Vlad Nora
Support the fight of the Ukrainian people for freedom
Welcome!
I am Assistant Professor of Economics in Nazarbayev University in Astana. I do micro theory, industrial organization, and applied game theory.
Here you can find some of my papers with brief explanations. These are neither abstracts, nor rigorous descriptions of papers (I hope to add the remaining work soon).
You can reach me at vladyslav.nora@nu.edu.kz.
Research
We are swayed by others: friends, colleagues, lifestyle bloggers, and political pundits. Not surprisingly, some may wonder how to exploit social influence to promote their own agenda. In a recent Netflix documentary about social media, "The Social Dilemma", writer Jaron Lanier says “We've created an entire global generation of people who were raised within a context where the very meaning of communication, the very meaning of culture is manipulation". So, here we take a stab at explaining the mechanics of such manipulation, and uncover social networks that are the easiest to manipulate.
Imagine that a social media platform wants to convince Alice, Bob, Charlie, and Dave to buy a particular product or join a political movement. Of course, Facebook or Instagram might not be interested in such behaviours per se. However, their business model is built upon selling the possibility of such influence to the interested parties. We shall model convincing by saying that a platform promises individuals a reward if they act as it desires. Individuals are prone to social influence - they are more inclined to act if more of their friends do.
First, we ask how can a platform convince everyone using as little resources as possible? The answer is intuitive. Generally, it must promise higher rewards to popular individuals, and rely on them to influence the others, who are promised lower rewards. For example, in the star graph on the left, it must spend more resources convincing Alice, and can spend less on others because they are influenced by Alice.
Having established the simple mechanics of social influence, we then ask which networks are the most susceptible to manipulation. The answer depends on the nature of social influence. One example is when people care about a fraction of their friends who act. Then a star network requires the least resources to convince everyone to act. Another example is when people care about the absolute number of friends who act. Then convincing individuals in a complete network is the easiest.
Generally, we model social influence from each friend as a decreasing function of the total number of friends, hence capturing the idea that people with many connections are swayed less by each one of them. Then, under some natural assumptions, we show that the easiest networks to manipulate are core peripheries. A core-periphery network partitions nodes into stars and periphery, with every star being linked to all nodes, and every periphery node being linked only to stars. The figure on the left lists all such networks with four nodes, a star, and a complete network being the special cases.
The main takeaway is that social networks which are very unequal in terms of individual connections are also the most vulnerable to manipulation. This is, perhaps, the message that internet regulators should keep in mind.
"Dynamic Assignment Without Money: Optimality of Spot Mechanisms" with Julien Combe and Olivier Tercieux, R&R in Theoretical Economics, 2023.
We often want to allocate goods to people without making them pay real money. The reason is that we are repulsed by the idea of trading things like kidneys or places in public schools. How much money one has, should not decide whether she can get a new kidney, right? So, we invent artificial markets with “virtual” money where with a bit of luck everyone stands a chance. This paper is about the design of such markets for the dynamic allocation of goods.
Imagine a large crowd of people who want to party in Berlin over a weekend. Each one has two nights to spend visiting one of the famous techno clubs. Let’s say there are two clubs, B and K. Everyone has some preferences over where they want to go each day. There are two days and two clubs, so four possible alternatives and let's say each clubber has a ranking over these four alternatives. For example, I might rank first night K and second night B as the top alternative (KB), second best might be B and then K (BK), etc. So, the good here is the sequence - where you spend the first and where the second night. How should we design an artificial-money market to allocate the rights to enter the clubs each day?
There are multiple ways to do it, but generally, they work like this: we give clubbers some budgets of virtual money (the budgets might be random), set prices on club visits, and let them choose how to spend their budgets. For example, one can give each clubber two separate budgets, one for every night. Alternatively, we can give a single budget transferable across nights. In addition to budgets, another design dimension is the prices. Should there be a separate price for each club every night? Or should there be a price for each two-night sequence?
Each design choice entails certain efficiency properties of the resulting allocations. Our main result is that the only efficient way to design this market is to have a single transferable budget, and separately price each club every night so that the price of the two-night sequence is simply the sum of two prices. This result has implications for many real-world markets and provides a justification for a simple and natural market design.
