good evening to everyone it is my great honor and pleasure tonight to present you paul milgrom the shirley and leonard ali professor at stanford university paul milgrom is among the leading experts in the world of game theory which is the set of mathematical models that economists use to study strategic interactions for instance firm pricing decisions and throughout his career he has contributed enormously to broaden the set of tools that game theory has to model these situations and to apply these tools to a wide variety of problems from financial market to firms internal organization to market structures and for his contributions and specifically for his theoretical contribution in auctions and for the development of new auction systems he was awarded the nobel memorial prize in economics together with bob wilson in 2020 so let me tell you that i've been a professor of economics for more than 10 years now both in the united states and in italy and it is absolutely impossible to teach modern microeconomics without spending substantial amount of times on many of the works of paul milgrom because there are so many areas in economics that it touched and so many models that he created and that later became the toolbox for us to understand and study these markets that he is really off one of the fathers of modern microeconomics and this would be already a lot an incredible contribution but there is more there is the fact that paul milgrom was able to speak with policymakers and to bring to reality some of his ideas so back in 1993 he was one of the persons that worked together with the fcc and played a crucial role in constructing a beautiful new market called the auction for spectrum licenses so all our modern phones and all our economy is based on the exchange of data and this is possible thanks to the good allocation of this car's resource that is the spectrum of telecommunications and it is thanks to paul milgrom that it was possible to invent a market system for the efficient allocation of this car's resource and the more recent times between 2016 and 2017 he designed an auction that took place in the united states again for the location of auctions and in the k in this case the auction system that he had invented was able to relocate spectrum from one set of players local tv stations to another set of players and the telephone operators that were eager to use the spectrum for their services to consumers so when i teach my class in milan and i tell to students how do you give to the government seven billion dollars i say what do you do do you raise taxes do you sell assets do you go to the market selling bond government bonds i say well milgram created for the us government seven billion dollars by constructing a market that was able to create efficiency by real locating resources that were previously not efficiently allocated and so the topic of this year festival year entrento is the return of the state and so since the state is indeed returning is crucial that the state adopts the proper system to allocate economic resources so let me give you now the floor to professor milgram who will tell us about his research and how the state can be helped to work better thank you well thank you um hello everybody um so my talk is is titled modern auction design some novel challenges but um given who i was told was in the audience i decided actually to break this talk into two pieces i want to talk to you for a few moments about recent trends in the world economy which um i know everybody who follows economics is uh definitely interested in and then the second half of this talk i'll talk a little bit about modern auction market design what is it that makes this different from any other option why is this different from going into an auction house and bidding for a piece of art or or an antique or something like that so what is it what is it about modern auction design that's novel and we'll talk about that for a few moments after we talk about some trends in the world economy now i want to talk about the world economy because it's on everybody's mind now the all the disruption that's occurred on account of the uh pandemic and and the sudden loss of jobs and if you take a look here um the loss of jobs and recovery this is uh a statistic from the the graph here is the u.s unemployment rate um global economy shrunk by over three percent in 2020 and most countries went into recession with of course china being different but you can see if you see the um the recovery from or the increase in unemployment that occurred during the um uh during 2008 you can see that it occurred gradually and much more gradually and was much smaller than the impact that occurred in um in 2020 when we were hit by the coven pandemic so so the current global economy is just obviously dominated by what's going on an account of uh on account of coping and it's led to unprecedented uh disruptions in global supply chains ever since uh beginning with the factory shutdowns in wuhan which was a major supplier of quite a few companies internationally we saw chip shortages that emerged that are expected to last beyond 2022 we've had a recent resurgence in demand and the people who it takes time to build capacity demand can surge quickly and we had shortfalls that resulted from these shutdowns and it's it's changed the way businesses are thinking about their uh the just-in-time manufacturing about maintaining very low levels of inventories and being vulnerable um to these kinds of disruptions so these kinds of disruptions are changing the way the whole world economy works that's one of the things that i think every every economist is thinking about uh thinking about these days a lot of the uh chip manufacturing now we're relying on china on taiwan where tensions with china may arise and lead for lead to uh further disruptions in the supply chain for electronics um covet has created a lot of uncertainty and this is another factor that probably hasn't gotten as much attention as it should have uncertainty is really bad for investment if you are a producer and you're not sure that