Living in London means that half of the year the weather is awful. As an academic, the conference circus provides some relief. This January I spent a week at the Society for the Advancement of Economic Theory in Santiago, Chile. Flying over the Andes mountains is a thing to behold already.
But Chile also boasts some very fine economists (apart from being one of the most advanced economies in South America). I will flag only two talks here.
The first talk that I particularly enjoyed was on frictional labor markets: Carrillo-Tudela, Clymo, Coles (2024) – Equilibrium Job Turnover and the Business Cycle.
Consider this: There are two reasons as to why a firm-worker match may end. Firms may lay off the worker. Or the worker may choose to pursue employment elsewhere. In the data, the layoff rate and the voluntary quit rate are nothing alike. Layoffs are anti-cyclical (rise in lockstep with the unemployment rate), whereas voluntary quits are pro-cyclical (they occur more frequently when the unemployment rate is low). Critically, if a worker voluntarily leaves a job, that job need not vanish. It may be refilled and as such be fundamentally different from a job that has not yet come into existence. However obvious to casual observers, recent estimates in the literature (including from this paper) then invite a break from tradition: job vacancies are not perfectly elastic as under the traditional free entry assumption. Rather, we should think of vacancies as a stock that only sluggishly adjusts to market changes. Why does this matter? If existing vacancies can be quickly refilled, there is an asymmetry in market tightness. Low unemployment rates need not hinder new job creation as new vacancies can be filled by poaching employed workers, setting off a virtuous hiring chain that down the line comes to benefit the unemployed. A low number of filled firm vacancies paired with unemployment, on the other hand, is likely to persist as the unemployed cannot in a similar way poach the employed workers’ jobs. Post scriptum. Looking back at my notes, I notice a detail of their model that is perhaps not sufficiently flagged in the paper: if the economy is down, i.e., passing through a temporary low productivity state, firms that are comparatively more productive have a stronger incentive to not lay off their workforce. In fact, to the extent that credit constraints prevent them from doing so, this may well explain why some countries view state-subsidized temporary reductions in working hours (‘Kurzarbeit’) as a key policy tool in times of crisis.
The second talk that I found intriguing was a very neat observation by Kun Zhang on Bergemann, Brooks, and Morris’ (2015) celebrated paper “The Limits of Price Discrimination.”
Consider the most basic problem in economics: a monopolist seller setting the price at which to sell. The potential buyer’s willingness-to-pay is some number v_k in \{v_1,...,v_K\}. If the seller was perfectly informed about this value, setting a price no less would maximize profit. Consumer surplus would be zero. In response, one may venture that the less the seller knows, the better off the buyer will be. But that is not true. If the seller knew nothing, the seller would still know the prior distribution, i.e., the probability F(v_k) that the potential buyer’s true willingness-to-pay is less or equal to v_k . For most distributions, by solving the standard problem \max\limits_{v_k} v_k(1-F(v_k)), the seller would then set a price exceeding the lowest possible valuation. If valuations are 1,2 or 3 and all equally likely, then charging price 2 maximizes the seller’s profit. Trade, despite being efficient would not take place in the event where the buyer’s valuation happens to be low. Could providing the seller with some information in such a contingency improve efficiency without lowering consumer surplus across the board?
Bergemann, Brooks, and Morris (2015) sweepingly affirm this question: there exists an information structure that holds the seller’s surplus down to the expected profit under the uniform monopoly price yet guarantees that efficient trade always takes place. Designing this information structure relies on what feels like a magical property: Every subset of valuations supports a distribution for which any price within its support is profit-maximizing. Here, since indifferent, the seller may just as well select the lowest price. An example (cf. (4) in Bergemann, Brooks, and Morris (2015)) may be more revealing: Consider the valuations 1,2 or 3 from above so that ½ of all buyers have a willingness-to-pay of 1, 1/6 a willingness-to-pay of 2, and 1/3 a willingness-to-pay of 3. This means, for instance, that at price 2 only half of all buyers—those whose willingness-to-pay exceeds 1—will buy. It is easy to check that in this case any price, 1,2 or 3, achieves an expected profit of 1. Similarly, if 1/3 of buyers have a valuation of 2 and 2/3 a valuation of 3, any price, 2 or 3, achieves an expected profit of 2. But what is truly magnificent is that any distribution for which some price v_k is profit-maximizing is the convex combination of distributions of the kind described above whose support includes v_k. In the initial example where the buyers’ willingness-to-pay is uniformly 1,2, or 3 one can imagine that a willingness-to-pay of 1 results in message A; and a willingness-to-pay 2 results in message B with probability ½. In every other case either message A or C so that C is twice as likely as A is reported to the seller. This may be hard to read. A Bayesian seller, however, will notice immediately that message B is fully revealing that the buyer’s willingness-to-pay is 2 whereas messages A and C instill beliefs about said willingness-to-pay that correspond to the distributions just described. Having thus associated message A with the profit-maximizing price 1 and messages B and C with price 2 then guarantees the same expected seller profit as before: 4/6. The key difference is that here the item is always sold. There is no ex-post inefficiency. And since the seller’s surplus remains unchanged, providing more information increases the buyer’s surplus.
