*Harry DeAngelo and Richard Roll (2008 JF)*

**Summary**

- Lemmon, Roberts, and Zender (2008) and others make the argument that the cross-section of corporate capital structure is quite stable over long horizons. LRZ show that, among a selection of determinants that are believed to be linked to capital structure, firm fixed effects are by far the most powerful predictor. They also sort firms into leverage quartiles, and show that the high- (low-) leverage portfolios at time
*t*also have the highest (lowest) leverage as much as 20 years into the future. - One interpretation, bluntly stated, is that 50 years of capital structure research has been barking up the wrong tree. Researchers should go “back to the beginning” and rethink their approach.
- Another interpretation of these findings is that the cross-section is stable over time, and therefore not interesting. Researchers should confine themselves to explaining within-time variation in capital structure.
- DeAngelo and Roll (this paper) present compelling arguments that (1) LRZ’s analysis masks variation of the cross-section over time, and so is somewhat biased toward finding stability and (2) capital structure at the firm level is wildly unstable, and the evolution of the cross-section is at least as interesting a research topic as the snapshot at each point in time.

**Methods and Findings**

- The sample consists of 15,096 CRSP/Compustat firms over 1950-2008.
- The authors test for firm-level stability by measuring the length of each firm’s “stable leverage regimes” – i.e. a period of time where the firm’s leverage stays within a narrow band of 0.05, 0.1, or 0.2.
- Among firms that are in the sample for at least 20 years, only 20% have stable (using the bandwidth definition of 0.05) leverage for a ten-year period, only 4% have stable leverage for a 20-year period, and the median length of firms’ “longest stable regime” is only 6 years.
- Among firms that are in the sample for the entire 59-year period, 52% have stable leverage (again, bandwidth of 0.05) for some 10-year period, 0% have stable leverage for a 40-year period, and the median length of “longest stable regime” is only 10 years.

- They then show that, during firm’s stable regimes, leverage tends to be quite low, often less than 0.1.
- A significant portion of the paper is dedicated to explaining why the results in Lemmon, Roberts, and Zender (2008) and MacKay and Phillips (2005) are misleading.
- Use a creative specification that allows firm-time fixed effects with firm-time observations. A textbook application of this would result in one fixed effect for each observation and tells the researcher absolutely nothing.
- Following Scheffé (1959), they get around this by imposing some additional structure. They assume that firm-time interaction effects are stable over longer periods – they arbitrarily choose 10 years – so that they run regressions with firm-
*decade*fixed effects.- Using firm-decade interactions significantly improves R-squared over a specification with firm fixed effects only.
- ANOVA reveals that

- Following Scheffé (1959), they get around this by imposing some additional structure. They assume that firm-time interaction effects are stable over longer periods – they arbitrarily choose 10 years – so that they run regressions with firm-
- DeAngelo and Roll (this paper) use a longer sample than the papers they criticize. This is important, because if firm leverage changes slowly (this can be represented by adjustment costs to leverage), then the power of firm fixed effects will be overstated in short samples. The problem is similar if many firms only appear in the sample for a few years.
- This is also important because, as these authors argue, the LRZ sample begins in 1970 – after economy-wide increases in leverage as firms took advantage of post-war investment opportunities. Thus, the later sample misses crucial time-series variation in leverage.

- Use a creative specification that allows firm-time fixed effects with firm-time observations. A textbook application of this would result in one fixed effect for each observation and tells the researcher absolutely nothing.
- In another set of analyses, DeAngelo and Roll show that the cross-section varies meaningfully over time.
- They measure the correlation between several pairs of cross-sections – the pairs {t,t+1}, {t,t+2},…,{t,+40}.
- The correlation falls quickly from 0.8 for {t,t+1} and approaches zero.

- They show that firms in one leverage quartile at time
*t*moves across quartiles a lot in the ensuing years. LRZ miss this because they look at*average*leverage of quartile groups.

- They measure the correlation between several pairs of cross-sections – the pairs {t,t+1}, {t,t+2},…,{t,+40}.
- The final part of the paper tests various theories of capital structure: (1) Miller’s (1977) random variation, (2) speed-of-adjustment (SOA) models, (3) flexible target ratio models, and (4) time-varying target (TVT) models.
- They simulate a model that nests all these as special cases, and see which model(s) appear to best fit the data.
- The TVT and flexible-target models seem to be the best, while Miller’s random leverage model is not supported.

The authors’ primary conclusion is that the cross-section is far from stable, and that within-firm variation of leverage over the time series is probably a response to the firm’s investment needs.

**Comments**

- I liked this paper. It documents new facts about the variability of firm-level debt ratios over time, and uses creative analysis.
- The paper also leaves many questions unanswered. It doesn’t explain
*why*firms have such low leverage during stable regimes (though DeAngelo, Stulz, and Gonçalves currently have a working paper that tries to answer this question). It doesn’t explain why, if macro factors (post-war environment) are so important, the cross section still changes so much from year to year. - Sample-selection bias is an accusation commonly made to papers whose data start after the post-war period. David, Fama, and French (2000) made a similar criticism of Daniel and Titman’s (1997) argument that firm characteristics matter and risk-factor loadings don’t. But it is not clear that the post-war decades are relevant for all analyses.
- Fama’s and French’s three-factor asset pricing model seeks to explain only cross-sectional difference in returns, and is not concerned with how the cross-section changes over time. High-return stocks in any period have different risk-loadings, but the risk loadings are allowed to change over time in an unspecified manner. In this case, extending the sample backwards should be safe.
- But this paper is concerned with how the cross-section varies over time. Firms’ “wholesale abandonment of conservative capital structure” in the 1950s and 1960s, as they took advantage of (potentially very rare) investment opportunities may not tell us much about how they have managed their debt ratios in the last 40 years.