«Department of Accounting and Finance Working Paper Series AF2014/15WP05 ...»
capitalization of some R&D by international firms will depress their profitability when investment growth declines. Thus, a negative coefficient on the interaction term will provide support for hypothesis 2.
of R&D in country i and year t, and 0 otherwise. Other variables are as previously defined.
3.2. Adjusting ROE for effects of R&D accounting To more directly estimate the effect of R&D accounting on reported ROE, we adjust earnings and equity of pharmaceutical firms for distortions caused by the immediate expensing of R&D. This procedure requires tracking back R&D outlays, capitalizing them in periods where R&D outlays were incurred, and expensing them over subsequent years via amortization.
Amortization assures that R&D outlays are matched to future revenues they help generate.
Capitalizing and amortizing R&D outlays will undo distortions to earnings and equity caused by immediate R&D expensing and will result in profitability that is more descriptive of the underlying business reality (i.e. economic income) (Dechow, 1994; Dechow et al., 1998).
As a starting point of our analysis, we note that the economic rate of return (ERR) is given as economic profit (EP) divided by the employed capital (C). The economic profit can be stated as revenues (Revenues) less variable costs (Variable costs), less economic depreciation of tangible assets (λTA TA), less the portion of R&D outlays applicable to the period (λR&D R&D).
The employed capital is invested into net tangible assets (TA) and capitalized R&D investments (R&D) & & (3) & The formula follows the simple intuition: Companies advance their expenses such as R&D outlays to generate revenues at the later point in time. Therefore, to determine how effective the business operates, it makes sense to relate revenues to expenses that were advanced to help generate it. Because R&D outlays have a long-term character the nominator in (3) relates revenues only to a portion of R&D outlays that contributed to generating this revenue. This is done by employing an economic amortization rate (λR&D). The resulting profit is then related to invested capital – the sum of R&D investments and other assets.
The accounting counterpart of the ERR is ROE measured as net income (Income) over equity (Equity). When accounting rules prescribe R&D outlays to be expensed as incurred (U.S.
GAAP), the amortization parameter λR&D is set to zero, and current R&D outlays (R&D outlays) are substituted for λR&D R&D. Because R&D is expensed rather than capitalized, the R&D
investment equals zero. As a result, ROE measures the economic rate of return with error:
& (4) Whether accounting rules understate or overstate ROE relative to ERR in case of R&D intensive firms depends on the level of current R&D investments relative to the past R&D investments (that is, R&D outlays vs. λR&D R&D), and on whether the nominator effect (subtracting R&D outlays instead of λR&D R&D) outweighs the denominator effect (excluding R&D from denominator).
Thus, to account for the effects of R&D accounting we can use reported ROE and recast it using equation (3) and (4) into our estimate of ERR, that is an economic rate of return adjusted for the effects of R&D accounting. We use three accounting techniques to match R&D outlays to revenues they help generate.
First, we assume that the benefits from R&D investments accrue as a steady stream of revenues (or cost cuts) over a 10 year period after the year of initial investment. This leads to a use of straightline amortization for capitalized R&D outlays.
Second, we use a declining balance amortization, where the benefits (revenue increases or cost cuts) decline over the 10 year period. This approach follows the long tradition of using declining balance depreciation in this line of literature (Grabowski and Mueller, 1978; Mansfield, 1968; Nerlove and Arrow, 1962; Schmalensee, 1972). Furthermore, this approach is somewhat more descriptive of the business practice in pharmaceutical industry as substitute products and expiring patents lead to declining revenues and profit margins over time. We use a constant amortization rate, because prior studies show that obtained results are not sensitive to this choice (Nadiri and Prucha, 1996). As a result, the R&D stock reported on the balance can be computed
& & 1 & (5) &
The recursive substitution leads to the following equation:
∑∞ & 1 & (6) & We set the amortization rate λR&D at 10%, which is consistent with numerous previous studies (Baily, 1972; Grabowski and Mueller, 1978), and very close to the R&D amortization rate of 9.2% used by Haneda and Odagiri (1998). Finally, we set k to equal 10 years, due to data availability.
