class: center, middle, inverse, title-slide .title[ # Lecture 03 ] .subtitle[ ## Coasean Bargaining ] .author[ ### Ivan Rudik ] .date[ ### AEM 4510 ] --- exclude: true ``` r if (!require("pacman")) install.packages("pacman") ``` ``` ## Loading required package: pacman ``` ``` r pacman::p_load( tidyverse, xaringanExtra, rlang, patchwork, tweetrmd ) ``` ``` ## Installing package into '/opt/homebrew/lib/R/4.6/site-library' ## (as 'lib' is unspecified) ``` ``` ## Warning: package 'tweetrmd' is not available for this version of R ## ## A version of this package for your version of R might be available elsewhere, ## see the ideas at ## https://cran.r-project.org/doc/manuals/r-patched/R-admin.html#Installing-packages ``` ``` ## Warning: 'BiocManager' not available. Could not check Bioconductor. ## ## Please use `install.packages('BiocManager')` and then retry. ``` ``` ## Warning in p_install(package, character.only = TRUE, ...): ``` ``` ## Warning in library(package, lib.loc = lib.loc, character.only = TRUE, ## logical.return = TRUE, : there is no package called 'tweetrmd' ``` ``` ## Warning in pacman::p_load(tidyverse, xaringanExtra, rlang, patchwork, tweetrmd): Failed to install/load: ## tweetrmd ``` ``` r source("R/video_helpers.R") options(htmltools.dir.version = FALSE) knitr::opts_hooks$set(fig.callout = function(options) { if (options$fig.callout) { options$echo <- FALSE } knitr::opts_chunk$set( cache = TRUE, cache.extra = list(tools::md5sum("R/video_helpers.R")), echo = TRUE, fig.align = "center" ) options }) ``` ``` ## Warning in xaringanExtra::style_panelset(panel_tab_color_active = "red"): 'xaringanExtra::style_panelset' is deprecated. ## Use 'style_panelset_tabs' instead. ## See help("Deprecated") ``` ``` ## Warning in style_panelset_tabs(...): The argument names of `style_panelset()` ## changed in xaringanExtra 0.1.0. Please refer to the documentation to update to ## the latest names. ``` ``` ## NULL ``` --- # Roadmap 1. Can we achieve the efficient outcome .hi[without] government intervention? 2. What does the Coase theorem say? 3. What are the limits to Coasean bargaining? 4. Can data centers and their neighbors bargain their way out of the AI boom's externalities? --- # This lecture's running example: the AI boom next door <video width="900" controls preload="auto" style="display:block;margin:auto;"><source src="videos/03-coase-data-centers-utility-bills.mp4" type="video/mp4"></video> --- # Data centers are a textbook externality problem Data centers bring jobs and tax revenue, and their neighbors bear real costs: -- - Cooling fans and diesel backup generators hum around the clock - Generator testing burns diesel next to homes - Water use competes with local supplies - Electricity demand pushes up everyone's bills -- The parties are easy to name: one operator, identifiable neighbors -- .hi[Today's question:] when can the operator and the neighbors bargain to the efficient outcome, and when do transactions costs get in the way? --- class: inverse, center, middle name: coase # Coase Theorem <html><div style='float:left'></div><hr color='#EB811B' size=1px width=796px></html> --- # Pigou vs Coase We have argued that efficient allocations are not achieved in the presence of externalities -- Why? -- There are no markets through which the source of the externality must pay/be compensated for its effect on society -- i.e. they're not priced -- This means there's a role for government to create this market or price the externality --- # Ronald Coase (1910-2013) .pull-left[ <center> <img src="files/03-coase.png" alt="" width="100%" /> </center> ] .pull-right[ In a famous paper (”The Problem of Social Cost”), 1991 Nobel prize winner Ronald Coase made people rethink this Do we actually .hi[NEED] government intervention? ] --- # The Coase Theorem If there are: 1. No wealth effects on demand 2. No transactions costs 3. Well-defined and enforceable property rights then: -- the socially efficient or optimal economic activity will occur regardless of who holds the property rights -- The right to pollute (a resource) will end up in the hands who value it most through negotiation --- # Property rights <video width="900" controls preload="auto" style="display:block;margin:auto;"><source