The Limits of DeFi for Financial Inclusion

3. FinTech applications with potential benefits for financial inclusion

ASEAN is the fastest growing internet market in the world, with an estimated 125 000 new internet users every day (WEF, 2023[26]), while the wider Asian region is home to an increasingly urban, literate, and young population that is keen to adopt new technologies across sectors, including in finance (OECD, 2021[27]). This provides fertile ground for FinTech applications to fill in some gaps in the provision of financial services to underserved parts of the region.

Indeed, the increased usage of digital finance applications during the COVID-19 pandemic (Fu and Mishra, 2020[28]) has been a great example of the possible benefits of FinTech applications for financial inclusion. According to the CCAF ASEAN FinTech benchmarking study, FinTechs in the region tend to cater to traditionally excluded populations in line with the Global Findex rankings of the respective countries (Figure ‎3.1). Although FinTech customers in ASEAN are more likely to be already banked individuals, important part of FinTech adopters are underbanked or unbanked individuals. In parallel, investment is flowing into FinTech development, with ASEAN’s share of global FinTech funding standing at 7% in 2021, up from 2% in 2018, although Singapore and Indonesia account for more than two thirds of this funding (UOB, PWC and Association, 2022[29]).

This chapter discusses some examples of use cases that could be conducive to better servicing underbanked parts of the population, such as MSMEs or populations in rural parts of the region and provides some examples of practical applications from the ASEAN region.

Figure ‎3.1. Underserved part of ASEAN population and banked status of FinTech customers

Source: Euromonitor, World Bank, Bain and Temasek, CCAF.

3.1. Regulated DLT-based finance: tokenisation and regulated digital assets

Although decentralised finance markets at their current stage expose investors to significant risks, concepts derived by DeFi protocols and regulated/compliant digital assets could potentially become a relevant part of the financial system at the institutional (i.e. non-retail) level in the future. Technologies and mechanisms associated with decentralised finance that may need to be further explored could potentially produce efficiencies and drive productivity gains in financial markets by improving the effectiveness and safety of trading and settlement at the infrastructure level (OECD, 2020[24]). These include atomic or programmable settlement of securities and payments at the post-trade, with the possibility for seamless delivery versus payment (DvP) or payment versus payment (PvP), increased automation and corresponding efficiency gains and enhanced transparency, as well as increased use of automation and programmability as is the case with smart contracts functionalities.

Policymakers are considering ways to allow for safe and responsible innovation, anticipating and addressing emerging risks for participants and markets. Technical level experimentation is already taking place with innovative pilots involving public-private co-operation, with the participation of traditional and decentralised finance entities. Project Mariana, for example, is a proof of concept launched by the Bank of International Settlements (BIS) Innovation Hub together with the Banque de France, Monetary Authority of Singapore and Swiss National Bank in co-operation with DeFi protocols, and explores the use of Automated Market Makers (AMMs) such as those deployed by DEXs in foreign exchange markets using CBDCs (BIS Innovation Hubs, 2023[30]). Project Guardian is a collaborative initiative led by the Monetary Authority of Singapore and with the participation of other regulators such as the Financial Services Agency of Japan, which seeks to test the feasibility of applications in asset tokenisation and DeFi while managing risks to financial stability and integrity (MAS, 2022[31]; FSA, 2023[32]).

The tokenisation of assets is indeed one of the most promising applications of DLTs in finance with numerous potential benefits (OECD, 2020[24]). It involves the digital representation of rights to real assets on distributed ledgers, or the issuance of traditional asset classes in tokenised form. Potential benefits include efficiency gains driven by automation and disintermediation; transparency; improved liquidity potential and tradability of assets which currently have near-absent liquidity; and faster and potentially more efficient clearing and settlement. When it comes to the impact on retail investors, tokenisation offers an additional tool for fractional ownership of assets1 allowing for small minimum investments, which, in turn, could lower barriers to investment and promote more inclusive access to previously unaffordable or insufficiently divisive asset classes, such as real estate (see Box ‎3).

