R&D PRODUCTIVITY IN 
 

THE Chemical INDUSTRY 
 
 
 
 
 

BY 
 
 
 
 
 
 

David Aboody and Baruch Lev^* 
 
 
 
 
 
 
 
 
 
 
 
 
 

      Corresponding Author: 
 

                                                      Baruch Lev

                                                      Tel:   (212) 998-0028

                                                      e-mail: blev@stern.nyu.edu

                                                      website: baruch-lev.com 
 
 

March 2001 
 
 

I. Origins of the Study 
 

      The Chemical industry is among the largest and most prominent economic sectors in the U.S. and in several other developed countries.  Chemical production amounts to about 2% of annual U.S. GDP and 11% of the total product of all manufacturing companies.  Chemical companies employ close to 1.5 million people in the U.S., and as a group are the largest exporter, generating over 10% of all U.S. exports.¹  These few statistical highlights, chosen from amongst many, demonstrate the central economic and social roles played by Chemical companies.

      The prowess of the Chemical industry has been in a large measure driven by research and development (R&D), conducted by corporations, universities and national laboratories.  The industry currently produces more than 70,000 different Chemical substances, generated by over a century of intensive R&D effort.²  In fact, the Chemical industry was the first to establish formal industrial R&D laboratories in the late 19^th century.  A staggering number of path breaking innovations emerged during the 20^th century from Chemical laboratories: plastics, PVC, polyethylene, corfam (synthetic leather), Lycra, polyester, silicone oxide, liquid crystal, and quartz crystal, among others.³  In addition to fostering Chemical innovations, Chemical R&D provided much of the scientific and industrial foundations in such diverse sectors as agriculture, transportation, housing, communications, pharmaceutics and biochemistry.    A relentless pace of innovation has been the outcome of Chemical R&D.

      So far so good, the Chemical industry is undoubtedly very large (e.g., global Chemical production exceeded in 1998 $1.5 trillion⁴), pervasive – involved in almost every aspect of life and commerce, and highly innovative, due to persistent and successful R&D activities.  However, economic history teaches us that complacency often causes the demise of success.  Innovative companies (e.g., IBM in the1960s and 1970s) tend to rest on their laurels, after a successful innovation period.  Temporary setbacks, such as currently experienced by genetically–engineered crop developers, often lead to disillusionment of investors and managers with radically new research and development.  Furthermore, since R&D outlays are fully charged (expensed) against earnings, it’s hard for managers to resist the temptation (particularly during hard times) to slow the growth of investment in innovation in order to meet short-term earnings targets.

      Indeed, evidence suggests the presence of a certain complacency, and perhaps even disillusionment with investment in innovation in the Chemical industry.  For example, over the 10-year period 1989-1998, the R&D spending of the major Chemical companies stagnated at an annual level of $3.25 billion, while the R&D spending of the major pharmaceutical companies, for example, increased at an average annual rate of 22% per year (from $3.35 billion in 1989 to $10.08 billion in 1998) ⁵  The total number of utility patents issued annually to the major Chemical companies in fact decreased from 2,942 in 1989 to 2,722 in 1998, while the patent activity of the major pharmaceutical companies has increased from 800 in 1989 to 1,115 in 1998.⁶  Similarly, while the number of R&D scientists and engineers in the Chemical industry increased by 14% during 1989-1998 (from 78,300 to 89,500), the corresponding increase in the drug industry was 32% (from 34,400 to 45,300).⁷

      The apparent slowing of investment in innovation by Chemical companies during the 1990s.—a period of unprecedented innovation and growth in the U.S.—is clearly reflected by the volume of “Intangible Capital,” or intellectual assets of these companies. As elaborated in Section II below, in terms of intangible capital, the Chemical industry ranks roughly in the middle of all major industries, lagging behind such innovative sectors as electronics, software, pharmaceutics, and even oil & gas.

