R&D investments, capital expenditures, and earnings thresholds

FASB standards relating to research and development are currently found under ASC 730 in the FASB’s Accounting Standards Codification.

Phone interviews with a CFO and controller from two Fortune 500 companies suggest the capitalization of R&D is very common in practice.

Compustat data reveals that 90% of firms that report R&D expense also report significant capital expenditures (i.e., at least 1% of total book value of assets).

One example of this type of shifting is substituting the hiring of additional research personnel (where salaries would be reported as R&D expense) with purchases of additional laboratory devices/equipment to improve the productivity of existing personnel (which would be reported as capitalized R&D).

Discretion over R&D investments may be exercised in different ways. For example, firms may identify alternative future uses for an R&D investment, which would permit the expenditure to be capitalized as a tangible asset (SFAS No. 2). Similarly, firms may deliberately classify an R&D expenditure that should be expensed as an R&D capital expenditure. We cannot observe the specific actions that managers take when they cut R&D expense. However, managers may be less likely to engage in reclassification using accounting discretion (as compared to real earnings management) because this action could be subject to auditor scrutiny.

Interviews with a CFO and controller of two Fortune 500 corporations indicate that managers know very early in the year that firms may have an earnings shortfall because finance and the commercial side of the business have a good handle on the pulse of the market and whether demand is lower for the year and even beyond. Therefore managers can plan ahead to shift resources from expensed R&D to capital expenditures.

Other common R&D expenses firms can temporarily defer include travel and training for R&D personnel, the execution of clinical trials, and costs for developing prototype models. In addition, both interviewees indicate that managers are reluctant to lay off scientists because of the tacit knowledge scientists have and the high startup costs of hiring new scientists. The controller also commented that there is a reluctance to lay off because a scientist who is laid off will likely seek employment with a competitor.

Consistent with prior literature on R&D and capital expenditures (e.g., Jackson et al. 2009; Richardson 2006; McNichols and Stubben 2008; Darrough and Rangan 2005), we use average total assets as the scaler for all of the variables. To show that our results are not unique to the choice of the specific scaler, we also use sales and lagged total assets as alternative scalers. Untabulated results show that our inferences are qualitatively similar.

Some prior studies scale the year-to-year change in net income by the market value of equity (Burgstahler and Dichev 1997). We scale INCNI_it by average total assets, rather than the market value of equity, because other variables in our regressions are scaled by average total assets. However, when we scale INCNI_it by the market value of equity our inferences and conclusions do not change.

The range of values used in the threshold variables is similar to prior research. If the range of values for INCNI_it is changed from 0 to 0.005 to 0–0.010, our inferences and conclusions do not change. Likewise, if the range of values for MBE_it is changed from 0 to 1 cent per share to 0–3 cents per share, our inferences and conclusions do not change.

Following Bhojraj et al. (2009), we also use analysts’ median forecast in the second month of the fourth fiscal quarter as our proxy for the market’s expectation. Our inferences and conclusions do not change when we use this alternative timing to calculate MBE_it.

Both threshold variables, INCNI_it and MBE_it, are defined such that they are inclusive of a zero earnings increase and a zero earnings surprise, respectively.

Our inference remains unchanged when we use the Wooldridge (1995) test of endogeneity of the two variables. In addition, we perform diagnostic tests on our system of equations. First, we perform an F-test for the joint significance of our instruments to determine whether the additional instruments have significant explanatory power after controlling for the endogenous regressor. The results show that the instruments in the R&D model and capital expenditure model are jointly significant in explaining the endogenous variables (p-values <0.001). Furthermore, the related F-statistic far exceeds the critical value of 10, as suggested by Stock, Wright, and Yogo (2002). Second, we perform the Stock and Yogo (2005) test and can reject the hypothesis that our set of instruments are weak instruments.

We also use log transformed values of our dependent variables in Equations (1) and (2). Untabulated results show that our conclusions are robust to this alternate specification, as we find that firms reduce R&D expense and simultaneously increase capital expenditures when they narrowly achieve an earnings threshold.

