Networks

Cascading Political Unity Through Information Cascades

Recently, we have been hearing the same story with every election that comes around — that our country is becoming increasingly divided politically. Indeed, political polarization is a phenomenon that is deeply rooted in the United States’ history. The advent of social media and a boom in the number of online information-sharing sites has only grown this political division. Nowadays, it is common to see images of social networks highlighting this divide between people, and revealing how the interaction between people of opposing political views has broken down. The following diagram of a network, for example, shows how American Facebook users after the 2016 presidential election shared little common interests:

 

Thus, many of us have turned our blame on media companies, blaming them for radicalizing consumers through their content. We can apply what we have learned in class about information cascades to address this issue. Information cascades can occur when many people make a decision based on the decisions of other people, even if their own information contradicts the decision other people have made. Using a model, Tokita et al. explore whether social connections and networks are reordered due to partisan media coverage. The model featured an information ecosystem of polarized media sources, with sources differing in the amount of coverage they give to each topic. It predicted that this sort of ecosystem would result in ties breaking down between people with different ideologies. Social media users first receive a news story through any of their ties but only use the coverage of their preferred source when later reacting to a news story. After reacting, social media sites normally offer users to make a decision. Twitter users, for instance, would have the decision of remaining followed by some account or breaking ties with some account by unfollowing it. As coverage increasingly differs, social media users are more likely to prefer news sources that disagree with how much coverage they should give to certain topics, and subsequently unfollow sources that have differing coverage. This is shown below:

 

 

As shown above, our social network would most resemble the leftmost diagram that depicts high polarization, and a scenario when information sources differ in coverage. Towards the right of the diagram, we have little polarization, as news sources have similar coverage, and thus are less likely to upset their followers, who might prefer a different source. To check the prediction of this model, Tokita et al. then conducted a study where they observed 1000 Twitter users, and their following activity of factual news outlets, and outlets that offer more opinion in their stories. This study reinforced the results of the model, as it found that cross-ideology unfollows for opinionated news outlets were greater than they were for factual outlets. Thus, Twitter users “sorted” themselves out by cutting ties with outlets that differed in the coverage of their preferred outlet, splitting themselves into clusters based on ideology. Users observe the Twitter following patterns of other users and are more likely to create a tie and follow users who share the same ideology.

Source: https://www.pnas.org/doi/full/10.1073/pnas.2102147118

LUNA coin and its information cascade

Terra(Luna) coin, created by the user Do Kwon, was one of the most popular types of cryptocurrency in the crypto market. Now, it is known for its largest collapse ever, with an estimated 60 million dollars of wipeout, significantly affecting the people who invested mass amounts of money into the company. Through analyzing sudden attention and the collapse of this company, we can analyze the information cascade factors we have learned in our lecture. 

 

Let us first know what was so attractive about this coin to the crypto enthusiasts. Because the ancestor of crypto Bitcoin required too much energy to extract, the alternative solution known as Altcoin came out. The problem of Altcoin was the lack of credibility due to its unstable movement of prices each day. Terra created a solution by creating its sister coin Luna coin. Luna coin operated as a Stablecoin, thus creating a symbiotic, and stable relationship in case of Terra’s failure. This Stablecoin solution was a known solution to many crypto industries as well, but what thoroughly induced the people was its promise to give 20 percent worth of interest once invested. The collapse occurred once 2 billion worth of Terra was unstaked, decreasing its value to approximately 0.9 USD per Terra. The reason still remains unknown, whether it is due to a rising interest rate or an intentional attack towards Terra blockchain. As a result, traders started to exchange its $0.9 worth of Terra to $1 of Luna. This caused panic to other investors, creating more exchange for Luna, and due to its over-production, the value of Luna essentially became worthless, creating one of the largest collapses of crypto in history. 

 

Now let us examine the information cascade of these occurrences. 

