Keywords

JEL Classification

1 Introduction

It was clearly revealed by the US mortgage crisis in 2008 that financial institutions have considerable weaknesses in evaluating consumer credit risk. The ensuing economic effects spilled over globally and attention turned to consumer credit in emerging markets (Küçüktalasli et al. 2012). Thanks to financial innovations during the last two decades, households in emerging markets have gained easier access to consumer loans (Livshits et al. 2014). Therefore it has become more significant for banks to tailor the consumer credits according to the risk characteristics of households. Perraudin and Sorensen (1992)’s explicitly show that banks mainly depend on demographic characteristics of consumers while assessing credit applications and constructing the consumer loans. Swain (2008) also emphasizes the role of these characteristics on household default risk because they provide useful information in times of change. Consequently, the objective of this study is to determine the impact of demographical and financial attributes of Turkish households aftermath the 2008 global financial crisis.

This paper aims to increase the information for financial institutions that target to be lenders for the consumer loan market for Turkish households, hence contribute to the well functioning of the Turkey’s credit market. In other words, through the results of the publicly available survey in Turkey, we aim to determine the risk profile of Turkish consumers. Therefore, this paper informs banks, building societies, credit card institutions and large retailers about the risk characteristics of their potential borrowers. In specific, the institutions are acknowledged about particular socioeconomic and demographic features of Turkish households and informed on which households they must either target or avoid. Moreover, our work also contributes to efforts for maximizing wealth of Turkish consumers given that borrowers face credit constraints when lenders are not able to obtain sufficient or necessary information to evaluate their credit risks correctly. Nonetheless, this improvement in the demand side of the credit market is called as the democratization of credit lending by Lyons (2003).

In recent years, banks have increasingly been using credit-scoring techniques based on these characteristics to evaluate the loan applications they receive from consumers (Blöchlinger and Leippold 2006; Karan and Arslan 2008). For Turkey, Küçüktalasli et al. (2012) examine the 5120 Turkish households during the economic expansion period of Turkey, which is 2006 and 2007 and right before the global crisis, and point out the importance of socioeconomic and demographic characteristics of consumers on quality of services in the financial sector.

However, in Turkey, the global financial crisis caused dramatic restrictions on the supply side of the consumer loans. Banks started to build reserves and hoard cash and hence limit the pool for extending consumer loans. Right after the global financial crisis Turkey experienced a significant rise in household liabilities reaching up to 10.8 % and the saving rate has declined to the historically low level (Duman 2013). This has caused Turkish Banking Regulation and Supervision Agency (BRSA) to restrain private credit growth in Turkey aftermath the global crisis followed by this sharp drop in the household savings rate and simultaneous explosive growth in private credit in Turkey. Alarmed by such a trend BRSA decided to take action to curb the unsustainable credit growth, and in December 2010, it officially announced a set of steps to be implemented as a part of credit restraint policy. Agarwal et al. (2015) indicate that the credit tightening policy has an immediate and economically strong effect on consumer spending in Turkey.

Altogether, we need to understand which socioeconomic and demographic characteristics of households are likely to lead to default at a specific time period when access to external finance for households are more restricted however household spending (saving) markedly rises (falls) compared to the pre-crisis period.

Being one of the world’s leading emerging economies and a candidate for the European Union membership, Turkey’s consumer credit market offers vast opportunities to foreign financial institutions. Besides, Turkey’s consumer finance industry has been rapidly growing since the end of the domestic financial crisis in the year 2001. This outcome is mainly owed to the rising disposable income on back of robust economic growth and increasing employment since the end of the crisis. Moreover, since the last quarter of the year 2002, foreign banks have been intensely merging, taking over or acquiring Turkish domestic banks. Specifically, the equity ownership of foreign banks in the Turkish banking system rose from less than 4 % in 2002 to more than 40 % by 2013. The swift entry of foreign banks into Turkish financial system has instantly escalated competition in domestic financial market, particularly in consumer lending. Thus, the quantity of increase in Turkish consumer credits since the last quarter of the year 2002 aligns with the rise of the share of foreign banks in the Turkish banking system (The Banks Association of Turkey 2014).

