Intangible Measurement Matters:

Price Informativeness and Opaque Signals

Abstract

Empirical finance commonly deflates nominal measures using the book values of assets; however,
this practice omits internally generated intangible capital. Recent literature utilizes multiple proxies
to measure these off-balance sheet (OBS) intangibles, suggesting they harbor informational content
about future performance that departs significantly from that of the book values of assets. We apply
an extended q-theory framework that includes divergent measures of intangibles to test price infor-
mativeness. Our framework generates multiple predictions, demonstrating that intangible capital
significantly alters the information content of market prices and should be incorporated into em-
pirical asset pricing frameworks. Results indicate that prices deflated by OBS intangibles provide
incremental information beyond the book values of assets for firm fundamentals such as cash flows,
R&D, patents, and returns. However, the heterogeneous and opaque nature of intangible assets im-
plies no single proxy captures their full informational content, and thus tests based on a single asset
deflator are susceptible to measurement error. We show that firms with higher intangible intensity
exhibit lower price informativeness, consistent with the opacity and valuation challenges associated
with intangibles. By contrast, firms with higher physical capital intensity exhibit greater price infor-
mativeness attributable to the greater transparency of tangible assets.

JEL Codes: G11, G12, G17

Keywords: Intangible Assets, Price Informativeness

1 Introduction

Well-functioning financial markets lead to stock prices that “provide accurate signals for resource allo-

cation” (Fama, 1970, p. 383), implying prices are informative for future firm fundamentals. Bai et al.

(2016) present a theoretical and empirical framework that derives and tests a welfare-based measure of

price informativeness (PI). Using this framework, Carpenter et al. (2021) examine PI changes in China,

Kacperczyk et al. (2021) show that foreign institutional owners improve PI, and Koijen et al. (2024)

investigate capital flows and regulatory changes on PI. These papers apply q-theory and evaluate PI

as the coefficient of the log of market capitalization deflated by total assets. However, the book val-

ues of assets exclude internally generated intangible capital, which has become increasingly large in a

service-oriented economy that relies less on physical capital and more on knowledge and organization

capital.1

Belo et al. (2022), Crouzet et al. (2022), Eisfeldt et al. (2022), and Peters and Winters (2025)

establish intangible capital as a central driver of firm value, exposing the limits of traditional, tangibles-

focused measures in valuation and performance metrics. Furthermore, the rapid growth of intangibles

has occurred largely off-balance sheet (OBS), as accounting standards prohibit the capitalization of in-

ternally generated intangible assets such as knowledge and organizational capital. Indeed, Crouzet et al.

(2022) note that intangible assets are inherently difficult to quantify and interpret, posing significant

challenges for external investors. Eberly (2022) documents that intangible capital is less transparent,

more scalable, more challenging to measure, and arguably more forward-looking than physical capital.

Does a lack of a universally accepted valuation metric for these complex assets obscure the information

contained in OBS intangibles, rendering prices less informative about future fundamentals?

Using four prominent approaches to measuring OBS intangibles, we investigate whether these assets

possess the same informational content as the book values of intangibles and physical capital. Peters

and Taylor (2017) estimate internally generated intangible capital as the sum of knowledge and orga-

nizational capital. These estimates are derived using the perpetual inventory method and capitalizing

SG&A and R&D expenditures. Eisfeldt et al. (2022) propose an alternative measure using different as-

sumptions to capitalize only SG&A. By contrast, Woeppel (2022) values intangible capital using patents

as a direct measure of corporate innovation. Finally, Ewens et al. (2025) estimate knowledge capital
1
Following Ewens et al. (2025) and Gulen et al. (2025), the book values of assets and intangibles refer to the values of
total assets and intangibles recognized on the balance sheet. Similarly, physical capital is the tangible capital recognized
on the balance sheet as Net Property, Plant and Equipment. OBS intangibles are internally created intangible capital that
is not included on the balance sheet.

1
and organizational capital using capitalization parameters based on market-based exit prices. These

papers highlight both the challenge of measuring intangibles and the significant value omitted from the

balance sheet.

Accurate measurement of intangible capital is critical for financial decision-making, as the transition

to an intangible-intensive economy reshapes financial systems and asset valuations (Li, 2025b). Correctly

accounting for intangibles resolves several empirical anomalies, including the investment puzzle (Crouzet

and Eberly, 2019), corporate cash policies (Falato et al., 2022), the profitability of “star” firms (Ayyagari

et al., 2024), and the declining relevance of book-to-market ratios (Park, 2022; Gulen et al., 2025; Li,

2025a). We extend this literature by investigating how OBS intangibles affect price informativeness,

focusing on the market’s ability to anticipate future fundamentals such as cash flows, innovation output,

and returns.

We begin by documenting fundamental differences between the book values of intangibles and four

prominent OBS intangible capital metrics. Pairwise t-statistics reveal that the means, variance, and

change in means over time are significantly different – not only between book values and OBS metrics,

but also between different OBS metrics. Box plots, violin plots, and histograms further highlight that

the means, interquartile ranges, and distributions are distinctly different across all five metrics. We

further find that the economic magnitude of OBS intangible assets is considerable, averaging 189% to

356% of the book values of intangibles. The gap between OBS intangibles and both book values and

physical capital is growing. For instance, Ewens et al. (2025) note that incorporating their measure

of intangibles into the market-to-book ratio reduces its upward trend by 68%, while documenting that

OBS intangibles exceed physical capital. This economically significant increase highlights the urgency

of accurately measuring intangibles.

To clarify how heterogeneity in information and multiple measures of OBS intangibles influence price

informativeness, our paper employs a conceptual q-theory framework adapted from Bai et al. (2016)

to motivate four testable predictions. The first prediction is that OBS intangibles provide incremental

information beyond the book values of assets and therefore influence forecasting price efficiency (FPE).2

The results show that all four estimates of OBS intangible assets significantly predict firm fundamentals,

including cash flow, R&D, patents, and returns. Further, we construct an incremental measure of OBS

intangibles and show its significance in forecasting firm fundamentals. Hence, the traditional book values

of intangibles omit relevant information, and forecast-encompassing tests additionally demonstrate that
2
Forecasting price efficiency and price informativeness are used interchangeably in the present paper.

2
OBS intangibles contain significant information beyond the book values of intangibles.

We also assess the degree of informational overlap among alternative measures of intangibles. Our

second prediction is that, given the lack of consensus in measuring intangibles and their substantial

heterogeneity across firms, testing price informativeness with a single metric is subject to significant

measurement error. Forecast-encompassing tests reveal that the informational content of OBS intangi-

bles is multidimensional. Specifically, multiple principal components offer incremental predictive power

for cash flow, R&D, patents, and returns. Furthermore, we document that the information content in

the OBS intangible measures is statistically significant and distinct. Alternative measures of OBS assets

exhibit incremental predictive power for future firm fundamentals, indicating that no single proxy fully

captures the complexity of intangible capital.

Our third prediction is that price informativeness is negatively related to intangible intensity. Indeed,

the opacity of intangible-intensive firms can hinder the incorporation of fundamental information into

market prices, providing a reason to examine the heterogeneity of PI between high and low intangible

intensity levels. Supporting this hypothesis, we find that firms with high levels of intangible intensity

(top 30%) exhibit significantly lower price informativeness for future cash flows compared to firms with

low levels of intangible intensity (bottom 30%).

