the effects of taxation on investment: new ... - National Tax Association

i

National Tax Journal, Vol. 45, no. 3, (September, 1992), pp. 243-51

THE EFFECTS

OF TAXATION ON INVESTMENT: NEW EVIDENCE FROM FIRM LEVEL PANEL DATA** JASON G. CUMMINS* AND KEVIN A. HASSETT*

"lu discrepancy between theory and e-Piriw work n perhaps nowhere in macroeconomics so obvious a! in the case of the aggregate investment function, (Olivier Blanchard) 'One of the best established facts in macmeconoma is that business fixed investment and output move dmngly together over the business cycle. By contrut, investment and the cost of capital are either unwrrelated or only weakly correlated." (Matthew D. &&Piro)' One of the most important paradoxes in both modem empirical public finance and macroeconomics is the inability of remarchers to find a convincing effect of the cost of capital on business fixed investment, despite strong theoretical motivation for the importance of the cost of capital and the apparent confidence of policymakers that changes in the cost of capital can substantially alter the path of domestic investment. 2 The belief that the cut of capital has little effect on investment was seemingly buttressed by the response of aggregate investinent to the Tax Reform Act of 1986 (TRA86),. TRA86 was generally believed to have eased the tax favored treatment of new investment, and this was especially true for equipment investment, which no longer qualified for the investment tax credit after the Act. Despite this, the share of GNP devoted to fixed investment increased after 1986, and, edntrary to the predictions of theory, there was a substantial increase in equipment investment, which soared from 8.4 percent of GNP in 1985 to 9.3 percent in 1989. Investment in structures, which, as we will see below, one might have suspected would increase after the Act, actually decreased, from 4.1 percent of GNP in 1985 to 3.0 percent in 1989. Thus, one might be tempted to conclude hat the response to TRA86 has confirmed the conventional wisdom that the cost of capital is unimportant for empirical modeling, and that new theories of investment must be sought which more 'Columbia University, New York, NY 10027. 243

adequately explain the lack of investment response to tax policy. Auerbach and Hassett (1991), using industry and asset level data, have recently shown that this conclusion is premature. They have argued that the TRA86 experience provides both a natural experiment for evaluating the impact of tax reform on investment and a clue to the possible econometric causes of the historically weak empirical relationship between the cost of capital and investment. They develop an econometric technique for estimating the effect of changes in tax policy on investment which relates forecast errors in investment to forecast errors in the cost of capital.' Using aggregate asset and industry level panels, they find a large, significant effect of the cost of capital on equipment investment, but not on structures investment. In this paper, we discuss this technique and use it to analyze the responses of more disaggregated firm level investment to TRA86. A companion paper, Cummins and Hassett (1992), provides a more extensive discussion of the theoretical issues, and develops the econometric methodology more rigorously. Using Compustat data, we find strong confirmation of the results found in Auerbach and Hassett (1991). We conclude that there is a significant relationship between the cost of capital and equipment investment. Highlighting the power of the approach adopted here, we also find, for the first time in the literature, a strong relationship between the cost of capital and structures investment. In the first section we provide a discussion of the important tax changes affecting investment introduced in TRAS6. In the next section, we discuss the investment model we seek to estimate and the econometric techniques used. The third section describes the data. The final two sections review our empirical results and discuss conclusions.


