Determinants of the conditional probability that a household has informal loans given liquidity constraints regarding access to credit banking channels

Journal of Behavioral and Experimental Finance 13 (2017) 16–24

Contents lists available at ScienceDirect

Journal of Behavioral and Experimental Finance journal homepage: www.elsevier.com/locate/jbef

Full length article

Determinants of the conditional probability that a household has informal loans given liquidity constraints regarding access to credit banking channels Luca Zanin Prometeia, G. Marconi 43, Bologna 40122, Italy

article

info

Article history: Received 12 December 2016 Accepted 10 February 2017 Available online 20 February 2017 JEL classification: C14 D14 E26 Keywords: Conditional probability Liquidity constraints Informal loans Semiparametric bivariate probit

abstract We use a semiparametric bivariate probit model to explore the determinants of the conditional probability that a household has informal loans given objective or subjective liquidity constraints regarding access to credit through banking channels. In our empirical study, we use Italian microdata on household income and wealth covering the 1995–2014 period. Our results emphasize that the most important trigger factors influencing the conditional probability of interest are debts in the form of both mortgage(s) and loan(s) and the unemployment status of the household head. Other trigger factors include a young age of the household head, residence in a large municipality, no home ownership (paying rent or free use), an equivalent income lower than 10,000 Euro, and a ratio of liquid assets to net annual income very close to zero. Understanding the factors associated with a household’s probability of taking out informal loans is important to gain knowledge about a phenomenon that is not tracked by official statistics. This knowledge is also useful to practitioners and policy-makers interested in providing new tailored financial services or solutions for reducing poverty risk. © 2017 Elsevier B.V. All rights reserved.

1. Introduction A household can decide to finance its consumption from a variety of sources, including labour income, accumulated monetary wealth, monetary inheritance or borrowing (e.g., Rubaszek and Serwa, 2014; Zanin, 2016b). Households that decide to borrow can often choose among the following options in obtaining a loan: (a) formal credit (i.e., from a financial institution) or informal credit (i.e., from a network of friends or relatives) as a preferred solution or unique choice; (b) informal loans as a substitute for formal channels when liquidity constraints are a factor (see also Benvenuti et al., 2015 and references therein); and (c) complementarity between both formal and informal credit. In developed countries, the banking system represents the main channel of access to credit for households (e.g., Modigliani and Brumberg, 1954; Friedman, 1956; Benvenuti et al., 2015). Unsurprisingly, the literature on the forms of financial intermediation for such countries is thus rich in studies both that reveal the factors associated with the entry and sustainability of households regarding bank credit and that explore events such as unemployment that trigger increased probabilities of default or delays in the re-payment of debt (e.g., Banasik et al., 2003;

E-mail address: [email protected]. http://dx.doi.org/10.1016/j.jbef.2017.02.002 2214-6350/© 2017 Elsevier B.V. All rights reserved.

Aristei and Gallo, 2016; Tian et al., 2016; Thomas et al., 2016). Other studies have documented the difficulties that some families face in finding lenders in the presence of adverse selection and moral hazard, as these factors can lead to information asymmetry between borrowers and lenders (e.g., Berger et al., 2011; Benvenuti et al., 2015; Agarwal et al., 2016). In this regard, when issues such as liquidity constraints preclude access to bank credit, informal channels can play an important role for households seeking credit. Informal lenders are likely to be able to gather more information about such a household than financial institutions, which can contribute to reducing problems involving both adverse selection and moral hazard (e.g., Ghatak, 1999; Giné, 2011; Lee and Persson, 2016). Notably, households that seek loans from networks of friends or relatives frequently encounter either benefits and pitfalls. As a non-exhaustive list, the benefits include the following possibilities: 1. Obtaining loans without security (or without physical collateral) or with less security than is required by banks, as informal loans benefit from ‘social collateral’ typically associated with kinship and/or friendship between the lender and borrower (Karaivanov and Kessler, 2015). 2. Obtaining interest-free loans or loans at lower rates than can be found in the banking system.

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

3. Repaying informal lenders more flexibly than formal lenders. 4. Obtaining loans faster and more easily than through formal credit channels because informal lenders often do not require an assessment process (in terms of time and bureaucracy) that is as onerous as that required from a formal lender. Among the pitfalls are the following factors: 1. Informal lenders have limited resources, particularly in comparison to financial institutions. 2. Borrowers may not have an available network of friends or relatives to lend money. 3. The borrower’s default can jeopardize lender–borrower kinship ties or friendship (Karaivanov and Kessler, 2015). The literature on informal loans has mainly focused on case studies of households in developing countries (e.g., Madestam, 2014). To the best of our knowledge, few studies have focused on developed countries in this regard, and most of these have focused on Italy (e.g., Benvenuti et al., 2015). Among the available studies, Benvenuti et al. (2015) provided some of the initial interesting evidence regarding the relationships between liquidity constraints and informal loans in Italy. These authors considered the following two typologies of liquidity constraints, in particular: (a) the household has applied for a loan or a mortgage and has been rejected by a financial intermediary (objective event traceable from financial institutions); or (b) the household had considered applying for a loan or mortgage from a financial institution, but decided against it based on the belief that it would be rejected (subjective event not traceable from financial institutions). Their empirical analysis was performed using a Bank of Italy survey on household income and wealth. Notably, the survey represents an important point of reference for long-term analyses of informal transfers because information is available from 1995. Examining the 1995–2012 period, the authors used both a fully parametric linear and logit model and found a positive and statistically significant relationship between liquidity constraints and informal loans. We will extend the Benvenuti et al. (2015) study from the following perspectives:

• First, we include the micro-data that refer to 2014 to enrich the descriptive analysis of the time series of informal loans and liquidity constraints. Specifically, we have decided to consider the same liquidity constraints as Benvenuti et al. (2015). • Second, we estimate the conditional probability that a household has taken out informal loans given liquidity constraints by considering the possible dependence between the two events. Conditional probability is an interesting measure of the probability of an event ‘A’ given that another event ‘B’ has occurred. This measure is appealing when using cross-sectional data because a temporal or causal relationship between event ‘A’ (informal debt) and event ‘B’ (liquidity constraints) is not required. To achieve this aim, we specify a flexible bivariate probit model, as proposed by Radice et al. (2016). Such a modelling method let us explore the possible non-linear dependence between the variables of interest using several copula functions, whereas the relationship between continuous covariates and response is modelled using a penalized spline approach. In a classic, fully parametric modelling method, continuous covariates typically enter the model as linear, polynomial, or categorical terms based on the a priori assumptions imposed by the researcher. However, the correct functional form is rarely known a priori, and there is a risk that the model will incur incorrect specifications, which in turn can lead to inconsistent parameter estimates (e.g., Zanin, 2015). Splines thus represent a valid solution when the functional shape is not known a priori, while the penalty prevents overfitting.

17

• Third, using the proposed flexible bivariate probit model, we provide novel empirical evidence for the factors associated with the probability that a household has objective or subjective liquidity constraints. • Fourth, we produce evidence regarding the marginal effects of the selected explanatory variables on the outcomes of interest, particularly the effects that such factors have on the conditional probability that a household has taken out informal loans given liquidity constraints (or not). We have included in the models some new explanatory variables not considered in the studies discussed above. These variables include the ratio between liquid assets and net annual income, the equivalent income, the ratio between the unemployment rate at the regional level and the unemployment rate at the national level, and the typology of formal debt (if present). Improving knowledge in this manner might help practitioners from banks and insurance companies to assess the characteristics of households that are likely to be interested in new tailored financial services. For example, insurance companies might want to assess the potential market for an insurance policy tailored to informal transfers (loans) to cover creditors against borrower default. Economists, sociologists and policy-makers might be interested in the sphere of informal loans to improve knowledge about the management of family budgets, including poverty issues. The remainder of this article is organized as follows. In Section 2, we present the microdata used for the proposed analysis. In Section 3, we briefly describe the econometric model, and we discuss the main results in Section 4. Section 5 provides the main conclusions of the study. 2. Data The Bank of Italy conducts a biennial national survey of household income and wealth using a structured questionnaire to collect information about the socio-demographic characteristics of household members, labour conditions, income sources, wealth composition, and debt, among other data. Since 1995, the questionnaire has included a section covering informal loans. The cross-sectional data used in our analysis consider the 1995–2014 period. The microdata consist of a sample of 79,137 households (approximately 8,000 households for each survey). We restricted our analysis to the sample of households with a household head between 20 and 95 years of age because of the small number of observations outside of this age range. The household questionnaire was addressed to a person of reference (typically the household head), who responded on behalf of all the household members. Below, we describe the variables of primary interest. 2.1. Informal loans and liquidity constraints In this section, we describe the variables used to determine the conditional probability that a household has informal loans given liquidity constraints regarding access to banking channels to credit. Informal loans from friends or relatives: The survey includes a question regarding the presence of debts among Italian households in the form of informal loans. Specifically, the question asks, ‘‘At the end of the last year, did the household have any debt owed to relatives or friends?’’ Based on this question, we defined a binary variable with a value of 1 if the household declared any debts with friends or relatives and 0 otherwise. Plot (a) in Fig. 1 shows that the percentage of households with informal loans was decreasing from 1995 to 2002. Beginning in 2002 (the year in which the Euro become the official currency in Italy), we observe a progressive return to growth in the group of families

18

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

(a) Informal loans.

(b) Liquidity constraints. Fig. 1. Time series of the percentage of households with informal loans and liquidity constraints.

with informal loans (peaking at 3.5% in 2012). We perform a macro-area analysis and note a higher incidence of informal loans in southern Italy than in northern parts of the country. This evidence reflects the typical north–south dichotomy observable in Italy for several macroeconomic and credit market indicators (e.g., northern Italy has a higher GDP per capita, higher net wealth and higher disposable household income than southern Italy; D’Alessio, 2012; Mauro, 2004; Cracolici et al., 2009). Therefore, we can assume that informal loans might act as a substitute for or alternative to bank credit in those areas of the country that are most economically disadvantaged and where access to bank credit might be most constrained. Households in central Italy have evinced only a minor interest in informal loans, particularly during the 2000–2008 period. Notably, we also found a higher average value of private transfers in northern Italy (approximately 9,000 Euro) than in southern Italy (approximately 5,000 Euro). This gap reflects the spatial differences in the wealth distribution of households (see, e.g., Zanin, 2016a). Liquidity constraints: As proposed by Benvenuti et al. (2015), we consider two typologies of liquidity constraints: rejected applications (where the household has applied for a loan or a mortgage and has been rejected by a financial intermediary) and discouraged households (where the household has considered the idea of applying for a loan or mortgage from a financial company but then decided against it, believing that their application would be rejected). This information is collected by dedicated questions present in the family questionnaire. We have defined a binary variable that takes the value of 1 if the household has a liquidity constraint due to a rejected application or if the household is discouraged and 0 otherwise. Plot (b) in Fig. 1 shows at the national level that the incidence of households with liquidity constraints reached a peak during the global financial crisis (2008–2010), with higher levels in northern and southern Italy than in central Italy. Focusing on the nature of liquidity constraints, Fig. 2 reports the trends in the incidence of discouraged households (plot 2(a)) and rejected applications (plot 2(b)). We note that ‘discouragement’ plays a dominant role in the overall measure of the incidence of households with liquidity constraints. This information is important because it is not accounted for in any official credit institution or governmental statistics. Specifically, plot (a) provides evidence regarding the sharp rise in the incidence of discouraged households in 2006 (particularly in the northern regions), which coincides with the maximum expansion of the credit market (see

