Graduate School
Master Degree Project in Economics
The Effect of Budget Support on Private Sector Credit in Malawi
Author: Magdalena Andersson Paul, 19830617-5665 Supervisor: Professor Dick Durevall
Abstract
In order to study the impact of budget support disbursement on credit to the private sector in Malawi, I use an error correction framework and apply it to monthly data. I first identify a
cointegrating vector including real Private Sector Credit, real GDP, the Interest Rate Spread and the real Treasury Bill rate. I then specify the model based on observations from 2002 to 2014, in which the error correction term is found statistically significant suggesting that the real Treasury bill rate have a significant negative impact on credit to the private sector. The results also show that there is a crowding out effect caused by government borrowing. Adding budget support to the error
correction models shows the significance of the cointegration vector, and find that has a statistically significant negative impact on credit to the private sector in Malawi.
2
Acknowledgements
This thesis would not have been possible without the encouragement, patience and support I have received through the process.
To Elof Hansson Foundation, for awarding me a travel grant which allowed me to travel to Malawi and interview key stakeholders on the ground. A sincere Thank You!
To Ann Veiderpass, for your mentorship along the way. It is easy to mentor students at their best but it is when they are at their worst it takes someone with a huge heart, incredible knowledge and the wisdom only life can bring, to bring someone out from the woods, even at a business school.
To Per-Åke Andersson, for the moral support and guidance in finishing the actual paper, and the many chats about life in Africa. Thank you!
To Dick Durevall, from econometric approach and discussions about Malawi and its economy to having the patience to see this paper to the end. Thank you for your continuous support!
To my husband David, for pushing and pulling, for endless discussions about the private sector in Malawi, for not losing faith and for being there, thank you!
Finally, to Mamma och Pappa, who will never read this paper, yet to whom I owe this work. You laid the foundation for it many years ago by giving me the confidence to go out into the world with open eyes and open mind. You taught me not to be discouraged by difficulties but always remember “the kite needs upwind to rise” - something particularly applicable in development economics. This paper is devoted to you – the Thank You I never got to say.
3
Table of Contents
Abstract ... 1
Acknowledgements ... 2
1. Introduction ... 4
2. Background ... 6
3. Literature Review ... 10
3.1. Economic Development and Credit to the Private Sector ... 10
3.2. Economic Development and Budget Support ... 13
4. Theoretical Framework ... 14
4.1. Determinants of Credit to the Private Sector ... 14
4.2. Determinants of Crowding Out and the Impact of Budget Support ... 14
5. Data and Descriptive Statistics ... 15
5.1. Descriptive Statistics ... 16
6. Empirical Framework ... 18
6.1. Co-Integration Analysis ... 20
6.1.1. Engle-Granger Cointegration Test ... 20
6.1.2. Johansen Test for Cointegration ... 20
6.2. The Error Correction Model ... 20
7. Empirical Analysis ... 21
7.1. Unit-root Testing - Results ... 21
7.2. Engle-Granger - Results ... 21
7.3. Johansen Test - Results ... 23
7.4. Error Correction Model – Results ... 24
8. Discussion ... 27
8.1. Results ... 27
8.2. Limitations... 28
8.3. Further Research ... 28
9. Conclusion ... 29
10. References ... 30
Appendix ... 32
4
1. Introduction
The seventeen Sustainable Development Goals set up by the United Nations have as their main objective to reduce poverty and enable long term sustainable development through food security, education, health and gender equality by a global partnership for development.1 To achieve this, economic growth in general and private sector growth in particular are fundamental, as a sustainable economy depends on production to provide employment and raise taxes for
investments in infrastructure and for redistribution. For private sector growth, access to credit is an essential financial requirement as short term working capital for operational needs as well as for long term investment. The demand for credit however, is not only limited to the private sector but the public sector also turns to the domestic market for financing. This aspect of demand for credit increases when other sources of government funding fall short. For many governments in poor countries budget support is a significant source of revenue aimed at supporting economic and social development. However, the conditionalities for continuance tend to make the assistance volatile over time, as a breach in agreements can lead to suspension of the support and a negative cash flow for the government. This increases the government’s demand for domestic credit which in turn increases the total demand for credit in the domestic market causing interest rates to rise and subsequently causes an increase in the cost of borrowing for private companies and restricts private sector development.
The environment for the private sector in Malawi has been extremely adverse over the last two decades. Faced with strict foreign exchange controls and high inflation rates as well as high nominal interest rates, foreign as well as local investments have been heavily inhibited. The nominal lending rates over the period 2000 to 2015 have been marked by high volatility and have fluctuated between a low of 6.28 percent in December 2010 and a peak of 37.91 percent in February 2013 implying significant fluctuations in the cost of borrowing for the private sector. This volatility is also recognised in the budget support disbursements data. For Malawi, budget support is a significant part of government revenue. For the budget year 2000/2001, total grants in which budget support grants are included, corresponded to 33 percent of the total revenue that year. It peaked in 2008/2009 at 43 percent only to drop to 13 percent for 2014/20152.
Economists have detected a link between budget support, interest rates, domestic borrowing and hence credit to the private sector. In 2001, the IMF suspended financial assistance to Malawi
1 https://sustainabledevelopment.un.org/?menu=1300
2 IMF article IV 2002, 2008 and 2015. Table 2. Malawi: Central Government Operations. (respective figures from the respective versions).
5 following failure to meet IMF programme targets. Following IMF’s actions and recommendations, bilateral donors also stopped budget support disbursements leaving the Government of Malawi with an increased budget deficit. The joint staff assessment by IMF/IDA concludes “The suspension of budgetary assistance by donors did not result in reduced overall expenditure but instead in increased domestic borrowing, which contributed to high interest rates and a rise in domestic debt” (IMF/IDA, 2003).
