Featured faculty: Kim Keongtae and Michael Zhang
Written by Pan Jingyi

AI-powered coding assistants are changing how software is built. These tools can be highly effective in helping programmers and software developers, even those with less experience, to write code in a programming language. Almost everyone can now make an app or software from scratch by simply asking an AI chatbot, a practice known as vibe coding.

OpenAI’s release of Codex in 2021 was the first AI model powerful enough to build entire programmes from simple text prompts. The company then partnered with GitHub, a platform where developers worldwide can store and collaborate on their software code, to launch GitHub Copilot as an autocomplete tool. In early 2023, Copilot was transformed into Copilot X, a full AI coding assistant.

Nowadays, Copilot X is not the only tool in the market. Other major AI companies, such as Anthropic and DeepSeek, as well as big tech like Google, Amazon, and Meta, have also launched their AI-assisted software development tools. As millions of developers start relying on AI, burning questions loom: have these tools changed how developers code, and can they actually boost productivity?

“AI technology may increase the developers’ productivity and push the limits of what developers feel able to attempt, but these gains come with trade-offs for code quality and originality.”

Professor Kim Keongtae

Kim Keongtae, Professor in the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School, collaborates with Michael Zhang, the Wei Lun Professor of Business AI in the same department, their PhD student Li Xinyu, and Francis Joseph Costello of Nova School of Business and Economics, to seek answers. In the study titled Exploring altered open source software development patterns in a time of generative AI, they examine 1,350 open-source software developers on GitHub before and after Copilot X was released.

The analysis finds that CopilotX significantly helps occasional contributors who were previously less active on GitHub. These casual developers participated in 48 per cent more open-source projects and became far more likely to experiment with new programming languages after using the AI tool.

Meanwhile, highly active or more seasoned developers who used Copilot X updated their code 12 per cent more often and contributed to six per cent more projects compared to those who didn’t use the tool. AI not only helps developers work faster but also spreads their effort across a larger set of projects and switches between different projects more easily.

Contributing to open-source projects sometimes requires developers to learn an unfamiliar codebase and figure out how the software is structured before making a small update. For occasional contributors, these barriers can be enough to stop them from participating at all.

“AI coding assistants can lower these barriers and make the first step easier,” says Professor Kim. “After conversational AI coding assistance became available, developers became more willing to explore new programming languages and projects.”

The hidden costs of easy coding

While productivity rose, a noticeable increase in copy-pasted code appeared. Although more projects are being started, they are not necessarily leading to more original work. “AI technology may increase the developers’ productivity and push the limits of what developers feel able to attempt, but these gains come with trade-offs for code quality and originality,” Professor Kim adds.

“Copied code is not always harmful, as some reuse is normal in software development. The concern is that if developers insert larger blocks of code without fully understanding how they interact with the rest of the project, the project may accumulate technical debt or a shaky foundation.”

The build-up of hidden problems could make software harder to maintain, debug, or improve later. It can also lead to compatibility issues, inconsistent coding styles, and more rework down the line. For project managers, this finding calls for treating AI-generated code as a draft, not as final output.

AI can boost careers

Open-source software development relies on voluntary effort, so developers have to contribute their personal time. While some programmers and software developers participate for their own interests, GitHub functions much like a public portfolio to showcase recent work and collaborations. Employers also often look at GitHub profiles to assess software developers’ skills and experience.

Therefore, to further examine whether changes in how developers worked with AI affected their careers, the researchers matched developers’ GitHub histories with their LinkedIn profiles and found encouraging signs. Developers with access to Copilot X were more likely to experience internal transitions and promotions in the following six to 12 months.

It appears that increases in open-source contributions and proficiency with new technologies are associated with positive career outcomes. “Visible output on GitHub can also signal employers that a developer is adapting to new tools and can leverage them to deliver more work,” Professor Kim says.

“However, the benefits of AI coding tools for career improvement are temporary. When only a few developers know how to use these tools really well, they’ll be noticeably more productive than others. As these tools become widely used, simply using AI won’t make a developer stand out as much anymore.”

How can developers stand out in the AI age?

Given the short-term benefits and the increasingly prevalent copied code, Professor Kim warns developers that AI should only be used as a learning partner and assistant. Although AI may accelerate software development, it cannot replace the developer’s responsibility to understand and evaluate the system. “Take the AI tool as a tutor rather than a shortcut. The human developer should remain responsible for the final judgement.”

This advice is especially crucial for beginners, as less experienced software developers are more likely to use code suggested by AI tools, even if the code is not very good or has mistakes.

RELATED ARTICLE

How retail investors use GenAI to navigate stock markets

As technologies evolve rapidly, AI tools can play a vital role in ensuring software quality and in helping spot critical issues during development, including reviewing updates that might introduce risks and identifying areas that need more testing. Highly skilled developers can leverage these tools to validate code outputs, establish clear coding rules, and integrate code into complex systems.

Ultimately, Professor Kim anticipates that the more sustainable career advantage will likely come from higher-level capabilities, such as understanding software’s underlying problems, evaluating AI output, designing maintainable systems, and coordinating human-AI workflows.

“The career advantage from knowing how to use AI coding tools will not vanish but shift towards those who can use it in more advanced, strategic, and complex ways.”

The post Is AI creating more productive but dull programmers? first appeared on China Business Knowledge.

Featured faculty: Wu Jing and Michael Zhang
Written by Putro Harnowo

Natural disasters have become more frequent and severe each year. In 2025 alone, devastating wildfires struck California, powerful hurricanes ravaged the Atlantic, and floods and cyclones over South and Southeast Asia inflicted profound economic losses. Across the US, the record-breaking natural disasters last year caused US$115 billion in total damage.

Businesses now operate in an increasingly unpredictable environment. Artificial intelligence (AI) is poised to help navigate turbulence, but its value is more evident in optimising business under stable circumstances. This is not surprising since corporate AI investment, such as from big pharma to tech giants, has historically prioritised building competitive advantage over resilience.

“AI undoubtedly helps productivity in normal times. However, given the current high-velocity environment characterised by disruptive upheavals, a better understanding of how to deal with such unrest becomes more urgent,” says Wu Jing, Professor in the Department of Decisions, Operations, and Technology at the Chinese University of Hong Kong (CUHK) Business School.

