11
What causes household debt to increase in South Africa? Christelle Meniago a, b, 1 , Janine Mukuddem-Petersen a, c, , 2 , Mark A. Petersen a, d, 3 , Itumeleng P. Mongale a, b, 4 a Faculty of Commerce and Administration, North-West University, Makeng Campus (NWU-MC), Private Bag X2046, Mmabatho, 2735, South Africa b Department of Economics, South Africa c Graduate School of Business and Government Leadership (GSBGL), South Africa d Faculty of Commerce and Administration, South Africa abstract article info Article history: Accepted 9 April 2013 JEL classication: C32 H31 C51 C52 Keywords: (Global) Financial crisis Household indebtedness Cointegration Vector Error Correction Model South Africa The 20072008 US subprime mortgage crisis evolved into a nancial crisis that negatively affected many economies in the world and was afterwards widely referred to as the global nancial crisis. Since the beginning of this nancial crisis of 20082009, South Africa experienced a signicant increase in its household debt to income ratio. In the main, this paper investigates the prominent factors contributing to the rise in the level of household debt in South Africa. Specically, we construct a model for South African household debt through the application of a Vector Error Correction Model (VECM). We employ quarterly time series data throughout the timeline 1985 Q1 to 2012 Q1 and all the econometric tests are analyzed using the statistical software package EViews 7. Our results conrmed the existence of a long run cointegrating relationship between household debt and other macroeconomic deter- minants. Increasing household debt was found to be signicantly affected by positive changes in consumer price index, gross domestic product and household consumption. Also, house prices and household savings were found to positively contribute to a rise in household debt but this relationship was found to be statistically insignicant. Alternatively, household borrowing was found to be signicantly and insignicantly affected by negative changes in income and prime rate, respectively. Ultimately, the existence of a long run cointegrated relationship enabled us to build an error correction model for household debt which will facilitate future forecasting. © 2013 Elsevier B.V. All rights reserved. 1. Introduction An economy with a good nancial situation is associated with low debt levels in the household sector. To manage low debt levels among households, nancial institutions which act as the primary source of credit in several countries and specically in South Africa must be very prescriptive and discerning when it comes to the provision of loans to customers. With escalating debt, the household sector may run the risk of being too exposed to several adverse surprises like un- employment shocks, asset price shocks and shocks from income, just to name a few. In this regard, nancial crises have historically been perceived to emit devastating shocks to vulnerable economies. In lieu of this, it is unfortunate to observe that South Africa records a very high debt level. According to the SARB (2012) quarterly bulle- tin (June 2012), the ratio of household debt to disposable income decreased slowly to 74.7% in the rst quarter of 2012 compared to 74.8% in the fourth quarter of 2011. Though this is seen as a slow de- crease, this gure still remains very high and such high levels of household debt may have adverse effects on the economy. Without an effective investigation to discover what might cause this high level of debt among households in South Africa, the household sector might still remain very susceptible to shocks. On a global scale, South Africa records low debt levels compared to the most inuential economies in the world (see Fig. 1). The ratio of household debt to disposable income has briskly augmented in recent years in both Denmark and the Netherlands. In contrast, France alongside with Finland did not experience such signicant increases as compared to those other countries. Although high, South Africa has a very good household debt to income ratio compared to the biggest economies in the world. These big economies are interconnected with the US economic and nancial system, and therefore, it can be concluded that their high debt to income ratios are primarily due to the US eco- nomic recession of 20072008. Presently, the high debt ratios can be explained by the continuing adverse effects of the 20082009 nancial crisis and the arrival of the European sovereign debt crisis. The US subprime mortgage crisis (SMC) which started in 20072008 is considered by many economists to be the most signicant incident that occurred since the great depression of the 1930s. Petersen et al. (2012) indicated that the SMC shook the foundations of the nancial Economic Modelling 33 (2013) 482492 Corresponding author. Tel.: +27 18 389 2615; fax: +27 18 389 2090. E-mail addresses: [email protected] (C. Meniago), [email protected] (J. Mukuddem-Petersen), [email protected] (M.A. Petersen), [email protected] (I.P. Mongale). 1 Cell.: +27 78 436 9966. 2 Tel.: +27 18 389 2615; fax: +27 18 389 2090. 3 Tel.: +27 18 389 2622; fax: +27 18 389 2090. 4 Tel.: +27 18 389 2620; fax: +27 18 389 2090. 0264-9993/$ see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econmod.2013.04.028 Contents lists available at SciVerse ScienceDirect Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

What causes household debt to increase in South Africa?

Embed Size (px)

Citation preview

Page 1: What causes household debt to increase in South Africa?

Economic Modelling 33 (2013) 482–492

Contents lists available at SciVerse ScienceDirect

Economic Modelling

j ourna l homepage: www.e lsev ie r .com/ locate /ecmod

What causes household debt to increase in South Africa?

Christelle Meniago a,b,1, Janine Mukuddem-Petersen a,c,⁎,2,Mark A. Petersen a,d,3, Itumeleng P. Mongale a,b,4

a Faculty of Commerce and Administration, North-West University, Mafikeng Campus (NWU-MC), Private Bag X2046, Mmabatho, 2735, South Africab Department of Economics, South Africac Graduate School of Business and Government Leadership (GSBGL), South Africad Faculty of Commerce and Administration, South Africa

⁎ Corresponding author. Tel.: +27 18 389 2615; fax:E-mail addresses: [email protected] (C. Meniag

[email protected] (J. [email protected] (M.A. Petersen), ItumelengMo(I.P. Mongale).

1 Cell.: +27 78 436 9966.2 Tel.: +27 18 389 2615; fax: +27 18 389 2090.3 Tel.: +27 18 389 2622; fax: +27 18 389 2090.4 Tel.: +27 18 389 2620; fax: +27 18 389 2090.

0264-9993/$ – see front matter © 2013 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.econmod.2013.04.028

a b s t r a c t

a r t i c l e i n f o

Article history:Accepted 9 April 2013

JEL classification:C32H31C51C52

Keywords:(Global) Financial crisisHousehold indebtednessCointegrationVector Error Correction ModelSouth Africa

The 2007–2008US subprimemortgage crisis evolved into afinancial crisis that negatively affectedmany economiesin theworld andwas afterwardswidely referred to as the globalfinancial crisis. Since the beginning of this financialcrisis of 2008–2009, South Africa experienced a significant increase in its household debt to income ratio. In themain, this paper investigates the prominent factors contributing to the rise in the level of household debt inSouthAfrica. Specifically,we construct amodel for SouthAfrican household debt through the application of a VectorError CorrectionModel (VECM).We employ quarterly time series data throughout the timeline 1985Q1 to 2012Q1and all the econometric tests are analyzed using the statistical software package EViews 7. Our results confirmedthe existence of a long run cointegrating relationship between household debt and other macroeconomic deter-minants. Increasing household debt was found to be significantly affected by positive changes in consumer priceindex, gross domestic product and household consumption. Also, house prices and household savingswere foundto positively contribute to a rise in household debt but this relationship was found to be statistically insignificant.Alternatively, household borrowingwas found to be significantly and insignificantly affected by negative changesin income and prime rate, respectively. Ultimately, the existence of a long run cointegrated relationship enabledus to build an error correction model for household debt which will facilitate future forecasting.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

An economy with a good financial situation is associated with lowdebt levels in the household sector. To manage low debt levels amonghouseholds, financial institutions which act as the primary source ofcredit in several countries and specifically in South Africa must bevery prescriptive and discerning when it comes to the provision ofloans to customers. With escalating debt, the household sector mayrun the risk of being too exposed to several adverse surprises like un-employment shocks, asset price shocks and shocks from income, justto name a few. In this regard, financial crises have historically beenperceived to emit devastating shocks to vulnerable economies.

