Believe those who are seeking the truth. Doubt those who find it. Andre Gide

Tuesday, June 14, 2016

DSGE Theory

This post is for my students, and whoever else is interested in what DSGE theory is and why I find it useful.

Dynamic Stochastic General Equilibrium (DSGE) theory refers to a methodology employed by macroeconomists to build DSGE models -- mathematical representations of the macroeconomy. DSGE models, like all models, are used for a variety of purposes. They are used to help organize thinking. They are used to interpret data. They are used to help make conditional forecasts. They are used to predict and evaluate the possible consequences of government policies (especially useful for policies that have never been tried before). They are used to help make policy recommendations.

The use of DSGE theory is often criticized in ways that reflect what I view as a deep misunderstanding of the research program, how it fits in with the evolution of macroeconomic theory over time, and how it is actually applied by (say) central bank policy makers. This is, I think, to some extent the fault of DSGE practitioners who, accustomed to speaking in their specialized trade language, find it difficult to translate core ideas and findings in the vernacular. (This is an issue with most trade associations, of course, but is especially acute in economics because so many non-specialists take an interest in the subject.)

Let me first provide some context for my views. We are all scientists trying to understand the world around us. We use our eyes, ears and other senses to collect data, both qualitative and quantitative. We need some way to interpret/explain this data and, for this purpose, we construct theories (or hypotheses, or models, or whatever term you prefer). Mostly, these theories exist in our brains as informal "half-baked" constructs. This is not meant to be a criticism (as long as we recognize the half-baked nature of our ideas and why some humility is always in order). Often it seems we are not even aware of the implicit assumptions that are necessary to render our views valid. Ideally, we may possess a degree of higher-order awareness--e.g., as when we're aware that we may not be aware of all the assumptions we are making. It's a tricky business. Things are not always a simple as they seem. And to help organize our thinking, it is often useful to construct mathematical representations of our theories--not as a substitute, but as a complement to the other tools in our tool kit (like basic intuition). This is a useful exercise if for no other reason than it forces us to make our assumptions explicit, at least, for a particular thought experiment. We want to make the theory transparent (at least, for those who speak the trade language) and therefore easy to criticize. Constructive criticism is the fuel that fires the furnace of new ideas in academia. [ End of philosophical rant :) ]

Now let me turn back to DSGE theory. I think it will be useful to break the acronym into its parts and discuss each component separately.

The "D" stands for dynamic--as in--the phenomena in question involve a time element. The opposite of dynamic is static. While static models have their uses, who's going to argue that a dynamic element isn't desirable? Almost all decisions like consumption and saving, deficit-finance, human capital investments, have a time dimension to them. No controversy here, I hope.

The "S" stands for stochastic--as in--societies appear subject to random events, like unforeseen technological breakthroughs, unexpected changes in government policy regimes, or just random acts of nature. Again, I don't think there's much controversy with this idea. Note, however, many DSGE models do not have the S, in which case we might instead employ the acronym DGE. (For a history of the evolution of these acronyms, see here.)

The "G" stands for general--as in--well, it's not entirely clear. There is a traditional distinction in economics between partial and general equilibrium theory. The partial equilibrium approach (associated with Alfred Marshall) refers to the supply-demand curve analysis that most people are familiar with. The analysis is "partial" in the sense that it typically restricts attention to a particular market--like the market for motor vehicles, taking the price of other goods as given. In contrast, the general equilibrium approach (associated with Leon Walras) strives to model the economy as a closed system, paying particular attention to how markets interact with each other and how prices are determined jointly. Importantly, the "G" insists on giving an explicit account of the government budget constraint (i.e., a government is not to be modeled as Jesus feeding the multitude.) Another way to think about "G" is that it means to capture the possibility of "feedback effects." The notion of feedback effects in macroeconomic systems is not, I do not think, controversial.

This leaves us with the "E," which stands for equilibrium. Here lies the controversy. But why? For all sorts of reasons, some of which are based on legitimate concerns, and some of which are based on simple misunderstanding.

Let me first address the misunderstanding. The concept of "equilibrium" in economics has evolved to mean something quite specific and something quite different from the notion of a "system at rest" (which is closer to what economists label a steady-state). Technically, an equilibrium is simply a set of conditions imposed by the theorist to help determine the outcome of an hypothetical social interaction. In this sense, an equilibrium is probably better thought of as a solution concept. There is no unique way to specify an equilibrium solution concept. In the game theory, there is plethora of alternatives, beginning with the Nash equilibrium. The classical theory of Walras uses the concept of a competitive equilibrium. In my own view (probably not representative), I even think of general disequilibrium as just another type of equilibrium concept. Every theorist has to have a solution concept in mind when deducing the likely outcome of an hypothetical social interaction. There is no right or wrong way to specify an equilibrium concept--there are just more or less useful ways in doing so.

