Organizations and Markets

5 Comments Add your own

• 1. spostrel  |  7 March 2007 at 8:21 pm

  Way back in 1991 (SMJ 12 153-155) I wrote that “…game theory does not, of itself, contain a substantive account of behavior. Game models are extremely sensitive to assumptions about information, the order of moves, constraints on action, and players’ beliefs. Yet these assumptions, not game logic itself, are the real substance of a theory of business competition. Game theory is no more a theory of business behavior than calculus is a theory of consumer behavior.”

  The complaints mentioned in the paper are neither new nor particularly damaging to the use of game theory as a modeling tool. As someone who was trained in the “See if you can build a good model where phenomenon X happens” tradition, I agree that you will not generate that many precise empirical predictions just by assuming common knowledge of rationality on arbitrary strategy spaces. But I will also tell you that a lot of stories that you might want to tell fall apart when you try to model them game-theoretically, in that you end up needing assumptions you never imagined in order to get the outcome you’re shooting for.

  I’d also point out that Ghemawat’s Games Businesses Play makes a pretty good argument (by demonstration) that applying game theoretic tools to specific business cases can lead to much better understanding and prediciton about the cases. In addition, trying to apply game theory to these cases shows where the faults and weaknesses of our models really bind, and point the way to improvement.

• 2. John Quiggin  |  8 March 2007 at 4:48 am

  There’s a much more fundamental problem here. The problem is that game theorists routinely write down a problem description in which each players action can be summarised by a single choice variable, then derive a Nash equilibrium in which the strategy space consists of settings of this choice variable. [We give a lot of examples] But there are infinitely many such descriptions, and there are usually reasonable ones that yield any plausible outcome.

  The problem of deriving a strategic representation from a description of possible actions is not mentioned in your list, and is implicitly assumed away in your post. But its much more fundamental than concerns about information, order of moves and so on.

• 3. How dead is Game Theory? « Yet Another Sheep  |  8 March 2007 at 3:09 pm

  […] Flavio Menezes and John Quiggin have a draft paper in which they argue that problems with the exact specifiction of the strategy space may invalidate many economic uses of game theory. […]

• 4. spostrel  |  8 March 2007 at 5:52 pm

  I enjoyed reading your paper. I have no problems with your suggestion to use extrinsic evidence to constrain choices of strategy space. This is no different from using extrinsic evidence to pin down beliefs, move orders, etc.

  We have known that strategic representation is a problem since the 19th century, with the choice of Bertrand vs. Cournot models. These indeterminacies are in no way “more fundamental” than the ones usually noted. Just more of the same. It is worth reminding people to pay attention to this issue–generally people do in industrial organization. I can’t speak to the folks who do Tullock contests in political economy, because I’m not familiar with that work.

  I do think that not all strategic representations are a priori reasonable. The cube root of output plus twenty-three is not likely to be in the other guy’s Book of Nash. That may leave multiple reasonable options in play, just as there are multiple reasonable options for modeling information structures, move orders, and all the rest, but it is less wide open than you imply.

  Finally, on methodology–the Popperian bogeyman has been dead for a long time, as has the Demarcation Problem. Falsifiablility is certainly desirable in a theory, and it may be a necessary check on certain applications of theory–If game theorists tried to use game theory the way Marxists used Marxism or Freudians used Freud, they would indeed be subject to Popper-style critiques–but it is not a holy grail. Shannon’s communciation theory is a good description of the real world, but you couldn’t go out and falsify it empirically.(Offered only as a counterexample–I’m not claiming that anything in game theory is as good as that!)

  (One minor clarification issue: In your paper, the point being made when you have one firm choosing prices and the other choosing quantities could be clearer. Was an explanatory paragraph cut during editing? I apologize if I am being obtuse.)

• 5. John Quiggin  |  8 March 2007 at 6:40 pm

  Thanks for these comments. I agree that not all strategic representations are a priori reasonable, but I also think its unlikely that strategic representations that work nicely for economic modellers (for example, prices and quantities as decision variables) are necessarily going to be plausible representations of the way in which decisionmakers in firms think about decisionmakers in other firms.

  I’m probably more attached to falsifiability than you are, but I don’t suppose we’ll resolve this in a comments thread.

  We’ll go back and take a look at the price-quantity example. Obviously we were unclear on this.

Trackback this post  |  Subscribe to the comments via RSS Feed