Decreasing returns to shareholders from R&D spending in the computer industry

Introduction

Corporate R&D expenditures represent major, controllable, elective spending with the management goal of creating future revenue and profits for the firm. At the national level, total annual R&D expenditures were $221 billion in 1998, of which 65% ($143 billion) was provided by industry. This investment represented 2.61% ofthe gross domestic product (GDP) (Payson, 1999). Confidence in these investments is reflected in the average increase of more than 5% per year during the past 23 years for industry financed, real R&D expenditures (Payson, 1999). Even though there is widespread agreement that technological innovation is key to economic growth, the value generated by R&D investments has been seriously challenged. The yields on these investments are perceived by many to be unacceptably low and very slow to appear (Rouse and Boff, 1998).

The production of patents is one of the proxies used to assess the value generated by the R&D investments and the creation of inventive output. Metrics for this production have included patent counts and patent intensity (the number of patents generated divided by the firm's assets). Over time, there has been a documented drop in the ratio of patents to R&D spending (Kortum, 1993), an indication that the productivity of R&D might be decreasing.

An issue that has not been widely addressed by researchers is the impact to stockholder wealth from R&D over periods of a year or more. This is an important issue, since in modern financial theory, the primary goal of management is the maximization of stockholder wealth. Addressing this need, this study analyzes the short- and long-term effects of R&D expenditures on shareholder returns in a high-technology industry. The research correlates computer firm R&D intensity with future short-term (1-year) and long-term (5-year) returns to shareholders. The computer industry was chosen for the study because of the fast-paced rate of technology change and the competitive nature that characterizes the industry. Generally, the effects of R&D spending on growth are difficult to measure because the lag times between current spending and future results are variable and long, while the effects are also subject to many potential influences (Griliches, 1998). The rapid rate of technology change in the computer industry is expected to reduce the lag time between R&D and its growth effects, and limit the effects of outside influences, The intense competitive environment in the computer industry is also expected to motivate rapid management action to generate stockholder returns and make it more likely that the R&D impacts be measurable. The computer industry also was chosen because the needed data was available.

The following sections present an analysis of R&D spending decisions, life cycle implications, and an overview of previous studies analyzing the impact of R&D expenditures on corporate financial results, as well as a description of a conflict that arises between life cycle expectations and documented management practices. Also described is the data collection method, the choice of variables, and the analysis method used in this study, followed by the research results and conclusions.

R&D Spending Overview

At the industry level, industries with higher corporate funded R&D intensity have been shown to grow faster than industries with lower rates of corporate funded R&D (Branch, 1973). Corporate funded R&D has also been found to vary widely by industry (Scherer and Ross, 1990). However, the effects of R&D spending intensity on individual firms within an industry are less clear.

There is evidence for the practice of holding the level of R&D intensity fixed at the firm level (Scherer and Ross, 1990) and specifically for the use of the practice in the computer industry (Link, Seaks and Woodberry, 1988). This has been termed the "R&D as a fixed percentage of revenue" rule (Link, Seaks and Woodberry, 1988). Using this rule, firms establish an R&D budget by estimating revenue and then multiplying it by a fixed percentage factor, i.e., the R&D intensity factor. The application of the rule makes the R&D budget determination formulaic and stabilizes the intensity factor over time, which provides a level of continuity to research programs and avoids the excessive hiring and firing of valuable research professionals over the short term. Over the longer term, the firm R&D intensity may be adjusted in response to external events and economic conditions (Grabowski, 1970). Evidence has found clearly discernible R&D spending patterns among competing firms (Frumau, 1992). In other words, competing firms within an industry are known to have different levels of R&D intensity.

Studies have positively and significantly related announced increases in R&D expenditures in high-technology companies to share price increases (Chan, Martin and Kensinger, 1990). Interestingly, this was not true for lowtechnology firms. Research shows that when low-technology companies announce R&D expenditure increases, their stock prices drop (Chan, Martin and Kensinger, 1990). R&D spending can also be viewed as an investment problem. In this perspective, the value of R&D investments is equivalent to. the present value of the expected future profits generated by the research (Scherer and Ross, 1990). The share price movement of high-technology companies following announcements of R&D increases implies that the investors believe that the net present value of future earnings and cash flows will be higher as a result of the increased expenditures. The contrasting result for low-technology firm R&D investment increases would then reflect the expectation of diminished future returns from the planned research. This differentiates the expectation by the market on the returns of R&D investments based on the type of industry. Note that in these studies, the stock price effects for the announced plans to change R&D expenditure levels were measured over very short time periods prior to and after the announcements. The longer-term returns to shareholders were not reported.