 

Information cascades are caused when people make decisions without looking at their own observations. What would have been the first people’s observation of Terra(just like the first person to pick up the marble from the urn)? They knew that Do Kwon was studying engineering in Stanford University and had experience of being an engineer in Apple and Microsoft. They also heard about his hyped promise for a guaranteed 20 percent interest. They also know that Do Kwon gathered around 200 million dollars from investment firms, though critics still questioned his technological plausibility. These factors were convincing enough to make decisions for many observers then blasted the value of Luna to 40 billion dollars, creating excitement and hype to regular traders as well as wealthy investors. There were over 280 thousand people in Korea who invested in Luna Coin, yet most of them were not able to listen to the critics during its rise of attention. Without knowing much information about the functionality of this crypto, people started to make blinded decisions, hoping to earn the profit like other successful crypto investors did. There were plenty of flaws with Do Kwon’s 20 percent interest promise and has been questioned and challenged by many professional analysts. Yet for the investors, what they saw was the numbers: 40 billion dollars and 20 percent interest. When Terra eventually fell below a dollar, it created another massive information cascade of people looking for an exchange without knowing the occurrences. This in fact can happen to many investing markets, where people look at the short term results (like the observation of first two people in the urn research) and finalize their decisions. Hence, the Luna coin incident both started and ended with an information cascade, with people focusing only on the graphs and not on the rudimentary underpinnings of Do Kwon’s flawed technology. 

 

References

https://www.nytimes.com/2022/05/18/technology/terra-luna-cryptocurrency-do-kwon.html

 

https://www.forbes.com/sites/qai/2022/09/20/what-really-happened-to-luna-crypto/?sh=5b39890d4ff1

5 Monkeys and an Information Cascade

https://www.psychologytoday.com/us/blog/games-primates-play/201203/what-monkeys-can-teach-us-about-human-behavior-facts-fiction

You’ve probably heard of the (usually rhetorical) question, “If all your friends jumped off of a bridge, would you too?”, as it’s often used to reprimand people who draw inferences from the actions of others rather than using logic to make a decision. The phenomenon this question refers to is a direct consequence of actors being connected in a network. From the clothes we buy to the restaurants we dine at, our decisions are constantly influenced by others in some way. To be more specific, it’s often that the information we get from observing others overpowers our own private information when it comes to making a decision. Let’s examine this phenomenon further through the 5 monkeys experiment.

For context, the 5 monkeys experiment generally went as follows:

• A cage contained 5 monkeys, and inside this cage, experiments hung a banana with a set of stairs leading up to it.
• As soon as any monkey reached the base of the stairs, each of the monkeys would be sprayed with ice-cold water.
• Eventually, all of the other monkeys would physically prevent any monkey from climbing the stairs.
• One by one, the experimenters replaced the monkeys with one who had never experienced the ice-cold water. Still, monkeys would prevent others from climbing the stairs despite not knowing why.
• Even with each of the original monkeys replaced and the threat of the hose gone, no monkey would allow another to reach the stairs to the banana.

The results of this experiment indicate issues with the use of tradition, but perhaps more importantly the potential dangers of information cascades. In class, we learned that in theory, all it would take is the decisions of two individuals to create a powerful and unvarying chain of decision making. Ultimately, while it is nice to have empirical data, all of it should be highly scrutinized with reason if it’s going to support our beliefs. With the voting season underway, keeping these considerations in mind are more important than ever.

 

Information Cascade: Gamestop Incident

Information cascade has become more prevalent in society than ever before as the Internet provided the means for a faster, more effective communication platform for users. People are easily influenced by other people’s decisions as opinions and choices by others are easily accessible on the web. According to the textbook, an information cascade occurs when a sequential choice of information is decided by users where choices of users later down the road get affected by previous users. 

One example of this concept is the Gamestop Reddit incident. Drawing information from the CNBC article, we can understand how the information cascade has shaped a significant shock to the investors and hedge funds at Wall Street. To briefly overview this incident, it first occurred when individual investors on an online platform called “Reddit” started gathering shares and call options to the GameStop stock for which the Wall Street investors piled short options for. In the article, Jim Paulsen, a chief investment strategist at the Leuthold Group, analyzed the effect of this incident as “retail investors with the help of technology acting as a union in attacking is a new phenomenon.”