Previous research has focused on credit to the private sector, its determinants, and the impact of budget support separately. There is therefore a gap in the literature with regards to Malawi where, as far as I am aware, no research with this focus has been conducted. The working hypothesis is that there is a positive relationship between credit to the private sector and budget support
disbursements which implies that an increase in budget support corresponds to an increase in private sector credit while withdrawal of budget support will result in a decrease in credit. The chain of events starts with the budget process in which the Government allocates funds for the budget year to come. Budget support donors then agree to provide general budget support3, which is included in the financial management system, given certain conditions. A breach in the conditions for the support leads to suspension of disbursements. While the donors expect the suspension in budget support to lead to a reduction in government expenditure observations suggest that the government tends to seek alternative funding instead. By issuing treasury bills to the domestic market as well as direct borrowing from local banks, the government meets the shortfall in the short term. This increases government borrowing which causes interest rates to rise and crowds out the private sector. The result is that the credit to the private sector is reduced, implying an increase in the cost of working capital and a reduction in investment.
The aim of this thesis is to analyse the link described above and answer the question: How is credit to the private sector in Malawi affected by budget support disbursements?
I do this by testing the following three hypotheses:
1. Economic activity measured in real GDP has a significant positive relationship to Private Sector Credit while the cost of borrowing measured by the Interest Rate Spread, has a significant negative impact.
2. Government borrowing causes crowding out of Private Sector Credit.
3. Budget Support has a significant impact on credit to the private sector.
3 Budget support normally includes general budget support as well as project budget support however the latter is outside the scope for this paper.
6 I estimate an error correction model based on time series data from the Reserve Bank of Malawi, the IMF and the World Bank data bases respectively. The sample is based on 150 monthly observations from May 2002 to October 2014. Due to a limited number of observations for budget support, the regressions which include budget support are based on 76 observations, starting in May 2008 to October 2014.
I find that there is a statistical significant long-run relationship indicated by the estimated
cointegration vector which includes credit to the private sector, economic activity, measured in GDP, the cost of borrowing, measured in the Interest Rate Spread, and the real Treasury Bill rate. The estimated cointegration vector suggest that an increase in economic activity has a significant positive impact on private sector credit while an increase in the cost of borrowing as well as in the Treasury Bill rate, has a significant negative impact. The first difference results for Treasury Bill rate as well as for real Government Borrowing are also statistically significant which confirms the hypotheses that an increase in government borrowing has caused a crowding out effect on the private sector. Applying the model on the short time period, 2008 to 2014, I find that real Budget Support is statistically significant and negative which implies that suspension of budget support leads to an increase in demand for credit from the private sector.
This paper is structured accordingly: A section on Background describes the circumstances in Malawi and the economic and political context on which this paper is based. The Literature Review provides a summary of previous general literature on Credit to the Private Sector and the Impact of Budget Support respectively to be followed by a section on the respective areas with focus on Malawi. In the Theoretical Framework, I lay out the foundation on which my model is based and describe the model derived for the analysis. The Data section contains the data sources and descriptive statistics. The Empirical Framework presents the empirical method applied and the Empirical Analysis takes the reader through the econometric testing, from a preliminary graphical analysis, on to co-integration analysis to the final regressions. The main findings are examined in Discussion in the context of the Background and previous literature. It also discusses the limitations of the study and ideas for further research. Finally, the Conclusion summarizes the paper and comments on the implications of the results.
2. Background
In order to understand the mechanisms at play within the Malawi economy is important to be aware of the major political and economic events which have occurred during the time period of interest.
7 While political leadership and foreign exchange policy in itself is not the scope of this paper, they do play a role when trying to comprehend the political turbulence which affected both donors and economic indicators such as the inflation rate and demand for credit.
The period starts with Bakili Muluzi as the sitting president (2000) and a floating exchange rate regime with broad money as the intermediate target. In 2002 Malawi experienced a food shortage caused by a combination of poor harvest due to drought in 2001 and food reserve mismanagement by the National Food Reserve Agency. The mismanagement led to suspension of donor funds while suspected corruption was investigated. In 2004, Bingu wa Mutharika won the elections. Under his leadership the exchange rate was pegged to the US dollar in an attempt to keep inflation rates under control and interest rates down. In cooperation with donors he also introduced the Fertilizer Subsidy Programme which had a significant positive impact on agricultural production boosting the
economic growth in the country. During Mutharika’s second term, pressure from the IMF to devalue the currency and criticism of the leadership of the country, led to infected international relationships culminating in the British High Commissioner being asked to leave the country. Once again, this resulted in the suspension of aid from several donors following the IMF recommendations. The unexpected death of Bingu wa Mutharika in April 2012, led to a change in governance and vice president Joyce Banda took over the leadership and managed to improve the relationship with donors. The most significant policy change was to allow for a devaluation of the currency followed by an exchange rate regime change, from a fixed exchange rate to a floating exchange rate. The discovery of substantial fraud and theft of public means, also referred to as the “cashgate scandal” in 2013 was again a setback for the improved relationships and the investigation that followed
revealed that top leaders and politicians were involved. Subsequently budget support donors suspended their funding with immediate effect. In May 2014 Peter wa Mutharika, the younger brother of the late Bingu wa Mutharika, won the elections and is the sitting president at present.
8 Table 1: Time line Political Leadership, Exchange Rate Policy and Economics Events
Year President Foreign Exchange
Policy
Key Economic Events
1994 – 2004 Bakili Muluzi (UDF) Fixed exchange rate 2001/2002 – Food crisis
2004 – 2009 Bingu wa Mutharika (UDF/DPP)
Exchange rate pegged to the US$
2007 – 2011 - Fertilizer subsidy programme
2011/2012 – Foreign exchange and fuel crisis
2012 - 2014 Joyce Banda (DPP) Floating exchange rate (introduced May 2012)
Sep 2013 – “Cash gate” scandal
2014 - present Peter wa Mutharika (DPP) Floating exchange rate 2015 – Severe floods affecting the country
Source: http://africanelections.tripod.com/mw.html, and IMF Article IV papers.