244

NATIONAL TAX JOURNAL

The Tax Reform of 1986 TRA86 contained several provisions which potentially could affect investment. The investment tax credit was repealed, depreciation lifetimes were generally extended, and the corporate tax rate was reduced. From the outset, policymakers viewed the Act as a fundamental and long lasting reform. This view may have been built into the expectations of investors, whose plans in the past had been disrupted virtually every third year by one tax policy change or another, and who frequently found themselves either racing to invest in order to claim an investment tax credit before the rule expired, or delaying investment so as to qualify for a credit after it became available. The perceived permanence of the tax change is an important aspect of the Act, which allows us to identify the effect of tax changes on investment. We return to this point below. The change in depreciation lifetimes had the potential to significantly alter firms' incentives to invest. Equipment lives, which had been five years for most types of equipment, and three years for light equipment, increased in most cases to seven years. Business structures, which had been depreciated over 19 years prior to TRA86, had their asset lives increased to 31.5 years. The investment tax credit for purchases of new equipment was reduced from 10 percent to zero. The corporate tax rate was reduced from 46 percent to 40 percent in 1987 and 34 percent for 1988 and thereafter. On net the removal of the investment tax cr;dit and lengthening of asset lives increased the tax burden of purchases of equipment. Structures, on the other hand, which did not qualify for the investment tax credit prior to 1986 had their marginal tax burden reduced, as the effect of decreasing the corporate rate overpowered the increase in the cost of investment due to longer asset lives.' We now explore a model which relates tax changes such as these to a firm's investment decision and discuss an econometric strategy for estimating the effect of tax policy in this setting.

[Vol. XLV

The Empirical Relationship Between Investxnent and the Cost of Capital Consider an investment model with perfect certainty and no adjustment cost, As is well known, under these circum. stances, an optimizing firm sets the mar. ginal product of capital equal to a shadow price of capital: c@q(l-F)

r+b-

q(l - r)

where q is the price of capital goods relative to output, r is the real interest rate, 8 is the rate of economic depreciation,,r @ the corporate tax rate, A is a differencing operator, and F is the present value of tax savings from depreciation and other investment incentives. If there is also an in. vestment tax credit, for example, r at time period t is: rt = kt + (1 + r + 7r)-tT. D. (s - t).

(2)

Here, kt is the investment tax credit, D(a) is the depreciation allowance permitted an asset of age a, and this is discounted at a nominal rate which includes the inflation rate, 7r. If we add adjustment costs and uncertainty to this model, then the firm bases current investment on current and future expected values of a term similar to (1). For an individual firm, the in. vestment rule becomes: iit = @t+ (@Et Kit-,

p- c@

(3)

where i may index the industry, or the as. set if asset investment is separable, t indexes time, p is a discount rate which may depend on technology parameters, and C, is defined as in (2) but is subscripted in recognition of the fact that its components may time vary. 5


XLV

No. 31

veen w

Auerbach and Hassett (1992) have estirated (3) with aggregate time series data W have found that the cost of capital does affect investment significantly, although model validation attempts were unable to rule out ad hoc alternative specifications, and the relative size of the cost of capital coefficient was small, implying implausibly large adjustment costs. There are many reasons why the neoclassical model might not be expected to fare well using aggregate data. The compwition of the aggregate capital stock thenges rapidly over time. Computer investment for example has risen from .9 percent of total investment in 1953 to 25 percent of real investment in 1989. This rapid compositional change induces high levels of serial correlation in the observed ratio of investment to capital, making it extremely likely that other serially correlated variables, e.g. cash flow, will appeu correlated with investment, and that the relatively serially uncorrelated cost of capital will explain little of the time series variation of investment, even if one is careful to adjust the cost of capital over time for the changing composition of investment. If one adds to this the necessity of using instrumental variables in order to construct the expected cost of capital terms on the right hand side of (3), and the fact that tax changes tend to be discrete and difficult to linearly predict, then it is unsurprising that a convincing empirical relationship has been elusive. If we seek to estimate (3) with panel data, we run into several other serious problems. Since the cost of capital depends on current and expected future taxes and productivity (Auerbach and Hassett, 1992) which are very difficult to measure at the level of individual machines or firms, one quickly discovers that the measurement problems are significantly worse at this level of aggregation and that the relationship, at least using traditional techniques, is even more elusive. In recognition of the serious difficulties an econometrician faces in observing the unr cost of capital, one may alternatively consider treating the cost of capital as mrely" observable and develop an estimation strategy which allows for identi.