This incident had a significant impact on Wall Street where individual investor’s effort congregated to make a huge output. This has very close ties with the concept we learned in class: information cascade. The initial choices of individuals on Reddit investing more in call options for Gamestop affected the choices of individual investors later on. The medium, Reddit, provided access for the platform users to access choices made by other people who chose earlier. The individual decisions all came to a single goal of investing more into Gamestop. Ultimately, this information cascade occurred at Reddit proved the extent of individual investors on the stock market. 

Source: 

CNBC News: https://www.cnbc.com/2021/01/27/gamestop-mania-explained-how-the-reddit-retail-trading-crowd-ran-over-wall-street-pros.html

Girls Who Invest: A Road To Success

A network effect is a concept whereby the value of the service provided by a platform increases as more users join. Consider a platform like Facebook as an example. For each additional user joining Facebook, Facebook becomes more valuable to advertisers as they can find diverse individuals interested in their advertisements. This is an illustration of a two-sided network effect.

Many platforms have or will have a strong network effect, but they might be hard to see initially. Because the advantages aren’t immediately apparent, people frequently underestimate these platforms with slower networks. In typical circumstances, such as social media platforms, network effects are readily perceived and recognized. In this article, I use network effects to examine Girls Who Invest, an educational non-profit,  assessing the types of nodes and the total value of its network as time passes.

In order to increase the number of women in senior leadership and portfolio management positions in the asset management industry, a nonprofit organization called Girls Who Invest (GWI) was established in 2015 [1]. The organization’s summer program offers four weeks of educational training led by prominent university professors to sophomore women in college interested in investment management. This is followed by a seven-week paid internship with a partner company. This program has a strong emphasis on empowering women from socioeconomically disadvantaged backgrounds and those who belong to historically underrepresented minorities.

 

We observe three things when we analyze how network effects fuel this educational program. GWI’s user base of students and partners demonstrates classic network effects. As Girls Who Invest gains students of high caliber, they would, in return, be able to:

1. Expand their network value since more alumni lead to prospective connections who may give current students mentorship on job opportunities or professional advice.
2. Find more partner companies who want to choose to hire GWI scholars as interns or full-time employees.
3. Gain partnerships with universities and educational institutions who are willing to host education sessions to educate GWI’s students further

GWI gains credibility by accepting highly qualified sophomore women interested in a career in finance. These sophomores apply to the GWI application, which opens once a year. These women showed promising futures as they would then gain internship experience at firms that partnered with GWI. Interns would then go on to accept full-time roles later on in their careers at reputable companies, which is where GWI closes a singular loop.

Here we can see that programs such as GWI or educational organizations have long feedback loops and user cadence spans over two years. In contrast, social media platforms like Facebook have fast networks where users can gain access to the forum and see the value of their participation immediately. 

GWI benefits from its ability to identify students that are made to succeed in asset management. This loop further plays into the ability of GWI to maintain and grow a solid alum base, as alums can help incoming GWI members and encourage others to also find reputable full-time offers. Companies that partner with GWI would want to continue this collaboration, and more firms would want to join as they feel candidates are highly skilled and may add to their company’s value. Colleges would continue to support GWI’s educational presence on their campus as they would feel as though they are contributing to this social goodness. It takes two years for GWI to close its loop, as that is when its members would get full-time positions. This added value, while delayed, is able to give the non-profit quality connections. 

GWI’s network is characterized by a lag between the creation of the network and the emergence of value. Higher education and other programs of education progress considerably more slowly than even GWI since their benefits are seen over a much longer period of time. The benefit of GWI networks is that, once established, it would be typically difficult to replace. This reality can be seen as it has partnered with several prestigious colleges including Notre Dame and the University of Pennsylvania since 2015 [2]. In the end, GWI’s business model is able to not only create a reputable brand for itself but also empower women’s voices across corporate America.