The political events described above affected the private sector both directly and indirectly. Private firms are closely linked to the government budget in many areas; banks, insurance companies and pension funds, construction companies and medical suppliers are only to mention a few sectors, which are directly linked to the government through contracts and/or indirectly by dependency on fiscal as well as monetary policies such as the exchange rate policy. The private sector and public sector are also competitors in the demand for credit on the domestic financial market. As budget support is a significant source of revenue for the country and the annual budget, suspension of budget support creates a significant shortfall in the budget. The government can either reduce government spending or find alternative funding. While the suspension is aimed to target the administration of the country and/or the individuals responsible for such breaches in agreements, the rational choice for the decision makers is rarely what would make them unpopular among their voters, hence they seek alternative funding where possible and delay payments where possible.
From the recent history of Malawi as described above, a pattern occurs which enables a chain of events to be derived. The chain starts with the budget process in which the Government allocates funds for the budget year to come. Budget support donors agree to provide general budget
support4, which is included in the financial management system, given certain conditions. A political or economic event takes place, such as the food crisis in 2001 or the Cashgate scandal in 2013 which triggers a response by the donors and the breach in the conditions leads to suspension of
disbursements. The government must then decide between reducing government expenditure or
4 Budget support normally includes general budget support as well as project budget support however the latter is outside the scope for this paper.
9 seeking alternative funding. The literature suggests that the government tends to seek alternative funding, at least on a short-term basis. By issuing treasury bills to the domestic market as well as direct borrowing from local banks, the government meets the shortfall in the short term. Over time, the increased government borrowing causes interest rates to rise and crowds out the private sector.
The result is that credit to the private sector is reduced, implying an increase in the cost of working capital and a reduction in investment.
The purpose of Figure 1 below is to present a hypothetical and simplified chain of events and show how shocks in one sector affect other players. It is important to highlight that this chain of events takes the variables included in this paper into account and omits important factors which are beyond the scope of this paper. The figure presents the events according to observations during the work on this thesis suggesting that important factors can have been omitted, misinterpreted or accentuated.
occur divided onto the three players, Government of Malawi, Donors and the Private Sector.
10
Table 2: Hypothetical Chain of Events
Time Government of Malawi Donors Interest Rates Private Sector
t=1 Annual government budget is developed in agreement with development partners.
Conditions x,y,z are agreed upon and the budget is approved in Parliament.
Donors agree to provide general budget support given a number of conditions x,y,z.
t=2 Annual budget is enrolled; wages of public servants are set procurement contracts are signed etc.
General budget support in accordance with agreement is disbursed.
Contracts signed.
Projects enrolled.
Expectations for the budget year set on which companies decide their need for financing and investments.
t=3 Breach in agreement due to unexpected circumstances, ie:
- Food crisis (2001) - British High Commissioner
requested to leave the country (2011)
- Cash Gate Scandal (2013
General budget support is suspended.
t=4 Negative cash flow as a result of suspended budget support:
The government choose between:
- Postponement of payments/wages
- Increase domestic borrowing - give out Treasury Bills to finance shortfall
- Increase in taxes
Awaiting investigation.
- Postponement of payments leads to a negative cash flow for companies. Leading to increased borrowing.
- Increase in taxes affects companies and employed people
t=5 - Increased demand
for credit leads to an increase in interest rates.
Increase in interest rates leads to an increase in cost of borrowing which leads to reduced private sector demand for credit
t=6 When the cost of borrowing is
increased, the private sector reduces investments. Together with a negative cash flow due to postponed payments, private sector
development inhibited.
t=7 As a result of an unprofitable private sector, real tax revenue is reduced and which further aggravates the situation.
Source: Compiled by the author based on theory and interviews with key stakeholders within the Private Sector in Malawi.
3. Literature Review
Previous literature has been divided into research on credit to the private sector and the impact of budget support respectively.
3.1. Economic Development and Credit to the Private Sector
One of the early economists to develop a theory on development and credit was Joseph Schumpeter who in 1911 presented his arguments that financial intermediaries, such as banks and other financial institutions play an essential role for technical innovations and economic growth. King and Levine (1993) examines if this theory still holds in modern times when analysing the impact of financial
11 development on long term economic growth using the four indicators; liquid liabilities as share of GDP, deposit banks versus central banks importance when allocating domestic credit, credit (to non- financial firms) as a share of total credit and credit (to non-financial firms) as share of GDP. They find that there is a significant positive relationship between financial development and economic growth as well as the accumulation of physical capital and economic efficiency. They also find that financial development has a significantly positive impact on the following 10-30 years of economic
development and that higher financial development is positively correlated with future accumulation of capital and its efficient use.
Calderón and Liu (2002) tests the causality of the relationship between financial development and economic growth on data from 109 countries and confirms that financial development in general leads to economic development. They find that capital accumulation and technological innovation are catalysed by financial development. They also find the co-existence of Granger causality that goes from financial development to economic growth as well as from economic growth to financial development. This supports the theory that financial development is important for economic development and one part of this is to understand what drives demand for private sector credit in general and especially so in poor countries where economic growth is crucial for development.
Most studies on credit to the private sector include a measure for economic activity, often
represented by GDP, as well as variables to explain the cost of credit, such as the lending rate. The alternative cost in defined as the interest rate spread complements the lending rate in some studies and replaces it in other. In addition to this, in countries with limited credit available, government lending has been included in order to capture potential crowding out.
Calza et al (2010) applies the Vector Error Correction Model (VECM) framework when studying the demand for credit to the private sector in the euro area. In this study, more than sixty percent of loans to the private sector are to be considered long-term credit, hence have a maturity longer than five years. They find that there exists a long run relationship between credit and real GDP, real short- term and long-term interest rates. Real GDP is found to be positively related to credit while interest rates in the short as well as in the long run are found to have a negative relationship. Abouka and Egesa (2007) also find a long-term positive relationship between credit to the private sector and GDP per capita.
