Banking Transformation – SME credit decisioning

Banking Transformation – SME credit decisioning

December 2022

Small and medium-sized enterprises (SMEs) contribute significantly to global economies, in both advanced and emerging markets. For instance, in the European Union (EU), SMEs represent 99% of all businesses, employ two thirds of the work force and account for more than half of the region’s GDP. These numbers can be even higher in emerging markets.

In addition to these impressive numbers, SMEs, as they are generally run by entrepreneurs, contribute significantly to innovation and shaping future economies. They are present in all industries where barriers to entry can be relatively low and do not require a large group of people at the beginning. Recently, the SMEs that have drawn the most attention are Fintech firms.

Historically, SMEs face significantly more difficulties in obtaining bank loans when compared with their larger counterparts. They have to rely more on internal funding or funding from family and friends. This is because (1) SMEs generally possess higher default risk than individuals or large corporates due to their higher failure rate and (2) there is generally a lack of quality data for banks to assess the creditworthiness of SMEs to support their lending decisions.

However, thanks to Fintechs and government support, the availability of both higher quality traditional data and alternative data is now growing. Banks are therefore reconsidering SME lending. In this paper, we will focus on discussing SME credit scoring, as a critical tool for SME lending.

 

KEY TRENDS IN SME LENDING MARKET

SMEs account for the vast majority of business establishments globally, and most of them seek loans from banks for business development. They represent a unique segment in the lending market, as they require very different services from retail customers or large corporations. For instance, an SME may not have a large finance team to provide comprehensive financial records and business data for banks to make credit decisions. It may not have the time or the resources to go through a long loan application process. As a result, they have become a tough nut to crack and therefore a somewhat underserved segment.

However, in an era of technology enablement and financial inclusion, things are changing fast. With technology that enables fast customer onboarding and screening and the availability of alternative data for risk assessment and monitoring, SMEs are coming into the spotlight.

We have observed five key trends in the SME lending market, including changing SME business models, faster banking processes, greater regulatory support on Fintech adoption, rising competition from newcomers and growing service offerings.

 

Figure 1: Key trends in SME lending market

Source: Accuracy

 

Changing SME business models

SMEs across various industries are being forced to alter their business models drastically in order to stay in business. As technology adoption and digitisation have become global trends, there is no exception for SMEs. The development of new business models (i.e. asset-light, online-based) makes it difficult for banks to assess their credit quality from a traditional perspective.

Faster banking processes

With the help of technology and alternative data, Fintechs and some traditional banks are now able to process SME loan applications much more quickly. For instance, Liberis, a UK Fintech that provides finance for small businesses, can interface with a company’s bank account and dashboards to enable access to instant funding, which may be retrieved in a matter of minutes for SMEs that need it.

Regulatory support on Fintech

Financial technologies have driven global innovation in financial services. At the same time, they are altering the nature of commerce and end-user expectations for financial services. Regulatory bodies are increasingly open to innovations and supportive of the adoption of Fintech solutions. As such, we have seen a number of Fintech companies specifically targeting SME finance in the past few years.

Rising competition from newcomers

Alternative lending providers such as tech giants are entering the battlefield. Big players like Google, Amazon and Tencent, as well as their more regional counterparts, have been putting pressure on banks for some time. There is a good probability that this pressure will increase as Techfins increasingly use their potent consumer franchises and advanced digital capabilities to outbid banks, especially in SME lending. Competition in the field will be increasingly intense.

Increase in service offerings

The SME banking sector has transformed as a result of changes brought by Fintechs, Techfins, government and regulatory support, and challenger banks. SME clients now have more options than ever to obtain access to financing. Banks must modify their SME offerings to compete in an environment where SMEs are looking for a suite of services (e.g. invoicing, corporate credit cards, payroll management) and one-stop-shop experiences.

 

 

TREDING THEMES IN SME CREDIT SCORING

To tap into and expand their business in SME lending, banks needs a series of tools, ranging from fast customer onboarding platforms to accurate risk assessment and monitoring tools. In this whitepaper, we dedicate our discussion to the use of credit scoring for risk assessment and customer acquisition.