      This situation raises various intriguing and important questions for Chemical manufacturers, their partners in innovation—universities and national laboratories—and given the pervasiveness of the Chemical industry, to the U.S. and global economy as well:

• What is the productivity (return on investment) of Chemical R&D? A slow growth investment in R&D, currently characterizing the Chemical industry, is an appropriate policy when the return on R&D is close to the firm’s cost of capital.  If, on the other hand, the return on R&D is substantially higher then the cost of capital, a low growth policy is detrimental to corporate and shareholder value growth, reflecting misallocation of corporate and investor resources.  Assessment of the return on Chemical R&D is, therefore, crucial for optimal resource allocation at both the corporate and national levels.
• Are all forms of R&D born equal?  The Chemical industry is very heterogenous: products can be broadly classified into commodity and specialty chemicals; and further into basic chemicals, organic chemicals, plastics and fertilizers.  The nature of R&D conducted by Chemical companies can be categorized into product development, process (R&D aimed at enhancing production efficiency), and customer support (R&D aimed at addressing specific customers problems).  It stands to reason that the productivity of Chemical R&D varies by product and type of research.  It is, therefore, important to penetrate the “R&D blackbox” and estimate the productivity of different types of R&D, in order to assist and direct the allocation of resources, as well as the research at universities and national laboratories.
• What are the drivers of successful Chemical R&D?  The previous two questions dealt with the primary outcome of the R&D process—return on investment, in R&D.  If one wants to change the outcome (e.g., enhance R&D productivity), a thorough understanding of the drivers (casual factors) of R&D and the value linkages (e.g., the effect on R&D productivity of an increase in the number and quality of scientists) is required.  Accordingly, it’s of major importance to identify the central drivers of R&D and quantify the cost-benefit linkages.  Optimal allocation of corporate and national resources hinges on a through understanding of R&D drivers and their impact on innovation and growth. 
• Lastly, What are the externalities (spillovers) of Chemical R&D?  Case studies and anecdotal  evidence indicate that Chemical R&D historically had and continues to have considerable “positive externalities,” that is contributions to the scientific and technological development of other industries, such as pharmaceutics, biotech, transportation, agriculture, semiconductors, food, and apparel.  A comprehensive assessment of the contribution of Chemical R&D (return on investment) must therefore extend beyond the measurement of R&D contribution to the productivity of Chemical companies, to encompass the contribution of Chemical R&D to other industrial sectors and society at large (the social return on Chemical R&D).

      The Council for Chemical Research embarked in 1999 on an ambitious research program aimed at addressing empirically the aforementioned questions.  Given the complexity of the issues and the size as well as heterogeneity of the Chemical industry, such an investigation is obviously a multi-phase, multi-year endeavor.  The study reported below constitutes the first phase of the investigation—an empirical assessment of the overall productivity of Chemical R&D—addressing the first of the four questions posed above.

      The following section (II) provides a bird’s-eye view of the knowledge (intangible) capital generated by the Chemical industry, relative to other major economic sectors.  Section III elaborates on the sample of Chemical companies used in this study and Section IV discusses the statistical methodologies underlying the study.  Section V presents the major findings, while Section VI provides further results, based on partitioning of the sample companies.  Section VII presents concluding remarks and charts the course of future research on Chemical R&D. 
 

II. Intangible Capital

      Corporate wealth and growth is generated by the deployment of physical (property, plant & equipment, inventory, etc.), financial (working capital, equities, bonds), and intangible capital (patents, brands, human resources).  During the last 20-30 years, much of corporate growth was generated by intangible assets, particularly in the developed economies.⁸  Intangible assets can be broadly classified into those related to discovery and innovation (e.g., new products, patents), human resources (e.g., specific compensation and work practices enhancing employee productivity), and organizational capital.  The latter intangibles are unique organizational designs, such as Cisco’s web-based product installation and maintenance system, Wal-Mart’s integrated inventory and supply operations, and Dell’s built-to-order computer distribution channels, which create considerable and sustained value. For example, Cisco’s web-based product installation system was estimated by its CFO to save $1.5 billion in three years.⁹ 

      The valuation of intangible assets is complicated, in part due to the nature of these assets (high risk, not traded in organized markets, often associated with incomplete property rights), and in part due to archaic accounting rules which deny them the status of assets presented on corporate balance sheet.  However, one of the authors of this study has recently developed a methodology to estimate the value of corporate intangible assets and the earnings derived from  these assets.¹⁰