Our analysis also directly addresses tangible capital. Our fourth prediction is that firms with higher

physical-to-intangible capital ratios are expected to exhibit greater price informativeness than firms

with lower ratios. Regression analysis supports this relationship, which is consistent with the theoretical

prediction that more transparent forms of capital facilitate more accurate market pricing. These results

show both that the transition to an economy less reliant on physical capital has important implications

for price informativeness and that measures that ignore OBS intangible information are significantly

biased.

We address whether our findings are the result of a correction of the measurement error present

in book values or whether OBS intangibles capture unique informational signals about future cash

flows. We find that the results are driven by the latter; intangible intensity and differential information

significantly impact price informativeness, whereas the influence of the deflator is negligible. These

findings are robust across principal component and encompassing tests. Furthermore, using sales as a

deflator, a measure with minimal measurement error, confirms the relevance of four OBS measures and

intangible intensity for forecasting cash flows.

3
 Our findings have important implications beyond price informativeness. The results highlight that

there is no ‘true’ definition of assets, as estimates of OBS intangible assets vary significantly across

measurement methodologies and differ materially from the book values of intangibles. We demonstrate

that the book values of intangibles are incomplete proxies for forecasting future firm fundamentals,

and furthermore, that no single OBS metric fully captures the omitted information. The inherent

complexity and opacity of intangible assets warrant consideration of multiple deflator specifications in

financial ratios. For investment analysts, our findings indicate that incorporating OBS intangibles into

asset measures yields superior insights into future corporate performance.3

The remainder of this paper proceeds as follows. Section 2 describes the challenges and approaches

to measuring intangible capital. Section 3 presents a conceptual framework based on q-theory that

incorporates OBS intangible assets and presents testable predictions for the relationship between OBS

intangibles, intangible intensity, physical capital intensity, and price informativeness. Section 4 describes

our data and methodology for evaluating intangible assets and price informativeness. Sections 5 and 6

discuss our main empirical results and the robustness of our findings. Section 7 provides our conclusions.

2 Measuring Intangibles

Our investigation is motivated by the growing dominance of intangible assets and the consequent mis-

measurement of corporate capital. Belo et al. (2022) show that knowledge capital is an increasingly

important driver of firm value, a finding that holds across industries. Yet, Lev (2018, 2019) and Gulen

et al. (2025) argue that balance sheets grossly underestimate intangible capital. Crouzet et al. (2022)

and Ewens et al. (2025) find that intangible assets now comprise roughly half of total corporate capital.

Eisfeldt et al. (2022) identify a growing measurement gap, noting that because investments in employees,

brands, and knowledge capital are expensed, they are absent from corporate balance sheets.

The significance of this mismeasurement is amplified as intangible assets possess attributes distinct

from physical capital, and these characteristics likely influence their price informativeness. Crouzet

et al. (2022) identify two fundamental properties of intangible assets: non-rivalry in use and limited ex-

cludability. Each of these properties has implications for returns on corporate investment in intangibles.

Non-rivalry in use indicates that intangibles are scalable. For firms with a broad scope, intangibles can

be deployed across a larger range of operations at a lower marginal cost than tangible capital. Limited
3
Lev (2001) argues that the “risk assessment of firms containing intangibles is (or should be) at the core of investment
analysis, particularly given the deficient public information about intangibles.”

4
excludability implies that intangibles are subject to spillovers to competitors that inhibit firms from

capturing profits from innovation. Therefore, intangible assets are characterized by greater uncertainty

regarding their useful economic life due to risks of obsolescence, technological disruption, and infringe-

ment of ownership rights. Consistent with the riskiness of intangibles, Hou et al. (2022) find that R&D

intensity is associated with increased return volatility and a higher probability of default.

Anderson et al. (2025) argue that assets in the intangible category have fewer shared character-

istics than other balance sheet categories, and intangible assets are subject to greater psychological

biases even from professional financial analysts. Intangibles have greater information asymmetry, lower

liquidity, lower pledgeability as collateral, and greater investment indivisibility than tangible assets.

Because intangibles are inherently difficult for external stakeholders to value (Eberly, 2022), investment

in such capital fundamentally alters the firm’s information environment. Whereas physical capital has

more homogeneous and transparent value throughout an industry, Aboody and Lev (2000) argue that

intangible assets exacerbate information asymmetry due to their firm-specific nature, facilitating larger

insider trading profits in R&D-intensive firms.

Gu and Wang (2005) and Barron et al. (2002) demonstrate that the complexity of intangible as-

sets contributes to analyst forecast errors and disagreement. Similarly, Chan et al. (2001) and Cohen

et al. (2013) illuminate investor struggles to price intangible assets, leading to systematic underreac-

tion, mispricing, and anomalies related to past R&D performance. More recently, Wu and Lai (2020)

document a positive association between intangible intensity and information asymmetry, arguing that

the opacity of intangibles facilitates managerial concealment of negative news, resulting in higher crash

risk and return volatility. Such informational frictions are structurally reinforced by the exclusion of

internally generated knowledge and organizational capital from the book values of assets (total assets).

We next discuss the limitations of total assets and how measures of OBS intangible capital address

these weaknesses.

2.1 Total Assets (BV)

Total assets are traditionally measured using Compustat item at. This book value of total assets

(ABV ) includes the book values of tangible assets (ppent) and intangible assets (intan), which is the sum

of goodwill (gdwl ) and other intangible assets (intano). The measure of intangible assets on the balance

sheet has numerous deficiencies. First, the definition of goodwill has evolved over time, leading to an

5
inadequate measurement of externally acquired intangible assets between firms.4 When a business is

acquired, goodwill is recognized as the excess of the purchase price over the net identifiable book assets.

Although classified as an intangible asset on the acquirer’s balance sheet, goodwill could arise from

paying a premium on the purchase of the tangible assets of a target.

Second, accounting rules treat externally and internally created intangible assets differently. Other

intangible assets (intano), such as copyrights, patents, and trademarks, represent externally created

intangible capital recognized on the balance sheet during business combinations. However, most in-

tangible capital created internally is expensed rather than capitalized. When expensed, investments in

knowledge and organizational capital – such as R&D, advertising, employee development, and IT/data

– are not reflected in the book values of assets.5

The weaknesses of total assets inhibit meaningful longitudinal and cross-sectional comparisons be-

tween companies that make different choices regarding accounting treatment and internal vs. external

investment decisions. Yet, studies on PI typically use total assets as a deflator, even as intangible capital

has become an increasingly important component of corporate value in our knowledge-based economy.

To address this issue, Sections 2.2–2.5 define methods of estimating OBS intangible assets that we

use to construct comprehensive measures of corporate assets. Our contribution lies in highlighting the

significant effect of these estimates on price informativeness.