src="videos/03-coase-property-rights.mp4" type="video/mp4"></video> Property rights are only as good as prevailing norms or enforcement --- # Coasean arguments Coase versus Pigou: externalities are reciprocal in nature -- A power plant produces emissions that nearby residents breathe and those people incur the external costs -- By breathing the air, the nearby residents help create the externality (i.e. if they weren’t there, there would be no external cost from emitting pollution) --- # Coasean arguments What is more valuable to society? -- The externality-producing good generated by the power plant or letting people live nearby? -- To make the reciprocity concrete, consider a doctor working next door to a data center whose equipment creates noise --- # The Doctor and the Data Center More noise = more computing services and less medical services -- Less noise = fewer computing services and more medical services -- Which is better from a social point of view depends upon the relative values of computing and medical services -- Is the net benefit to society better at no noise, 0, or the level of noise that maximizes the data center's profit, N<sub>0</sub>? --- # The Doctor and the Data Center .pull-left[  ] .pull-right[ MC is the marginal cost imposed on the doctor by noise MB is the marginal benefit to the data-center operator from running the equipment that creates noise ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ It is important to establish that someone has the property rights Otherwise, trade will not happen Give the right to create noise to the data-center operator Initial outcome will be N=N<sub>0</sub> What happens next? ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ The doctor can pay the data center to reduce operation of its noisy equipment Why? Because MC to the doctor is higher than MB to the data center for noise at N<sub>0</sub> ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ The doctor is willing to pay more (MC) than the data center is willing to accept (MB) until noise is reduced to N<sup>*</sup> This is where total benefit is maximized (blue area) ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ The doctor and data center can split the .hi-red[bargaining surplus], the red area This is just the avoided deadweight loss from the noise externality ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ Instead of assigning the right to create noise to the data center, we could give the doctor the right to quiet In this case what happens? ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ First, we start at N<sub>d</sub> now since the doctor does not like noise The data center pays the doctor to be allowed to make noise The data center is willing to pay (MB) more than the doctor is willing to accept (MC) until we reach N<sup>*</sup> ] --- # Coase: Point 1 .pull-left[  ] .pull-right[ We now maximize surplus (.hi-blue[blue]) and gain bargaining surplus (.hi-blue[blue]) that is split between the doctor and data center .hi[It didn't matter who had the property rights, we managed to get to N<sup>*</sup>] ] --- # Coase: Point 2 .pull-left[  ] .pull-right[ The initial assignment of property rights does matter for the distribution of surplus If the data center has the right to create noise, it receives a payment from the doctor up to the total size of the red area (bargaining surplus) ] --- # Coase: Point 2 .pull-left[  ] .pull-right[ The initial assignment of property rights does matter for the distribution of surplus If the doctor has the right to quiet, the doctor receives a payment from the data center up to the total size of the blue area (bargaining surplus) ] --- # Coase: Point 2 .pull-left[  ] .pull-right[ This means that property rights are valuable! If you have property rights, others have to incentivize you in order to deviate from your privately optimal choice You will only change the level of noise if your welfare/surplus improves ] --- # Coase Caveats Coasean bargaining does not always work like you think There are two key pieces we need to have satisfied: 1. No Transactions costs 2. No wealth/income effects --- # Coase Caveats: Transactions costs Suppose the doctor owns the right to quiet Noise `\(N\)` imposes cost `\(C(N)\)` on the doctor and produces benefit `\(B(N)\)` for the data center -- Status quo is no contract: no additional noise and no transfer payment -- This is a one-shot game: the data center makes one take-it-or-leave-it offer in which the doctor accepts noise `\(N\)` in exchange for payment `\(P\)` --- # Coase Caveats: Transactions costs Timing: 1. The data center pays transaction/setup cost `\(tr\)` to propose one contract `\((N,P)\)` 2. Doctor accepts or rejects once 3. If rejected, they stay at the status quo (no additional noise, no transfer) -- There is no renegotiation after the accept/reject decision --- # Coase Caveats: Transactions costs When does the doctor accept the contract? -- The doctor is weakly better off accepting the contract if the payment `\(P\)` is at least the cost of noise `\(C(N)\)` -- Let `\(C(N)\)` denote the doctor's total cost of noise, so `\(C'(N)\)` is the marginal cost of additional noise -- What contract does the data center offer in equilibrium? -- i.e. what contract proposal maximizes the data center's profit? --- # Coase Caveats: Transactions costs The data center will choose to offer `\(P = C(N)\)` -- Why? -- In a one-shot take-it-or-leave-it offer, the data center offers the minimum the doctor is willing to accept: `\(P = C(N)\)` -- If bargaining were unlimited, for a given `\(N\)`, the feasible payment range is between the minimum the doctor is willing to accept and the maximum the data center is willing to pay -- The realized payment in that interval depends on bargaining power and bargaining protocol --- # Coase Caveats: Transactions costs The data center's total profit is: -- `$$\pi(N) = B(N) - \underbrace{C(N)}_{P} - tr$$` where `\(B(N)\)` is the data center's operating benefit from noise -- Its optimal choice of `\(N\)` satisfies: -- `$$B'(N) = C'(N)$$` which is the condition for efficient noise -- .hi[Did transactions costs actually cause any problems?] --- # Coase Caveats: Transactions costs .pull-left[  ] .pull-right[ Yes! Why? ] --- # Coase Caveats: Transactions costs .pull-left[  ] .pull-right[ Yes! Why? To have a mutually beneficial contract we still need the total gain in surplus (.hi-blue[blue]) to be greater than `\(tr\)` ] --- # Coase Caveats: Transactions costs .pull-left[  ] .pull-right[ Yes! Why? To have a mutually beneficial contract we still need the total gain in surplus (.hi-blue[blue]) to be greater than `\(tr\)` Otherwise the total cost of the bargaining is greater than the total benefit from bargaining `\(\rightarrow\)` bargaining makes us worse off ] --- # Coase Caveats: Transactions costs If the data center has the right to create noise, we just flip the script -- The doctor makes a one-shot take-it-or-leave-it contract offer `\((N, P)\)` in which the data center reduces noise from `\(N_0\)` in exchange for a payment --- # When does the data center accept? The data center accepts or rejects once; if rejected, noise stays at `\(N_0\)` with no transfer When does the data center accept the contract? -- The data center accepts if `\(P \geq B(N_0) - B(N)\)` (payment exceeds its lost operating benefit) -- Under one-shot take-it-or-leave-it bargaining, the doctor offers the minimum required: `\(P = B(N_0) - B(N)\)` --- # The feasible payment range With unlimited bargaining, for a given `\(N\)`, the feasible payment range is between the minimum the data center is willing to accept and the maximum the doctor is willing to pay -- Again, the realized payment in that interval depends on bargaining power and bargaining protocol --- # Coase Caveats: Transactions costs The doctor chooses `\(N\)` to minimize total cost: `$$\min_{N} C(N) + \underbrace{\left[B(N_0)-B(N)\right]}_{P} + tr$$` -- `\(C(N)\)` is the remaining noise cost, `\(P\)` is the payment, and `\(tr\)` is the transaction cost -- The first-order condition is: `$$C'(N) = B'(N)$$` --- # Coase Caveats: Transactions costs .pull-left[  ] .pull-right[ We again reach the social optimum! To have a mutually beneficial contract we still need the total gain in surplus (.hi-red[red]) to be greater than `\(tr\)` Otherwise the total cost of the bargaining is greater than the total benefit from bargaining `\(\rightarrow\)` bargaining