The example of tokenisation of real estate assets is a noteworthy one given that retail investors are usually restricted in their ability to participate in such investments due to capital constraints and limited access to commercial projects. Some real estate tokenisation platforms have emerged in Asia, such as KASA in Korea and ADDX in Singapore (Chow and Tan, 2022[33]). Nevertheless, this market is still nascent and there are important challenges to real estate tokenisation which include, inter alia, insufficient evidence of investor demand for tokenisation of single real estate assets and the related difficulty in using digital platforms for such investment, as well rules governing the fractionalisation of ownership of assets (Chow and Tan, 2022[33]).

Box ‎3.1. Real estate tokenisation: The example of Thailand

The first regulated asset-backed tokenisation backed by real estate in Thailand took place in 2021, with the issuance of SiriHub Token by SPV77 Co. Ltd., a Special Purpose Vehicle (SPV) set up to issue and offer the tokens. The offering consisted of a commercial real estate asset (Siri Campus Office Building valued at THB 2.4 billion, or USD 73 million) as the underlying asset, which was under a long-term lease commitment of 12 years to a single tenant at the time of the offering. The issuer filed a white paper with the Thai Securities and Exchange Commission (SEC) and complied with the SEC’s requirements for initial coin offerings (Emergency Decree on the Amendment of the Revenue Code No. 19 B.E. 2561 A.D. 2018) (Okanurak and Koohasaneh, 2018[34]).

The token was designed for a term of four years and tokenholders were entitled to revenues from both the rentals and the potential sale of the asset through an auction at the end of the term. The SiriHub Token was offered in two tranches to cater for investors with different risk appetite. The minimum investment ticket for this offering was THB 10 (about USD 0.30) and 240 million tokens have been issued. Subscription for the tokens was via an online app (XSpring) available 24/7. The SiriHub Token was approved by the SEC and began trading at the digital exchange managed by ERX Company Limited in October 2021 (Chow and Tan, 2022[33]). Also in 2021, the Thai SEC granted a license to Fraction, a wholly owned subsidiary of Hong Kong based FinTech company Fraction Group, to host and trade tokens for fractional ownership of physical or digital assets.

The possible benefits of the transaction include fractionalisation and the possibility for retail investors to participate in a commercial real estate investment with as little as THB 10, potential economic efficiencies and possibilities customisation of the transaction. Thailand’s real estate-backed tokenisation offering is conceptually similar to a real estate investment trust, with a passive trust mechanism whereby the SPV acts as the trustee holding the title deed of the property on behalf of the unit holders of the trust, and has to be a legal entity that ensures the underlying asset is not transferred, disposed of or encumbered without the tokenholders’ approval.

In addition, and in order to prevent asset bubbles from being created by a possible future increased use of such offerings, a 2021 amendment specified that the real estate to be tokenised must not be a single condo unit or house and must constitute more than 80% in number or investment value of the project or in aggregate of not less than THB 500 million (USD 16.4 million) in order to be eligible for tokenisation (Securities Exchange Commission, 2020[35]).

Figure ‎3.2. SiriHub real estate tokenisation transaction structure

Source: Public information (https://spv77.digital/?lang=en).

The flow of private financing from capital owners to SMEs could also be eased and facilitated through regulated Security Token Offerings (STOs). Indeed, several countries in ASEAN have opted for tailor-made regulatory frameworks for STOs in addition to Thailand, including Malaysia, the Philippines (proposal stage) and Singapore (see Box ‎3.2) (OECD, 2020[36]). Frameworks for STOs have also been implemented in Asia beyond ASEAN, with the most prominent examples being that of Korea (FSC, 2023[37]) and Japan (FSA, 2019[38]). These frameworks all require issuers and intermediaries of such regulated tokenised security offerings to comply with securities regulations in respective jurisdictions, given the technology neutral approach to regulation. Tailor-made frameworks and guidance may promote legal certainty and assist market participants in understanding the requirements, equipping them to participate in such endeavours in a compliant and safe manner.