2.2 Peters and Taylor, 2017 (PT)

Peters and Taylor (2017) address the challenge of valuing intangible capital by developing estimates

of internally generated knowledge and organizational capital. Their results indicate that, on average,

firms acquire less than 20% of their intangible capital from external sources, implying that a majority

of intangible assets are not reported on the balance sheet. They estimate knowledge capital (Gi,t ) by

capitalizing R&D expenditures using the perpetual inventory method:

Git = (1 − δR&D )Git−1 + R&Dit , (1)

4
Appendix 1 defines the variables used in our study, and the description of gdwl explains the change in accounting
rules in 2001 for business combinations.
5
The complexity of corporate expenditures on intangible capital is illustrated by the inconsistent accounting treatment
of R&D and advertising expenses. While firms have been required to disclose R&D expenses since 1975, they are also
required to separately report advertising expenses (an investment in brand value) when these expenses are material. Yet,
Liang (2024) finds that almost 70% of firm-years in Compustat lack this expense data.

6
where δR&D is the R&D depreciation rate from the BEA industry guidelines (Li and Hall, 2020). A

depreciation rate of 15% is used for companies in industries without guidelines.

They estimate organizational capital (Si,t ) using a similar approach:

Sit = (1 − δSG&A )Sit−1 + γS (SG&Ait − R&Dit ), (2)

where the depreciation rate (δSG&A ) = 20% and the portion of the expense being capitalized (γS )

= 30%.6 Compustat typically includes R&D expense in SG&A. Therefore, it is subtracted from the

formula for organizational capital because it is already included via knowledge capital. Lastly, total

assets (AP T ) are the sum of book assets, knowledge capital, and organizational capital.

2.3 Eisfeldt et al., 2022 (EKP)

Based on work by Eisfeldt and Papanikolaou (2013) and Eisfeldt and Papanikolaou (2014), Eisfeldt

et al. (2022) develop an alternative method of estimating OBS intangible capital. Their estimate is

similar to Equation (2) but for two important differences. First, they use γS = 100% because some

industries rely heavily on intangible capital, and the assumption that γS = 30% is based on dated

studies with small sample sizes. Second, they treat R&D expense the same as advertising expense,

which is not subtracted from SG&A in Equation (2). In computing intangible capital, goodwill (gdwl )

is subtracted to exclude merger activity and avoid double-counting this intangible capital. Therefore,

total assets using this approach (AEKP ) equal book assets plus OBS intangible capital minus goodwill.

2.4 Woeppel, 2022 (W)

To address underreporting of patent market value, Woeppel (2022) estimates intangible capital using

patents as a direct measure of corporate innovation. While this represents a narrower measure than

knowledge and organizational capital, patents can be directly valued using the market reaction to the

patent announcement. Furthermore, patent capital may be more durable; unlike organizational capital,

which is often contingent upon employee retention, patents remain with the firm regardless of labor

turnover. Woeppel shows that a modified Tobin’s Q, reflecting patent capital, explains more variation

in investment than the standard Tobin’s Q and the Peters and Taylor (2017) version of Q.
6
These assumptions follow the literature. The depreciation rate of 20% is used by Eisfeldt and Papanikolaou (2013)
and Falato et al. (2022), and γS = 30% is used by Hulten and Hao (2008) and Eisfeldt and Papanikolaou (2014).

7
 Using this approach, total assets (AW ) equal book assets plus patent capital. Patent capital is the

three-day idiosyncratic return following the announcement of a newly granted patent. The firm-level

patent capital is determined using the perpetual inventory approach, which includes the value of new

patents and depreciates old patents at the industry depreciation rate for capitalized R&D. This approach

to estimating total assets does not include explicit proxies for knowledge and organizational capital.

2.5 Ewens, Peters, and Wang, 2025 (EPW)

Ewens et al. (2025) builds on prior research by using market prices to estimate the parameters of

the intangible capitalization model. Like Peters and Taylor (2017), they view OBS intangible capital

as composed of knowledge capital and organizational capital. Their approach estimates these items

using Equations (1) and (2) with several modifications. First, they follow Eisfeldt et al. (2022) by not

subtracting R&D from SG&A. Second, they use market-based exit prices to estimate δR&D and γS . This

allows their estimates to reflect different industry parameters.7 Using this methodology, total assets

(AEP W ) equal the sum of book assets, knowledge capital, and organizational capital.

2.6 The (Non)Equivalence of Intangible Measures

Table 1 presents summary statistics for the intangible asset measures, highlighting material differ-

ences across methodologies. The average (median) book value (BV) of intangible assets is $2,463.7

million ($299.2 million), whereas the four OBS estimates are 1.9 to 3.6 times larger in terms of means

and 2.4 to 9.1 times larger in terms of medians. Tests reject the equality of both means and variances

between each OBS measure and BV, indicating that OBS intangibles are significantly greater in both

magnitude and dispersion. For comparison with tangible capital, the final column reports the average

Net Property, Plant, & Equipment (NPPE) of $2,962.9 million, which is below all four OBS intangible

measures. Across the full sample period, the average NPPE is 20% greater than the average BV in-

tangibles; however, this relationship reverses by the end of the period, with BV intangibles exceeding

NPPE by 15% in 2022.

The change in mean (∆µ) over time shows that growth in OBS intangibles substantially outpaced the

increase in both the BV of intangibles and NPPE. The differences in ∆µ between the OBS measures

and BV are significant at the 99% confidence level; the ∆µ of BV is less than half of three of four
7
Some research suggests low depreciation rates for certain intangible assets. Emphasizing the long-term advantages
gained by first movers in a product category, Bronnenberg et al. (2012) estimates a 2.5% depreciation rate for brand capital
associated with consumer packaged goods.

8
OBS measures. Panel B presents summary statistics for the five measures of assets. Tests of mean

and variance also reject the equality between BV and each total asset measure incorporating OBS

intangibles at the 99% confidence level. On average, including OBS intangibles increases total assets by

48% relative to book values, with the magnitude of the adjustment ranging from 21% to 80%, depending

on valuation methodology.

Table 2 presents results of pairwise t-tests comparing five alternative measures of intangible assets.

Panel A reports tests for the equality of means, and all t-statistics are highly significant, rejecting the

null hypothesis of equal means at the 1% level across all comparisons. Column (1) shows that each

OBS intangible measure differs statistically from the BV measure, while columns (2)–(4) indicate the

OBS measures are also statistically distinct from one another. Panel B reports the ratios of standard

deviations for each pair of measures. The results show large variations in variance; e.g., the standard

deviation of PT is 2.6 times greater than BV. All associated tests reject the null hypothesis that the

standard deviation ratio equals one at the 1% significance level, confirming that they differ significantly

across methods. Panel C evaluates the mean change (∆µ) in each intangible measure over the sample

period. Except between PT and EKP, t-tests reject equality of mean changes, implying that most

metrics experience significantly different growth rates over the sample.

Collectively, these results support three conclusions. First, book values consistently and significantly

understate the value of intangible assets. Second, the five intangible asset measures exhibit statistically

distinct distributions, differing in both magnitude (Panel A) and dispersion (Panel B). Third, this

mismeasurement (relative to BV) appears to be growing over time (Panel C). Taken together, we

conclude that these five approaches yield substantially and statistically different estimates of intangible

assets.