"Inefficient Screening in Online Rental Markets" with Irina Kirysheva, Journal of Industrial Economics, 70, 2022. [Working paper]
“Sofas were outside. The TV was in my bathroom. Objects … photos, memorabilia … Pictures of my nieces that are on my fridge that were on the floor, trampled on and torn, with shoe prints all over their cute, little faces”, complains one unlucky Airbnb host after guests trashed her house. Inconsiderate guests are a nightmare of anyone who rents out a flat. Although Airbnb found a simple solution. Hosts are free to reject someone whom they are not comfortable with based on public profiles and past reviews of users. In other words, they “screen” potential guests.
Screening is common. Mortgage lenders, credit card issuers, health insurers all screen their customers, and firms routinely conduct tests and interviews prior to making job offers. Such screening is intended to improve efficiency, for instance by allowing lenders to restrict credit to subprime borrowers or Airbnb hosts to refuse inconsiderate guests. But does it really work?
Our finding is surprising: screening is never fully efficient and sometimes can be harmful. Imagine consumers differ with respect to the cost they impose on a firm who serves them, and firms commit to prices before observing their customers. We observe that when firms compete in prices, there is typically no pure equilibria and sometimes all firms set prices at such a low level, that they must reject even some efficient trades. We show that when there are relatively few firms, all market equilibria are inefficient. Moreover, sometimes these welfare losses due to excessive rejections outweigh the benefits from filtering out subprime customers. Hence introducing screening can reduce overall welfare.
"Does Vertical Integration Enhance Non-price Efficiency? Evidence from the Movie Theater Industry" with In Kyung Kim, Rev Econ Design 24, 143–170 (2020). [Working paper]
Kim Ki-duk once complained: "Countless films won’t get a chance to screen due to politics at multiplex theater chains". If you did not hear about Kim, this beautiful movie can be a great intro. So, what does the famous filmmaker mean?
A new movie is distributed to theaters by a distribution company (20th Century Fox, Universal etc) that also does marketing and advertisement. Distributers and theaters agree how to share the box office (usually 50/50, movie demand is uncertain so revenue sharing allows a risk sharing). But it might be a good idea for a distributor to buy a theater: You get to keep all the revenues from own movies, and make money on showing movies of other distributors. Here is a controversial part. You can show the rival movies less often in your theaters (aka foreclose). This is a problem if many multiplex chains are integrated with one of the major distributors. Independent movies get squeezed away from screens and consumers suffer. That is what Kim Ki-duk is saying. In the paper with In Kyung we explore these concerns.
Our findings are surprising. Although, we confirm that integrated theaters favor own movies when allocating seats, there is something else. It appears that despite this, integrated theaters might be doing a better job in allocating seats than independent ones (those that are not owned by distributors). Here is a simple example of how it may happen. Imagine that 36 people want to see movie A, 34 - movie B, and 10 - movie C, and the theater has 100 seats. But, of course, the exact demand is not known in advance. Then something like the following happens. If a theater is integrated with movie A, it allocates seats 50-35-15 (50 to movie A, 30 to movie B, 20 to movie C, assume seats can be allocated in any way). If it is independent it allocates 34-33-33. So, integrated one turns away no one, whereas independent turns away 2 people from movie 1, and 1 person from movie 2. The two panels show a percentage of turned away consumers for theaters with 8 screens, under the assumption that the aggregate demand reaches 70 percent of all seats in a theater (of course we never observe real theater level demand and therefore have to use indirect estimates). We see that on average integrated theaters turn away roughly 30 percent fewer consumers: they choose a seat allocation that matches the demand better.
It appears that independent theaters might just suck in forecasting demand! Indeed, for integrated theaters forecasting demand is more important because they have more to loose when misallocating the seats.
This time Kim Ki-duk is not telling us the full story.
"Saddle Functions and Robust Sets of Equilibria" with Hiroshi Uno, Journal of Economic Theory 150, 866-877 (2014).
Many of my favorite applications of game theory touch upon the surprising effects of changes in players' information. Electronic Mail Game of Rubinstein is a fun example (the idea alludes to Two Generals' Problem formulated in CS literature in 70s). The observation is that "tiny" changes in the information structure of a game can dramatically change equilibrium behavior. It appears that it's hard to guarantee that an equilibrium is "robust" to such changes. The paper with Hiroshi explores the notion of "robustness to incomplete information" in games which we call (in a bluntly non-sexy way) games with Saddle Function. It brings together the literature on robustness, team-maximin equilibria, and potential games. Below is a simple example.
