This paper is a response to a major inconsistency between the theory of capital investment and the results of simple and traditional econometric analysis. Economic theory (such as the q theory discussed in class) suggests that the cost of capital is a major determinant of the amount of capital and investment demanded. However, empirical analysis of the usual kind struggles to find any effect of the user cost (or rental price) of capital on investment. There are likely many reasons for this:
error in measuring the real interest rate due to unobserved inflation expectations
impossibility of measuring and controlling for firms' expected marginal product of capital
long "gestation" lags between changes in market conditions and the actual changes in investment spending that they may cause
possible reverse causality: changes in investment demand may affect interest rates and thus the cost of capital
fact that real interest rates don't vary very much in the available data sample (and the unobserved expected marginal product probably varies a lot more).
Cummins, Hassett, and Hubbard (CHH) use changes in tax policy as large, exogenous, and easily identifiable shocks to the cost of capital and attempt to measure the resulting changes in investment spending.
The "user cost of capital" and the "rental price of capital" are two names for the same concept.
Our derivation of q in class obscures an important distinction in the empirical literature. The q that is relevant for investment decisions is "marginal q," which depends on the marginal product of capital. The q that is most easily measured from stock prices or actual productivity is "average q," which depends on the average product of capital. In Romer's derivation, the π (pi) function plays both roles. It is the operating profit from a unit of capital, which does not depend on κ (kappa), the amount of capital that the firm has. Thus, the marginal product of capital for the firm is assumed to be flat and equal to the average product. This assumption means that Romer's q is both average and marginal q.
Try to grasp the intuition of the equations on page 20, but do not stress about understanding the details of every term.
The "generalized method of moments" estimators are a way to deal with the bias that may result from reverse causality (investment affecting the cost of capital) and with expectations. Don't worry about the details.
The authors have a somewhat confusing change of convention within the paper.
In Tables 3 and 4, the number in parentheses below each coefficient is its standard error. A variable whose coefficient is twice as large (in absolute value) as the standard error is typically said to be statistically significant: the relationship between the variables is too large to be likely through random chance.
In Tables 5 through 10, the number in parentheses is a t-statistic, which is the coefficient divided by its standard error. In these tables, a value in parentheses larger than 2 indicates conventional statistical significance.
The section "Some Additional Considerations" that begins on page 44 is largely robustness checks and can safely be read very lightly.
Questions for analysis
Discuss the design of the "experiment" undertaken in the paper. Why do the authors believe that tax reforms are useful for identifying effects of the cost of capital on investment? Are their arguments persuasive?
The authors (following Fazzari, Hubbard, and Petersen) include cash flow as a determinant of investment in some equations. What is the rationale for including this variable? If capital markets were perfect and the Modigliani-Miller Theorem held, would cash flow have an effect?
Describe the results of the "traditional" regressions in Table 3. In what ways are the results consistent or inconsistent with traditional investment theory?
Most empirical studies of investment use time-series data. Each regression in Tables 5-10 has a sample of firms from only a single year. How does the cost of capital in a particular year vary across firms in a way that allows the authors to identify its effect on investment?
What results do the authors observe in years of tax changes and years without tax changes? Based on these regressions, what do they conclude about the sensitivity of investment to capital costs or q? How do these results compare to those from Table 3 that you discussed above?