How Big Tech Lenders Will Disrupt The Business Cycle

The Study in a Nutshell

A new BIS paper finds that Big Tech lenders using big data to assess creditworthiness hold informational advantages over traditional banks. This could disrupt credit markets by shifting the emphasis away from collateral. It could also weaken the transmission of monetary policy.

In a recent Deep Dive, we discussed how innovations in information and communication are challenging the traditional bank business models. This week we follow that thread with a paper that looks at more than two million Chinese firms that received credit from both an important Big Tech firm (Ant Group) and traditional commercial banks. It asks, do informational advantages, gleaned by fintech firms using machine learning and big data, have implications for collateral and monetary policy transmission?

The answer is yes. The credit offered to businesses via Ant Group correlates strongly to changes in the actual borrower firm’s characteristics. By gathering firm-specific data obtained by direct interactions on their platforms, Big Tech firms are able to assess the credit worthiness of a borrower based on their activities (transaction volumes and network scores) rather than the amount of collateral they can offer up.

Meanwhile, the credit that commercial banks offer correlates significantly to local economic activity and house prices. The banks require their SME borrowers to pledge tangible assets, such as real estate, to lessen ex-ante adverse selection problems. This collateral is far more susceptible to the local economic environment, and hence credit reacts more strongly to developments in activity and house prices.

This has important implications for monetary policy. Traditionally, business lending is tied to the net worth of individuals and collateral (often housing), which makes lending very pro-cyclical as represented by the ‘financial accelerator’ theory in economics. Should Big Tech lending start to dominate, then this link could be broken, which changes the dynamics between interest-rate sensitive sectors like housing and business lending.

How Big Tech Lenders Make Credit Decisions

The Big Tech business model rests on enabling direct interaction between many users, which creates a large stock of user data. Big Tech firms use this data as an input to offer a range of services that exploit natural network effects, generating further user activity. Increased user activity then completes the circle, as it generates yet more data.

The mutually reinforcing data-network-activity (DNA) feedback loop helps Big Tech firms to identify the characteristics of their clients and offer them financial services that best suit their needs. They can then conduct monitoring almost in real time and quickly adjust credit scoring.

Data gathering and enhanced monitoring of borrowers once they are within a Big Tech’s ecosystem gives these firms a competitive advantage over banks and helps them serve those that otherwise would remain unbanked. In this new way to conduct financial intermediation, data and monitoring (i.e. information) can substitute for collateral when the latter is relatively more expensive.

The consequences of Big Tech credit:

• More firm-specific data reduces asymmetric information and provides financial products for all SMEs.
• E-commerce platforms make credit scoring and enforcing loan repayments easier.
• Credit decisions are made on the basis of borrow characteristics, not the economic environment.
• If credit responds less to asset prices, such as real estate, monetary policy transmits less well.

The Data Used in the Study

BIS uses a large dataset to test the difference in lending behaviour of Big Tech and commercial banks. It is constructed at the firm-month level over the period 2017:01 to 2019:04, using more than 2mn Chinese firms. The study’s sample of firms contains not only firms on Alibaba’s e-commerce platforms (online firms) but also those that use more traditional business channels (offline firms). The latter use the Alipay app for mobile payments, through the so-called Quick Response (QR) code, but are not fully integrated into the e-commerce platform.

The paper uses the dataset to compare the characteristics of loans provided by MYbank, one of the brands under Ant Group (one of the most important Big Tech companies in China and affiliates to Alibaba) with loans supplied by traditional Chinese banks. For all loans provided by traditional banks, they further distinguish between collateralised credit (secured bank credit) and uncollateralised credit (unsecured bank credit).

Data is collected on three key areas:

1. Firm characteristics: transaction volumes, credit data and network scores from Ant Group. These have a crucial role in the credit scoring analysis of MYbank. Evidence from the paper shows that Big Tech borrowers are typically granted $975 over a 1m to 1yr period, while the median secured (unsecured) bank credit is $42,400 ($8,500) for 1-3 years. Despite the differences in credit, firm size between Big Tech and bank credit users (measured as transaction volumes) is not large.
2. Entrepreneurial information: age and level of income from Ant Group and Alipay. For SMEs, information about the entrepreneur (typically the owner of the firm or the store) is very important for risk assessment. The paper shows that borrowers who access Big Tech credit are slightly younger (the median age is 31 years) than the owners of firms that use unsecured bank credit (36 years) or secured bank credit (38 years).
3. Economic activity: house prices, GDP and monetary policy (land supply and mortgage rate). Housing data is obtained from China Index Academy, while land supply is hand-collected for each local government scaled by urban construction land.

