The Price of Everything, the Value of the Economy: A Clark Medal for Emi Nakamura!

Fantastic and well-deserved news this morning with the Clark Medal being awarded to Emi Nakamura, who has recently moved from Columbia to Berkeley. Incredibly, Nakamura’s award is the first Clark to go to a macroeconomist in the 21st century. The Great Recession, the massive changes in global trade patterns, the rise of monetary areas like the Eurozone, the “savings glut” and its effect on interest rates, the change in openness to hot financial flows: it has been a wild twenty years for the macroeconomy in the two decades since Andrei Schleifer won the Clark. It’s hard to imagine what could be more important for an economist to understand than these patterns.

Something unusual has happened in macroeconomics over the past twenty years: it has become more like Industrial Organization! A brief history may be useful. The term macroeconomics is due to Ragnar Frisch, in his 1933 article on the propagation of economic shocks. He writes,

“The macro-dynamic analysis…tries to give an account of the whole economic system taken in its entirety. Obviously in this case it is impossible to carry through the analysis in great detail. Of course, it is always possible to give even a macro-dynamic analysis in detail if we confine ourselves to a purely formal theory. Indeed, it is always possible by a suitable system of subscripts and superscripts, etc., to introduce practically all factors which we may imagine…Such a theory, however, would have only a rather limited interest. It would hardly be possible to study such fundamental problems as the exact time shape of the solution, [etc.]. These latter problems are just the essential problems in business cycle analysis. In order to attack these problems on a macro-dynamic basis…we must deliberately disregard a considerable amount of the details of the picture.

And so we did. The Keynesians collapsed the microfoundations of the macroeconomy into a handful of relevant market-wide parameters. The Lucas Critique argued that we can collapse some things – many agents into a representative agent, for instance – but we ought always begin our analysis with the fundamental parameters of tastes, constraints, and technologies. The neoclassical synthesis combined these raw parameters with nominal ridigities – sticky prices, limited information, and so on. But Frisch’s main point nonetheless held strong: to what use are these deeper theoretical parameters if we cannot estimate their value and their effect on the macroeconomy? As Einstein taught us, the goal of the scientist should be to make things as simple as possible, but no simpler.

What has changed recently in macroeconomics is twofold. First, computational power now makes it possible to estimate or calibrate very complex dynamic and stochastic models, with forward looking agents, with price paths in and out of equilibrium, with multiple frictions – it is in this way that macro begins to look like industrial organization, with microeconomic parameters at the base. But second, and again analogous to IO, the amount of data available to the researcher has grown enormously. We now have price scanner data that tells us exactly when and how prices change, how those changes propagate across supply chains and countries, how they interact with taxes, and so on. Frisch’s problem has in some sense been solved: we no longer have the same trade-off between usefulness and depth when studying the macroeconomy.

Nakamura is best known for using this deep combination of data and theory to understand how exactly firms set prices. Price rigidities play a particularly important role in theories of the macroeconomy that potentially involve inefficiency. Consider a (somewhat bowdlerized) version of real business cycle theory. Here, shocks hit the economy: for instance, an oil cartel withholds supply for political reasons. Firms must react to this “real” supply-side shock by reorganizing economic activity. The real shock then propagates across industries. The role of monetary policy in such a world is limited: a recession simply reflects industries reacting to real change in the economic environment.

When prices are “sticky”, however, that is no longer true. The speed by which real shocks propagate, and the distortion sticky prices introduce, can be affected by monetary policy, since firms will react to changes in expected inflation by changing the frequency in which they update prices. Famously, Golosov and Lucas in the JPE argued, theoretically and empirically, that the welfare effects of “sticky prices” or “menu costs” are not terribly large. Extracting these welfare effects is quite sensitive to a number of features in the data and in the theory. To what extent is there short-term price dispersion rather than an exogenous chance for all firms in an industry to change their prices? Note that price dispersion is difficult to maintain unless we have consumer search costs – otherwise, everyone buys from the cheapest vendor – so price dispersion adds a non-trivial technical challenge. How much do prices actually change – do we want to sweep out short-term sales, for example? When inflation is higher, do firms adjust prices equally often but with bigger price jumps (consider the famous doubling of the price of Coca-Cola), or do they adjust prices more often keeping the percentage change similar to low-inflation environments? How much heterogeneity is there is the price-setting practice across industries, and to what extent do these differences affect the welfare consequences of prices given the links across industries?

Namakura has pushed us very far into answering these questions. She has built insane price datasets, come up with clever identification strategies to separate pricing models, and used these tools to vastly increase our understanding of the interaction between price rigidities and the business cycle. Her “Five Facts” paper uses BLS microdata to show that sales were roughly half of the “price changes” earlier researchers has found, that prices change more rapidly when inflation is higher, and that there is huge heterogeneity across industries in price change behavior. Taking that data back to the 1970s, Nakamura and coauthors also show that high inflation environments do not cause more price dispersion: rather, firms update their prices more often. Bob Lucas in his Macroeconomic Priorities made a compelling argument that business cycle welfare costs are much smaller than the costs of inflation and inflation costs are themselves much smaller than the costs of tax distortions. As Nakamura points out, if you believe this, no wonder you prioritize price stability and tax policy! (Many have quibbled with Lucas’ basic argument, but even adding heterogeneous agents, it is tough to get business cycles to have large economic consequences; see, e.g., Krusell et al RED 2009.) Understanding better the true costs of inflation, via the feedback of monetary expansion on pricesetting, goes a great deal toward helping policymakers calibrate the costs and benefits of price stability vis-a-vis other macroeconomic goals.

Though generally known as an empirical macroeconomist, Nakamura also has a number of papers, many with her husband Jon Steinsson, on the theory of price setting. For example, why are prices both sticky and also involve sales? In a clever paper in the JME, Nakamura and Steinsson model a firm pricing to habit-forming consumers. If the firm does not constrain itself, it has the incentive to raise prices once consumers form their habit for a given product (as a Cheez-It fan, I understand the model well – my willingness to pay for a box shipped up from the US to the Cheez-It-free land of Canada is absurdly high). To avoid this time inconsistency problems, firms would like to commit to a price path with some flexibility to respond to changes in demand. An equilibrium in this relational contract-type model involves a price cap with sales when demand falls: rigid prices plus sales, as we see in the data! In a second theoretical paper with Steinsson and Alisdair McKay, Nakamura looks into how much communication about future nominal interest rates can affect behavior. In principle, a ton: if you tell me the Fed will keep the real interest rate low for many years (low rates in the future raise consumption in the future which raises inflation in the future which lowers real rates today), I will borrow away. Adding borrowing constraints and income risk, however, means that I will never borrow too much money: I might get a bad shock tomorrow and wind up on the street. Giving five years of forward guidance about interest rates rather than a year, therefore, doesn’t really affect my behavior that much: the desire to have precautionary savings is what limits my borrowing, not the interest rate.

Nakamura’s prize is a well-deserved award, going to a leader in the shift in macro toward a more empirical, more deeply “microeconomic” in its theory, style of macro. Her focus is keenly targeted toward some of the key puzzles relevant to macroeconomic policymakers. There is no way to cover such a broad field in one post – this is not one of those awards given for a single paper – but luckily Nakamura has two great easily-readable summaries of her core work. First, in the Annual Review of Economics, she lays out the new empirical facts on price changes, the attempts to identify the link between monetary policy and price changes, and the implications for business cycle theory. Second, in the Journal of Economic Perspectives, she discusses how macroeconomists have attempted to more credibly identify theoretical parameters. In particular, external validity is so concerning in macro – remember the Lucas Critique! – that the essence of the problem involves combining empirical variation for identification with theory mapping that variation into broader policy guidance. I hesitate to stop here since Nakamura has so many influential papers, but let us take just more quick tasters that are well worth your more deep exploration. On the government spending side, she uses local spending shocks and a serious model to figure out the national fiscal multiplier from government spending. Second, she recently has linked the end of large-scale increases in female moves from home production to the labor force has caused recessions to last longer.

Advertisement

The 2018 Fields Medal and its Surprising Connection to Economics!

The Fields Medal and Nevanlinna Prizes were given out today. They represent the highest honor possible for young mathematicians and theoretical computer scientists, and are granted only once every four years. The mathematics involved is often very challenging for outsiders. Indeed, the most prominent of this year’s winners, the German Peter Scholze, is best known for his work on “perfectoid spaces”, and I honestly have no idea how to begin explaining them aside from saying that they are useful in a number of problems in algebraic geometry (the lovely field mapping results in algebra – what numbers solve y=2x – and geometry – noting that those solutions to y=2x form a line). Two of this year’s prizes, however, the Fields given to Alessio Figalli and the Nevanlinna to Constantinos Daskalakis, have a very tight connection to an utterly core question in economics. Indeed, both of those men have published work in economics journals!