In lieu of this, it is unfortunate to observe that South Africa recordsa very high debt level. According to the SARB (2012) quarterly bulle-tin (June 2012), the ratio of household debt to disposable income

+27 18 389 2090.o),tersen),[email protected]

rights reserved.

decreased slowly to 74.7% in the first quarter of 2012 compared to74.8% in the fourth quarter of 2011. Though this is seen as a slow de-crease, this figure still remains very high and such high levels ofhousehold debt may have adverse effects on the economy. Withoutan effective investigation to discover what might cause this highlevel of debt among households in South Africa, the household sectormight still remain very susceptible to shocks.

On a global scale, South Africa records low debt levels compared tothe most influential economies in the world (see Fig. 1). The ratio ofhousehold debt to disposable income has briskly augmented in recentyears in both Denmark and the Netherlands. In contrast, Francealongside with Finland did not experience such significant increasesas compared to those other countries. Although high, South Africa hasa very good household debt to income ratio compared to the biggesteconomies in the world. These big economies are interconnected withtheUS economic andfinancial system, and therefore, it canbe concludedthat their high debt to income ratios are primarily due to the US eco-nomic recession of 2007–2008. Presently, the high debt ratios can beexplained by the continuing adverse effects of the 2008–2009 financialcrisis and the arrival of the European sovereign debt crisis.

TheUS subprimemortgage crisis (SMC)which started in 2007–2008is considered by many economists to be the most significant incidentthat occurred since the great depression of the 1930s. Petersen et al.(2012) indicated that the SMC shook the foundations of the financial

Page 2: What causes household debt to increase in South Africa?

DenmarkNetherlandsNew ZealandAustraliaUnited KingdomIrelandUnited StatesSwedenJapanCanadaGermanySpainFinlandFrance South Africa

0 50 100 150 200 250 300

Source: OECD and SARB

Fig. 1. Level of household debt to income ratio in selected countries.Source: OECD and SARB.

483C. Meniago et al. / Economic Modelling 33 (2013) 482–492

industry by causing the failure of many iconic Wall Street investmentbanks and prominent depository institutions. In mid-2008–2009, theSMC cascaded into a financial crisis that spread into many regions ofthe world. This crisis stymied credit extension to households and busi-nesses thus creating credit crunches and, ultimately recessions. Allenand Giovannetti (2010) reiterated that the seeds of this crisis can betraced to the low interest rate policies adopted by the Federal ReserveBank and other central banks after the collapse of the technologystock bubble. In this context, Masilela (2009) asserted that this financialcrisis was to a large extent a debt crisis.

Consequently, the question regarding the degree to which SouthAfrica and other African countries were affected arises. Despite theinitial assumptions about the South African economy not beingaffected significantly by this crisis, this economy was plunged intoa recession for the first time in 17 years. On the other hand, Naude(2009) accentuated that the overall effects of the financial crisis ondeveloping countries and especially African countries will certainlybe negative.

In surveying the literature, it is apparent that the existing studiesinto the development of debt have been largely theoretical. Conse-quently, studies which examined the causes of increasing debt levelsin the household sector are very limited. To the best of our knowledge,this study is the first of its kind to build amodel for South African house-hold debt using the Vector Error Correction Model (VECM) frameworkto investigate the main reasons why South African households enterinto debt and records high debt levels in their balance sheet. Also, thiseconometric model is estimated using quarterly data from 1985 Q1 to2012 Q1 with the aid of the statistical software package EViews 7.Consequently, from our results, measures will then be proposed to alle-viate credit conditions among households thereby encouraging a stablefinancial household sector in South Africa. The structure of this paper isas follows. The current section is of an introductory nature, Section 2 re-views the theoretical framework and the empirical literature related tohousehold debt. Section 3 provides a brief explanation of the methodsused in this paper to build a VECM model for household debt in SouthAfrica, while Section 4 will discuss the results. Finally, we concludewith Section 5.

2. Theoretical model and empirical literature

2.1. Theoretical model

Above all, this paper is mainly supported by the theoretical frame-work of the life cycle hypothesis (LCH) formalized by the economists

Irving Fisher, Rod Harrod, Albert Ando and FrancoModigliani, but wasmainly initiated from the latter. The main idea behind the LCH is thathouseholds mainly go in for large amounts of debt to smooth theirconsumption and for the possession of long-lasting commodities(houses, cars, etc.). The model assumes that a household can maximizeutility over its life-time subject to an intertemporal budget constraint.This implies that by smoothing their consumption, households canmaximize utility over their life-cycle. Clearly, the model foresees thatconsumption in each period is dependent on expectations about lifetime income. Assuming that household income is upward sloping, wecan say that during the early stage of their working life, householdswill have a negative saving rate. However, as they will grow older to-getherwith their income, their savingswill increasewhile indebtednesswill decrease. Upon retirement, that is when households are no longerworking, households will again dissave as in the early stage of theirworking life. At this stage of their lives, their consumption will be prin-cipally financed by the income they earned during their working age.Households may then enter into debt in periods where their income isextremely low, mainly because they need to finance their existing con-sumption. For that reason, they will then repay these loans in periodswhen their income will be relatively high.

In part, the LCH framework includes house prices, inflation, consumerprice index, household income, interest rates, economic growth, andhousehold consumption as determinants of increasing household debtlevels.

In addition, we also consider the Permanent-Income Hypothesis(PIH). In 1957, the economist Milton Friedman developed the PIHin an attempt to explain consumer behavior. In particular, Friedmanargues that consumption should not depend on current incomealone. This model emphasizes that consumers use saving and borrow-ing behavior to smooth their consumption in response to random andtemporary changes in their incomes from year to year. In essence, thistheory conveys the message that households look at the future on de-ciding upon their current consumption. Friedman's PIH complementsthe LCH of Modigliani.

2.2. Empirical literature

This literature review encapsulates the relevant studies pertainingto household indebtedness in South Africa as well as in other countries.

For instance, Kotzè and Smit (2008) asserted that the high level ofhousehold debt in South Africa is due to a lack of a comprehensivesaving culture among South Africans. This is certainly caused byfinancial illiteracy on the part of the consumers as they spend almost

Page 3: What causes household debt to increase in South Africa?

484 C. Meniago et al. / Economic Modelling 33 (2013) 482–492

all of their income on consumption leaving little money for savingsand investments (Lorgat, 2003).