Third, we use empirically-derived amortization rates using approach proposed by Lev and Sougiannis (1996). Current R&D outlays increase future revenues or decrease future costs. The empirical approach aims at estimating the contribution of each dollar spent on R&D to future profitability. These estimates help us in the next step to reallocate R&D expenses to the periods in which they help generate revenues or lead to cost savings. In other words, we obtain empirical amortization rates. This approach does not rely on any assumptions regarding the amortization rate and amortization method (e.g., straightline, declining balance), instead the amortization rate is empirically derived using the cross-section of pharmaceutical firms. That is, the empirical approach allows for amortization rates that are adopted to firm-specific circumstances. Similar to Lev and Sougiannis (1996), we empirically assess the relation between R&D outlays and future operating income to determine how current R&D outlays increase future sales or decrease future costs. We regress operating income on lagged R&D outlays using the following regression
∑ & (7) The dependent variable Operating incomeit is operating income of firm i during year t.
Operating income captures expected benefits from R&D – primarily revenue increases, but also cost savings. The independent variables are represented by a vector of past R&D outlays. In this set-up, the estimated regression coefficients on past R&D outlays (αk) reveal how $1 of past R&D outlays contributes to current operating income. We estimate regression (7) using all available data points and 9 years of past R&D outlays. We use the obtained coefficient estimates to derive amortization rates in equation (8).
3.3. Profitability and regulatory activity Hypothesis 3 and 4 predict that accounting profitability of pharmaceutical firms attracts regulators attention and increases regulatory activity. To test this prediction, we employ two measures of regulatory activity as our dependent variable and use lagged ROE reported by pharmaceutical firms as independent variable. We also decompose reported ROE into adjusted ROE using empirical amortization rates as shown above and accounting bias. We are particularly interested in finding out whether regulators attach the same weights in their decision making to the biased part of earnings as they do for the unbiased earnings. In other words, we examine
whether regulators fixate on aggregate earnings:
We limit this analysis to U.S. firms because they report abnormally high profitability and because regulatory data is only available for our U.S. sample. We include years 1997-2012 because 1997 is the first year for which regulatory data is available. We use two proxies for changes in regulation of pharmaceutical firms. ΔRestrictionst is the log change in the number of restrictive words used in the laws pertaining to the pharmaceutical industry and issued by U.S.
regulators during year t. ΔReg_activityt is the log change in the word-length of laws in the pharmaceutical industry issued by U.S. regulators during year t. ROEt–1 is the weighted-average ROE of pharmaceutical firms in year t–1. Hypothesis 3 predicts a positive coefficient on lagged ROE. To control for the effects of the ruling party on the regulatory activity, we employ an indicator Partyt, which is set to 1 if Republicans dominate the U.S. House, and 0 otherwise.
Controlling for the party of the U.S. President or the U.S. Senate majority does not change our inferences.
In the next step, we decompose reported ROE into its components in equation (10): ROEt– = ROE_adjt–1 + ROE_biast–1. ROE_adjt–1 is ROE adjusted using empirical amortization rates in year t–1. ROE_biast–1 is the difference between observed ROE and adjusted ROE in year t–1.
We report results for the empirical amortization rates, as we see this approach superior to using normative assumptions about the amortization rates. Hypothesis 4 predicts that regulators fixate on earnings and do not adjust for accounting bias; that is α2 = α3.
4. Data Our primary sample comprises pharmaceutical firms from U.S. and 6 major pharmaceutical markets based on the number of listed pharmaceutical firms. We obtain financial data for our analysis from Compustat North America and Compustat Global. Our U.S. sample spans years 1972-2012. Due to data availability, our non-U.S. sample includes only the years 1999 to 2012. We identify pharmaceutical firms based on their SIC code (2833, 2834, 2835, and 2836), which results in a comprehensive sample of 2,281 unique firms or 26,528 firm-year observations. We delete firms with missing data for main test variables and firms with negative book equity, as calculation of ROE is meaningless when book equity is negative. To control for outliers, we delete firm-year observations in the top and bottom 1 percentile of ROE distribution.