Brzoza-Brzezina (2005), estimate private sector credit in Poland, Hungary and the Czech Republic by using a VECM framework. They find that a decline in lending rates have a significant positive effect on loans as this reduces the total cost of lending, which enables companies to increase borrowing at
12 the same nominal cost. Abouka and Egesa (2007) model credit to the private sector in East Africa, using credit to the private sector as share of GDP as the dependent variable. The study includes GDP per capita, inflation, interest rate spread as well as infrastructure, urbanisation and bank credit to the government as explanatory variables. They find that the nominal lending rate has a significant and positive impact on private sector credit. The authors consider this finding counter intuitive as one would expect that an increase in interest results in an increased cost for the individual company which according the fundamental supply and demand theory would lead to a reduction in the demand for credit. The results are therefore interpreted as the influence of high inflation rates in some of the included countries.
Other variables included in models in previous literature are credit to government and the interest rate spread. Sacerdoti, 2007, finds that credit to government has a significant negative impact on private sector credit. As the government face a budget deficit it must seek financing in the domestic banking system resulting in an increase in interest rates and decrease in available funds for the private sector. The dependent variable used is credit to the government as part of GDP and while this ratio was low for a large number of HIPC countries5 Nigeria, Kenya and Zambia all exceed 18%
with the Gambia reaching as high as 31%. Abouka and Egesa (2007) use the interest rate spread as an indicator for financial sector liberalization in East Africa but does not find this variable significant in explaining the demand for credit.
There appears to be a gap in the literature when it comes to specific research on credit to the private sector in Malawi, however there are a few papers which are touching the subject. Chirwa and Mlachila (2004) looks at the commercial banking sector in Malawi with focus on interest rate spreads and financial reforms between 1990 and 2000. They use an alternative definition for interest spreads in the analysis and find that the regardless of definition, the spreads have increased from approximately 10 percent in the beginning of the period to close to 40 percent at the end. These findings indicate that the liberalization of the financial markets in Malawi resulted in a significant increase in the spread and that this has happened at the expense of the commercial bank
customers. Their statistical analysis also indicates that the high interest rate spreads are a result of high central bank discount rates.
A later paper by Chirwa, Kaluwa and Chirwa (2015), on competition and banking industry regulation in Malawi, addresses the issue of high market concentration in Malawi as a potential explanation for the high lending rates averaging between 27 percent and 46.6 percent between the years of 2005 and 2014. They find that the high margins on credit have been sustained by a combination of factors,
5 HIPC=Heavily-Indebted-Poor-Countries
13 among these lending to the government which is considered a comparatively safe and profitable client causing a crowding out effect on private sector credit.
3.2. Economic Development and Budget Support
Budget support is one of the tools employed by donors to assist poor countries. Molenaers,
Cepinskas and Jacobs (2010), describe the use of budget support as a long-term channel to increase the levels of human development, consolidate democratic institutions and improve socio-economic standards for people in low-income countries. Koeberle et al (2006) describe budget support as a financial instrument used by donors to assist countries with poverty reduction by targeting
improvements in financial management as well as the specific socio economics needs in the country at stake, for example within education and health. The support is designed in accordance with the specific strategic plans agreed between donors and recipients and is part of a country’s budget procedure and included in the financial management system.
As a mean of providing aid has budget support received increased attention lately as an alternative to traditional project support, Koeberle, S. et al (2006). Where project support suffers from low disbursements rates and short-term thinking, general budget support is considered a more strategic, medium term form of economic assistance. Budget support also provides a partnership-based alternative to traditional conditionality and enables improvements in mutual accountability between donors and recipients. Molenaers, Cepinskas and Jacobs (2010) find that budget support is
considered flexible and encourages ownership. It is also political and often directly linked to specific leadership and can be considered a symbol of trust between the receiving government and the donor organisation. As the assistance is part of a receiving government’s overall budget it emphasizes the importance of local ownership and use of country specific structures and budget processes.
While there are many positives related to budget support, Molenaer et al. (2010) highlight the weaknesses and the impact of suspension of support due to a breach in the agreement. They consider budget support as high risk due to its fungible nature where funds can easily be relocated within a budget or displaced and where this can trigger a breach in trust between donor and recipient. Koeberle, S. et al (2006) recognizes that budget support can have a disruptive role when the disbursements are unpredictable and volatile. Delays in disbursements or sudden surges can have a negative impact on the macroeconomic stability in the recipient country.
Nilsson (2004) analyses the volatility of budget support in a number of Sub-Saharan countries between 1994 and 2002, including Malawi and finds that there is significant volatility in budget
14 support over the time period. The authors suggest that it is the unpredicted volatility that is most harmful to the economy.
4. Theoretical Framework
As far as I am aware, no previous research with specific focus on credit to the private sector using budget support as an explanatory variable has been conducted. Modelling of private sector credit however, has been done frequently in Europe as well as Africa.
4.1. Determinants of Credit to the Private Sector
Calza et al (2010), examines the long-run demand for credit to the private sector in the Euro area and includes GDP and interest rates according to the following relationship presented below:
𝐿𝑂𝐴𝑁𝑆 = 𝛼 + 𝛽1∗ 𝐺𝐷𝑃 + 𝛽2∗ 𝑆𝑇 + 𝛽3∗ 𝐿𝑇 (1)
The variable 𝐿𝑂𝐴𝑁𝑆 denotes real loans to the private sector, 𝐺𝐷𝑃 denotes real GDP and 𝑆𝑇 and 𝐿𝑇 are the real interest rates on short-term and long-term basis respectively6. The general conclusion in the literature suggests that an increase in economic activity leads to an increase in private sector borrowing, hence 𝛽1> 0 and that an increase in the interest rate results in an increased cost of borrowing leading to a reduction in demand for private sector credit, 𝛽2< 0.
