food, and 4.46 percent for computers, to 0.06 percent for nonferrous metals, 0.24 percent for ferrous metals, 0.43 percent for lumber, wood and furniture, and 0.62 percent for motor vehicles and parts.

Table 3.12 makes clear the heavy concentration of R&D expenditures among the largest companies as measured by net sales. In our set of 592 firms, $20 billion of the $31.9 billion total R&D expenditures in 1982 were made by the 52 firms with net sales of $5 billion and over. As may be noted in Table 3.13 these firms, which accounted for 66.8 percent of 1982 sales of all of those in the compilation, accounted for only 58.4 percent of the total credit currently claimed. Thus, as seen in Table 3.14, since the percent increases in R&D were least among the firms with sales of $5 billion and over, their shares of the total tentative credit of the firms in the compilation were considerably less than their shares of total R&D: 56.5 percent and 57.3 percent for 1981 and 1982, respectively, for the tentative credit, as against 63.3 percent and 62.6 percent for 1981 and 1982 for R&D.

A concentration of R&D expenditures among the largest firms is also indicated in Table 3.15 where companies are categorized in terms of number of employees. Here we note that 120 companies with 25,000 or more employees, or 20.3 percent of a total of 592 companies in the compilation, accounted for 1982 employment of 8,645,000 or 76.8 percent of total employment, and 1982 R&D expenditures of $25,869 million or 81.0 percent of total R&D. We can see further, in Table 3.16, that the companies with employment of 25,000 or over accounted for 82.0 percent of the total credit currently claimed by firms in our compliation. This last represents a higher proportion than for the tentative credit, which was some 77 and 78 percent in 1981 and 1982, respectively, as shown in Table 3.17. There, we also note that the rates of growth of R&D, both from 1980 to 1981 and from 1981 to 1982 seemed generally lower in the largest firms.

The tax credit estimates presented thus far from Compustat data have all been constructed on the assumption of a constant eligibility ratio. In the special supplementary R&D survey run for us by McGraw-Hill, however, we have received responses regarding qualified expenditures and total company funds for research and development which enable us to estimate eligibility ratios industry-by-industry and year-by-year. We have utilized annual aggregate eligibility ratios derived from these McGraw-Hill responses to estimate qualified expenditures and consequent tax credits with the Compustat data.

Since the McGraw-Hill responses suggest a relatively low eligibility ratio in 1980 and a relatively high one in 1981, estimates of tax credits which are based upon them are generally higher for 1981 than those utilizing a constant eligibility ratio. In Table 3.18, we note that the tentative credit claimed for 1981, making use of these variable eligibility ratios is $586 million, rather than the $465 million estimated on the assumption of a constant eligibility ratio. The credit currently claimed for 1981 rises to $507 million from the $402 million calculated with the constant eligibility ratio. The estimates of $1,028 million and $670 million for tentative credit and credit currently claimed in 1982, respectively, differ little, however, from the estimates prepared on the assumption of a constant eligibility ratio. Table 3.19 presents a distribution of eligible R&D and the tentative credit and percent increases in each by category of 1982 R&D, applying these variable aggregate eligibility ratios. It is thus analogous to Table 3.5, calculated with a constant eligibility ratio. Similarly, Table 3.20, cross-classifying firms by their amount-of-1982-R&D category and their 1981 tax credit situation, it is analogous to Table 3.6, but uses the variable eligibility ratios.

We have also prepared estimates of the tax credit from the aggregates of Compustat R&D expenditures for the years 1980, 1981 and 1982, based to the 1980 NSF total of Company Funds for R&D. Working with the aggregates implies ignoring the limitations: that credits are not negative for companies for which qualifying expenditures are less than their base and that the base can never be less than 50 percent of current qualifying expenditures. The first limitation implies that our aggregate estimates will be less than estimates from individual company expenditures because in the aggregate firms with expenditrues below base will offset some of those with expenditures above base. Failure to take account of the 50 percent base limitation, however, will tend to make the aggregate estimates higher.

In fact, the estimates from aggregates, ignoring limitations, tend to be somewhat less than those calculated by summing estimated credits for individual firms. This is particularly true in 1982, where a larger proportion of firms showed qualifying expenditures below base.

