Contemporary trends in Data science in Banking sector

1. Traditional credit risk assessments in banking sector:

One of the main functions of banks is a source of providing loans to customers (Moradi & Rafiei, 2019). In 2021, Commonwealth bank served hundreds of thousand loan customers and provided over AUD$50bn worth of new business loans (Commonwealth Bank, 2021). This can be seen that the borrowing demand is huge, especially in the time of the Covid-19 Pandemic, where there are many enterprises short of funds to remain their business (Johnston, 2021). The loan application process in general has been simplified for a better customers experience as customers can now apply online with a shorter application time (Westpac, 2022a). In particular, Westpac’s personal loan application process only takes 60 seconds to get a response (Westpac, 2022a).

The current method of credit risk assessment focus on the traditional ‘5Cs’ model, which assesses character, capacity, collateral, capital and conditions of the borrowers (Connolly, Cava & Read, 2015). Evaluating borrowers’ characters means examining their “overall trustworthiness, personality and credibility” to assess their probability of default (Treece & Tarver, 2021). This would involve looking at borrowers’ credit history (Treece & Tarver, 2021) or face-to-face interviews (Connolly, Cava & Read, 2015). The capacity of the borrower is the ability to make full repayments on time by analysing borrowers’ income (Treece & Tarver, 2021). A borrower’s capital is also an important criterion that shows the fund amount the borrower has contributed to the project for a business loan or to pay additional purchases for mortgages and personal loans (Treece & Tarver, 2021). Collateral is an asset that the borrower uses to secure the loan, which can be used to repay the debt in case of defaults (Treece & Tarver, 2021). In addition, lenders would assess other conditions relating to the repayment probability such as the industry overview of the borrower’s business or the overall economy (Treece & Tarver, 2021).

2. Challenges: Difficulties for potential borrowers in receiving desirable loans due to their lack of history credit profile.

While each bank has its own internal credit rating strategy (APRA, 2017), through observing the three main Australian banks namely ANZ, Westpac and NAB, it is indicated that Australian banks share similar major criteria to assess borrowers’ creditworthiness. To make a personal loan application, the borrower would normally have to provide a list of personal documents such as personal identification, financial income and credit history (NAB, 2022a), (ANZ, 2022a), (Westpac, 2022a). For a business loan, the main criteria would include having a valid ABN/ACN for at least 12 months and being registered for GST with a turnover of at least $75K per annum (NAB, 2022b), (ANZ, 2022b), (Westpac, 2022b). While credit risk assessment should be conducted in the most careful way as it is related to the bank’s chance of receiving back its loan, this traditional method may miss out on other crucial information that could bring banks more potential borrowers (Kumar, Sharma & Mahdavi, 2021). Particularly, it would be a challenge to seek a loan for borrowers with no loan history such as small and new businesses (OnDeck Australia, 2021) or borrowers with “limited banking transactions” such as rural residents (Kumar, Sharma & Mahdavi, 2021). The use of only credit history could omit other relevant variables than can affect the repayment capacity such as health status or employment experience (Avery et al., 2000). Another problem with this method is that it is unable to assess a borrower’s chance of default (Kumar, Sharma & Mahdavi, 2021). The traditional risk assessment determines the chance of default of customers by examining only previous default loans (Fenerich et al., 2020). Therefore, banks need a more efficient credit risk assessment strategy in order not to lose potential borrowers, hence leveraging data science for credit scoring would benefit the current credit risk assessment, providing a more holistic credit profile of borrowers.

3. Opportunities & Impact: Various Big Data and AI-ML technique applications could tackle the shortcoming of traditional credit risk assessment of banks hence generating better insights into borrowers’ credit profiles.

The use of mobile phone data: The increasing use of mobile phones and social networks have provided a “new Big Data source” that can be exploited to improve the creditworthiness of borrowers (Óskarsdóttir et al., 2018). According to Óskarsdóttir et al. (2018), positive information, which includes all information that shows positive financial behaviour, extracted from call-detail records could help increase credit availability. This would most benefit borrowers with no or limited credit history as they can resort to their call datasets available from their phone to enhance their profile (Óskarsdóttir et al., 2018). The research also indicated that these datasets can also help banks to better predict default probability (Óskarsdóttir et al., 2018).