Another misunderstanding is that insisting on equilibrium analysis necessarily implies that one assumes markets always "clear" in the sense prices adjust to ensure supply equals demand at all times. This is understandable because many DSGE models (especially the RBC variety) do in fact make this assumption. But, of course, there's a large class of DSGE models that do not (e.g., the NK variety). More to the point, it's important to understand that the concept of equilibrium is not wedded to the concept of competitive market-clearing models. In DSGE models that replace centralized Walrasian markets with decentralized search markets, conventional "supply and demand" curves do not even exist. In search models, prices are determined through bilateral negotiations and the "clearing" mechanism operates through quantity variables, like labor-market tightness (the ratio of vacancies to unemployment).

A more legitimate concern relates to the equilibrium concept of "rational expectations." Because of the "D" element, the theorist must take a stand on how expectations are formed and updated over time. Macroeconomic theorists have grappled with this question for over a century, if not longer (see Laider, 1999). There is little controversy that people are forward-looking. But exactly how are they forward-looking? John Muth (1961) suggested that, in the context of a model, we might begin by assuming that our modeled agents (somehow) form model-consistent expectations (i.e., "rational" expectations). Intuitively, the idea is that we should not model people as forming expectations that are wildly at odds with the reality unfolding around them and, that as a limiting case, we might even begin by assuming that expectations are formed in a manner that is perfectly consistent with the surrounding reality. Among other things, model agents are assumed to possess common knowledge (see, Geanakoplos, 1992).

Now, if all of this sounds like a bit of a stretch, it no doubt is. The relevant criticism and response is recorded in section 6.4 Stationary Models and the Neglect of Learning in Lucas and Sargent (1979). I'm not going to get into it here, but suffice it to say that there's been a large and vibrant literature on non-rational-expectations "learning" models since Lucas and Sargent wrote that piece. And you'd be very wrong to think it hasn't had any influence in the way policymakers, central bankers in particular, think about policy and its effects. St. Louis Fed president James Bullard, for example, is among those who have made significant academic contributions in this area (you can view his works here).

In terms of their use in policy making, DSGE models are no different than their predecessors. Some applications entail large scale quantitative models to make conditional forecasts. But their main value is the manner in which they (along with other models) are used to organize thinking in policy deliberations. I think I disagree with Narayana Kocherlakota here when he suggests that DSGE models are built purposely not be useful for day-to-day policy making--for example, in helping to answer the question of whether the interest rate should be changed in the upcoming FOMC meeting. Instead, he views DSGE models as useful for thinking about policy rules (which I agree with). But his view here seems inconsistent with a view he has expressed elsewhere, namely, that isolated changes in the policy rate are largely irrelevant--that what is important is how the path of interest rates is expected to evolve over time (I agree with this too). I think that the decision of whether to move rates today has to be made in the context of what the policymaker views as wise policy principles based on some combination of theory, evidence, and experience. These principles should no doubt make allowances for the necessity of discretionary and ad hoc policy actions. But this allowance does not mean that reference to a DSGE model (or any other model) cannot be useful for thinking through the likely consequences of a contemporaneous policy action. [Note: I may have misunderstood the point NK was trying to make.]

In terms of a defense of the use of DSGE theory for policy, I can do no better than Chris Sims here (video, highly recommended). See also this interview with Tom Sargent, who defends modern macro theory. Finally, I have my own related post: In Defense of Modern Macro Theory.


  1. With your description of "Equilibrium" above, it's possible that equilibrium in DSGE is unrelated to equilibrium as discussed in science. This probably leads to scientists asking questions that are incomprehensible to the DSGE practicing economist:

    Is the equilibrium stable or unstable?

    Are there multiple equilibria? Is there an actual equilibrium?

    Is the world actually in an equilibrium state? If not, how is it behaving?

    Something Keynes said is relevant: oceanography goes far beyond noting that the usual (long-term, perhaps) state of the ocean is calm. He has to study waves and storms, and understand them. Other people have to understand how to deal with storms to minimize danger to people and minimize or repair damage to structures.