also Benvenuti et al., 2015), followed by a decline beginning in 2010. In southern Italy, a new important increase in the incidence of discouraged households occurred in 2014. From plot (b) of Fig. 2, we note instead that the incidence of households receiving a ‘rejection’ is much smoother over time than the incidence of discouraged households, with a higher incidence in the southern than in the northern parts of the country (see also D’Alessio, 2012; Benvenuti et al., 2015). 2.2. Explanatory variables In estimating the conditional probability that a household has taken out informal loans given liquidity constraints, we have included in the proposed model a number of socio-economic and demographic variables related to the household head, to the household as a whole, and to a certain macroeconomic context. In Table 1, we provide a detailed description of these variables, while a descriptive analysis is presented in Table 3. Among the variables characterizing the household head, we included age, marital status and professional status. Specifically, the assumption is that the probability that a household is saddled with liquidity constraints and informal loans decreases with the age of the household head, whereas an increase is expected for households with unemployed and divorced heads. Compared to married and employed household heads, divorced and unemployed household heads have a higher probability of having liquidity constraints and of having taken out informal loans as a consequence of worsening economic well-being (e.g., Mazzucco et al., 2009; Zanin, 2016b). These reasons are associated with an increase in economic uncertainty at the household level and thus an increase in difficulty accessing the formal credit market (e.g., Zanin, 2016a). In both cases, the network of relatives or friends can be fundamental to economic support. Earnings from labour and monetary wealth accumulated by the household (e.g., liquidity, government bonds, mutual funds, stocks) are further factors affecting the outcomes of interest. Specifically, these variables are essential for formal lenders to assess the credit rating of a household. At this juncture, it is notable that we have considered equivalent income (deflated by the consumer price index) and a measure of the liquidity ratio. The first measure allows us to compare the incomes of different households (in terms of size and age composition), whereas the second indicator is useful for measuring the number of months during which a household can meet its consumption needs after an income loss. We have

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

(a) Discouraged households.

19

(b) Rejected application.

Fig. 2. Time series of the percentage of households believing that their requests would be rejected (a) or that applied for a loan or a mortgage and were rejected by a financial intermediary (b). Table 1 Descriptions of the outcome variables and explanatory factors. Variable Outcome Liquidity constraints

Definition

Informal loans

1 = The household has received a rejection from a bank after having applied for credit (loan or mortgage) in year T , or the household considered applying for a loan or a mortgage at a bank or another financial intermediary but then decided against it based on the expectation that the application would be rejected. 1 = At the end of year T , the household has debts with relatives or friends.

Socio-demographic information about the household head Age Professional status Marital status

Age of the household head (years) 1 = Employed; 2 = Unemployed; 3 = Retired; 4 = In another condition. 1 = Married; 2 = Single; 3 = Divorced; 4 = Widower.

Variables at the household level Equivalent income (Euro/1,000)

Liquidity ratio Debt with banks Dwelling of residence Size of the municipality of residence Macro-area of residence (area) Macroeconomic variable Ratio (unemployment rate)

The equivalent income is calculated using the OECD modified scale (Hagenaars et al., 1994; De Vos and Zaidi, 1997) as follows: Net Annual Household Income / index of consumer price (base year 2014) Equivalent income = 1+0.5×(no. adults−1)+0.3×(no. child under 14) Measured as the ratio between household liquid assets and net annual income (De Vaney, 1994). 1 = No debt with banks (loan or mortgage); 2 = Debt with banks—only loan(s); 3 = Debt with banks—only mortgage(s); 4 = Debt with banks—loan(s) and mortgage(s). 1 = Home ownership; 2 = No ownership, rents; 3 = No ownership, free use. 1 = Up to 40,000 inhabitants; 2 = From 40,000 to 500,000 inhabitants; 3 = More than 500,000 inhabitants. 1 = Northern Italy; 2 = Central Italy; 3 = Southern Italy. Measured as the ratio between the annual unemployment rate at the regional level and the annual unemployment rate at the national level.

also considered factors regarding the dwelling of residence and the size of the municipality in terms of inhabitants. Both factors are expected to be positively correlated with both outcomes of interest (liquidity constraints and informal loans). As discussed in the introduction, the informal channel can be considered as exclusive or as complementary to formal credit (e.g., Myers and Majluf, 1984; Guiso and Jappelli, 1991; Benvenuti et al., 2015; Lee and Persson, 2016). To explore this aspect of informal credit, we have included a categorical variable describing the possible position of a household with respect to formal credit: no debt with a financial institution (loan or mortgage); only loan(s); only mortgage(s); and both loan(s) and mortgage(s). As a factor of a macroeconomic nature, we considered the ratio between the annual unemployment rate at the regional level and the annual unemployment rate at the national level to capture the tensions in the labour market that can vary in magnitude over time, depending on the economic cycle and the productive characteristics of an area (see, e.g., Okun’s law; Zanin and Marra, 2012a; Zanin, 2016c).