Other research provides contradictory findings of the effects of R&D spending on the net income of a firm (Leonard, 1971). Some studies have suggested that a firm's R&D activity causes revenue to grow at a steady profit level (Scherer, 1965). However, none of these studies tied R&D expenditures to their effects on market valuation.

Life Cycle Implications

The life cycle is a well-accepted model that describes the dynamics of products and firms as they mature through the stages of introduction, to growth, to maturity, to decline. R&D has a major role in the introduction and growth stages, as it develops the breakthrough product technologies and applies them to new market opportunities. Manufacturing breakthroughs are particularly valuable during the growth stage to improve productivity and yields. However, during the mature stage of a product or business life cycle, the profit margins are expected to decrease. This steady downward pressure on profitability might exert pressure to reduce the firm's R&D intensity as management struggles to deal with reduced margins.Simultaneously, the firm is faced with a reduced opportunity for breakthrough research. The best projects with the best risk/return profiles have already been selected and spending is directed toward projects with higher risk and lower return profiles. The resulting higher than optimum R&D spending levels generate "unnecessarily high costs and low incremental returns" (Moore, 1995), and it is during this phase that the "power of the R&D function comes under challenge." The actual profit dollars and R&D expenditures may continue to rise during this phase, but the intensity of both, expressed as a percentage of revenue, is expected to fall.

The practice of holding the R&D intensity fixed at the firm level can create a bias toward slow adaptation of R&D policies consistent with changing conditions as the firm matures. It suggests that firms that continue to spend funds at previously established levels of intensity, when they should be lowering their R&D spending, may be diminishing returns to shareholders.

The object of this research is to test whether the computer industry is reducing its research intensity and if firms that have higher R&D intensities are providing a greater or smaller return to their stockholders. Based on this analysis, a number of hypotheses can be made and tested. These hypotheses include:

1. Most of the firms in the computer industry are entering the mature stage. Supporting evidence would be decreasing R&D intensity rates by the firms in the industry.

2. A bias exists toward policies that retain excessively high R&D intensity. Firms that have higher R&D intensities should provide lower shareholder returns.

3. It takes considerable time for R&D policies to affect shareholder returns. Five-year returns should have a stronger signal and a more statistically significant negative correlation between the two variables than the 1-year return.

4. Many other key factors affect shareholder returns. Extremely high statistical correlation should not be expected.

5. Not all firms in the industry are equally mature. Firms with extremely high growth rates should reflect firms in the growth stage. If these firms are eliminated from the group of firms analyzed, correlation should improve.

Data Sources

In order to select an impartial list of firms for the analysis, we used the Wall Street Journal ' "Shareholder Scoreboard." This list provided the names of computer industry firms and their shareholders' compound annual returns data. The Wall Street Journal listing "incorporated all U.S. stocks included in the Dow Jones Global Indexes" or 718 firms representing 94 industries-with the additional "largest 282 U.S. corporations, based on market capitalization, that weren't included in the Dow Jones Global Indexes" (Wall Street Journal, R:6, 1998). The resulting list of firms represented over "85% of the market value of all publicly traded U.S. stocks as measured by the Wilshire 5000 index" (Wall Street Journal, R:6, 1998).

Other financial information was sourced from the Standard and Poor's Compustat Database and from the United States Security and Exchange Commission (SEC) Electronic Data Gathering, Analysis and Retrieval (EDGAR) System database.

The computer companies in the "Shareholder Scoreboard" were segregated into three groups. The first were the "superstars." Superstars were defined as firms that had over two times the industry's average revenue growth during the years studied. The second group included firms eliminated from specific parts of the study. To be included in the study, sufficient data was required for the entire time period being studied. In some cases, companies could be included in the 1 year analysis but not in the 5-year analysis. Firms that had significant revenue restatements due to acquisitions or divestitures during the period of the research were also eliminated. A restatement of greater than +/- 30% of revenue was considered significant. The third group was called the "mere mortals." These were the firms that were not the superstars in their industry in terms of growth. The relationship between R&D spending intensity was studied for the mere mortals alone, and for the mere mortals together with the superstars. The firms that comprised the basis for this study are shown in Exhibit 1. In addition to those companies listed in the exhibit, Seagate Technology, Creative Technology, and Gateway 2000 Inc. were in the Wall Street Journal Shareholder Scoreboard but were eliminated from the analyses due to either lack of data over the time period or to significant restatements. NCR could be included in the 1-year regression but not the 5-year regression.