Assuming the usual constant eligibility ratio, the estimate of the tentative tax credit from the aggregates, ignoring limitations, was $451 million for 1981, as against $465 million from the individual firm data, as shown in Table 5.1. For 1982,

the estimate from the aggregates is $989 million, as against $1,046 million in the data from individual firms.

Working with the variable eligibility ratios taken from the McGraw-Hill supplementary R&D survey, our estimate from the aggregates of the credit without limitations is $579 million in 1981, almost identical with the $586 million estimate from the individual firm data. For 1982, our estimate from the aggregates is $960 million, as against $1,028 million from the individual firm data.

We have undertaken some preliminary exploration of cross-sectional relations among R&D, sales and net income. Regressions of R&D on lagged R&D, changes in R&D, sales, sales changes and net income are shown in Table 3.21 for 1980, 1981 and 1982.

The cross-sectional differences in R&D are overwhelmingly explained by previous R&D and the previous change in R&D. To the extent that the tax credit is encouraging firms with increasing R&D to increase their R&D still more, we might expect higher coefficients of the lagged change-in-R&D variable in 1981 and 1982 than in 1980. It is apparent that this is not so, although the change-in-R&D variable does have a higher coefficient in 1982, when the credit was in effect for the entire year and increased expenditures did less to lower the future base and hence reduce future credits than they did in 1981.

One might have expected that if the tax credit were effective in increasing R&D expenditures the coefficients of the current net income variable, which would include tax credits, would be higher in 1981 and 1982 than in 1980. This too was not the case.

We observe also that both the sales change and net income variables are positively related to R&D expenditures, suggesting accelerator and profitability determinants of R&D spending. To the extent that these determinants were dominant, the possible role of marginal tax credits might be reduced.

As we have noted in section I, while the tax credit may have some effect of increasing R&D expenditures for firms over base, it may also induce firms under base to lower qualified R&D spending in the interest of securing credits in the future. To test this hypothesis, we ahve compared R&D spending below and above "base" for the years 1976 to 1980, before the incremental tax credit was in effect and the years 1981 and 1982 when the credit was operative. We define the "base" for all years alternatively as that applicable for 1981, that is, half of the (qualified) R&D expenditures of the previous year, and as that in effect for 1982, half of the (qualified) R&D expenditures of the two previous years. In the comparisons in terms of the 1981-type base, we take as R&D expenditures qualified for the credit the imputation of secondhalf-of-year expenditures defined in (3.1).

As may be seen in Table 3.22, we can find no evidence of effect of the incremental tax credit in this test. The differences between the mean excess of R&D over "base" and the mean shortfall of R&D below "base" are expressed as percents of previous R&D. They turn out to be no greater, and indeed somewhat smaller, by either measure of base, in the years 1981 and 1982 than in the five years before 1981 when there was no incremental tax credit.

A substantial amount of further statistical analysis of the Compustat data, utilizing both time series and cross sections, will be undertaken in an effort to detect possible effects of the tax credit on R&D.

IV. MC GRAW-HILL DATA

The special supplementary R&D survey undertaken on our behalf by McGrawHill brought responses from 81 firms who indicated the R&D tax credit that they claimed in 1981 along with the R&D expenditures. Blowing up these responses in accordance with (the reciprocal of) the ratio of the sum of R&D expenditures reported and the NSF estimate of company funds for R&D in each industry for 1980 (NSF, 1983b, Table B−5), yields an estimate of the tax credit of $684 million, as shown in Table 4.1.

It is unfortunately not clear whether the tax credits reported were restricted to those currently claimed or included as well credits carried back or forward because of lack of current federal taxes payable. Interpreting these answers as relating to the "tentative credit," whether or not there were tax liabilities, the resultant estimate is still high in comparison with the estimate of $465 million from Compustat data and even with the estimates from OTA data. There the tentative credit, it may be recalled, came to $408 million from a sample thought to be 80 percent complete although, as noted above, Treasury Assistant Secretary Chapoton presented an estimate of $600 million.