      In essence, this methodology estimates a company’s intangible capital by a multi-stage process: (a) the company’s annual performance is estimated as a function of both historical and expected (growth potential) core earnings.  Expected earnings are derived from the consensus forecasts of the financial analysts following the company.  (b) A “normal return” on the physical and financial assets of the company (stated on its balance sheet) is subtracted from the estimated annual performance (previous stage), to yield the part of the company’s performance contributed by the third asset category--intangible capital.  This residual performance is termed “intangibles-driven earnings.”  (c) The future stream of these earnings is capitalized (i.e., the present value of the stream is computed) to yield an estimate of the company’s intangible capital. 

      The value of intangibles-driven earnings is thus derived from a “production function,” which relates a company’s performance to the three major asset groups generating this performance—physical, financial and intangible assets.  The only unknown in this equation (the residual) is the value of intangible capital.  The other values are either given (physical and financial assets) or estimated (company’s performance, and the normal returns on physical and financial assets).  The value of intangibles-driven earnings is thus derived as a residual, after “physical and financial earnings” are subtracted from the total performance of the company.

      Extensive empirical examination (Gu and Lev, 2001) has established that intangibles-driven earnings are more strongly correlated with changes in corporate market values (stock returns) than widely used performance measures, such as corporate earnings and cash flows.  Strength of correlation with value changes is a commonly used indicator of the informativeness of a performance measure or other signals (e.g., a corporate acquisition announcement).  Furthermore, the estimated value of intangible capital—the major missing asset from corporate balance sheets—when combined with book value (the balance sheet value of net assets), provides an effective yardstick for the estimation of corporate value and predicting future stock performance (Gu and Lev, 2001).

      Figure 1 presents median values of intangible capital (for the year 1998) for 19 industries, derived from the 1998 CFO magazine’s ranking.  Each industry is represented by the five largest public companies operating in the industry.  There are three distinct groups of industries in Figure 1: Those with intangible capital per company below $10 billion (e.g., airlines, specialty retail, forrest/paper, motor vehicles), those with intangible capital between $10 and $20 billion (semiconductors, scientific instruments, oil & gas, aerospace), and the third group—industries with intangibles values per firm exceeding $20 billion (software, entertainment, computers, telecom, and the highest—pharmaceutics). 
 

Insert Figure 1 here 
 
 

      The Chemical industry is, as evident from Figure1, situated in the middle group—median intangible capital per firm of roughly $16 billion, with large variability within the industry.¹¹  A sample of some leading Chemical companies’ intangible capital (in 1998) is: Dupont--$41B, Monsanto--$22B, Dow--$16B, PPG Industries--$9B, and Union Carbide--$4B.

      A different perspective of intangibles’ value and contribution is provided in Figure 2 (derived from Gu and Lev, 2001), which portrays the growth rate of intangibles-driven earnings, by industry, over the 1990s.  This figure is based on a much larger sample then Figure 1--roughly 2,000 public companies (Figure 1 is based on 100 companies).  We can once more classify the industries in Figure 2 into three classes: Low growth rate of intangibles earnings (0-10 percent annual growth), medium growth rate (11-15 percent annual growth), and high growth rate (16 and higher percent annual growth).  As indicated in Figure 2, the Chemical industry is at the high end of the low growth group, with 8.2 percent annual growth rate.  Also in this group are oil and gas companies (9.9 percent), insurance (8.3 percent), and primary metals (3.7 percent).  In the intermediate group we find drugs (13.7 percent), medical instruments (13.1 percent), and telephone communication (12.2 percent).  The high intangibles earnings growth group includes special machinery (24.3 percent), computers (19.4 percent), and software (17.6 percent).

      Summarizing, the message emerging from the intangible measures presented in Figures 1 and 2 is that the intangible capital of Chemical companies ranks at about the median (mid-point) of nonfinancial industries (Figure 1).  However, in terms of Growth in the contribution of intangible assets to overall corporate performance over the 1990s, Chemical companies reside among the low rate of growth group (Figure 2).  The latter finding is consistent with (perhaps, the outcome of) the slow growth during the 1990s in the investment in innovation by the Chemical industry, noted in the preceding section.