Disruptions brought by natural disasters impair business operations and erode investor confidence, leading to lower stock prices for affected companies. Professor Wu’s study reveals that companies that hire more AI talent see a smaller drop in stock value and recover more quickly after disasters as opposed to companies with less AI investment.

The interesting part is that companies with limited budgets are found to benefit more from AI investment during crises. However, their productivity gains still couldn’t match their wealthier peers in normal times due to a lack of organisational systems and infrastructure to fully reap AI potentials.

“Resilience becomes more and more important in today’s ever-changing environment. If crisis is the new norm, infusing AI into firm productions is no longer a luxury,” he adds.

How much AI investment is enough?

Along with Michael Zhang, the Wei Lun Professor of Business AI at the same department, as well as Han Miaozhe at the Hong Kong University of Science and Technology and Shen Hongchuan at the University of Macau, Professor Wu’s latest study, Artificial intelligence and firm resilience: Empirical evidence from natural disaster shocks, assess AI’s impact on firm resilience during challenging periods.

The study focuses on 3,137 firms in the US across agriculture, mining, utilities, construction, manufacturing, trade, transportation, and warehousing sectors. To measure AI investments, the researchers identify AI-related job postings and find that the number of job advertisements seeking AI-related skills was low relative to the overall job postings, but has increased from 8.12 per cent in 2010 to 16.41 per cent in 2019.

“Resilience becomes more and more important in today’s ever-changing environment. If crisis is the new norm, infusing AI into firm productions is no longer a luxury.”

Professor Wu Jing

The team also looks into the International Disaster Database and spots 141 disasters that directly affected sample companies, mainly storms and floods. The researchers then merge the dataset with the stock prices data from the Centre for Research in Security Prices and S&P Compustat.

Stock returns represent adjustments in the general expectation of a firm’s performance, reflecting the firm’s ability to mitigate the damage amid catastrophes. Firms that hire more AI-related positions see moderate losses and higher stock returns during and after the disaster. They can fully recover quickly if at least 2.4 per cent of their job postings require AI-related skills, such as deep learning, image processing, AI tool operation, and the like.

This positive impact is greatest at the peak of the disaster, with the most effective AI-empowered roles focused on cognitive tasks, decision-making, and supply chain coordination. “AI generates significant resilience for firms facing natural disaster shocks primarily by optimising supply chains and production inputs,” says Professor Wu.

Natural disasters create tangible challenges that AI can get around, such as rerouting shipments or optimising output from remaining machinery. In the stock market, this benefit relies on shareholders believing the firm can continue operations amid natural disasters.

However, Professor Wu notes that such resilience may not persist in the face of human-induced shocks, such as cyberattacks, labour strikes, or industrial accidents. The damage in human-caused disasters is often reputational or contractual, and AI-driven operations cannot fully offset it. In this case, AI can only serve as a risk detector, providing data to support human operators rather than mitigating the damage.

“AI serves as complementary support for tangible operations and works more efficiently if it targets physical assets, such as factories,” he adds. “When AI is used on financial assets, intellectual property, or market access, its effects are limited.”

How to invest in AI wisely

Many may conflate AI with information technology (IT), as both often go hand in hand, so the researchers seek to examine them more deeply. IT is meant to improve efficiency by coordinating, communicating, and monitoring a wide range of activities, whereas AI focuses on applications where data and algorithms generate predictions to assist decision-making.

The team then measures investments in non-AI technologies by weighing job postings with general IT skills, such as robotics, data analytics, or cloud-related skills. The analysis finds that while IT is great for enhancing day-to-day operations and cutting costs, AI plays a distinct role in helping firms to be more resilient during crises.

Professor Wu suggests that AI investment should not be spread evenly across all technical functions. Instead, focus on encouraging managers who can maximise output with AI when resources are scarce. By concentrating AI capabilities in high-level cognitive and operational roles, firms can improve their resilience.

“Therefore, roles like supply chain coordinators should be prioritised to be empowered with AI skills for better predicting materials arrivals and planning alternative routes to address the disruptions directly,” says Professor Wu. “Strategic decision-makers and production operations managers should also be equipped with AI tools to make quicker, better decisions in resource allocation.”

For financially constrained firms, rather than investing solely in AI tools, the most critical takeaway is to prioritise organisational design, such as training manpower and setting up procedures to ensure AI can make a real difference. “These firms should view AI investment as an insurance premium for resilience instead of an immediate profit engine, ensuring they have the IT backbone to support it,” he adds.

“They should also focus on deploying AI in areas that immediately reduce operational costs or mitigate certain risks, such as vendor monitoring or financial planning. This will allow them to generate the savings needed to fund further resilience tools while avoiding the trap of investing in technology they cannot utilise efficiently.”

The post Can AI help businesses weather any storm? first appeared on China Business Knowledge.

Featured faculty: Michael Zhang
Written by Sally Ho

Social media has levelled the playing field of access to financial knowledge, granting institutional and retail investors alike equal access to timely news and sentiment that shape their decisions. Take the collapse of Silicon Valley Bank in 2023, for example. A panic spiralled on social media turned a bank with US$212 billion in assets into bankruptcy in just two days.

Meanwhile, environmental, social and governance (ESG) performance has become one of the most talked-about corporate lingos, not just among consumers but also as a signal in modern investing. Companies are held accountable for their sustainability claims and practices, and failure to do so may put them in the spotlight or, even worse, lawsuits.

Retail investors nowadays increasingly express their views online, influencing other investors’ behaviour and market movements. On social media, a company’s stance on climate change, labour practices, or corporate governance can spark heated discussions. However, little is known about how much retail investors actually value ESG practices among corporates.

“Sustainable investing has accelerated in the past few years, but most discourses focus on how ESG affects firm value or analyst recommendations,” says Michael Zhang, Wei Lun Professor of Business AI at the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School.

“There is a significant gap between the rising importance of ESG in the corporate world and the lack of understanding regarding how retail investors actually process ESG-related information.”