    One last thing: it appears that with your discussion of "equilibrium" that there is no basic theory or model, experimentally or observationally verified (even in approximation) that has developed a consensus in economics. One is free to introduce whatever equations one chooses, and they're all considered valid.

    1. My understanding is that in science, "equilibrium" refers to the rest point of a dynamic system. In economics, this is called a "steady state."

      Because an equilibrium concept leads to a system of dynamic equations, we can ask all the usual questions about stability, etc. See my notes here, for example:

      Note that the "S" is operative, equilibrium is defined in terms of a fixed point in a set of stochastic processes.

      The question of whether a particular equilibrium is "learnable" or not, by the way, is a completely separate issue.

      With respect to your concluding statement, that's basically correct, except that "validity" is usually restricted to mean "logically consistent." If the theory is then this is satisfying at an intellectual level, no matter how crazy it seems. But then it has to be tested against observation, either formally or informally. Informal tests include the "smell test."

    2. Science equilibrium: static vs. dynamic equilibrium.

      Example of static equilibrium: total force on a stationary object is zero, and total torque on a non-rotating object is zero.

      Example of dynamic equilibrium: a chemical reaction is progressing both ways at an equal rate, so that the total overall amounts of each chemical is constant. It's also probably stochastic, in the sense that individual molecules break up and recombine at random, but the huge numbers usually render the random aspect negligible.

      Another dynamic equilibrium example: hold volume constant. Internal energy (and hence temperature) is constant if energy in equals energy out. (Again, things occur microscopically at random.)

      How do these compare with "equilibrium" in DSGE? Perhaps these are "steady state"? Actually, related to the last example, "steady state" may refer to cyclic cases, where we should be at equilibrium on average. (Being slightly off average is bringing us global warming.)

    3. John, please refer to the notes I linked to.

      You will find there a linear first-order difference equation in the unemployment rate. For a given job separation rate and for a given job finding rate, we can deduce the dynamics very easily. There is a unique steady-state (in one version of the model) where the job destruction rate just offsets the job creation rate. Call this the steady-state unemployment rate.

      Is the steady-state unemployment rate above an "equilibrium?" Well, it depends. Consider a theory that explains the job finding rate as the outcome of the strategic recruiting intensity of firms. Their choices depend on all sorts of things including (possibly) the contemporaneous unemployment rate (since the unemployment rate may influence the wage bargain, which then influences the attractiveness of creating jobs, etc.). We can impose a solution concept here that includes, among other things, the condition that vacancies are created to the point that profits are driven to zero. Imposing this allows us to deduce the entire sequence of recruiting activity over time which, when combined with the first-difference equation above, determines the time path of unemployment. The whole thing all together is called an 'equilibrium' outcome.

      Not sure how much of this makes sense, but let me know.

    4. Okay, I think that I got the gist of the paper, although it's hard for me without actually getting out pencil and paper and doing the work.

      Equation (2) had a typographical error, although the error didn't propagate into the rest of the analysis. (Comparing eqs. (1) and (3) confirms what eqn. (2) should have been.)

      I'm not at all sure if highly relevant issues can be incorporated in this kind of analysis. I'm thinking of a couple situations: y (the value produced by a worker) for (say) the local Burger King restaurant (not the entire corporation) may be a value for each of (N-1) workers, a fraction of that for the Nth worker, and zero for further workers. (I call this the downsizing model, because similar thinking is behind downsizing. If you can distribute a worker's work among the others, his value for you is zero.)

      Then y must also depend on the demand for the goods produced. (If demand increases, you will try to produce more. But you don't want to produce what will be spoiled. And when demand drops, you not only will produce less, you might also want to reduce your warehouse of goods.) That would depend on wages (and debt one gets and debt one pays off). There's the Prisoner's Dilemma issue, in that demand for your goods depends on the wages of many other companies, and minimally on the wages you pay.

      Thanks for the response and the link.

  2. I'm now even more confused about equilibrium than I was before. Can you explain what you mean by "solution concept"?

    1. Draw a supply curve and a demand curve. These are both *theoretical* restrictions. One describes optimal behavior of sellers, the other describes optimal behavior of buyers. What is the outcome of this interaction?

      To answer this last question, we need another restriction. One such *theoretical* restriction is that S = D. But its just a restriction--the concept of eqm is a list of restrictions.

      1. Sellers are optimizing.
      2. Buyers are optimizing.
      3. S = D

      With this solution concept, we can use the theory to make a prediction.

      Does this help?