We further require that a pharmaceutical firm has at least 10 consecutive observations to model the effects of R&D accounting on ROE. Finally, we eliminate countries with less than 20 observations to obtain stable estimates of our test statistics. Our selection procedure leads to 3,000 data points or 413 unique pharmaceutical firms located in 7 countries: Australia (34), Canada (149), India (45), Japan (105), Sweden (22), U.K. (46) and U.S. (2,599). As a control group we obtain Compustat data for firms from all other industries and come up with 39,414 firm-years: Australia (1,327), Canada (1,929), India (1,597), Japan (1,802), Sweden (343), U.K.
(1,214) and U.S. (24,585). GDP data are from OECD (2014). Regulatory data are obtained from Mercatus Center’s RegData database that quantifies the federal regulations using text analysis (Al-Ubaydli and McLaughlin, 2014).
5.1. ROE of pharmaceutical and non-pharmaceutical firms Table 2 reports the accounting rates of return of pharmaceutical firms and nonpharmaceutical firms. We obtain a weighted average ROE for each country and year, and set weights to the share of book equity in the total equity of pharmaceutical firms in each countryyear. We find that in the U.S., the average accounting ROE of pharmaceutical firms over the last 40 years is 19.8%, and is substantially higher than the accounting ROE reported by firms from other industries (11.1%). Table 2 also shows that U.S. pharmaceutical firms report 12.5 p.p.
higher R&D intensity than firms from other industries.
Figure 1 further reveals that U.S. pharmaceutical firms report higher ROE in each of the sample years. The average ROE of U.S. pharmaceutical firms was in the order of 15-20% in the 1970ies and the first half of the 1980ies, ROE increased to 20-25% in the 1990ies, and then declined to 15-20% in the last decade (2000-2012). Despite this decline, U.S. pharmaceutical firms reported about 5 p.p. higher ROE in 2012 than firms from other industries. Overall, the pharmaceutical firms’ ROE is between 3.3 p.p. (year 2010) and 18.4 p.p. (year 2001) higher than ROE of non-pharmaceutical firms during the second half of our sample. This descriptive evidence supports hypothesis 1 by showing that accounting for R&D investments may bias ROE.
We find that ROE of non-U.S. pharmaceutical firms is low (4.7%) and that on average ROE of international pharmaceutical firms is 4.9 p.p. lower than ROE of firms from other industries. However, Figure 2 reveals that the country and year of analysis determine whether pharmaceutical firms “outperform” their peers from other industries. For example, Australian, Indian and Japanese pharmaceutical firms “outperform” their peers in 2012, while pharmaceutical firms in Canada, Sweden and U.K. report lower ROE than firms from other industries. The evidence that more mature firms from our international sample report lower profitability is consistent with hypothesis 2.
Table 3 presents the regression results for equation (2) where we explain the difference in the accounting rates of return between pharmaceutical and non-pharmaceutical firms with economic up- and downswings of business cycle and R&D intensity. We find that our model fits data well and explains about 30% of variation in ROE differences between pharmaceutical firms and firms from other industries. We find that in the U.S. the upward bias of ROE documented in our descriptive analysis is related to R&D intensity (coef. 0.431; t-stat. 2.51). Furthermore, we find that U.S. and international pharmaceutical firms report relatively lower ROE during economic upturns and relatively higher ROE when the economy contracts (U.S. coef. –0.399; tstat. 2.19; international coef. –0.887; t-stat. 2.78). These results support hypothesis 1.
We employ model (2) to shed further light on what makes R&D intensity coefficient insignificant in our international sample. Recall that international pharmaceutical firms report lower ROE than their counterparts from other industries. Column (4) of Table 3 shows that decreasing investment growth is responsible for the lower ROE of international pharmaceutical firms (coef. –0.845; t-stat. 2.74). This evidence is consistent with hypothesis 2. In contrast, we do not observe this effect in the U.S. data where it is not predicted (coef. –0.089; t-stat. 0.31).