with cost. leummar. adow

rel. rate, T is

Lcing f tax r inn intime

(2) D(a) tted nted in !Osts the rent sim. in-

(3) as. inmay i C, i iu IPO-

JASON

G. CUMMINS

AND

KEVIN

A. HASSETT

245

fication of the cost of capital effect when this is the case. Since investment depends on future expected costs of capital which are extraordinarily difficult to predict in an uncertain policy environment, one could credibly argue that firms' expectations about future tax policy cannot be described well with linear projection techniques before 1986. After 1986, linear projection techniques are unnecessary, for future tax policy is known both to the investor and the econometrician following the "permanent" tax reform. To see how this can lead to identification of the tax effects, consider a permanent change in r from above. In this case, the tax components of C. are constant through time and, assuming that r and q are fixed, we can factor out the cost of capital term. Equation (3) then reduces to the product of a traditional user cost term, and a forward looking infinite sum of current and future expected weights, which again will depend on future expected productivity. Separating out the tax term leads to the expression:

Kit-,

Xitp

+ Cit'y

+

Eit

(4)

where Xit is a vector of instruments in the current information set with which we construct forecasts of future productivity (X may well include lagged values of investment) and Cit, which reduces to: (I - rit) r + bi -

(I

rt)

A(i - rit (i - rit) (5)

given the above assumptions. Cit may vary over firms or assets even without an investment tax credit because of differing depreciation. Since we are concerned here with isolating the effects of tax policy changes on investment, we will use (5) as our measure of the cost of capital below, with an assumed constant real interest rate of .04.


246 If we cannot stead estimate:

NATIONAL TAX JOURNAL observe Cit, then we in-

[Vol. XLV

asset is booming. If we remove an inves@ ment tax credit for asset j, increasing the cost of capital, the boom in investment in asset j will be softened, but observed iniit Xitb + (oit. (6 vestment may still inaease, if not as much Kit- I @ as it would have had the ITC remained in place. Even if we cannot explain the boom Since there is an omitted variables in asset j well with a structural model, problem, the estimate of b will not con- because of the difficulty in observing the verge to P, but rather to P + 7ry, where it forces driving the boom, we might still be is the vector of coefficients from the auxable to identify the effect of the ITC by iliary regression of C on X. This obserusing time series techniques to forecast vation leads to a strategy for identificainvestment in asset j past the tax change, tion of -y. and then establishing whether the size and Assume that we have a T period panel direction of the forecast error (wit) is conwith observations spanning a permanent sistent with the theoretical impact of the tax reform in the beginning of period T- new tax -i(tit). If we have many assets with k. Notice, using both (4) and (5), that in different tax treatments, then we can period T-k the forecast error is: construct a cross section of forecast emn and relate these to changes in taxes in or. (7) der to estimate the structural response of Wit-k @ Eit-k + tit-ky investment to tax policy. where 4it-k is (Cit-k - Xil-krr). If we beWe proceed by applying this approad lieve we can observe Cil-k with confidence to Compustat firm level data which, along for many i in period T-k, as might be the with our measure of the costs of capital, case just after a tax reform which is bewe describe in detail in the next section. lieved to be permanent, we can utilize (7) to estimate -y with a two-step process. In the first step, estimate b and ff with sep- The Data arate regressions for Iit/Kit-I and Cit using both the time series and cross section As already discussed, Auerbach and variation for identification. Second, use the Hassett (1991) applied the above tech. estimated coefficients to calculate o)it and nique to a panel of aggregate investment @itand construct a second k period panel data supplied by John Musgrave of BEA with which we can estimate (7). We note and discovered a large effect of the cost of in passing that Pagan (1984) has shown capital on equipment investment. Many that the second stage standard errors are of the asset investment series they studcorrect in this case, since wit and @itare ied exhibited powerful and even accelertrue "surprise" terms. ating trends (e.g. investment in comput. Notice that the identification of the ef- ing equipment), and the VAR investment fect occurs only when we experience a pe- forecast errors (wit) used to estimate (7) riod where the agent observes a change in were at times quite large. This has lead the cost of capital which cannot be preto the criticism that implausibly large indicted well with linear econometric techvestment forecast errors are drivi niques. If the econometrician could per - results.' A natural response to [email protected],s ligistle. fectly predict the cost of capital, then the estimate (7) on a completely different data effect would be unidentified, as should be set. We turn to a panel of firms from the obvious from the definition of t.' Compustat industrial file 1989 which pro. In words, the approach is as follows. vides data from 1970-1989. This panel ha Consider a disaggregated investment se- an advantage over the BEA data in that ries, asset j. Because of rapid unobserved there are many more cross sectional ob. productivity changes or changes in the servations, which should substantially profitability in the industry which uses j improve the power of our test and reduce heavily, assume that investment in the the likelihood that a small number of ex-