The problem of interest concerns how best to sell an object. If you are a monopolist hoping to sell one item to one consumer, where the consumer’s valuation of the object is only known to the consumer but commonly known to come from a distribution F, the mechanism that maximizes revenue is of course the Myerson auction from his 1981 paper in Math OR. The solution is simple: make a take it or leave it offer at a minimum price (or “reserve price”) which is a simple function of F. If you are selling one good and there are many buyers, then revenue is maximized by running a second-price auction with the exact same reserve price. In both cases, no potential buyer has any incentive to lie about their true valuation (the auction is “dominant strategy incentive compatible”). And further, seller revenue and expected payments for all players are identical to the Myerson auction in any other mechanism which allocates goods the same way in expectation, with minor caveats. This result is called “revenue equivalence”.

The Myerson paper is an absolute blockbuster. The revelation principle, the revenue equivalence theorem, and a solution to the optimal selling mechanism problem all in the same paper? I would argue it’s the most important result in economics since Arrow-Debreu-McKenzie, with the caveat that many of these ideas were “in the air” in the 1970s with the early ideas of mechanism design and Bayesian game theory. The Myerson result is also really worrying if you are concerned with general economic efficiency. Note that the reserve price means that the seller is best off sometimes not selling the good to anyone, in case all potential buyers have private values below the reserve price. But this is economically inefficient! We know that there exists an allocation mechanism which is socially efficient even when people have private information about their willingness to pay: the Vickrey-Clarke-Groves mechanism. This means that market power plus asymmetric information necessarily destroys social surplus. You may be thinking we know this already: an optimal monopoly price is classic price theory generates deadweight loss. But recall that a perfectly-price-discriminating monopolist sells to everyone whose willingness-to-pay exceeds the seller’s marginal cost of production, hence the only reason monopoly generates deadweight loss in a world with perfect information is that we constrain them to a “mechanism” called a fixed price. Myerson’s result is much worse: letting a monopolist use any mechanism, and price discriminate however they like, asymmetric information necessarily destroys surplus!

Despite this great result, there remain two enormous open problems. First, how should we sell a good when we will interact with the same buyer(s) in the future? Recall the Myerson auction involves bidders truthfully revealing their willingness to pay. Imagine that tomorrow, the seller will sell the same object. Will I reveal my willingness to pay truthfully today? Of course not! If I did, tomorrow the seller would charge the bidder with the highest willingness-to-pay exactly that amount. Ergo, today bidders will shade down their bids. This is called the “ratchet effect”, and despite a lot of progress in dynamic mechanism design, we have still not fully solved for the optimal dynamic mechanism in all cases.

The other challenging problem is one seller selling many goods, where willingness to pay for one good is related to willingness to pay for the others. Consider, for example, selling cable TV. Do you bundle the channels together? Do you offer a menu of possible bundles? This problem is often called “multidimensional screening”, because you are attempting to “screen” buyers such that those with high willingness to pay for a particular good actually pay a high price for that good. The optimal multidimensional screen is a devil of a problem. And it is here that we return to the Fields and Nevanlinna prizes, because they turn out to speak precisely to this problem!

What could possibly be the connection between high-level pure math and this particular pricing problem? The answer comes from the 18th century mathematician Gaspard Monge, founder of the Ecole Polytechnique. He asked the following question: what is the cheapest way to move mass from X to Y, such as moving apples from a bunch of distribution centers to a bunch of supermarkets. It turns out that without convexity or linearity assumptions, this problem is very hard, and it was not solved until the late 20th century. Leonid Kantorovich, the 1975 Nobel winner in economics, paved the way for this result by showing that there is a “dual” problem where instead of looking for the map from X to Y, you look for the probability that a given mass in Y comes from X. This dual turns out to be useful in that there exists an object called a “potential” which helps characterize the optimal transport problem solution in a much more tractable way than searching across any possible map.

Note the link between this problem and our optimal auction problem above, though! Instead of moving mass most cheaply from X to Y, we are looking to maximize revenue by assigning objects Y to people with willingness-to-pay drawn from X. So no surprise, the solution to the optimal transport problem when X has a particular structure and the solution to the revenue maximizing mechanism problem are tightly linked. And luckily for us economists, many of the world’s best mathematicians, including 2010 Fields winner Cedric Villani, and this year’s winner Alessio Figalli, have spent a great deal of effort working on exactly this problem. Ivar Ekeland has a nice series of notes explaining the link between the two problems in more detail.

In a 2017 Econometrica, this year’s Nevanlinna winner Daskalakis and his coauthors Alan Deckelbaum and Christos Tzamos, show precisely how to use strong duality in the optimal transport problem to solve the general optimal mechanism problem when selling multiple goods. The paper is very challenging, requiring some knowledge of measure theory, duality theory, and convex analysis. That said, the conditions they give to check an optimal solution, and the method to find the optimal solution, involve a reasonably straightforward series of inequalities. In particular, the optimal mechanism involves dividing the hypercube of potential types into (perhaps infinite) regions who get assigned the same prices and goods (for example, “you get good A and good B together with probability p at price X”, or “if you are unwilling to pay p1 for A, p2 for B, or p for both together, you get nothing”).

This optimal mechanism has some unusual properties. Remember that the Myerson auction for one buyer is “simple”: make a take it or leave it offer at the reserve price. You may think that if you are selling many items to one buyer, you would likewise choose a reserve price for the whole bundle, particularly when the number of goods with independently distributed values becomes large. For instance, if there are 1000 cable channels, and a buyer has value distributed uniformly between 0 and 10 cents for each channel, then by a limit theorem type argument it’s clear that the willingness to pay for the whole bundle is quite close to 50 bucks. So you may think, just price at a bit lower than 50. However, Daskalakis et al show that when there are sufficiently many goods with i.i.d. uniformly-distributed values, it is never optimal to just set a price for the whole bundle! It is also possible to show that the best mechanism often involves randomization, where buyers who report that they are willing to pay X for item a and Y for item b will only get the items with probability less than 1 at specified price. This is quite contrary to my intuition, which is that in most mechanism problems, we can restrict focus to deterministic assignment. It was well-known that multidimensional screening has weird properties; for example, Hart and Reny show that an increase in buyer valuations can cause seller revenue from the optimal mechanism to fall. The techniques Daskalakis and coauthors develop allow us to state exactly what we ought do in these situations previously unknown in the literature, such as when we know we need mechanisms more complicated than “sell the whole bundle at price p”.

The history of economics has been a long series of taking tools from the frontier of mathematics, from the physics-based analogues of the “marginalists” in the 1870s, to the fixed point theorems of the early game theorists, the linear programming tricks used to analyze competitive equilibrium in the 1950s, and the tropical geometry recently introduced to auction theory by Elizabeth Baldwin and Paul Klemperer. We are now making progress on pricing issues that have stumped some of the great theoretical minds in the history of the field. Multidimensional screening is an incredibly broad topic: how ought we regulate a monopoly with private fixed and marginal costs, how ought we tax agents who have private costs of effort and opportunities, how ought a firm choose wages and benefits, and so on. Knowing the optimum is essential when it comes to understanding when we can use simple, nearly-correct mechanisms. Just in the context of pricing, using related tricks to Daskalakis, Gabriel Carroll showed in a recent Econometrica that bundling should be avoided when the principal has limited knowledge about the correlation structure of types, and my old grad school friend Nima Haghpanah has shown, in a paper with Jason Hartline, that firms should only offer high-quality and low-quality versions of their products if consumers’ values for the high-quality good and their relative value for the low versus high quality good are positively correlated. Neither of these results are trivial to prove. Nonetheless, a hearty cheers to our friends in pure mathematics who continue to provide us with the tools we need to answer questions at the very core of economic life!

