)(0.04044)(0.17898)(0.03623)
[-0.23030][-0.04874][ 0.62752][ 4.47359]
R-squared0.432840.0581470.7940760.941532
Sum sq. resids0.0547040.0502770.9845620.040354
Mean dependent0.0008770.02870.03692−0.024326
S.D. dependent0.0677720.0504180.4771530.18129
Sources: IMF staff calculation, Central Bank of Swaziland.
Sources: IMF staff calculation, Central Bank of Swaziland.

To examine the channel how the hike in risk premium would affect the real economy, we experiment with the specification with the change in real interest rate (Δreal interest rate). The impulse response functions in Figure 3 demonstrates that an increase in sovereign spread (above the historical trend), caused by negative shock in gross international reserves, tend to last longer and increase real interest rate. This then leads to a reduction in private credit growth and results in lower consumption growth.

Box 2.Satellite Model: Estimate on the Macroeconomic Shock on Credit Quality

The following table shows estimates of the fixed effect regression as specified in Eq. (2). The preferred specification includes the deviation of each country’s sovereign spread (above South African T-bill), private sector credit growth, and the crisis dummy (which is one if the country-quarter observation experienced a spike in NPL ratios by more than 30 percent points). The crisis dummy well captures a spike in the NPL ratio at the onset of the fiscal crisis.

In the satellite cross-country panel regression, the coefficients of sovereign spread is significantly positive, highlighting the spread is a critical factor in determining the asset quality. The marginal effect is also large, showing that a spike in the country’s spread by 10 bps beyond the trend increases the NPL ratio by about 1.1 percent.

Box Table 2.Panel Satellite Model: Macroeconomic Impacts of NPL Ratio
Δ NPL
Spread^0.108***
(0.011)
L.Δ Credit to private sector0.123
(0.134)
SACU^0.016
(0.042)
Crisis dummy0.549**
(0.080)
Constant−0.035*
(0.009)
Observations63
Number of groups3
R-square (within)0.625
R-squared (between)0.858
Country and quarter fixed effectY
Sources: IMF staff calculation, Central Bank of Swaziland.Note: Dependent variable is the percentage point change in NPL ratio. Spread^ and SACU^ are the detrended sovereign spread (above South African T-bill rate) and SACU disbursement which remove the respective historical trends. The first-differenced credit to private sector indicates their growth rates. “L” denotes the one period lag of the variable. Robust standard errors are reported. Hasuman test rejects the null hypothesis that random effect is more consistent, supporting our adoption of fixed effect model.Significance level: *significant at 10 percent; ** significant at 5 percent, *** significant at 1 percent.
Sources: IMF staff calculation, Central Bank of Swaziland.Note: Dependent variable is the percentage point change in NPL ratio. Spread^ and SACU^ are the detrended sovereign spread (above South African T-bill rate) and SACU disbursement which remove the respective historical trends. The first-differenced credit to private sector indicates their growth rates. “L” denotes the one period lag of the variable. Robust standard errors are reported. Hasuman test rejects the null hypothesis that random effect is more consistent, supporting our adoption of fixed effect model.Significance level: *significant at 10 percent; ** significant at 5 percent, *** significant at 1 percent.

The financial stability unit of CBS has also been compiling the balance sheet and income statement of four individual banks. We run the satellite regression using the bank-level panel data of NPLs covering 2009–2014 in quarterly frequency. The bank-level satellite regression includes same macroeconomic variables as the cross-country regression, but it also allows us to control for the electricity consumption growth rate in Swaziland.

As summarized in Box Table 3 below, an increase in sovereign risk premium and electricity consumption significantly increases the NPL ratio at the individual-bank level. This supports the evidence that the risk premium is a key determinant of the Swazi bank’s asset quality. It also demonstrates that the growth rate in the real economy matters for the change in NPLs.

Box Table 3.Bank-level Satellite Model. Macroeconomic Impacts of NPL Ratio
Δ NPL
Spread^0.009**
(0.002)
Δ electricity consumption−0.695***
(0.093)
L.Δ Credit to private sector0.371
(0.571)
SACU^−0.002
(0.051)
Crisis dummy0.306**
(0.082)
Constant−0.079
(0.034)
Observations86
Number of groups4
R-square (within)0.271
R-squared (between)0.553
Bank and quarter fixed effectY
Sources: IMF staff calculation, Central Bank of Swaziland.Significance level: *significant at 10 percent; ** significant at 5 percent, *** significant at 1 percent.
Sources: IMF staff calculation, Central Bank of Swaziland.Significance level: *significant at 10 percent; ** significant at 5 percent, *** significant at 1 percent.
References

Prepared by Manabu Nose (AFR).

Swaziland’s banking system is predominantly foreign-owned. Three commercial banks (which comprise 85 percent of total banking system asset) are subsidiaries of South African Banks and operate similar business models. The remaining 15 percent of asset is owned by a state-owned bank which plays a development banking role.

The deviations of international reserve, sovereign spread, and SACU payments are computed as the gap of actual gross official reserve, sovereign spread, and SACU disbursements from the trend component. The trend component is computed by applying the Hodrick-Prescott (HP) filter to the quarterly series with the smoothing parameter of 1,600.

In the absence of quarterly GDP data, quarterly electricity delivered and consumed by households and industries is the best available quarterly series in conducting the macro-financial analysis, which appears to track short-term fluctuations in growth well. See recent literature (Henderson, Storeygard, and Weil, 2012) for the application of night light (and electricity consumption if available) data as an alternative measure of growth in Sub-Saharan African countries where the quality of national account data is weaker than developed and emerging countries.

The VAR model is backward-looking, which estimates the macro-financial linkage based on the historical data and does not take into account future policy reactions to prospective SACU revenue fall. The analysis uses higher frequency data (quarterly) to increase the number of observations in the regression and employs a proxy measure of quarterly growth (electricity consumption) which is not a perfect measure for data availability. For these reasons, the calibrated series of growth and private sector credit based on the VAR model show larger volatilities than the annual projection in the staff report, though the overall trends are consistent between two frameworks.

Botswana is excluded from the sample as Botswana’s quarterly NPL data is not available from the FSI database.

In the panel regression, electricity consumption growth cannot be included as quarterly series are not available for Lesotho and Namibia.

The risk sharing agreement stipulates that the parent bank will recapitalize the subsidiary for losses exceeding 25 percent of the regulatory capital of the subsidiary. Given some uncertainty o the ability and willingness of the South African parent banks to honor this agreement, we cap the losses at 50 percent of a bank’s regulatory capital rather than 25 percent as agreed in the risk sharing document.

Other Resources Citing This Publication