July 2009

Ivo Welch, Brown/NBER/Brown: Updates to 2004 JPE paper

Background

When I look back on my published work, I do not think there is a single paper that I would not like to tweak in some way or another. I have learned many new things since I have published them. Sometimes, corrections are also due. Thus, it is my intent to write short notes eventually for all my published papers. Obviously, this is easier if I happen to work on the same subject right now—which is why this is the first note I am posting (as of 2009).

Updates

• The dependent variable in Table 5 has been multiplied by 100 (i.e., is quoted in percent). The table should have stated this. If you do not know this, you cannot understand the economic magnitudes. Mea culpa.
• There was no reason to use the Fama-Macbeth specification in the cross-sectional regressions. My intent at the time was to produce standard errors that were a little wider. With more than a hundred-thousand firm-years, the accuracy is super-precise, anyway. It does not hurt to do Fama-Macbeth, but it does not help, either.
• I would change the terminology of the paper. Whereas Table 5 (and most preceding papers) were interested in testing deeper causes of capital structures (such as theories of debt capacity and adverse selection), my own paper is testing a shallower cause. In fact, it would make sense to call most of what I am writing about mechanisms of capital structure change. Stock returns are somewhere in between a mechanism and a deeper cause.
• The original paper was nonchalant in its identification (interpretation of causality). This deserves more explanation. The identification in my paper comes at a cost of ignoring one form of reverse causality: if debt issues cause negative stock returns, then the no-readjustment leverage ratio would go up, and the paper could mistakenly attribute part of managers' intentional levering-up as being due to the no-readjustment hypothesis. The same holds if equity issues cause positive stock returns.
  
  Let me give an example. Take a firm with $50 in debt and $50 in equity (an ADR of 1/2). Now assume the firm had a stock return of -50%. Thus, its IDR is 2/3. The problem arises in the interpretation: if the $50 debt issue is the cause of the -50% rate of return, then managers were active, after all. The manager manipulated the future debt ratio via the debt issue that in turn caused the return (i.e., the IDR). In this case, my paper would mistakenly claim that the managers did nothing to manage their capital structures, when in fact they did. They just used their capital structure changes in order to manage their stock returns, which they used in order to adjust to next year's capital structure.
  
  Fortunately, the empirical evidence (and plausibility) does not suggest that this is a big problem. The equity issue stock price announcement evidence suggests the opposite, however. Debt issues are usually accompanied by very little announcement response. Thus, this reverse causality story is unlikely. In a sense, I got lucky here. I should have pondered these issues in the original paper. (Association rather than reverse causality is not a problem due to the way my evidence is interpreted.)
  
  However, even if I had pondered it, it would not have been easy to correct. Standard IV does not work in my context. See, because the identification is so specific (based an a quantitatively exact ratio on own stock returns), it is both less sensitive to this reverse-causality problem—unless the stock return and net issuing activity happen in exactly the right proportion, IDC is unlikely to pick it up strongly—and difficult to correct. Such a correction would be difficult because any stock return other than the actually observed exact one (i.e., the fitted value from an IV regression) should not work under the functional identification. Unlike common IV techniques in linear models, in my paper even a small constant added to the stock return would destroy the usefulness of the IDC identification.
• Peter Iliev and I have authored a follow-up paper (2009) that revisits the speed of adjustment estimations in my paper and in a number of followup papers.