there's going to be demand you delay uh building your factories and that delay itself um leads to reductions in demand and can be very bad for the economy the the graph here that's on the left shows the the consensus forecast we look at the degree of disagreement the standard deviation and we take a database that includes the all the forecasts of the leading forecasters and you can see that the there's always some disagreement but in 2020 the disagreement among the forecasters spiked and what that means is you know really there's people disagree and there's a lot of uncertainty if you're a businessman and trying to decide what to do about the uh what the prospects are and policy uncertainty the graph on the right shows you what's happening in terms of over time in terms of news articles that are written about policy uncertainty um how is the government going to respond and again we're at historic highs right now in terms of the degree of policy uncertainty and that too is bad for investment and leads to mistakes of course since we could get these things wrong the other thing the other thing that you uh i think we all need to see is how fast the recovery can be in in the us here you see a comparison between the um what happened in 2008 when we had a recession and a recovery that began over a period of years and what happened in 2020 where you can see just a much sharper drop and a v-shaped recovery it bounced right back as the economy began to open up and we're getting now as you all know uh excessive demand but this is only part of the story it's not just that the recover that the economy closed and nope and and reopened it's also that there was a huge shift in where the economic activity is taking place in fact we've had the biggest reallocation of labor since world war ii as people move into different kinds of work so you can see here that amazon now employs more than a million people they added 400 000 people in the last year huge increases facebook had a huge increase in the number of people meanwhile and in transportation and restaurants and hotels there were you know huge amounts of job loss the tourism industry faced about a trillion dollar loss during the course of this according to uh un reports and um this kind of thing well it's bad enough the way it is but the problem is that the inequality across workers is reinforcing the trend that um that has already been underway in our economy so what you're looking at here is the lower skilled and younger workers the percentage of new unemployment claims you can see that people with lower levels of education that is with the high school diploma or the equivalent had many more new unemployment plans people with graduate degrees are hardly losing jobs at all and the more educated you are the less impacted you are by what's going on in the economy and this is uh especially troubling because it reinforces existing pre-existing trends surging inequalities we have seen here the household income gap over a long period of years this is our u.s numbers yet the the the gap for uh between the difference in 2012 dollars between the high school educated and college educated has nearly doubled from about 30 000 to about 60 000 and that's typical of what's going on around the world according to the oecd and in 2018 post-secondary degrees earned 50 more than high school lonely so what we're seeing is is a continued growth and huge inequality and those are the very the people at the bottom are the very same people who are being hurt in the pandemic and it looks like it's going to only get worse if you take a look at job openings and you take a look at what new technologies are doing well here i'm looking at ai and perhaps it's no surprise that in information technology and uh and you know finance and consulting and management that's where the new jobs are and job openings this is about job openings and retail and wholesale and services as that follow as a result of new technologies there are very few new jobs being created down here the most of the new jobs that are being created are for highly educated people that's going to make these things only these inequalities only worse i felt like i just you know couldn't be giving an economics talk that uh um at a seminar like this one without beginning by saying something about that but i realized my main charge was to talk to you guys about modern market design and um so what i'm going to do now is is talk to you a little bit about the kinds of things that are going on in auction market design and why there might be a nobel prize why it's any different than just submitting a bid or or just competing against one another in an open auction at a typical auction house so this actually is quite a lot going on in the modern economy that's making use of new kinds of auction rules so um here's just the sample of a few uh just to give you an idea of um what's happening so in on the internet every time you conduct a search on the internet there's an auction that's run um it determines uh you know who gets to show their ad to you and or what position their ad is in on the page i'll show you a little bit of a difference about that so so what makes that different from anything else well one of the differences is that these auctions are run using automated bidders uh another one of the differences um these auctions take place in you know milliseconds they're over very quickly and they're all entirely automated when you program your bidder to say how much it wants the bid to show your ad um another thing that's novel about it is the basis of comparison so the bidders will often bid in terms of the price per click what they'll pay if somebody on a particular search page clicks on their ad and of course the the uh the person the publisher is not selling clicks the publisher is selling space on its page it's interested in the amount that it's paid uh per impression or per thousand impressions that are there so there's a conversion that has to take place and there's uh the uh bidders are concerned about uh clicks partly because they cannot eclipse i can't if if i'm bidding