makes us worse off ] --- # Coase Caveats: Transactions costs Main takeaway: transactions costs can prevent Coasean bargaining from achieving the efficient allocation -- Why? -- In cases where the gains from bargaining are small, transactions costs may exceed the benefits and prohibit bargaining from occurring --- # Coase Caveats: Wealth effects Changing the initial endowment can change our outcome if wealth changes the doctor's preferences for noise -- The property right is valuable: the owner is richer even before trading Quiet makes the doctor more productive: owning the property right raised his wealth before negotiating -- With diminishing marginal utility of wealth, extra dollar matters less: - If the doctor owns quiet, he requires a larger payment to accept noise - If he does not own quiet, he is not willing to pay as much to reduce the noise --- # Coase Caveats: Wealth effects Suppose the data center gains $90 from noisy operation When the doctor owns the right to quiet he is wealthier and values the noise at $100, but when the doctor does not own the right to quiet he is poorer and values the noise at $80 - Doctor owns quiet (wealthier): WTA to accept noise = $100 `\(\rightarrow\)` no trade `\(\rightarrow\)` quiet - Data center owns the right to create noise (doctor poorer): WTP for quiet = $80 `\(\rightarrow\)` no trade `\(\rightarrow\)` noise -- Different initial rights `\(\rightarrow\)` different final allocation --- # Coase Caveats: Income effects .pull-left[  ] .pull-right[ Suppose the data center is given the right to create noise (start at N<sub>0</sub>) The doctor pays the data center to reduce noise and move to N<sub>1</sub><sup>*</sup> The data center receives some bargaining surplus What can the data center do with it? ] --- # Coase Caveats: Income effects .pull-left[  ] .pull-right[ The data center can buy more servers, increasing its marginal benefit from operating the noisy cooling and power equipment This changes the optimal noise to N<sub>2</sub><sup>*</sup> We have a new equilibrium! If they contracted to reach N<sub>1</sub><sup>*</sup>, it is now .hi[inefficient] ] --- # Ways to alleviate transactions cost One common issue is incomplete information -- Disseminating information can make it easier to know each other's costs and benefits which makes beneficial trades more likely to occur --- # Ways to alleviate transactions cost Emergency Planning and Community Right-to-know Act (1986) set-up the Toxic Release Inventory (TRI): http://www.epa.gov/tri/ Green (“eco”) labeling - Allows companies to learn which firms took positive steps to reduce pollution, and to reward them in the marketplace - Firms need to be able to increase demand by enough to offset higher costs and raise profits -- Is there a role for the government to be involved in verifying “green” claims? What really constitutes “organic”? --- # Coase theorem While transaction costs can be the downfall of Coasian bargaining, there are costs to government policy: - Bureaucratic costs - Government might set the policy incorrectly -- Coase and Pigou both have trade-offs to their approaches to solving environmental issues --- class: inverse, center, middle name: coase # An example: The Cheshire transaction <html><div style='float:left'></div><hr color='#EB811B' size=1px width=796px></html> --- # The Cheshire transaction Cheshire, Ohio: a small town with a population 221 before 2002 -- In the 1970’s the town welcomed the construction of a power plant nearby -- Gavin Power plant owned by American Electric Power (AEP): 2.6 GW Enough power for 2 million people (completed 1975) -- In the 1990’s, property values in the village plummeted -- Why would this be the case? --- # The Cheshire transaction Acid rain fallout damaged cars, odors nauseated residents and thick plumes of smoke sometimes blocked the sun -- In August 2000 the Environmental Protection Agency declared the Gavin plant in violation of the Clean Air Act -- A later study found that the air in Cheshire was five times the level necessary to cause an asthma attack -- What happened next? Some real world Coasean bargaining --- # The Cheshire transaction Fall 2001: the village selected a law group from Washington, DC to pressure the