Talk Notes

Here is where I keep notes about the talks that I have attended.

Milena Almagor May 26

Apr 14: Michela Tincani - Preferential Admissions under Biased Beliefs

• Human capital - disadventaged students have worse schools so trouble getting admitted.
• Colleges moving to programs which garuntee admittance for top x% of students from each school.
• Students from disadvantaged backgrounds have incorrect beliefs about their admission chances.

Mar 2nd: Kris Iyer - Persuasion with risk-conscious agents: A geometric approach

• Online, the payoff to a user depends on the state of the network. Congestion, other users, remainging itesm, etc.
• How should a platform share information with users to persuade them to "better" outcomes? Operations Research
• Uncertain wait times are percieved to be longer. Customers are risk averase over wait time.
• Straightforward expected utility doesn't capture the fact that consumers tend to dislike variance.
• My Thought: asymetric time costs due to delays necessitating schedule restructuring??
• Bayesian Persuasion is a hot field. Platform has commitment power.
• Basic Model
• State of the world \(\omega\in\Omega\). e.g. waiting times
• Reciever chooses an action \(a\in A\). Like choosing ot wait.
• Sender commits to a signaling scheme \(S,\sigma\), where S is the set of possible signals. \sigma is a mapping from state and signal \(\sigma(\omega,s)=P(sendsignal \; s | state=\omega \)
• Reciever utility for an action under belief \(\mu\) is given by \(\rho(\mu,a)\). EX: mean variance utility, conditional value at risk, risk measures, cumulative prospect theory.
• With expected utility maximizers, utility is linear in belief. With risk-concious agents, utility need not be linear in belief.
• Lookup later: Equilibirum topology: weak star. Alea pardox? Ellesburg Paradox?
• The reciever upon getting signal s, chooses the action to maximize \(\rho(\mu_s, a)\\)
• Sender's goal is to select a signalling machanism \(s,\sigma\) to maximize expect sender's expected utility.
• Linear Programming formulation fails in this setting.
• Coalescing signals will not change optimal actions if spaces which induce each optimal action are convex. Happens with expected utiltiy.
• Sender can control action directly. Given obeidience constraint
• Relax assumption of expected utility. Then coalescing can change actions. Revalation principle fails. action recomendations may not suffice.
• Outside idea: Apply this to patrick's farming thing. Non-expected utility, I mean.
• Set up a persusion scheme so the reciever will always choose something other than their prior. (myexample:choosecareer) cant do that with expected utiilty
• Lemma: if two signals induce the same beliefm the coalescing those signals induces that belief. CAn use belief recomendatons. Need that mean mean posterior is the prior. Baye's plasubility.
• Instead of optimizing over the entire space of ditributions of beliefs. Instead we can compute over \(m_a\)s where each \(m_a\) is the mean posterior mean over the cluster of bbelifes which induce action \(a\)
• Sender's optial signaling mechanism can be found by solving a convex optimization problem.
• My thought: what if sender might not be credible.???
• At most (\Omega\) actions per signal. convex analysis.
• Binary Persuasion:
• Two actions (1 and 0), Sender prefers 1 over 0.
• Assume that ht eset of beliefs which induces action 0 is negative.
• If the prior is in the convex hull of \(P_1\) and in \(P_0\), then the reciever can be fully persuaded.
• Robust Persuasion: Sender Minimaxes against rciever with private type
• According to Lingebrunk an dIyer 2017, with expected utility customers, the optimal signaliing mechanism sends "join" or "leave" signals, and has a threshold structure.

Feb 14: Firm Creation and Local Growth, Conor Walsh

• New firms create new ideas, they creatively destroy old stuff, and they provide idea diffusion. Lots of important stuff.
• This paper focus on new varieties, and how expanding on the variety margin contributes to local growth.
• Facts: Startuups account for most local employment growth, exit rates unrelated to employment growth, incumber firms mostly invariant. So local growth is mostly about high entry rates.
• Census LBD Longitudianl Business Database, sales data from IRS records
• Growth, he argues, is mostly about extensive entry margin.
• Creative destruction would have high entry be associated with high exit, but don't see that.
• Idea diffusion would have density related to firm growth.
• Model has Workers who can move, Capitalists fixed in place, and Landlords
• Model has entry modelled as combining land and finalgood in the construction sector
• Rust belt story: Firms die all the time. The problem is simply that new firms aren't being made. That's what's causing these towns to die.

2020 Jan 29 Micro Workshop

Talk from that Indiana person

• Married women have 40-50% lower volatility in hours worked than other demographics.
• Compared to 1960, married women work 40% more, and hours worked went from 16->20% women.

So if married women have less labor volatility and there are more married women working, will this lead to less overall volatility in women working? It depends.

• If married women are joining the labor force because their behavior is becoming more similar to a man's, then no.
• If married women wokring more because of less wage gap, then sure.
• Some other third thing I can't recall?

Also interesting note about how married women often exit the labor force during good times, but don't exit the labor force during bad times. This offsets the married women laid off.

Patrick's farm thing.

Poor countries tend to have less efficient agriculture sectors, but food is cheaper in these countries. They also seem to have a comparative advantage in manufacturing. But they don't import food. There aren't super high trade costs for grain and stuff, so why don't these poor countres import their food?

Maybe they can't use capital so wages go way down somehow? Maybe they eat the low-quality food. Or export cash crops and import staples. Or uhh...