    2. "To answer this last question, we need another restriction. "

      No, we don't. The only solution that satisfies both supply and demand curves is when D = S.

    3. Well, yes, that's the only solution that satisfies a competitive equilibrium with flexible prices.

      But suppose I replace [3] with an alternative restriction, for example, [3'] that the price is administered by a government agency (think of an interest rate, or a minimum wage, etc.). Then conditions [1], [2] and [3'] imply something different.

    4. A third condition is overdetermining in the presence of [1] and [2], unless you mean that [3] can change the supply or demand curve.

      My training was in mathematics, where the matter is simple: a solution is a system that satisfies the restrictions, period. You seem to be implying that in Economics one determines the restrictions simultaneously with determining the solution. That seems to me like playing tennis without a net.

      I'm still completely mystified by the concept of "solution concept". Is it really anything other than a way for DSGE theorists to get any answer they want?

    5. Yes, the administered price and the short side of the market (in this case, the labor demand curve if the price is an administered wage) determines price and quantity. The supply curve is then used to determine the desired supply (an excess supply, in this case). Here is the diagram:

      Also note that in the case of a competitive equilibrium, there is no notion built into the theory of how the economy actually gets there. People can tell ad hoc (and plausible) stories based on a tatonnement process (look it up). But these are just stories that exist outside the theory.

      Let's try another example. I presume you are familiar with the prisoner's dilemma game. There are 4 possible outcomes. Which outcome is likely to transpire? We cannot answer this question without imposing a solution concept like Nash equilibrium. The unique Nash equilibrium of a one-time PD game is to rat out your partner. But there is no God-given reason to impose Nash.

    6. "Also note that in the case of a competitive equilibrium, there is no notion built into the theory of how the economy actually gets there."

      Yes, the D in DSGE. But your example didn't address this.

      "We cannot answer this question without imposing a solution concept like Nash equilibrium."

      No: we cannot get a *unique* answer to this question without making further assumptions. Is that what a "solution concept" does, impose extra (maybe ad hoc or even tendentious) conditions in order to guarantee a unique solution?

      Perhaps Economics is trying too hard to emulate mathematics, which studies much, much simpler systems. Differential equations with boundary or initial conditions will have unique solutions, but the world economy is much too complicated for anyone to imagine being able to characterize it uniquely. Economists would be better off facing this squarely, and admitting the best they can do is enumerate possibilities, rather than make predictions.

      On the other hand, DSGE as I understand it (which is not at all well) may be an exploration of all these possibilities, with the sole problem that individual researchers tend to take their own systems too seriously.

    7. A solution concept imposes restrictions on what the answer (the outcome) needs to satisfy. Sometimes there is a unique answer. Sometimes there are multiple answers. And sometimes, there is no answer at all (non-existence of equilibrium).

      I'm not sure you're in a position to lecture economists on how to best practice their trade. What would be welcome instead is for you to give it a try and show us how it can be done better! :)

    8. Point taken. But there's no need to introduce the new notion of "solution concept" if all you want to do is impose restrictions. Can't you agree that "solution conecpt" was a failed imrpovisation?

    9. Well, let's be clear here: It's not my improvisation. It's standard to first define an equilibrium (a list of restrictions) before describing the properties of the solution. I linked to this:

      I find the concept useful. You evidently do not. We can leave it here. Cheers.

    10. My apologies, I missed your link to wikipedia. But from that article it seems that "solution concept" is the term used in economics for what mathematicians would generally call a constraint. It introduces nothing novel into the discussion.

      So when you say

      "Technically, an equilibrium is simply a set of conditions imposed by the theorist to help determine the outcome of an hypothetical social interaction. In this sense, an equilibrium is probably better thought of as a solution concept. "

      I now take you to mean that in the context of DSGE "equilibrium" has completely lost it's ordinary meaning and here is just a synonym for "constraint". That's needlessly confusing terminology, but coherent.

  3. I worked on DSGEs in my job for a provincial government. Here’s my take.

    That’s a good defence of DSGEs as they should be, but there is one thing you don’t mention. My and many people’s complaint is with the accumulation of ad hoc assumptions. Fine if you absolutely need investment adjustment costs, fine if you need habit formation, but the accumulation of these makes the results meaningless. The models become no better than some curve fitting machine; their sheer size precludes it from answering both policy questions and making short- or medium-term projection (long-term projection comes from the steady state which is a different model in itself).