Box ‎3.2. Singapore framework for digital token offerings

Singapore clarified its regulatory position around the offering of digital tokens as early as 2017, and has issued a detailed guide to digital token offerings a year later (Monetary Authority of Singapore, 2017[39]; Monetary Authority of Singapore, 2020[40]).

According to the Singapore framework, digital tokens may represent ownership or a security interest over an issuer’s assets or property and the Monetary Authority of Singapore (MAS) will examine the structure and characteristics of, including the rights attached to, a digital token in determining if the digital token is a type of capital markets product under the Securities and Futures Act (SFA). Digital tokens may also represent a debt owed by an issuer and be considered a debenture under the SFA, or a securities-based derivatives contract in case any derivatives are included.

Where digital tokens fall within the definition of securities, issuers of such tokens would be required to lodge and register a prospectus with MAS prior to the offer of such tokens, unless exempted based on proportionality (offering size lower than SGD 3 million and offer made to less than 50 investors in any period of 12 months, or offer made to accredited and institutional investors). Issuers and intermediaries are all subject to licensing requirements, while platforms facilitating secondary trading of such tokens would also have to be approved or recognised by MAS as an approved exchange or recognised market operator respectively under the SFA.

Source: Monetary Authority of Singapore.

3.1.1. FinTech lending and the use of artificial intelligence for credit scoring

FinTech lending accounts for the majority of digital finance activity in ASEAN driven by strong growth in consumer spending, increased needs for MSME financing and low banking penetration (as of 2019, in number of FinTech companies) (CCAF and ADBI, 2019[41]). Artificial intelligence (AI)-based models and big data are increasingly being used by FinTech lenders to assess the creditworthiness of prospective borrowers and make underwriting decisions. Machine Learning (ML) models are used to predict borrowers’ defaults with superior forecasting accuracy compared to standard statistical models (e.g. logic regressions) especially when limited information is available (Bank of Italy, 2019[42]; Albanesi and Vamossy, 2019[43]). They can also improve both the efficiency and the speed of credit decision making and improve the risk prediction accuracy overall (Gambacorta et al., 2019[44]), although their use comes with a number of risks and challenges.

The use of AI-based models for credit scoring could have potential beneficial impacts on the extension of credit to MSMEs, both through a reduction in the cost of underwriting, and through improvements in the analysis of creditworthiness of clients with limited credit history and/or without collateral (‘thin files’) (OECD, 2021[45]). It can therefore enable the extension of credit to young start-ups and MSMEs that cannot prove their viability through historical performance data or tangible collateral assets, potentially enhancing access to credit and supporting the growth of the real economy by alleviating constraints to MSME financing.

Although there are many potential gains from AI in terms of cost reduction and financial inclusion, there are also risks stemming from its use. Risks involve data quality and confidentiality, limited explainability of the models used and possible risks to financial consumers (OECD, 2021[45]). In particular, the use of credit rating models based on such techniques can raise risks of disparate impact in credit outcomes and the potential for discriminatory or unfair lending (US Treasury, 2016[46]). Well-intentioned ML models may inadvertently generate biased conclusions, discriminate against certain classes of people (e.g. based on race, gender, ethnicity, religion), or simply reinforce existing biases, while making discrimination in credit allocation more difficult to identify (OECD, 2021[45]). Regulations, guidance and frameworks, such as the OECD Principles on Artificial Intelligence, are increasingly put in place to guide the use and governance of AI systems in a safe, responsible and trustworthy manner (OECD, 2019[47]).

Box ‎3.3. Use of artificial intelligence-based models for credit scoring: the case of Viet Nam

Algorithmic credit scoring solutions powered by AI and using Machine Learning (ML) models use massive amounts of traditional and alternative data sources referred to as ‘big data’. These include conventional credit information, social media data, digital footprints, payment behaviour and other transactional data accessible through Open Banking data sharing initiatives.