Reference:

1. https://www.girlswhoinvest.org/sip
2. https://news.nd.edu/news/notre-dame-girls-who-invest-partner-to-advance-women-in-finance/

Cascade Network Effects and International Policy Diffusion

Information cascades have important implications for international policy diffusion. The theory of international policy diffusion has attributed the reason for the spread of certain policies to different countries to different causes – while some simply cite the success of a policy as the reason for its spread to other countries, others claim it is due to other less objective measures, one being a perceived validity of a policy due to its existing popularity. This effect is often connected to the middle accelerating section of the S curve that is known to model the spread of international policy diffusion. This effect is described by Weyland’s 2005 piece entitled Theories of Policy Diffusion: Lessons from Latin American Pension Reform. After the policy shows some signs of success, and spreads to some other countries, it tends to have an explosion of popularity where many other countries pass similar versions of the bill in a somewhat short timeframe. Additionally, the spread tends to follow a geographic pattern, where countries that have more exposure and geographic proximity to the origin country pass the mimic bill sooner.
The diffusion of a bill thus can likely be modeled with a network cascade effect, where a policy, choice A, spreads to its neighboring nodes first if the q, or perceived benefit of choice A, is low enough, and then spreads more rapidly as more countries pass the bill in a highly connected system such as the global network we have today. Policies often spread through trade agreements or market pressure, thus emphasizing the role of strong ties to the spread of this policy. However, these policies often spread beyond the geographic region of the origin country and spread to other geographic regions over a longer period, thus indicating that weak links and bridges still can carry the policy change, also indicating a lower q value, and a higher a value, the perceived benefit of passing the policy. These network cascades can shed light on the mechanisms of policy diffusion, and how much international interconnectedness and perception of a benefit of a certain policy play in this historic but evolving pattern in international politics.

https://www.jstor.org/stable/25054294#metadata_info_tab_contents

Rich Get Richer in Socioeconomic Class at Universities

Research: https://huolab.psych.ucla.edu/wp-content/uploads/sites/5/2021/08/Ni-Goodale-Huo-SOE-2019.pdf

We often hear the phrase ” the rich get richer and the poor get poorer” in a number of situations, such as web links, social media, medical health, wealth, etc. In their research, Dr. H. Wenwen Ni, Dr. Brianna M. Goodale, and Dr. Yuen J. Huo analyzed this concept in the context of academic performance for students with varying socioeconomic status backgrounds. In their research, they split their sample size into two groups, categorizing them into high socioeconomic status (high-SES) and low socioeconomic status (low-SES). They then tested both groups in two types of testing environments, one with affluence cues and one without. The rooms with affluence cues contained common items usually found in study rooms however they were modeled to follow older decorative trends, signifying generational wealth. Rooms without affluence cues were often decorated with simpler designed items, simulating a lower-cost environment. The test itself was a standardized math test as most humanities and English exams contained some level of cultural influence.

From running this experiment, the results were that high-SES test-takers performed better when in an environment with affluence cue in direct comparison to other high-SES test takers who were placed in a room without any affluence clues. However for low-SES students they saw no particular difference in those who took the exam in a room with or without affluence cues. Then they directly compared the results of high-SES and low-SES students and saw that when affluence cues were present, high-SES students outperformed low-SES students, whether the low-SES students themselves tested surrounded by affluence cues or not. It was concluded that this may be the case due to the affluence cues making the environment more comfortable to high-SES students who had grown up in a similar environment so these cue’s were in parallel to their own social identity.

From this experiment we can learn that, when low-SES students by pass the obstacles placed on their path due to their socioeconomic status to simply get into college and pursue a higher education, they may still continue to be disadvantaged as even the environment serves as a stepping stone for students with higher socioeconomic status.

In class we saw this concept of rich-gets-richer as pages that are already popular (as in have a higher distribution of in-links) continue to amass links despite the presence of new pages. This concept extends into higher education as those with parents who have pursued higher education are able to create a more familiar environment for their children as they grow up that they continue to reap the benefits of their status while pursuing higher education.

Cascading Behavior: How Jio brought 4G Cellular Data into Rural India

Cascading Behavior refers to the manner in which people tend to influence each other’s decisions when they are connected in a social network. Cascading behavior contributes to the diffusion”of a new technology or product within a network. As a new product or technology spreads from the few early adopters to others in the network, a cascade is created whereby people begin to adopt it by virtue of their network neighbors using it. Thus, cascading behavior is the foundation behind the viral marketing strategies employed by businesses. Viral marketing exploits cascading behavior by giving people the impression that everyone in their social sphere is adopting a new product, thus encouraging them to adopt it themselves.