Adjusting the relationship above to the Malawi context, which involves low financial development reflected in no differentiation between long term and short-term interest rates, the long-term lending rate 𝐿𝑇, can be omitted and the model include a “cost of borrowing”-variable which gives me the following theoretical relationship:
C = α + β1∗ GDP + β2∗ CB (2)
In the equation above, C denotes real credit to the private sector in Malawi, GDP denotes real GDP in Malawi and 𝐶𝐵 denotes the cost of borrowing, which is defined as the real lending rate or the interest rate spread alternatively.
4.2. Determinants of Crowding Out and the Impact of Budget Support Credit to the government is found to be a determinant to private sector credit. Given a small financial sector and a limited supply of credit, an increase in government borrowing leads to an increase in interest rates and a reduction of credit available as well as an increased cost of lending for the private sector, hence a crowding out of private sector credit.
6 The notations using capital letters are taken directly from Calza et al (2010).
15 As per the discussion in Background, budget support disbursements (and the suspension of the same) appear to have both an indirect impact on private sector credit through the channel of increased government borrowing and potentially a direct impact as well.
In order to test if the findings in previous literature hold for Malawi, I include the variables of interest in the fundamental model above in order to establish if government borrowing has caused crowding out of private sector credit and determine the impact of budget support on credit to the private sector. Including government borrowing, the Treasury bill rate and budget support
disbursements into the model, gives me the following relationship:
C = α + β1∗ GDP + β2∗ CB + βi∗ Χ 𝑖 (3)
In the equation above, Χ 𝑖, denotes a vector of explanatory variables including real government borrowing, GB, the 91 days Treasury Bill rate, TBR, and real budget support disbursements, BS, respectively.
5. Data and Descriptive Statistics
The primary data used for the empirical analysis is collected from three main sources; the Reserve Bank of Malawi, the IMF and the World Bank Group.
The data set consists of monthly observations starting January 2002 to October 2014. The econometrical long-term analysis to allow for a long-run cointegrating vector is based on 150 observations starting in May 2002 while the short-term analysis which include budget support disbursements is based on 78 observations due to the lack of available data for budget support disbursements. All data is based on current local currency, thus Malawi Kwacha.
The dependent variable, Credit to the Private Sector and the explanatory variable Credit to Government are retrieved from the “Commercial Bank Assets and Liabilities” survey downloaded from the Reserve Bank of Malawi’s (RBM) online database7. Bank Rate, Deposit rate, Lending rate and Treasure Bill Rate (91 days) are retrieved from the data set “Interest Rates” while the variable Inflation comes from the data set “Inflation” and indicates overall inflation, also from the RBM online data base.
The data for Budget Support is collected from the International Monetary Fund (IMF) and the International Financial Statistics (IFS) online data base where it is defined under “Budgetary Central
7 www.rbm.mw
16 Government, Revenue” as Grants8. The variable for economic activity is approximated by
interpolated annual GDP collected from the World Bank Group9 online data base.
In addition to the variables above the Consumer Price Index (CPI) is used to calculate real values and the figures are retrieved from the IMF and the IFS online data base10. CPI was normalized and set to 100 on 1st January 2010.
5.1. Descriptive Statistics
The data set consists of 150 monthly observations starting in May 2002 to October 2014.
Table 3: Descriptive Statistics – Nominal values
Variable: Unit: Mean: Standard Dev: Min: Max: No. Obs:
Credit to the Private Sector (C) MWK billions 95.97 86.28 9.40 288.04 150
Gross Domestic Product (GDP) MWK billions 80.33 54.98 22.28 265.37 150
Credit to the Government (GB) MWK billions 27.65 19.55 3.332 97.23 150
Budget Support (BS) MWK billions 7.03 6.94 0.358 34.42 76
Interest Rate Spread (IRS) Percentage units
19.83 4.93 14.00 29.86 150
Deposit Rate (DR) Percent (%) 9.88 7.35 3.50 30.5 150
Lending Rate (LR) Percent (%) 29.72 10.92 17.75 52 150
Treasury Bill Rate 91 days (TBR) Percent (%) 19.75 11.20 5.66 45.5 150
Inflation Rate (IR) Percent (%) 13.57 7.39 6.28 37.91 150
Source: IMF, the World Bank, the Reserve Bank of Malawi
The dependent variable, Credit to the Private Sector is measured in Malawi Kwacha (MWK) billions, with a mean of MWK 95.97 billion. The figures are nominal values and the increase from MWK 9.40 billion (minimum) to MWK 288.04 billion (maximum) suggest that there has been a significant growth in credit in nominal terms, throughout the period, most likely driven by inflation and economic growth.
GDP is included as a proxy for economic activity and the data reveals that a significant growth in nominal terms has taken place with an increase in GDP from the minimum MKW22.28 billion to the maximum of MWK 265.37 billion. The original data was on annual basis and was interpolated to monthly figures to enable analysis.
Figure 2 shows the real figures of the variables plotted over time and suggest that also in real terms, credit to the private sector as well as government borrowing and GDP has seen a positive growth
8 http://data.imf.org/?sk=5dabaff2-c5ad-4d27-a175-1253419c02d1
9 http://data.worldbank.org/
10 http://data.imf.org/?sk=5dabaff2-c5ad-4d27-a175-1253419c02d1
17 during the time period. The data for budget support disbursements reveal a trend that payments seemed to be made on a quarterly or half year basis rather than evenly distributed on a monthly basis. They were therefore converted to indicating a three-month moving average. Despite the conversion the graph shows significant fluctuations during the time period with a maximum MWK 34.42 billion and a minimum of MWK 0.358 billion.