Intangibles earnings and capital are driven, in part, by investment in R&D.¹²  We accordingly proceed in the following sections to analyze the return on Chemical R&D. 
 

Insert Figure 2 here 
 
 

III. Sample Characteristics

      The sample of companies whose data were used in this study to estimate the productivity of Chemical R&D was carefully chosen to represent the broadest cross-section of Chemical companies.  To secure data availability, we restricted the sample to publicly traded companies, since these enterprises publish annually audited financial statements.  We further restricted our sample to companies whose main activity involves commodity and/or specialty chemicals.  Thus, for example, oil and gas companies with Chemical divisions are not included in our sample.¹³  Finally, the sample had to be restricted to companies whose financial data are included in COMPUSTAT, the major electronic database we used.  These sample selection criteria yielded 83 Chemical companies listed in the Appendix.

      The data used for estimation of R&D productivity cover the 20-year period 1980-1999.¹⁴  Some sample companies have shorter time series than the 20 years examined.  This causes the number of companies in each year analyzed to be smaller, sometimes substantially so, than 83.  Table 1 provides summary statistics characterizing our sample.  The next to left column in the top panel of Table 1 indicates that the average R&D intensity (the ratio of annual R&D expenditures to sales) of the sample companies increased from 2.47% in 1980 to 4.70% in 1999, a robust increase.¹⁵  However, compared with other economic sectors, the overall R&D investment of Chemical companies is less impressive.  While the average R&D intensity of Chemical companies in 1999 was 4.70% (Table 1), the average R&D intensity of other sectors were: pharmaceutics—12.14%, software—11.06%, computers—9.16%, and oil and gas—3.02%.  The average R&D intensity of all U.S. public companies having R&D operations was 4.84% in 1999.¹⁶

      The bottom panel of Table 1 breaks the sample to large and small firms—above or below the sample median of market capitalization.  It is evident that the R&D intensity of large companies (4.86% in 1999) is higher than that of small companies (2.75% in 1999), as was the rate of growth of R&D intensity over the sample period (1980-1999).

      While the ratio of R&D to sales in the Chemical industry is modest relative to some other R&D intensive sectors, the ratio of R&D to operating earnings (third column from left) is quite high: 56.7% in 1999 for the whole sample, and 46.7% for small companies.  This high ratio of R&D to operating earnings (i.e., earnings before finance and tax expenses), which of course is even higher relative to net earnings, indicates the existence of a serious constraint on substantial increases in R&D budgets.  Stated differently, earnings (profits) of Chemical companies are quite sensitive to R&D expenditures, an issue which must weigh heavily on Chemical executives. 
 

Insert Table 1 here

      The right column in the top panel of Table 1 (RDCAP/BV) provides an indication of the value of R&D relative to other corporate investments.  RDCAP (from R&D capitalized), represents the value of corporate R&D investment, if the annual R&D expenditures were capitalized (i.e., treated as an asset, rather than a regular expense), and then amortized annually according to the economic amortization rates generated by our estimation procedure (Table 2).  The data in Table 1 indicate that if R&D were treated as an asset, it would have constituted, on average, in 1999, 33.9% of book value (the net value of physical and financial assets, as stated on corporate balance sheets).¹⁷ 
 
 

      Summarizing, the average R&D intensity of Chemical companies has increased over the last quarter century, yet in the late 1990s it stands slightly below the U.S. average, and below  that of several R&D intensive sectors.  While low in intensity, R&D expenditures constitute about 57% of the operating earnings of Chemical companies, and an even higher percentage of net earnings—a serious constraint concerning significant increases in R&D budgets, unless such R&D promises a reasonably quick return on investment. 
 
 
 
 

IV. Statistical Methodology¹⁸

      1. The Model

      Our estimation of R&D productivity is derived from a “production function,” reflecting the fundamental relation between the value of corporate assets and the earnings, or operating income generated by them.  Accordingly, we define the operating income (OI_it) of firm i in year t, as a function of tangible, TA_it, and intangible assets, IA_it, where the latter includes the R&D capital:

                        OI_it = g(TA_it,IA_it).      (1)

While the values of operating income and tangible assets (at historical costs) are reported in financial statements, the intangible capital, IA, is not reported and therefore has to be estimated.