Figures 1 and 2 provide box plots to illustrate that the mean, standard deviation, and skewness differ

considerably across the various intangible asset measures. Figure 1 deflates the metrics by BV, whereas

Figure 2 uses the corresponding total assets measure for each approach (e.g., PT intangible assets are

deflated by total assets that include PT intangible assets). The plots confirm that all four OBS metrics

exhibit significantly higher means and wider interquartile ranges compared to BV. These comparisons

highlight the importance of the deflator; outliers are less extreme in Figure 2, as large values influence

both the numerator (intangibles) and the denominator (total assets). The wide difference in the figures

implies that most of the variation in assets is not in the book value of assets, which includes current

9
assets and physical assets, but in OBS intangibles.

Figure 3 presents a violin plot that further confirms the distinct distributions across measures.

The BV distribution is tightly concentrated near zero, reflecting its omission of internally generated

intangibles. In contrast, PT shows a relatively uniform distribution, EKP’s central mass is positive,

and W exhibits a large mass at zero (similar to BV) but with a thicker positive tail. Figure 4 examines

the incremental intangible measures (OBS metric minus BV), as defined in Equation (15). If the OBS

metrics were similar to BV, these differences would cluster near zero. However, the plots demonstrate

that the incremental components for PT, EKP, W, and EPW are non-trivial. They exhibit larger means

and wider distributions than BV, implying the extra information in OBS intangibles is larger and has

more variance than book assets.

Finally, Figure 5 presents overlapping histograms. The top panel contrasts BV’s concentration of

near-zero values, consistent with its assumption of no organizational and knowledge capital, with the PT

metric. Because PT estimates internally generated intangibles using SG&A and R&D, its distribution

is more dispersed with fewer zeros. The middle panel compares distributions of EKP and EPW, while

the bottom panel contrasts BV and W, showing that although many firms report no patent capital,

others record substantial intangible values. In summary, the Figures and Tables 1 and 2 demonstrate

that intangible asset metrics differ significantly in both magnitude and distributional properties, which

leads to material differences in their corresponding deflators.

2.7 Illustrative Examples

Analysts compare firms using financial ratios, which often include the book values of assets. As

these measures omit OBS intangible assets, they increasingly fail to capture variations in corporate

strategies and investment decisions. To illustrate this deficiency, Appendix 2 compares measures of in-

tangible intensity for Best Buy and Ross Stores, which both operate within the Consumer Discretionary

Distribution and Retail Industry. From an accounting perspective, both firms appear to have relatively

stable intangible capital utilization. Best Buy maintains an intangible intensity of approximately 10%,

while Ross Stores has negligible intangible book assets throughout the 2001–2023 period.

However, these firms employ fundamentally different business models concerning intangible capital,

which are not reflected in their financial statements. As detailed in Appendix 3, Best Buy made

substantial investments in brand equity and human capital, whereas Ross Stores pursued a low-cost

10
strategy with minimal advertising and brand development. While the book value of intangibles fails to

reflect these divergent strategies, measures including OBS intangibles (PT, EKP, and EPW) indicate

increased investment by Best Buy and a relative decline in intangible focus for Ross Stores.

A second comparison examines Astronics and Oshkosh to further underscore the differences in the

intangible capital measurements. Both of these industrial firms operate in the Capital Goods Industry

and make significant sales to government and military clients. Astronics’ business strategy involves

manufacturing highly engineered electronic systems with a skilled workforce, while Oshkosh’s model

is oriented towards tangible capital and its significant manufacturing scale. Appendix 4 plots the

intangible intensity measures. All measures indicate an increasing intangible intensity for Astronics and

a declining intensity for Oshkosh.

The comparison of these firms supports two conclusions. First, relying solely on book values to

measure intangible intensity can be misleading. For example, BV intangibles in 2023 indicate that

Astronics and Oshkosh have similar intangible intensity (approximately 20% each). In contrast, OBS

measures reveal a significant disparity, showing that Astronics’ intangible intensity is substantially

higher than Oshkosh’s (e.g., 62% versus 31%, using the PT methodology). Second, the five measures

of intangible assets are not equivalent or substitutable. The measures exhibit significant differences in

both variation and magnitude, as shown in Appendix 5.

3 Conceptual Framework

This section introduces the conceptual basis for incorporating OBS intangible capital measures that

provide incremental information beyond the book values of assets. We examine how the inclusion of

OBS intangibles affects the extent to which market prices reflect or aggregate firm information.8

Based on the framework of Bai et al. (2016), we focus on forecasting price efficiency (FPE), also

referred to as price informativeness (PI), which quantifies the amount of information about future firm

fundamentals embedded in market prices. Specifically, FPE is defined as the variance of the predictable

component of firm value v/k̄, given the market-to-book ratio q = p/k̄, where v/k̄ ≈ 1 + z, z is the firm’s

production shock, p is price, and k̄ is current assets. That is

VF P E = V ar(E[z|q]). (3)
8
For brevity, we do not present the full model but instead highlight the key components underlying our empirical
analysis. The micro foundation of the price informativeness measure is provided in Appendix 6.

11
 Bai et al. (2016) empirically implement this measure by regressing cash flow (EBIT), scaled by the

book value of total assets A, on the log market-to-asset ratio (LM A), where market capitalization M

is also scaled by A. However, as Gulen et al. (2025) emphasizes, failing to incorporate OBS intangible

capital into asset measures introduces significant measurement errors; specifically, they show that using

Ewens et al. (2025) measure of OBS intangibles improves the Fama and French factor models, and the

value factor is no longer redundant in empirical factor models.

To address this, a firm’s current assets can be defined as k̄ = k̄ B + k̄ O , where k̄ B refers to balance-

sheet assets and k̄ O captures OBS intangible capital (e.g., Gulen et al., 2025). To more accurately

estimate the latter, we employ several established OBS measures from the literature, each designed to

capture distinct dimensions of intangible value. For instance, Peters and Taylor (2017) defines intangible

capital as the sum of knowledge and organizational capital. Eisfeldt et al. (2022) employ an alternative

capitalization approach with a distinct treatment of R&D and goodwill, while Woeppel (2022) use

patent-based proxies to represent innovation directly. Similarly, Ewens et al. (2025) extends Peters

and Taylor (2017) by incorporating market-based exit prices and industry-specific depreciation rates to

refine their estimation.

These four approaches differ significantly in construction, reflecting the diverse and complex nature of

intangible capital. As shown in Table 1, OBS measures have substantially higher means and variances

compared to traditional balance-sheet intangibles. t-tests confirm that the differences in both mean

and variance are statistically significant, rejecting the null hypothesis of equivalence. This finding is

also supported by the box plots in Figures 1 and 2, which illustrate that the book values of assets

systematically understate intangible values.

Since price informativeness is measured as the variation in future cash flows explained by market

prices (both deflated by assets), the composition of the deflator becomes critical. Relying solely on the

book values of assets underestimates firm values by ignoring the significant contribution of OBS intan-

gibles. Incorporating these measures enhances the informativeness of the deflator, thereby improving

the explanatory power beyond what the book values alone can offer.

