SMBC Mobit's screening closes at 9 PM: If being a night owl is genetic, who is being locked out of payments?

Is a habit of staying up all night a valid excuse?

I can't stop staying up late. I've tried to fix it and failed, so eventually, I started saying it's just my constitution. It's usually taken as me being lazy.

I started looking into whether it's really a matter of constitution, and halfway through, the topic drifted toward payments. The drift was more interesting, so I'll write about that instead.

To what extent is being a night owl genetic?

A genome-wide association study published in 2019 identified 351 genetic loci associated with the trait of being a morning person or a night owl, covering 697,828 people from the UK Biobank and 23andMe combined. Of these, 327 were new discoveries¹.

There is a range in the estimated heritability. Twin and family studies suggest 12% to 42%², while SNP-based estimates from the UK Biobank alone suggest 13.7%¹. The numbers fluctuate because the measurement methods differ, but in any case, it cannot be said that 'it's all the fault of genes.' There are certainly parts that are influenced by lifestyle habits and light exposure.

However, the reverse conclusion remains firm: being a night owl is not a phenomenon that can be explained solely by an individual's lack of discipline, but a trait that continues to exist in a certain percentage of the population. From here on, I will use only this premise.

The real reason why the incidence of fraud increases late at night

Research on fraud detection using supervised learning reports that fraudulent transactions tend to concentrate between 10 PM and 4 AM (GMT). This is because it is a time when the majority of victims are asleep and the banks do not have monitoring systems in place³. The same study also states that the number of fraudulent transactions at night is actually lower than during the day³.

This is where it is easy to misread; fraud is not surging late at night. The number of cases is actually decreasing. It's just that the way it decreases is different.

The time distribution of all card transactions is a mountain shape peaking at noon, with small peaks at mealtimes, and a significant drop toward late night⁴. On the other hand, the fraud side continues to work even at night because they are targeting people who are asleep. Since only the denominator disappears first, the ratio increases.

In other words, high risk late at night does not mean 'the time when bad people increase,' but 'the time when decent people decrease.'

And returning to the premise from earlier, there are people who are always awake within that reduced denominator. As long as there is a certain percentage of the population that functions at night as a matter of constitution, legitimate transactions late at night will never reach zero. From a statistical perspective, it is negligible noise, but for the individuals involved, it is just daily life.

Card company design is actually friendly to night owls

I want to defend them here.

Sumitomo Mitsui Card's monitoring operates 24 hours a day, 365 days a year, and is designed to detect transactions that do not match the user's tendencies after accumulating data on each user's payment habits and the countries they usually use⁵. It is not cut off by absolute time, but looks at the deviation from that person's own normal state.

With this design, even if a habitual night owl makes a payment at 3 AM, it will not theoretically be blocked as long as it is that person's normal state. The common complaint 'it was stopped because it was late at night' is more accurately 'it was stopped because it was an unusual time for you.'

The problem lies beyond that. What happens to people whose 'normal' cannot be defined?

Newly issued cards, users with thin transaction histories, and first-time EC merchants. When there is no individual baseline, the system has no choice but to rely on the overall average. And the overall average is created by a morning-oriented lifestyle.

Night owls with established credit are protected, while those yet to build it are viewed with suspicion. As a sequence, it is a bit harsh.

Lending closes before it even suspects.

If fraud detection is a matter of probability, lending is much simpler. It closes by the clock.

While card loans accept applications 24 hours a day, the screening process itself has specific operating hours. In the case of SMBC Mobit, while web applications are generally accepted 24 hours a day, screening is only conducted from 9:00 to 21:00. If you apply after 21:00, the screening will not begin until 9:00 the next business day, meaning you cannot receive same-day financing⁶. Aiful's explanation also explicitly states that if you want to shorten the screening time, you should apply early in the day, such as in the morning⁷.

This is not discrimination; it is simply an operational necessity. Screening involves human intervention, and in the case of bank-affiliated services, it also involves inquiries to the National Police Agency. There is no rationale for keeping offices open at night.

However, as a result, financial windows are closed during the hours when night owls are at their sharpest. Before they can even be suspected, they cannot even get in line.

Time is a variable. However, the weights are not disclosed.

So, is time itself used as a variable in the score?

Regarding fraud detection, it is used. In research using real-world transaction data, 'time (1:00 to 24:00)' is explicitly included in the list of features input into the model⁸. Merchant-side analysis tools also provide standard features for tracking payment trends by time of day⁹. The fact that time is a variable is not particularly hidden.

What is unknown is the weight. How many points shift during which time slots, and to what extent individual history corrections cancel them out? This is the core of the scoring mechanism itself, so it is not disclosed. Naturally, disclosing it would reveal the strategy to fraudsters.

There are no clues in official statistics either. The Japan Credit Association only publishes quarterly damage amounts, breakdowns by method, and annual fraud occurrence rates¹⁰; there is no breakdown by time of day. There is effectively no material available from the outside to verify how 'late-night payments' are handled in Japan.

As a result, there is no way for the parties involved to verify whether night owls are structurally disadvantaged. They might be, or they might not be. They continue to use the services without ever knowing.

The premise of 'normal waking hours'

The design that assumes a specific time is statistically correct. The fraud rate at night is indeed high, and the cost of maintaining a screening system at night is not justifiable. Both are rational decisions.

However, that rationality is built on the premise that being awake at night is a personal choice. To what extent can a trait with a heritability of 10% to 40% be called a choice?

There is no malice in financial design. Without malice, costs are quietly imposed on the minority divided by their biological constitution. This structure is not limited to credit and fraud detection; it is common to all service designs that assume specific time slots. The fact that the pros and cons of 24-hour operations are discussed solely in terms of cost efficiency likely shares the same root.

I started researching to find an excuse for staying up all night, but I ended up talking about a place where excuses don't work. It's late again tonight.

Sources

1. Jones, S. E. et al. "Genome-wide association analyses of chronotype in 697,828 individuals provides insights into circadian rhythms" Nature Communications, 2019 https://www.nature.com/articles/s41467-018-08259-7

2. Lane, J. M. et al. "Genome-wide association analysis identifies novel loci for chronotype in 100,420 individuals from the UK Biobank" Nature Communications, 2016 https://www.nature.com/articles/ncomms10889

3. "A supervised machine learning algorithm for detecting and predicting fraud in credit card transactions" ScienceDirect, 2023 https://www.sciencedirect.com/science/article/pii/S2772662223000036

4. "A Bayesian Simulator for Payment Card Fraud Detection" Finance and Economics Discussion Series, Federal Reserve Board, 2025 https://www.federalreserve.gov/econres/feds/files/2025017pap.pdf

5. Sumitomo Mitsui Card "What is the mechanism of the credit card fraud detection system?" https://www.smbc-card.com/nyukai/magazine/knowledge/creditcard_fraud-detection.jsp

6. Minna no Mobit "Can I apply for or borrow from a card loan 24 hours a day?" https://www.mobit.ne.jp/media/1027/index.html

7. Aiful "How long does card loan screening take? Explaining the screening process and tips for shortening it" https://www.aiful.co.jp/media/consumer-finance/column-0106.html

8. "Credit card fraud detection through parenclitic network analysis" arXiv, 2017 https://arxiv.org/pdf/1706.01953

9. YTGATE Co., Ltd. "51 billion yen in credit card fraud damage in 2025. A thorough explanation of the current situation, causes, and countermeasures!" https://ytgate.jp/news/trends/20260317-001/

10. Japan Credit Association "Credit-related statistics" https://www.j-credit.or.jp/information/statistics/index.html