Tuesday, April 24, 2012


Sometimes impacts are big enough in magnitude that they shine through the fog of endogeneity. In that sense, there may be something to this. I'm convinced enough by the Keynes-Hicks argument that I personally think that's probably exactly what this is: impact shining through the fog of endogeneity.

But damn, Paul. It sure makes it tough to talk about good empirical practice when you're so loose with throwing up the scatterplots.


  1. if I may interrupt

    here is an interesting post on capital flight from Southern Europe


    How can the Fed raise our interest rates; doing such will only increase pressure on the South, will it not?

    IOW, we are all in the same life boat.

  2. I'm not sure I understand what you mean by "impact shining through the fog of endogeneity". Either there is a significant degree of endogenity or there is not. If there is, the graph is meaningless at best, or perhaps fun to make but deliberately misleading. PK is not a 19 year old sophomore just accepted into the econ program at Princeton. Dropping a best fit line across a bunch of dots is nothing like science.

    In his own words, I think this is another example "the occasional desire of Democrats to put evidence in a more favorable light" (Feb 29, 2012)

    1. What I mean is that I wouldn't take a point estimate off that fit, but I wouldn't be surprised at all if it is indicative of a strong multiplier that is just washed out by endogeneity.

      We really have two sources of endogeneity here that work in opposite directions. First, countries with a worse recession are going to be the most likely to want to do stimulus if they've got a Keynesian orientation. Second, countries with a worse recession are going to have lower tax revenues, and will therefore cut spending to the extent that a balanced budget is enforce.

      So to me the impact of endogeneity is ambiguous (it's less ambiguous if we consider just the U.S. federal government, which doesn't have those sorts of balanced budget constraints). An ambiguous bias suggests to me that we may be observing a real effect there. But of course you wouldn't want to take the point estimate.

  3. I understand. Thanks. I'd have found the graph to be far more compelling if there were a cloud or football drawn around the points, instead of a line. The line is the center-point of a WAG, and not actually a crisp value. I see this quite a bit at work, where a stack of assumptions will be used to generate a point value or line or curve that is believed to be precise. I don't care for it, to say the least.


All anonymous comments will be deleted. Consistent pseudonyms are fine.