Note that of the 2mn MYbank borrowers from which the data is gathered, just under 10% have obtained credit from traditional banks.

Answering the Main Questions

Do Big Tech and bank credit react differently to collateral value, local economic conditions and firm-specific characteristics?

In short, yes. In their first model (controlling for entrepreneurial information), the authors conclude that Big Tech credit does not correlate with house prices or local economic conditions. Instead, lending is strongly correlated with firm-specific variables (transaction volumes and network score) and borrow-specific variables (age and income).

By contrast, the authors find that bank credit is only weakly correlated with firm-specific characteristics, but significantly correlated with house prices (especially secured credit) and local economic activity (only unsecured bank credit). They also find that secured bank credit is not correlated with borrower-specific variables.

Source: Page 31 in Gambacorta et al. (2020)

The positive correlation between unsecured bank credit and house prices could reflect higher demand in cities with higher asset prices. Another explanation may be that banks do not have enough granular information on the firm, so local house price dynamics turn out to be one relevant indicator to identify a firm’s creditworthiness.

Next, the authors control for unobservable client characteristics and divide results into offline borrowers (those with a QR code, but not trading on the e-commerce platform) and online borrowers (those integrated in the e-commerce platform). This model shows that borrower fixed effects (the R-squared rises) capture around 40% of MYbank credit variability.

Once again, Big Tech credit is not correlated with house prices but does become significantly correlated with local economic conditions. The positive correlation between Big Tech credit and city-level GDP is significant only for firms that work offline. Perhaps this is because offline firms (i.e. a restaurant or a shop) depend on local business conditions (i.e. demand side factors).

Interestingly, collateralised bank credit does show some signs of correlation with firm-specific characteristics, but only with respect to transaction volumes for firms that work online. Nevertheless, the correlation of bank credit (secured and unsecured) with respect to firm-specific characteristics is weaker than that displayed by Big Tech credit, especially when considering the network score.

The paper subsequently considers two specific credit contracts issued by MYbank to address any biases from aggregating credit contract. Product 1 is a credit line for each merchant that is based on their specific risk profile; Product 2 is credit offered based on the overall value of orders and receivables in the Taobao platform.

Again, neither form of Big Tech credit is correlated with house prices nor with local economic conditions, and they are highly correlated with borrower-specific characteristics (age and income). The correlation is higher for Product 2, used by firms that work online, while the correlation for Product 1, used by firms that work offline, is lower.

How could the increased use of big data and machine learning in solving asymmetric information problems, in lieu of collateral, impact the financial accelerator?

To control for issues around endogeneity (more credit leads to higher house prices), the authors instrument house prices. The local government has a great influence on housing prices through the land supply in China – more land supply has a negative effect on house prices.

When using the log house prices as an instrument, the authors continue to find that only bank credit is significantly correlated with house prices. This result underscores that in the case of an (exogenous) increase in the value of collateral triggered by an expansion in the supply of land by the government, there is no positive effect on Big Tech credit. This result is interesting because it indicates a reduction of the effectiveness of the financial accelerator (which amplifies the effects of shocks to the real economy by means of loan supply shifts caused by changes in collateral values) in case of Big Tech credit.

Do Big Tech platforms matter? Are there differences between credit granted to firms that operate in the e-commerce platform (online) and credit granted to firms that operate on traditional business channels (offline)?

Yes, they do. If we look at the correlations described above, Big Tech credit to online firms is more strongly correlated with transaction volumes and network scores than it is in the case of offline firms. That means that Big Tech firms are able to more efficiently collect and process information from online lenders that are integrated in the Big Tech ecosystem. Therefore, they have access to a rich set of additional data to be combined with traditional transaction volumes obtained from payments.

Source: Page 29 and page 30 in Gambacorta et al. (2020)

Bottom Line

If we move towards a world in which fintech firms predominantly provide financial services that commercial banks traditionally offered, we should expect monetary policy transmission to weaken. In credit markets, this is because Big Tech firms can overcome information asymmetries by using their own borrower-specific microdata.

Moving away from collateral will have many upsides – fostering greater entrepreneurship, supporting otherwise unbanked SMEs and promoting a more efficient use of resources. What is less clear is how it may affect the ability of central banks to stabilise SME financing.

To view the full paper, please click here