Creditworthiness Prediction in Microfinance using Social Network and Mobile Data
COM2 Level 4
Executive Classroom, COM2-04-02
Abstract:
In many emerging markets, lack of sophisticated credit reporting systems or credit bureaus makes creditworthiness assessment challenging. In the hopes of boosting credit growth, other forms of credit and credit assessment have emerged, such as microfinance. Nevertheless, in the absence of formal credit-scoring methods for microfinance institutions, the return from such ventures is often fraught with high risk and uncertainty. An interesting observation, however, is that in these markets, the explosive growth of Internet penetration and mobile device adoption provides plentiful social and locational information. Drawing on these technologies, in a networked context I propose two studies leveraging two important data sources, social network data from online social network sites and geo-location data from mobile devices, to circumvent the problem of predicting creditworthiness in resource-poor communities. In particular, my studies propose to use network theory to characterize latent borrower information. In the first study, I propose a Bayesian method for social network-based credit scoring that helps to address network sparsity and data scarcity, which is common with ego-centric networks as found in the credit scoring context in emerging and undeveloped markets. The empirical results from this study suggest that the proposed method can improve the creditworthiness prediction in microfinance by up to 300% as compared to traditional methods. In the second study, I propose to predict micro-loan defaults using spatial networks based on co-location events (i.e. borrowers appear at roughly the same place at approximately the same time). The co-location network is established in a way that does not require borrowers to interact or communicate directly. I find a significant and positive relationship between co-located borrowers' loan default behaviors. The empirical results suggest that incorporating location information improves the creditworthiness prediction by at least 100%. My dissertation makes a number of important methodical and network-construction related contributions by relaxing the assumption of availability of complete interaction data. More importantly, the results from my research can not only illuminate how social and mobile data might be used in assessing creditworthiness, but also empower microfinance companies in resource-poor communities with novel methods for credit scoring. The use of financial technology (FinTech), including Internet and mobile based financial services, helps expand access to financial services among unbanked people, thereby increasing financial inclusion.
In many emerging markets, lack of sophisticated credit reporting systems or credit bureaus makes creditworthiness assessment challenging. In the hopes of boosting credit growth, other forms of credit and credit assessment have emerged, such as microfinance. Nevertheless, in the absence of formal credit-scoring methods for microfinance institutions, the return from such ventures is often fraught with high risk and uncertainty. An interesting observation, however, is that in these markets, the explosive growth of Internet penetration and mobile device adoption provides plentiful social and locational information. Drawing on these technologies, in a networked context I propose two studies leveraging two important data sources, social network data from online social network sites and geo-location data from mobile devices, to circumvent the problem of predicting creditworthiness in resource-poor communities. In particular, my studies propose to use network theory to characterize latent borrower information. In the first study, I propose a Bayesian method for social network-based credit scoring that helps to address network sparsity and data scarcity, which is common with ego-centric networks as found in the credit scoring context in emerging and undeveloped markets. The empirical results from this study suggest that the proposed method can improve the creditworthiness prediction in microfinance by up to 300% as compared to traditional methods. In the second study, I propose to predict micro-loan defaults using spatial networks based on co-location events (i.e. borrowers appear at roughly the same place at approximately the same time). The co-location network is established in a way that does not require borrowers to interact or communicate directly. I find a significant and positive relationship between co-located borrowers' loan default behaviors. The empirical results suggest that incorporating location information improves the creditworthiness prediction by at least 100%. My dissertation makes a number of important methodical and network-construction related contributions by relaxing the assumption of availability of complete interaction data. More importantly, the results from my research can not only illuminate how social and mobile data might be used in assessing creditworthiness, but also empower microfinance companies in resource-poor communities with novel methods for credit scoring. The use of financial technology (FinTech), including Internet and mobile based financial services, helps expand access to financial services among unbanked people, thereby increasing financial inclusion.