      Given our focus on R&D, we single it out of all intangible assets and define its value, RDC_it, as the sum of the unamortized past R&D expenditures.  Those are the expenditures that are expected to generate current and future income:

                        RDC_{it = ik}RD_{i,t – k,}      (2)

where a_ik is the contribution of a dollar R&D expenditure in year t – k (k = 0, …,N) to subsequent earnings (i.e., the proportion of the R&D expenditure in year t – k that is still productive in year t).  Substituting expression (2) into (1) yields:

                        OI_it = g(TA_it,_ikRD_{i,t – k},OIA_it),    (3)

Where OIA_it are other (than R&D) intangible assets (e.g., brand values).

      The variables in relation (3) are defined thus.  Operating income, OI_it, is measured as reported operating income (sales minus cost of sales) before depreciation and the expensing of R&D and advertising.  Operating income is used as a measure of R&D benefits, since R&D investment and its consequences seem largely unrelated to nonoperating items, such as administrative expenses and financing charges. Depreciation, R&D, and advertising expenses were excluded from (added back to) operating income since they represent, largely ad hoc write-offs of the independent variables in (3) – tangible and intangible assets.

      Tangible assets, TA_it in (3), consist of all assets reported on the balance sheet, including, among others,  plant and equipment and inventories.  The major intangible asset, R&D capital, is represented here by the “lag structure” of annual R&D expenditures, expression (2), where R&D expenditures stretch over the preceding nine years.  Advertising expenditures on product promotion and brand development may create an additional intangible asset for some sample firms.  This may raise an omitted variable problem in expression (3), if R&D capital were the only intangible asset included.  Conceptually, advertising capital can be estimated from its lag structure (current and previous expenditures), similarly to the procedure applied to R&D (expression 2).  However, inspection of our data source, revealed that annual advertising expenditures were occasionally missing for may sample firms, straining the requirement for reasonable length of lag structure for reliable estimation.  We therefore employed a procedure frequently used by economists (e.g., Hall, 1993), in which the advertising intensity (advertising expenses over sales) is substituted for advertising capital.  Empirical evidence (e.g., Bublitz and Ettredge, 1989; Hall, 1993), indicated that, in contrast to R&D, the effect of advertising expenditures on subsequent earnings is short-lived, typically one to two years only.  Accordingly, an advertising proxy based on annual expenditures may account reasonably well for the brand value in expression (3).

      The following model (4) is used to estimate the returns on R&D, by means of least squares regression.  The variables are scaled (divided) by sales to mitigate the econometric problem of heteroscedasticity, due to different sizes of sample companies.  We also use the Almon lag procedure (for details see Johnston 1984), to alleviate the multicolinerity problem due to the relative stability of firms’ R&D expenditures over time.  The estimated model is: 
 

                  (OI/S)_it = a_o + a₁(TA/S)_i,t-1 + _2,k(RD/S)_i,t-k+ a₃ (AD/S)_i,t-1 + e_it,        (4)

where:

OI  =  annual operating income, before depreciation, advertising and R&D expenses, of firm i in year t,

S    = annual sales in t,

TA = the balance sheet value of total assets at year t,

RD  = annual R&D expenditures in t,

AD  = annual advertising expense in t. 
 

2. Intuitive Interpretation

      Following is an intuitive, nontechnical interpretation of the estimation model (4) and its parameters.  We assume that the productivity of a company’s R&D expenditures is manifested by the contribution of these expenditures to current and future (up to eight years) operating income.  This underlies model (4), where operating income in a given year is related to (a function of ) the firm’s R&D expenditures in that year, as well as R&D expenditures in each of the preceding eight years.  The nine R&D coefficients to be estimated by our econometric technique, a_2,k, reflect the contribution to current operating income of each vintage of R&D expenditures.  Thus, for example, an a_2,0 of 0.372 (Table 2) indicates that a dollar R&D spent in the current year (year 0) increases current operating income by $0.372.  Similarly, an estimated value of a_2,4 of 0.206 (Table 2) indicates that a dollar R&D spent four years ago increases current operating income by $0.206. 

