Periodic Reporting for period 1 - IMEDMC (Information and Misinformation Economics: Design, Manipulations and Coutermeasures)
Reporting period: 2021-05-01 to 2022-10-31
The IMEDMC project aims to understand the impact and optimal design of information production when accounting for how it shapes downstream actions of those at the receiving end of the process, but also upstream actions of agents whose interests are at stake, including effort, investment or manipulations. For example, creating a green label for businesses generating few emissions is informative for, and may affect consumer behavior. But it also provides incentives for businesses to invest in emission reduction technologies, and to possibly engage in green washing to appear cleaner than they really are. With this general objective in mind, using the analytical frameworks of information design, mechanism design and game theory, the project is divided in five broad areas.
The first area studies the strategic production of fake news by malevolent agents seeking to influence the beliefs or actions of a heterogeneous population. We write a model where the fake news producer faces both naïve and sophisticated receivers that react differently to fake news: while naïve receivers believe any news, sophisticated receivers are aware of the prevalence of false information and discount any news accordingly. Receivers are interconnected in a network and take two decisions based on their information: which position to adopt on a binary issue, and whether they wish to transmit the information they have received to their connections. This model allows us to study the strategic production of fake news, their diffusion over social media, and can be used to test the effect of different policies and their welfare consequences.
The second area studies the design of tests, or information structure, when the tested agent can manipulate the test. A typical example is car manufacturers manipulating the results of emissions tests. We ask how tests can be optimally designed when accounting for such manipulations. Our model can also be used to study allocation mechanisms without transfers when manipulations are possible. This applies to the allocation of social housing, promotions or green labels based on characteristics of agents that they can manipulate. Our (now published) work shows that optimal tests (or allocation mechanisms) rely on inducing manipulations by the agents. In an extension of this work, we study optimal tests or mechanisms under the constraint of not inducing manipulations. We discuss why such a constraint might be desirable.
The third area studies delegated information design, where the principal seeks to incentivize an agent to produce information according to her interests, but can only contract on the actions taken by receivers on the basis of this information.
The fourth area studies the design of tests (or allocation mechanisms) when agents can improve their type beforehand. For example, businesses make costly investment to reduce emissions in order to pass emission tests. The role of testing in fostering such investments is of course quite prominent among the goals of such tests. However, the question of how to design them optimally has not been fully studied. Intuitively, randomizing the outcome of the test may seem useful to spread effort across agents with different levels of the tested characteristic. However, we prove that, in a basic and focal model of the problem, simple pass-fail tests at a threshold level remain optimal when agents can improve their characteristic.
Finally, a fifth area has emerged from the work of members of the team. It uses information design to better understand the impact of consumer data on markets and provides insights on what regulation can achieve. One way in which businesses can use consumer data is as a tool to segment and price discriminate consumers. The range of outcomes that can be attained has been studied in the literature. In recent and ongoing work, our team studies two variations of this problem. The first one introduces redistributive concerns to the normative perspective on different segmentations. Assuming poor consumers have a lower willingness to pay, this new work shows that redistributive concerns never lead to departures from efficiency but may imply leaving additional rents to the monopolistic seller. The second one studies how accounting for consumer entry might lead platforms controlling the access of sellers to consumer data to opt for segmentations that are less prone to price discrimination.
The first area studies the strategic production of fake news by malevolent agents seeking to influence the beliefs or actions of a heterogeneous population. We write a model where the fake news producer faces both naïve and sophisticated receivers that react differently to fake news: while naïve receivers believe any news, sophisticated receivers are aware of the prevalence of false information and discount any news accordingly. Receivers are interconnected in a network and take two decisions based on their information: which position to adopt on a binary issue, and whether they wish to transmit the information they have received to their connections. This model allows us to study the strategic production of fake news, their diffusion over social media, and can be used to test the effect of different policies and their welfare consequences.
The second area studies the design of tests, or information structure, when the tested agent can manipulate the test. A typical example is car manufacturers manipulating the results of emissions tests. We ask how tests can be optimally designed when accounting for such manipulations. Our model can also be used to study allocation mechanisms without transfers when manipulations are possible. This applies to the allocation of social housing, promotions or green labels based on characteristics of agents that they can manipulate. Our (now published) work shows that optimal tests (or allocation mechanisms) rely on inducing manipulations by the agents. In an extension of this work, we study optimal tests or mechanisms under the constraint of not inducing manipulations. We discuss why such a constraint might be desirable.
The third area studies delegated information design, where the principal seeks to incentivize an agent to produce information according to her interests, but can only contract on the actions taken by receivers on the basis of this information.
The fourth area studies the design of tests (or allocation mechanisms) when agents can improve their type beforehand. For example, businesses make costly investment to reduce emissions in order to pass emission tests. The role of testing in fostering such investments is of course quite prominent among the goals of such tests. However, the question of how to design them optimally has not been fully studied. Intuitively, randomizing the outcome of the test may seem useful to spread effort across agents with different levels of the tested characteristic. However, we prove that, in a basic and focal model of the problem, simple pass-fail tests at a threshold level remain optimal when agents can improve their characteristic.