IV

No. 31

JASON G. CUMMINS

sthe in n-

AND KEVIN A. HASSETT

TABLE I EQUIPMENT

INVESTMENT RESIDUALS BY INDUSTRY

Ch

in )M el, he be by kst pt nd )nhe ith an )rs DrOf

Agriculture Mining Construction Manufacturing Trans.,Comm. Wholesale

ich ng al, )n.

Total

nd

Note:

)nt RA Of .ny iderut@nt (7) ,ad Inle to ata le roiumm ut obIly ice eux-

247

FIRE

Util. and Retail Trade

and Services

Standard

deviations

IN

1987,

With Trend

Without Trend

.179 (.396) -.107 (.240) .041 (.047) -.020 (.221) -.115 (.274) -.030 (.238) -.045 (.296) -.027 (.234)

-.113 (.358) -.241 (.280) -.066 (.045) -.010 (.210) -.082 (.231) .004 (.215) -.070 (.216) -.019 (.215)

in parentheses.

treme observations will influence the results. The variables used are defined as follows. Gross equipment investment is the change in the net stock of machinery and equipment (Compustat data item #156) grossed up by the appropriate asset depreciation rate reported in Jorgenson and Sullivan (1981) and Jorgenson and Yun (1989). Gross structures investment is similarly constructed as the change in the net stock of buildings (Compustat data item #155) grossed up by the appropriate asset depreciation rate, again as reported in Jorgenson and Sullivan (1981) and Jorgenson and Yun (1989).a The above variables are divided by their own beginning of period capital stocks. Our X vector of explanatory variables includes la@ values of investment, a time trend,

and cash flow, which is defined as after tax income before extraordinary items plus depreciation and amortization (Compustat data item # 18 plus data item # 14).9 Cash flow is also divided by the appropriate beginning of period capital stock. Cost of capital measures for equipment and structures are constructed for each firm by aggregating the 36 asset level costs of capital derived in Auerbach and Hassett (1991) using each firm's industry specific capital stock weights."

Results We first estimate linear forecasting equations for the sample period 19701985, and use these to generate forecasts of investment and the cost of capital past TRA86. In order to allow for substantial


248

NATIONAL TAX JOURNAL TABLE STRUCTURES

Mining Construction Manufacturing Trans.,Comm. Wholesale

Util.

and Retail Trade

FIRE and Services Total

Note: Standard

2

INVESTMENT RESIDUALS BY INDUSTRY

Agriculture

deviations

[Vol. XLV

IN

1987,

With Trend

Without Trend

-.100 (.236) -.060 (.259) .012 (.237) .006 (.194) -.043

-.120 (.212) -.141 (.211) .017 (.210) .028 (.171) -.018

(.151)

(.195)

-.027 (.212) -.084 (.228) -.010 (.201)

.016 (.227) -.049 (.190) .016 (.185)

in parentheses.