and you tell me yeah you've showed my ad a thousand times in a newspaper i can check that easily i just buy a copy of the newspaper but on the internet you can't check whether you know the the guy down the street has viewed your ad so so you want to pay on a basis that you can audit and where your value comes from clicks and the newspaper wants to sell or the publisher wants to sell on the basis of its limited resource which is space on the page so that involved all kinds of innovation radio spectrum i'm going to give you some examples of that too um uh the the multiple licenses that are for something that are for sale this varies by country but especially in um in larger areas it may be that uh the united states for example that you only want to buy a license in california if you can also buy licenses for all the western states so the licenses are are compliments and you care about not just an individual item that you're buying but the whole package of things and how it fits into your business plan or there can be substitutes there can be different kinds of licenses that serve the same purpose and you may want to may want to compare the prices and buy one but not the other this extensive pattern of substitutes and complements is is something that's been very important in radio spectrum of auction design the third of the applications i want to tell you about briefly i'll do a little bit more detail in the first two because we'll have a few moments for that but the third one i'll mention is is food banks where the issue is how do we allocate when donors give food the right food to the right places where they need it we don't want to send a donated potatoes to idaho and we uh we we the people in alaska might find that what they need is not jars of pickles but fresh produce and and so how do we get this allocation done right without uh when these guys don't have any money because the whole point is that there are that the gifts they're getting are in terms of goods so this is really a barter exchange um in which uh gifts are being distributed and and what's been created um candace prendergast at the university of chicago was the economist who led this particular project what was arranged was a system in which people bid credits and they were given credits according to their perceived needs and any credits that were spent to buy for example a truckload of produce would then be redistributed among the uh the other food banks so that they could use to make purchases the next day so this is again something that involves some new elements no real money but they wanted to run an auction in order to promote an efficient allocation of resources environmental resources cap and trade well i think many of you have heard of cap and traded especially in connection with carbon but there are lots of other uses of cap and trade that are mined in in new south wales and australia fishing rights were were sold by kemp they they had over fishing they kept the uh they kept the level of of of fishing to a sustainable level and then um reduced the uh the catch that was associated with rights and some of the fishing boats needed to go out of business and other fishing boats needed to acquire more rights so they could fill their capacity and this was a complicated market because of for those of you who are economists because of non-condensed cities basically we had some fishing boats that would be interested in selling the rights if they could sell all of them but they wouldn't we're not interested in selling selling the rights just piecemeal um so that's that's one kind of thing in in california we have cap and trade on uses of groundwater where the uh right now there's excessive use of groundwater um the the uh it's not sustainable and uh the government has created a mandate for sustainable markets and groundwater uh in the near future and markets are being set up where we uh cap the amount of groundwater that can be taken and we do trade so that it can go to toward the more uh more efficient uses a lot of you are probably excited about what's going on in the way of uh digital currencies and cryptocurrencies uh including uh government-issued uh cryptocurrency digital currencies and the the new contracts that are being used for decentralized finance where there's automated pricing based on supply and demand information and again the rules of the market um are being determined on the fly and some of the stuff works well and some of it does not but these are all new kinds of auction markets that are very different from just submitting a simple sealed bid or going into an auction house and bidding live against a competitor so i'm going to say it for a few more words about the first two of these things about internet advertising and radio spectrum and then i guess i will join the conversation uh with my host so um so this is a sponsored search if you go to google the other day i was looking for a new bike pump and when i put in bike pumps the first things that appear on the top of the page as you can all see are um are a bunch of advertisements for buy bike pumps um in addition to men's clothing and shoes and so on um but this is uh uh these are advertisements and you can see this is a large ad and this this ad is uh includes a graphic and um uh and this ad is a smaller ad well how do they decide which ads to show and how much space to give to each other there's been a whole series a whole evolution of the way these took place in the early days of internet advertising on google all the ads were the same size little three or four line ads there were simple text ads that appeared on the right hand side of the page they were all perfect substitutes for one another google would just compare what it thought was the price per impression for each that is what it could get the bids were made in terms of price per click uh google would estimate the click rate and figure out what is that really worth to me in terms of price per impression it would rank the ads that way and it had a pricing rule to determine the prices but now we see the ads have become more complex the