Importantly, both cash flows (capturing the firm value component, v/k̄) and market prices (capturing

the firm’s market-to-asset ratio, p/k̄) in Equation (3) are influenced by changes in the asset deflator,

k̄, as in the empirical implementation in Bai et al. (2016). When OBS intangibles are included, the

composition and scale of the deflator change, which influences forecasting efficiency. Incorporating OBS

12
intangibles has two opposing effects on FPE. On one hand, it may increase the precision of insider

information (i.e., a lower ση2 ) by capturing more firm-specific information. On the other hand, due

to the complexity and opacity of intangible valuation, it may reduce trader signal precision (i.e., a

higher σs2 ), as information acquisition and processing become more costly and noisy (Crouzet et al.,

2022). According to Bai et al. (2016), a decrease in ση2 raises FPE, while an increase in σs2 reduces it.

Therefore, the net effect of incorporating OBS intangible capital on price informativeness is theoretically

ambiguous. We summarize this idea in the following prediction.

Prediction 1. OBS measures of intangibles provide incremental information beyond the book values of

assets and therefore influence price informativeness.

As discussed above, the various approaches to measuring intangible assets differ substantially in

construction, reflecting their complex and multidimensional nature. This variation is empirically evident

in the summary and t statistics in Tables 1-2 as well as the box and violin plots and histograms in Figures

1-5; they also show substantial differences between book value and OBS intangible measures.

Given this heterogeneity, how OBS intangibles are incorporated into asset proxies influences both

the type and amount of information traders produce. Since each proxy captures different dimensions of

intangibles, no single measure is likely to fully reflect the underlying firm value. As a result, relying on

a single OBS metric may introduce measurement error and limit the explanatory power. This motivates

the following prediction:

Prediction 2. The lack of consensus on how to measure intangibles, along with substantial firm-level

heterogeneity, suggests that relying on a single metric to test price informativeness introduces measure-

ment error, as no single approach can fully capture the information embedded in OBS intangibles.

As noted by Crouzet et al. (2022), intangible assets are inherently difficult to quantify and inter-

pret, posing significant challenges for external investors. This complexity reduces the transparency and

precision of financial information, particularly for firms with high intangible intensity. Such firms are

therefore more likely to produce opaque information compared to those with predominantly tangible

assets. Farboodi et al. (2022) shows that when average investors have less precise or accessible infor-

mation, market prices become less informative. Similarly, as demonstrated in Prediction 1 of Bai et al.

(2016), when traders produce less precise information due to the difficulty of valuing intangible-heavy

firms, the efficiency of prices in forecasting future fundamentals declines. Taken together, these insights

13
suggest that firms with greater intangible intensity may be at a disadvantage in how effectively mar-

ket prices reflect their underlying fundamentals, providing a basis for examining heterogeneity in price

informativeness as a function of intangible intensity.

Prediction 3. Firms with higher intangible intensity exhibit lower price informativeness than firms

with lower intangible intensity.

Extending this logic, we also consider the relative composition of physical versus intangible capital. A

high physical-to-intangible asset ratio indicates a greater reliance on tangible assets, which are generally

easier for investors to observe, verify, and value. This transparency reduces information asymmetry and

improves investors’ ability to interpret firm-specific information, thereby enhancing the informativeness

of market prices. In contrast, when firms rely more heavily on intangible capital, information becomes

more opaque and harder to process. As a result, the relative mix of physical and intangible assets is

likely to affect how efficiently markets incorporate information into prices, which is consistent with the

framework proposed by Farboodi et al. (2022).

Prediction 4. Firms with higher physical-to-intangible capital ratios exhibit greater price informative-

ness than firms with lower ratios.

4 Data and Methodology

4.1 Data

Our sample period spans 1980–2022, allowing for the analysis of subsamples with comparable lengths

before and after the 2001 accounting standard that eliminated goodwill amortization. The starting

year of 1980 is broadly consistent with prior studies on intangible assets (Peters and Taylor, 2017;

Eisfeldt et al., 2022; Ewens et al., 2025). These studies focus on the period following 1975, when the

FASB mandated the expensing of R&D costs. Each year, we construct a sample of large- and mid-

capitalization firms by including constituents of the S&P 500 Index and the 400 largest firms outside

the S&P 500. Although the S&P MidCap 400 Index, established in June 1991, is a commonly used

benchmark for U.S. mid-cap equities, its later inception precludes its direct use over our full sample

period.

Following Bai et al. (2016) and Carpenter et al. (2021), our methodology calculates stock returns

starting at the end of March and utilizes financial statement data from the prior calendar year, ensuring

14
sufficient time for investors to incorporate publicly available information. Following the methodology of

Peters and Taylor (2017), we exclude financials, regulated utilities, and non-operating establishments.

We also adopt their assumption that missing values of the Compustat item intan are set to zero.

Similarly, missing SG&A expenses are treated as zero, following Eisfeldt and Papanikolaou (2013).

We employ intangible capital estimates provided by the authors of each methodology. The PT

estimates of knowledge capital and organizational capital are available on WRDS. Edward Kim provides

the EKP estimates of OBS intangible capital, Michael Woeppel supplies the patent capital estimates

(W), and Michael Ewens provides intangible capital estimates for the EPW methodology.

Market prices are retrieved from the Center for Research in Security Prices (CRSP), and market

capitalization is calculated at the end of March for each year. This date is the starting point for the

subsequent one-year stock return. Financial statement data are obtained from Compustat for the prior

fiscal year, and Appendix 1 provides variable definitions. The natural logarithm of share turnover

(number of shares traded divided by the number of shares outstanding) is calculated over the 12-month

period corresponding to the fiscal year. Idiosyncratic volatility is measured as the standard deviation of

daily abnormal returns over the 252 calendar days prior to March 31, requiring a minimum of 21 trading

days (approximately one month) of return data for each stock (permno). Analyst coverage is measured

as the natural logarithm of the number of analysts, scaled by total assets. Finally, all variables are

winsorized at the 1% level to mitigate the influence of outliers.

4.2 Methodology

Following Bai et al. (2016), we estimate the informativeness of market prices in forecasting firm i’s

future cash flow using EBIT as a proxy for cash flow. Specifically, we examine the extent to which market

prices at time t incorporate information about cash flows realized in the subsequent year (CFi,t+1 ). The

empirical specification is given by:

   
CFi,t+1 Mi,t CFi,t
= α + β1 ln +ρ + γXi,t + ds 1s + ϵi,t+1 , (4)
Aj,i,t Aj,i,t Aj,i,t

where Mi,t denotes firm i’s market capitalization at time t, and Aj,i,t represents total assets measured

using five alternative asset definitions indexed by j: BV, PT, EKP, W, and EPW. Our measure of price

informativeness (PI) is the coefficient on the log of the market-to-asset ratio, ln(Mi,t /Aj,i,t ), which we

refer to as LMA in the subsequent analysis. The vector Xi,t includes firm-level control variables, and 1

15
denotes year and industry dummy variables.9

To account for dynamic firm behavior and unobserved heterogeneity, we include a lagged dependent

variable and fixed effects. To address the Nickell bias documented in dynamic panel models (Nickell,

1981), we follow Kacperczyk et al. (2021) and employ the system GMM estimator developed by Blundell

and Bond (1998), which extends the difference GMM approach of Arellano and Bond (1991). The

GMM estimator applies lagged levels as instruments for the differenced equation and lagged differences

as instruments for the level equation.