Uncertainties in the labour market can increase income risk and thus the probability that a family has either objective or subjective liquidity constraints, contributing, in turn, to increased difficulty in accessing the formal credit market. Because of data constraints, we are not able to control for other factors that might potentially be associated with informal loans, such as the characteristics of the network of friends available as lenders, time preferences, and the psychological traits of the household head, to name a few. 3. Econometric model Let y1 and y2 be a pair of binary variables determined as follows: y1 = 1 identifies a household with liquidity constraints, while y2 = 1 identifies a household with informal loans. The focus is on the estimation of the measure of conditional probability, P(y2 = 1|y1 = 1). By assuming dependence between the two marginal distributions, P(y1 ) and P(y2 ), a convenient setting for estimating

20

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

the conditional probability is to use a bivariate model within a flexible framework, as proposed by Radice et al. (2016). Formally, the general specification of the bivariate probit model is written as follows: y1i = η1i = y2i = η2i =

K1 

xk1 i αk1 +

J1 

k 1 =1

j 1 =1

K2

J2 

 k2 =1

xk2 i αk2 +

s2j2 (z2j2 i ) + ε2i ,

j 2 =1

(1)

where n is the sample size, and η1 and η2 are the linear predictors of the respective equations. The term x represents the binary and categorical variables (including the intercept), as listed in Table 1, while α is the associated parameter to be estimated. s represents the unknown one-dimensional smooth functions of the continuous covariates (i.e., equivalent income, liquidity ratio, age, year of survey, and the ratio between the annual unemployment rate at the regional level and the annual unemployment rate at the national level). For a generic continuous covariate z, the smooth function is given as a linear combination of known (cubic) spline bases, bk (z ), and unknown regression parameters, δk . In other K words, s(z ) = k=1 δk bk (z ), where K is the number of bases. Calculating bk (zi ) ∀ k and i yields K curves that encompass different degrees of complexity, which are multiplied by some real valued parameters δk and then summed to give an estimated curve for s(z ) (e.g., Wood, 2006; Zanin and Marra, 2012b; Zanin, 2015). Smooth  components are subject to identifiability constraints, such as i s(zi ) = 0. Determining the appropriate number of base splines is a relatively complex task. For this reason, the use of a penalized estimation approach allows for the suppression of that part of the smooth term complexity that the data do not support. One important advantage of using a flexible model over a fully parametric approach is that continuous predictors can be modelled without making any a priori assumptions of linearity or non-linearity. In this manner, the modelling method allows for the relaxation of the assumptions of specific functional forms to reduce the risk of model mis-specification (e.g., Zanin, 2015). We have also included the interaction between the year of the survey and the macro-area of residence of the household in considering spatiotemporal variations during the model fitting process. The joint cumulative distribution function (cdf) of events investigated can be studied using a different copula structure (Radice et al., 2016; Winkelmann, 2012). As a general representation:

P(y1i , y2i ) = C (P(y1i ), P(y2i ); θ ) with P(y1i = 1) = Φ (η1i ) and P(y2i = 1) = Φ (η2i ), which represent the two marginal outcome variables, where η is the linear predictor as described in (1) and Φ (.) is the cdf of the univariate standard normal distribution. C is the chosen copula function, while θ is a parameter that measures the association between the two marginals. We have considered the following types of dependence: Gaussian, Clayton, Gumbel, Joe, and Frank (see Radice et al., 2016; Trivedi and Zimmer, 2009). For practitioners, the parameter θ cannot be easy to interpret, and a Kendall’s τ measure in the range of −1 to +1 can be derived (Brechmann and Schepsmeier, 2013). This measure is zero if there is independence between the two equations. The log-likelihood function for a bivariate probit can be expressed as follows: l =

n 

y1i y2i log P(y1i = 1, y2i = 1) + y1i (1 − y2i )

i=1

× log P(y1i = 1, y2i = 0) + y2i (1 − y1i ) × log P(y1i = 0, y2i = 1) + (1 − y1i )(1 − y2i ) × log P(y1i = 0, y2i = 0)

P(y1i = 1, y2i = 0) = P(y1i = 1) − P(y1i = 1, y2i = 1); P(y1i = 0, y2i = 1) = P(y2i = 1) − P(y1i = 1, y2i = 1); P(y1i = 0, y2i = 0) = 1 − [P(y1i = 1) + P(y2i = 1) − P(y1i = 1, y2i = 1)] .