R&D (CAPX) comprises approximately 65 (35) percent of total investment for the top quartile subsample, which suggests that these firms indeed have a significant amount of both R&D and capital expenditures that ensures flexibility to shift.

As low investment-intensity firms presumably have smaller investments in R&D, we conjecture that these firms may not be apt to manipulate R&D expense to achieve earnings thresholds. Rather, the positive coefficient on the INCNI_it variable suggests that these firms make additional investments in R&D when they achieve an earnings threshold, possibly to grow their R&D investment programs.

We do not scale the variables and instead control for the scaler, average total assets, on the right side of the equation. We do not scale the variables in order to estimate a total amount of capital expenditures shifted from R&D so that, in the second stage, we can scale the shifted amount using average total assets and obtain a firm-specific measure of shift. Alternatively, if we use scaled variables to estimate Equation (2), then, in the second stage, we would not be able to scale the estimated shifted investment, as it would become the estimated coefficient on THRESH from Equation (2). However, we obtain similar results when we scale variables in the first stage and then use unscaled variables in the second stage.

We are grateful to the reviewer for helpful suggestions with this empirical design.

In untabulated tests, we also test the equality of the coefficients on SHIFTED_CAPX_it across the two equations using a likelihood ratio test. We find that the coefficients on SHIFTED_CAPX_it are significantly different (p-value = 0.001 and p-value <0.001 for the small earnings increase and small positive earnings surprise samples, respectively).

R&D expense in year t + 1 may capture the reversal of R&D expense cuts in year t. However, R&D expense in year t + 2 to year t + 5 is less likely to be affected by the reversal of R&D expense cuts in year t. Our inferences are similar for these longer horizons. For instance, for the small earnings increase (small positive earnings surprise) sample, the coefficients on SHIFTED_CAPX_it for Equation (4) are −0.000 and −0.000 (−0.002 and −0.001) for year t + 2 and t + 5, respectively, and they are not statistically significant. In contrast, the same coefficients for Equation (5) are 0.025 and 0.031 (0.083 and 0.082) for year t + 2 and t + 5, respectively, and both coefficients are positive and significant at the 10% level or better (1% level).

Thirty-five percent of observations in our samples report a positive depreciation difference and the mean difference averages approximately $3.4 million.

A controller of a Fortune 500 corporation explained how the difference could arise in the financial statements: R&D assets are maintained in the same fixed asset sub-ledger as all other fixed assets. At the end of each accounting period, the accounting system mechanically feeds depreciation expense to the financial statements (i.e., statement of cash flows, income statement, and balance sheet to update accumulated depreciation). However, reclassification entries are made to move depreciation expense arising from R&D cost centers to the R&D expense line on the income statement, resulting in depreciation differences reported on the separate financial statements. The controller also explained that, when there is no observed depreciation difference in the financial statements, it most likely means the company does not make a reclassification because the amounts are not material.

Our findings and conclusions should be interpreted with some caution, as we cannot directly observe all the reasons firms may have a positive depreciation difference. Furthermore, firms are not required to disclose the amount of capitalized R&D.

We do not partition based on R&D intensity because low R&D intensity could indicate that firms capitalize a significant part of their R&D as capital expenditures. We also do not partition the sample based on total investment intensity because total investment intensity will not allow us to identify firms that are more likely to capitalize their R&D investment. In Section 6.3, we argue that firms with high total investment intensity would have more flexibility to shift; however, the shift could be made to non-R&D capital expenditures, capitalized R&D, or both, which we do not distinguish in Section 6.3.

For the small positive earnings surprise threshold, we state R&D expense on a per share basis with common shares used to calculate earnings per share. Since the small earnings increase threshold uses total Compustat earnings, it is not necessary to report R&D on a per share basis.

Alternatively, we also estimate the capital expenditure model using OLS, and results remain qualitatively unchanged. As there is a functional negative relation between the alternative measures of threshold and R&D for the R&D equation, results estimating the R&D equation (Equation 1) are not discussed.