Social capital, phone call activities and borrower default in mobile micro-lending

Highlights

• Phone call activities can be used effectively as a measure of social capital to explain and predict borrower default.

• Individual-level transaction data provided by one of the world's largest mobile lending platforms are utilized.

• Incoming and outgoing calls have different implications for social capital, and thus borrower default.

• Calling activities associated with stronger social ties have greater predictive power for loan defaults than those associated with weaker ties.

Abstract

This study examines how the social capital of borrowers affects loan defaults in the burgeoning mobile micro-lending market. We analyze the individual-level transaction data provided by one of the world's largest mobile lending platforms. Focusing on identifying behavior-based predictors of financial transactions, we propose that mobile phone calling activities constitute a valuable measure of social capital. Drawing on the theoretical foundation of social capital theory, we identify and study two types of calling activities: incoming calls and outgoing calls, strong ties and weak ties. Our analysis shows that the more incoming calls a borrower usually receives, the less likely he or she is to default on a loan. However, the more outgoing calls a borrower makes, the more likely default will occur. We further find that calling activities associated with stronger social ties have greater predictive power for loan defaults than those associated with weaker ties. These findings demonstrate the relevance of phone call activities in consumers' financial decisions. They provide micro-lending companies with valuable alternatives for assessing borrower creditworthiness beyond “hard information”, such as credit scores and income, and further help them make more effective loan decisions.

Introduction

As a key component of Fintech applications, micro-lending has emerged in recent years as a popular funding channel for consumers [21,39,55,56,62]. Micro-lending enables individuals to borrow directly, often in relatively small amounts, from lenders via the internet or mobile platforms without traditional financial institutions acting as intermediaries [55,62]. It is expected that the loan origination of this market in the United States will soon amount to $90 billion and the global market value will reach $290 billion.¹

Predicting consumer credit risk is a crucial task for financial institutions when making decisions regarding loans and other transactions in the traditional credit market, P2P lending, and online consumer credit service markets [48]. It has further attracted substantial attention in decision support systems. This is particularly true for micro-lending because customers on these platforms often lack credit history and belong to low-income groups that are vulnerable to credit risks [14]. In fact, the default rate is much higher in micro-lending than in the conventional credit market, where borrowers are often required to provide collateral [18,20]. The default rate on Prosper.com, the largest micro-lending platform in the United States, is over 8.48% on average and can reach 30% if the economy experiences a recession.² In contrast, the default rate of consumer loans by commercial banks in the United States is, on average, 2.79%.³ If not managed properly, a high default rate has the potential to hamper and severely damage the micro-lending market, which benefits millions of consumers.

Financial institutions generally use “hard information” such as income, assets, and debt history to verify borrowers' creditworthiness. For example, the Fair Isaac Corporation (FICO) combines a borrower's credit history and debt level to determine his or her credit score, which subsequently influences the provision of credit [60]. However, borrowers in the micro-lending market vary widely in terms of their background and credit history. Further, micro-lending companies often have very limited information about borrowers, and it is well known that hard information does not always provide a full picture of creditworthiness, especially for prediction purposes [45]. These realities have prompted lenders in these markets to increasingly turn to “soft information” to better understand borrowers [38].⁴ Existing research suggests that gathering soft information is crucial for limiting lending risk when hard information is insufficient [50]. Various aspects of “soft information”, such as descriptive loan texts, lender-borrower communication, and social connections, can be extracted to predict a borrower's credit risk [58,61]. For example, the micro-lending company Lenddo uses social network data (i.e., the number of followers a borrower has and with whom they are friends) to evaluate loan applicants.

Previous studies have increasingly focused on the role of another source of soft information, that is, calling activities, in predicting default probability in online micro-lending platforms (e.g., [42,63]). This study aims to identify common behavior-based soft information as a predictor of loan default, establish a causal link between calling activities and borrower default, and look deeper into the heterogeneous effects of calling activities. Our study is situated in mobile micro-lending, a fast-growing segment of micro-lending that enables borrowers to initiate and finalize transactions on mobile devices such as smartphones and tablets. Theoretically, we propose that as a source of soft information, the social capital of borrowers can be measured through daily user behaviors that occur routinely on mobile devices. These measurements, together with the information traditionally used in borrower assessment, can then be utilized to predict default risk.

Specifically, we focus on the connection between mobile phone calling activities and the likelihood of default on mobile microloans. The core argument is that social capital comes from social connections [40,51], which prior research has demonstrated as a typical example of soft information (e.g., [39]). Calling activities on mobile phones, which have become a hub that facilitates social connections, constitute key indicators of resources and social capital [8,32]. By leveraging borrowers' calling activities to measure their social capital, we discern the different types of calling activities and elucidate their heterogeneous effects on borrower default. In a nutshell, our study uniquely links financial behavior on mobile devices (i.e., mobile loan default) to the primary function of these devices (i.e., phone calls).

Our results show that with any sensitive information of users protected, the pure patterns of incoming versus outgoing calls on mobile phones in the past are strongly correlated with the likelihood of borrower default in the future. Before borrowers sign up as users on the mobile micro-lending platform, those who receive more incoming calls in their daily lives are less likely to default in future transactions than those who make more outgoing calls. We also find that mobile phone calling activities associated with strong ties, measured by connections with people who are on the borrower's contact list, are a more powerful predictor of default than calling activities with weak ties (i.e., people who are not on the borrower's contact list).

This study has important theoretical contributions and practical implications. First, by advancing the knowledge of common and easily accessible behavior-based predictors of financial activities, we provide evidence for the efficacy of “soft information” in explaining and predicting borrower default. Most previous research relies on hard information, such as demographic data or credit history, to analyze borrower defaults in the context of traditional lending. However, mobile micro-lending customers tend not to provide adequate hard information because of insufficient credit records and low income [14]. Fintech applications can obtain more soft information [56], especially concerning borrowers' social connections. In the context of mobile micro-lending, this study identifies a novel measure of soft information—mobile phone calling activity—and demonstrates how it is connected to borrower default.

Second, our study extends the literature on soft information augmentation to provide more accurate credit risk predictions. Drawing on social capital theory, our study establishes the causal connection between soft information and the behavior of borrowers through empirical design, in addition to revealing the role of social capital in explaining how different types of phone calls are connected to borrower default. Previous studies investigated the correlation and predictive power of soft information, and mostly adopted a data-driven approach without providing a theoretical underpin when modeling the credit risk of consumers.

Third, our study adds to the social capital literature on micro-lending. A small but growing body of research exploits the role of social capital in online micro-lending platforms such as Prosper.com (e.g., [20,39]). In contrast to these platforms where the social network of a user can be traced based on his or her user groups, mobile micro-lending platforms do not provide users with tools to socialize with each other; user activity on these platforms is highly individualized behavior.⁵ Therefore, it is important to identify individual behaviors that provide insights into social capital. Furthermore, we explore heterogeneities in the impact of social capital on borrower default in terms of social tie strength.

In conclusion, this study provides decision support for risk management. Using available and valuable information from phone calls helps lenders distinguish the risk level of the borrower through a more fine-grained partitioning of different types of phone calls, and consequently make more effective loan decisions, which further helps them improve the quality of loan approval and reduce loan default and costs.

The remainder of this paper is organized as follows. In the next section, we discuss the related literature and provide theoretical arguments that relate mobile calling activities to the likelihood of loan default. Next, we introduce the data, variables, and the empirical model. We then present and discuss our results. The paper concludes with theoretical and managerial implications, as well as several directions for future research.