s1j1 (z1j1 i ) + ε1i

i = 1, . . . , n,

where

Fully computational details on the estimation process of model (1) can be found in Radice et al. (2016). The model was estimated using the R package SemiParBIVProbit (Marra and Radice, 2016). After estimating model (1), the effects of the covariates on the outcome are computed as the average change (increase or decrease) in the probability of the outcome given a change in the explanatory variable, such as a one-unit change in the case of binary variables, holding the other covariates at their observed values. 4. Results Before analysing the effects of the determinants of P(y2 = 1|y1 = 1), we comment on the results obtained estimating model (1) using different copula functions. Table 2 reports the results for the AIC, BIC, Vuong and Clarke tests (Vuong, 1989; Clarke, 2007) and the Kendall’s τ of all the models fitted. The Vuong (1989) and Clarke (2007) tests are likelihood-ratio-based tests for model selection that can be employed to choose between two non-nested bivariate models. The Vuong and Clarke tests are computed for models ‘‘(1)(b)’’–‘‘(1)(e)’’. The comparison is performed with respect to a traditional model specification, i.e., ‘‘(1)(a)’’. In general, the computed indicators do not indicate a preference for a model specification that uses a copula function different from Gaussian. For this reason, we comment on the results obtained from model ‘‘(1)(a)’’. The results obtained from models ‘‘(1)(b)’’–‘‘(1)(e)’’ are available upon request. The Kendall’s τ measurement indicates a positive and statistically significant association between the two equations (0.27; CI: 0.25, 0.29), which thus supports the assumption of dependence between P(y1 ) and P(y2 ) and a preference for the use of a bivariate model to estimate and study the determinants of the measure of interest: P(y2 = 1|y1 = 1). Table 3 presents a descriptive analysis of the selected explanatory variables, the estimated parametric components of both equations of model (1), and the marginal effects of Pr(Y2 = 1|Y1 = 1) and Pr(Y2 = 1|Y1 = 0). We also included the marginal effects of Pr(Y1 = 1) and Pr(Y2 = 1). Figs. 3 and 4 report the smooth function estimates of the spatio-temporal component and the other continuous covariates, respectively. We begin by discussing the marginal effects obtained from the univariate (marginal) predicted probability of success Pr(Y1 = 1). We observe that households with liquidity constraints are likely to have at least one loan obtained from a financial intermediary (debt with banks—only loan(s): ME equals 0.032), not to live in a house of their own (ME equals 0.026), to have household heads who are unemployed (ME equals to 0.016), and to be economically poor. Focusing on household economics, we find a linear decreasing relationship between the probability of liquidity constraints and the measure of liquidity ratio (ME equals −0.002). In other words, the higher the ratio between savings and household income, the higher the probability that a household is protected from the risk of incurring liquidity constraints of either objective or subjective nature. The annual equivalent income instead shows a non-linear impact on outcome with a vertiginous increase in the probability of incurring liquidity constraints for households below the line of poverty risk (less than 10,000 Euro). We also observe a decreasing non-linear pattern for the relationship between the age of the household head and outcome, with a reduction that is accentuated for heads of households

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

21

Table 2 Model selection based on different copula models. 95% confidence intervals for  τ are obtained using a Bayesian posterior simulation (Radice et al., 2016). Model

Copula

AIC

BIC

(1)(a) (1)(b) (1)(c) (1)(d) (1)(e)

Gaussian Gumbel Clayton Frank Joe

36,223 36,229 36,226 36,278 36,294

37,202 37,206 37,210 37,254 37,268

Vuong’s test

Clarke’s test

a

b

a

c

b

b

b

b

 τ (95% CIs) 0.27 (0.25,0.29) 0.37 (0.34,0.39) 0.12 (0.11,0.13) 0.40 (0.37,0.44) 0.60 (0.58,0.63)

We compare each estimated model with Model (1)(a) based on Vuong’s test and Clarke’s test. a It is not possible to discriminate with regard to Model (1)(a). b Model (1)(a) is preferred. c Model (1)(c) is preferred.

60 years old and over. This finding is coherent with the model of the life-cycle profile of income and consumption.1 The ratio between the unemployment rate at the regional level and the unemployment rate registered at the national level presents a non-linear increasing pattern, suggesting that liquidity constraints are most likely in areas in which the unemployment rate is a crucial issue (typically in southern Italy as opposed to northern Italy; see, e.g., Mauro, 2004; Cracolici et al., 2009; Zanin and Calabrese, 2017). The higher the gap is between the unemployment rate of an area and the value registered at the national level, the higher the probability is that a household can suffer issues associated with liquidity constraints (in particular, as discussed in Section 3, liquidity constraints of a subjective nature). The smooth functions that capture the spatio-temporal patterns of liquidity constraints show an increasing trend during recent years for southern Italy and the islands, a decreasing trend for northern Italy, and a substantially stable trend for central Italy. Regarding the determinants of informal loans, we focus on the main factors affecting the probability that a household has debts with friends or relatives given objective or subjective liquidity constraints regarding access to credit through banking channels. The most important factor is represented by having both mortgage(s) and consumer credit (loans). In this regard, the probability of having informal loans given liquidity constraints (Pr(Y2 = 1|Y1 = 1)) increases by 13.8 percentage points, compared with households without bank credit. An increasing probability of 7.5 and 3.8 percentage points is instead estimated for households with only mortgage(s) or only loan(s), respectively. The unemployment status of the household head is another important factor associated with the probability of informal loans. The conditional probability that a household with an unemployed head has informal loans given liquidity constraints is 9.7 percentage points higher than that of a household with an employed head, whereas the increase is only 2.7 percentage points in the absence of liquidity constraints. The loss of household income (sometimes unique in single-income families) contributes to economic uncertainty in the family budget (e.g., Giarda, 2013; Brunetti et al, 2016; Zanin, 2016a). This factor represents a crucial obstacle to access to formal credit because it weakens one or more of the guarantees required by the financial intermediary. In this

1 During the initial periods, individuals have relatively low income levels, and they borrow with formal lenders having the expectation that their income will increase in the future. However, if low income levels are accompanied by factors of uncertainty from the labour market (for instance, if they are employed under temporary contracts), it is likely that issues of liquidity constraints will arise (i.e., receiving a rejection from a financial intermediary or believing that a loan application would be rejected). With advanced age, greater work experience and stabilization in the labour market, individuals are expected to earn higher incomes, and they tend to accumulate financial assets to protect against expected income decreases during retirement (see also, Zanin, 2016b). After retirement, individuals likely use their financial savings to maintain consumption above their pension level, and they become less inclined to take loans. Under this scenario, an individual/household is less likely to incur issues of liquidity constrains. For further details on the life-cycle model and labour market issues, please refer to Rubaszek and Serwa (2014) and Zanin and Calabrese (2017) and the references therein.

