Depreciation of Business R&D Capital

• Published date: 01 March 2020
• Author: Wendy C. Y. Li,Bronwyn H. Hall
• Date: 01 March 2020
• DOI: http://doi.org/10.1111/roiw.12380

© 2018 Internationa l Association for Res earch in Incom e and Wealth

1

DEPRECIATION OF BUSINESS R&D CAPITAL

By Wendy C. y. Li*

U.S. Bu reau of Economic An alysis

and

BronWyn H. HaLL

University of Ca lifornia at Berkeley

We develop a forward-looking profit model to estimate the depreciation rates of business R&D capital.

By using U.S. government data between 1987 and 2007, and the newly developed model, we estimate

both constant and time-varying industry-specific R&D depreciation rates. The results comprise a set of

R&D depreciation rates for major U.S. high-tech industries. They align with the main conclusions from

recent studies that the rates are in general higher than the traditionally assumed 15 percent and vary

across industries. The relative ranking of the constant R&D depreciation rates among industries is

consistent with industry observations and the industry-specific time-varying rates are informative about

the dynamics of technological change and the levels of competition across industries. Lastly, we also

present a cross-country comparison of the R&D depreciation rates between the U.S. and Japan, and

find that the results reflect the relative technological competitiveness in key industries.

JEL Codes: O30, C20, C80

Keywords: depreciation, intangible capital, research and development, technological change

1. introduCtion

In an increasingly knowledge-based U.S. economy, measuring intangible

assets, including research and development (R&D) assets, is critical to obtaining a

complete picture of the economy and explaining its sources of growth. Corrado et

al. (2007) pointed out that after 1995 intangible assets reached parity with tangible

assets as a source of growth. Despite the increasing impact of intangible assets

on economic growth, it is difficult to capitalize intangible assets in the national

income and product accounts (NIPAs) and therefore to measure their impacts on

economic growth. The difficulties arise because the capitalization involves several

critical but difficult measurement issues. One of these is the measurement of the

depreciation rate of intangible assets, including R&D assets.

The depreciation rate of R&D assets is required for capitalizing R&D invest-

ments in the NIPAs for two reasons. First, the depreciation rate is needed to construct

knowledge stocks—it is the only asset-specific element in the commonly adopted

Notes: We are grateful to Ernie Berndt, Wesley Cohen, Erwin Diewert, Carol Moylan, and Brian

Sliker, and three anonymous referees for very helpful comments. The views expressed herein are those

of the authors and do not necessarily reflect the views of Bureau of Economic Analysis.

*Correspondence to: Wendy C. Y. Li, Bureau of Economic Analysis, 4600 Silver Hill Road

Suitland, MD 20233, USA. (Wendy.Li@bea.gov)

Review of Inc ome and Wealth

Series 66 , Number 1, March 2020

DOI : 10.1111 /roi w.123 80

bs_bs_banner

162

161

2

user cost formula. This user cost formula is used to calculate the flow of capital ser-

vices (Jorgenson, 1963; Hall and Jorgenson, 1967; Corrado et al., 2007; Aizcorbe et

al., 2009), which is important for examining how R&D capital affects the productiv-

ity growth of the U.S. economy (Okubo et al., 2006). Second, the depreciation rate

is required in order to measure the rate of return to R&D (Hall, 2005).

As Griliches (1996) concludes, the measurement of R&D depreciation is the

central unresolved problem in the measurement of the rate of return to R&D.

The problem arises from the fact that both the price and output of R&D capital

are generally unobservable. Additionally, there is no arms-length market for most

R&D assets and the majority of R&D capital is developed for own use by the firms.

Therefore it is difficult to independently compute the depreciation rate of R&D

capital (Corrado et al., 2007; Hall, 2005). Moreover, unlike tangible capital which

depreciates partly due to physical decay or wear and tear, R&D capital depreciates

mainly because its contribution to a firm’s profit declines over time. The driving

forces are obsolescence and competition (Hall, 2005), both of which reflect individ-

ual industry technological and competitive environments. Given that these environ-

ments can vary immensely across industries and over time, the resulting (private)

R&D depreciation rates should also vary across industries and over time.

In response to these measurement difficulties, previous research has adopted

four major approaches to calculate R&D depreciation rates: patent renewal, pro-

duction function, amortization, and market valuation (Mead, 2007). As noted

by Hall et al. (1986) and summarized by Mead (2007), all approaches encounter

the problem of insufficient variation in R&D spending over time and thus cannot

separately identify R&D depreciation rates without imposing strong identifying

assumptions. As discussed in Mead (2007), estimates from amortization models

(Lev and Sougiannis, 1996; Ballister et al., 2003) are derived from a more general

set of models that attempt to explain the returns on R&D investment. However, the

estimates are subject to concerns related to strong assumptions such as an assumed

relationship between the amortization rate of R&D capital and its earnings, and

operating income serving as a proxy for R&D benefits (Lev and Sougiannis, 1996).

Given the fact that firms’ propensities to patent vary across industries and

technology areas, the patent renewal approach cannot capture all innovation activ-

ities (Hall et al., 2014). Moreover, innovations may remain valuable even if their

patents have expired, given the other ways in which firms capture returns to R&D

(Levin et al., 1987). The patent renewal approach also suffers from the failure to

observe the right hand tail of a very skewed value distribution due to the relatively

low level of renewal fees. The identification problem can be mitigated by using

citation-weighted patent data, but there is a truncation bias problem arising due to

an incomplete observed citation life of patents (Hall et al., 2000).

Using the production function and market value approaches has the advan-

tage of incorporating all R&D rather than just that which is patented. However,

these approaches generally rely on the assumption that the average realized rate

of return is the same as the expected rate of return (Hall, 2005). This assumption

allows one to back out the depreciation rate which makes the two consistent. We

use a similar approach here, in that we assume a normal rate of return to R&D

when computing the profit function, although we do not explicitly require it to be

equal to the realized rate.

163

162

Review of Income and Wealth, Series 66, Number 1, March 2020

© 2018 Internationa l Association for Res earch in Incom e and Wealth