Figure 1: Private Sector Credit, GDP, Government Credit and Budget Support
Deposit rate, Lending rate and Treasure Bill Rate (91 days) are measured in percentage and are the monthly averages for a given time period. The Deposit rate, Lending rate and Treasure Bill Rate (91 days) have a mean of 9.88 percent, 29.27 percent and 19.75 percent respectively. Figure 3 shows the interest rates over time. It reveals high levels of inflation, starting from 2012, which corresponds to the devaluation of the currency and the change to a floating exchange rate policy. As the graph shows this increase had a significant impact on the real lending rate, defined as the nominal lending rate minus inflation. The interest rate spread, measured as the lending rate minus the deposit rate is therefore found to be an alternative proxy for the cost of borrowing since the change in exchange rate regime had a subtler impact on deposit rates than on inflation rates.
050100150200
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date Billions kwacha
real Private Sector Credit
406080100120140
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date Billions kwacha
real GDP
1020304050
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date Billions kwacha
real Government Credit
051015
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date
3 month moving average (Billions kwacha)
real Budget Support
Source: Authors calculations
18 Figure 2: Interest Rates
6. Empirical Framework
Different applications of error correction frameworks are employed in previous literature to capture cointegration properties. The specific models vary due to availability of data and country-specific context however the general approach, as described in the literature review, include explanatory variables for economic activity, most commonly measured as GDP, the cost of credit, measured as interest rates. In addition to this, credit to the government and the treasury bill rate are included separately to measure crowding out. My addition to the literature is to include budget support in the model.
Following previous literature, I develop an error correction model, able to capture co-integration properties, to analyse the relationship between Credit to the Private Sector and the explanatory variables:
C = α + β1∗ GDP + β2∗ CB + βi∗ Χ 𝑖 (4)
01020304050
2008m1 2012m1 2014m1 2006m1
2002m1
2002m1 2004m12004m1 2010m12010m1
Date
real Lending Rate nominal Lending Rate
real and nominal
Lending Rate
15202530
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date
Lending Rate minus Deposit Rate
Interest Rate Spread
-20 0204060
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date
real Treasury Bill Rate nominal Treasury Bill rate
real and nominal
Treasury Bill Rate (91 days)
010203040
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date
Inflation Rate
Source: Authors calculations
19 In the equation above, Χ 𝑖 denotes the Treasury Bill rate, Government borrowing and Budget
support respectively.
I wish to test the following hypotheses:
1. The impact of the fundamental determinants to credit to the private sector in Malawi, that is economic activity measure in real GDP and the cost of borrowing, are consistent with previous studies, hence:
𝛽1> 0 and 𝛽2< 0
2. Government borrowing cause crowding out of private sector credit, hence:
𝛽3< 0 or 𝛽4< 0
3. Budget support disbursement have a significant impact on credit to the private sector, hence:
𝛽5≠ 0
The variables are transformed to their natural logarithms of the real values with exception for the lending rates which continue to be presented in real terms and the following time series model is fitted to the data:
𝑙𝑜𝑔 (𝑐𝑝𝑖C𝑡
𝑡) = α0+ β1∗ 𝑙𝑜𝑔 (GDP𝑐𝑝𝑖𝑡
𝑡) + β2∗ 𝐶𝐵𝑡+ βi∗𝑐𝑝𝑖Χ 𝑖𝑡
𝑡+ 𝜀𝑡 (5)
Denoting the log of real values with lower case letters in italics we get the following equations:
𝑐𝑡 = 𝛼0+ 𝛽1∗ 𝑔𝑑𝑝𝑡+ 𝛽2∗ 𝑐𝑏𝑡+ βi∗ x 𝑖𝑡 (6)
The dependent variable, c𝑡, is the natural logarithm of real credit to the private sector where 𝑡 denotes time. 𝑔𝑑𝑝𝑡 is the natural logarithm of real GDP, 𝑐𝑏𝑡is the real cost of borrowing and Χ 𝑖𝑡 denotes the vector which includes the natural logarithm of real government borrowing, gb𝑡, the real treasury bill rate, 𝑡𝑏𝑟𝑡, and the natural logarithm of real budget support disbursements, 𝑏𝑠𝑡,and 𝜀𝑡
is the error term, all at a point in time, 𝑡.
Time series analysis is based on the assumption that the variables are stationary or that a combination of variables are cointegrated. The first step is to test for unit roots, which I do by applying the Augmented Dickey Fuller. The null hypothesis in the test is that a unit root exists, denoted I(1), i.e. the variable is integrated of order one. Based on specific variable characteristics a deterministic trend is included in the test where applicable.
20 6.1. Co-Integration Analysis
If the variables are non-stationary, there is still a possibility that a linear combination of them is stationary, hence that the variables are co-integrated and contains a long-run equilibrium
relationship. To test for co-integration the Engle-Granger test is performed followed by the Johansen test. The econometric theory and methodology follow the chapters on co-integration and error correction models in Enders, 2015.
6.1.1. Engle-Granger Cointegration Test
The Engle-Granger test is performed in three steps. The first step is to estimate the long-run equilibrium relationship. An Ordinary Least Square (OLS) regression is run, including the non- stationary variables. In the case of co-integration, the OLS regression results are “super-consistent”
indicated by high R values. The residuals are the estimated values of the deviation from the long-run relationship wherefore the second step is to save and plot the residuals over time. If the residuals move around a mean of zero, this suggests that the residuals are stationary and that a long-run co- integration relationship exists. The third step is to perform the Augmented Dickey Fuller test on the residuals using adjusted critical values. If the residuals are found stationary, one can conclude that a cointegrating vector exist between the variables. As there might be several variables that are found non-stationary in the initial testing procedure, different combinations are tested in order to assess the number of cointegrating vectors.
6.1.2. Johansen Test for Cointegration
The Johansen maximum likelihood test for cointegration provides a similar test to Engle-Granger with the difference that it allows for the existence of several cointegrated vectors. The cointegrating vector is interpreted as the long-run relationship between the variables included in vector. The null hypothesis in the test is that there are p cointegrating vectors.