power plant and its owner to heed their concerns and clean up the plant’s emissions -- Others wanted the plant to compensate them for diminished property value and to address health concerns -- April 16, 2002: AEP announced its plan to acquire the incorporated town for $20 million -- September 24, 2002: AEP announces that it has finalized the buyout. About 90 percent of town residents have participated in the buyout offer and have signed the health waivers and the confidentiality agreements --- # The Cheshire transaction Property owners in town receive 3.5x assessed value Outside town: 2x assessed value Renters receive $5k for each year lived in Cheshire, up to $25k Must sign a health waiver prohibiting them from suing AEP for future health problems --- # The Cheshire transaction Must also sign a confidentiality agreement -- Cheshire residents over age of 71 able to remain in homes rent free until death -- Original population: 221. Current population: <20 -- Total settlement disbursed by AEP: $20 million -- Attorneys take about 1/3 of settlement money --- # The Cheshire transaction: was it a good thing? We know the efficient pollution control decision was made, why? -- AEP could have abated instead of compensating! Basically all the involved parties agreed to the contract -- Now imagine that AEP Gavin’s control costs were low, and the efficient outcome would be to install additional control equipment -- Would it matter whether we granted the “right to clean air” to the town or to AEP? --- class: inverse, center, middle name: datacenters # Data centers: Coasean bargaining in the AI boom <html><div style='float:left'></div><hr color='#EB811B' size=1px width=796px></html> --- # Cheshire, twenty years later The Cheshire transaction was one power plant buying out one small town -- The AI boom is running the same experiment hundreds of times at once -- Data centers need land, power, and water next to real neighborhoods -- Sometimes the operator and the neighbors bargain, sometimes they end up in court -- The difference is exactly what our transactions cost model predicts --- # A bargaining attempt: Ashburn, Virginia <video width="900" controls preload="auto" style="display:block;margin:auto;"><source src="videos/03-coase-ashburn-buyout-offers.mp4" type="video/mp4"></video> --- # A bargaining attempt: Ashburn, Virginia Developers discussed buying all 143 homes in Ashburn's Regency neighborhood -- The reported target was about $4.4 million per acre---more than $500 million in total -- But the HOA president told NBC4 there was .hi[no current offer], and no sale closed .footnote[Source: NBC4 Washington, March 25, 2026.] --- # Why didn't a deal close? - .hi[Many parties:] 143 homeowners must coordinate - .hi[Holdouts:] each owner may demand a larger share of the surplus - .hi[Regulatory risk:] purchase does not guarantee data center approval --- # When the deal doesn't close: holdouts <video width="900" controls preload="auto" style="display:block;margin:auto;"><source src="videos/03-coase-farmland-holdout.mp4" type="video/mp4"></video> --- # When the deal doesn't close: holdouts The sisters turned down $26 million for the family farmland -- Is this a failure of the Coase theorem? -- Not necessarily: if their true willingness to accept exceeds the developer's willingness to pay, .hi[no deal is the efficient outcome] -- But bargaining with many owners is fragile: - Is a "no" true WTA, or a holdout for a bigger share of the surplus? - With `\(N\)` owners who must all say yes, transactions costs rise fast --- # Why bargaining fails: Southaven <video width="900" controls preload="auto" style="display:block;margin:auto;"><source src="videos/03-coase-xai-turbines-residents.mp4" type="video/mp4"></video> --- # Why bargaining fails: Southaven xAI operates a gas-turbine power plant in Southaven, Mississippi, to supply its Colossus 2 data center across the state line in Memphis, Tennessee -- Why is there no Cheshire-style transaction here? -- - .hi[Many parties:] thousands of households breathe the emissions - .hi[Contested rights:] who owns the air? Permits are disputed, not settled - .hi[Free riding:] every resident hopes the others fund the fight - .hi[Information:] health