The blowup of 77 McGraw-Hill firms reporting tax credits for 1982 came to only $884 million, however, which is less than the $1,046 million tentative credit estimated from the Compustat tape, although more than the $682 million currently claimed. In terms of internal consistency, the rate of growth of the reported credit from 1981 to 1982, only 29 percent, is extremely low in relation to both analytical predictions and calculations from the Compustat data on assumptions of constant eligibility ratios.

Results substantially similar to those in direct reports of the credit are derived when we calculate it on the basis of qualified expenditures, reported by a slightly smaller number of respondents. These expenditures when blown up to national aggregates, as shown in Table 4.2, would appear to have grown only by 7.89 percent in 1981 and 7.56 percent in 1982. The survey questionnaire asked, however, for "qualified research expenses used in base period calculations" and further stipulated, "Please enter annual or annualized amounts; for the 1981 credit only half-year base percents were used." Thus, as in our estimates with the Compustat annual data, we imputed second-half-of-year figures from year-over-year growth. In this instance we compute 1981 qualified R&D expenditures after June 30 as (4.1) RDQ81II=RDQ81/[RDQ80/RDQ81)-5 1], where RDQ81 and RDQ80 are qualified R&D expenditures in 1981 and 1980, respectively.

As indicated in Table 4.3, with these imputations for 1981 expenditures after June 30, the tentative credit in 1981 comes to $634 million. It then rises only to $907 million in 1982, falls to $809 million in 1983 and rises again to $952 million in 1984. In Table 4.4 we present total R&D projections derived from responses in the R&D section of the main McGraw-Hill capital expenditure survey. In Table 4.5, assuming again a constant eligibility ratio of 0.6329144 for all years, we show tentative credits from a sample of 205 firms blown up to NSF nationwide totals. For 1981, the tentative credit estimate of $442 million is close to the Compustat figure of $465 million but well below the $634 million from the reported qualifying expenditures of the smaller number of 78 firms in the McGraw-Hill supplementary R&D survey. For 1982, however, our projected tentative credit from the 205 firms in the capital expenditure surey is $974 million, consistent on the one hand with the greater lag in the 1982 base and, on the other, the decline in the year-over-year rate of growth of R&D expenditures, from 16.51 percent from 1980 to 1981 to 11.68 percent from 1981 to 1982. The forecast expenditures in the McGraw-Hill surveys would then indicate additional growth in the tentative credit to $1,477 million in 1983, essentially because of a further lengthening in the lag of the base, but a moderate decline to $1,308 million in 1984 and a further decline to $1,238 million in 1985, reflecting the cumulative consequences of a slackened projected growth of R&D.

We next assume that eligibility ratios for all firms were identical in each year but changed from year-to-year as did the ratios of the aggregates of qualifying expenditures and total company-funded expenditures from intersecting supplementaryR&D-survey subsets of 74 firms reporting this information for the years 1980 to 1982 and 52 firms reporting for the year 1982 to 1984. Results of the consequent calculations, shown in Table 4.6, suggest a tentative tax credit of $586 million in 1981, $954 million in 1982, $1,388 million in 1983, $1,307 million in 1984 and $1,281 million in 1985 (this last on the assumption of the same eligibility ratio in 1985 as in 1984). We have utilized NSF estimates of industrial R&D for the years 1980 through 1984 to supplement the estimates of future tax credits from the McGraw-Hill data. The results are estimates of the credit without limitations calculated from these aggregative NSF data as shown in Table 4.7. We may note first that the estimates from these aggregates for the years 1981 and 1982 are quite close to the Compustat estimates for these years. Estimates for future years of course depend critically on forecast growth in nominal R&D. Estimates from the NSF projections, assuming our constant eligibility ratio, were $491 million of tax credit without limitations in 1981, $1,019 million in 1982, $1,354 million in 1983 and $1,513 million in 1984. Estimates prepared on the basis of the variable eligibility ratios from the McGraw-Hill supplementary R&D survey prove sensitive to the high eligibility ratio in 1981. The estimated credit without limitations would rise from $620 million in 1981 to $995 million in 1982, more slowly to $1,197 million in 1983 and then more rapidly again to $1,529 million in 1984.