industry heterogeneity, we estimate separate forecasting equations for each fourdigit industry, and then pool the resulting 1987 forecast errors to construct the post tax reform cross section needed to estimate -y. Tables 1 and 2 provide summary statistics for the forecast errors for equipment and structures investment, by industry. The firm level equipment forecast

viations. The structures forecasts, reported in Table 2, generally follow the same pattern as those for equipment, although the point estimates for the mean appear to be more sensitive to inclusion of the trend term. Given this apparent sensitivity, we report second stage regres. sions for both sets of forecast errors be- I low. It is important to note, however, that the two-step procedure applied here doeo

error average is -.027 if we use a forecasting specification without a trend, and -.019 if we use a specification with a trend. Both numbers suggest that investment was on average lower than would have been predicted using the pre-tax reform information set, although the presence of severe outliers is evidenced both by the size of the errors for mining and agriculture, and the large standard de-

not rely upon precise first stage forecasts. The imprecision of these forecasts will only lead to low second stage R"s. Table 3 reports the estimation of equation (7) using the 1987 forecast errors based on projection equations both with I and without trends. Columns 1 and 3 report the estimates of the basic model. Given the theory, the equations fit remarkably well. The constant term, which


No. 31

JASON G. CUMMINS AND KEVIN A. HASSETT

249

TABLE 3 EXPLAINING CROSS-SECTION BEHAVIOR IN 1987,

Trend in lst Stage Explanatory Variable: Constant User Cost Cash Flow ]ft2

Note: t-statistics

EQUIPMENT INVESTMENT ASSET CLASS

By

Yes

Yes

No

No

.025 (1.03) -1.06 (-2.36) .01

.008 (.324) -1.06 (-2.37) .034 (2.64) .02

.011 (.638) -1.25 (-2.00)

-.018 (-1.01) -1.33 (-2.14) .068 (5.03) .05

.01

in parentheses.

the neoclassical model would predict to be zero, is insignificant in both specificadons, and the coefficient on the cost of capital, -y, is significant at a 95% confidence level in both cases and ranges from -1.06 to -1.25. The structures estimabon is reported in Table 4. Once again, the equations seem to fit very well. The constant terms in columns 1 and 3 are inswicant, and the coefficients on the cost

cost of capital term (.188 for equipment, .111 for structures) and @ is the coefficient of the adjustment cost function. 12 Summers'(1981) estimate of * is roughly 32, an implausibly large estimate which suggests that the marginal adjustment cost for installing a new machine may exceed the purchase price of the machine. Our estimates suggest substantially more plausible values of @. Our best estimate

of capital are large and significant, ranging from -.575 to -.712. These point esfimates imply an elasticity of gross investment to the cost of capital of roughly -1.1 for equipment and -1.2 for structures. These are slightly larger than the elasticities reported in Auerbach and Hassett (1991), and are substantially larger than estimates reported in the past." The large response of structures investment to the cost of capital is unprecedented, and suggests that the gains from turning to firm level investment data have been large. In order to give a structural interpretation to the parameters, we note that under constant returns to scale the coefficient on the cost of capital converges to 1/ (c$o), where c* is the mean value of the

of @ for equipment is roughly 4.6 and that %@for structures is roughly 14. These estimates, combined with average net investment numbers from our sample (.061 for equipment, .025 for structures) imply marginal adjustment costs substantially less than 1: 1 dollar of equipment investment would lead to roughly 28 cents of adjustment costs, 1 dollar of structures investment would lead to roughly 35 cents Costs. 13 in adjustment One criticism of cost of capital models has traditionally been that variables which should not matter given the theory are frequently found to explain investment over and above the effects of the cost of capital. Columns 2 and 4 in both tables report alternative estimates which include cash flow surprises in (7)."' Auer-


250

NATIONAL

TAX JOURNAL

TABLE

[Vol. XLV

4

EXPLAINING CROSS-SECTION STRUCTURES INVESTMENT BEHAVIOR IN 1987, By ASSET CLASS

Trend in Ist Stage

Yes

Yes

No

No

Explanatory Variable: Constant

-.005

-.015

.007

-.001

User Cost

(-.543) -.575

(-1.42) -.584

(.805) -.712

(-.884) -.702

(-2.38)

(-2.19)

(-2.38)

(-2.36)