6.2. The Error Correction Model
Based on the existence of one or more cointegrating vectors, one or more cointegration vectors are introduced in the model to capture the long-run relationship. The coefficient before the co-
integrating vector is interpreted as the speed of adjustment and indicates the time period it takes to get back to equilibrium after a shock in the system. In the case no cointegrating vector is identified or the error correction term is statistically insignificant, the model is reduced to a difference-in- difference model.
21
7. Empirical Analysis
7.1. Unit-root Testing - Results
Table 4 presents the Augmented Dickey Fuller test. Based on a graphical analysis of the real variables, the decision to include a trend and/or a constant has been made and the variables are tested with 2 lags.
Table 4: Augmented Dickey Fuller test results
*= 10% level of significance, **=5% level of significance, ***=1% level of significance
Note 2: Number of observations. # =Data only available from 2008, number of observations are therefore 75.
The results suggest that I cannot reject the null hypothesis of a unit root for Private Sector Credit, GDP, Loans, Interest Rate Spread and Treasury Bills however when testing the first differences they are all found to be stationary. For Government Credit and Budget Support the results show that I can reject the null hypothesis of a unit root for hence conclude these variables as stationary within the sample.
7.2. Engle-Granger - Results
The results from the Augmented Dickey Fuller test on the individual variables showed that Private Sector Credit, GDP, the Lending rate, the Interest Rate Spread and the Treasury Bill rate are non- stationary. Using the log of the real variables, the real lending rate, the interest rate spread and the real Treasury Bill rate, different combinations are tested for co-integrations. All combinations include
Augmented Dickey Fuller
Variable: Trend/Constant: ADF t-statistic: 5 % critical value:
c (t)
∆ c (t)
Constant -1.116
-11.659***
-2.887 -2.887 gdp (t)
∆ gdp (t)
Trend and Constant -2.930
-7.930***
-3.443 -3.444 gb (t)
∆ gb (t)
Constant -2.792*
-14.097***
-2.887 -2.887 bs (t)
∆ bs (t)
Constant -7.343***
-13.004***
-2.910 -2.911 tbr (t)
∆ tbr (t)
Trend and Constant -2.872
-10.301***
-3.443 -2.887 irs (t)
∆ irs (t)
Constant -0.662
-10.324***
-2.887 -2.887 lr (t)
∆ lr (t)
Constant -1.956
-10.625***
-2.887 -2.887
Number of observations 150
22 Private Sector Credit as this is our independent variable of interest. Table 3, presents the most significant combination which includes Private Sector Credit, GDP, the Interest Rate Spread and the Treasury Bill Rate.
Table 5: Engle-Granger – OLS regression
Dependent variable: Private Sector Credit (c (t)) Cointegrating Vector
Variable: Coefficient: t-value: (P>t):
gdp(t) 2.45 34.91
irs (t) -0.017 -7.51
tbr (t) -0.010 -6.04
Constant -5.983 -19.79
Number of observations 150
R-squared 0.9701
Source: Authors Calculations
The results from the initial OLS regressions on the potential co-integrated vectors show that all three combinations are super-consistent with R-square values of 0.9701, which indicate that there is cointegration.
Figure 3 plots the residuals from the regression. The plot suggests that the residuals are stationary though they deviate from the mean of zero towards the end of the period.
Figure 3: Engle-GrangerGraphical Analysis
Table 6 reports the results from the Augmented Dickey Fuller test on the residuals.
-.6-.4-.2 0.2.4
Residuals
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1
Date
Source: Authors Calculations
Private Sector Credit, GDP, IRS and TBR
Residuals
23 Table 6: Engle-Granger –Augmented Dickey Fuller on residuals
Engle-Granger Regression: Test Statistics: Observations:
Cointegrating Vector:
- c(t) - gdp (t) - irs (t) - tbr (t) 2 lags
-3.232** 147
Critical values: (n=3) 1% = -3.6215, 5% = -3.0048, 10% = -2.6744 Source: Authors calculations.
Since the test statistic in the Augmented Dickey Fuller test on the residuals is greater than
MacKinnon critical value, I can conclude that the residuals are stationary and co-integration exists.
The results show that the residuals from the cointegration vector is significant at a five percent level.
7.3. Johansen Test - Results
The following semi log-linear relationship is tested for co-integration where the cost of borrowing, 𝑐𝑏𝑡, is defined as the Interest Rate Spread in Regression and the alternative cost is defined by the real Treasury Bill rate.
Table 7: Johansen test for cointegration
Cointegrating Vector:
- c(t), gdp (t), irs (t) and tbr(t) Null hypothesis:
Rank = p
Eigen-value: Trace Statistics: 5 % crit value:
p = 0 - 40.84* 47.21
p ≤ 1 0.134 19.51 29.68
p ≤ 2 0.088 5.88 15.41
p ≤ 3 0.024 2.29 3.76
p ≤ 4 0.015 - -
Number of observations: 148
Source: Authors calculations.
In the cointegrating vector, the trace statistic is 40.84 is smaller than the critical value. However, it is close enough to the 10 percent level of significance so I could likely conclude that there is a co- integrating vector.
The estimates of the coefficients in the cointegrating vector are presented in Table 8.
24 Table 8: Cointegrating Vector
Estimation of Cointegrating Vector:
Variable: Coefficient: z-statistics: (P>Z):
c (t) 1 - -
gdp(t) -1.306 -3.66 0.000
irs (t) 0.037 3.24 0.001
tbr (t) 0.031 3.72 0.000
Constant 0.603 - -
Number of observations: 148
Source: Authors calculations.