Variables Considered

The return to shareholders for the 5-year analysis was calculated as the compound annual return to shareholders, including both share price changes and reinvested dividends for the years 19931997 (Wall Street Journal, 1998). Returns to shareholders was chosen as the dependent variable because it includes both the scale and timing that are necessary to adequately judge any investment. The return for the 1-year analysis was similarly calculated for the year 1997.

R&D intensity was calculated as the 3-year average of firm R&D expenditures, divided by the 3-year average revenues over those same years. For the 5-year return analysis, the years 1992, 1993, and 1994 were used. For the 1-year return analysis the years 1994, 1995, and 1996 were used. The average R&D intensity was used to establish a routine level of expenditure to smooth year-to-year variations. R&D intensity was chosen as an independent variable based on the results of previous work that found that research intensity, measured by R&D spending/ net sales, was the most consistently correlated measure of research intensity to growth-however measured (Leonard, 1971).

Microsoft Excel and the Computerized Business Statistics software package were used for the analyses.

Analysis

The R&D intensity trend for the industry was first analyzed using a Student's T test to determine if the R&D intensity of the firms within the industry had changed from 1992 to 1997. Firm R&D intensities for 1992 and 1997 were used for the analysis.

A second regression was done in each case that eliminated the superstar firm or firms from the industry population. In the 5-year returns analysis, this resulted in the elimination of one firm (Iomega). The superstar criteria eliminated 3 firms from the second regression for 1-year returns (Iomega, Dell Computer, and EMC Corp).

Results

Ten (10) out of the 14 firms for which data was available over the time period recorded a drop in R&D intensity. One firm had remained constant and only 3 showed some increase-1 of these 3 doubled. This suggests that the implied life cycle change measured by decreasing R&D spending intensity is occurring in this industry. The average R&D intensity of these 14 firms decreased from 8.73% in 1992 to 7.39% in 1997. The Student's T test confirmed this observation with an 80% confidence that industry R&D intensity had decreased. The actual data is included in Exhibit 1.

The first regression for returns over the 5-year period, shown in Exhibit 2, included all firms in the sample that were not eliminated for lack of data or significant restatements. This regression, including the superstar firm, generated a negative correlation coefficient, but it was not statistically significant. A second regression, after the elimination of the superstar firm, found a statistically significant negative correlation for the mere mortal firms' R&D intensity and stockholder returns over the 5-year period. The results were significant to the 1 level as shown in Exhibit 2.

Mere mortal computer companies with higher R&D spending intensity in 1992-1994 tended to have lower shareholder returns over the 5-year period, 1993-1997. This relationship is shown in the scatter diagram and regression line in Exhibit 3.

The regression results for all firms and for the mere mortal firms were then performed for the 1-year period in 1997. The analysis of all firms over the 1-year period, shown in Exhibits 2 and 4, indicated a statistically significant result at the .05 level. It found a negative correlation between R&D spending intensity and shareholder returns. The similar regression for the mere mortal firms' 1-year returns also showed a significant, negative correlation, as shown in Exhibits 2 and 5. The results in this case were significant at the .2 level.

All of the analyses show a negative correlation between firm R&D intensity and shareholder returns. There was a stronger correlation for the 1-year period than for the 5-year period, which was unexpected. In addition, the 1-year returns analysis provided significant correlation for both populations, for "all firms" and for the mere mortals. It was also noteworthy that the 1-year regression for "all firms" had the strongest correlation. However, on the 5-year returns, only the analysis with the mere mortal firms produced statistical significance.

The correlation coefficients for all regressions are also shown in Exhibit 2. They ranged from -.43 to -.62 on the statistically significant regressions and are remarkably consistent. The B coefficients ranged from -4.1 to -13.1 for the significant regressions.

This analysis is part of an ongoing research project into the returns to shareholders of corporate R&D spending intensity decisions in high-technology industries. Preliminary analysis of the software and semiconductor industries supports the finding of negative correlation between firm R&D intensity and shareholder returns.

Conclusion

Consistent with the expectations, the computer industry is undergoing a reduction in their R&D intensity. This might be an indication that most of the firms in the industry are becoming more mature and focusing more of their investments in marketing and other areas instead of R&D. However, it could also be a result of changes in the structure of many of the firms in the industry. For example, if these firms acquire more of their innovation through mergers instead of internal R&D, there would be a reduction in R&D intensity, but not necessarily in their acquisition of innovative capabilities.

