Tuesday, April 22, 2014

DSGE: the sinking Titanic of economic methodology

What's the future of macroeconomics? Does it lie in further development of the old-style models of rational optimizers and equilibrium, the dynamic stochastic general equilbrium (DSGE) models? Or will it instead be a new breed of agent-based computational economics (ABM), i.e. in computational simulations which don't restrict themselves to rational optimizing behavior, or to equilibrium?

From what I see, the DSGE people -- the old guard, if you will, as this IS the current mainstream approach -- don't take the opponent very seriously. They seem to sneer and chuckle at ABM for its lack of mathematical rigor; they don't even prove theorems! BUT, I suspect, this is only because the DSGE people secretly know very little at all about ABM, about its potential, its power and flexibility, and especially, about how far it has been developed already, for exploring banking stability,  monetary policy and so on. DSGE, as I see it, is doomed for sure. It's not going to be a fair fight.

DSGE is the Titanic of economic methodology, already taking on water, its bow looming high in the air. Message to young economists: Don’t let your career go down with it! Read more at Medium.


  1. I believe one of the reasons that ABM is not popular, at least in the academic circle, is that you cannot really do any "fancy" mathematics. Pretty much all that could be done is to specify how the agent behaves both over time and under external stimulus and then let the computer run it 1000 times with different random seeds. After that, conclusions are drawn based on the simulation results. How can this be published in a "renowned" journal? Where is the intellectual challenge to go through the maze of 20 steps of complicated mathematical derivations? Nevertheless, the seemingly "naive" ABM can yield way more realistic results if the behavior of the agents are programmed realistically.

    1. I disagree. Think of all the automated quantitative trading systems out there. They do math as fancy as you can imagine and they are each agents.

      All canonical economic models, including DSGEs, can be represented in agent models if you can think abstractly and have some coding chops.

      I've been laying out a theoretical ABM framework in which we can accomplish this. I think traditional economists would be more receptive to ABMs if they can begin in the comfort zone of their orthodox models, and then have the ability to begin relaxing assumptions.

  2. yes, I think that may well be right. However, I think it is also something of an illusion. There's no reason lots of fancy mathematics might not be done in the context of an ABM. For example, it will be important to do more than just run simulations. It would be nice to try to derive from the large scale model some lower dimensional models which capture a lot of the same dynamics, though of course not all. This is a big challenge where mathematics is very useful.

    1. I agree. The other month I discussed a couple different ABMs in a post http://zacharydavid.com/2014/03/on-hft-assumptions-agent-based-modeling-and-a-philosophy-of-error/#Secret-AgentBased-Man

      One was a system using extremely granular individual loan data in a single city (John Geanakoplos et al) to try and derive empirical foundations of behavior and assess systemic risk.

      The other is what happens when you rely too much on assumptions and simulation and get the answer you wanted to get.

  3. Isn't the real failure of Economics it's reluctance to consider the interaction between economic models and the workings of the physical systems that investors invest in? "Growth" occurs in economies much like any other place in nature, as nature's most ever-present process of inventing new organization, very often seen with an appearance of new forms emerging in a dazzling array, then settling down to either stabilize everything created or break it all down in some way.

    There's no equation for that, of course, like there's also no equation for how to drive on a crowded highway other than needing to watch the traffic. So it’s odd that its only "non-quants" who ever seem to carefully study it. It’s clearly where the things investors invest in come from... so not studying that seems a fairly significant omission I think.

    Your interesting 2011 article on "Efficiency v. Stability" suggested that success in achieving market efficiency might make it more likely for market bubbles to develop. As a scientist who studies these kinds of repeated phases of organizational change found in all kinds of emerging systems, I think it’s fairly likely to be true, not for a statistical reason but a behavioral one. Economic models don't behave like physical systems...

    From an investment risk point of view there's an inherent conflict, due to their points of minimum and maximum risk being misaligned. It’s a problem easy to understand if you compare how they work. When economic models are most stable is when they are smoothly multiplying. That happens to be when physical systems are actually least stable, when they are changing form in increasingly unexpected and rapid ways.

    Perhaps what fools us is that there generally is a period of smoothly multiplying change as part of nearly any process of natural growth. **For that system**, though, it's a physical process of changing its own working internal and external relationships ever faster, which is not stable. So, it's naturally "just after" a growth system has been behaving in the economically "most stable" way, that a physical growth system becomes "least stable".

    The opposite is also true, unfortunately, that when physical systems are “most stable” is when they are not multiplying, and that's when economic models are “least stable”. So investors are then inclined to do anything they can, to make their environment start multiplying returns again... like letting down their guard and creating bubbles perhaps.


  4. I've come at this from philosophy and had the same experience you describe in your post on "Just So Stories", it just doesn't make sense. I got into a discussion with a well known econo blogger over email about whether algebraic manipulation of the equation used to calculate GDP--created a priori and true by definition--could possibly tell us anything about the world (other than GDP). It was a strange experience. Meanwhile, I actually thought that all the talk of models meant ABM or something like it. I've read pretty much everything Mark Thoma has linked to for a year and a half and not once has anyone even tried to make any sense of what "equilibrium" is supposed to be.

    I'm amazed that that word itself doesn't ring alarm bells. If you know what it means then you know it can't describe anything about human behavior. So strange.

    The whole field is cruising for a bruising.

  5. Mark, are there any UK universities doing research into ABM or similar?


  6. I personally believe that finance homework help in various economic affairs.