Professor Zhang’s latest study, Attention or sentiment: How social media react to ESG?, aims to address this gap and finds that, when it comes to sustainability, retail investors’ talk and walk don’t always go hand in hand. While retail investors do care about ESG, they may not be able to determine if such practices are financially material in the short term.

Classic dilemma of profit vs. planet

To explore how retail investors react on social media about companies’ ESG performance, Professor Zhang, along with Kalok Chan at City University of Hong Kong, Xu Dapeng of South China University of Technology, and Hong Hong of Tongji University, examines tens of thousands of posts on Seeking Alpha, a popular online community platform for retail stock investors.

With more than 17 million monthly visitors, Seeking Alpha publishes content written by diverse contributors and features comment sections, forums, and groups where users discuss investment strategies. Given these characteristics, this platform has been dubbed social media for retail investors.

The researchers focus on firms in the S&P 500 index, or the top US-listed companies. After including only firms discussed on Seeking Alpha and having ESG scores, the final sample consists of 429 firms.

“Managers should realise that in the short term, ESG gets you noticed, but financial metrics likely still drive the sentiments.”

Professor Michael Zhang

Given that online attention and sentiment are often drivers of stock market volatility and trading volume, the team categorised the posts into two key responses: social media attention and social media sentiment. Attention refers to how many posts are about a certain company, and sentiment refers to the emotional tone, including optimistic or pessimistic attitudes, towards the company.

After matching and analysing the social media posts with ESG performance scores from Sustainalytics, a global provider of ESG ratings and data, the researchers found that firms with strong ESG scores attract more social media attention. ESG score upgrades are associated with increased social media attention, while downgrades result in decreases in the following months.

However, retail investors on the platform do not necessarily express more positive feelings toward firms with high ESG scores. Investors also do not show more positive sentiment after ESG upgrades or more negative sentiment after downgrades. This suggests a disconnect where sustainability sentiment does not always align with visibility.

Talk is cheap, but vibe speaks volumes

The study further examines the three major components of ESG separately and their individual impact on social media attention and sentiment. The environmental and social components represent the main drivers of increased attention online, while the governance aspect did not have any significant impact.

For online sentiment, social and governance aspects have no significant impact, but surprisingly, the environmental score alone is associated with less favourable sentiment. This finding might seem counterintuitive, but it aligns with a specific stream of financial theory: many retail investors treat environmental efforts as a cost, rather than a clear financial benefit.

“It is likely that retail investors in these online communities often view high environmental commitment as a financial burden,” Professor Zhang says. “From a utilitarian investment perspective, extensive spending on environmental initiatives, like emission cuts or resource reduction, can be seen as an expense that might hurt financial performance.”

Despite being beneficial to wider society, investing in sustainability initiatives can raise questions about short-term profitability and provoke scepticism due to a perceived threat to a company’s bottom line. It appears that retail investors on online forums tend to be driven primarily by financial concerns, rather than environmental causes.

Research implications and future directions

ESG activities may increase visibility among retail investors, but firms shouldn’t assume that higher visibility automatically translates into a positive image, since investor sentiment is driven by profit. “Companies need to manage their expectations regarding the return on their ESG investments when it comes to public perception,” Professor Zhang adds.

“Maintaining ESG standards is crucial for relevance and visibility, even if it doesn’t immediately result in a warmer reception from retail investors. Managers should realise that in the short term, ESG gets you noticed, but financial metrics likely still drive the sentiment.”

For market analysts and regulators, the disconnect between attention and sentiment raises questions about how ESG information is communicated and interpreted. Social media may amplify awareness, but it doesn’t guarantee confidence in sustainability practices or in their financial implications.

Looking ahead, Professor Zhang expects to examine whether the findings hold true for smaller-cap companies, not just S&P 500 firms. “We also plan to explore other digital channels such as X [formerly Twitter], Yahoo! Finance, or Google search to see if the attention and sentiment dynamics might differ on general social media platforms,” he adds.

For now, the key takeaway is clear. In the digital age, seeing isn’t always believing. ESG draws attention, but steering investor sentiment requires more than just doing good.

The post Do retail investors really care about sustainability? first appeared on China Business Knowledge.

Featured faculty: Michael Zhang
Written by Putro Harnowo

Hungry while travelling, but couldn’t decide which restaurant to go to? No worries. Just grab your phone and check the ratings for nearby restaurants. These user-generated reviews appear to be more trustworthy than traditional advertising due to their open submissions, making them handy not only for tourists, but also for locals craving something special.

No wonder platforms like Google Reviews and Yelp are cherished. Local apps like Japan’s Tabelog, South Korea’s Naver and Kakao, Chinese Mainland’s Meituan, and Hong Kong’s OpenRice also thrive on their community-based and more curated ratings. However, these reviews should be taken with a grain of salt. Ratings often vary across different platforms, and a high rating doesn’t always guarantee quality.

“Online ratings can be biased, leading to misleading information for readers,” says Michael Zhang, the Wei Lun Professor of Business AI at the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School.

In his latest study, Why is the grass always greener on the other side? Tourist bias in online restaurant ratings, Professor Zhang finds that one key source is tourist bias in their reviews. “Tourists are about 13.4 per cent more likely to give restaurants a higher rating than locals. If most reviews come from tourists who rate places more positively, locals might feel the restaurant is overrated.”

This tendency is evidently found across all kinds of restaurants, whether chains or independent, cheap or high-end, and among tourists from both large and small cities, across all genders. The study also looks into the very different experiences and expectations about restaurants from local customers and tourists.

Understanding this bias would help businesses better manage customer satisfaction levels and tailor their offerings to meet their customers’ expectations. Meanwhile, customers can leverage the findings to assist their judgment when looking for restaurants.

“Tourists are about 13.4 per cent more likely to give restaurants a higher rating than locals. If most reviews come from tourists who rate places more positively, locals might feel the restaurant is overrated.”

Professor Michael Zhang

Why do tourists tend to give higher ratings?

Along with Xu Da Peng at South China University of Technology, Hong Hong at Tongji University, and Ye Qiang at the University of Science and Technology of China, Professor Zhang collected data from Chinese Mainland’s most popular restaurant review platform, focusing on 10 cities famous for their local cuisines, namely Changchun, Changsha, Guangzhou, Guiyang, Haikou, Suzhou, Tianjin, Xi’an, Xiamen, and Zhengzhou.