context, a network of friends or relatives can play a crucial role in supporting the household at risk of poverty or social exclusion. The issue is particularly accentuated in those areas of the country with the highest unemployment rates compared to the value registered at the national level. The profile of the household more inclined to be eligible for an informal loan given liquidity constraints is also characterized by a young age of the household head (up to 40 years old), residence in a large municipality (more than 500,000 inhabitants), and not owning a house (paying rent or free use). Moreover, the profile also includes that the household’s equivalent income level is less than 10,000 Euro, and the measure of the liquidity ratio is very close to zero. Specifically, we found that the household’s equivalent income follows a non-linear decreasing pattern that it is not possible to capture assuming a priori a linear relationship with outcome variables. For an average equivalent income of 17,000 Euro, we estimate that a one standard deviation change contributes to decreases of 2 percentage points in the Pr(Y2 = 1|Y1 = 1). Regarding the measure of the liquidity ratio, a value of the indicator less than 0.25 indicates that the household can maintain the same standard of living in the absence of labour income and can cover its expenses for less than 3 months; at a value between 0.25 and 0.5, the household can cover its expenses for 3–6 months; whereas a value greater than 0.5 indicates that the household can cover its expenses for more than 6 months. Here, we note that the highest impact of the variable on Pr(Y2 = 1|Y1 = 1) is for values of the indicator very close to zero. Based upon these results, we can conclude that when a household has issues of liquidity constraints in accessing new banking credit, the network of friends or relatives might represent a valid substitute. However, what occurs if a family does not find support from this informal network? Why might an informal lender not grant a loan? Due to data availability limitations, we are not able to explore such issues. We can, however, assume some of the motivations associated with this adverse scenario. For example, an informal lender might perceive a higher risk of loss on a loan that he or she might not be willing to accept. Alternatively, the household might have a network of friends or relatives in similar economic conditions. From this perspective, how does the household react to constraints in access to both formal and informal credit channels? Future research is required to close this important gap in the literature. The findings that emerged provide important insights for both policy-makers and practitioners interested in understanding the factors that trigger the probability that a household has liquidity constraints and thereby is motivated to contact its network of relatives and friends for financial support. In particular, the results can support practitioners in proposing new financial services tailored to a specific population or sub-population of households. 5. Conclusions and policy recommendations Bank credit is the channel of financing most used by families in developed countries. However, in the presence of liquidity

22

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

Table 3 Estimated effects of explanatory variables on outcome (Y1 = liquidity constraints; Y2 = informal loans) obtained by estimating model (1) with a Gaussian copula. The covariates’ effects on the outcomes are provided as average marginal effects. For each discrete variable, the marginal effect measures the change (increase or decrease) in the probability of the outcome, given a one-unit change in the explanatory variable. For each continuous variable, the marginal effect is measured as the change (the increase or decrease) in the probability of the outcome, given a one standard deviation change in the explanatory variable from the mean value. The standard deviations for the continuous variables are reported in square brackets. Variables

Descriptive (proportion/mean)

Linear predictor

Marginal effects

Y1

Y2

Pr(Y1 = 1)

Pr(Y2 = 1)

Pr(Y2 = 1|Y1 = 1) Pr(Y2 = 1|Y1 = 0)

Socio-demographic factors of household head Age

56.07 [16.42]

–

–

−0.015**

−0.009**

−0.019**

−0.006**

Professional status Employed Unemployed Retired In another condition

58.3% 2.1% 38.1% 1.6%

– 0.189** −0.108** −0.096

– 0.434** −0.077* 0.093

– 0.016** −0.007** −0.006

– 0.033** −0.004** 0.005

– 0.097** −0.007** 0.033

– 0.027** −0.003** 0.005

Marital status Single Married Divorced Widower

17.4% 60.5% 6.3% 15.8%

−0.029

0.007 – 0.109** 0.100**

−0.002

– 0.049 −0.045

– 0.003 −0.003

0.000 – 0.006** 0.005**

0.005 – 0.021** 0.030**

0.001 – 0.005** 0.005**

Variables at the household level Equivalent income (Euro/1,000) Liquidity ratio

17.05 [10.49] 0.89 [1.71]

– –

– –

−0.003** −0.002**

−0.004** −0.002**

−0.020** −0.009**

−0.004** −0.002**

No debt with banks (loan or mortgage) Debt with banks—only loan(s) Debt with banks—only mortgage(s) Debt with banks—loan(s) & mortgage(s)

79.5% 9.2% 9.1% 2.2%

– 0.383** 0.042* 0.181**

– 0.328** 0.313** 0.580**

– 0.032** 0.003* 0.013**

– 0.019** 0.018** 0.043**

– 0.038** 0.075** 0.138**

– 0.014** 0.016** 0.037**

Dwelling of residence Home ownership No ownership, rent No ownership, free use

67.9% 21.7% 10.4%

– 0.347** 0.228**

– 0.208** 0.172**

– 0.026** 0.015**

– 0.011** 0.009**

– 0.014** 0.017**

– 0.008** 0.007**

Size of the municipality of residence Up to 40,000 inhabitants From 40,000 to 500,000 inhabitants More than 500,000 inhabitants

61.1% 26.1% 12.8%

– 0.029 0.025**

– 0.083** 0.121**

– 0.002* 0.002**

– 0.004** 0.006**

– 0.017** 0.026**

– 0.004** 0.005**

Macroeconomic variable Ratio (unemployment rate)

1.03 [0.58]

–

–

0.028**

0.008**

0.002**

0.004**

−1.877**

−2.194**

Intercept * **

P-value: < 0.1. P-value: < 0.05.