In the estimations, the co-integrating equation is statistically significant as well as the coefficients for GDP, Interest Rate Spread and Treasury Bill Rate. As per the definition of the equation, Private Sector Credit is normalised to 1. GDP is estimated to -1.306, which is interpreted as a positive relationship suggesting that in the long run, a 1 percent increase in GDP corresponds to a 1.306 percent increase in credit to the private sector. The coefficient for the Interest Rate Spread is 0.037, which suggests a negative relationship and that a one unit increase in the Interest Rate Spread corresponds to a 0.037 percent reduction in private sector credit in the long-run respectively. The coefficient for the Treasury Bill rate is 0.031 implying that a one percent increase in the Treasury Bill rate leads to a 0.031 percent decrease in credit to the private sector.
7.4. Error Correction Model – Results
The results from the error correction model which includes the cointegration vector are presented in the Table 9. Different combinations of lags are tested for significance and the most significant results are presented. Robust standard errors are used in order to adjust for heteroscedasticity.
The basic error correction model is described by the following equation:
∆c𝑡= 𝛿0+ 𝛾𝐴2∗ (c𝑡−1− 1.306 ∗ 𝑔𝑑𝑝𝑡−1+ 0.037 ∗ irs𝑡−1+ 0.031 ∗ tbr𝑡−1) + 𝛿1∗ ∆c𝑡−1+ 𝛿2∗ ∆irs𝑡+ 𝜀𝑡 (7)
Specification 1 includes the cointegrating vector, (1 ∗ c𝑡−1− 1.306 ∗ 𝑔𝑑𝑝𝑡−1+ 0.037 ∗ irs𝑡−1+ 0.031 ∗ tbr𝑡−1) which is interpreted as the long-run relationship and where 𝛾𝐴 denotes the speed of adjustment parameter. It also includes the first difference of the interest rate spread. The first difference of real GDP is excluded in the specification as the monthly observations are interpolated hence the first differences are the function of the interpolation key, not observed economic activity.
25 In Specification 2, real Treasury Bills Rate (tbrt) is added to the model as per Specification 1 and in Specification 3, real Government Borrowing (gbt) is included.
Table 9: Error Correction Model Results (2002 - 2014)
Specification 1: Specification 2: Specification 3:
Variable: Coefficient: Standard
Error:
(robust):
Coefficient: Standard Error:
(robust):
Coefficient: Standard Error:
∆ c (t)
ect_1 (t-1) -0.054*** 0.017 -0.048*** 0.017 -0.050*** 0.017
∆ c (t-1) 0.024 0.075 0.029 0.073 0.049 0.077
∆ irs (t) 0.003 0.004 0.002 0.004 0.001 0.004
∆tbr (t) - - 0.004** 0.002 - -
∆gb (t-5) - - - - -0.088** 0.037
Constant -0.017* 0.010 -0.013** 0.010 -0.138** 0.010
Test:
Number of observations: 148 148 144
F-test (Prob>F) 3.60
(0.0152)
4.72 (0.0013) 4.22 (0.0029)
R-squared 0.063 0.013 0.089
Cointegration Vector A2: (ect_1) c GDP irs tbr Constant
1 -1.306*** 0.037*** 0.031*** 0.603
*=significant at 10%. **=significant at 5%. ***=significant at 1%.
Source: Authors Calculations
In all three specifications, the first lag of the error correction term, ect_1, which denotes the cointegrating vector, is statistically significant indicating a long-run relationship between Private Sector Credit and real GDP, the Interest Rate Spread and the Treasury Bill rate. A one percent increase in real GDP results in a 1.306 percent increase in credit to the private sector, a one
percentage increase in the interest rate spread results in a 0.037 percentage reduction in credit and a one percent increase in the treasury bill rate in a 0.031 percent. The speed of adjustment
parameter is estimated to -0.054, -0.048 and -0.50 in Specification 1, 2 and 3 respectively, which implicates the time of adjustment per month it takes to get back to the long-run equilibrium after a shock in the system. The first difference of the interest rate spread is statistically insignificant.
When adding the first difference of Treasury Bills Rate (tbr𝑡) in Specification 2 the coefficient is statistically significant and suggest that a one percent increase in the real Treasury Bill rate increase, results in a 0.004 percent increase in real Private Sector Credit.
The results when including real Government Borrowing (gb𝑡) in Specification 3, suggest that there is a delay in the effect of Government Borrowing indicated by a statistical significant fifth lag. The
26 results imply that a one percent increase in Government Borrowing leads to a 0.099 percent
reduction in Private Sector Credit five months later. The full results of 1-6 lags are included in the Appendix.
Given the limited number of budget support data, the second part of the analysis is based on the shorter time period (2008 – 2014) with a total of 76 observations. Specification 4 corresponds to Regression 1 above but based on the shorter time period. Specification 5 includes real Budget Support (three months moving average).
Table 10: Error Correction Model Results (2008 - 2014)
Specification 4: Specification 5:
Variable: Coefficient: Standard Error
(robust):
Coefficient: Standard Error (robust):
∆ c (t)
ect_1 (t-1) -0.039* 0.023 -0.040* 0.023
∆ c (t-1) 0.056 0.094 0.109 0.106
∆ irs (t) 0.005 0.004 0.004 0.003
∆ bs (t-2) - - -0.026** 0.013
Constant -0.009 0.011 -0.010 0.011
Test:
Number of observations: 76 73
F-test (Prob>F) 1.55 (0.2095) 2.36 (0.062)
R-squared 0.046 0.104
Cointegration Vector A2:
(ect_1)
Private Sector Credit (logREPrivSecCred)
GDP (logreGDP)
Interest Rate Spread (IRS)
Treasury Bill rate Constant
1 -1.306*** 0.037*** 0.031*** 0.603
*=significant at 10%.
**=significant at 5%.
***=significant at 1%.
**=significant at 5% ***=significant at 1%
The results from Specification 4 and 5, based on the shorter time period, confirms that the long-run relationship defined by the cointegration vector and reflected in the error correction term ect_1 is statistically significant and negative. The results in Specification 5 also reveals that real budget support is statistically significant at a five percent level and that a one percent increase in budget support leads to a 0.026 percent decrease in private sector credit two months later, as indicated by the second lag.