Credit scoring is a statistical method for determining a borrower’s creditworthiness by combining a number of risk factors into a single score. We have observed several trends in SME credit scoring: (1) the use of alternative data and the adoption of data sharing platforms; (2) the adoption of advanced analytic solutions; and (3) the streamlining and automation of credit approval processes.

 

 

Figure 2: Key trends in SME credit scoring

Source: Accuracy

 

Alternative data and data sharing platforms

Traditionally, banks use a limited set of data to perform credit decisioning. These can be grouped into financial variables and non-financial variables. In recent years, the banking industry has seen a rise in the use of alternative data, which adds value throughout the customer banking lifecycle, particularly during the credit evaluation process. As the value of alternative data has gradually attracted more attention from banks and regulators, so has the concept of Open Banking and data sharing platforms. For example, the Commercial Data Interchange platform of the Hong Kong Monetary Authority (HKMA) aims to connect numerous SME data owners and providers with financial institutions for easier, faster and better credit assessment.

 

Figure 3: Types of credit scoring data

Source: Accuracy

 

Alternative data

Up until now, the majority of banks still assess potential borrowers’ creditworthiness using traditional credit data and methods. However, traditional data only captures the tip of the iceberg when it comes to the borrower’s information. The use of alternative data presents two attractive opportunities for banks. Firstly, alternative data help banks enhance model performance. Secondly, they help banks to expand the total addressable market (TAM) as the data make sound credit assessments possible. Meanwhile, the growth in computational power has effectively lifted the barrier of collecting and processing big data.

Alternative data are generated everywhere in the digital footprint of a company. The following figure describes typical examples of alternative data.

 

Figure 4: Alternative data providers and sources

Source: Accuracy

 

Data Sharing and Open Data

Governments and industries across the world are promoting the concept of open data and data sharing. The idea is to facilitate data transmission among various stakeholders to enhance overall efficiency. In some cases, banks are being urged to exchange customer data in a machine-readable format so that customers can access and securely transmit their banking information to reliable parties. This makes it easier for borrowers, especially SMEs, to switch financial service providers seamlessly, increasing availability of funding sources and opportunities.

One recent example of the use of alternative data in SME financing is the launch of the Commercial Data Interchange (CDI), a core pillar of HKMA’s “Fintech 2025” strategy. The HKMA officially launched the CDI in October 2022, with a proof-of-concept study dating back to November 2020. During the pilot launch phase, the CDI facilitated over HKD 1.6 billion of SME loans or 800+ loan cases. This initiative aimed to enable more efficient financial intermediation in the banking system and to facilitate the innovative use of commercial data to enhance financial services.

The CDI connects five types of stakeholder, namely, data owners (i.e. SMEs), data consumers (i.e. financial institutions), analytics service providers, solutions providers and data providers (i.e. commercial entities that collect the digital footprint of data owners). With the CDI, each bank and data provider has connections to the platform, making it simple for them to link their systems to the infrastructure for data access. This allows SMEs to share their digital footprints with their banks. The data help banks in a number of ways, including KYC, credit underwriting, product development, customer acquisition and credit monitoring. As we have highlighted, the use of alternative data is especially important for credit decisioning.

 

Figure 5: HKMA CDI initiative

Source: HKMA CDI, Accuracy

 

Hong Kong is not alone. The Data Governance Act, as approved by the European Parliament in April 2022 and applicable from September 2023, aims to increase data sharing in the EU so that businesses and start-ups can access more data. The regulations will allow greater use of data gathered in various public sector domains. They also enable the construction of shared European data platforms in various fields, including finance.

Additionally, policymakers in India, Japan, Singapore and South Korea are proposing a number of initiatives to encourage and accelerate the adoption of data sharing frameworks in the banking industry. For instance, the Monetary Authority of Singapore and the Association of Banks in Singapore have released an API playbook to promote data interchange and communication between banks and Fintechs.

Advanced analytics solutions

Banks are searching for better credit scoring techniques to improve the predictive power of their models. For example, they are developing or considering AI and machine learning for their credit scoring. These methodologies are more sensitive to real-time indicators of an SME’s creditworthiness than traditional credit scoring methods.