Fig. 3. Estimated smooth function component of the temporal variable for each macro-area. The results on the scale of the linear predictor are obtained from estimating model (1) with a Gaussian copula. The shaded areas represent 95% confidence intervals. The rug plot at the bottom of each graph shows the covariate values. The number in brackets on the y-axis report the estimated degrees of freedom (edf) of the smooth curves. P-values associated with each smooth term are less than 0.05.

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

23

Fig. 4. Estimated smooth function components for each continuous covariate. The results of the scale of the linear predictor are obtained from the estimations of model (1) with a Gaussian copula. Shaded areas represent 95% confidence intervals. The rug plot at the bottom of each graph shows the covariate values. The number in brackets on the y-axis reports the estimated degrees of freedom (edf) of the smooth curves. P-values associated with each smooth term are less than 0.05.

constraints, informal loans obtained from friends or relatives can serve as a possible alternative or substitute. Using microdata from the Italian survey on household income and wealth for the 1995–2014 period, we explored the determinants of the conditional probability that a household has one or more informal loans given objective (application rejected) or subjective (discouraged household) liquidity constraints. To achieve this aim, we specify a bivariate probit model within a flexible framework, as proposed by Radice et al. (2016). The modelling method employed allows us to explore the possible non-linear dependence between the two outcomes of interest by assessing several copula functions (Gaussian, Clayton, Gumbel, Joe, and Frank), whereas the relationship between continuous covariates and the response is modelled using a penalized spline approach. The findings suggest that the profile of the household that is most interested in informal loans given liquidity constraints is mainly characterized as having obtained both mortgage(s) and loan(s) obtained from the banking system (whereas a less positive effect is registered for households that have with only mortgage(s) or loan(s)), have unemployed household heads, have young household heads (up to 40 years old), and reside in large municipalities (more than 500,000 inhabitants). Regarding economic factors, the profile of households that are likelier to obtain informal loans favours households making an equivalent income level of less than 10,000 Euro and a measure of liquidity ratio that is very close to zero. In comparing the estimated marginal effects for Pr(Y2 = 1|Y1 = 1) and Pr(Y2 = 1|Y1 = 0), we emphasize the important role played by liquidity constraints in increasing the probability that a household has taken out one or more loans from its network of friends or relatives. The planning of effective policies to prevent (or alleviate) events of default is important in the sphere of informal loans to avoid the deterioration of a household’s economic well-being and ties of kinship or friendship between lenders and borrowers. The implementation of preventive actions thus represents an important aim for economic agents. However, this is a challenging task because there are no statistics or in-depth knowledge about those factors that most affect households that decide to borrow informally as a substitutive, preferred or complementary solution

to formal credit. The information available mainly comes from sample surveys. We believe that practitioners and policy-makers can benefit from the evidence that emerged in this study, which provides a starting point to determine the characteristics of the households most likely to seek informal loans, particularly conditioned by the presence of liquidity constraints. This information is useful for developing the most appropriate prevention strategies, such as new tailored financial services, or for creating points of access to informal credit markets outside of the network of friends and relatives. Importantly, some factors included in the model are known to practitioners and policy-makers (such as typologies of formal credit, employment status, and residence) and might be considered important markers for identifying the target households that are most likely be interested in informal loans and insurance policies tailored to cover events due to borrower defaults (see also Eurofound, 2013). Further exploration might be oriented towards answering certain open questions regarding informal loans, such as (a) the role of arrears (in the repayment of formal credit, the payment of rent for the house of residence, payment of utilities); (b) the weight and importance of the informal sphere in the definition of a household as over-indebted; and (c) the characteristics of households that are (un)willing to offer informal loans. Acknowledgment We would like to thank one anonymous reviewer for suggestions that have contributed to improving the presentation and quality of the article. The opinions expressed herein are those of the author and do not reflect those of the institution of affiliation. Conflict of Interest The authors declare that they have no conflict of interest. References Agarwal, S., Chomsisengphet, S., Liu, C., 2016. An empirical analysis of information asymmetry in home equity lending. J. Financ. Serv. Res. 49, 101–119. Aristei, D., Gallo, M., 2016. The determinants of households’ repayment difficulties on mortgage loans: evidence from Italian microdata. Int. J. Consum. Stud. 40, 453–465.