this is all an ad for specialized and it's a much larger ad than the ad that appears next and over here we have some color ads so how does um how does google determine all of that well the space on the page isn't isn't fixed and it isn't a single thing that's being sold these require uh complicated new kinds of option rules similar to victory auctions actually that are used to determine uh this allocation now after i searched for bike pumps i then went over to cnn to download a page to sign up for this i did this last week and as you can see the crypto crash had just wiped out a trillion dollars worth of value but um what do i see on the cnn page but an ad for uh for a bicycle um an electric bike they figured i guess uh cnn uh now knows that i was searching for bike pumps and thought maybe i would be interested in a new bike too and uh and again there's an auction that takes place um bidders who know something about me are um are seeking to personalize an ad that um that's shown directly to me so again these are real-time uh auctions that take place in milliseconds and where we try uh the the auctioneer tries to show the most valuable most relevant ad to the uh uh to the consumer who appears and the publisher is selling this ad space in this case to a bicycle company to uh to try to show an ad to me okay let's see we have um now i want to turn um to the broadcast incentive auction which was mentioned in my introduction uh what was it why was it developed well what was going on at this in this period this is uh during the last years of the obama administration and the united states had been going through what the rest of the world was going through and exploding demand for mobile broadband for which there weren't sufficient frequencies available and we were trying to figure out where was the three where were those frequencies going to come from that we would use for um for mobile broadband for 4g and now 5g kinds of technologies so what was going on at the same time was that consumer viewing of over-the-air television broadcast was plummeting people were beginning to use alternative means like cable television satellite television watching tv over the internet and just real-time over-the-air broadcast was declining and a lot of frequencies had been devoted to that so the idea would be to try to get some of the frequencies that had been used for over the year broadcast and make them available for mobile broadband and for which we thought there would be a high value of the frequencies and the uhf television spectrum which was being used for tv was especially good for what uh the telecom guys call coverage spectrum which means that it has great propagation characteristics you can have it in a rural area and um and a cell tower in a rural area can serve an area of many miles um where the uh because these frequencies have such good uh propagation characteristics and even in cities its ability uh the propagation there that's important is the ability to penetrate uh indoors and you know into buildings with thick walls that um and still be able to receive signals so so this was really um very good uh spectrum for those uses now what made this um the auction of this sword hard well let me give you an idea um here we're going to take a look at this this is looking at a television station and in the center you see this red dot which represents the location of a tv station and this graph represents each each dot or node on this graph represents the location of some uhf tv station somewhere in the northeast of the united states and each edge in this graph that connects two nodes in the graph represents a pair of stations that can't both be assigned to the same td channel without creating an acceptable interference so what we're trying to do is reassign but take some of the stations off the air reassign the remaining stations to channels in a way that lets them continue to broadcast without interference uh the nodes that are connected to the central red node are colored in this pink or salmon color here and and so this graph is we call this a co-channel interference graph it's a graph where again any two stations that can't be assigned to the same channel are connected by some edge and why is that interesting well by the way this was a project that was done in coordination with canada so you see here the interference graph for the united states and in canada and there's about 2500 tv stations that are shown here there's about 130 000 of these constraints that is of these edges in this graph and a set of stations can all continue to broadcast if there's a way to assign channels to the stations without interference we're trying to figure out which sets of stations can be left on the air that's part of what the auction has to decide and what channel should they be assigned so there's about 130 000 of these constraints there's some other constraints too but this is the this is the heart of it and um this is a famous problem in computer science and operations research it's called a graph coloring problem if you have a graph that is a bunch of nodes and edges and you want to assign a color to each node so that no two connected nodes are the same color that's called the graph coloring problem and here the problem of assigning channels to tv stations is exactly a graph coloring problem you want to assign a channel to the station so that no two connected stations are the on the same channel and that's what's called an np complete problem which roughly speaking means it's hard even for computers the this is not a problem that we can guarantee even to solve if you give me a set of stations um and a set of colors and say is there a way to assign uh channels to these stations so that they're not interfering and we run any known algorithm um the chances are if you you uh you give it uh some amount of time a day a minute whatever the amount of time you choose sometimes the computer is just going to come back and say sorry that wasn't long enough i can't solve this problem this is a problem that might take millennia to solve i mean it's it's this problem sometimes