    1. I just reread my post. My critique is not aimed at abandoning the project, your post pretty much sums up what I believe is good about it. It's more of a critique of the direction of research. Instead of adding stuff to the model, we should try to find ways to reduce the number of stuff in it. There is no way that (I'm citing ToTEM here) "investment adjustment cost + habit formation + variable capital utilization + hand-to-mouth consumers + Calvo on everything" is a description of anything useful. Some of these assumptions are only there to repair what other assumptions broke.

    2. Gilles, absolutely. But what you say is true of any theory, even the half-baked ones in our heads. A tool can be crafted to be useful for specific purposes. When the tool is modified to the point of being unwieldy, like a Swiss Army knife, it becomes less useful some purposes.

      In fairness though, they're looking for ways to make quantitative projections. That's a different ball game.

  4. This is a well-argued piece. To me, one of the most important points you make here is that we frequently make assumptions without realising we are doing so. Do you think that the dominance of DSGE has a tendency to shape the way economists tend to look at things and the questions they seek to answer? Whilst it is of course entirely possible to consider issues relating to balance sheet developments within the context of DSGE models, I don't think there is a natural fit there and it must be tempting to assume away such complications for most purposes. Could this perhaps have meant that economists paid less attention to such issues pre-crisis than they should have?

    1. As I mentioned in my "Defense of Macro" post, there was a ton of work being done on financial frictions prior to the financial crisis. (Note that even Sims points this out.)

      The question is why were central bankers not paying attention to that literature? I don't think it was because they were not aware of it. I think they thought, based on the evidence of the great moderation, that these types of concerns were not pertinent for well-developed economies. It nothing to do with the use of DSGE modeling per se.

  5. David,

    "... there's a large class of DSGE models that do not (e.g., the NK variety)."

    " In DSGE models that replace centralized Walrasian markets with decentralized search markets, conventional "supply and demand" curves do not even exist. "

    Having written the above, how is it that these models can continue to be called "general equilibrium" models?


    1. "General" because they are closed systems, and government budget constraint must hold (there is no manna from heaven).

      "Equilibrium" because they rely on a well defined solution concept, like "competitive search eqm," etc. See:

    2. When you say "no manna from heaven", I presume you mean "conservation" of income, money and assets?

      So if all these models are closed, all agents are have a utility function and face a budget constraint. Does this also apply to the NK models where sticky prices/wages feature? It seems to me that sticky prices/wages can be accommodated within the general equilibrium framework via the effect of stickiness on agents' budget constraint - allowing equilibrium conditions to pertain. In the case of the labour market, this would mean there is only unemployment of the voluntary kind (putting aside the effect of frictions) - i.e. no involuntary unemployment. Is this an accurate characterization of these models?

      In the competitive search model referenced, can you explain how the work/leisure trade off evident in normal GE models is realized? And given the CSE is optimal, there is only voluntary unemployment. Is that correct?


    3. Yes, that's what I mean by "no manna from heaven."

      Yes, the closed property is present in a standard (closed economy) NK model.

      I do not like the labels "voluntary" and "unvoluntary" unemployment; see here:

      I do not understand your question about work/leisure trade off.

      Note: CSE results in optimal allocation in *that* model. In other search models with bargaining, the so-called Hosios condition (Google it) is necessary for optimality. But equilibria are generically suboptimal.

    4. "No manna from heaven." "Conservation of income, money, and assets."

      1. Combine this with population growth. What do you get?

      2. Eliminate the assumption of conservation of money. Suppose that the government is a currency-creator. (Even better: assume that tax revenues all go into the bit bucket, and that the government including the Fed, pays by increasing numbers in an account.) Of course, you have to assume a choice for how much government creates -- or make that something you solve for.

    5. "First, suppose there is only one job and three workers. Are you kidding me? What world are you describing? The situation is the exact opposite. It is as Alchian is reported to have said (according to Jim Rose below):

      "`Alchian said there are always plenty of jobs because to suppose the contrary suggests that scarcity has been abolished.'

      "So, first off, let's get serious. There are always more things to be done than there are people to do them. Throughout history it is *leisure* that has always been wanting, not work."

      Excuse me? "One job but three searchers" might be a reasonable model illustrating certain circumstances.

      Scarcity has been abolished if there aren't plenty of jobs? Where the heck did that idea come from? My guess: substituting labor for whatever else is scarce; a special case of the infinite substitutibility fallacy. And of course, infinite substitutibility means scarcity has been abolished.