The sample consists of 70,950 reviews written by 747 users for nearly 40,000 restaurants with a wide range of prices and popularities. The researchers identify whether a reviewer is a local or traveller by analysing their location information relative to the location of the reviewed restaurants.

Professor Zhang and his team discovered that the positive emotional state while travelling is the reason for tourist bias. Travelling boosts their mood and makes tourists more willing to express their feelings instead of pondering on analytical or detailed reasoning. Reviews from tourists are also shorter but accompanied by more photos, suggesting they care more about sharing their experiences rather than commenting on taste and price. They are also more forgiving of restaurant flaws.

Tourists on the lookout for more than just functional dining are more sensitive to the emotional aspects, such as the restaurant’s vibe and the hospitality they experience, as they mention in their reviews. Meanwhile, locals write more about locations, cooking methods, and prices. The researchers further identify the distinctive words used by each group

Focusing on experience may give the impression that tourists seek more expensive or higher-quality restaurants, but a deeper analysis finds that both locals and tourists who write reviews often avoid fancy establishments. It’s just that reviews from tourists tend to be more positive.

How can restaurants and customers leverage tourist bias?

Given the different interests between locals and tourists, Professor Zhang suggests restaurants apply a dual-focus promotional strategy that addresses the distinct needs of both groups. Restaurants can emphasise service quality to attract tourists, create photo-worthy environments and highlight experiential aspects that tourists can share with others, while also training staff to provide attentive service that generates positive emotional experiences.

To retain local customers, restaurants should focus on fundamentals, such as competitive prices and consistent quality and taste, while offering value propositions that appeal to the rational mind and build a reputation through authentic cooking techniques that locals prefer to mention in reviews.

“To integrate both target markets, restaurants may separate seating areas or dining times to cater to different atmospheres, or differentiate menus by highlighting signature experiences for tourists and the most favourite dishes for local patrons,” he says. “Both groups value quality, so maintaining high standards across all dimensions benefits all target markets.”

For the customers, Professor Zhang suggests taking extra steps to find reviews that better match their preferences and weigh the ratings differently. “Tourists wanting to eat like a local should prioritise longer reviews that highlight the restaurant’s location, cooking methods, value for money, with more analytical wordings. Meanwhile, those seeking a memorable experience should consider reviews highlighting service and ambience, with more photos and emotional expressions.”

As restaurants located in touristy areas may have their overall ratings inflated by travellers, the actual local feedback might be closer to lower-rated reviews.

Taking bias into account for better reviews

To ensure the users receive more accurate reviews, Professor Zhang suggests platforms identify which reviews are submitted by tourists and locals, then calculate separate average ratings for each group and provide functionality that allows users to view ratings from either locals or tourists.

With ever-evolving technologies, machine learning can be deployed to train algorithms to scan and detect reviewer types. Algorithmic adjustments can also be carried out to moderate ratings from identified tourists, considering their tendency to inflate their rating submissions.

For transparency, platforms can add labels mentioning locals or tourists to their reviews or display notes or disclaimers about potential bias in highly rated restaurants with predominantly tourist reviews. Platforms can also show trend lines of ratings separated by reviewer type over time.

Furthermore, platforms can suggest personalised recommendations. For example, platforms can recommend restaurants emphasising service and environment for travellers, while suggesting restaurants highlighting location, cooking quality, and value for local users.

“Overall, we find that the tourist bias is consistent across restaurant types, whether chains or independent, as well as across price levels and cities, making the findings broadly applicable,” Professor Zhang adds.

The post The hidden bias of restaurant reviews first appeared on China Business Knowledge.

Featured faculty: Michael Zhang
Written by Putro Harnowo

There is an app for anything nowadays. Digital platforms have made it easier to shop, work, find jobs, travel, and even date by bridging the two sides in need at their fingertips. With no more face-to-face interaction, searching for the perfect match is as easy as observing pictures and numbers on a screen, as many would imagine.

Perhaps the nitty gritty of the virtual matching is the fact that each user has different preferences. Regardless of the detailed information, a mismatch naturally occurs when preferences from both sides differ, as it prevails in many aspects of life. Imagine when you crave ethnic cuisine but all you see are fine dining eateries, or search for a one-bedroom apartment but all listings offer houses for families.

Information management and strategy become crucial for digital platforms to sustain their businesses. This can explain why Airbnb gives away much data on its interface, but Uber only relies on user ratings and conceals the passengers’ destination until the ride starts. However, less is known about how much information should be shown to alleviate the misalignment.

“Conventional wisdom often suggests that users on platforms should try their best to obtain more information, but we find circumstances where more information can actually hurt the users,” says Michael Zhang, the Wei Lun Professor of Business AI at the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School. “We call this the ‘less information is more’ effect.”

To this extent, dating sites provide a natural setting to analyse such a phenomenon. In a study titled Mr. Right or Mr. Best: The role of information under preference mismatch in online dating, Professor Zhang and his co-authors analysed data from a major dating site in China, and their initial data already suggests notable preference mismatches: men generally prefer a woman 10 cm shorter than themselves, but women like a man 20 cm taller. Such mismatches also exist in other attributes, from income and homeownership to classic astrology.

“The dating market is characterised by a high degree of heterogeneity in terms of both user characteristics and their preferences over potential partners. Therefore, online dating is an ideal context to study preference mismatch,” he adds.

Why is less information more useful?

In the context of dating preferences, it is worth noting that some may argue, “beauty lies in the eyes of the beholder.” While this statement stands true, the preference mismatch in the study describes the preference discrepancy between the two sides of the market, rather than variation within a single side or subjective preference, as the famous phrase refers to.

For two-sided platforms where both sides have preferences and engage in search and matching, the main challenge is creating an effective and efficient search process. Professor Zhang, along with Shen Hongchuan of the University of Macau and Dang Chu of the University of Hong Kong, found that the amount of information influences matching outcomes.

“Conventional wisdom often suggests that users on platforms should try their best to obtain more information, but we find circumstances where more information can actually hurt the users. We call this the ‘less information is more’ effect.”