Philbrick and Gustafsson (2010) explored the determinants ofthe household debt to disposable income ratio in Australia usingboth the long-run cointegration analysis and a short run error cor-rection model. These authors considered the theoretical view ofthe LCH proposed by Ando and Modigliani. The results of thelong-run cointegration analysis showed that the change in thedebt ratio depends positively on house prices and negatively on in-terest rates. On the other hand, the error correction model (ECM)affirmed that in the short-run, the change in the debt ratio dependspositively on the change in house prices and the consumer senti-ment index, and negatively on inflation and the error term. In accor-dance with these authors, our study is based on the theoreticalframework of the LCH and the PIH. These fundamental theoriesplay a critical role in our choice of variables and the constructionof our model. Specifically, we adopt the long-run cointegrationanalysis and the short-run ECM to analyze the significant long andshort-run relationship between household debt and house prices,inflation, household income, prime rate, GDP, household expendi-tures and household savings. The approach of the cointegrationand short-run ECM analysis has been chosen because it has beenwidely recognized to model macroeconomic and several financialdata (Hendry and Juselius, 2000).

Meng et al. (2011) explored the possible causes of Australianhousehold debt using a Cointegrated Vector Autoregression (CVAR)model. The study employs seven variables to analyze the main reasonswhy Australian households record high debt levels. They used GDP,number of new dwelling approvals (NDWELL), house price index(HPI), interest rate, unemployment, consumer price index (CPI) andpopulation. The CVAR approachused byMeng and co-workers is a close-ly related model type to the VECM employed in our study (Juselius,2007; Zaystev, 2010). The study by Meng et al. (2011) reiterated thatthe most substantial cause of increasing debt levels was associatedwith house prices and the number of new dwellings. This paper foundthat interest rate, unemployment and CPI contribute to a negative influ-ence in explaining Australia's high household debt levels whereas GDP,HPI, NDWELL and population contribute positively. Also, wewill use thesame variables as this study used except for NDWELL and population.Notwithstanding, other studies also investigated why householdsenter into debt but mainly using a qualitative type of assessment (forinstance Debelle, 2004; Keese, 2012; Nieto, 2007).

Ultimately, when considering the outcome of the existing literature,the quantitative VECM framework appears to be the appropriate methodfor our study (Anderson et al., 2002; Babatunde and Adebafi, 2005;Carrascal and Rio, 2004; Chu, 2011; Gimeno and Martinez-Carrascal,2010; Ni et al., 2011; Oikarinen, 2009; Omisakin et al., 2009; Pattersson,2011).

3. Methodology

In this study, we employ the VECM to estimate the relationshipbetween household debt and other macroeconomic variables. Thisapproach allows for contemporaneous and lagged interconnectionbetween the variables of interest to exist (Zaystev, 2010). In the sequelwe provide a brief description of the steps taken to develop our debtmodel.

3.1. Unit root tests

Many macroeconomic time series contain unit roots, which tend tobe dominated by stochastic trends (Dickey and Fuller, 1979, 1981;Phillips and Perron, 1988; Kwiatkowski et al., 1992). Among theavailable tests we selected the Augmented Dickey–Fuller (ADF) andthe Phillips and Perron (PP) tests to investigate the presence of unitroots in time series.

3.1.1. Augmented Dickey–Fuller testThe two unit root tests mentioned above are computed for each

time series variables to check whether the time series variables areintegrated of the same order. The ADF test, tests the following threeequations:

ΔXt ¼ α þ δXt−1 þ∑ni¼1λiΔXt−1 þ εt ð1Þ

ΔXt ¼ α þ βtþ δXt−1 þ∑ni¼1λiΔXt−1 þ εt ð2Þ

ΔXt ¼ δXt−1 þ∑ni¼1λiΔXt−1 þ εt : ð3Þ

Eqs. (1), (2) and (3) represent the equations for the ADF test for theconstant (intercept) only, constant and trend and none of the determin-istic components, respectively. The ADF tests the null hypothesis that Xthas a unit root against the alternative hypothesis that Xt does not have aunit root.

3.1.2. Phillips–Perron (PP) testAn alternative unit root test, the Phillips–Perron (PP) test (Phillips

and Perron, 1988) was conducted to ensure the stationarity of thedata series as this test uses non-parametric correction to deal withany correlation in the error terms. PP tests estimate the following re-gression:

ΔY t ¼ θ0 þ δY t−1 þ εt : ð4Þ

Similarly to the ADF test, the PP test, analyzes the null hypothesisthat Xt has a unit root against the alternative hypothesis that Xt doesnot have a unit root. The decision rule for the two stationarity tests(ADF and PP) is therefore the same. In particular, in both tests if thecorresponding test statistics is greater than the critical value at thegiven significance level, then we do not reject the null hypothesisand conclude that there exists a unit root in the series. Conversely,if the corresponding test statistics is less than the critical value atthe corresponding test statistics, then we reject the null hypothesisand conclude that there is no existence of a unit root in the timeseries.

Once a unit root has been confirmed for a data series, the questionis whether there exists some long-run equilibrium relationshipamong variables. The existence of a long-run equilibrium relationshipamong economic variables is referred to as cointegration. Before weproceed to the cointegration test, the suitable lag length is estimatedto see which number of lags best fits the time series data. The estima-tion of the appropriate lag length is based on different informationcriteria for the selection of a model (Liu, 2007; Meng et al., 2011;Philbrick and Gustafsson, 2010) such as Akaike information criterion(AIC) (Akaike, 1973), Schwarz information criterion (SIC) (Schwarz,1978) and the Hannan–Quinn information criterion (HQ) (Hannanand Quinn, 1978).

3.2. Cointegration test

When the order of integration of the variables has been identifiedthrough the stationarity test, the next step is to perform the cointegrationtest. Cointegration test is often used as a pre-requisite for determiningwhether a standard VAR or VECM should be employed to study therelationship between the variables. When cointegration is present inour variables, the VECMwill be the appropriate specification to achieveour objectives. The concept of cointegration provides a sound method-ology for modeling both long run and short run dynamics in a system(Dunis and Ho, 2005).

In this paper, we employ the Johansen's cointegration test becauseof its proven suitability when dealing with multivariate time seriesdata. Johansen and Juselius (1990) hypothesized that this test examinesthe null hypothesis of no-cointegration in the variables against the

Page 4: What causes household debt to increase in South Africa?

485C. Meniago et al. / Economic Modelling 33 (2013) 482–492

alternative that there exist cointegration. In this regard, if cointegrationis found among the variables, this implies that a long run relationshipexists and thereby granger causality among them in at least one direc-tion. Consequently, VECM will be used to correct the disequilibrium inthe cointegrating relationship captured by the error correction term(ECT).

By employing the Johansen cointegration test, we use both the trace(refer to Eq. (5)) and the max-eigenvalue (refer to Eq. (6)) tests to seethe number the cointegration equations that exist among our variables.These tests are summarized by the following equations:

λtrace rð Þ ¼ −TIn∑ni¼rþ1 1−λið Þ ð5Þ

λmax r; r þ 1ð Þ ¼ −TIn 1−λrþ1� � ð6Þ

where λs are the estimated eigenvalues of the characteristic equations,T is the number of usable observations and r is the rank of the long-runmatrix (π). The number of cointegration vectors (r) is always smaller orequal to the number of endogenous variables (n). The trace test teststhe null hypothesis of r cointegrating vectors against the alternativehypothesis of n cointegrating vectors. The maximum eigenvalue test,on the other hand, tests the null hypothesis of r cointegrating vectorsagainst the alternative hypothesis of r + 1 cointegrating vectors.