Alice and Bob are in love. To send a secret message to Bob, Alice chooses one of the communication channels, and to receive the message Bob has to listen to the right channel. Unfortunately, lovers did not agree in advance which channel to use. To make things worse, jealous Eve is determined to intercept the message, and chooses a channel to listen to. Alice and Bob maximize the probability of secretly transmitting the message, whereas Eve minimizes this probability.
Of course, if in equilibrium Alice and Bob always choose the same channel, they cannot keep their communication private. However, there is a pair of mixed strategies, one for each, which guarantees them the highest probability of secret communication. Moreover, these strategies are part of some equilibrium. Is this "best" equilibrium robust to small changes in the information that players have? Our main result based on the techniques from potential games implies, in particular, that whenever the game has almost complete information, Alice and Bob can obtain payoffs close to the ones in the best equilibrium.
The sketch is from the series by Me Kyeoung Lee, depicting traditional convenience stores throughout Korea. Me Kyeoung says: "Many of these shops are now closing down and it makes me sad. I have a lot of fond memories of visiting them with my mother and grandmother when I was younger. To me, they are warm and show a lot of love."
Surely, Korea is not the only country that succumbed to the efficiency of modern retail chains. Walmart, Target, Carrefour, Aldi, Lidl etc. are behind every corner in the developed countries. Many producers rely heavily on these chains for sales, and retailers use it by squeezing producers’ margins.
Using Bayesian persuasion framework, we ask how increasing buyer power affects one aspect of producer-retailer relationships – sharing of product information. Often, a producer knows more about its product and can help to set the right retail price. However, transparency when dealing with powerful retailers might be a bad idea. Subsequent aggressive bargaining threatens to leave a producer with next to nothing. So, in contrast to when mainly selling through mom-and-pop shops, producers should be much more reluctant to share information when selling through large chains. We show that in equilibrium producers fully disclose their demand information only when the market share of powerful retailers is small.
We empirically explore this idea using non-binding retail price recommendations (RPRs) as a proxy for information sharing, and a fraction of a product’s sales through chain retailers as a measure of buyer power in a product’s market. The interesting fact about RPRs is that you can find them on some products, but not on others. So, we went around grocery stores in Korea and recorded which products had RPRs, and which did not. It appears, that when sales of a product rely heavily on chain retailers, this product is less likely to have a recommended price. Seems like buyer power may discourage use of RPRs, and hence, potentially, hinder information sharing. Indeed, along with the steady demise of mom-and-pops shops, RPRs are becoming an oddity these days.
Alice, Bob, and Charlie work in the econ department. They develop original research projects, and join projects of colleagues. Each one decides how many projects to develop, and randomly joins projects of each colleague. Economists’ enjoy having more projects, but developing projects is costly. Although random, joining existing projects is free. So, projects are local public goods. One day the department hires a research assistant who makes developing new projects cheaper. Now a department head faces a problem: whom to allocate the RA in order to increase welfare?
We focus on a network aspect of the problem: how does the structure of spillovers shape efficient allocation? For instance, suppose Alice creates the most spillovers: both Bob and Charlie are likely to join her projects and vice versa. However, Bob and Charlie do not work together that often. Should Alice get the RA?
Here is a crux tradeoff: Alice creates the highest spillovers, but also free-rides the most, developing fewer original projects. Our result, based on a first-order approximation to equilibrium, characterizes efficient allocation in terms of the degree of relative risk aversion and Bonacich centralities of agents in a spillover network. In particular, it implies that Alice should get the RA only if our economists are not very risk averse. Otherwise Bob or Charlie get the RA.
Curriculum Vitae
Assistant Professor of Economics in Nazarbayev University since 2014.
Ph.D. in Economics, 2010-2014 from Center for Operations Research and Econometrics (CORE), UCL.
M.A. in Economics, 2007-2009 from Kyiv School of Economics.
You can download my CV here.
Photo by Andrey Korotich. Check out his work at andreykorotich.com.