      Often, there are plenty of things to be done, but the people in the position to decide to have them done don't. (This nation's infrastructure for example. Fix the damn bridges! Fix the main water lines! Upgrade our electric grid! Then there's education.)

      Come to think of it, certain neoclassic economic theories assume perfect information and perfect rationality. Therefore, education doesn't exist.

  6. "I do not like the labels "voluntary" and "unvoluntary" unemployment"

    Yes, I understand that you would not like the distinction - I will read your referenced blog later.

    Regarding the work/leisure trade off, I apologize for my lack of clarity. Is it not the case in the standard GE model that the labour supply function (and ultimately the equilibrium real wage) is derived in part from the worker consumption-leisure/work utility function? Having scanned and selectively read Moen's paper, I can see no similar formulation. I am wondering where the worker's work/leisure utility function fits into the model, if at all.


    1. There is no labor-leisure trade-off in the Moen model. Workers either work a fixed number of hours, or not at all. I have a version of the search model where the firm and worker pair can also choose how many hours to work; see

      But the hours choice is determined through a bilateral negotiation--there is no conventional labor supply and labor demand curve. The wage rate is determined by relative bargaining power, among other factors, including the worker's utility for consumption and leisure.

  7. "I do not like the labels "voluntary" and "unvoluntary" unemployment;.... "

    Having read your blog on this matter, it was pretty much what I expected you to say. I am sure nothing I could say would make a difference.

    However, I will say that after 40 years of analysing and studying business and economic conditions, many of those years in a professional capacity, I cannot understand how any one could argue that involuntary unemployment does not exist. In those 40 years I have witnessed 3 serious recessions and of course the slump of 2008. Generally there is a shock to the system which initiates a sequence of events. More often than not it is an "economic manager" deciding an economy is overheated and needs reigning in. Interest rates are pushed up, finance is difficult to find, businesses sailing close to the wind because of high debt close up - and there begins the start of employment retrenchment. There is no question of workers wages being too high. The business is busted and the jobs evaporate. This unemployment causes a fall in consumption. This loss of spending causes businesses to cutback production and re-evaluate investment opportunities - the retrenchment rolls on. Consumer and business confidence and expectations collapse. And on top of all this, new participants enter the labour market through natural growth. None of this has anything to do with wage levels. Jobs just aren't available at any wage. This is involuntary unemployment.


    1. You did not understand my post, evidently. The circumstances that drive people to unemployment are involuntary, but the actual choice itself is generally voluntary (for example, there is almost always an option to drop out of the labor force, as discouraged workers do). If we start attaching labels like this, then pretty soon we see a lot of choices are "involuntary"...having to stay home with the kids, going to work, etc. This is just not a useful path.

      Btw, I've had my own experience to draw on as well. I was a journeyman drywaller/taper for many years and lost my job in the 1981 recession. That experience fits in perfectly well with the interpretation I offer above and in my post.

    2. David,

      I think I do understand your post - what I don't understand is why anyone would believe there is no such thing as involuntary unemployment. When jobs evaporate and the labour force grows relentlessly, there is no choice, there are no jobs. Reducing wages doesn't cure the problem, it adds to the problem through income effects.


  8. David,

    Here's what I don't understand. You lost your job in 1981. I presume you lost it not because you were overpaid and you were unwilling to reduce your wage to keep your job. I presume you lost your job because your employer no longer had contracts to fulfill sufficient to keep you employed. I presume you wanted to work had your employer had contracts to fulfill. How is your retrenchment not involuntary on your part? This is where you begin to twist heavily the logic of your explanation. You argue that because a worker has the choice, once retrenched, to either find other work or drop out, that his retrenchment is in some way voluntary. He may have no choice after his retrenchment if there is general retrenchment, as happens in recessions, and jobs disappear, just as yours did. Your logic is crazy and twisted, designed to allow you to deny the possibility of involuntary unemployment.


    1. Here's what you don't understand.

      I was working at a union job, making a decent hourly wage. At the same time I (along with most of my coworkers) would perform the occasional "side job." These side jobs were usually residential renos (basements, upgrades, etc.). Nobody laid me off of my side job employment. I could have worked as long as I wanted, had I been willing to price my services lower (we would use piece rate to price side jobs, so wage was highly flexible--alarmingly flexible in fact).

      So you see, my own experience has shown me that there's always work to be done. That doesn't mean it's worthwhile doing it. I didn't think it was at the time, which is why I went back to school (another voluntary choice).