3.3. Vector Error Correction model

In general, when variables are cointegrated, there must also be anECM that will describe the short-run adjustments of the cointegratedvariables. Then, after having determined the number of cointegrationvectors, the corresponding VECM can be estimated. This model(VECM) is a restricted VAR designed for use with non-stationary seriesthat are identified to be cointegrated. Hassan (2003) described theVECM as a model which describes how the system is adjusting in eachtime period towards its long run equilibrium state. The author assertedthat since there is a long run relationship among the variables, thereforein the short term, deviations from this long-run equilibrium will feedback on the changes in the dependent variables in order to force theirmovements towards the long-run equilibrium state. The author thenconcluded that the coefficients of the error-correction terms derivedfrom the cointegrating vectors represent the proportion by which thelong-run disequilibrium in the dependent variables is corrected ineach short-term period.

Sourcing from Naranjo and Toevs (2002), the VECM can berepresented as:

ΔXt ¼ μ þ Γ1ΔXt−1 þ Γ2ΔXt−2 þ…þ Γk−1ΔXt−kþ1 þΠXt−k

þ �t ; t ¼ 1;2; ; ; ; Tð7Þ

where Xt is a vector of p I (1) variables, μ is a p × 1 vector of inter-cepts, Γ1, Γ2, Γk,Π is a p × p vector of parameters, t is the error termor vector of impulses representing unanticipated movements in Xt

and Δ is the difference operator. Furthermore, a stepwise summaryof the steps of a VECM is illustrated in Fig. 2.

4. Results and discussion

4.1. Data

This paper uses time series data transformed into natural loga-rithms and the following household debt model for South Africawas estimated:

LRHD ¼ f LRHPI; LCPI; LRIN; LRPR; LRGDP; LRCON; LHSAVð Þ:

The regression equation of household debt will therefore be of theform:

LRHDi ¼ β0 þ β1LRHPIizfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{þ

þ β2LCPIizfflfflfflfflffl}|fflfflfflfflffl{−

þ β3LRINi

zfflfflfflfflfflffl}|fflfflfflfflfflffl{−

þ β4LRPRi

zfflfflfflfflfflffl}|fflfflfflfflfflffl{−

þ β5LRGDPizfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{þ

þ β6LRCONzfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{þ

þ β7LRSAVi

zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{−

þ μ i

where LRHD is the natural log of real household debt, LRHPI is thenatural log of real house price index, LCPI is the natural log of consumerprice index, LRIN is the natural log of real income, LRPR is the natural logof real prime rate, LRGDP is the natural log of real GDP, LRCON is thenatural log of real household consumption expenditures and LRSAV isthe natural log of real household savings.

The variable LCPI is used here as a proxy to measure inflation. TheLRHPI data comes from the ABSA bank South Africa. As for the othervariables, they are all sourced from the South African Reserve Bank(SARB) except for inflation (CPI) which was extracted from StatisticsSouth Africa. The time periods extend from 1985 Q1 to 2012 Q1.

4.2. Results and interpretation

The descriptive statistics of the variables are provided in Table 1.The results of the descriptive statistics indicate that all the vari-

ables exhibit a positive skewness with values greater than 0 exceptfor CPI and LRSAV which indicates a negative skewness. In otherwords, the positive skewness indicates that the series are skewedto the right. The value of the kurtosis which measures the flatnessand peakness of the distribution shows that the variables havevalues less than 3. The Jarque–Bera test which is used to test whetherthe variables follow a normal distribution reveals that at the 5% sig-nificance level, all the variables show forms of non-normality exceptfor LRPR.

The growth trend of the variables confirms that all the variables aretrending and are non-stationary (see Appendix A) and therefore need-ed to be differenced to be made stationary (see Appendix B). This con-clusion will be confirmed with the unit root test results presented inthe following section.

4.2.1. Unit root testThe ADF and PP unit tests were conducted to determine the order

of integration of the variables. From the model of intercept and trendwhich appears to be more realistic compared to the other twomodels, the results confirm that the series are all of I (1) series (seeAppendices C and D).

4.2.2. Cointegration and VECMThe conclusion that the time series variables are cointegrated allows

us to proceedwith cointegration, but, we first determine the appropriatelag length that will be suitable for our study. Table 2 presents the resultsof the lag length determination test and we selected two lags proposedby the Final Prediction Error (FPE) and the Hannan–Quinn (HQ) infor-mation criteria in building our model but specifically to avoid somemisspecification problems in our analysis.

The results of the Johansen cointegration test are shown in Table 3.As reported in Table 3, the null hypothesis of no cointegration

is rejected from both test statistics. The trace test suggests fourcointegrating equations while the maximum eigenvalue suggeststwo cointegrating equations at the 5% significance level. Becauseboth tests differ in their results, we prefer the maximum eigenvalue(Banerjee et al., 1993). The two cointegrating equations proposed bythe maximum eigenvalue test are not fully identified unless weimposed some random normalization. Therefore, the cointegratingvectors have been normalized on household debt because it is con-sidered as the dependent variable in this study and in this way we

Page 5: What causes household debt to increase in South Africa?

ECONOMIC DATATRANSFORMATION OF VARIABLES IN LOGS

VISUAL INSPECTION OF THE VARIABLES

UNIT ROOT TEST

AUGMENTED DICKEY-FULLER TEST

PHILLIPS PERRON TEST

ARE THE VARIABLES STATIONARY?

YES NO

PROCEED ESTIMATION WITH

VECTOR AUTOREGRESSION

ARE THE VARIABLES COINTEGRATED?

TEST FOR NUMBER OF COINTEGRATION

RELATIONSHIPS USING JOHANSEN

APPROACH

TRACE STATISTIC

MAXIMUM EIGENVALUE

STATISTIC

IMPOSE RESTRICTIONS BASED

ON ECONOMIC THEORY

PERFORM DIAGNOSTIC TESTS

ON THE MODEL

IS THE MODEL GOOD?

IF YES, GOOD FOR POLICY ANALYSIS

IF NO, RE-ESTIMATE THE

MODEL

PROCEED ESTIMATION WITH

VECTOR ERROR CORRECTION MODEL

YESNO

LAG LENGTH DETERMINATION

KPSS TEST

SCHWARTZ INFORMATION

CRITERIA

AKAIKE INFORMATION

CRITERIA

HANNAH-QUINN CRITERIA

DESCRIPTIVE STATISTICS

Fig. 2. Steps for estimating a VECM for household debt.

Table 1Descriptive statistics.