Nobel Prize 2016 Part II: Oliver Hart

The Nobel Prize in Economics was given yesterday to two wonderful theorists, Bengt Holmstrom and Oliver Hart. I wrote a day ago about Holmstrom’s contributions, many of which are simply foundational to modern mechanism design and its applications. Oliver Hart’s contribution is more subtle and hence more of a challenge to describe to a nonspecialist; I am sure of this because no concept gives my undergraduate students more headaches than Hart’s “residual control right” theory of the firm. Even stranger, much of Hart’s recent work repudiates the importance of his most famous articles, a point that appears to have been entirely lost on every newspaper discussion of Hart that I’ve seen (including otherwise very nice discussions like Applebaum’s in the New York Times). A major reason he has changed his beliefs, and his research agenda, so radically is not simply the whims of age or the pressures of politics, but rather the impact of a devastatingly clever, and devastatingly esoteric, argument made by the Nobel winners Eric Maskin and Jean Tirole. To see exactly what’s going on in Hart’s work, and why there remains many very important unsolved questions in this area, let’s quickly survey what economists mean by “theory of the firm”.

The fundamental strangeness of firms goes back to Coase. Markets are amazing. We have wonderful theorems going back to Hurwicz about how competitive market prices coordinate activity efficiently even when individuals only have very limited information about how various things can be produced by an economy. A pencil somehow involves graphite being mined, forests being explored and exploited, rubber being harvested and produced, the raw materials brought to a factory where a machine puts the pencil together, ships and trains bringing the pencil to retail stores, and yet this decentralized activity produces a pencil costing ten cents. This is the case even though not a single individual anywhere in the world knows how all of those processes up the supply chain operate! Yet, as Coase pointed out, a huge amount of economic activity (including the majority of international trade) is not coordinated via the market, but rather through top-down Communist-style bureaucracies called firms. Why on Earth do these persistent organizations exist at all? When should firms merge and when should they divest themselves of their parts? These questions make up the theory of the firm.

Coase’s early answer is that something called transaction costs exist, and that they are particularly high outside the firm. That is, market transactions are not free. Firm size is determined at the point where the problems of bureaucracy within the firm overwhelm the benefits of reducing transaction costs from regular transactions. There are two major problems here. First, who knows what a “transaction cost” or a “bureaucratic cost” is, and why they differ across organizational forms: the explanation borders on tautology. Second, as the wonderful paper by Alchian and Demsetz in 1972 points out, there is no reason we should assume firms have some special ability to direct or punish their workers. If your supplier does something you don’t like, you can keep them on, or fire them, or renegotiate. If your in-house department does something you don’t like, you can keep them on, or fire them, or renegotiate. The problem of providing suitable incentives – the contracting problem – does not simply disappear because some activity is brought within the boundary of the firm.

Oliver Williamson, a recent Nobel winner joint with Elinor Ostrom, has a more formal transaction cost theory: some relationships generate joint rents higher than could be generated if we split ways, unforeseen things occur that make us want to renegotiate our contract, and the cost of that renegotiation may be lower if workers or suppliers are internal to a firm. “Unforeseen things” may include anything which cannot be measured ex-post by a court or other mediator, since that is ultimately who would enforce any contract. It is not that everyday activities have different transaction costs, but that the negotiations which produce contracts themselves are easier to handle in a more persistent relationship. As in Coase, the question of why firms do not simply grow to an enormous size is largely dealt with by off-hand references to “bureaucratic costs” whose nature was largely informal. Though informal, the idea that something like transaction costs might matter seemed intuitive and had some empirical support – firms are larger in the developing world because weaker legal systems means more “unforeseen things” will occur outside the scope of a contract, hence the differential costs of holdup or renegotiation inside and outside the firm are first order when deciding on firm size. That said, the Alchian-Demsetz critique, and the question of what a “bureaucratic cost” is, are worrying. And as Eric van den Steen points out in a 2010 AER, can anyone who has tried to order paper through their procurement office versus just popping in to Staples really believe that the reason firms exist is to lessen the cost of intrafirm activities?

Grossman and Hart (1986) argue that the distinction that really makes a firm a firm is that it owns assets. They retain the idea that contracts may be incomplete – at some point, I will disagree with my suppliers, or my workers, or my branch manager, about what should be done, either because a state of the world has arrived not covered by our contract, or because it is in our first-best mutual interest to renegotiate that contract. They retain the idea that there are relationship-specific rents, so I care about maintaining this particular relationship. But rather than rely on transaction costs, they simply point out that the owner of the asset is in a much better bargaining position when this disagreement occurs. Therefore, the owner of the asset will get a bigger percentage of rents after renegotiation. Hence the person who owns an asset should be the one whose incentive to improve the value of the asset is most sensitive to that future split of rents.

Baker and Hubbard (2004) provide a nice empirical example: when on-board computers to monitor how long-haul trucks were driven began to diffuse, ownership of those trucks shifted from owner-operators to trucking firms. Before the computer, if the trucking firm owns the truck, it is hard to contract on how hard the truck will be driven or how poorly it will be treated by the driver. If the driver owns the truck, it is hard to contract on how much effort the trucking firm dispatcher will exert ensuring the truck isn’t sitting empty for days, or following a particularly efficient route. The computer solves the first problem, meaning that only the trucking firm is taking actions relevant to the joint relationship which are highly likely to be affected by whether they own the truck or not. In Grossman and Hart’s “residual control rights” theory, then, the introduction of the computer should mean the truck ought, post-computer, be owned by the trucking firm. If these residual control rights are unimportant – there is no relationship-specific rent and no incompleteness in contracting – then the ability to shop around for the best relationship is more valuable than the control rights asset ownership provides. Hart and Moore (1990) extends this basic model to the case where there are many assets and many firms, suggesting critically that sole ownership of assets which are highly complementary in production is optimal. Asset ownership affects outside options when the contract is incomplete by changing bargaining power, and splitting ownership of complementary assets gives multiple agents weak bargaining power and hence little incentive to invest in maintaining the quality of, or improving, the assets. Hart, Schleifer and Vishny (1997) provide a great example of residual control rights applied to the question of why governments should run prisons but not garbage collection. (A brief aside: note the role that bargaining power plays in all of Hart’s theories. We do not have a “perfect” – in a sense that can be made formal – model of bargaining, and Hart tends to use bargaining solutions from cooperative game theory like the Shapley value. After Shapley’s prize alongside Roth a few years ago, this makes multiple prizes heavily influenced by cooperative games applied to unexpected problems. Perhaps the theory of cooperative games ought still be taught with vigor in PhD programs!)

There are, of course, many other theories of the firm. The idea that firms in some industries are big because there are large fixed costs to enter at the minimum efficient scale goes back to Marshall. The agency theory of the firm going back at least to Jensen and Meckling focuses on the problem of providing incentives for workers within a firm to actually profit maximize; as I noted yesterday, Holmstrom and Milgrom’s multitasking is a great example of this, with tasks being split across firms so as to allow some types of workers to be given high powered incentives and others flat salaries. More recent work by Bob Gibbons, Rebecca Henderson, Jon Levin and others on relational contracting discusses how the nexus of self-enforcing beliefs about how hard work today translates into rewards tomorrow can substitute for formal contracts, and how the credibility of these “relational contracts” can vary across firms and depend on their history.

Here’s the kicker, though. A striking blow was dealt to all theories which rely on the incompleteness or nonverifiability of contracts by a brilliant paper of Maskin and Tirole (1999) in the Review of Economic Studies. Theories relying on incomplete contracts generally just hand-waved that there are always events which are unforeseeable ex-ante or impossible to verify in court ex-post, and hence there will always scope for disagreement about what to do when those events occur. But, as Maskin and Tirole correctly point out, agent don’t care about anything in these unforeseeable/unverifiable states except for what the states imply about our mutual valuations from carrying on with a relationship. Therefore, every “incomplete contract” should just involve the parties deciding in advance that if a state of the world arrives where you value keeping our relationship in that state at 12 and I value it at 10, then we should split that joint value of 22 at whatever level induces optimal actions today. Do this same ex-ante contracting for all future profit levels, and we are done. Of course, there is still the problem of ensuring incentive compatibility – why would the agents tell the truth about their valuations when that unforeseen event occurs? I will omit the details here, but you should read the original paper where Maskin and Tirole show a (somewhat convoluted but still working) mechanism that induces truthful revelation of private value by each agent. Taking the model’s insight seriously but the exact mechanism less seriously, the paper basically suggests that incomplete contracts don’t matter if we can truthfully figure out ex-post who values our relationship at what amount, and there are many real-world institutions like mediators who do precisely that. If, as Maskin and Tirole prove (and Maskin described more simply in a short note), incomplete contracts aren’t a real problem, we are back to square one – why have persistent organizations called firms?