Decision tree and random forest are among the most commonly considered machine learning techniques that can be applied in SME credit scoring. Researchers are also exploring new solutions such as hybrid BWM and TOPSIS, when facing issues of insufficient data.[1] A detailed discussion on various advanced analytics solutions can be found in the next part of this paper.

Leveraging the enormous amounts of data gathered, Shopify, a leading all-in-one e-commerce platform that powers millions of businesses globally, has become a leader in using machine-learning techniques. Not surprisingly, it has launched Shopify Capital, a data-powered product that enables merchants to secure funding and accelerate their business growth. According to the company, Shopify has constructed Shopify Capital using a version of a recurrent neural network (RNN) that analyses more than 70 million data points across the Shopify platform to understand trends in merchants’ growth potential and provide cash advances that match their business needs. Since its inception, Shopify Capital has provided over USD 3.8 billion in funding.

Process streamlining & automation

In this digital era, another rising trend is the demand for seamless and automated credit approval processes. This trend is prominent in all phases of credit application, from client interactions to data collection (e.g. use of API, open banking, the CDI initiative), credit decision-making (automated models) and result communication (workflow streamlining and integration). Automation’s ultimate goal is to speed up the banking services for clients while reducing decision-making time, saving money, and improving productivity and efficiency for banks. In some cases, SME loan processing is part of the broader banking CRM suite, making it easier for banks to manage the whole customer lifecycle digitally.

Let’s take NeoGrowth, for example. NeoGrowth is a pioneer in SME lending in India, with a unique underwriting model based on digital transactions. The company has used technology to provide consumers with a smooth and seamless digital experience, where the entire process flow – from lead generation to loan origination, approval, disbursement and collections – is handled digitally.

[1] BWM – best worst method; TOPSIS – technique for order of preference by similarity to ideal solution

 

Figure 6: Digitally integrated operations in SME loan lifecycle

Source: NeoGrowth annual report, Accuracy

 

SME CREDIT SCORING APPROACHES

SME credit scoring refers to risk models to help financial institutions gauge SMEs’ creditworthiness and risk level. Traditionally, most banks use credit scorecards developed based on logistic regression for its simplicity to use and interpret. However, as we mentioned above, with the rapid development of data and analytics fields, some banks have started to adopt more advanced and dynamic models. In particular, decision models based on machine-learning techniques (e.g. decision tree, random forests) have gained popularity in recent years.

 

Figure 7: SME credit scoring flow

Source: Accuracy

 

Logistic regression-based credit scorecards

Credit scorecards developed by a combination of weight of evidence (WoE) transformation and logistic regression are among the most commonly used credit decision tools in banks. These credit scorecards were widely used in the past few decades; they have been well tested and have proved their effectiveness. Today, they are still the most popular scorecards used and maintained by banks, thanks to their simplicity to use and explain, while remaining effective.

There are seven important steps in developing a logistic-regression-based SME credit scorecard, including data processing, variable transformation and selection, logistic regression, performance inference, segmentation analysis, scorecard scaling and scorecard validation. In contrast to retail credit scorecards, segmentation analysis is usually performed for SME scorecards as companies in different sectors may exhibit very different risk characteristics. A detailed discussion on retail credit scorecards can be found in our previous whitepaper – FINANCIAL SERVICES & BANKING: RETAIL BANKING TRANSFORMATION – CREDIT SCORING.

 

Figure 8: major steps in SME credit scorecard development

Source: Accuracy

 

Decision tree

Machine learning now makes a substantial contribution to making credit decisions thanks to the rapid growth in data availability and computer capacity. One of the most popular supervised learning techniques is tree-based machine learning.

A decision tree is made up of two parts: branches and nodes that use various features from a dataset at each node to recursively partition a training sample. The algorithm iterates through all conceivable binary splits in search of the feature and related cut-off value that best distinguishes one side as having predominantly higher credit quality and the other as having relatively low credit quality. As an example, we can build a decision tree as below:

 

Figure 9: Illustrative decision tree

Source: Accuracy

 

• The most crucial factor is the interest-expense-to-sales ratio. The sample model decides that when data in the root node is divided into instances with the ratio < 2.5% and those with the ratio ≥ 2.5%, the figure of merit is optimised.