LRHD LRHPI LCPI LRIN LRPR LRGDP LRCON LRSAV

Mean 4.803 4.018 3.900 13.717 3.420 14.186 14.278 4.9752Median 4.675 3.819 4.050 13.661 3.572 14.094 14.090 9.112Maximum 6.117 4.679 4.799 14.232 5.319 14.762 15.313 11.135Minimum 4.118 3.600 2.504 13.244 2.002 13.762 13.859 −10.099Std. Dev. 0.521 0.366 0.629 0.304 0.805 0.314 0.391 7.854Skewness 0.788 0.679 −0.546 0.193 0.092 0.391 1.073 −1.141Kurtosis 2.617 1.789 2.252 1.731 2.134 1.745 2.947 2.403Jarque–Bera 11.952 15.043 7.966 7.991 3.558 9.943 20.945 25.296Probability 0.002 0.000 0.018* 0.018* 0.168* 0.006 0.000 0.000Sum 523.53 438.02 425.15 1495.19 372.819 1546.335 1556.40 542.301Sum sq. dev 29.375 14.543 42.779 10.025 69.992 10.687 16.512 6663.37Observations 109 109 109 109 109 109 109 109

*, **, *** represent significance at 1%, 5%, 10% respectively.

486 C. Meniago et al. / Economic Modelling 33 (2013) 482–492

Page 6: What causes household debt to increase in South Africa?

Table 2Lag length determination criteria results.

Lag LogL LR FPE AIC SIC HQ

0 589.7853 NA 1.37e−15 −11.520 −11.313 −11.4361 1821.658 2244.204 1.24e−25 −34.646 −32.782⁎ −33.8912 1947.212 208.842 3.76e−26⁎ −35.865 −32.344 −34.440⁎

3 1998.127 76.624 5.16e−26 −35.606 −30.428 −33.5104 2047.239 66.131 7.78e−26 −35.311 −28.476 −32.5445 2117.043 82.934 8.45e−26 −35.426 −26.933 −31.9886 2203.120 88.634⁎ 7.49e−26 −35.863 −25.713 −31.7547 2273.586 61.396 1.07e−25 −35.991 −24.184 −31.2128 2376.931 73.671 1.02e−25 −36.779⁎ −23.306 −31.320

LR: Sequential modified LR test statistic (each test at 5% level).FPE: Final prediction error.AIC: Akaike information criterion.SIC: Schwarz information criterion.HQ: Hannan-Quinn information criterion.⁎ indicates lag order selected by the criterion.

487C. Meniago et al. / Economic Modelling 33 (2013) 482–492

are able to express the long run relationship. Nevertheless, accordingto the theory, we expected to have only one cointegrating equation.However, it is worth acknowledging that the first cointegratingequation best reflected the theory though not all the coefficientsbear the normal sign. Consequently, the first cointegrating equationis considered in this paper as our long run equation.

The results of the normalized cointegration coefficients whichrepresent the long run equation and the corresponding error correc-tion terms results are presented in the VECM in Table 4.

The results of the long run exclusion and weak exogeneity testswere carried out and the outcome respectively maintained that at5% significance level, all the variables belong and should thereforebe included in the model and none of them were seen to be weaklyexogenous to the whole system of equations. The diagnostic testswere also performed and show that the estimated model is goodexcept for the normality test which has a P-value of 0.0000. However,as confirmed in the literature, this not a very important conditionespecially if we want to use our estimated model for forecasting. AsStengos and Wu (2006) affirmed, the normality is very sensitive tothe number of observations.

With a coefficient of determination of 0.40, we conclude that theoverall system indicates a fair goodness of fit. As household debtwhich appears to be on the same side as other variables in thecointegrating vector, the opposite signs of the explanatory variableswill be considered to get the right interpretation. However, we thenrecognize that because our variables have been transformed intonatural logarithms, the estimated coefficients will be interpretedas elasticity.

A positive linear dependence was found to exist between houseprices but this relationship is found to be statistically insignificant.For each percentage increase in LRHPI, LRHD will increase by 0.52% inthe long run. The estimated coefficient of LRGDP is significant andbears the expected sign as suggested by the theory. The results confirm

Table 3Johansen cointegration test results.

Hypothesized no of CE(s) Eigenvalue Trace statistic 0.05 critical value

None* 0.477041 226.9401 159.5297At most 1* 0.386190 158.2254 125.6154At most 2* 0.279720 106.4899 95.75366At most 3* 0.231220 71.70967 69.81889At most 4 0.180727 43.83694 47.85613At most 5 0.141881 22.70719 29.79707At most 6 0.058399 6.487819 15.49471At most 7 0.001032 0.109436 3.841466

Trace test indicates 4 cointegrating eqn(s) at the 0.05 level.Max-eigenvalue test indicates 2 cointegrating eqn(s) at the 0.05 level.* Denotes rejection of the hypothesis at the 0.05 level.**MacKinnon et al. (1999) P-values.

that a 1% increase in LRGDP will cause household debt to increaseby 9.85%. This outcome correspondswith the theory of the LCH and con-sequently implies that the relationship between household debt andGDP in South Africa is positive.

Still from the results, we noticed that household debt is also pos-itively influenced by household consumption. The outcome of theresults confirms that a 1% increase in LRCON will induce LRHD tochange by 15.53%. This finding confirms what was expected fromthe theory and this tells us that in South Africa, the more householdsincrease their consumption/expenditures, the more they go intolarge amounts of debt.

Contrary to the theory, LCPI and LRSAV were also revealed to pos-itively contribute to the rise in the level of household debt in SouthAfrica. The outcome shows that a 1% change in LCPI increases LRHDby 16.95% in the long run. This conclusion allows us to postulatethat the negative relationship between household debt and CPI sug-gested by the theory does not hold in South Africa for our estimatedtimeline. On the other hand, though LRSAV was found to be statisti-cally insignificant, the existence of a positive relationship may berepresenting a tendency for the South African households to still goin for more debt even when their savings records are positive.

Still from the results, there is evidence of a negative relationshipbetween LRHD with LRIN and LRPR, respectively. We conclude thata 1% increase in LRIN will significantly change LRHD by 5.26%. Thisoutcome was expected and allows us to confirm that in South Africa,households borrow more when their income is lowered. On the otherhand, the coefficient of LRPR is having the predictable sign but it isinsignificant. The insignificant negative relationship between thesevariables implies that when banks reduce a high interest rate, SouthAfrican households are encouraged to borrow more because the costof repayment will be lower.

The second part of Table 4 presents the results of the short runequation. The result of the error-correction analysis gives some insightinto the deviations from the long run relationship described above.In the short run, the most important part is the analysis of the ECT.Concerning the coefficients of the short run equation, only the variablesLRHPI, LRGDP and LRCON bear the correct negative signs but LRCONappears to be statistically insignificant.

The sign of the coefficient is negative as expected by the theorybut turns out to be statistically insignificant. Though statistically insig-nificant, the estimated coefficient implies that about 3.6% of the disequi-librium is corrected between each quarter (since we are dealing withquarterly data). That is, from any exogenous shock to the system, itwill take 3.6% each quarter for household debt to get back to equilibriumwhich is a very slow movement.

5. Conclusion

We examined the long run and short run relationships betweenhousehold debt and seven macroeconomic variables over the timeline

Hypothesized no of CE(s) Max-eigen statistic 0.05 critical value

None* 68.71474 52.36261At most 1* 51.73548 46.23142At most 2 34.78024 40.07757At most 3 27.87272 33.87687At most 4 21.12976 27.58434At most 5 16.21937 21.13162At most 6 6.378383 14.26460At most 7 0.109436 3.841466

Page 7: What causes household debt to increase in South Africa?