What should we do? Some theorists have tried to fight off Maskin and Tirole by suggesting that their precise mechanism is not terribly robust to, for instance, assumptions about higher-order beliefs (e.g., Aghion et al (2012) in the QJE). But these quibbles do not contradict the far more basic insight of Maskin and Tirole, that situations we think of empirically as “hard to describe” or “unlikely to occur or be foreseen”, are not sufficient to justify the relevance of incomplete contracts unless we also have some reason to think that all mechanisms which split rent on the basis of future profit, like a mediator, are unavailable. Note that real world contracts regularly include provisions that ex-ante describe how contractual disagreement ex-post should be handled.

Hart’s response, and this is both clear from his CV and from his recent papers and presentations, is to ditch incompleteness as the fundamental reason firms exist. Hart and Moore’s 2007 AER P&P and 2006 QJE are very clear:

Although the incomplete contracts literature has generated some useful insights about firm boundaries, it has some shortcomings. Three that seem particularly important to us are the following. First, the emphasis on noncontractible ex ante investments seems overplayed: although such investments are surely important, it is hard to believe that they are the sole drivers of organizational form. Second, and related, the approach is ill suited to studying the internal organization of firms, a topic of great interest and importance. The reason is that the Coasian renegotiation perspective suggests that the relevant parties will sit down together ex post and bargain to an efficient outcome using side payments: given this, it is hard to see why authority, hierarchy, delegation, or indeed anything apart from asset ownership matters. Finally, the approach has some foundational weaknesses [pointed out by Maskin and Tirole (1999)].

To my knowledge, Oliver Hart has written zero papers since Maskin-Tirole was published which attempt to explain any policy or empirical fact on the basis of residual control rights and their necessary incomplete contracts. Instead, he has been primarily working on theories which depend on reference points, a behavioral idea that when disagreements occur between parties, the ex-ante contracts are useful because they suggest “fair” divisions of rent, and induce shading and other destructive actions when those divisions are not given. These behavioral agents may very well disagree about what the ex-ante contract means for “fairness” ex-post. The primary result is that flexible contracts (e.g., contracts which deliberately leave lots of incompleteness) can adjust easily to changes in the world but will induce spiteful shading by at least one agent, while rigid contracts do not permit this shading but do cause parties to pursue suboptimal actions in some states of the world. This perspective has been applied by Hart to many questions over the past decade, such as why it can be credible to delegate decision making authority to agents; if you try to seize it back, the agent will feel aggrieved and will shade effort. These responses are hard, or perhaps impossible, to justify when agents are perfectly rational, and of course the Maskin-Tirole critique would apply if agents were purely rational.

So where does all this leave us concerning the initial problem of why firms exist in a sea of decentralized markets? In my view, we have many clever ideas, but still do not have the perfect theory. A perfect theory of the firm would need to be able to explain why firms are the size they are, why they own what they do, why they are organized as they are, why they persist over time, and why interfirm incentives look the way they do. It almost certainly would need its mechanisms to work if we assumed all agents were highly, or perfectly, rational. Since patterns of asset ownership are fundamental, it needs to go well beyond the type of hand-waving that makes up many “resource” type theories. (Firms exist because they create a corporate culture! Firms exist because some firms just are better at doing X and can’t be replicated! These are outcomes, not explanations.) I believe that there are reasons why the costs of maintaining relationships – transaction costs – endogenously differ within and outside firms, and that Hart is correct is focusing our attention on how asset ownership and decision making authority affects incentives to invest, but these theories even in their most endogenous form cannot do everything we wanted a theory of the firm to accomplish. I think that somehow reputation – and hence relational contracts – must play a fundamental role, and that the nexus of conflicting incentives among agents within an organization, as described by Holmstrom, must as well. But we still lack the precise insight to clear up this muddle, and give us a straightforward explanation for why we seem to need “little Communist bureaucracies” to assist our otherwise decentralized and almost magical market system.

“Does Regression Produce Representative Estimates of Causal Effects?,” P. Aronow & C. Samii (2016)

A “causal empiricist” turn has swept through economics over the past couple decades. As a result, many economists are primarily interested in internally valid treatment effects according to the causal models of Rubin, meaning they are interested in credible statements of how some outcome Y is affected if you manipulate some treatment T given some covariates X. That is, to the extent that full functional form Y=f(X,T) is impossible to estimate because of unobserved confounding variables or similar, it turns out to still be possible to estimate some feature of that functional form, such as the average treatment effect E(f(X,1))-E(f(X,0)). At some point, people like Angrist and Imbens will win a Nobel prize not only for their applied work, but also for clarifying precisely what various techniques are estimating in a causal sense. For instance, an instrumental variable regression under a certain exclusion restriction (let’s call this an “auxiliary assumption”) estimates the average treatment effect along the local margin of people induced into treatment. If you try to estimate the same empirical feature using a different IV, and get a different treatment effect, we all know now that there wasn’t a “mistake” in either paper, but rather than the margins upon which the two different IVs operate may not be identical. Great stuff.

This causal model emphasis has been controversial, however. Social scientists have quibbled because causal estimates generally require the use of small, not-necessarily-general samples, such as those from a particular subset of the population or a particular set of countries, rather than national data or the universe of countries. Many statisticians have gone even further, suggestion that multiple regression with its linear parametric form does not take advantage of enough data in the joint distribution of (Y,X), and hence better predictions can be made with so-called machine learning algorithms. And the structural economists argue that the parameters we actually care about are much broader than regression coefficients or average treatment effects, and hence a full structural model of the data generating process is necessary. We have, then, four different techniques to analyze a dataset: multiple regression with control variables, causal empiricist methods like IV and regression discontinuity, machine learning, and structural models. What exactly do each of these estimate, and how do they relate?

Peter Aronow and Cyrus Samii, two hotshot young political economists, take a look at old fashioned multiple regression. Imagine you want to estimate y=a+bX+cT, where T is a possibly-binary treatment variable. Assume away any omitted variable bias, and more generally assume that all of the assumptions of the OLS model (linearity in covariates, etc.) hold. What does that coefficient c on the treatment indicator represent? This coefficient is a weighted combination of the individual estimated treatment effects, where more weight is given to units whose treatment status is not well explained by covariates. Intuitively, if you are regressing, say, the probability of civil war on participation in international institutions, then if a bunch of countries with very similar covariates all participate, the “treatment” of participation will be swept up by the covariates, whereas if a second group of countries with similar covariates all have different participation status, the regression will put a lot of weight toward those countries since differences in outcomes can be related to participation status.

This turns out to be quite consequential: Aronow and Samii look at one paper on FDI and find that even though the paper used a broadly representative sample of countries around the world, about 10% of the countries weighed more than 50% in the treatment effect estimate, with very little weight on a number of important regions, including all of the Asian tigers. In essence, the sample was general, but the effective sample once you account for weighting was just as limited as some of “nonrepresentative samples” people complain about when researchers have to resort to natural or quasinatural experiments! It turns out that similar effective vs. nominal representativeness results hold even with nonlinear models estimated via maximum likelihood, so this is not a result unique to OLS. Aronow and Samii’s result matters for interpreting bodies of knowledge as well. If you replicate a paper adding in an additional covariate, and get a different treatment effect, it may not reflect omitted variable bias! The difference may simply result from the additional covariate changing the effective weighting on the treatment effect.

So the “externally valid treatment effects” we have been estimating with multiple regression aren’t so representative at all. So when, then, is old fashioned multiple regression controlling for observable covariates a “good” way to learn about the world, compared to other techniques. I’ve tried to think through this is a uniform way; let’s see if it works. First consider machine learning, where we want to estimate y=f(X,T). Assume that there are no unobservables relevant to the estimation. The goal is to estimate the functional form f nonparametrically but to avoid overfitting, and statisticians have devised a number of very clever ways to do this. The proof that they work is in the pudding: cars drive themselves now. It is hard to see any reason why, if there are no unobservables, we wouldn’t want to use these machine learning/nonparametric techniques. However, at present the machine learning algorithms people use literally depend only on data in the joint distribution (X,Y), and not on any auxiliary assumptions. To interpret the marginal effect of a change in T as some sort of “treatment effect” that can be manipulated with policy, if estimated without auxiliary assumptions, requires some pretty heroic assumptions about the lack of omitted variable bias which essentially will never hold in most of the economic contexts we care about.