      So, in conclusion, I have no idea why you call my logic crazy. It's not crazy. I lived it. And it makes perfect sense, to me, at least! :)

  9. David,

    A few comments up you said "I....lost my job in the 1981 recession." That sounded like you were retrenched because of the recession.

    I've seen plenty of businesses close or reduce output during recessions, laying off workers in the process. I don't think the workers that were laid off would say their retrenchment was voluntary. They were paid off (if they were lucky) and shown the gate.

    And thinking about this a little more, I would even argue that frictional unemployment could have an involuntary component. The fact that someone is looking for a job begs the question - how is it they are unemployed in the first place?

    1. Yes, I lost my union job during the recession. That doesn't mean I lost every opportunity I might have before me to work.

      I'm afraid you do not understand my post. I'll repeat: the circumstances that lead to unemployment may be involuntary, but the choice of how to allocate time across competing uses is not. That's all. I don't want to make too much of it. I just wanted you to understand an alternative perspective. Thanks.

    2. Maybe the concern here is that if all unemployment is considered voluntary, then under what economics statistic does the human suffering from worker redundancy fall?

    3. If you read my post, you'll see that just because unemployment (or employment, for that matter) is "voluntary," it doesn't mean that people are not suffering. The *circumstances* that drive people to make certain choices *are* involuntary.

    4. But I repeat: where does it show up in the model? The concern is that anything not in model is ignored.

    5. First off, note that individual welfare does NOT show up in models like IS/LM. But it does show up in DSGE models because this class of models makes explicit reference to the preferences (or objective function) of individuals. A part of the model's predictions depend on the properties of these preferences. AND, we can use the very same preferences to ascertain whether individuals are made better or worse off in different circumstances.

      You might want to have a look here:

    6. "If you read my post, you'll see that just because unemployment (or employment, for that matter) is "voluntary," it doesn't mean that people are not suffering."

      Isn't that a strange thing to say? In a GE model, frictions aside, equilibrium is the optimal point. All economic agents have maximized their objective functions relative to their budget constraints. There can be no suffering as all action is given by choice. If the p/w ratio changes, the worker's budget constraint moves to set up a new equilibrium and a new work/leisure position on his indifference curve.

      Everybody is happy!

      On the other, involuntary unemployment has necessarily no such connotations.


    7. "The *circumstances* that drive people to make certain choices *are* involuntary."

      And this is a strange thing to say also. What choice does a worker who has been laid off by a bankrupt business have in the matter? Absolutely none. He may have options regarding finding new work or not. If he wants to continue to work and the general condition is one of recession, I will assert he has effectively no choice. I know you will say I do not understand your perspective, but I believe I do. (And here we go around the mulberry bush again. :-) )


  10. How can any economic analysis using a Macroeconomic Paradigm (i.e. the economy is one big market) achieve anything but "partial" equilibrium since it leaves so much activity outside of its view? Of course this is a rhetorical question. It can't. You need a workable paradigm with the market placed along side other factors such as government (at the several levels of its existence) that may participated in but are detached from markets. The decisions here are not based on supply and demand the market criterion.

    1. Show us how it should be done. We're listening.

    2. Showing how it should be done is a big, complicated project, but I can give a soundbite description of the approach that I'd take: build a brute-force simulation of the economy, with a hybrid knowledge-based and data-based architecture.

      This of course assumes that your goal is a scientific one, namely to build a model with external, empirical, predictive validity rather than play games with elegant math where the term "validity" has been distorted to mean internally self-consistent. Meteorologists realized 40 years ago that their elegant six-parameter model of the atmosphere didn't have a stable equilibrium, and was in fact subject to chaotic dynamics. Their response wasn't to stop studying real weather and try to defend simpler models that have equilibrium solutions as the proper subject for meteorology.

      Now, real economies are far more complicated than atmospheric dynamics, but we've learned a lot about how to build complex simulations since the Simula programming language was invented in the mid-1960s. There is important computer science research to be done at the interface between knowledge-based and data-based systems, but you can get a fine start with tools like scipy Python.

      Albert Einstein is said to have stated that "theories should be as simple as possible, but no simpler". It's time for economics to stop oversimplifying in the name of tractability. Computation is cheap, use it!

    3. I have nothing against the use of large scale models. There have been attempts to do so in the past, unfortunately, with limited success. Does not mean we shouldn't continue to try though.

      One problem with large scale models, however, is that that it seems very hard to know why things happen. This may not be so bad if prediction is all one wants. But small scale models can often isolate a particular force that permits us to understand what is going on. The idea here is that by examining the properties of small scale models, we can build some intuition for the way large scale models (including reality) work. Using one tool does not preclude using others.