Table 4VECM.

Variables LRHD LRHPI LCPI LRIN LRPR LRGDP LRCON LRSAV

Long run 1.000 −0.515 −16.950 5.625 0.546 −9.850 −15.526 −0.0191Standard error (0.787) (4.077) (3.085) (0.364) (3.606) (4.124) (0.018)T-stat [−0.683] [−4.940] [2.348] [1.978] [3.131] [−4.460) [−1.299]

Variables ΔLRHD ΔLRHPI ΔLCPI ΔLRIN ΔLRPR ΔLRGDP ΔLRCON ΔLRSAV

Short run −0.036 −0.073 0.025 0.003 0.093 −0.047 −0.002 1.455Standard error (0.030) (0.018) (0.009) (0.024) (0.053) (0.017) (0.013) (3.414)T-stat [−1.218] [−4.082] [2.744] [0.144] [1.757] [−2.789] [−0.1673] [0.426]

R-squared: 0.402657.Adj. R-squared: 0.279069.VEC residual serial correlation LM test shows no form of autocorrelation at all orders.VEC Heteroskedasticity test (no cross terms): P-value = 0.0625.Normality test: P-value: 0.0000, Stability test: VECM is stable.

488 C. Meniago et al. / Economic Modelling 33 (2013) 482–492

1985 Q1 to 2012 Q1. These macroeconomic variables are real houseprice index, consumer price index, real income, real prime rate, realgross domestic product, real household consumption expendituresand real savings.

The main aim of this paper was to investigate what causes thehigh level of debt among South African households. The Johansencointegration analysis confirmed that the selected macroeconomicvariables all move together in the long run. The outcome of the anal-ysis established that increasing household debt in South Africa canbe associated to positive changes in house prices, inflation (CPI),GDP, household consumption expenditures and household savings.Specifically, this implies when house prices, inflation, GDP, house-hold consumption and savings increase, this will encourage house-holds to borrow more and in response cause household debt toincrease. Though some variables were found insignificant (LRHPI,LRPR, LRSAV), we tend to believe that at the practical level, theyplay a major role in the variation of household debt levels in SouthAfrica. On the other hand, negative factors that caused householddebt to increase in South Africa are household income and primerate. In other words, when the income of households and the primerate decrease, South African households tend to borrow more. Theerror correction term also confirms that in the short run; about3.6% of the disequilibrium is corrected each quarter for the wholesystem to be reestablished back to equilibrium.

Based on the estimated outcome, we propose some sugges-tions to alleviate the high debt levels among the South Africanhouseholds.

The inverse connection found between household debt andhousehold income supports that low income which surely resultsfrom high unemployment rate in the country leads to higherdebt. We therefore suggest that debt in the household sector canbe lowered if more jobs are created in the economy which willdrive more income into the household sector and consequentlyless debt.

Higher GDP was found to contribute significantly in the in-crease in household debt levels. The theory supports that withhigher GDP which implies higher economic growth followed byhigher income; households and creditors will respectively feelvery confident in taking on and issuing more debt. One of thebenefits of higher economic growth is higher income enjoyedamong households except that most often this benefit is notalways equally shared among households. South Africa for exam-ple is considered as one of the countries with the highest incomeinequality in the world. The government however should try toreduce the income inequality in the country so that the benefitsof economic growth could be enjoyed by all and as such householddebt will be decreased.

The theory supports the hypothesis that higher household consump-tion expenditures lead to increased household debt levels. We concludethat if households are faced with lower income in their pockets, in orderto meet all their expenses, they must go in for more debt. Therefore,appropriate attention should be engaged on how households spendtheir money. We recommend that households should spend carefullyby meeting their basic needs so that in return their debt will decrease.

Though house price index and prime rate were found to be insignif-icant but having the correct sign, we believe that they play a great rolein affecting the development of household debt in South Africa. There-fore, we propose closer supervision of the housing market because,housing debt represents the main component of household debt inSouth Africa. In particular, the government should be very strict whennew homes are issued to households. Also, we recommend that in issu-ing new homes to households, their credit conditions have to be exam-ined very closely and houses should however be given to householdswith good credit conditions. If this is properly done, this will boost notonly the money lenders to be more professional in the giving out ofloans but also households will be more responsible when they go infor debt and in return will develop an attitude of keeping good creditconditions.

On the other hand, interest rates which act as official instrumentsused by every country to control the amount of credit extended to theprivate sector should therefore be regulated carefully. In particular,because of the negative relationship found between these two vari-ables, care should be taken to avoid large economic shocks sinceshocks on this variable can have a significant influence on other mac-roeconomic variables and household debt in particular. Dynan (2012)quoted this “Even households with positive net worth often choose tohold some debt. This behaviour arises in part because of the convenienceof using credit cards, but more significantly when a household wants toown a home and its desired housing services correspond to a propertywith value exceeding the household's wealth. In this case, the householdnot only has a motivation for borrowing but also can use the home ascollateral in order to create an ability to borrow that would not other-wise exist”. In this context, we postulate that the dismissal of creditcard debt could contribute to a reduction of a large amount of debtin the household sector in South Africa.

For future research, because South Africa lacks quarterly unem-ployment data, we recommend that this variable should be analyzedtogether with other variables in a VECM using annual data. This anal-ysis will be very significant because we will then capture the role thatunemployment has to play in affecting household debt levels in SouthAfrica. We would like to encourage more national and internationalstudies using this type of econometric approach for future researchto facilitate researchers in the interpretation and comparison ofsuch complex economic relationships.

Page 8: What causes household debt to increase in South Africa?

489C. Meniago et al. / Economic Modelling 33 (2013) 482–492

Appendix A. Graph of variables at level

0.4

0.6

0.8

1.0

1.2

1.4

86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10

86 88 90 92 94 96 98 00 02 04 06 08 1086 88 90 92 94 96 98 00 02 04 06 08 1086 88 90 92 94 96 98 00 02 04 06 08 10

86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10

LRHD

3.4

3.6

3.8

4.0

4.2

4.4

4.6

4.8LRHPI

2.0

2.5

3.0

3.5

4.0

4.5

5.0LCPI

13.2

13.4

13.6

13.8

14.0

14.2

14.4LRIN

1

2

3

4

5

6LRPR

13.6

13.8

14.0

14.2

14.4

14.6

14.8LRGDP

13.6

14.0

14.4

14.8

15.2

15.6LRCON

-12

-8

-4

0

4

8

12LRSAV

(LRHPI) D(LCPI)

Appendix B. Graph of variables at first difference

D(LRHD) D

-.15

-.10

-.05

.00

.05

.10

86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10

86 88 90 92 94 96 98 00 02 04 06 08 1086 88 90 92 94 96 98 00 02 04 06 08 1086 88 90 92 94 96 98 00 02 04 06 08 10

86 88 90 92 94 96 98 00 02 04 06 08 10 86 88 90 92 94 96 98 00 02 04 06 08 10

-.08

-.04

.00

.04

.08

-.02

.00

.02

.04

.06

-.08

-.04

.00

.04

.08

.12D(LRIN)

-.3

-.2

-.1

.0

.1

.2

.3D(LRPR)

-.04

-.02

.00

.02

.04

.06D(LRGDP)

-.08

-.06

-.04

-.02

.00

.02

.04D(LRCON)

-20

-10

0

10

20D(LRSAV)

Page 9: What causes household debt to increase in South Africa?