As indicated by Getter (2006), a well functioning credit market is constituted once borrowers are treated differently in terms of credit availability and charged rates. Consequently, the well functioning of the consumer credit market in Turkey depends on the amount of information that enables financial institutions to differentiate good quality borrowers from bad quality ones. Nevertheless, in the theoretical framework of Cutts et al. (2000) the expansion of consumer credit markets is shown to depend on the availability of information about borrowers.

Our analyses targeting at detecting demographic and financial characteristics of Turkish households are applied on two stages. First we use logistic regressions to detect which features are likely to lead to default. Second we make robustness test with Survival Cox Analysis and evaluate the impacts of these features again. Our verified results show that the features that affect the default in consumer credit are; household income, volatility of the household income, being a home owner, age, education and gender.

The remainder of this paper is as follows. In the next section we review the related literature. After that, we describe the data and methodology, followed by the explanation of the results. In the final section our paper provides the concluding remarks.

2 Literature

Demographic and socioeconomic characteristics of households are extensively studied in the previous literature. One of the most elaborated characteristics is the gender of the household head. The literature generally stresses on the fact that women have a lower disposable income compared to men (See, among others, Bajtelsmit and Bernasek 1996). Besides, Anbar and Eker (2009) stress on that, compared to women, man are more knowledgeable about financial concepts. Despite earning a lower income and being less informed about the financial issues, the literature has not found a significant difference in financial risk taking between women and men (Embrey and Fox 1997; Harrison et al. 2007; Schubert et al. 2000). Overall, we expect to find women in our analyses to be more susceptible to consumer credit default.

Age is also one of the characteristics highly studied in the previous literature. Ardehali (2004) show that financial risk tolerance falls as the age of an individual increases. Wagner (2011) explicitly indicate that wealth of an individual increases as his age and he becomes more financially savvy as well. Consequently, we expect the likelihood of default to be higher for households with younger heads.

Education is also found to be an important attribute influencing the probability of consumer credit default. Initially, as the education level of an individual rises, thanks to the enhanced skills, the likelihood of being employed in a high-income job increases as well (Hallahan et al. 2003; Sultana and Pardhasaradhi 2011). Besides, education also contributes to the enhancement of valuation and assessment skills and therefore individuals become more disciplined and efficient in their investments and spending (Grable and Lytton 1999 and Ardehali 2004). As such, we expect the education to be adversely related with consumer credit defaults.

For the period before the global financial crisis, Arslan and Karan (2010) finds that being a home owner in Turkey imposes not only a discipline on spending but also stabilizes the household income. That home is also a collateraziable fixed asset for a household. We also expect that owning the house, which is actually lived in by the household, is negatively related with consumer default during the post-crisis period as well.

Finally household income is a very important determinant of borrower creditworthiness (See, Getter 2006 for a detailed discussion). Consumer’s income not only extends the borrowing capacity but also repayment potential (Arslan and Karan 2010). Therefore we expect the households in the lower income brackets to be more prone to credit defaults. We also take into account of the stability of the household income and extend our prediction that a higher volatility in household income has a positive influence in credit default.

3 Data

The data on this study is obtained from the “Income and Living Conditions Survey” conducted by the Turkish Statistical Institute (henceforth; TURKSTAT) between the years 2008 and 2012. The survey covers the entire members of the households that live within the borders of Turkey. The survey is conducted consistently each year between April and July. The surveys are carried out through face to face interviews. The survey employs the rotational design whilst a group of households stay in the sample frame from one year to another, new households enter the sampling frame. Around 75 % of the sample stays in the panel from year to year.

Individuals that are 13 years and older in the selected basic sampling household are monitored for 4 years. The variables on income, house-ownership and financial failure belong to the entire household. However the data on age, education and gender belong to the household head. The total number of observations is based on an unbalanced panel of 13,979 households between the years 2008 and 2012

Definitions of each variable are provided in Table 1. All variables are the arithmetic mean of the sample.

