instead some of the impact comes from changes in the set of entrepreneurs who are able to borrow.

The third implication of our analysis is that many variables that do not affect investment when capital markets are perfect matter when capital markets are imperfect. Entrepreneurs wealth provides a simple example. Suppose that and W are heterogenous across entrepreneurs. With perfect financial markets, whether a project is funded depends only on y. Thus the projects that are undertaken are the most productive ones. This is shown in Panel (a) of Figure 8.15. With asymmetric information, in contrast, since W affects the agency costs, whether a project is funded depends on both and W. Thus a project with a lower expected payoff than another can be funded if the entrepreneur with the less productive project is wealthier. This is shown in Panel (b) of the figure.

The fact that financial-market imperfections cause entrepreneurs wealth to affect investment implies that these imperfections can magnify the effects of shocks that occur outside the financial system. Declines in output arising from other sources act to reduce entrepreneurs wealth; these reductions in wealth reduce investment, and thus increase the output declines (Bernanke and Gertler, 1989; Kiyotaki and Moore, 1995).

Two other examples of variables that affect investment only when capital markets are imperfect are average tax rates and idiosyncratic risk. If taxes are added to the model, the average rate (rather than just the marginal rate) affects investment through its impact on firms ability to use internal finance. And risk, even if it is uncorrelated with consumption, affects investment through its impact on agency costs. Outside finance of a project whose payoff is certain, for example, involves no agency costs, since there is no possibility that the entrepreneur will be unable to repay the investor. But, as our model shows, outside finance of a risky project involves agency costs.

Fourth, and potentially most important, our analysis imphes that the financial system itself can be important to investment. The model implies that increases in c, the cost of verification, reduce investment. More generally, the existence of agency costs suggests that the efficiency of the financial system in processing information and monitoring borrowers is a potentially important determinant of investment.

This observation has implications for both short-run fluctuations and long-run growth. For short-run fluctuations, it implies that disruptions to the financial system can affect investment, and thus aggregate output. For example, Bernanke (1983b) argues that the collapse of the U.S. banking system in the early 1930s contributed to the severity of the Great Depression by reducing the effectiveness of the financial system in evaluating and funding investment projects. Similarly, many observers argue that an important factor in the 1990-91 recession in the United States was a "capital crunch" at banks that reduced their ability to make loans. Their argument is that because banks had little capital of their own in this period, they were

l + r

l + r

FIGURE 8.15 The determination ofthe projects that are undertaken under symmetric and asymmetric information

unusually dependent on external finance; this raised the opportunity cost of funds to them, and thus made them less willmg to lend (see, for example, Bernanke and Town, 1991).

With regard to long-run growth, McKiimon (1973) and others argue that the financial system has important effects on overall investment and on

the quahty of the investment projects that are undertaken, and thus on economies growth over extended periods. Because the development of the financial system may be a by-product, rather than a cause, of growth, this argument is difficult to test. Nonetheless, King and Levine (1993a, 1993b) present some evidence that financial development is important to growth.

8.8 Empirical Applications

The Investment Tax Credit and the Price of Capital Goods

As we saw in Section 8.1, the costs of adjusting the capital stock can be either external or Internal to firms. If they are external, they take the form of an increase in the relative price of capital goods when firms desired capital stocks rise. Since there are data on the relative price of capital goods, it is possible to test for this effect.

Such a test is carried out by Goolsbee (1994). He focuses on how the investment tax credit affects capital-goods prices. His basic specification is

lnP,f = bnt + c;x„ + e,t, (8.35)

where P,f is the relative price of capital of type in year t, is a measure of the investment tax credits subsidy to the purchase of capital good z in year t (as a percentage of the purchase price), and X,i is a vector of control variables whose effects are allowed to vary across the goods. Goolsbee argues that the government may have a tendency to increase the investment tax credit in times, such as recessions, when investment demand is otherwise weak. If this is correct, the estimate of wHl tend to understate the true effect of the investment tax credit on capital-goods prices. To address this problem, Goolsbee includes dummy variables for each year in some specifications; when he does this, he is controlling for the variation over time in the overall investment tax credit and focusing only on the differences in the credit across different types of capital.

Goolsbee considers 22 capital goods over the period 1962-1988. His basic regression includes dummy variables for each capital good, a time trend, and dummy variables for the years of the Nixon price controls. This specification yields an estimate of b of 0.55, with a standard error of 0.12. Thus the results provide strong evidence of external adjustment costs, and the\ suggest that about half of the investment tax credits subsidy to investment is passed into capital-goods prices.

When dummy variables for each year and time trends for each good are included, the estimate of b rises, which is consistent with Goolsbees argument that leaving out the year dummies biases the estimate of b downward. The estimate of b is now 0.95 (with a standard error of 0.35); thus this specification suggests that the investment tax credit is reflected essentially one-for-one in prices and has essentially no effect on firms incentives to invest.