Cascading behavior in the real world can be analyzed via two mathematical models:

• Threshold models: In this model a person adopts the technology if the weighted sum of its neighbours who have already adopted this behaviour is greater than the threshold.
• Independent cascade model:s In this model there is a probability that a person will decide to adopt a technology every time one of their neighbours decided to adopt the technology.

Reliance Jio Infocomm Ltd (Jio) is the largest telecommunications company in India. They offer LTE network and cellular services. In 2016, Jio revolutionized India’s digital markets by bringing cheap 4G data to the rural markets of India. Since its launch in 2016, Jio’s 4G subscriber base has grown to a whopping 398 million with a majority of its subscribers being located in rural areas. One could say that the cascading behaviour caused by Jio’s viral marketing is responsible for its success in India’s rural markets.

Jio’s initial marketing strategy was of pricing their services at a much lower price than their competitors in order to appeal to India’s rural markets. These low prices in combination with large-scale advertising campaigns pushed Jio off the ground in rural India. This rural market had previously been dormant because they had been sceptical of adopting cellular internet. Jio’s launch-off campaigns consisted of the biggest celebrities in India endorsing it in its widescale television advertisements. These celebrities have a massive fan-following in rural India and were instrumental in nudging Jio’s rural early adopters towards using the service.

Jio’s early adopters were primarily young adults who were enamoured by their celebrity stars and were eager to adopt technologies they endorsed. However once early adopters started using 4G cellular data for social media in rural areas, a cascading effect took over whereby the close-knit younger population in villages and small towns started getting influenced into buying Jio’s cellular data.

The cascading behaviour of young people in rural India can be analysed through the threshold model. India’s population is such that young people in small towns or villages can be considered as a close-knit cluster of nodes. When a node (or young person) observed that their friends were utilising social media and digital platforms more frequently because they had Jio’s 4G data subscription, they tended to also adopt Jio themselves because their peers in local schools and colleges had the cellular service and access to broadband wifi was limited. Thus, Jio capitalised on India’s young rural population wanting to access digital platforms.

Jio’s success with the middle-aged rural population can be analysed through the lens of the independent cascade model. Since joint families tend to be the norm in India, it was observed that younger individuals were helping their older relatives with the process of signing up for Jio. Thus, the probability of older individuals in rural areas adopting Jio could be closely linked to the probability to their younger neighbour nodes being familiar with Jio’s sign up policies and processes, since they signed up for the cellular data services themselves.

In this manner, the spread of Jio’s 4G services in rural India can be attributed to its successful viral marketing campaigns which effectively roped in rural India’s young population as Jio’s early adopters. Jio’s rapid spread throughout India’s rural population was due to cascading behaviour taking over and causing a diffusion of Jio’s technologies within the rural social networks of India.

Sources:

http://www.cs.cmu.edu/~jure/pub/diffusion-paper.pdf

https://iide.co/case-studies/reliance-jio-marketing-strategy/

Information Cascade and Share Market Volatility in Chinese Share Market

https://koreascience.kr/article/JAKO201616759692439.page

The research paper “Information Cascade and Share Market Volatility: A Chinese Perspective” discusses the large volatilities in Chinese Share market are because of information information blockage, which impedes share prices to timely respond to economic conditions. Investors have asymmetry information to the share market which leads to information cascade. An information cascade has been defined as behavior tendency to follow the lead of other traders when investors experience high information asymmetry. Imagine we have limited knowledge in crypto currency, and the crypto market is highly profitable. We would ignore our personal information about the market and just simply follow what everybody else is doing, hoping to get a high return. This action leads to tendency to follow predecessor’s actions and starts herding. However, such a process cannot sustain as investors will finally realize that the accuracy of the information aggregated from predecessors’ actions is open to question. Even a tiny shock could cause sudden change of popular behavior, causing market instability.