you're lucky and they solve fast but some instances you discover you can't solve at all and that makes it hard to design an auction we can't even tell which sets of state sometimes can't tell whether a set of stations can be simultaneously packed unto the air and we had to create an auction that you know bidders would understand we didn't want to show them some complicated mathematics that made them doubt uh bidders who were already disinclined to trust the government and made them doubt this so we decided that we would hide all the complexity under the hood and here's how we ran the auction each tv station was offered a tentative price which might later be adjusted downward so i say to a tv station owner would you take 100 million dollars for your station and the tv station owner looks and says to himself oh my station's worth 20 million dollars to me so the answer to that question is yes and the station owner can say yes and which and continue in the auction or the station owner can say no and exit the auction keeping its broadcast rights then i go around and i take a look and i say well gee it looks like i have more offers than i need so the price that i offer to the station in each round is reduced i'm gonna not describe the exact formula for reduction let's call it five percent um uh and uh and i'll reduce the offer that i make to a station so one is i can compute a channel assignment for that station and others that have already exited so i know when i reduce your price that if you say no there's space for you to continue to broadcast and if there's no space for you to continue to broadcast then i don't lower your price and say fine you accepted my offer of 75 million dollars you got it um i'm going to pay you 75 million dollars for your station and uh so that's basically how the auction runs all the hard computation is under the hood and if i can't compute a channel assignment if the computer comes back and says no i don't know whether there's a way to put that station on the air and i say fine um then i can't guarantee that there's a station a channel for that station i'm going to buy that station so that was the rule we used to run this auction and we did some other stuff the outcome was pretty good we cleared channels 38 to 51 from television broadcast um resulting in 70 megahertz which is quite a lot of new spectrum for mobile broadband and 14 megahertz for other low-powered applications it generated 20 billion dollars of revenue from the sale of broadcast licenses that we created and we paid about 10 billion dollars in total to broadcasters to uh relinquish their broadcast rights so that was the auction that was mentioned earlier there was about three billion dollars in expenses leading to about seven billion dollars that went to the government well that's the amount of time i have so um thank you all i hope you enjoyed my talk that's it for now so thank you so much paul if there are questions from the audience you're welcome to go there to the podium and ask your question in the meantime to break the eyes let me start by asking the first question paul um it would be interesting for all of us here to understand how you managed to interact with government officials with policy makers to describe them the benefits of engaging with something that seems complex even to experts this seems to complex and you need a big leap of faith to do something like this and then in the end you managed to convince them and great benefits came from this so what were the tricks we can learn from your experience so so actually um a lot of this began much earlier in 1993 my first interaction with the federal communications commission was it was in 1993 and at that point the federal communications commission had never run an auction before it didn't have an auctions division and uh the people inside uh didn't know what to do they uh you know they they weren't sure how to do it so they put together a proposal but the the task of of designing the auction was handed to what that time was the economic analysis group and those guys were phd economists and they didn't know what to do so they were very open and expected to get ideas from industry and industry was open too i got called by what was then pacific bell and uh they said you know here's this proposal we don't understand it at all it's so complicated so so i was able to go to washington and explain an alternative proposal and uh at the time because there was no entrenched bureaucracy um the uh the fcc was very open to suggestions and they basically adopted what was the milgram wilson design for the first u.s spectrum auction which later was copied around the world and evolved and then i was not so much an insider until again it what happened in 2010 and 11 it was 2011 the legislation was passed and this was passed up to the fcc and they looked at this and they shrugged they had no idea what to do and you know it's very hard to get a bureaucracy where the people think they know what to do to change what they're doing but they talked about this internally and decided they didn't know what to do and so they reached out to me in um 2011 and asked me if i would uh sign a design and build contract um they said we want something we can actually do um so the contract is going to say you design it subject to our oversight and you build it because we have no idea what kind of software you're going to need how to do this so it was a very practical problem that they didn't know how to do and so they looked for expertise and in my case um it was two times uh uh when when people recognized that the people who were there didn't know what to do didn't have a standard way of doing it and they were willing to they were open to new ideas and in fact second time they reached out to me and said we have no clue can you make a suggestion thank you so much so now one question yes uh thank you very much professor milgram um so we have talked both in the introduction by professor de carolis and in your speech of markets that have been successfully created um i was thinking during your speech about one example of not