Table 1 Variables
Full size table

The first variable is “Time” and it shows the total number of years a household is monitored. This is the critical variable in the Survival Analysis as it follows the financial failure of households within the years. If a household fails its financial obligations, its monitoring is ceased. As such, for the households with no financial failure, “Time” represents the total number of years it is followed since the beginning of our time period. “Default” is a dummy variable taking the value of 1 if a household fails a payment of consumer credit, credit card, installment, loan or rent, and zero otherwise. 1274 households in our data set is found to be “Default” whereas 12,704 households have not defaulted any of their financial obligations between 2008 and 2012. “Household Income” is the natural logarithm of the total household income. “Income Deviation” is the standard deviation of the household income within the sample period and it measures the variability of the income. “Age” and “Education” are the average age and education categories of the household head, respectively. “Gender” is a dummy variable taking the value of 1 if a household head is a woman and zero if man. Lastly “House Owner” is a dummy variable taking the value of 1 if the household owns the house currently being lived, and zero otherwise.

Table 2 shows descriptive statistics of the variables. On average a household is followed for 2.73 years and majority of the sample has not defaulted in any of the financial obligations. On average, the age group of the household head falls between 20 and 24 years and a middle school graduate. Moreover 62 % of the households in our study own their own houses. Finally, a female heads 15 % of the sample household.

Table 2 Descriptive statistics
Full size table

Table 3 provides the correlation coefficients across the variables. The results in the table show that the correlation between the variables does not exceed 50 % and therefore we are under no threat of multicollinearity.

Table 3 Correlation matrix
Full size table

Table 4 reports univariate results on the differences between default and non-default group. As expected, compared to the non-default group, default households have a higher deviation in their income during the sample period and their household head is younger. The education level of default-group is slightly lower than the non-default group. This can be suggested by the lower spending discipline or lower income with lower number of schooling of the household heads. It is also observed that more of the non-default group’s head is female than that of the default-group. Lastly, household income is significantly lower in default group as the home ownership is less as well.

Table 4 Univariate tests
Full size table

4 Methodology

For our multivariate analyses we first employ logistic regression methodology followed by the survival analysis for validation of the findings from the previous estimations. Specifically, logit model is used for estimating the probability of participants failing in financial obligations included in the TURKSTAT database given that most of the econometric models are usually thought of as only being suitable for target variables that are continuous. With logistic regression, we are estimating a divided outcome. This condition creates problems for the assumptions of ordinary least squares that the error variances (residuals) are normally distributed. Therefore at a usual linear regression equation, an algebraic conversion is needed.

Linear regression models are similar with other general linear equations like;

$$ \mathrm{Y} = {\mathrm{a}}_0+{\mathrm{a}}_1{\mathrm{X}}_1+{\mathrm{a}}_2{\mathrm{X}}_2+....+{\mathrm{a}}_{\mathrm{n}}{\mathrm{X}}_{\mathrm{n}}+\mathrm{e} $$

Where, Y is regressing for the probability of a categorical outcome. In simplest form, this means that we're considering just one outcome variable and two states of that variable—either 0 or 1. The following is the equation for the probability of Y=1;

$$ P\left( Y=1\right)=\frac{1}{1+{e}^{-\left({a}_0+{\displaystyle \sum {a}_i{X}_i}\right)}} $$

Independent variables (Xi) of logistic regression can be continuous or binary and perfectly fits our approach, which is aimed to identify personal variables (continuous and binary) of two groups namely early leavers and faithful participants of pension company.

Survival analysis is special technique dealing with the time it takes for something happen or failure. The main goal of the survival analysis is to explain the proportion of cases, which are observed at various times (Tabachnick and Fidell 2013). Similar with the logistic regression, survival analysis use hazard rates, which is the probability of any event occurring at any point in time. Despite many advantages of logistic regression, survival analysis deals with time dependent variables better than logit models (Wang et al. 2013). If some of the variables change over time, using survival analysis leads more precise estimations. One of the most interesting characteristics of the analysis is that survival time is unknown for many cases. Predictor variable of the survival analysis is always the time. The criterion variable, which is often called as the status variable is the Default, which is the dependent variable of the logistic regression. In this case, there are two possible outcomes: Default (1) or not default (0).

