Machine Learning Algorithms: The Unforeseen Contributor to Financial Inclusion
By Lynn Musembi
Think about the countless times you’ve paid for a good or service using a credit/debit card let alone a mobile wallet on your phone, the times you’ve sent money to a friend via mobile. Now imagine, suddenly and somehow, you don’t have your credit/ debit card, digital wallets are merely a construct yet to be explored, and banking services simply ceased to exist. So what’s left? Well, good old cash (and yes, all the money you would have had stored in bank accounts is now available to you in the form of cash — stored/ saved somewhere…just not in the safety of a bank). Now, how do you pay your rent? What about the loan you wanted to take out to pay for college? Wait, with just cash, how would you even pay for college? Notice the problem…we need to deviate from cash! Digital? Mobile? Keep reading to find out!
The Reality of Financial Services Around the World
The ease and safety that comes with traditional banking services is a privilege that we take for granted yet many do not get to revel in. Financial products and services remain inaccessible and unaffordable (insufficient income) to many.
1.4 billion adults are currently unbanked. That’s nearly a fourth of the global population. This unfortunately is the reality for many.
The unbanked: Individuals lacking access to traditional banking services: think savings accounts, credit services — including access to credit and loans, formal payment services such as digital payment strategies, insurance products and many other services.
Factors Contributing to a Population’s Lack of Access to Banking Services
The financial inclusion issue, particularly with the prevalence of unbanked communities, stems from a variety of factors that limit the accessibility and affordability of formal financial services. Among the underlying propagators of limited access to banking services are:
Limited Financial Infrastructure: The scarcity of banks and the lack of ATMs in some regions — especially rural communities — make it difficult for individuals to access financial services.
Lack of documentation: The required documents and forms of identification may serve as a barrier for those who lack these materials.
Informal Economy: Economic activities that are not regulated by the government such as street vending, day labour and unregistered businesses often if not always involve cash transactions rather than formal banking services, contributing to a lack of financial transparency.
Low Income: Traditional banking services are usually accompanied by charges and fees while requiring individuals to constantly meet the set minimum balance level in their accounts. As low-income individuals often live paycheck to paycheck, such services aren’t affordable.
Machine learning algorithms: Expanding access and affordability of financial services
How can we solve these issues to make available to underserved communities the financial services that they need? The solution lies in Machine learning algorithms. Among the major problems that contribute to limited financial inclusion is the thin file problem.
Thin file problem: Tied to low income, cash-based transactions and a lack of documentation, the majority of the unbanked also face the thin file problem. This is when they lack adequate information to determine their creditworthiness, limiting their access to credit among other financial opportunities.
Traditional credit scoring models focus on data primarily based on an individual’s income level and credit history — and for unbanked individuals, this forms a barrier to receiving credit services as income levels are often volatile with incompetent credit histories if even existent. As lenders are risk averse, they steer away from lending to those with little to no credit data as it is conventionally impossible to develop a credit score and thus determine the risk associated with lending to the credit invisible, giving rise to the thin file problem. Financial technologies are coming up with innovative solutions to work around the issue to promote financial inclusion as there is unutilized potential that stems from this solvable problem. The solution? Alternative data analysis using machine learning (ML algorithms).
By allowing for the use of more than just the traditional methods of credit scoring which, although effective in most cases, overlook alternative data sources, machine learning analyzes additional information, creating inclusive credit scoring models. In this way, financial exclusion is minimized as possibly credit-worthy individuals can now receive a useful credit score using information they have that may not have been considered using the traditional credit scoring methods.
The role of ML in credit scoring
Machine learning (ML) for credit scoring addresses many of the limitations associated with traditional credit scoring models, especially the orthodox data used to develop credit scores. Specific to credit scoring and credit decisions, supervised learning is the primary strand of machine learning used. Among the subsets are: logistic regression, decision trees, random forests, neural networks and support vector machines
Supervised learning: A form of machine learning whereby a model is trained on labelled data and used to predict an outcome based on this labelled input data.