• Then, until it reaches the leaf node where the stopping requirement is met, this process is repeated for each new daughter node, i.e. loan size, working-capital-to-debt ratio, firm age, and cash to sale in this case.

• Finally, it provides the probability of default for each leaf node, of which the threshold is 0.2 in this case. Any borrowers that fall within the <0.2 criterion shall be granted the loan, while the others are rejected. This threshold is at the discretion of the lender.

Random forest

A random forest combines many different decision trees to get a prediction that is more precise and reliable. When compared with a single decision tree, a random forest avoids overfitting concerns, especially when there are enough trees in the forest.

To improve performance, numerous decision trees should be created in a random forest. The distinctness of each decision tree in the random forest is ensured by the random selection of data subsets and features. Overfitting is prevented since the model outcome is based on the combined predictions from each individual decision tree model. However, it is crucial to note that the individual decision tree models should not correlate highly with one another in this situation. The random forest approach does not make any assumptions about the data or its distribution, unlike many other algorithms (such as linear regression, SVM, etc.). Consequently, it typically only needs minor data transformations. As the random forest technique uses random feature subsets, it can work well with high-dimensional datasets (a dataset with a large number of features).

The random forest is especially effective compared with other models under the following circumstances:

• When there are outliers in the dataset, the random forest technique is unaffected by them.

• Many algorithms may take noise in the dataset as patterns (or extra manual power is required to remove outliers); however, the bagging method employed in random forest ensures that the noise in the dataset is not mistaken as signals or patterns.

• The random forest includes efficient methods to estimate missing values and preserve accuracy when there are missing values in the dataset, even when a sizable fraction of the data is missing.

Figure 10: Illustrative random forest structure

Source: Accuracy

 

Hybrid BWM and TOPSIS method

In addition to logistic regression and machine-learning techniques, an alternative method to develop an SME credit-scoring model is the hybrid BWM and TOPSIS.

• BWM is a decision-making logic tool that requires fewer data and less effort in development. It aims to find the optimal weights by minimising the gaps between actual weights and business judgement. Specifically, this process can be described as a linear model.

• TOPSIS is a tactic to determine an SME applicant’s relative position in contrast to a pool of borrowers. It finds the relative rank by calculating the weighted normalised matrix, obtaining the positive and negative ideal solutions, computing the Euclidean distance of the applicant between the positive and negative solutions and computing the relative closeness to the ideal solution.

This hybrid BWM and TOPSIS method requires judgement that is more subjective, but it offers an unparalleled performance in terms of cost, ease of development and implementation, and flexibility.

 

Figure 11: Illustrative steps in BWM and TOPSIS method

Source: Accuracy

 

Comparison of different modelling approaches

Overall, different modelling techniques, including but not limited to the those stated above, can be used to facilitate credit decisioning. These methodologies have their relative strengths and weaknesses, which we summarise in the below table.

 

Figure 12: Comparison of different modelling approaches

Source: Accuracy

 

FINAL REMARKS

SMEs make up a sizeable portion of the economies of both developed and emerging markets. Providing access to financing for small businesses has been a challenging task due to a variety of factors, including the expense and difficulty involved in determining the creditworthiness of small businesses with a lack of sufficient quality data.

However, banks are now able to lower the costs of originating and underwriting loans to SMEs while also increasing the performance of their SME loan portfolios, leveraging alternative data and new modelling techniques. These developments have led to an overall increase in the financing accessible to SMEs, and in time, will drive employment and economic growth.

At Accuracy, we have created our own SME rating model, Accur’Rating®. We originally developed the rating model using logistic regression but have recently migrated it to be a random forest model. We use this model to help clients evaluate investments in private debt, understand the credit quality of various corporates, etc.

We are in an era in which SMEs will become even more important for global economic development and driving innovation. With the right incentives, technology and knowledge, now is the time for banks to tap into and expand SME banking.

 

 

 

Download the article

More accuracy Perspectives