Because both dependent and independent variables vary across asset definitions, we standardize all

variables by subtracting their means and dividing by their standard deviations. As a result, conven-

tional model comparison metrics such as adjusted R2 or information criteria are not directly applicable.

Instead, we assess the informativeness of market prices by evaluating the magnitude and statistical

significance of the PI coefficient across asset definitions, with particular emphasis on the incremental

role of intangible capital measures.

4.3 Intangible Assets and Cash Flow

Previous studies that examine PI using the approach in Equation (4) do not include a separate

term for intangible assets and typically deflate variables using the book values of total assets. This

specification has two potential issues. First, it combines physical and intangible assets into a single

variable, despite their distinct roles in explaining future firm fundamentals. Second, book value based

total assets omit OBS intangible assets, which represent an increasingly important component of overall

corporate capital. To assess the impact of these issues, we examine how different measures of intangible

assets and intangible intensity affect price informativeness using several approaches. First, we test

whether adding a measure that includes OBS intangibles into total assets improves the forecasting

power of market prices for cash flows deflated by the book values of assets (BV), as measured by

Compustat item at. Specifically, we estimate:

     
CFi,t+1 Mi,t Mi,t CFi,t
= α + β1 ln + β2 ln +ρ + γXi,t + ds 1s + ϵi,t+1 , (5)
ABV,i,t ABV,i,t Aj,i,t ABV,i,t

This specification includes two LMA ratios: one based on book value assets (j = BV), and the other

based on total assets including OBS intangibles (j = PT, EKP, W, or EPW). However, because both
9
We adopt the industry classification modifications used by Ewens et al. (2025).

16
variables share the same numerator (market capitalization), this specification is susceptible to multi-

collinearity. To address this, we construct an alternative measure of incremental intangible capital that

does not depend on market capitalization. Specifically, we compute the difference between the two LMA
 
Aj,i,t
terms: LM ABV – LM Aj = ln ABV,i,t . This expression isolates the contribution of OBS intangibles

to total assets. It equals zero for firms without OBS intangibles and increases with the amount of OBS

intangibles. The model specification is:

     
CFi,t+1 Mi,t Aj,i,t CFi,t
= α + β1 ln + β2 ln +ρ + γXi,t + ds 1s + ϵi,t+1 . (6)
ABV,i,t ABV,i,t ABV,i,t ABV,i,t

A significant β2 supports Prediction 1, demonstrating that OBS intangibles provide incremental ex-

planatory power for future cash flows beyond what is captured by the book value of assets.
Second, to evaluate the role of intangible intensity and test Prediction 3, we sort firms within each
Fama-French 5 Industries using Ewens et al. (2025) based on their intangibles-to-assets ratio. We identify
firms in the top 30% and bottom 30% of intangible intensity and construct the corresponding dummy
variables.10 These dummies are then interacted with LM ABV to examine whether the informativeness
of market prices varies with intangible intensity, both in the full sample and across the two subgroups:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α+β1 ln +β2 ln +βH IH ×ln +βL IL ×ln +γZi,t +ϵi,t+1 , (7)
ABV,i,t ABV,i,t ABV,i,t ABV,i,t ABV,i,t

where IH and IL equal 1 for firms in the top and bottom 30% of intangible intensity, respectively, and

0 otherwise. The vector Zi,t includes the lagged dependent variable, firm-level controls, and year and

industry fixed effects, for brevity.

SHOULD BRIEFLY MENTION FEEDBACK

5 Results

5.1 Summary Statistics

Table 3 presents summary statistics for ratios. Panel A uses a common denominator (book values

of assets, ABV ), and confirms that the four ratios incorporating OBS intangible assets are substantially

and statistically different from the ratio based only on the book values (BV) of assets. The OBS
10
The use of this dummy variable structure is consistent with prior research. Gu and Wang (2005) find that analyst
forecast errors increase with intangible intensity that diverges from the industry median. These forecast errors are a proxy
for the informational complexity of intangible assets.

17
ratios exhibit significantly greater means and standard deviations than the BV ratios; specifically, their

standard deviations are 1.5 to 4 times larger. Furthermore, the average change over time in the OBS

ratios exceeds that of the BV ratio.

Panel B underscores the importance of the deflator choice by presenting ratios where each intangible

asset measure is deflated by its corresponding total asset measure (e.g., INP T /AP T ). A comparison

between Panels A and B shows that the specification of the asset measure materially influences the

magnitude of the ratio. For instance, the mean PT ratio is 0.52 when deflated by book assets (ABV )

but declines to 0.37 when deflated by its corresponding assets (AP T ), a difference exceeding 40%.

Panel C examines the market premium on assets, measured as the log market-to-assets ratio (log(M/Aj )),

which is used in the estimation of PI. The choice of asset measure (Aj ) impacts the ratio’s sign. Notably,

only the ratio based on book values (log(M/ABV )) has a positive mean, indicating that market capital-

ization exceeds book value. In contrast, all four OBS ratios are negative, implying that the measures,

including OBS intangible assets, consistently exceed market capitalization.

Finally, Panel D shows that these differences have a significant effect on the cash flow to assets ratio.

The mean and variance of the ratios that incorporate OBS assets are significantly lower than the ratio

using BV. This finding has important consequences for corporate performance analysis, which often

relies on profitability or efficiency metrics that divide earnings or cash flows by total assets. When a

measure of total assets omits a significant component of capital, corporate performance is overstated,

potentially by a substantial margin. In this case, the average cash flow return is 17%–64% higher when

total assets exclude OBS intangibles. Overall, the evidence highlights that incorporating OBS intangible

assets yields economically distinct and statistically significant differences in asset measurement.

5.2 Price Informativeness for Cash Flow

Table 4 presents the estimation results for Equation (4), where the dependent variable is the one-

year-ahead cash flow to assets ratio. Columns (1)–(5) report the five different measures of assets,

denote by CFjj , where the superscript denotes the dependent variable denominator and the subscript the

independent denominater. The specifications are built incrementally across the panels. Panel A includes

industry fixed effects, Panel B adds year fixed effects, and Panel C incorporates the full specification

with firm-level controls for institutional ownership, idiosyncratic volatility, industry concentration, and

analyst coverage. Appendix 1 defines these variables. Panel C serves as the baseline specification for all

18
subsequent regression analyses. The final row reports the number of observations, which we omit from

later tables for brevity, as the sample size remains consistent.

Results in Table 4 shows that the proxy for PI, the coefficient on LMA represented by β1 , is statisti-

cally significant across all five asset measures. However, a direct comparison of the coefficient estimates

across Models (1)–(5) is problematic, as the dependent variable (CF/Aj ) is constructed differently in

each specification. Furthermore, interpreting these comparisons is confounded by potential model mis-

specification. If OBS intangible assets and physical assets differ in their levels of price informativeness,

Equation (4) may be misspecified, making inference less reliable.

To address this issue, we estimate a model using Equation (6) that incorporates an additional term

that reflects information in the difference between LMA using BK and LMA deflated by OBS intangibles.