Responses to the McGraw-Hill supplementary R&D survey permit certain suggestive tests of the effect of the incremental tax credit on R&D expenditures. As shown in Table 4.8, among 71 reporting firms, the mean growth in total R&D in 1981 was 15.8 percent. For 64 firms reporting, the ratio of the change in qualifying R&D to total previous R&D was 21.3 percent. The similar ratio for nonqualifying expenditures was -4.4 percent. If we were to treat these as a sample, the differences in rates of change in qualifying and nonqualifying expenditures would clearly be sta

tistically significant as the standard error of that difference of 25.7 percent is only 5.0 percent.

The regression coefficients indicate an analogous story for 1981. The cross-section coefficient of 5.051 of the change in qualified expenditures on the tax credit is clearly significant. The regression coefficient of the tax credit for nonqualified R&D expenditures is in fact negative although the difference from zero is not statistically significant.

The differences for 1982 are not nearly as striking. The mean rate of growth of R&D was only 10.2 percent. The change in qualified R&D divided by lagged total R&D was 8.6 percent while a corresponding ratio for nonqualified R&D was 2.1 percent. Where there had been a difference of 25.7 percent in 1981 there was thus a difference of only 6.5 percent in 1982. The cross-section regression coefficient of qualified R&D on the tax credit was a smaller but still highly significant 3.467 while that for nonqualified R&D was virtually zero.

What these relationships suggest is that there was a substantial surge in R&D reported to be qualified in 1981. The tax credit was of course, as its definition requires, greater the greater was the growth in qualified R&D. That growth in qualified R&D, however, was considerably offset by a reduction in non-qualified R&D. One might infer that the tax credit was effective in inducing an increase in qualified R&D and a substitution of qualified R&D for non-qualified R&D. One might also infer, alternatively, that there is little evidence of any real impact on R&D, but rather a reclassification of R&D so that a considerably larger portion of the total was classified as qualified in 1981 than in 1980.

With the big increase in the eligibility ratio in 1981, however, there was little room for further growth in qualified R&D relative to total R&D. Hence the difference in rates of increase between qualified and nonqualified R&D very greatly narrowed in 1982.

V. CONCLUSIONS

Our analytical section has indicated a number of the prime characteristics of the incremental tax credit for R&D.

For one thing, it can be expected to grow rapidly in amount over its first three years, 1981, 1982 and 1983. Much of the growth from 1981 to 1982 is obviously foreordained by the fact that the 1981 credit applies only to expenditures after June 30. The progressive lengthening of the lag of the base over all of the first three years, however, offers predictable consequences of increases in the credit, magnified in the context of generally growing expenditures.

Second, the amount of the credit is senstitive to nominal rates of growth of R&D, whether due to real growth or inflation. A credit may be earned by firms which are merely paying more for the same or even less R&D than previously.

Third, the amount of credit is sensitive to changes in the eligibility ratio, that is the proportion of total R&D deemed or accepted as qualified. Proportionally small reclassifications of expenditures as qualified for credit will cause proportionately large changes in the credit.

Fourth, the shifting, company-specific nature of the base reduces drastically the incentive effect of the credit. Since increasing current expenditures reduces future credits, the effective rate of credit, taking into account the present value of future losses of credit, is far lower than the nominal 25 percent rate of current credit. With 1981 expenditures counting as half of the base for 1982 as well as one-third of the bases for 1983 and 1984, the effective rate of credit in 1981 is indeed generally close to zero. For later years the effective rate of credit for firms with tax liabilities and increasing R&D is in the order of only a few percent.

Fifth, for a number of firms the effective rate of credit is negative. This occurs in particular where R&D is already below base and the credit is not sufficient to induce an increase of expenditures to a level above base. Firms in such a situation maybe induced by the existence of the credit to lower current expenditures, on which no credit is earned in any event, in order to lower the base for future years and thus earn a credit or greater credit then.

A second category of firms for which the effective rate of credit may be negative includes companies where qualifying expenditures are more than 100 percent over base. Since these firms in effect raise their current base by 50 cents with each dollar of additional expenditures their current tax credit is actually at a rate of only 12.5 percent instead of 25 percent. With future bases being fully raised by current expenditures, their loss in future credits with any reasonable rate of discount is more than the current gain.

