Creating Better Economic Models

A former student of mine recently sent me a link to this excellent article in the Wall Street Journal: "Economists' Grail: Finding a Post-Crash Model", by Mark Whitehouse (the same Mark Whitehouse who wrote "How a Formula Ignited Market that Burned Some Big Investors", which I mention in my old post "Finance's Gaussian Copulas: The New Frankenstein Monster" from March 2009). The article discusses whether mathematical models can hope to be of any sort of relevance in a world that has proved extremely complex and populated by actors who are not completely rational.

Given the recent crash, there is growing concern that these models are built on inaccurate assumptions. The Institute for New Economic Thinking, launched last year with the help of $50mil donated by financier George Soros, is funding new research attempting to tackle such issues and create better models. 

The article does an excellent job summarizing the flaws of current models:

1. "If the equations [defining the dynamic stochastic general equilibrium model] get too complex, or if there are too many elements, the models have a hard time finding the point at which all the players' preferences meet."
2. "Before the crisis, most models didn't have banks, defaults or capital markets, a fact that proved problematic when the financial crisis hit."
3. "They also commonly use a single equation to represent each player [i.e., they consider households, firms, central banks as monolithic groups], impairing the models' ability to explain the unexpected outcomes that can emerge when millions of different people interact."

A possible solution to address these flaws would be, the journalist explains, to resort to agent-based modeling; this would yield a fine-grained description of reality with individual actors and their sometimes irrational behavior (via prespecified rules). Computer-based simulations would be used to produce tractable answers. Such models are, unfortunately, a long way off for now.

My favorite idea in the article was the concept of "imperfect knowledge economics" proposed by New York University professor Roman Frydman, who explains that "[t]he main flaw in the dominant models... is the same feature that makes them so attractive to policy makers: Their ability to make precise predictions" and "believes economists and policy makers must come to terms with the limits of their knowledge." Frydman's eponymous book was highly recommended by The Economist in the article "A new fashion in modeling: what to do when you don't know everything" when it came out in 2007 and in a blog post entitled "New economics" dated February 2009. Time will tell if the idea catches on.

Comments

[Account Deleted]

Hi,Aurelie,

First of all, Merry X'mas!

1. Agent-based method (always with simulation tool, like Arena, Netlogo) is suitable for the complex system which consists of millions of irrational agents (people). It will be useful for performance evalution among multiple alternative decision schemes. However, our mathematical intelligence cannot be injected into the model.

2. I think the concept of "imperfect knowledge economics" is consistent with the emerging concept of "Behavioral Operations Research/Operations Management" (BOR/BOM) in the field of OR/OM.
We had a BOR/BOM workshop in Tsinghua University (Beijing,China) a couple of days ago. Welcome to the workshop next year.

PS: I add your blogsite into my blog as a external link, which makes my blog more informative. Your Chinese students are welcome to my blog (http://blog.sina.com.cn/duyuquan). Thank you.

[Aurelie C. Thiele]

Hi Yuquan,
Thanks a lot for the pointer on Behavioral Operations Management. For interested readers, the following paper is a good introduction to the topic. http://www.bschool.nus.edu.sg/staff/bizwyz/BOM_LochWu.pdf
Merry Christmas to you too and best wishes for 2011! I'm sure my Chinese students will be happy to visit your blog.