24

L. Zanin / Journal of Behavioral and Experimental Finance 13 (2017) 16–24

Banasik, J., Crook, J., Thomas, L., 2003. Sample selection bias in credit scoring models. J. Oper. Res. Soc. 54, 822–832. Benvenuti, M., Casolaro, L., Ciani, E., 2015. Informal loans, liquidity constraints and local credit supply: evidence from Italy. Working Paper. Berger, A.N., Espinosa-Vega, M.A., Frame, W.S., Miller, N.H., 2011. Why do borrowers pledge collateral? New empirical evidence on the role of asymmetric information. J. Financ. Intermed. 20, 55–70. Brechmann, E.C., Schepsmeier, U., 2013. Modeling dependence with c- and d-vine copulas: the R package CDvine. J. Stat. Softw. 52, 1–27. Brunetti, M., Giarda, E., Torricelli, C., 2016. Is financial fragility a matter of illiquidity? An appraisal for Italian households. Rev. Income Wealth 62, 628–649. Clarke, K., 2007. A simple distribution-free test for non-nested model selection. Polit. Anal. 15, 347–363. Cracolici, M.F., Cuffaro, M., Nijkamp, P., 2009. A spatial analysis on Italian unemployment differences. Stat. Methods Appl. 18, 275–291. D’Alessio, G., 2012. Ricchezza e disuguaglianza in Italia. Occasional Papers, n. 115, Bank of Italy. De Vaney, S.A., 1994. The usefulness of financial ratios as predictors of household insolvency: two perspectives. Financ. Couns. Plann. 5, 5–26. De Vos, K., Zaidi, M.A., 1997. Equivalence scale sensitivity of poverty statistics for the member states of the European community. Rev. Income Wealth 43, 319–333. Eurofound - European Foundation for the improvement and working conditions, 2013. Household over-indebtedness in the EU: the role of informal debts. Dublin. Friedman, M., 1956. A Theory of the Consumption Function. Princeton University Press, Princeton NJ. Ghatak, M., 1999. Group lending, local information and peer selection. J. Dev. Econ. 60, 27–50. Giarda, E., 2013. Persistency of financial distress amongst Italian households: Evidence from dynamic models for binary panel data. J. Banking Finance 37, 3425–3434. Giné, X., 2011. Access to capital in rural Thailand: an estimated model of formal vs informal credit. J. Dev. Econ. 96, 16–29. Guiso, L., Jappelli, T., 1991. Intergenerational transfers and capital market imperfections. Eur. Econ. Rev. 35, 103–120. Hagenaars, A.J.M., De Vos, K., Zaidi, M.A., 1994. Poverty statistics in the late 1980s: research based on micro-data. Technical report. Office for Official Publications of the European Communities, Luxembourg. Karaivanov, A., Kessler, A., 2015. A friend in need is a friend indeed: theory and evidence on the (dis) advantages of informal loans. SFU working paper. Lee, S., Persson, P., 2016. Financing from family and friends. Rev. Financ. Stud. 29, 2341–2386. Madestam, A., 2014. Informal finance: a theory of moneylenders. J. Dev. Econ. 107, 157–174. Marra, G., Radice, R., 2016. SemiParBIVProbit: Semiparametric Bivariate Probit Modelling, R package version 3.7–1.

Mauro, L., 2004. The macroeconomics of Italy: a regional perspective. J. Policy Model. 26, 927–944. Mazzucco, S., Meggiolaro, S., Ongaro, F., 2009. Economic consequences of union dissolution in Italy: findings from the European community household panel. Eur. J. Popul. 25, 45–65. Modigliani, F., Brumberg, R., 1954. Utility analysis and the consumption function: An interpretation of cross-section data. In: Kurihara, K.K. (Ed.), Post-Keynesian Economics. Myers, S., Majluf, N., 1984. Corporate financing and investment decisions when firms have information that investors do not have. J. Financ. Econ. 13, 187–221. Radice, R., Marra, G., Wojtys, M., 2016. Copula regression spline models for binary outcomes. Stat. Comput. 26, 981–995. Rubaszek, M., Serwa, D., 2014. Determinants of credit to households: an approach using the life-cycle model. Econ. Syst. 38, 572–587. Tian, C.Y., Quercia, R.G., Riley, S., 2016. Unemployment as an adverse trigger event for mortgage default. J. Real Estate Financ. Econ. 52, 28–49. Thomas, L.C., Matuszyk, A., So, M.C., Mues, C., Moore, A., 2016. Modelling repayment patterns in the collections process for unsecured consumer debt: A case study. European J. Oper. Res. 249, 476–486. Trivedi, P.K., Zimmer, D., 2009. Pitfalls in modeling dependence structures: explorations with copulas. In: Castle, J., Shephard, N. (Eds.), The Methodology and Practice of Econometrics. Oxford University Press. Winkelmann, R., 2012. Copula bivariate probit models: with an application to medical expenditures. Health Econ. 21, 1444–1455. Wood, S.N., 2006. Generalized Additive Models: An Introduction with R. Chapman & Hall. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57, 307–333. Zanin, L., 2015. Determinants of risk attitudes using sample surveys: the implications of a high rate of nonresponse. J. Behav. Exp. Financ. 8, 44–53. Zanin, L., 2016a. On Italian households’ economic inadequacy using qualiquantitative measures. Soc. Indic. Res. 128, 59–88. Zanin, L., 2016b. The effects of various motives to save money on the propensity of Italian households to allocate an unexpected inheritance towards consumption. Qual. Quant. http://dx.doi.org/10.1007/s11135-016-0364-8. Zanin, L., 2016c. The pyramid of Okun’s coefficient for Italy. Empirica - J. Eur. Econ. http://dx.doi.org/10.1007/s10663-016-9343-5. Zanin, L., Calabrese, R., 2017. Interaction effects of region-level GDP per capita and age on labour market transition rates in Italy. Available at SSRN: https://ssrn.com/abstract=2916018. Zanin, L., Marra, G., 2012a. Rolling regression versus time-varying coefficient modelling: an empirical investigation of the Okun’s law in some Euro area countries. Bull. Econ. Res. 64, 91–108. Zanin, L., Marra, G., 2012b. A comparative study of the use of generalized additive models and generalized linear models in tourism research. Int. J. Tourism Res. 14, 451–468.