In addition, firms with no federal income taxes currently payable and also unable to take advantage of the three-year carry-back provision may lose the advantage of the credit. If a carry-forward of one or two years is necessary, the present value of tax savings is reduced. If the carry forward is three years, the effective rate of credit is brought to zero.

We have analyzed vast amounts of data from the Office of Tax Analysis of the U.S. Treasury Department, Standard and Poor's Compustat tape and tabulations, and individual company responses from both the regular McGraw-Hill spring surveys and a special supplementary R&D survey undertaken for us. Results of the analysis serve generally to flesh out our analytical predictions.

The "tentative tax credit," ignoring the issue of current tax liabilities against which it would have to be offset, appeared to run in the broad neighborhood of $500 million in 1981 and roughly double that in 1982, despite the slackening rate of growth in R&D in that year. Forecasts of the future tax credit suggest some further increase in 1983, but the forecasts for 1984 are mixed.

A considerable number of firms, accounting for substantial proportions of R&D expenditures, fall into categories for which the effective credit is negative or zero. In 1982, for example, 128 firms out of 592 in our Compustat sample, accounting for 15.3 percent of R&D expenditures in that group, appeared to have R&D equal to or below base. These firms could benefit by reducing current expenditures. An additional 154 firms, accounting for no less than 32.1 percent of total R&D expenditures, had no federal income taxes payable in 1982. What they will gain from the credit will depend upon in which direction and how far it will have to be carried.

The OTA data indicated a huge falloff for 1981, of some 43 percent, from tentative credit to credit currently claimed. The corresponding figure for 1981 estimated from the Compustat data was only about 14 percent. If the OTA data are in fact correct for 1981, the trend in the Compustat data for 1981 to 1982 suggests that perhaps well over half of 1982 R&D would have been undertaken by firms unable to claim the credit in 1982 because of lack of tax liabilities in that year.

The sensitivity of the credit to eligibility ratios is supported in several bits of evidence. First, the direct estimates of the 1981 credit from both OTA and McGrawHill data are significantly larger than is indicated in either Compustat or McGrawHill R&D tabulations. This suggests that firms did indeed increase their eligibility ratios in 1981 as against 1980. That may have been accomplished by counting as qualified R&D in 1981 expenditures that had not been so classified in 1980. It may also have been accomplished by crediting to the second half of 1981 expenditures for goods and services which in 1980 were spread relatively evenly over the year. Firms in a position to do this could enjoy a greater 1981 tax credit without increasing the base and thus reducing credits in future years. The answers to explicit questions in the McGraw-Hill supplementary R&D survey dealing with qualified and total company-financed R&D expenditures do in fact suggest a sharply higher eligibility ratio in 1981.

The R&D tax credit, like R&D expenditures themselves, is concentrated heavily among firms which are large, whether measured in terms of R&D expenditures, sales or number of employees. The concentration of the credit is somewhat less among large firms than is R&D itself, however, because R&D is apparently growing proportionately somewhat less among large firms.

Efforts to find in the data measures of the effectiveness of the tax credit are still very preliminary. Analytical considerations suggest that the effectiveness may indeed be hard to detect, partly because the credit is by its definition tied to increases in R&D, whether it causes them or not, and partly because the effective rate of credit is generally small, and in many instances zero or negative. Our data appear consistent with the view that levels of current R&D expenditures are largely accounted for by previous R&D levels and broadly determining economic conditions such as sales and profitability.

By one hypothesis the credit, if it were affecting firms' decisions, might have been expected to result in greater increases in R&D by firms with R&D above base and lesser increases, or greater decreases, for firms with R&D below base. A comparison of such increases and decreases in R&D for the years 1976 to 1980, before the adoption of the tax credit, and the years 1981 and 1982, did not indicate that the credit was having this impact.

A comparison of changes in qualified and nonqualified R&D and their relations with the tax credit indicates a substantial upsurge in R&D reported as qualified in 1981, significantly offset by a decline in nonqualified R&D. Qualified R&D apparently increased somewhat more rapidly than nonqualified R&D in 1982 but the growth in total R&D decreased sharply.

40-356 O 85 - 8