On the face of it, Bergemann, Brooks, and Morris provide a proof of concept that consumer surplus may rise if the seller is better informed. Yet the intricacies of the design may be troubling to some observers. Information design requires a pre-commitment on the side of the buyer to reveal information in a probabilistic way. (Moreover, the buyer must convincingly convey to the seller how private information will subsequently be sliced, tranched, and re-packaged.)
In Kun’s talk, evocatively titled ‘From Design to Disclosure’, the buyer does not have the ability to commit to an information design. The buyer can however, in light of his valuation, report any set of valuations as long as his true valuation belongs to this set. In the absence of commitment, the above design can then impossibly emerge in equilibrium: whenever the buyer’s true valuation is 2 or 3, the buyer could, by sending message A, induce the seller to lower its price from 2 to 1. Does this mean that in the disclosure paradigm, we can revert to our initial guess that more seller information unambiguously hurts the buyer? Not necessarily. But let us examine the bad news (from the buyer’s perspective) first: Full disclosure is clearly an equilibrium. This is driven by the observation that the buyer will always insist on a price of 1 if his willingness-to-pay is 1 and accept no price greater than 2 if his willingness-to-pay is 2. Then upon receiving the unexpected message that the true willingness-to-pay is either 1 or 2, the seller can interpret this message as almost surely coming from a buyer whose willingness-to-pay is 2. In effect, the buyer is condemned to revealing nothing but the whole truth. Kun’s observation is that as the set of possible buyer valuations increases, more sophisticated message structures exist that are only partially revealing and thereby allow the buyer to extract the entire surplus. Instilling such a belief is not an easy feat. As in Bergemann, Brooks, and Morris (2015), it requires a fine and somewhat incredible partitioning of valuations. In addition, the information design must be pure with no randomization over messages taking place. The willingness-to-pay fully determines the message. To illustrate, consider a buyer that sends message A if his willingness-to-pay is extremely low, somewhat low, somewhat high, or extremely high whereas that buyer sends message B if his willingness to pay is very low and very high. The seller, of course, is sophisticated enough to see through all of that and update his belief accordingly. Crucially, if the chosen message sets replicate the magical distributions described earlier, the seller will be indifferent between all prices in the distribution’s support. So he may just as well offer prices on the lower end. A bit fantastic? Sure. But I view it as an important proof of concept, showing that it does not require the buyer to hold commitment power to appropriate significant (possibly all) gains from trade.
To grasp all the details, you must wait a little longer. Kun has not yet posted (pardon I mean disclosed) the paper. He has posted another paper on disclosure, however, titled “Withholding Verifiable Information” which may be a good starting point for this literature.. (For even more research on disclosure, you may wish to check out my future colleague Paula Onuchic’s work on “Disclosure in Teams.”)
SAET boasted many more insightful sessions. I will stick to the anecdotal. One taxi driver explained to me that Chileans curse too much and speak poor Spanish because the country started as a convict colony. Another taxi driver explained to me that Chileans, unlike Brazilians, speak English like the Americans. Make of that what you will. There is the off-chance that you may want to know that there exist two Avenidas Cardenal José María Caro. Only one is located in the city center. Oh well. And did I mention the breathtaking views on my return flight to Rio de Janeiro, Brazil? Incidentally, that’s where SAET is headed to next. See for yourself!
Nice read (and cool pictures), many thanks for sharing your conference impressions!