Professor Michael Zhang

In practice, users receiving a message on the dating app will have the sender’s short profile attached, showing basic information like nickname, age, education and home city. Users can find more details by clicking on a long profile page, which contains information on height, income, Chinese zodiac sign, astrological sign, lifestyle, house ownership and even smoking habits.

Users on the initiation side are called the focal users for easier reference, and those on the other side are called the candidates in whom the focal users are interested. The total data comprises 1,198,943 clicks from 33,504 focal users with search and matching history, including profile-checking, messaging and the reactions from the candidates. A match is considered successful if the focal user and a candidate have exchanged messages at least once.

To measure the preference mismatch, the researchers construct attractiveness scores of each candidate based on the estimated preferences from each focal user. If the candidates whom the focal users like reply to the messages and favour the users back, their preferences are matched, but if the preferences are misaligned, candidates with a higher attractiveness score are less likely to reply.

The analysis found that the preference mismatch happens more often if focal users and candidates click long profiles. Alas, obtaining more information does not enhance the likelihood of matching. On the contrary, candidates approached based on their short profiles are more likely to reply to focal users. The profile photos may influence the outcomes, but after running another set of experiments, the researchers ruled out this factor as a significant concern.

The researchers initially suspected that candidates whose long profiles were clicked may be pickier and, hence, less likely to match. To answer this, they constructed a popularity measure for each candidate based on the number of focal users reaching out. The results show that the average popularity of the candidates in the long profile is lower than in the short bio. Candidates with a lower popularity score are also more likely to be in the long profile group.

“More details about the other side lead to a greater mismatch, likely because when the users learn more about the candidates, misaligned preferences in each attribute start to weigh in and increase the level of mismatch,” says Professor Zhang.

More information can be useful sometimes

While there is no doubt that more information helps make informed decisions, complete disclosure may lead to market failure, especially for two-sided platforms where misaligned preferences are pervasive. However, there are several settings where the “less information is more” effect may be less applicable.

In decentralised platforms like Tinder, Upwork, and Airbnb, where user preferences play a key role in the search process, preference mismatch is likely to exist. Conversely, in a centralised platform like Uber, passengers value time efficiency, and the drivers’ preferences do not matter much, so the platform hides the destination until the ride request is accepted, making preference mismatch negligible.

Considering the mismatch cost—the negative impact a user experiences when failing to find a match that closely aligns with their ideal—a low cost will result in more matches within mutually acceptable preferences, like finding a part-time job. In a setting where mismatch cost is relatively high, like in online dating, users who spend more effort understandably expect to get the best match possible.

Bargaining power between the two sides also matters. “When imbalanced bargaining power exists, the final outcome will depend on the preference of the side with more power,” Professor Zhang says.

While users can reach out to all candidates regardless, time and effort are still wasted on the no-brainer approach. Inevitably, some degree of information helps users screen out candidates who are far from the criteria. “An optimal amount of information is needed before proposing a match,” he adds.

The post How much info is too much to get the right match? first appeared on China Business Knowledge.

Featured faculty: Michael Zhang
Written by Putro Harnowo

A good decision requires critical thinking and time but many of us don’t have enough privilege or patience to ponder sometimes. To illustrate, if you ever bought a sale product right after seeing its before-discount price, then you have carried out a fun heuristic exercise or mental shortcut.

Heuristic refers to a mental process when individuals seek to reduce complex tasks to simpler operations to process information with less effort. Although the results may not be optimal, heuristics can be good enough when solving problems with limited timeframes in routine minutiae.

In peer-to-peer lending platforms, this pragmatic method is not uncommon among borrowers. The most conventional one is using round numbers. This is not surprising as when dealing with precise numbers like 29 or 15.15, people tend to round it up or down to get simple figures. Putting round numbers requires less thinking and is associated with feeling-based decision-making.

“Consequently, the use of round numbers may lead to suboptimal decisions,” says Michael Zhang, Wei Lun Professor of Business Artificial Intelligence at the Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School. “When using the round-number heuristic, decision-makers may sacrifice accuracy for convenience.”

In a study titled Numerological heuristics and credit risk in peer-to-peer lending, Professor Zhang, along with Maggie Hu of the City University of New York, Li Xiaoyang of the Hong Kong Polytechnic University, and Shi Yang of Deakin University, found that loan applications on Chinese lending platform using round amounts have lower success rates and are more likely to be overdue.

Heuristic often mingles with superstition. A notable example in China is applying for loans by incorporating number eight as the word sounds like “wealth”, and surprisingly, loans incorporating this lucky number are more likely to get funded and have lower delinquency rates. “Borrowers who use lucky numbers make more effort to understand the lenders’ preferences and make their loan applications more attractive,” he adds.

As digital finance grows, lending platforms are likely to integrate behavioural data into their artificial intelligence-driven risk assessment tools. Professor Zhang suggests platforms leverage the study by tailoring loan recommendations based on users’ financial behaviour and heuristic tendencies.

“By identifying borrowers’ tendencies to use heuristics like round or lucky numbers, businesses can segment their customers into different risk profiles.”

Professor Michael Zhang

Finding the right mental shortcut

The researchers gathered data from a leading Chinese peer-to-peer lending platform from October 2010 to January 2016. Before submitting a loan application, borrowers must register and provide information on their gender, age, education, and income. A registered borrower can submit loans in multiples of 50 Chinese yuan (US$7.62), ranging from 1,000 to 3 million Chinese yuan, and the lenders will decide whether to bid or not.

The researchers obtained a sample of 611,079 successful loans, with 64.85 per cent of them using round numbers like 1,000, 5,000, 10,000 and the like. Only 9.45 per cent used lucky numbers containing the number eight, such as 8,000, 28,000, or 88,000, etc., but not number four as it sounds similar to “death” in Chinese.

The analyses show that round-number loans have a 73.79 per cent lower chance and take 1.05 hours longer to get funded with a 6.37 per cent higher chance of being overdue than non-round numbers. Conversely, loans with lucky numbers are 42.12 per cent more likely to be funded and spend 0.19 hours less on bidding time with a 1.80 per cent lower delinquency rate. For borrowers who have never used round and lucky numbers, adopting round numbers would drop the chance of securing loans by 66.73 per cent, but lucky numbers would raise their odds by 69.47 per cent.