Now consider the causal model, where y=f(X,U,T) and you interested in what would happen with covariates X and unobservables U if treatment T was changed to a counterfactual. All of these techniques require a particular set of auxiliary assumptions: randomization requires the SUTVA assumption that treatment of one unit does not effect the independent variable of another unit, IV requires the exclusion restriction, diff-in-diff requires the parallel trends assumption, and so on. In general, auxiliary assumptions will only hold in certain specific contexts, and hence by construction the result will not be representative. Further, these assumptions are very limited in that they can’t recover every conditional aspect of y, but rather recover only summary statistics like the average treatment effect. Techniques like multiple regression with covariate controls, or machine learning nonparametric estimates, can draw on a more general dataset, but as Aronow and Samii pointed out, the marginal effect on treatment status they identify is not necessarily effectively drawing on a more general sample.

Structural folks are interested in estimating y=f(X,U,V(t),T), where U and V are unobserved, and the nature of unobserved variables V are affected by t. For example, V may be inflation expectations, T may be the interest rate, y may be inflation today, and X and U are observable and unobservable country characteristics. Put another way, the functional form of f may depend on how exactly T is modified, through V(t). This Lucas Critique problem is assumed away by the auxiliary assumptions in causal models. In order to identify a treatment effect, then, additional auxiliary assumptions generally derived from economic theory are needed in order to understand how V will change in response to a particular treatment type. Even more common is to use a set of auxiliary assumptions to find a sufficient statistic for the particular parameter desired, which may not even be a treatment effect. In this sense, structural estimation is similar to causal models in one way and different in two. It is similar in that it relies on auxiliary assumptions to help extract particular parameters of interest when there are unobservables that matter. It is different in that it permits unobservables to be functions of policy, and that it uses auxiliary assumptions whose credibility leans more heavily on non-obvious economic theory. In practice, structural models often also require auxiliary assumptions which do not come directly from economic theory, such as assumptions about the distribution of error terms which are motivated on the basis of statistical arguments, but in principle this distinction is not a first order difference.

We then have a nice typology. Even if you have a completely universal and representative dataset, multiple regression controlling for covariates does not generally give you a “generalizable” treatment effect. Machine learning can try to extract treatment effects when the data generating process is wildly nonlinear, but has the same nonrepresentativeness problem and the same “what about omitted variables” problem. Causal models can extract some parameters of interest from nonrepresentative datasets where it is reasonable to assume certain auxiliary assumptions hold. Structural models can extract more parameters of interest, sometimes from more broadly representative datasets, and even when there are unobservables that depend on the nature of the policy, but these models require auxiliary assumptions that can be harder to defend. The so-called sufficient statistics approach tries to retain the former advantages of structural models while reducing the heroics that auxiliary assumptions need to perform.

Aronow and Samii is forthcoming in the American Journal of Political Science; the final working paper is at the link. Related to this discussion, Ricardo Hausmann caused a bit of a stir online this week with his “constant adaptation rather than RCT” article. His essential idea was that, unlike with a new medical drug, social science interventions vary drastically depending on the exact place or context; that is, external validity matters so severely that slowly moving through “RCT: Try idea 1”, then “RCT: Try idea 2”, is less successful than smaller, less precise explorations of the “idea space”. He received a lot of pushback from the RCT crowd, but I think for the wrong reason: the constant iteration is less likely to discover underlying mechanisms than even an RCT, as it is still far too atheoretical. The link Hausmann makes to “lean manufacturing” is telling: GM famously (Henderson and Helper 2014) took photos of every square inch of their joint venture plant with NUMMI, and tried to replicate this plant in their other plants. But the underlying reason NUMMI and Toyota worked has to do with the credibility of various relational contracts, rather than the (constantly iterated) features of the shop floor. Iterating without attempting to glean the underlying mechanisms at play is not a rapid route to good policy.

Edit: A handful of embarrassing typos corrected, 2/26/2016

“Forced Coexistence and Economic Development: Evidence from Native American Reservations,” C. Dippel (2014)

I promised one more paper from Christian Dippel, and it is another quite interesting one. There is lots of evidence, folk and otherwise, that combining different ethnic or linguistic groups artificially, as in much of the ex-colonial world, leads to bad economic and governance outcomes. But that’s weird, right? After all, ethnic boundaries are themselves artificial, and there are tons of examples – Italy and France being the most famous – of linguistic diversity quickly fading away once a state is developed. Economic theory (e.g., a couple recent papers by Joyee Deb) suggests an alternative explanation: groups that have traditionally not worked with each other need time to coordinate on all of the Pareto-improving norms you want in a society. That is, it’s not some kind of intractable ethnic hate, but merely a lack of trust that is the problem.

Dippel uses the history of American Indian reservations to examine the issue. It turns out that reservations occasionally included different subtribal bands even though they almost always were made up of members of a single tribe with a shared language and ethnic identity. For example, “the notion of tribe in Apachean cultures is very weakly developed. Essentially it was only a recognition
that one owed a modicum of hospitality to those of the same speech, dress, and customs.” Ethnographers have conveniently constructed measures of how integrated governance was in each tribe prior to the era of reservations; some tribes had very centralized governance, whereas others were like the Apache. In a straight OLS regression with the natural covariates, incomes are substantially lower on reservations made up of multiple bands that had no pre-reservation history of centralized governance.

Why? First, let’s deal with identification (more on what that means in a second). You might naturally think that, hey, tribes with centralized governance in the 1800s were probably quite socioeconomically advanced already: think Cherokee. So are we just picking up that high SES in the 1800s leads to high incomes today? Well, in regions with lots of mining potential, bands tended to be grouped onto one reservation more frequently, which suggests that resource prevalence on ancestral homelands outside of the modern reservation boundaries can instrument for the propensity for bands to be placed together. Instrumented estimates of the effect of “forced coexistence” is just as strong as the OLS estimate. Further, including tribe fixed effects for cases where single tribes have a number of reservations, a surprisingly common outcome, also generates similar estimates of the effect of forced coexistence.

I am very impressed with how clear Dippel is about what exactly is being identified with each of these techniques. A lot of modern applied econometrics is about “identification”, and generally only identifies a local average treatment effect, or LATE. But we need to be clear about LATE – much more important than “what is your identification strategy” is an answer to “what are you identifying anyway?” Since LATE identifies causal effects that are local conditional on covariates, and the proper interpretation of that term tends to be really non-obvious to the reader, it should go without saying that authors using IVs and similar techniques ought be very precise in what exactly they are claiming to identify. Lots of quasi-random variation generates that variation along a local margin that is of little economic importance!

Even better than the estimates is an investigation of the mechanism. If you look by decade, you only really see the effect of forced coexistence begin in the 1990s. But why? After all, the “forced coexistence” is longstanding, right? Think of Nunn’s famous long-run effect of slavery paper, though: the negative effects of slavery are mediated during the colonial era, but are very important once local government has real power and historically-based factionalism has some way to bind on outcomes. It turns out that until the 1980s, Indian reservations had very little local power and were largely run as government offices. Legal changes mean that local power over the economy, including the courts in commercial disputes, is now quite strong, and anecdotal evidence suggests lots of factionalism which is often based on longstanding intertribal divisions. Dippel also shows that newspaper mentions of conflict and corruption at the reservation level are correlated with forced coexistence.

How should we interpret these results? Since moving to Canada, I’ve quickly learned that Canadians generally do not subscribe to the melting pot theory; largely because of the “forced coexistence” of francophone and anglophone populations – including two completely separate legal traditions! – more recent immigrants are given great latitude to maintain their pre-immigration culture. This heterogeneous culture means that there are a lot of actively implemented norms and policies to help reduce cultural division on issues that matter to the success of the country. You might think of the problems on reservations and in Nunn’s post-slavery states as a problem of too little effort to deal with factionalism rather than the existence of the factionalism itself.

Final working paper, forthcoming in Econometrica. No RePEc IDEAS version. Related to post-colonial divisions, I also very much enjoyed Mobilizing the Masses for Genocide by Thorsten Rogall, a job market candidate from IIES. When civilians slaughter other civilians, is it merely a “reflection of ancient ethnic hatred” or is it actively guided by authority? In Rwanda, Rogall finds that almost all of the killing is caused directly or indirectly by the 50,000-strong centralized armed groups who fanned out across villages. In villages that were easier to reach (because the roads were not terribly washed out that year), more armed militiamen were able to arrive, and the more of them that arrived, the more deaths resulted. This in-person provoking appears much more important than the radio propaganda which Yanigazawa-Drott discusses in his recent QJE; one implication is that post-WW2 restrictions on free speech in Europe related to Nazism may be completely misdiagnosing the problem. Three things I especially liked about Rogall’s paper: the choice of identification strategy is guided by a precise policy question which can be answered along the local margin identified (could a foreign force stopping these centralized actors a la Romeo Dallaire have prevented the genocide?), a theoretical model allows much more in-depth interpretation of certain coefficients (for instance, he can show that most villages do not appear to have been made up of active resistors), and he discusses external cases like the Lithuanian killings of Jews during World War II, where a similar mechanism appears to be at play. I’ll have many more posts on cool job market papers coming shortly!