The top panel of Table 5 reports estimates using book assets as a common deflator. Consistent with

prior research, we find that the coefficient for LM ABV , the measure of PI, is positive and significant

across all five metrics. The coefficient for OBS intangible assets (β2 ) is large, positive, and statistically

significant when using the methodologies of PT, EKP, and EPW, even when controlling for LM ABV .

This provides strong support for Prediction 1, indicating that OBS intangible assets are incremental

informative about future cash flows.

We next turn to Prediction 3, which posits that price informativeness is lower among firms with high

intangible intensity. Equation (7) specifies the model, incorporating indicator variables for high (IH )

and low (IL ) intangible intensity. IH (IL ) takes the value of 1 if a firm falls in the top (bottom) 30%

of intangible intensity, and 0 otherwise. Panel B of Table 5 presents the estimates. In four of the five

models, βH is negative and statistically significant, consistent with the hypothesis that high intangible

intensity is associated with reduced price informativeness. The average coefficient for βH is –0.057 and

for βL is 0.043. F -tests reject the null hypothesis of equality (βH = βL ) for two of the five metrics at the

10% level. In addition, the joint restriction (βH = βL = 0) is rejected for four (five) metrics at the 5%

(10%) level. Taken together, the large and significant coefficients on the incremental intangible variable

(LM ABV − LM Aj ), combined with the findings for intangible intensity, imply that models omitting

intangible assets (e.g., Equation 4) are misspecified.

We examine the robustness of our results to different asset metrics (Aj ) with:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 . (8)
Aj,i,t Aj,i,t ABV,i,t Aj,i,t Aj,i,t

19
The vector Zi,t includes the lagged dependent variable, firm-level controls, and year and industry fixed

effects; these are omitted for brevity. Table 6 reports the estimation results. The coefficients on the

difference term LM ABV −LM Aj are large and statistically significant across all specifications except for

W, which uses patent capital as its OBS intangible asset proxy. These findings reinforce the predictive

power of intangible capital for one-year-ahead cash flows and provide further support for Prediction 1.

As shown in Panel B, F -tests indicate that the price informativeness dummies are significantly different

for three (four) of the measures at the 5% (10%) level, as the average difference between high and low

intangible intensity is –0.174 and hence relatively large. Moreover, the joint restriction βH = βL = 0 is

strongly rejected for all five intangible measures, implying that intangible intensity significantly affects

price informativeness and providing further support for Prediction 3.

Lastly, note the similarity of the PI coefficients; e.g., β1 are relatively similar across specifications.

This suggests the question which deflator is the most important for forecasting cash-flow is of secondary

importance; more important is the relatively large coefficients for LM ABV −LM Aj and the significance

of intangible intensity implies that OBS intangibles possess different important than the book value of

intangibles. In summary, our findings demonstrate that OBS intangible measures contain significant,

incremental information, and that firms with high intangible intensity exhibit lower price informative-

ness. These results indicate that the baseline model in Equation (4), which includes only a single LMA

term, is misspecified. This model fails to capture both the contribution of OBS intangibles and the

variation in informativeness across firms with differing intangible intensity.

Table 7 evaluates PI by deflating cash-flow by Sales; in this case, the denominator for cash-flow is

similar and hence we can more closely examine whether its the deflator or differential information driving

future cash-flow. Following Tables 5 and 6, we first evaluate the importance of the additional LM ABV −

LM Aj and then in Panel B also evaluate intangible intensity. The results show relatively similar β1

in all metrics, similar to Tables 5 and 6, as well as relatively large coefficients for LM ABV −LM Aj ,

which supports the importance of differential information in the intangibles of the OBS in forecasting

cash flow. Additionally, F tests reject the null that intangible intensity equals zero for four of the five

metrics. Overall, the findings indicate that OBS intangibles provide value-relevant information, beyond

what is captured by book value, for predicting firms’ fundamental performance.

20
5.3 Forecast Encompassing Tests

Which metric serves as a more effective proxy for intangible capital? The coefficient estimates from

the various specifications of Equation (8) are not directly comparable, as each uses a different scaling

variable (Aj ). To address this issue, we employ the forecast encompassing test proposed by Fair and

Shiller (1990) to statistically evaluate whether the information in one forecast is fully contained in a

competing forecast. This framework is represented by:

yi,t = α + β1 ŷ1,i,t + β2 ŷ2,i,t + ui,t , (9)

where yi,t is the realized outcome, regressed on the forecasts from two competing models (ŷ1,i,t and
ŷ2,i,t ). We first examine future cash flow scaled by the book values of assets (CFi,t+1 /ABV,i,t ). Model
1, shown in Equation (10), serves as the baseline and uses only the book values of assets (BV) to
\
generate forecasts, denoted as CF/A BV |BV , where the |BV denotes using BV to deflate the LMA

terms on the right-hand-side. This represents forecasts of cash flow deflated by BV that are generated
using the market cap and intangible intensity deflated by BV. Model 2, provided in Equation (11),
employs an alternative asset measure (j ) that incorporates OBS intangible assets to construct forecasts,
\
designated by CF/A , where LMA terms on the right-hand-side deflate by Aj . It forecasts the
BV |Aj

same independent variable as Model 1 but uses Aj to deflate market cap and intangible intensity. Thus,
the equations have different denominators on the right-hand side to facilitate direct comparisons as
shown in the following equations:
     
CFi,t+1 Mi,t Mi,t Mi,t
= α + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 . (10)
ABV,i,t ABV,i,t ABV,i,t ABV,i,t

     
CFi,t+1 Mi,t Mi,t Mi,t
= α + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 . (11)
ABV,i,t Aj,i,t Aj,i,t Aj,i,t

The forecast encompassing test of Fair and Shiller (1990) evaluates the significance of the coefficients

β1 and β2 in the following model:

CFi,t+1 \ \
= α + β1 CFi /A i BV |BV + β2 CF/ABV |Aj + ui,t . (12)
Ait ,BV |BV,i,t

CFi,t+1 \ \
= α + β1 CF/A BV |LM ABV,i,t + β2 CF/ALM AAj,i,t +ui,t .
ABV |LM ABV,i,t

If β1 is statistically significant and close to one, and β2 = 0 is not rejected, the baseline forecast (BV)

21
generated from Equation (10) encompasses the competing forecast (OBS), rendering the latter redun-

dant. However, if β2 is statistically significant, it indicates the model incorporating OBS intangibles as

the deflator provides predictive information incremental to that contained in the model deflated by the

book value of assets.

Table 8 reports the forecast encompassing tests. The first column in the top panel presents a Mincer-

Zarnowitz regression based solely on the forecast using the book values (BV ) of assets, yb1 . We find that

β1 ≈ 1, confirming that the forecast is unbiased. This specification serves as the baseline model, and

its adjusted R2 reflects the explanatory power of an information set limited to book values. Columns

(2)–(6) of Panel A present the Fair-Shiller encompassing tests for CF/ABV . In Column (2), both βBV

(0.52) and βP T (0.57) are statistically significant, suggesting that the PT deflator captures information

incremental to the book values of assets for forecasting future cash flows. Similarly, Columns (4) and

(5) reveal that the coefficients for W and EKP are positive and significant at the 1% level. In contrast,

Column (3) indicates that the coefficient for EKP is negative and only marginally significant at the 10%

level. The last column presents results using principal components derived from the four OBS intangible

forecasts. The first two principal components are statistically significant at the 1% level, indicating that

multiple OBS intangible deflator metrics provide significant information, consistent with Prediction 2.