To test the repayment performance, the researchers used 5,948,938 monthly repayment records from 217,237 funded loans. The results confirmed that borrowers with round numbers have lower funding success rates and higher delinquency, but those with lucky numbers have the opposite results. Borrowers who resort to round numbers are also more likely to experience difficulties in budgeting or make negligent mistakes when repaying the loans.

Implication for fintech businesses

Professor Zhang highlights the importance of peer-to-peer lending platforms in understanding user behaviour. Platforms can create loan packages or promotional campaigns that align with user preferences as superstitious borrowers may be more responsive to offers incorporating the number eight or avoiding the number four, especially in cultures where these numbers have significant meanings.

Financial technology or fintech firms can incorporate user preferences into marketing strategies to strengthen customer engagement and trust. They can also offer more personalised tools, educational content, or artificial intelligence-powered suggestions for borrowers submitting round-number loans, which may indicate less financial planning.

“By identifying borrowers’ tendencies to use heuristics like round or lucky numbers, businesses can segment their customers into different risk profiles,” says Professor Zhang. “This segmentation could guide marketing efforts, such as offering lower interest rates or faster approval processes for borrowers using lucky numbers, who are more likely to repay loans on time.”

The study could also help lenders enhance risk assessment models by incorporating heuristic preferences as indicators of creditworthiness. Stricter underwriting criteria (e.g., additional income verification) may be applied for loan applications with round numbers to mitigate potential risks. On the other hand, lenders can prioritise applications with lucky numbers for faster approval or offer preferential terms, as these borrowers are more likely to repay on time.

“By incorporating heuristic usage data into machine learning algorithms for loan performance predictions, this additional layer of information can improve the accuracy of delinquency forecasting and help lenders allocate resources more effectively,” he adds.

Forecasting the future

Professor Zhang sees that traditional credit scoring models may evolve to incorporate non-financial indicators, including heuristic-driven behaviours, as proxies for financial responsibility or sophistication. “As alternative lending models expand globally, platforms will need to adapt their practices to local cultural contexts, leveraging insights like superstitious number preferences or other decision-making heuristics.”

Furthermore, platforms could gamify loan application processes and encourage borrowers to move away from mental shortcuts by offering rewards for more precise financial planning. For regulators, fintech platforms may be required to disclose how heuristic-driven behaviours influence lending decisions, ensuring transparency in how borrowers are assessed and treated.

RELATED ARTICLE

Sustainable ways to grow cash-burning marketplaces

Considering cultural and contextual differences, Professor Zhang believes that the findings can likely be applied to other forms of crowdfunding. Given the influence of numerology is strong in some Asian markets like China and Japan but less prevalent in Western countries, other types of heuristics may play a more prominent role. Crowdfunding platforms should adjust their strategies based on local norms.

“While the specific numerological preferences may vary, the broader principle—heuristics influence decision-making—can be generalised across different cultural and economic contexts,” he adds. “Platforms worldwide can use these insights to better understand user behaviour and improve decision-making frameworks.”

The post How mental shortcuts guide loan success and repayment first appeared on China Business Knowledge.

By Ellis Ng

Blockchain technology has emerged as a transformative force in the digital economy, with cryptocurrencies like Bitcoin and Ethereum becoming household names. The past few years have also seen cryptocurrency bubbles burst into a crash, yet many survive and get stronger.

In mid-November, the value of Bitcoin reached record highs of around US$92,000 as part of the latest rally in cryptocurrencies and growing mainstream adoption. The rise comes as major trading platforms report millions of new users and the crypto market value hit a record US$3.2 trillion valuation, though the sector remains volatile and unpredictable.

“Blockchains came to the world as a novel technology and rapidly attracted an industry built around it,” says Li Tianyi, Assistant Professor at the Department of Decisions, Operations and Technology of the Chinese University of Hong Kong (CUHK) Business School. “Emerging technologies hold promises while often undergoing hype cycles of their utility.”

In a recent study titled Development trajectory of blockchain platforms: The role of multirole, Professor Li and the Wei Lun Professor of Business Artificial Intelligence at the same department, Michael Zhang, revealed how blockchain has enabled participants to act simultaneously as users, investors and labours, which engenders complex dynamics of the system that have influenced platform development.

Professor Li and Professor Zhang’s research analysed over 100 blockchain token price series and developed a novel model to explain platform development patterns. It suggested that the success of platforms is dependent on how well these roles of platform participants are balanced, with different roles dominating at different stages of platform development.

The research found that blockchain platforms demonstrate a three-phase development trajectory, characterised by the relative dominance of participants’ different roles. By studying the price data of platform tokens, the researchers created an effective parametric model that could be used to predict and explain how platforms grow.

Their model outperforms traditional scaling models in predicting network growth and is extended to analyse “forking”, those events in which blockchain platforms branch into competing versions. The creation of Bitcoin Cash in 2017 from a hard fork of Bitcoin, as well as Ethereum Classic and Ethereum from a 2016 hard fork, are prominent examples.

Professor Li and Professor Zhang’s findings suggest that the timing of a fork matters more than the size of the split. Counterintuitively, these splits can help the original platform by increasing its visibility in the market.

Moving from chaos to stability

The study shows that the initial phase of a blockchain platform begins with a small, dedicated community of labourers, which refers to participants who contribute to a blockchain ecosystem by performing various digital work, including supporting the blockchain’s operation, maintenance, and development. The second phase marks rapid growth driven by investment activities. A mature blockchain platform will then settle into stable market cycles.

“Blockchains came to the world as a novel technology and rapidly attracted an industry built around it. Emerging technologies hold promises while often undergoing hype cycles of their utility.”

Professor Li Tianyi

“In this trajectory, the beginning stage of the platform is characterised by intense labouring activities within a small participant group, the intermediate stage is characterised by increasing investment activities that rapidly boost the platform’s population base, and the mature stage is characterised by recurrent market oscillations between a high and a low participation mode,” says Professor Li.