“Minimal Model Explanations,” R.W. Batterman & C.C. Rice (2014)

I unfortunately was overseas and wasn’t able to attend the recent Stanford conference on Causality in the Social Sciences; a friend organized the event and was able to put together a really incredible set of speakers: Nancy Cartwright, Chuck Manski, Joshua Angrist, Garth Saloner and many others. Coincidentally, a recent issue of the journal Philosophy of Science had an interesting article quite relevant to economists interested in methodology: how is it that we learn anything about the world when we use a model that is based on false assumptions?

You might think of there being five classes which make up nearly every paper published in the best economics journals. First are pure theoretical exercises, or “tool building”, such as investigations of the properties of equilibria or the development of a new econometric technique. Second are abstract models which are meant to speak to an applied problem. Third are empirical papers whose primary quantities of interest are the parameters of an economic model (broadly, “structural papers”, although this isn’t quite the historic use of the term). Fourth are empirical papers whose primary quantities of interest are causal treatment effects (broadly, “reduced form papers”, although again this is not the historic meaning of that term). Fifth are descriptive work or historical summary. Lab and field experiments, and old-fashioned correlation analysis, all fit into that framework fairly naturally as well. It is the second and third classes which seem very strange to many non-economists. We write a model which is deliberately abstract and which is based on counterfactual assumptions about human or firm behavior, but nonetheless we feel that these types of models are “useful” or “explanatory” in some sense. Why?

Let’s say that in the actual world, conditions A imply outcome B via implication C (perhaps causal, perhaps as part of a simultaneous equilibrium, or whatever). The old Friedman 1953 idea is that a good model predicts B well across all questions with which we are concerned, and the unreality of the assumptions (or implicitly of the logical process C) are unimportant. Earlier literature in the philosophy of science has suggested that “minimal models” explain because A’, a subset of A, are sufficient to drive B via C; that is, the abstraction merely strips away any assumptions that are not what the philosopher Weisberg calls “explanatorily privileged causal factors.” Pincock, another philosopher, suggests that models track causes, yes, but also isolate factors and connect phenomena via mathematical similarity. That is, the model focuses on causes A’, subset of A, and on implications C’, subset of C, which are of special interest because they help us see how the particular situation we are analyzing is similar to ones we have analyzed before.

Batterman and Rice argue that these reasons are not why minimal models “work”. For instance, if we are to say that a model explains because it abstracts only to the relevant causal factors, the question is how we know what those factors are in advance of examining them. Consider Fisher’s sex ratio model: why do we so frequently see 1:1 sex ratios in nature? He argues that there is a fitness advantage for those whose offspring tend toward the less common sex, since they find it easier to procreate. In the model, parents choose sex of offspring, reproduction is asexual (does not involve matching), no genetic recombination occurs, there are no random changes to genes, etc: many of the assumptions are completely contrary to reality. Why, then, do we think the model explains? It explains because there is a story about why the omitted factors are irrelevant to the behavior being explained. That is, in the model assumptions D generate E via causal explanation C, and there is a story about why D->E via C and A->B via C operate in similar ways. Instead of simply assuming that certain factors are “explanatorily privileged”, we show that that model factors affect outcomes in similar ways to how more complicated real world objects operate.

Interesting, but I feel that this still isn’t what’s going on in economics. Itzhak Gilboa, the theorist, in a review of Mary Morgan’s delightful book The World in the Model, writes that “being an economic theorist, I have been conditioned to prefer elegance over accuracy, insight over detail.” I take that to mean that what economic theorists care about are explanatory factors or implications C’, subset of C. That is, the deduction is the theory. Think of Arrow’s possibility theorem. There is nothing “testable” about it; certainly the theory does not make any claim about real world outcomes. It merely shows the impossibility of preference aggregation satisfying certain axioms, full stop. How is this “useful”? Well, the usefulness of this type of abstract model depends entirely on the user. Some readers may find such insight trivial, or uninteresting, or whatever, whereas others may find such an exploration of theoretical space helps clarify their thinking about some real world phenomenon. The whole question of “Why do minimal models explain/work/predict” is less interesting to me than the question “Why do minimal models prove useful for a given reader“.

The closest philosophical position to this idea is some form of Peirce-style pragmatism – he actually uses a minimal model himself in exactly this way in his Note on the Economy of the Theory of Research! I also find it useful to think about the usefulness of abstract models via Economic Models as Analogies, an idea pushed by Gilboa and three other well-known theorists. Essentially, a model is a case fully examined. Examining a number of cases in the theoretical world, and thinking formally through those cases, can prove useful when critiquing new policy ideas or historical explanations about the world. The theory is not a rule – and how could it be given the abstractness of the model – but an element in your mental toolkit. In physics, for example, if your engineer proposes spending money building a machine that implies perpetual motion, you have models of the physical world in your toolkit which, while not being about exactly that machine, are useful when analyzing how such a machine would or would not work. Likewise, if Russian wants to think about how it should respond to a “sudden stop” in investment and a currency outflow, the logical consequences of any real world policy are so complex that it is useful to have thought through the equilibrium implications of policies within the context of toy models, even if such models are only qualitatively useful or only useful in certain cases. When students complain, “but the assumptions are so unrealistic” or “but the model can’t predict anything”, you ought respond that the model can predict perfectly within the context of the model, and it is your job as the student, as the reader, to consider how understanding the mechanisms in the model help you think more clearly about related problems in the real world.

Final version in Philosophy of Science, which is gated, I’m afraid; I couldn’t find an ungated draft. Of related interest in the philosophy journals recently is Kevin Davey’s Can Good Science Be Logically Inconsistent? in Synthese. Note that economists use logically inconsistent reasoning all the time, in that we use model with assumption A in context B, and model with assumption Not A in context C. If “accepting a model” means thinking of the model as “justified belief”, then Davey provides very good reasons to think that science cannot be logically inconsistent. If, however, “accepting a model” meaning “finding it useful as a case” or “finding the deduction in the model of inherent interest”, then of course logically inconsistent models can still prove useful. So here’s to inconsistent economics!

“Aggregation in Production Functions: What Applied Economists Should Know,” J. Felipe & F. Fisher (2003)

Consider a firm that takes heterogeneous labor and capital inputs L1, L2… and K1, K2…, using these to produce some output Y. Define a firm production function Y=F(K1, K2…, L1, L2…) as the maximal output that can be produced using the given vector of outputs – and note the implicit optimization condition in that definition, which means that production functions are not simply technical relationships. What conditions are required to construct an aggregated production function Y=F(K,L), or more broadly to aggregate across firms an economy-wide production function Y=F(K,L)? Note that the question is not about the definition of capital per se, since defining “labor” is equally problematic when man-hours are clearly heterogeneous, and this question is also not about the more general capital controversy worries, like reswitching (see Samuelson’s champagne example) or the dependence of the return to capital on the distribution of income which, itself, depends on the return to capital.

(A brief aside: on that last worry, why the Cambridge UK types and their modern day followers are so worried about the circularity of the definition of the interest rate, yet so unconcerned about the exact same property of the object we call “wage”, is quite strange to me, since surely if wages equal marginal product, and marginal product in dollars is a function of aggregate demand, and aggregate demand is a function of the budget constraint determined by wages, we are in an identical philosophical situation. I think it’s pretty clear that the focus on “r” rather than “w” is because of the moral implications of capitalists “earning their marginal product” which are less than desirable for people of a certain political persuasion. But I digress; let’s return to more technical concerns.)

It turns out, and this should be fairly well-known, that the conditions under which factors can be aggregated are ridiculously stringent. If we literally want to add up K or L when firms use different production functions, the condition (due to Leontief) is that the marginal rate of substitution between different types of factors in one aggregation, e.g. capital, does not depend on the level of factors not in that aggregation, e.g. labor. Surely this is a condition that rarely holds: how much I want to use, in an example due to Solow, different types of trucks will depend on how much labor I have at hand. A follow-up by Nataf in the 1940s is even more discouraging. Assume every firm uses homogenous labor, every firm uses capital which though homogenous within each firms differs across firms, and every firm has identical constant returns to scale production technology. When can I now write an aggregate production function Y=F(K,L) summing up the capital in each firm K1, K2…? That aggregate function exists if and only if every firm’s production function is additively separable in capital and labor (in which case, the aggregation function is pretty obvious)! Pretty stringent, indeed.