AI-based credit scoring is being used in many countries across Southeast Asia in consumer finance markets, and Viet Nam is one of the foremost AMS in which the usage of such models is spreading rapidly (Lainez and Gardner, 2023[48]). Consumer finance was virtually non-existent in Viet Nam ten years ago and has succeeded to grow considerably thanks to efforts made by local credit institutions to attract and service unbanked borrowers, inter alia through the use algorithmic credit scoring. Viet Nam has an official public credit registry the Credit Information Centre (CIC), gathering data from 30.8 million citizens, and a smaller private one, Viet Nam Credit Information Joint Stock Company (PCB) formed by 11 local banks in 2007.

FinTech companies, such as FE Credit, a consumer lender subsidiary of VPBank, are supplementing the official credit ratings with alternative scoring methods based on algorithmic models and big data. In 2018, it launched $NAP, an automated lending platform that aims at achieving a ‘fast and easy credit’ experience for its clients in Viet Nam (Lainez and Gardner, 2023[48]). Such industry activity has the potential to promote financial inclusion of underbanked parts of the population, however, emerging risks for consumers related to data privacy and to possible bias or discrimination may also need to be accounted for.

Financial institutions in ASEAN markets are leveraging AI to enhance their ability to proactively manage and mitigate risks for their customers, ultimately contributing to their financial well-being and indirectly promoting financial inclusion by sustaining their participation in the formal economy. For example, AI-driven models are being used in AML transaction monitoring, online sexual exploitation of children transactions2 detection, mule account detection, phishing transaction detection and card fraud detection (Box ‎3.4). These tools can contribute to the strengthening of the capacity of financial institutions to comply with anti-money laundering (AML) and countering the financing of terrorism (CFT) rules, while increasing the confidence and trust that customers have in their participation in the formal financial system.

Box ‎3.4. Strengthening transaction safety through AI tools promotes consumer trust: the case of AI-assisted mule account detection

So-called money mules are actors who transfer or move illegally acquired money on behalf of someone else, usually recruited by criminals’ help launder proceeds derived from online scams and frauds, or crimes like human trafficking and drug trafficking. Money mules add layers of distance between crime victims and criminals through the use of unrelated consumer accounts to move money at the direction of third persons involved in a criminal activity, which makes it harder for law enforcement to accurately trace money trails.

Financial institutions in the Southeast Asian region, such as UnionBank in the Philippines, perform mule account detection using AI tools. Such tools are designed to get ahead of fraudsters and detect them early on, so that appropriate actions can be taken when account owners have knowingly or unknowingly had their accounts used by criminals. The use of AI tools is a shift from the traditional reactive, reports-based approach with no automated process to detect mules, which was only effective once the customers reported a scam.

The ML-based detection is based on the patterns observed on (a) account opening data and (b) transaction behaviour data. The initial detection rate for UnionBank in the Philippines has been 42% of true money mules at a low rate of false positives and with 90% accuracy on non-mules prediction. This particular model is built and validated on real production data and real scenarios, and has to be continuously fine-tuned. In terms of efficiency, when comparing the traditional approach against the AI-based mule detection tool, money mule accounts are reported and endorsed for investigation for approximately a month after account opening, while the AI model has the capability to detect true money mules as early as four days after the account is opened.

Such models can support and enhance fraud prevention in an efficient way, while fostering customer engagement and promoting trust in formal financial services. The tools aim at protecting from fraud, accelerating investigation and providing an interface to view and analyse predicted mule activities, integrated into existing systems without disruption and through a transferable model and methodology which ensures reduced implementation time and effort.

Notes

← 1.  In addition to existing tools to promote fractional ownership in certain asset classes, such as real estate, for example, where REITs are a popular vehicle for retail participation.

← 2.  The model can monitor financial transactions to identify patterns and characteristics associated with child exploitation through payments to illegal websites then later can be flagged as suspicious transactions.