-

.01

.010 (1.61) .01

.015 (2.63) .02

Cash

Flow P2

Note:

t-statistics

.01

in parentheses.

bach and Hassett (1991) found little effect of cash flow variables in the second stage. We find in all cases that the cash flow variable is significant, although the coefficients are significantly smaller than has been found in the past. It is important to note that cash flow surprises would enter significantly if they were correlated with future investment conditions, and that the significance here is difficult to evaluate. On the other hand, it is only fair to note that the criticism that has been leveled against the parsimony of the neoclassical model in the past may apply here also. More formal treatment of these important model validation issues is the subject of our current research.

structural parameters identified here im. ply reasonable adjustment costs on the order of 25% of the purchase price of a new machine. Our evidence suggests that measurement problems, which are overcome here because of the unique features of the natural experiment provide by TRA86, may have contributed signifi. cantly to past failures to identify the im. r pact of changes in the cost of capital on investment.

Conclusion

ENDNOTES **We thank participants of the NTA Symposium, Alan Auerbach, Bob Chirinko, and Glenn Hubbard for helpful comments and suggestions. Any remain. in errors are ours. f'See Blanchard (1986), p. 153 and Shapiro (1986), P. 111. 2,@ Chirinko (1986, 1987) and Bemanke, Bohn, aW Reiss (1988).

We use firm level panel data to explore the relationship between the cost of capital and investment. Unlike previous work, we find that cost of capital innovations contributes to investment forecast innovations. We find that the effect of the cost of capital on equipment and structures investment is significant both economically and statistically. In addition, the

3Bosworth (1985) pursued a somewhat relaw strategy. He also studied investment forecast inno. vations following tax changes. His approach diffen from that adopted here, and these differences an & cussed below. 'Some utilities structures did qualify for the M. and for these the reform increased the marginal tax burden on new investment. 'See Auerbach and Hines (1988) and Auerbach aW Hassett (1992) for detailed derivations. The C tem may vary slightly depending on the specific production model chosen.


No. 31

JASON G. CUMMINS

AND KEVIN A. HASSETT

'Of mum, we assume the firm is still capable of faming its own expectations. It is important to recopize that the forecast error of the cost of capital is rm a true surprise to the investor, who, we assume, can observe future tax policy perfectly after 1986, and who alters his investment plans accordingly. We asswne only that the econometrician has difriculty predkiing the tax changes with linear projection techniques, using the pre tax reform information set as h4 instrument base. Together, these assumptions lead toour eatunation strategy, which is different from that of Bosworth (1985), who did not include the x[I term whenconstructing his cost of capital innovations. This omission makes his estimates difficult to interpret. 7SeeBosworth and Burtless (1992), p. 17 for a discmion of this point. Ve select the net to gross markup by taking the weighted sum of asset depreciation rates by industry. Our results are not sensitive to changes in this step. Ve also experimented with inclusion of macro variables such as the price of investment goods, oil pnees, and the interest rate, but found that the inclusion of these had little impact on the results reWW below, perhaps because these effects are capku-edin the time trend or the industry level variation Inparameters which we allow in the first step. l'In constructing this panel, we discovered that there m in outlier problem, with a few firms having very krp ratios of investment to capital. Such observaUousare deleted if the ratio of investment to capital awk is greater than I or less than negative 1. Very few observations are deleted as a result, but these outliers can effect the coefficient estimates if not the spirit of the results. In addition, observations are also Avietedif the firm's capital stock is less than $1 mill=, as these firms may either be near bankruptcy or jug starting up, and we found very large forecast ermrs in both directions to be possible. Our results are ad sensitive to the particular value of the cutoffs we 6m. "One of the larger previous estimates, by Feldstein 11963), was .4. I%w Auerbach and Hassett (1992) for a discussion wWderivation of this result. '%see estimates are the derivatives of a quadratic &4utment cost function evaluated at the implied valuesof @and the sample averages of net I/K (.061 equipment and .025 for structures). %w Fazarri, Hubbard, and Peterson (1987). REFERENCES Abel,A. B., 1983, @'Optimal Investment Under Unoertainty," American Economic Review 73, 228-233. Abel, A. B. and 0. Blanchard, 1986, "The Present Value of Profits and Cyclical Movements in Investment," Econometrica 54, 249-273. Auerbach, A. J., 1983, "Taxation, Corporate Financial Policy and the Cost of Capital," Journal of Economic Literature 21, 905-940. Auerbach, A. J., 1989, "Tax Refbrm and Adjustment C4&: The Impact on Investment and Market Value," International Economic Review 30, 939-962. Auerbach, A. J. and K. Hassett, 1992 "Tax Policy and Business Fixed Investment in the United States," Journal of Public Economics 47, 141- 170. Auerbach, A. J. and K. Hassett, 1991, "Recent U.S.