Fisher helps things just a bit in a pair of papers from the 1960s. Essentially, he points out that we don’t want to aggregate for all vectors K and L, but rather we need to remember that production functions measure the maximum output possible when all inputs are used most efficiently. Competitive factor markets guarantee that this assumption will hold in equilibrium. That said, even assuming only one type of labor, efficient factor markets, and a constant returns to scale production function, aggregation is possible if and only if every firm has the same production function Y=F(b(v)K(v),L), where v denotes a given firm and b(v) is a measure of how efficiently capital is employed in that firm. That is, aside from capital efficiency, every firm’s production function must be identical if we want to construct an aggregate production function. This is somewhat better than Nataf’s result, but still seems highly unlikely across a sector (to say nothing of an economy!).

Why, then, do empirical exercises using, say, aggregate Cobb-Douglas seem to give such reasonable parameters, even though the above theoretical results suggest that parameters like “aggregate elasticity of substitution between labor and capital” don’t even exist? That is, when we estimate elasticities or total factor productivities from Y=AK^a*L^b, using some measure of aggregated capital, what are we even estimating? Two things. First, Nelson and Winter in their seminal book generate aggregate date which can almost perfectly be fitted using Cobb-Douglas even though their model is completely evolutionary and does not even involve maximizing behavior by firms, so the existence of a “good fit” alone is, and this should go without saying, not great evidence in support of a model. Second, since ex-post production Y must equal the wage bill plus the capital payments plus profits, Felipe notes that this identity can be algebraically manipulated to Y=AF(K,L) where the form of F depends on the nature of the factor shares. That is, the good fit of Cobb-Douglas or CES can simply reflect an accounting identity even when nothing is known about micro-level elasticities or similar.

So what to do? I am not totally convinced we should throw out aggregate production functions – it surely isn’t a coincidence that Solow residuals for TFP match are estimated to be high in places where our intuition says technological change has been rapid. Because of results like this, it doesn’t strike me that aggregate production functions are measuring arbitrary things. However, if we are using parameters from these functions to do counterfactual analysis, we really ought know better exactly what approximations or assumptions are being baked into the cake, and it doesn’t seem that we are quite there yet. Until we are, a great deal of care should be taken in assigning interpretations to estimates based on aggregate production models. I’d be grateful for any pointers in the comments to recent work on this problem.

Final published version (RePEc IDEAS. The “F. Fisher” on this paper is the former Clark Medal winner and well-known IO economist Franklin Fisher; rare is it to find a nice discussion of capital issues written by someone who is firmly part of the economics mainstream and completely aware of the major theoretical results from “both Cambridges”. Tip of the cap to Cosma Shalizi for pointing out this paper.

“Epistemic Game Theory,” E. Dekel & M. Siniscalchi (2014)

Here is a handbook chapter that is long overdue. The theory of epistemic games concerns a fairly novel justification for solution concepts under strategic uncertainty – that is, situations where what I want to do depends on other people do, and vice versa. We generally analyze these as games, and have a bunch of equilibrium (Nash, subgame perfection, etc.) and nonequilibrium (Nash bargain, rationalizability, etc.) solution concepts. So which should you use? I can think of four classes of justification for a game solution. First, the solution might be stable: if you told each player what to do, no one person (or sometimes group) would want to deviate. Maskin mentions this justification is particularly worthy when it comes to mechanism design. Second, the solution might be the outcome of a dynamic selection process, such as evolution or a particular learning rule. Third, the solution may be justified by certain axiomatic first principles; Shapley value is a good example in this class. The fourth class, however, is the one we most often teach students: a solution concept is good because it is justified by individual behavior assumptions. Nash, for example, is often thought to be justified by “rationality plus correct beliefs”. Backward induction is similarly justified by “common knowledge of rationality at all states.”

Those are informal arguments, however. The epistemic games (or sometimes, “interactive epistemology”) program seeks to formally analyze assumptions about the knowledge and rationality of players and what it implies for behavior. There remain many results we don’t know (for instance, I asked around and could only come up with one paper on the epistemics of coalitional games), but the results proven so far are actually fascinating. Let me give you three: rationality and common belief in rationality implies rationalizable strategies are played, the requirements for Nash are different depending on how players there are, and backward induction is surprisingly difficult to justify on epistemic grounds.

First, rationalizability. Take a game and remove any strictly dominated strategy for each player. Now in the reduced game, remove anything that is strictly dominated. Continue doing this until nothing is left to remove. The remaining strategies for each player are “rationalizable”. If players can hold any belief they want about what potential “types” opponents may be – where a given (Harsanyi) type specifies what an opponent will do – then as long as we are all rational, we all believe the opponents are rational, we all believe the opponents all believe that we all are rational, ad infinitum, the only possible outcomes to the game are the rationalizable ones. Proving this is actually quite complex: if we take as primitive the “hierarchy of beliefs” of each player (what do I believe my opponents will do, what do I believe they believe I will do, and so on), then we need to show that any hierarchy of beliefs can be written down in a type structure, then we need to be careful about how we define “rational” and “common belief” on a type structure, but all of this can be done. Note that many rationalizable strategies are not Nash equilibria.

So what further assumptions do we need to justify Nash? Recall the naive explanation: “rationality plus correct beliefs”. Nash takes us from rationalizability, where play is based on conjectures about opponent’s play, to an equilibrium, where play is based on correct conjectures. But which beliefs need to be correct? With two players and no uncertainty, the result is actually fairly straightforward: if our first order beliefs are (f,g), we mutually believe our first order beliefs are (f,g), and we mutually believe we are rational, then beliefs (f,g) represent a Nash equilibrium. You should notice three things here. First, we only need mutual belief (I know X, and you know I know X), not common belief, in rationality and in our first order beliefs. Second, the result is that our first-order beliefs are that a Nash equilibrium strategy will be played by all players; the result is about beliefs, not actual play. Third, with more than two players, we are clearly going to need assumptions about how my beliefs about our mutual opponent are related to your beliefs; that is, Nash will require more, epistemically, than “basic strategic reasoning”. Knowing these conditions can be quite useful. For instance, Terri Kneeland at UCL has investigated experimentally the extent to which each of the required epistemic conditions are satisfied, which helps us to understand situations in which Nash is harder to justify.

Finally, how about backward induction? Consider a centipede game. The backward induction rationale is that if we reached the final stage, the final player would defect, hence if we are in the second-to-last stage I should see that coming and defect before her, hence if we are in the third-to-last stage she will see that coming and defect before me, and so on. Imagine that, however, player 1 does not defect in the first stage. What am I to infer? Was this a mistake or am I perhaps facing an irrational opponent? Backward induction requires that I never make such an inference, and hence I defect in stage 2.

Here is a better justification for defection in the centipede game, though. If player 1 doesn’t defect in the first stage, then I “try my best” to retain a belief in his rationality. That is, if it is possible for him to have some belief about my actions in the second stage which rationally justified his first stage action, then I must believe that he holds those beliefs. For example, he may believe that I believe he will continue again in the third stage, hence that I will continue in the second stage, hence he will continue in the first stage then plan to defect in the third stage. Given his beliefs about me, his actions in the first stage were rational. But if that plan to defect in stage three were his justification, then I should defect in stage two. He realizes I will make these inferences, hence he will defect in stage 1. That is, the backward induction outcome is justified by forward induction. Now, it can be proven that rationality and common “strong belief in rationality” as loosely explained above, along with a suitably rich type structure for all players, generates a backward induction outcome. But the epistemic justification is completely based on the equivalence between forward and backward induction under those assumptions, not on any epistemic justification for backward induction reasoning per se. I think that’s a fantastic result.

Final version, prepared for the new Handbook of Game Theory. I don’t see a version on RePEc IDEAS.

Dale Mortensen as Micro Theorist

Northwestern’s sole Nobel Laureate in economics, Dale Mortensen, passed overnight; he remained active as a teacher and researcher over the past few years, though I’d be hearing word through the grapevine about his declining health over the past few months. Surely everyone knows Mortensen the macroeconomist for his work on search models in the labor market. There is something odd here, though: Northwestern has really never been known as a hotbed of labor research. To the extent that researchers rely on their coworkers to generate and work through ideas, how exactly did Mortensen became such a productive and influential researcher?