    4. Building small scale, restricted models and assuming that they will work at full scale is an effective strategy only if the small models have some critical properties: compositionality, associativity, and transitivity for starters. You have no reason to believe that this is so, and good reasons to believe that there are important nonlinear interaction functions. For example, disputes over the existence of "involuntary unemployment" are a definitional artifact that only arises if the model assumes that people who are not in the workforce don't contribute to demand.

      If you're fascinated by math, you can look to modern physics (at a distance, not in close detail, and in journals like Physica D, not Physical Review Letters), which is full of notions like commutators that separate the nonlinear parts of the model from the linear parts, and structures like fibre bundles and categorical functors that allow you to prove theorems about how different subtheories fit together. I happen to think that economics is so complicated that capturing an economy in a single expression will make Einstein summation notation look like child's play. (The are are deep reasons to think that it's not possible even in principle.) Sean Carroll's Lagrangian with the core theory of everything on a T-shirt is not for economists, any more than it is for ecologists.

      Maybe the reason that it's so hard to understand why things happen in economies is because, like in biology, everything is connected to everything else at multiple levels of scale. If you're right, and small, simple, modular models provide actual insights into behavior of the whole system, that would be enormously valuable. But if you don't test that assumption by putting the small pieces together into a big model and testing against the reality, your insights are only insights about math, and not insights about the world. The appropriate phrase for those kinds of insights is "not even wrong". Building full-scale, full complexity models is not an activity that is "not precluded", it's necessary.

    5. I think there's something you need to get straight. We all use small scale models all the time in trying to understand the world around us. Sometimes, we try to verbalize these models. Experience shows that such verbal representations, while useful, are sometimes fraught with logical inconsistencies and/or with assumptions that are implicit. Small scale mathematical versions of these models has been, in my opinion, very useful for the purpose of clarifying our thinking and isolating precisely where our ignorance lies.

      Here are two examples. The first deals with the question of why it seems easy for central banks to lower inflation, but not to raise it:
      The second concerns understanding phenomena related to secular stagnation:

      Now, I invite you to criticize my interpretations. But I also invite you to try to improve on them with your large scale model. Maybe you say you can't, but that economists should try. Fair enough. But in the meantime, we are stuck with our small scale mental, verbal, mathematical models to help us try to understand parts of our world. I think I've learned a lot from my little models. Show me (don't tell me) how to do better.

    6. By the way, your claim that the disputes over involuntary unemployment are an artifact of small scale modeling procedures is false. It's easy to generate phenomena that look like "involuntary" unemployment; see the work of Roger Farmer, for example.

    7. This gets close to my point. It's easy to generate models that produce phenomena that look like involuntary unemployment, and are perfectly "valid" in the sense of self-consistency, and are even consistent with some broader theory or other. Every one of them provides a feeling of insight, at least to its creator. But are any of them "correct", in the sense of capturing what's going on in real-world economies? How do you tell? As someone trained in the natural sciences, I know that feelings of insight, however coherent and strong and clearly and forcefully expressed, are the second worst kind of evidence.

      It's very hard to find real-world natural experiments with sufficient isolation and other controls to even retrospectively test these models against, and it's generally unethical to artificially create convincing experiments.

      So you have to take a two-level approach, with a knowledge level that can alternatively plug in your model of involuntary unemployment, for example, or my model, or Roger Farmer's model, and a data level that creates a realistic environment and drives the response of the simulated economy to each unemployment model. The simulated behavior that deviates the least from the actual behavior indicates that its components are the best match to what's really going on. A simulation harness can also be partitioned or shocked or provided with other externalities (e.g. assumptions) to study how robust the simulation's component models are when pushed to their limits without ruining the lives of real people.

      Anyway, that's the plan. I only have resources to work on it at about 10% level, when it really needs about four people just to get started. Like I said, it's a big project.

  11. I am probably old fashioned, but I don't like DSGE models because they are completely confusing, obscure important points, and require absurd simpliciation assumptions.

    Lastly, it now appears that if you spend all your time and education learning them, you can really get any result you want, provided lots of even more complicated functional forms, modeling, and odd assumptions.

    Look, we have done the surveys. We know prices don't adjust quickly. What more do you need to do?

    1. Try reading this and then get back to me:

  12. Hi David,

    This is an excellent post on the DSGE mechanism. Thank you.