Nathan M. Palmer

Welcome to my woefully inadequate homepage - very much a work in progress! I am currently a PhD candidate in the Department of Computational Social Science at George Mason University, specializing in simulation techniques applied to macroeconomics (learning & consumption behavior) and finance (leverage, liquidity, & contagion). Rob Axtell is my advisor and I am having a blast working on interesting projects with a lot of very bright people.

Note to prospective CSS students:  If you are interested in pursuing economics or finance in the CSS program, please feel free to send me an email. I'm always happy to talk about the work we do and am happy to offer (often blunt) advise as you survey the PhD landscape. 

Publications & Work in Progress

  • Geanakoplos et al (2012), "Getting at Systemic Risk via an Agent-Based Model of the Housing Market." American Economic Review, 102(3): 53–58.
    • Paper here
    • More in Session 2 here; in particular, see the discussion by Sufi, Sims, Cochrane, Hansen, and others. Cochrane's questions in particular are what I hope to address in my own research.
    • Rob Axtell discusses this model briefly near the end of this talk, around 12:40.
  • My consumption research presented at CEF 2012.
    • Very preliminary version of the paper is linked here; an updated version is in progress and will be out any day now...
    • Slides can be found here.

  • My SSRN page.

  • In addition, I am working on interchangable components for multi-agent economic models (aka agent-based models) in Python and Java. The idea is to encapsulate key aspects of general economic models (macro and macro-finance in my case) in code objects such that they are modular and interchangeable -- similar to the so-called "Strategy Pattern." My end goal is to construct a set of libraries which does for agent-based modeling what Dynare does for DSGE modeling.
    • I do not have a working paper for this, as the research is in very early stages. My prototypes in progress are in Python, using the NumPy, SciPy, and SymPy libraries.


Aside from computational economics and agent-based modeling, I thoroughly enjoy being outdoors. As a native South Texan, I miss Tex-Mex and dance halls (not line dancing, although that can be fun -- I'm much more a fan of two-step, waltz, polka, and swing). I recently discovered that some activities I enjoyed in undergrad are nowadays described as "freerunning" or "parkour." I'm certainly not in shape enough for that now, but I'd love to get into it again someday...

On an academic note, I really enjoy programming and computational approaches, and the mathematics needed to understand and use computational models. (Heck I just think math is very interesting.) How people learn and make decisions in uncertain situations is fascinating to me. People are much smarter than they are often given credit, I beleive, and capturing their behavior in a highly complex and uncertain environments is a fascinating topic. Optimizing behavior may at times seem to vastly over-estimate an agent's ability, but rule-of-thumb behavior may vastly underestimate an agent's ability.

I strongly believe that agent-based modeling is simply the next step in computational economic modeling, incorporating insights from software engineering to construct models that "trace out" the effects of "well tested theory" (to use Kydland and Prescott's terminology). The "black box" of agent-based modeling need not exist. 

Ongoing research in areas such as boundedly rational dynamic programming, Q-learning & related tools, and N-Period-ahead learning are just a few examples of topics that interest me quite a bit. Another approach, of course, is to "mix" some optimal and non-optimal behavior, either in a single agent (as in Laibson et al.'s  Natural Expectations) or in a population of agents (as in Lansing's work and Wouter den Haan's work). None of these even touch on econometric or statistical learning approaches, or AI approaches, or game-theoretic approaches. Much exciting research taking place. For me, an ideal outcome of any of these is modular learning and expectations-formation procedures that are well-understood both at the individual level, in classic "simple" macroeconomic frameworks, which may then be used in any number of large-scale simulations.