Adding politico-economic factors to predict default loan: According to Moradi & Rafiei (2019), the static models that banks traditionally use to evaluate borrower credit risk might be inefficient in the time of crisis or economic fluctuations as customers previously recorded as good could possibly become bad customers without bank’s acknowledge. This is based on the reason that the traditional model is less responsive to sudden changes, especially in the time of the Covid-19 Pandemic. Moradi & Rafiei (2019) proposed a new model that utilises a fuzzy inference system based on monthly data of customer credit profiles. Fuzzy logic is an ML-based approach that examines human behaviour or judgement rather than purely numerical estimates (Bennouna & Tkiouat, 2018). The idea behind this is that this model would “learn” from a list of previous bad customers and then be used to assess future customers. Study shows that this model outperforms the traditional one because it could yield comparable results with real-life data while taking into account the politico-economic factors (Moradi & Rafiei, 2019).

KOALA — OnDeck’s innovative credit model: OnDeck is a lending company that focuses on offering loans to small businesses, with over US$13 billion loaned globally since 2007 (OnDeck Australia, n.d.). The company’s competitive advantage compared to a traditional bank is that it integrates data science into the credit risk assessment of customers with additional factors such as the actual performance of the business in addition to the business owner’s credit history (OnDeck Australia, 2021). KOALA is a lending algorithm launched by OnDeck’s data scientists in April 2021, which utilises a combination of sophisticated credit algorithms, statistical techniques and data from credit reporting agencies such as Illion and Equifax to improve its credit risk assessment, hence giving SMEs a higher chance of receiving loan (OnDeck Australia, 2021). As traditional banks mainly focus on commercial data, this innovation would benefit a wide range of borrowers including newer enterprises and sole traders (OnDeck Australia, 2021). In Q1 2021, the number of new loans issued by OnDeck increased by 11% after the launch of KOALA (OnDeck Australia, 2021), which indicates an attractive lending option for borrowers.

Using ML techniques to classify borrowers, hence improving the credit scoring model and minimizing losses for banks: Fenerich et al.’s study (2020) compared three different classifier algorithms of ML including Bayesian Networks, Decision Tree and Support Vector Machine to propose the best method that can improve credit risk assessment. The study shows that all three techniques can generate results at up to a 95% correct rate (Fenerich et al., 2020). In addition, this method also provides a “default index” instead of just labelling customers as “defaulters” or “non-defaulters” (Fenerich et al., 2020). Moreover, new borrowers with no history of credit lending or banking transactions can also benefit from this method as the model did not examine the previous behaviour of the borrowers (Fenerich et al., 2020).

4. Issues:

Data privacy: The implementation of different methods would require a huge amount of personal data, hence privacy concerns should be taken into account. Therefore, it is crucial that banks and other financial institutions ask for borrowers’ consent before collecting any personal information. In addition, borrowers should also be informed about how their data would be stored and processed, especially in the case where banks use third-party to conduct their credit risk assessments. Marriott & Robinson (2017) also suggested a higher level of data protection by implementing stricter regulation as they argued that the current UK legislation is unable to “address adequately the serious accuracy, transparency, and accountability concerns of individual data subjects”.

Credibility/eligibility: While additional soft data that is voluntarily submitted by customers can provide banks with more insights about their repayment capacity, the eligibility of the personal data collected should also be considered. While data from mobile phones could be a trustworthy source in a sense that it could be collected without the owner’s adjustment, the borrower could deliberately change their behaviour in terms of phone usage habit or call-detail on a long-term period to create a positive profile to lenders. An example of this is Anna Delvey, who was known for faking as a German heiress, then scamming New York socialites, stealing private jets and bilking banks (Pressler, 2018).

Digital discrimination: This issue concerned with biases or unfair assessments could arise as a result of personal data being processed by automatic algorithms (Criado & Such, 2019). It is suggested that to minimise discrimination, the data should be carefully examined at the early stage of collecting and developing the model (Kumar, Sharma & Mahdavi, 2021).

Challenge for policymakers: Australian law has long been known for its strict regulations which can be proved by its survival through the 2008 global financial crisis without any recessions or bank failures (Edey, 2014). However, the application of ML techniques for credit assessment should not be neglected. Regulators should find a balance in allowing banks to implement ML techniques while possibly requiring stricter supervision or even stricter regulation to protect customers’ rights when their personal data is being collected.

5. Conclusion:

Credit assessment is no doubt an important process that should be conducted with great care. While this process should abide by the law, there is still room for innovation by taking advantage of data science to improve the quality of the assessments. With all the issues mentioned above are taken into consideration, ML-techniques application to bank credit risk assessment could benefit both banks and borrowers with limited credit data.