490 C. Meniago et al. / Economic Modelling 33 (2013) 482–492

Appendix C. ADF unit root test results

Variable Level of test Model Number of lags ADF test statistics 5% critical value 10% critical value Order of integration

LRHD Level Intercept 0 −0.517 −2.888 −2.581Trend + intercept 6 −2.694 −3.452 −3.152None 0 0.881 −1.944 −1.615 I(0)**

Difference Intercept 0 −10.491 −2.888 −2.581 I(1)**Trend + intercept 0 −10.489 −3.452 −3.152 I(1)**None 2 −4.567 −1.944 −1.615

LRHPI Level Intercept 1 −1.109 −2.888 −2.581Trend + intercept 1 −1.877 −3.452 −3.152None 1 0.456 −1.944 −1.615 I(0)**

Difference Intercept 0 −3.578 −2.888 −2.581 I(1)**Trend + intercept 0 −3.444 −3.452 −3.152 I(1)**None 0 −3.547 −1.944 −1.615

LCPI Level Intercept 1 −4.312 −2.888 −2.581 I(0)**Trend + intercept 1 −2.919 −3.452 −3.152None 1 3.022 −1.944 −1.615 I(0)**

Difference Intercept 0 −4.788 −2.888 −2.581Trend + intercept 0 −6.201 −3.452 −3.152 I(1)**None 2 −1.747 −1.944 −1.615

LRIN Level Intercept 0 0.369 −2.888 −2.581 I(0)**Trend + intercept 0 −3.362 −3.452 −3.152None 0 3.961 −1.944 −1.615 I(0)**

Difference Intercept 0 −11.437 −2.888 −2.581Trend + intercept 0 −11.426 −3.452 −3.152 I(1)*None 0 −9.886 −1.944 −1.615

LRPR Level Intercept 1 −1.270 −2.888 −2.581Trend + intercept 1 4.617 −3.452 −3.152 I(0)**None 1 2.466 −1.944 −1.615 I(0)**

Difference Intercept 0 −5.254 −2.888 −2.581 I(1)**Trend + intercept 0 −5.245 −3.452 −3.152None 0 −4.688 −1.944 −1.615

LRGDP Level Intercept 0 1.540 −2.888 −2.581 I(0)**Trend + intercept 0 −2.345 −3.452 −3.152None 0 5.872 −1.944 −1.615 I(0)**

Difference Intercept 0 −9.286 −2.888 −2.581Trend + intercept 0 −9.480 −3.452 −3.152 I(1)**None 3 −2.647 −1.944 −1.615

LRCON Level Intercept 1 −3.330 −2.888 −2.581 I(0)**Trend + intercept 1 −1.946 −3.452 −3.152None 1 −2.725 −1.944 −1.615 I(0)**

Difference Intercept 0 −4.985 −2.888 −2.581Trend + intercept 0 −5.730 −3.452 −3.152 I(1)**None 0 −5.729 −1.944 −1.615

LRSAV Level Intercept 1 −1.062 −2.888 −2.581Trend + intercept 1 −2.231 −3.452 −3.152None 1 −1.336 −1.944 −1.615

Difference Intercept 0 −14.578 −2.888 −2.581 I(1)**Trend + intercept 0 −14.524 −3.452 −3.152 I(1)**None 0 −14.580 −1.944 −1.615 I(1)**

* Indicates that it is significant at 5%; ** Indicates that it is significant at 10%.

Appendix D. PP unit root test results

Variable Level of test Model Bandwidth PP test statistics 5% critical value 10% critical value Order of integration

LRHD Level Intercept 6 −0.768 −2.888 −2.581Trend + intercept 6 −2.159 −3.452 −3.152None 6 0.696 −1.944 −1.615 I(0)**

Difference Intercept 6 −10.632 −2.888 −2.581 I(1)**Trend + intercept 6 −10.621 −3.452 −3.152 I(1)**None 6 −10.609 −1.944 −1.615

LRHPI Level Intercept 8 −0.501 −2.888 −2.581Trend + intercept 7 −2.571 −3.452 −3.152None 7 0.44 −1.944 −1.615 I(0)**

Difference Intercept 6 −3.438 −2.888 −2.581 I(1)**Trend + intercept 6 −3.170 −3.452 −3.152 I(1)**None 6 −3.416 −1.944 −1.615

LCPI Level Intercept 5 −6.855 −2.888 −2.581 I(0)Trend + intercept 4 −3.574 −3.452 −3.152 I(0)None 8 5.515 −1.944 −1.615 I(0)

Page 10: What causes household debt to increase in South Africa?

(continued)

Variable Level of test Model Bandwidth PP test statistics 5% critical value 10% critical value Order of integration

Difference Intercept 4 −4.646 −2.888 −2.581Trend + intercept 3 −6.171 −3.452 −3.152None 3 −2.236 −1.944 −1.615

LRIN Level Intercept 2 0.534 −2.888 −2.581 I(0)**Trend + intercept 5 −3.391 −3.452 −3.152 I(0)**None 2 4.551 −1.944 −1.615 I(0)**

Difference Intercept 2 −11.632 −2.888 −2.581Trend + intercept 2 −11.617 −3.452 −3.152None 6 −10.094 −1.944 −1.615

LRPR Level Intercept 6 −1.638 −2.888 −2.581Trend + intercept 6 −4.075 −3.452 −3.152 I(0)None 6 −3.548 −1.944 −1.615 I(0)

Difference Intercept 2 −5.173 −2.888 −2.581 I(1)**Trend + intercept 2 −5.172 −3.452 −3.152None 1 −4.678 −1.944 −1.615

LRGDP Level Intercept 5 1.302 −2.888 −2.581 I(0)**Trend + intercept 5 −2.283 −3.452 −3.152None 5 4.987 −1.944 −1.615 I(0)

Difference Intercept 5 −9.361 −2.888 −2.581Trend + intercept 5 −9.521 −3.452 −3.152 I(1)**None 7 −8.115 −1.944 −1.615

LRCON Level Intercept 7 −5.205 −2.888 −2.581 I(0)Trend + intercept 7 −2.718 −3.452 −3.152None 7 −3.815 −1.944 −1.615 I(0)

Difference Intercept 6 −5.006 −2.888 −2.581Trend + intercept 5 −5.888 −3.452 −3.152 I(1)**None 5 −4.059 −1.944 −1.615

LRSAV Level Intercept 2 −1.289 −2.888 −2.581Trend + intercept 4 −2.775 −3.452 −3.152None 1 −1.517 −1.944 −1.615

Difference Intercept 3 −15.076 −2.888 −2.581 I(1)**Trend + intercept 3 −15.037 −3.452 −3.152 I(1)**None 2 −14.899 −1.944 −1.615 I(1)**

* Indicates that it is significant at 5%.** Indicates that it is significant at 10%.