Early participants are attracted by high labour rewards. For instance, blockchain miners receive block rewards for successfully mining new blocks, and the increasing value of the platform’s native tokens gains significant returns, among others. However, these rewards often decline quickly as more people enter the network. The platform then enters a stable, predictable development, with user numbers rising and falling in cycles that affect its value and use.

This phase will often set the foundation for later growth to the next phase, as investment takes centre stage. When a platform grows, it creates price differences that traders can profit from, making investment opportunities more attractive and turning many investors into active participants.

User numbers grow in waves, hitting new highs with each cycle, and the increased visibility draws in even more participants who find practical uses for the platform. Eventually, when joining offers fewer benefits, user growth slows down and stops. Rather than reaching higher peaks, user numbers rise and fall in stable patterns.

The study also found that two factors—the time delay in quitting the platform and the holding time of platform tokens—were particularly crucial in differentiating successful platforms from failing ones. A larger time delay indicated higher trust in the platform and higher switching costs, leading to more user “stickiness”, the research noted. Meanwhile, the longer an investor holds a token, the lower its price volatility or transaction velocity, which translates to more investment stability.

Engagement is king

Platforms need to effectively engage all three participation roles — users, who utilise the platform’s services, investors, who create the financial landscape, and labourers, who maintain the network — to achieve sustainable growth, the researchers suggested.

In the early stage, labourers — like miners or validators – are crucial in building the foundation. During the growth phase, investors drive rapid expansion, with regular users keeping the platform stable as it matures.

The research also showed that platforms supporting all three roles were resilient and showed better development trajectories than those focused on fewer roles. When one role becomes less active, others can help maintain a platform momentum.

“We hope that this study can bring some utility to different stakeholders interested in blockchains,” says Professor Li. One concrete application of this model could be in decision-support simulations to assist blockchain entrepreneurs with platform design and early-warning detection. The model can also aid in strategic investment analysis and regulatory policymaking, according to Professor Li and Professor Zhangy.

As blockchain technology continues to evolve and new platforms emerge, Professor Li hopes these findings will help stakeholders better understand and manage platform development.

“This model provides a ground for analysing the profitability of participants’ activities on tokenised platforms, [meaning] platforms with blockchain implementation, through their token investment and crypto labouring,” Professor Li says. Furthermore, he and Professor Zhang plan to extend the model to discuss the earnings on the platform and people’s strategic participation that maximises the earnings.

The post Critical paths to sustain blockchain growth first appeared on China Business Knowledge.

By Jaymee Ng, Principal Writer, China Buisiness Knowledge@CUHK

The COVID-19 epidemic has ravaged labour markets around the world. In April, the U.S. unemployment rate hit 14.7 percent, its highest since the Great Depression. Even mighty China, the global driver of economic growth, saw the official unemployment rate rise to a record 6.2 percent in February, after 5 million jobs were lost in the country in the first two months. Elsewhere, the United Nation’s International Labour Organization forecasts 1.6 billion people globally are in immediate danger of losing their livelihood.

However, every cloud has a silver lining (so the saying goes), and a recent study by The Chinese University of Hong Kong (CUHK) has outlined proof that rising unemployment may in fact encourage more people to engage in activities that are beneficial to society.

The research study, titled “Unemployment and Digital Public Goods Contribution”, was conducted by Michael Zhang, Associate Dean (Innovation and Impact) and Professor of Decision Sciences and Managerial Economics at CUHK Business School, Prof. Michael Kummer at University of East Anglia and Prof. Olga Slivko at Erasmus University.

“We found higher unemployment was associated with higher participation by volunteers on Wikipedia and an increased rate of content generation.”

Prof. Michael Zhang.

As a proxy for volunteering activity, the paper focuses on content contribution on Wikipedia, the open-source online encyclopedia, during the European debt crisis, when several countries in the bloc were unable to refinance their government debt. This was followed by a surge in unemployment.

“We found higher unemployment was associated with higher participation by volunteers on Wikipedia and an increased rate of content generation,” says Prof. Zhang. “Wikipedia articles were being read more frequently in areas where unemployment was going up, and this was followed by increasing numbers of edits that were made by anonymous or casual editors. Subsequently, the number of highly active users grew and the content on the platform increased as a result.”

Anaylsing Public Good

The researchers first focused on a district-level analysis for Germany. They chose German data as a starting point because of the Kurzarbeit programme implemented by the government there, which allowed businesses experiencing lower demand to reduce the working hours for their employees. Under this initiative, companies kept paying their employees according to their hours worked, and the government would pay workers 60% of the forgone income.

The results confirmed that rising unemployment in German districts led to higher participation by volunteers and increased content generation on the German edition of Wikipedia. The number of participating editors, the number of highly active users and the number of edits to Wikipedia articles all recorded increases.

Specifically, an increase in anonymous content contributions was recorded in affected districts that experienced an increase in unemployment, which suggests that these anonymous members are likely new users to the Wikipedia community. The amount of contributed information, as measured in kilobytes, rose by about 9 percent in January 2009 (when the crisis became significant for the German economy) and by almost 10 percent in the second month. Almost 4,340 additional edits and 55 megabytes, each equivalent to 1,000 kilobytes, were recorded over the next six-month period.

To examine the robustness of the results from the German dataset, the researchers replicated the study on European countries with unique languages that were affected by the crisis. The final sample contained 20 Wikipedia language editions, including Bulgarian, Greek and Italian. Other heavily affected countries, such as the U.K. and Spain were excluded as their languages are spoken all over the world. The researchers found even stronger effects on content generation because of the higher variation in unemployment in Europe during the crisis.

According to the study, there was a 14 percent increase in the number of casual Wikipedia users who made between five to 100 edits a month and a 13 percent increase in users who made more. The viewership of Wikipedia in the affected countries increased from between 8 percent to 20 percent more rapidly than in the unaffected countries in the 12 subsequent months after the start of the crisis. The growth of Wikipedia content, as measured in additional words, grew substantially more rapidly at the rate of between 12 percent and 22 percent and was clearly stronger in later months.

Practical Implications

Prof. Zhang explains that this research study has important practical and policy implications. “First, many volunteering organizations provide valuable public goods that range from disaster relief to publicly accessible knowledge, which matter to the wider economy” he says, adding that understanding the relationship between economic crisis and the propensity of people to do volunteer work is critical to them.

