The theory predicts that the effect of an increase in the credit on the relative price of capital goods should disappear over time as firms gradually increase their capital stocks; in addition, Goolsbee argues that capital-goods suppliers adjustment of their capacity also contributes to the adjustment process. To investigate whether the impact of the credit is temporary, Goolsbee examines its effect on the relative price of capital goods over various horizons. The estimated effect is 0.55 (with a standard error of 0.18) after one year, 0.60 (0.19) after two, and 0.45 (0.36) after three.

It is tempting to interpret these results as supporting the predictions of the theory: the estimated effect is significantly different from zero in the first two years but not significantly different in the third. This interpretation, however, commits the classic error of equating the failure to reject a hypothesis with accepting it. The hypothesis that the impact after three years is zero cannot be rejected only because the effect is estimated imprecisely: although one cannot reject the hypothesis that the effect is zero, one also cannot reject the hypothesis that it is 1. Thus the results support for the theory is slight: they suggest only a small decline in the effect over time, and they are more consistent with the view that the effect is constant than with the view that it disappears after three years.

Finally, returning to the contemporaneous effect of the investment tax credit, Goolsbee examines the variation in the effect across capital goods; specifically, he estimates a separate b for each type of capital. He finds that the price responses are large for goods that are purchased almost entirely by firms, such as mining machinery and railroad equipment, and small for goods that are purchased mainly by households, such as personal computers and furniture. Since households are not eligible for the investment tax credit, this suggests that the price effect of the credit is larger when it affects more of the buyers of a good. To investigate this idea, Goolsbee constructs estimates of the fractions of the purchasers of each good who are ehgible for the credit. He then includes in the regression not just the credit, but the product of the credit and his eligibility estimate. If the reason that the credit is associated with increases in the price of capital is through its impact on demand, then under plausible assumptions the association should depend on only this interaction term.

The results support this prediction. When the interaction term is included, the coefficient on the credit falls to -0.02, with a standard error of 0.27. Thus not only can the hypothesis of no effect not be rejected, but the point estimate suggests a negligible impact. The coefficient on the interaction term is 0.80, with a standard error of 0.35; thus the estimated effect of the interaction is quantitatively large and statistically significant.

Cash Flow and Investment

Theories of financial-market imperfections imply that internal finance is less costly than external finance. They therefore imply that, for a given level of interest rates, firms with higher profits invest more.

One complication to this argument is that It may be costly for high-dividend firms to reduce their dividends: there is evidence that reductions in dividends are interpreted by the stock market as a signal of lower future profitability, and that the reductions therefore lov\ er the value of firms shares. Thus it is possible that the test could fail to find differences between the two groups of firms not because financial-market imperfections are unimportant, but because they are important to both groups.

A naive way to test this prediction is to regress investment on measures of the cost of capital and on cash /7ow-loosely speaking, current revenues minus expenses and taxes. Such regressions can use either firm-level data at a point in time or aggregate data over time. In either form, they typically find a strong hnk between cash flow and investment.

There is a problem with this test, however. The regression does not control for the future profitability of capital, and cash flow is likely to be correlated with future profitability. We saw in Section 8.5, for example, that our model of investment without financial-market imperfections predicts that a rise in output that is not immediately reversed raises investment. The reason is not that higher current output reduces firms need to rely on outside finance, but that higher future output means that capital is more valuable. A similar relationship is likely to hold across firms at a point in time: firms with high cash flow probably have successful products or low costs, and thus have strong incentives to expand output. Because of this potential correlation between cash flow and current profitability, the regression may show a relationship between cash flow and investment even if financial markets are perfect.

A large literature, begun by Fazzari, Hubbard, and Petersen (1988), addresses this problem by comparing the investment behavior of different types of firms. Specifically, Fazzari, Hubbard, and Petersens idea is to divide firms into those that are likely to face significant costs of obtaining outside funds and those that are not (see also Hoshi, Kashyap, and Scharfstein, 1991). There is likely to be an associatton between cash flow and investment among both types of firms even if financial-market imperfections are not important. But the theory that financial-market imperfections have large effects on investment predicts that the association will be stronger among the firms that face greater barriers to external finance. And unless the association between current cash flow and future profitability is for some reason stronger for the firms with less access to financial markets, the view that financial-market imperfections are not important predicts no difference in the cash flow-investment link for the two groups. Thus, Fazzari, Hubbard, and Petersen argue, the difference in the cash flow-investment relationship between the two groups can be used to test for the importance of financial-market imperfections to investment.

The specific way that Fazzari, Hubbard, and Petersen divide their firms is according to their dividend payments as a fraction of income. Firms that pay high dividends can finance additional investment by reducing their di\ -idends. Firms that pay low dividends, in contrast, must rely on external finance.

The basic regression is a pooled time series-cross section regression of investment as a fraction of firms capital stock on the ratio of cash flow to the capital stock, an estimate of q, and dummy variables for each firm and each year. The regression is estimated separately for the two groups of firms. The sample consists of 422 relatively large U.S. firms over the period 1970-1984. Low-dividend firms are defined as those with ratios of dividends to income consistently under 10%, and high-dividend firms are defined as those with dividend-income ratios consistently over 20% (Fazzari, Hubbard, and Petersen also consider an intermediate-dividend group).

For the high-dividend firms, the coefficient on cash flow is 0.230, with a standard error of 0.010; for the low-dividend firms, it is 0.461, with a standard error of 0.027. The t-statistic for the hypothesis that the two coefficients are equal is 12.1; thus the hypothesis is overwhelmingly rejected. The point estimates imply that low-dividend firms invest 23 cents more of each extra dollar of cash flow than the high-dividend firms do. Thus even if we interpret the estimate for the high-dividend firms as reflecting only the correlation between cash flow and future profitability, the results still suggest that financial-market imperfections have a large effect on investment by low-dividend firms.

Many authors have used variations on Fazzari, Hubbard, and Petersens approach. Lamont (1993), for example, compares the investment behavior of the non-oil subsidiaries of oil companies after the coUapse in ofl prices in 1986 with the investment behavior of comparable companies that are not cormected with ofl companies. The view that internal finance is cheaper than external finance predicts that a dechne in oU prices, by reducing the availability of internal funds, should reduce the subsidiaries investment; the view that financial-market imperfections are unimportant predicts that it should have no effect. Lamont finds a statistically significant and quantitatively large difference in the behavior of the two groups; the point estimates imply that each doUar of lower income of a parent oil company reduces investment of the companys non-oil subsidiaries by 10 cents. Thus his results suggest that the barriers to outside finance are considerably larger than the barriers to finance between different parts of a company.

Gertler and Gilchrist (1994) carry out a test that is in the same spirit as these but that focuses on the effects of monetary policy (see also Kashyap, Lamont, and Stein, 1994, and Oliner and Rudebusch, 1994). They begin by arguing that smaU firms are likely to face larger barriers to outside finance than large firms do; for example, the fixed costs associated with issuing publicly traded bonds may be more important for smaU firms. They then compare the behavior of smaU and large firms inventories and sales following moves to tighter monetary policy. Again the results support the importance of imperfect financial markets. Small firms account for a highly disproportionate share of the declines in sales, inventories, and short-term debt following monetary tightening. Indeed, large firms borrowing increases after a monetary tightening, whereas smaU firms borrowing declines sharply.