such such a successful operation and it was the european emission trading scheme which was launched in the european union back in 2005 i guess you know creating trying to create a market for the rights to emit uh emissions and in order to to reduce them but it didn't actually take off uh properly uh so i was just wondering what's your stance on these whether you believe that it was a failure in the original design of the of this market or i don't know it was mainly because of other exogenous reasons that it failed thank you okay well thank you so um i don't know the details of that particular market but let me just describe how subtle it is to create some of these markets um you know it it's very interesting in economics that when we do economic theory we take the markets we take the commodities as given uh we say okay we're selling wheat for example or worse or we're buying carbon permits and we treat it as though as though the product design is obvious and the product design is not obvious um it's when if you are trading carbon emissions what are the rights exactly that you're trading um so you need to put a lot of thought into what the product is is this the right to admit in a given year uh is it a perpetual right to emit uh at a given rate what is what is it that's being bought and sold and then um this uh one of the things at least in the u.s that's been a a major problem sometimes is the use of offsets i don't know if that was built into what you're doing but some companies claim many companies claim in fact that they're offsetting their carbon footprint by buying and preserving let's say a plot of forest or buying some other carbon intensive activity in an attempt to offset the uh the carbon uses they have but those things themselves are also very complicated often if you find yourself buying a plot of land uh and saying okay then we're going to cut down the trees on this land and we're not going to cut them down anymore so that's an offset but it may be that the people who need the wood or whatever just cut down trees somewhere else the the figuring out what the product should be and what offsets will work well that's another complication also the you know the extent of the mandates the the how the whole thing fits in with the uh with the government regulations so yeah markets depend on looking at all the details and uh making sure that the regulations that surround them are right that the that the product design is right that the uh how often should the market be run is this an annual thing or something where trading takes place continuously um you know all of those details need to be thought through so i don't know the details that were used in the particular market that you described but i do want to emphasize that market design is subtle and it's remarkable how little attention we used to pay to this as economists about all these all the fine details that that it takes for america to succeed so building on your last comment maybe one thing that is that has been crucial in your research and we did not explicitly discuss this evening and it would be a nice way maybe to close and then um to leave you would be to hear about one of these crucial details that you were mentioning and so the role of information which so fundamentally so maybe if you have a couple of words to explain to the audience in simple words what's what is the role of information and how it can lead to things like the winners course sure well there's a lot of that that's important in market design so what the winner's curse that i was explicitly asked about and which was in some of my early research and a lot of bob wilson's early research is about the idea that you are more likely to win an auction as a bidder if you're trying to buy something you're more likely to bid high when your estimate is high and uh you're more likely to win when your estimate is high than when your estimate is low so if your estimates are unbiased on average there's an ad for there's a selection against you an adverse selection that you win more often when you've overestimated and you win less often when you've underestimated and that tendency is known as the winner's curse and the winner's course is alleviated or can be alleviated by information that is the winner's curse is least when um the bidders are all comparably informed and we see lots of examples where the winner's curse is important i mentioned internet advertising earlier today the the information that bidders have when they're trying to offer a bicycle to me or a bicycle pump um the bidders don't all have the same information about me they may not have we know that these demand side platforms that is the the uh the systems that inform the bidders give them information they need to help them bid they're all different and um and bidders may have a different information and may find themselves subject to the winner's curse there and you know some of the platforms are trying very hard to equalize that is to provide the same information um uh to bidders so that the allocation is is not driven by it's driven by values uh that are different across the bidders rather than by uh some being better informed than others and by the way the winner's curse is a is a reason for market failure just like ever since akron we have known that adverse selection can cause people not to trade because they're afraid of getting a bad deal the winner's curse can cause people not to bid and so the the really big companies google for example tries to do a good job of estimating on behalf of its clients what the values are of the various impressions that it's offering so that the bidders can bid effectively so changing the information that's available has been an important part of real world market design most recently i'd say a lot's been happening on the internet in that regard so thank you so much paul and on this note i think we can leave you and from all of us thanks again and we hope to see you soon in well i'd like italy get there too italy is one of my favorite countries and i would love to love to visit you all soon bye-bye bye-bye you