251

Investment Behavior and the Tax Reform Act of 1986: A Disaggregate View," Carnegie Rochester Conference Series on Public Policy 35, 185-215. Auerbach, A. J. and J. Hines, 1987, "Anticipated Tax Changes and the Timing of Investment," in: M. Feldstein, ed., The Effects of Taxation on Capital Accumulation (University of Chicago Press, Chicago) 163-196. Bemanke, B., H. Bohn and P. Reiss, 1988, "Alternative Non-Nested Specification Tests of Time Series Investment Models," Journal of Econometrics 37,1-38. Blanchard, 0., 1986, "Comment," Brookings Papers on Economic Activity 1, 153-158. Bosworth, B. and G. Burtless 1992, "Effects of Tax Reform on Labor Supply, Investment and Saving" The Journal of Economic Perspectives 6 1, 3-26. Bosworth, B., 1985, "Taxes and the Investment Recovery," Brookings Papers on Economic Activity 16, 1-38. Chirinko, R. S. 1987, "The Ineffectiveness of Effective Tax Rates on Business Investment: A Critique of Feldstein's Fisher-Schultz Lecture," Journal of Public Economics 32, 369-387. Chirinko, Robert S. 1986, "Business Investment and Tax Policy: A Perspective on Existing Models and Empirical Results," National Tax Journal 39, 137155. Cummins, Jason G. and K. Hassett 1992, "Taxes and Investment: An Innovation Based Approach," working paper, Columbia University, May. Fazzari, S. M., R. G. Hubbard and B. C. Peterson, 1988, "Financing Constraints and Corporate Investment," Brookings Papers on Economic Activity 19 1,141-196 Feldstein, M , 1982, "Inflation, Tax Rules and Investment: Some Econometric Evidence," Econometrica 50, 825-862, Hall, R. E. and D. W. Jorgenson, 1967, "Tax Policy and Investment Behavior," American Economic Review 57, 391-414. Hayashi, F., 1982, "Tobin's Marginal and Average q: A Neoclassical Interpretation," Econometrica 50, 213-224. Jorgenson, Dale W., 1963, "Capital Theory and Investment Behavior," American Economic Review 53, 247-59. Jorgenson, Dale W. and Martin A. Sullivan, 1981, "Inflation and Corporate Capital Recovery," in: C. R. Hulten, ed., Depreciation, Inflation, and the Taxation of Income from Capital (Washington: Urban Institute), 178-238. Jorgenson, Dale W. and Kun-Young Yun, 1989, "Tax Policy and the Cost of Capital," Harvard Institute of Economic Research Paper #1465. Pagan, Adrian, 1986, "Two Stage and Related Estimators and Their Applications," Review of Economic Studies 63, 517-638. Pagan, Adrian, 1984, "Econometric Issues in the Analysis of Regression with Generated Regressors," International Economic Review 25, 221-47. Shapiro, M., 1986, "Investment, Output and the Cost of Capital," Brookings Papers on Economic Activity, 1, 111-152. Summers, L., 1981, "Taxation and Corporate Investment: A q-Theory Approach," Brookings Papers on Economic Activity 12, 67-127.