Finally, a fifth area has emerged from the work of members of the team. It uses information design to better understand the impact of consumer data on markets and provides insights on what regulation can achieve. One way in which businesses can use consumer data is as a tool to segment and price discriminate consumers. The range of outcomes that can be attained has been studied in the literature. In recent and ongoing work, our team studies two variations of this problem. The first one introduces redistributive concerns to the normative perspective on different segmentations. Assuming poor consumers have a lower willingness to pay, this new work shows that redistributive concerns never lead to departures from efficiency but may imply leaving additional rents to the monopolistic seller. The second one studies how accounting for consumer entry might lead platforms controlling the access of sellers to consumer data to opt for segmentations that are less prone to price discrimination.
Area 1
1. Hiring of a research assistant to conduct a review of the literature on the subject.
2. We have the model and most theoretical results. We plan to write a first draft of the paper before the summer.
Area 2
1. Publication of the first paper at Econometrica with V. Skreta. Task 1 of the proposal for this area is now fully incorporated into this paper.
2. Presentation of the first paper in many venues (see list, some additional ones before the starting of the grant)
3. Work with my co-author on a new paper with a falsification-proofness constraint. We also reinterpret the problem as an allocation problem. We are finishing the first draft (available soon). This work has already been presented in many venues, and more are scheduled.
Area 3
1. We have already produced two working papers on this area, which are different interpretations of the same problem.
2. We are writing a third working paper in which the principal can choose the information structure to be delegated but the agent (a communication agency, for example) exerts efforts to diffuse the communication campaign across the population. Preliminary results have been presented at a workshop.
Area 4
1. With Ph.D. student V. Augias, we wrote a new paper in which we show that, even when agents can invest in their type, optimal tests or allocation mechanisms have a pass-fail structure. This is a stark result for a simple yet unexplored and difficult to answer question. We are in the process of streamlining the draft and prepare it for submission. This contributes to task 1 of this area.
Area 5
1. New working paper by V. Augias, D. Barreto and A. Ghersengorin on consumer optimal segmentations with redistributive concerns.
2. New research on profit maximizing segmentations by platforms, together with A. Salas.
1. Hiring of a research assistant to conduct a review of the literature on the subject.
2. We have the model and most theoretical results. We plan to write a first draft of the paper before the summer.
Area 2
1. Publication of the first paper at Econometrica with V. Skreta. Task 1 of the proposal for this area is now fully incorporated into this paper.
2. Presentation of the first paper in many venues (see list, some additional ones before the starting of the grant)
3. Work with my co-author on a new paper with a falsification-proofness constraint. We also reinterpret the problem as an allocation problem. We are finishing the first draft (available soon). This work has already been presented in many venues, and more are scheduled.
Area 3
1. We have already produced two working papers on this area, which are different interpretations of the same problem.
2. We are writing a third working paper in which the principal can choose the information structure to be delegated but the agent (a communication agency, for example) exerts efforts to diffuse the communication campaign across the population. Preliminary results have been presented at a workshop.
Area 4
1. With Ph.D. student V. Augias, we wrote a new paper in which we show that, even when agents can invest in their type, optimal tests or allocation mechanisms have a pass-fail structure. This is a stark result for a simple yet unexplored and difficult to answer question. We are in the process of streamlining the draft and prepare it for submission. This contributes to task 1 of this area.
Area 5
1. New working paper by V. Augias, D. Barreto and A. Ghersengorin on consumer optimal segmentations with redistributive concerns.
2. New research on profit maximizing segmentations by platforms, together with A. Salas.
Area 1: First working paper in progress addressing tasks 1 and 2. Tasks 3 and 4 will be addressed separately and still need launching.
Area 2: Task 1 is addressed in our now published paper. We have launched a new ambitious project for which a first draft is close to completion. Tasks 2 and 3 as planned are no longer priorities as we think our new work is more groundbreaking.
Area 3: Tasks 1 and 2 are addressed in the two working papers already written and the one in progress. We are working on related topics, thinking about how a principal can incentivize an agent to distort a given information structure, but that is preliminary.
Area 4: Task 1 is addressed in my work with V. Augias. A first draft is available. There are possible developments that go beyond the initial scope of the project. We want to think about questions of discrimination when designing selection procedures with inequalities in the cost of investment. Task 2 is still on the agenda as it is important to understand the design of selection procedures when agents can both invest and manipulate. Tasks 3 and 4 are still on the agenda but need launching.
Area 5: Was not part of the original project. The goal is to lead the two new projects to completion.
Area 2: Task 1 is addressed in our now published paper. We have launched a new ambitious project for which a first draft is close to completion. Tasks 2 and 3 as planned are no longer priorities as we think our new work is more groundbreaking.
Area 3: Tasks 1 and 2 are addressed in the two working papers already written and the one in progress. We are working on related topics, thinking about how a principal can incentivize an agent to distort a given information structure, but that is preliminary.
Area 4: Task 1 is addressed in my work with V. Augias. A first draft is available. There are possible developments that go beyond the initial scope of the project. We want to think about questions of discrimination when designing selection procedures with inequalities in the cost of investment. Task 2 is still on the agenda as it is important to understand the design of selection procedures when agents can both invest and manipulate. Tasks 3 and 4 are still on the agenda but need launching.
Area 5: Was not part of the original project. The goal is to lead the two new projects to completion.