Here’s an interpretation: Mortensen’s critical contribution to economics is as the vector by which important ideas in micro theory entered real world macro; his first well-known paper is literally published in a 1970 book called “Microeconomic Foundations of Employment and Inflation Theory.” Mortensen had the good fortune to be a labor economist working in the 1970s and 1980s at a school with a frankly incredible collection of microeconomic theorists; during those two decades, Myerson, Milgrom, Loury, Schwartz, Kamien, Judd, Matt Jackson, Kalai, Wolinsky, Satterthwaite, Reinganum and many others were associated with Northwestern. And this was a rare condition! Game theory is everywhere today, and pioneers in that field (von Neumann, Nash, Blackwell, etc.) were active in the middle of the century. Nonetheless, by the late 1970s, game theory in the social sciences was close to dead. Paul Samuelson, the great theorist, wrote essentially nothing using game theory between the early 1950s and the 1990s. Quickly scanning the American Economic Review from 1970-1974, I find, at best, one article per year that can be called game-theoretic.

What is the link between Mortensen’s work and developments in microeconomic theory? The essential labor market insight of search models (an insight which predates Mortensen) is that the number of hires and layoffs is substantial even in the depth of a recession. That is, the rise in the unemployment rate cannot simply be because the marginal revenue of the potential workers is always less than the cost, since huge numbers of the unemployed are hired during recessions (as others are fired). Therefore, a model which explains changes in churn rather than changes in the aggregate rate seems qualitatively important if we are to develop policies to address unemployment. This suggests that there might be some use in a model where workers and firms search for each other, perhaps with costs or other frictions. Early models along this line by Mortensen and others were generally one-sided and hence non-strategic: they had the flavor of optimal stopping problems.

Unfortunately, Diamond in a 1971 JET pointed out that Nash equilibrium in two-sided search leads to a conclusion that all workers are paid their reservation wage: all employers pay the reservation wage, workers believe this to be true hence do not engage in costly search to switch jobs, hence the belief is accurate and nobody can profitably deviate. Getting around the “Diamond Paradox” involved enriching the model of who searches when and the extent to which old offers can be recovered; Mortensen’s work with Burdett is a nice example. One also might ask whether laissez faire search is efficient or not: given the contemporaneous work of micro theorists like Glenn Loury on mathematically similar problems like the patent race, you might imagine that efficient search is unlikely.

Beyond the efficiency of matches themselves is the question of how to split surplus. Consider a labor market. In the absence of search frictions, Shapley (first with Gale, later with Shubik) had shown in the 1960s and early 1970s the existence of stable two-sided matches even when “wages” are included. It turns out these stable matches are tightly linked to the cooperative idea of a core. But what if this matching is dynamic? Firms and workers meet with some probability over time. A match generates surplus. Who gets this surplus? Surely you might imagine that the firm should have to pay a higher wage (more of the surplus) to workers who expect to get good future offers if they do not accept the job today. Now we have something that sounds familiar from non-cooperative game theory: wage is based on the endogenous outside options of the two parties. It turns out that noncooperative game theory had very little to say about bargaining until Rubinstein’s famous bargaining game in 1982 and the powerful extensions by Wolinsky and his coauthors. Mortensen’s dynamic search models were a natural fit for those theoretic developments.

I imagine that when people hear “microfoundations”, they have in mind esoteric calibrated rational expectations models. But microfoundations in the style of Mortensen’s work is much more straightforward: we simply cannot understand even the qualitative nature of counterfactual policy in the absence of models that account for strategic behavior. And thus the role for even high micro theory, which investigates the nature of uniqueness of strategic outcomes (game theory) and the potential for a planner to improve welfare through alternative rules (mechanism design). Powerful tools indeed, and well used by Mortensen.

“Price Formation of Fish,” A.P Barten & L.J. Bettendorf (1989)

I came across this nice piece of IO in a recent methodological book by John Sutton, which I hope to cover soon. Sutton recalls Lionel Robbins’ famous Essay on the Nature of Significance of Economic Science. In that essay, Robbins claims the goal of the empirically-minded economist is to estimate stable (what we now call “structural”) parameters whose stability we know a priori from theory. (As an aside, it is tragic that Hurwicz’ 1962 “On the Structural Form of Interdependent Systems”, from which Robbins’ idea gets its modern treatment, is not freely available online; all I see is a snippet from the conference volume it appeared at here). Robbins gives the example of an empiricist trying to estimate the demand for haddock by measuring prices and quantities each day, controlling for weather and the like, and claiming that the average elasticity has some long-run meaning; this, he says, is a fool’s errand.

Sutton points out how interesting that example is: if anything, fish are an easy good to examine! They are a good with easy-to-define technical characteristics sold in competitive wholesale markets. Barten and Bettendorf point out another interesting property: fish are best described by an inverse demand system, where consumers determine the price paid as a function of the quantity of fish in the market rather than vice versa, since quantity in the short run is essentially fixed. To the theorist, there is no difference between demand and inverse demand, but to the empiricist, that little error term must be added to the exogenous variables if we are to handle statistical variation correctly. Any IO economist worth their salt knows how to estimate common demand systems like AIDS, but how should we interpret parameters in inverse demand systems?

Recall that, in theory, Marshallian demand is a homogeneous of degree zero function of total expenditures and prices. Using the homogeneity, we have that the vector quantity demand q is a function of P, the fraction of total expenditure paid for each unit of each good. Inverting that function gives P as a function of q. Since inverse demand is the result of a first-order condition from utility maximization, we can restate P as a function of marginal utilities and quantities. Taking the derivative of P, with some judicious algebra, one can state the (normalized) inverse demand as the sum of moves along an indifference surface and moves across indifference surfaces; in particular, dP=gP’dq+Gdq, where g is a scalar and G is an analogue of the Slutsky matrix for inverse demand, symmetric and negative semidefinite. All we need to do know is to difference our data and estimate that system (although the authors do a bit more judicious algebra to simplify the computational estimation).

One more subtle step is required. When we estimate an inverse demand system, we may wish to know how substitutable or complementary any two goods are. Further, we want such an estimate to be invariant to arbitrary monotone increasing changes in an underlying utility function (the form of which is not assumed here). It turns out that Allais (in his 1943 text on “pure economics” which, as far as I know, is yet to be translated!) has shown how to construct just such a measure. Yet another win for theory, and for Robbins’ intuition: it is hopeless to atheoretically estimate cross-price elasticities or similar measures of substitutability atheoretically, since these parameters are determined simultaneously. It is only as a result of theory (here, nothing more than “demand comes from utility maximizers” is used) that we can even hope to tease out underlying parameters like these elasticities. The huge numbers of “reduced-form” economists these days who do not understand what the problem is here really need to read through papers of this type; atheoretical training is, in my view, a serious danger to the grand progress made by economics since Haavelmo and Samuelson.

It is the methodology that is important here; the actual estimates are secondary. But let’s state them anyway: the fish sold in the Belgian markets are quite own-price elastic, have elasticities that are consistent with demand-maximizing consumers, and have patterns of cross-price elasticities across fish varieties that are qualitatively reasonable (bottom-feeders are highly substitutable with each other, etc.) and fairly constant across a period of two decades.

Final version in EER (No IDEAS version). This paper was in the European Economic Review, an Elsevier journal that is quickly being killed off since the European Economic Association pulled out of their association with Elsevier to run their own journal, the JEEA. The editors of the main journal in environmental economics have recently made the same type of switch, and of course, a group of eminent theorist made a similar exit when Theoretical Economics began. Jeff Ely has recently described how TE came about; that example makes it quite clear that journals are actually quite inexpensive to run. Even though we economists are lucky to have nearly 100% “green” open access, where preprints are self-archived by authors, we still have lots of work to do to get to a properly ungated world. The Econometric Society, for example, spends about $900,000 for all of its activities aside from physically printing journals, a cost that could still be recouped in an open access world. Much of that is for running conferences, giving honoraria, etc, but let us be very conservative and estimate no income is received aside from subscriptions to its three journals, including archives. This suggests that a complete open access journal and archives for the 50 most important journals in the field requires, very conservatively, revenue of $15 million per year, and probably much less. This seems a much more effective use of NSF and EU moneys that funding a few more graduate research assistants.