The consistently negative correlation between firm R&D intensity and compound annual returns to shareholders suggest that firms in the computer industry are overspending on R&D. This appears to be true, even though the R&D intensity of the industry was decreasing during the time period studied. Perhaps the firms are recognizing the changes in their role for R&D, but not fast enough.

The stronger results over the 1-year period were surprising. However, that may be due to the fast industry cycles in the computer business. Taking into account all the other factors that impact shareholders returns, it is surprising to find such strong results, particularly with such a limited period of analysis.

The exclusion of the superstar firms did improve the correlation in the 5-year analysis. However, it did not improve the 1-year analysis. This might reflect that in the computer industry in which there is so much growth, even the firms that have greater growth are still subject to the forces of a maturing industry. It might also be an indication of other factors that affect shareholder returns.

The 5-year superstar firm was Iomega. Iomega's R&D intensity during the 1992-1994 period was the highest in the industry and was focused on a single key idea, the zip drive. While far from conclusive, the 5-year mere mortals regression, combined with the Iomega results, suggests that the competitive value of the idea being pursued is more important than the general R&D spending intensity. Perhaps firms should be willing to spend more on key technological advances while being more critical of"normal" research investments. This conclusion is indirectly supported by other studies that have found the value of patents to be highly skewed, with some patents having far greater value than others (Pakes, 1986).

The widely used, formulaic approach to determining the level of R&D spending at the firm level may also be contributing to these results. This "fixed percent of revenue" rule to determine the firm's R&D intensity may lead to systematic overspending of R&D dollars in life cycle phases where R&D intensity should decrease.

There are a number of possible explanations for these results. Overspending on evolutionary product improvements by internal R&D teams is one possibility. This overspending on evolutionary R&D might be accompanied by underspending on revolutionary R&D, and might explain the drop in the ratio of patents to R&D spending. The role of internal-to-the-firm R&D vis-A-vis joint ventures or purchased R&D is not well understood in this context.

Another cause for the negative correlation might be weaknesses in R&D management. One possibility is the difficulty of determining when to stop ineffective projects. Firms may find it easier to start new R&D projects than to shut down or redirect those that are not living up to expectations. The difficulty may be in the decision timing or in the criteria to identify unproductive projects. Given the risks inherent in any R&D project, any failure to identify or properly manage the re-targeting of projects that are not living up to initial expectations could add materially to R&D spending-with little or no returns.

As has been noted, research clearly shows that industries with higher R&D intensity grow faster than those industries with low R&D intensity. However, this study's results imply that the marginal R&D efforts do not supply sufficient shareholder returns and it sets up a real dilemma. The R&D investments are necessary for the industry to grow-but the average individual firm does not benefit sufficiently from their investment. To deal with this situation, different mechanisms might be required to sustain the required level of research that do not depend on internal R&D efforts. One option is to expand cooperative research efforts, such as SEMATECH in the semiconductor industry. In this semiconductor case, the participating firms spread some of the R&D risk while benefiting from the breakthrough technologies developed. Another mechanism might be the acquisition of technology through business mergers and acquisitions. These mechanisms might sustain the required level of R&D effort for the industry while reducing the reliance on internal R&D for the individual firms.

This research suggests that practitioners in industries with declining R&D intensity trends should look carefully at their R&D expenditures to determine the appropriate level of R&D spending. The R&D budgeting process should pay particular attention on a project-by-project basis to the returns that are possible. Given the inherent difficulty of picking the winners from the losers in proposed R&D projects, practitioners may want to pay particular attention to decisions involving continuing investments in ongoing projects, making sure that the original assumptions vis-A-vis returns are still valid before additional funds are committed.

A further implications of this research is that managers in high-technology firms should carefully reexamine their R&D expenditures, particularly in those product areas that are in or near the mature phases of their life cycle. The practice of holding the level of R&D spending intensity fixed as a percentage of revenue may be leading to investing in incremental innovation projects with marginal or negative rates of return. This research also suggests that firms should broaden their view of research and explore alternatives to in-house research projects to address the dilemma of R&D requirements for industry growth versus the negative shareholder returns for their firms. These alternatives may include industry cooperative research efforts, the purchase of technology through acquisitions, and the purchase of R&D output from third-party research firms. These alternative R&D strategies may help firms realize the rewards for innovation investments while avoiding decreasing returns to their shareholders.

Acknowledgments

The authors are grateful to PB. Thompson, D.D. Myers, and V. Allada, and especially H.E. Metzner, for their help, debate, and comments. All errors remain those of the authors.