 

The research paper used figure 2 to discuss the process of information cascade. It could either lead to invest or not invest for the cascade. We can prove the information cascade by setting up the payoff for each investors and use Bayes’ rule. We are not going to prove in this blog post, but in class we discuss how to using Bayes’ rule to prove. From the graph, we can tell information cascade starts with the individual who finds out that the number of predecessors who invested (rejected) exceeds that of predecessors who rejected (invested) by roughly two. Any subsequent investors will ignore his or her own signal and invest based on predecessor. It is similar to our class discussion about Sequential Decision-Making and Cascades.

There were several information cascade on Chinese Share market historically. In 2005, the People’s Daily published articles in solid support showing that strong belief to change the Chinese share market fundamentally. The new account user climbed from 4.48 million to 15.36 million in one year. Both the astonishingly growing number of new opening accounts and unusual good performance of the market index indicate that investors may have entered into an invest cascade. However, when US announced economy might go to recession in 2007, information cascades are born quickly and idiosyncratically, and shatter easily. When the devastating impact of the US recession on China’s manufacturing industry became more evident to the public, the investors realised the inaccuracy of the information that they had aggregated from predecessors’ action and went into a panic. The market began to collapse in October 2007.

From the class discussion we know that cascade could be wrong, cascade could have little information, and cascades are fragile. From the Chinese share market we can tell the information cascade model that is discussed in the paper suggests that effective policies should address the detrimental issue by reducing information procurement costs so that the negative impact of the priority to access information that benefits would be greatly reduced. By doing so, share prices will quickly back to their long equilibrium level, avoiding excessive volatilities. However, clearing out information blockage would infringe the interests of part of government bureaucrats. That is also reason that recent government policies are not effective to change the current situation in its true sense.

Multifactor Second-Price Auction in Google Search

Article: https://moz.com/blog/understanding-google-ads-auction

 

A well-known system for displaying advertisements on a search engine is an auction. In simple terms, the interested advertisers place their bids with the search engine in terms of a cost per click and the search engine runs some form of an auction to determine which advertiser gets a certain position in a user’s search results. Oftentimes, the form of the auction that search engines use to sell ad positions is a second-price auction. The guiding principle behind this form of auction is that the dominant strategy is to bid the advertiser’s true value since the advertiser knows that they will pay their bid or less. A popular implementation of this auctioning system is the Google AdWords system, which advertises on Google Search.

 

In contrast to the simple example above, Google uses values named “Ad Rank” and “Quality Score” to determine an advertiser’s position in an auction. The provided formula reveals that Google uses the Ad Rank of the advertiser directly below to determine the cost paid by the advertiser, proving that the system used underneath is, in fact, a second-price auction. By definition, “Ad Rank” is equal to the advertiser’s bid assuming that all advertisers have a “Quality Score” of 10, making it a simple second-price auction by ignoring the “Quality Score” factor. However, since the matching of advertisements to a given ad slot is significantly more complex than that of a simple item to a buyer, the usage of an extra factor such as a “Quality Score” is required.

 

According to the article, the factors used to determine the “Quality Score” are the expected clickthrough rate, the landing page experience, and ad relevance among other minor factors such as time of day, location, device, and others. The expected clickthrough rate is a value generated using the advertiser’s average clickthrough rate i.e. a quantitative measurement of an advertiser’s historical success. This exists to incentivize the advertiser to create high-quality content that urges users to click on their ad, earning Google more money and giving the high-quality advertiser more publicity. The next factor is the landing page experience which is the webpage that the advertisement leads to. The higher the relevance and usefulness of the page that the ad links to (determined using a hidden algorithm), the more likely the ad would be considered high-quality in this context. The ad relevance factor is determined in a similar way but with the contents of the ad itself. The search engine then uses the “Ad Rank” value to determine eligibility according to a predetermined cutoff and rank the ads on a search page. 

 

Overall, Google’s pricing and ranking method uses not only the advertiser’s bid, but also considers the quality of the ad, the historical performance of the advertiser, and the relevance of the ad in the context of a given search query. Evidently, this is a proven method since it makes up a large portion of Google’s income and continues to be used to place ads within their Search product.

« go back — keep looking »