damages are diffuse and hard to document --- # What replaces bargaining? Courts and regulators High transactions costs choke off bargaining `\(\rightarrow\)` the conflict moves to courts and regulators -- In April 2026, the NAACP and Mississippi NAACP sued xAI and its subsidiary MZX Tech under the Clean Air Act in federal court in northern Mississippi -- The lawsuit concerns 27 allegedly unpermitted turbines at the Southaven power plant -- Instead of bargaining over emissions, the parties are asking courts and regulators to define and enforce the rights --- # Data centers through Coase's lens | | Ashburn proposal | Farmland holdout | Southaven gas plant | |--------------------|:---------------:|:----------------:|:----------------:| | Parties | Few | Few | Thousands | | Property rights | Clear (land) | Clear (land) | Contested (air) | | Transactions costs | Low | Medium | High | | Outcome | No completed deal | No deal | Federal lawsuit | -- Same technology, same externalities, three different outcomes -- The Coase theorem tells us where to look: .hi[who holds the rights, and how costly is it to strike the deal?] --- # Your turn: a data center next door .pull-left[  ] .pull-right[ A data center tests its backup generators `\(G\)` hours per month Marginal benefit to the operator: MB = 20 - 2G Marginal noise cost to the residents: MC = 4 Left alone, the operator would run `\(G_0 = 10\)` hours; residents want zero ] --- # Your turn: a data center next door .pull-left[  ] .pull-right[ MB = 20 - 2G, MC = 4 Any deal needs lawyers: transactions cost `\(tr = 10\)` .hi[Your turn:] 1. Find the efficient hours `\(G^*\)` 2. Rights to the residents: what payments could work? 3. Under each rights assignment: does a deal happen? ] --- # Your turn: the efficient hours .pull-left[  ] .pull-right[ Efficiency: MB = MC 20 - 2G = 4 `\(\Rightarrow\)` `\(G^* = 8\)` The first 8 hours are worth more to the operator than they cost the residents; the last 2 are not ] --- # Your turn: residents hold the rights .pull-left[  ] .pull-right[ Start at G = 0; the operator pays for hours Residents accept `\(P \geq 4 \times 8 = 32\)`; the operator pays at most his benefit, `\(P \leq 96\)` Gains from trade: the .hi-blue[blue] triangle `\(= \frac{1}{2} \times 8 \times 16 = 64\)` `\(64 > tr = 10\)` `\(\rightarrow\)` .hi[the deal happens] and we reach `\(G^* = 8\)` ] --- # Your turn: the operator holds the rights .pull-left[  ] .pull-right[ Start at `\(G_0 = 10\)`; residents pay to cut hours Gains from trade: the .hi-red[red] triangle `\(= \frac{1}{2} \times 2 \times 4 = 4\)` `\(4 < tr = 10\)` `\(\rightarrow\)` .hi[no deal], and skipping the deal is the right call With transactions costs, .hi[who holds the rights changes the outcome] ] --- class: inverse, center, middle name: game # Let's bargain: the data center game <html><div style='float:left'></div><hr color='#EB811B' size=1px width=796px></html> --- # The data center bargaining game You will be randomly paired: one .hi[data center operator], one .hi[resident next door] -- Bargain in real time over .hi[generator hours] (0 to 6) and a .hi[payment]: send offers, counter, accept, reject, or walk away before the clock runs out -- - Round 1: the operator holds the rights (no deal = 6 hours) - Round 2: the resident holds the rights (no deal = 0 hours) - Round 3: same as round 2, but any paid deal costs 5 in legal fees -- Your payoffs across all three rounds go on the class leaderboard -- .hi[Join now:] course website `\(\rightarrow\)` Content `\(\rightarrow\)` Games 1: Coase --- # The data center bargaining game: what should happen? If Coase is right, what do we expect to see in the data? -- - Rounds 1 and 2 should land on the .hi[same] generator hours - The .hi[payments] should flow in opposite directions: rights are valuable - Round 3: think before you deal -- We will check the class results against these predictions after we play --- # Coase in practice Cap-and-trade is effectively Coase at large scale -- It allocates a number of rights to pollute (permits or allowances) -- The total number of rights is the cap -- Coase tells us that the initial distribution of permits does not matter -- The cap-and-trade system will then achieve the efficient outcome