      Once we have estimated the contribution to income of each vintage of R&D (we examine nine annual vintages), we can estimate the total contribution of a dollar R&D to current and future income by adding up the annual contributions.  From the yearly contributions we derive, as will be seen in the next section, the rate of return on R&D investment.

      Back to model (4), above.  R&D is, of course, not the sole contributor to Chemical companies’ operating income.  Physical assets and advertising (promotion, brands) contribute as well.  Accordingly, we include in the estimation model (4) the values of assets (TA/S) and advertising (AD/S), to enable us to focus on the incremental contribution of R&D to firm’s profitability.  Stated differently, in estimating the contribution of R&D to profitability, we control for the contribution of other productive factors to profitability.

      The parameters (contribution to profitability) in model (4)—the various α coefficients in the equation—are estimated by the widely used regression technique applied to our sample.  Recall that we have 83 companies in the sample, and a maximum of 20 years of data for each company.¹⁹  Thus, for example, one data point in the sample will be DuPont’s operating income (OI) in 1995.  This value is accompanied by DuPont’s total assets (TA) at the beginning of 1995, its advertising expenditures (AD) in 1995, as well as the series of nine annual R&D expenditures of DuPont, starting with 1995 and going back to 1987.  Each company in the sample has 20 similar annual data sets (some companies have less than 20 data sets).

      For those not versed in econometrics, we would like to emphasize that we don’t introduce any judgmental factors into the estimation, beyond the underlying assumptions (e.g., that operating income is generated by tangible and intangible assets).  In other words, we don’t estimate subjectively the contribution of R&D to income.  Rather, we let the data (our sample) “speak for itself.”  The estimated coefficients (contributions to income) to be reported in the next section, are the results of statistically estimating the fundamental model (4) from our sample data. 

      Finally, the important issue of causality.  So far, we have interpreted model (4) in a strictly causal manner—from R&D to income.  R&D expenditures (and other assets) were assumed to contribute to current and future income.  The fact that assets contribute to profits is undisputed, but a simultaneous reverse causation cannot be ruled out.  A decrease in current or expected profitability (due, say, to sharp increases in energy prices, or the onset of an economic recession) will undoubtedly have a dampening effect on firms’ willingness to invest in R&D.  To allow for such simultaneity (from R&D to income and from expected income to R&D), it is possible to employ a statistical technique known as simultaneous equations.  However, the experience of one of the authors with this technique applied to a similar assessment of R&D contribution to earnings (Lev and Sougiannis, 1996) indicated that it did not yield significantly different result than those estimated by model (4) using ordinary least squares regression.  Accordingly, at this phase I of the project we did not use simultaneous equations to estimate return on R&D. 
 
 

V. The Estimated Return on R&D

      Table 2 presents the main findings of this study.  It reports the results of estimating the coefficients (a) of model (4) which relates operating income to tangible assets, advertising expenses, and the series of current and past R&D expenditures.  Model (4) was estimated separately for every year, 1980-1999, and the coefficients reported in Table 2 are averages of the yearly estimated coefficients.  Following is a detailed interpretation of the estimates reported in Table 2.  We will focus first on the middle column of the table—Chemical companies.  The right column—estimates for software companies—is presented for comparison purposes.  
 

1. The return on physical and advertising capital

      The estimated coefficient, 0.070, presented at the top of Table 2, indicates that a dollar of total assets of Chemical companies contributes, on average, $0.070 annually to operating income.²⁰  This 7% estimated annual return on assets is close to the weighted average (equity and debt) cost of capital of Chemical companies.  A recent estimate of the cost of equity capital of Chemical companies (Fama and French, 1998) indicated a rate of 10.28%.  since the cost of debt is lower than the cost of equity, the weighted average cost of capital of Chemical companies is reasonably close to 7-8 percent.  Thus, balance sheet assets in the Chemical industry earn, according to our estimates, approximately the cost of capital. 
 

Insert Table 2 here