Appendix D (continued)

491C. Meniago et al. / Economic Modelling 33 (2013) 482–492

References

Akaike, H., 1973. Information theory as an extension of the maximum likelihood principle.In: Petrov, B.N., Csaki, F. (Eds.), Second International Symposium on InformationTheory. AkademiaiKiado, Budapest, pp. 267–281.

Allen, F., Giovannetti, G., 2010. Fragile countries and the 2008–2009 crisis. WorkingPapers. EUI-RSCAS.

Anderson, R.G., Hoffman, D.L., Rasche, R.H., 2002. A vector error correction forecastingmodel of the US economy. Journal of Macroeconomics 24, 569–598.

Babatunde, M.A., Adebafi, R.A., 2005. Long run relationship between education andeconomic growth in Nigeria: evidence from the Johansen cointegration approach.Regional Conference on Education in West Africa, Dakar, Senegal.

Banerjee, A., Dolado, J., Galbraith, J.W., Hendry, D.F., 1993. Co-integration, error correc-tion, and the econometric analysis of non-stationary data. Advanced Texts inEconometrics. Oxford University Press, Oxford.

Carrascal, C.M., Rio, A.D., 2004. Household borrowing and consumption in Spain: aVECM approach. Working Paper. Bank of Spain.

Chu, P.K.K., 2011. Relationship between macroeconomic variables and net asset values(NAV) of equity funds: cointegration evidence and vector error correction modelof the Hong Kong Mandatory Provident Funds (MPFs). Journal of InternationalFinancial Markets, Institutions and Money 21, 792–810.

Debelle, G., 2004. Macroeconomic implications of rising household debt. BIS WorkingPaper, 153.

Dickey, D.A., Fuller, W.A., 1979. Distribution of the estimation for autoregressive timeseries with a unit root. Journal of the American Statistical Association 79,355–367.

Dickey, D.A., Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time serieswith a unit root. Econometrica 49, 1057–1071.

Dunis, C.L., Ho, R., 2005. Cointegration Portfolios of European Equities for Index Track-ing andMarket Neutral Strategies. Liverpool JohnMoores University and Centre forInternational Banking, Economics and Finance.

Dynan, K., 2012. Is a Household Debt Overhang Holding Back Consumption? BrookingsInstitution.

Gimeno, R., Martinez-Carrascal, C., 2010. The relationship between house prices and housepurchase loans: the Spanish case. Journal of Banking and Finance 34, 1849–1855.

Hannan, E.J., Quinn, B.G., 1978. The determination of the lag length of an autoregression.Journal of the Royal Statistical Society 41, 190–195.

Hassan, A.M.H., 2003. Financial integration of stock markets in the Gulf: a multivariatecointegration analysis. International Journal of Business 8, 335–346.

Hendry, D.F., Juselius, K., 2000. Explaining cointegration analysis: part I. Energy Journal21, 1–42.

Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference oncointegration with applications to the demand for money. Oxford Bulletin ofEconomics and Statistics 52, 169–210.

Juselius, K., 2007. The cointegration VAR approach: methodology and applications.Advanced Texts in Econometrics. Oxford University Press Inc., New York.

Keese, M., 2012. Who feels constrained by high debt burdens? Subjective vs. objectivemeasures of household debt. Journal of Economic Psychology 33, 125–141.

Kotze, L., Smit, A.V.A., 2008. Personal finances: what is the possible impact on entrepre-neurial activity in South Africa. Southern Africa Business Review 12, 3.

Kwiatkowski, D., Phillips, P.C.B., SchmidT, P., Shin, Y., 1992. Testing the null hypothesisof trend stationarity. Journal of Econometrics 54, 159–178.

Liu, L., 2007. Consistent Testing for Lag Length in Cointegrated Relationships. (MastersDissertation, Faculty of Science) Nelson Mandela Metropolitan University, SouthAfrica.

Lorgat, M. 2003. Enabling worker finances: the effects of rising costs of living and debton worker finances. Unpublished paper (South African Sociological Association,Durban).

MacKinnon, J.G., Haug, A.A., Michelis, L., 1999. Numerical distribution functions of like-lihood ratio tests for cointegration. Journal of Applied Econometrics 14, 563–577.

Masilela, E., 2009. South African Savings Institute Chairman's Statement at the Occasion ofthe Launch of Savings Month 2009. The South African Savings Institute, Johannesburg(Available at www.savingsinstitute.co.za [accessed March 2011]).

Meng, S., Hoang, T.N., Siriwardana, M., 2011. The determinants of Australia householddebt: a macro-level study. Business, Economics and Public Policy Working Papers,pp. 2011–2014.

Naranjo, A., Toevs, A., 2002. The effects of purchases of mortgages and securitization bygovernment sponsored enterprises on mortgage yield spreads and volatility. Journalof Real Estate Finance and Economics 25, 173–195.

Naude, W., 2009. The financial crisis of 2008 and the developing countries. DiscussionPaper. UNU-WIDER.

Ni, J.-S., Huang, S.-S., Wen, Y., 2011. Interest rate, unemployment rate and house marketin the US. International Conference on Social Science and Humanity. IACSIT Press,Singapore.

Nieto, F., 2007. The determinants of household credit in Spain.Working Paper. Bank of Spain.Oikarinen, E., 2009. Household borrowing and metropolitan house-price dynamics—

empirical evidence from Helsinki. Journal of Housing Economics 16, 126–139.Omisakin, O., Adeniyi, O., Omojolaibi, A., 2009. A VECM of energy price volatility of oil

dependent economy. Pakistan Journal of Social Sciences 6, 207–213.Pattersson, B.H., 2011. Essays in Macroeconomics: A VECM of Household Debt and the

Macro-economy. (PhD. Thesis) Department of Economics, Finance and Legal Studies,Graduate School of The University of Alabama.

Page 11: What causes household debt to increase in South Africa?

492 C. Meniago et al. / Economic Modelling 33 (2013) 482–492

Petersen, M.A., Senosi, M.C., Mukuddem-Petersen, J., 2012. Subprime Mortgage Models.Nova Science Publishers, New York.

Philbrick, P., Gustafsson, P., 2010. Australian Household Debt: An Empirical Investigationinto the Determinants of the Rise in the Debt-to-Income Ratio. (Masters Thesis) LundUniversity Australia, Australia.

Phillips, R.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrika335–346.

SARB, 2012. Quarterly Bulletin. South African Reserve Bank, Pretoria.

Schwarz, G., 1978. Estimating the dimension of a model. The Annals of Statistics 6,461–464.

Stengos, T., Wu, X., 2006. Information-theoretic distribution test with application tonormality. Discussion Paper. Department of Economics, University of Cyprus.

The Organisation for Economic Co-operation and Development (OECD), 2012. OECD(Available at www.oecd.org/[accessed November 2012]).

Zaystev, O., 2010. The Impact of Oil Price Changes on the Macroeconomic Performanceof Ukraine. (Masters Dissertation) Kyiv School of Economics, Ukraine.