5 Results

Here we first present results from logistic regression analysis. The results are demonstrated in Table 5.

Table 5 Logistic regressions
Full size table

The table shows that as the variability of a household’s income increases it is found that the household is likely to default. The variability of household income increases the risk of the cash inflows and therefore increases the probability of not being able to service consumer debt obligations. Households headed by younger individuals are also likely to end up with a default. The result on the age is the reflection of the fact that right after the 2008 global crisis, youth unemployment has entered into a rising trend in Turkey. For instance the percentage of youth unemployment in the country in 2009 was 25.3 % and has not changed considerably as of now. Put differently, having an older household head increases the odds of cash flow security. We also find slight evidence that having a woman led household is likely to be associated with default. Woman’s participation to work force is 12 % of the population and it is the lowest among OECD countries. Moreover in Turkey women has the lowest leading role such as CEO, CFO or vice president etc., in corporations among OECD countries. Therefore being led by a female household head is likely to be associated with more volatile and lower income. Expectedly, household income and being a home owner is likely to be negatively related to the default. In Turkey, being a homeowner is also likely to be associated with not only having better saving propensity but also disciplined spending behavior (Arslan and Karan 2010). Lastly, lower is the education level, it is more likely that the household will end up with a default. This result for the after-crisis period in Turkey also holds for the pre-crisis period as well (Arslan and Karan 2010).

In order to verify our findings, we apply the Survival Analysis and the results are presented in Table 6. In the first model, the results for all the variables, except for the standard deviation of household income, are consistent with our previous findings. Figure 1 the survival rate falls faster in the first year and has been decreasing each year.

Table 6 Survival analysis
Full size table
Fig. 1
figure 1

General survival function of the model with all variables

Full size image

In the second model, age of the household head is used as the grouping variable. The ones older than 30 are denominated as old and the rest as young. Then it is found that the results for “Education”, “Gender”, “Household Income” and “House Owner” go hand in hand with the previous findings. Survival rates belonging to this group are presented at Fig. 2 showing that survival rate is lower in the younger population and has been falling through the years.

Fig. 2
figure 2

Survival function—age of the household head is the grouping variable

Full size image

In the third model, education of the household is used as the grouping variable. Households are grouped as; the ones with at least a college degree and those with a high school degree and less. Results for all the usual four variables are consistent with that of the logistic regressions except for the gender. This result can be attributed to the fact that women in Turkey usually have a homogenous level of education, being typically lower than that of men. Figure 3 contains the survival rates related to this model. The figure shows that survival rate is particularly lower for the ones that have a lower level of education and has been decreasing through the years.

Fig. 3
figure 3

Survival function—education of the household head is the grouping variable

Full size image

6 Conclusions

Right after the global financial crisis, Turkish consumer loan market is characterized by a demand side that contributed to the high growth of the market yet hand in hand with a significant drop in the household savings rate. Alarmed by such a trend, BRSA decided to take action to curb the unsustainable credit growth immediately after the crisis by officially announcing a set of steps to be implemented as a part of credit restraint policy. This situation also intersects with the stricter lending policies of banks for building reserves.

The objective of this study is to determine which demographical and financial features affect consumer credits risks of Turkish households aftermath the 2008 global financial crisis. This period had significant characteristics in terms of the forces of expansionary demand yet a contraction prone supply side of the Turkish consumer market.

Our analyses are built on the data from Turkish Statistical Institute (TURKSTAT) on an unbalanced panel of 13,979 households between the years 2008 and 2012. We apply our estimations on two stages. First we use logistic regressions to detect which features are likely to lead to default. Second we make robustness test with Survival Cox Analysis and evaluate the impacts of these features again. Our verified results show that the features that affect the default are; household income, volatility of the household income, age, being a home owner, education and gender.