The role of ML in credit scoring about financial inclusion
Machine learning (ML) for credit scoring addresses many of the limitations associated with traditional credit scoring models, especially the orthodox data used to develop credit scores. Specific to credit scoring and credit decisions, supervised learning is the primary strand of machine learning used. Among the subsets are: logistic regression, decision trees, random forests, neural networks and support vector machines
Supervised learning: A form of machine learning whereby a model is trained on labelled data and used to predict an outcome based on this labelled input data
The process of ML in enhanced credit scoring
Data preparation
- The first step involves the collection of a wide range of data relevant to credit scoring that is representative of the entire population. This spans demographic information, employment details, financial history and any other relevant information. To ensure accuracy, the data is refined and preprocessed to handle missing values and outliers to ensure the data is formatted such that it’s ready for use by the ML algorithm.
Feature Selection
- After the data is prepped, relevant features/ variables are created or transformed to enhance the predictive power of the model. The sole use of relevant features is key to ensure the model can not only perform effectively on trained data but also new data, avoiding overfitting. Relevant features are specific to the desired output but may include: credit history, debt-to-income ratio, employment status etc.
Overfitting: An issue characterized by the ML model overly learning the data that it captures random fluctuations rather than underlying patterns which is the overall goal. As a result, the model does exceptionally well on trained data but falls short when it comes to new data.
Model Training
- The dataset is then split into training and testing sets such that the training set trains the model while the testing set remains unused until the testing phase.
- At this stage, the ML model, whether decision trees, random forests, logistic regression or support vector machines, uses the training set to learn the relationship between the variables and the outcome
Model Evaluation
- The model is then assessed on its performance which determines whether the algorithm can proceed into deployment stages where it is utilized for credit scoring and credit decisioning while under continual monitoring to ensure effective performance.
With machine learning, alternative data that is often overlooked by traditional banks, thus gathering information about potential borrowers without solely relying on traditional credit data, thus allowing for financial inclusion by considering a broader set of factors in addition to credit history and income data when assessing creditworthiness, thus expanding to cater for the credit invisible.
Leading Companies Working Toward Financial Inclusion
M-Pesa
Quote: M-Pesa is of Swahili origin with “pesa” meaning money and the “M” standing for mobile hence the translation: Mobile Banking
Origins
Deemed the pioneer of the concept of mobile banking, M-Pesa was launched in 2007 by Safaricom as a mobile financial service initially aimed at facilitating financial transactions as there was limited access to banking services: the majority of the Kenyan population was unbanked, highly dependent on cash transactions and reliant on remittances to sustain their livelihood. The cash-based economy meant money was either sent home using trusted individuals — which was quite risky as you may imagine — or the sender had to travel home to physically deliver the highly inefficient money. M-Pesa sought to minimize the reliance on cash by introducing mobile banking, capitalizing on the high mobile penetration in the nation.
Current strides in financial inclusion
Today, the platform has branched out into other countries, offering a wide range of services from mobile banking, microfinance and loans, to international remittances. Primarily mobile, M-Pesa has enabled countless users to access financial services remotely, affordably and conveniently.
M-Shwari
M-Shwari is of Swahili origin which closely translates to “good”: a reflection of the platform’s goal of providing “good” if not “better” financial services to underserved communities through mobile technologies and innovative banking practices
Origins
M-Shwari, a Kenyan-based savings and loans service that is exclusively mobile, leveraging the fact that 94% of Kenyan adults own and use a mobile phone. In this way, they pride themselves in their accessibility and wide outreach as they don’t rely on the limited traditional banking infrastructure like bank branches and ATMs
Fun fact: M-Shwari does not rely on physical bank branches with users able to manage their accounts and access services primarily through mobile
Current strides in financial inclusion
M-Shwari enables individuals with minimal financial records to open mobile accounts where they can conduct transactions, thus enabling them to build a savings history which contributes to their creditworthiness. In this way, individuals who would have otherwise faced the thin file problem due to a lack of sufficient borrowing and repayment data, are now provided with the opportunity to develop a history, simply by utilizing the data they are able to provide.
Conclusion
Financial Inclusion is seeing great progress with machine learning algorithms penetrating the space and expanding the accessibility and affordability of financial services.
In future articles, I will continue to explore the prospects of Machine Learning and Artificial Intelligence in financial inclusion relative to their contributions toward financial equity.