There’s been debate in the media on both sides of the Atlantic whether people are more or less likely to do volunteer work when the economy is in trouble. However, such reports have been based on “physical” volunteer work in a formalised setting through a charity, and they ignore socially beneficial work done outside this context.

“Second, individual working hours have been decreasing for decades in many societies. With the increasing trend towards automation and digitisation, the trend to ever decreasing needs for human labor could continue,” Prof. Zhang comments.
uch social and economic restructuring brings new methods of production and consumption, but one unfortunate consequence of the progress is that workers with older skills are displaced from their jobs. Our results support that human beings respond to structural economic shifts by reallocating their time to peer production-related activities,” Prof. Zhang comments.

The valuable knowledge contributions that occurred due to worsened economic conditions also represent a beneficial side effect that has been overlooked by policy makers. It could be a novel channel through which private motivations can foster contributions to an online public good.

“Policymakers could enhance these beneficial effects by encouraging active groups in society to contribute online knowledge more systematically,” says Prof. Zhang. “While financial aid may jump-start economic recovery, government support on contributions to online public goods will provide valuable channels for knowledge exchange and help encourage skill upgrades in society, turning a crisis into an opportunity.”

The post Is There a Silver Lining in Unemployment? first appeared on China Business Knowledge.

By Jaymee Ng, Principle Writer, China Business Knowledge @ CUHK

We live in a social world where we are constantly sharing our opinions and exposed to other people’s opinions. Some of these opinions are useful especially when we are thinking about buying a new product or service. In fact, previous report stated that more than 80 per cent of consumers are likely to research online before making purchase.  In other words, online reviews of products and services can have a real effect on potential consumers.

We would like to think that online reviews are unbiased and neutral so that we can rely on them. However, a recent research study by Prof. Michael Zhang from the Department of Decision Sciences and Managerial Economics and Associate Dean (Innovation and Impact) at The Chinese University of Hong Kong (CUHK) Business School reveals that those online reviews could be influenced by our friends.

“Our paper examines friends’ social influence in online ratings using data from a large social network in China. And we find that online rating contributors are socially nudged by their friends when giving their ratings,” says Prof. Zhang.

“One challenge in this setting is that we cannot directly observe individuals’ original ratings, so how can we determine if individuals modified their ratings after observing their friends’ ratings? We are able to identify this effect with a clever research design,” he explains.

The Study

For the purpose of the study, Prof. Zhang and his collaborators from Peking University and University of Maryland obtained data from one of the most influential online rating websites for books, movies and music in China. The site has more than 8 million registered users and over 10 million page views per day.

In February 2008, the site introduced a friend network function enabling users to send friend requests and get updates about their friend’s ratings. The main analysis of the study focused on the observations of book ratings from February to August 2008, which amounted to a data set of 171,588 ratings by 33,605 users.

“Salient information about friends’ ratings is a critical condition for social influence to take place. On one hand, the salience of a friend’s rating-information feed enables a user to be aware of his friends’ expressed opinions. On the other, users are also aware that their friends will be able to access their expressed opinions easily,” says Prof. Zhang.

“As for marketing practitioners, it is important to identify early adopters and take their social influence into consideration when making plans to respond to online consumer ratings and reviews.”

Prof. Michael Zhang

Homophily Effect versus Social Influence

However, giving out similar ratings on the same product could mean that their friends share similar tastes with them – which is commonly known as the homophily effect. The researchers think it is important to distinguish whether similar ratings by friends are caused by the homophily effect or social influence since the two effects would lead to very different strategic implications for marketing managers.

“If homophily is the dominating force behind the similarity in ratings given by friends, then managers should not be concerned about the high correlation between friends’ ratings,” says Prof. Zhang.

“If, however, it is social influence that induces later reviewers to give ratings conditional on their friends’ earlier ratings, then the rating-score trajectory will be path dependent, and whoever leaves the first rating will influence his friends’ future ratings.”

To separate social influence from the homophily effect, the researchers examined the similarity in ratings before and after users become friends. The before ratings are used as a benchmark to show whether the users share similar tastes.

“Our results show that users’ earlier ratings do exert social influence on their friends’ later ratings; and the rating similarity increases by 190 per cent after the establishment of friendship,” he says.
e recent friendship has higher influence on social ratings. In addition to the explanation that new friends usually receive more attention than old friends, the result reveals that the social influence in online product reviews weakens over time.

It also shows introducing a friend has no significant impact on social rating similarity and that the social influence only takes place after the users became friends. In other words, one does not conform to another’s opinion by wanting to become friends with the other person.

“The increase in rating similarity takes place only with friends’ ratings, even when we introduce public ratings into the model. This result supports our finding that the identified social influence is indeed due to conformity pressure among friends,” Prof. Zhang explains.

“We also find that friends’ influence is more salient for older books. Yet, there is no significant relationship between rating intensity and social influence. This suggests that, rather than following the mainstream, users conform to their friends in an attempt to develop and manage relationships within their online community.”

Moreover, users with larger number of friends experience weaker social influence, given the challenge of attending to everyone in a big network of friends.

Other findings indicate that users usually read their friends’ ratings after they finished reading the books themselves, meaning the effect is more significant in the post-consumption stage. In addition, such social influence is stronger for extremely negative ratings, yet there is no evidence to show such influence works with extremely positive ratings.

Implications

The study offers important managerial implications to online rating-system designers and marketers.

“Systems designers can use our results to nudge their users. For example, ratings sites can make use of algorithms to recommend reviews not subject to the influence of social ties, highlight the reviews from users who do not have prior friends’ ratings, or even use a warning sign to tell users that some reviews might be influenced by friends’ ratings,” he says.

“As for marketing practitioners, it is important to identify early adopters and take their social influence into consideration when making plans to respond to online consumer ratings and reviews,” he adds.

“Our study shows that although social-networking features might help attract users and improve stickiness, social nudge as a result of interactions among friends may prevent users from giving independent evaluations of products,” Prof. Zhang warns.

The post Should We Trust Online Reviews? first appeared on China Business Knowledge.