This incremental information is economically relevant, increasing the adjusted R2 by 3.5%.

For robustness, we employ the forecast encompassing test of Chong and Hendry (1986). This test

determines whether a baseline model (Model 1) omits useful predictive content found in a competing

model (Model 2). We implement it by regressing the forecast errors from the baseline model on the

difference between the competing forecast and the baseline forecast:

d BV |Aj − CF
eBV = α + γ(CF d BV |BV ) + vt , (13)

where eBV is the forecast error from the baseline model from Equation (10), defined as CFBV −CF
d BV |BV

The independent variables CF

d BV |Aj and CF
d BV |BV are the forecasts from Models 1 and 2 above.

Chong and Hendry (1986) tests evaluate the incremental predictive content of alternative intangible

metrics, providing a direct test of Predictions 1 and 2. If information in the deflator is similar, γ = 0.

If γ > 0, it indicates that the baseline model’s forecast errors have different information, implying that

the baseline model fails to encompass the OBS model. Panel B of Table 8 reports the results, showing

that the OBS metrics PT, W, and EPW yield significant γ coefficients. This finding suggests that

22
OBS intangibles contain information that is incremental to book values, consistent with Prediction 1.

Furthermore, the results in Column (5) support Prediction 2; both principal components are positive

and significant in explaining baseline forecast errors, confirming that OBS measures enhance cash flow

forecasting.

While Table 8 demonstrates that OBS metrics offer incremental information for forecasting CF/ABV ,

Table 9 presents encompassing tests for forecasting CF/Aj . We present both sets of results because

a priori, the appropriate deflator is uncertain and BV may have relevant information for forecasting

CF/Aj . Note, this may seem counterintuitive as this measure includes BV. However, BV may be signifi-

cant if the information in OBS intangibles is incorrectly constructed due to incorrect model assumptions.

We find that LM ABV is insignificant in the PT and EPW specifications. In contrast, in Table 8 we find

that LM AAJ is significant in forecast PT, W and EPW. Combined with earlier results, this implies that

LM AP T and LM AEP W fully encompass the baseline model. In contrast, LM ABV remains significant

for EKP and W, indicating that neither model fully subsumes the other. The continued relevance of

LM ABV in these cases may reflect measurement errors, potentially arising from inaccurate assumptions

regarding the capitalization or depreciation of OBS intangible assets.

To further evaluate Prediction 2, we test whether any single OBS intangible metric encompasses

all relevant information, and Panel B of Table 9 presents the results. To mitigate multicollinearity,

we conduct a principal component analysis (PCA). When testing a specific metric (e.g., LM AP T ), we

extract principal components (PCs) from the remaining metrics (e.g., LM ABV , LM AEKP , LM AW ,

and LM AEP W ). We find that the first two principal components are statistically significant across all

specifications, with the sole exception of the first component in Column (3). This evidence suggests that

no single OBS metric captures the full spectrum of relevant information, aligning with Prediction 2.

Additionally, the Chong-Hendry encompassing tests reported in Panel C serve as a robustness check,

showing that BV only contains incremental information beyond that captured by EKP intangibles. The

results are broadly consistent with the findings reported in Panel A.

Collectively, the PCA results and pairwise encompassing tests indicate that no single intangible met-

ric is informationally complete. Incorporating information from multiple intangible metrics is important

for forecasting for both CF/ABV and CF/Aj .

23
5.4 Physical Capital Intensity

To test Prediction 4, we analyze whether price informativeness varies with physical capital intensity.
Firms are ranked within each industry based on the difference between physical capital (NPPE ) and
intangible assets, scaled by total assets (Aj ). Equation 14 follows a similar structure to equation 8, but
examines relative physical capital intensity. We define IHK and ILK as indicator variables for firms in
the top and bottom 30% of this distribution, respectively and estimate:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βHK IHK × ln + βLK ILK × ln + γZi,t + ϵi,t+1 . (14)
Aj,i,t Aj,i,t ABV i,t Aj,i,t Aj,i,t

Table 10 presents the regression results. We hypothesize BHK > 0 and BLK < 0 . βHK > 0 for all

five specifications and statistically significant in two models. Additionally, βLK is consistently negative

and significant in four cases, and supports our predictions. F -tests reject the null hypothesis of equality

between βHK and βLK in three (four) specifications at the 5% (10%) level, and reject the joint null

βHK = βLK = 0 in all cases. These results indicate that firms with more tangible capital are more price

informative, consistent with the view that tangible assets offer greater transparency than intangibles.

5.5 Price Informativeness for R&D, Patents, and Returns

To test the broader applicability of our findings on firm fundamentals and stock returns, we inves-

tigate whether measures of intangible assets influence the price informativeness in predicting a range

of firm outcomes beyond cash flows. We begin this extension by analyzing future R&D expenditures,

to determine whether the informational effects observed for cash flows also manifest in real corporate

investment decisions.

Panel A of Table 11 reports regression estimates for one-year-ahead R&D, scaled by five alternative

asset measures (Aj ). Across all five specification, we find that βH < 0 and βL > 0, indicating that firms

with high intangible intensity exhibit lower price informativeness, while those with low intensity exhibit

higher informativeness. F -tests reject the equality of these coefficients in four out of five cases at the

10% level.11

These findings also speak to the feedback role of stock prices in corporate decision-making. When

prices reflect relevant information, particularly about intangible capital, managers may learn from mar-

ket valuations and incorporate that information into their investment decisions. This mechanism, known
11
We further strengthen these findings by forecasting R&D/ABV , reported in Panel A of Table A.1. Across all specifi-
cations, βH is significantly negative, while βL is significantly positive and economically large. The statistically significant
difference between these coefficients indicates that models relying solely on book values of assets fail to capture key variation
in R&D forecasts.

24
as the feedback effect, has been well-documented in the literature (e.g., Bond et al., 2012, for a sur-

vey). In this context, lower price informativeness in high-intangible firms may hinder efficient capital

allocation, as market signals fail to fully convey intangible-related investment opportunities. Thus,

the informativeness of stock prices not only reflects investor understanding but may also shape real

investment behavior.

Taken together, these results provide evidence that firms characterized by high intangible intensity

operate in a distinct informational regime. We conclude that both the level of intangible intensity and

the incorporation of OBS intangible asset measures substantially influence the informativeness of stock

prices for forecasting R&D activity.

We next extend our analysis to future patents output, which represent the realized outcome of

innovation. This allows us to distinguish between inputs (R&D expenditures) and outputs (patents)

of intangible capital investment. Patents serve as non-financial, economically meaningful assets and

provide a test of whether stock prices can anticipate corporate innovation.

Panel B of Table 11 presents model estimates for changes in patents, scaled by Aj . F -tests reject

both the equality of coefficients (βH = βL ) and the joint restriction (βH = βL = 0) in four out of

five specifications. The OBS intangible proxy (LM ABV − LM AAj ) is statistically significant with

economically meaningful magnitudes. Panel B of Table A.1 reports results for specifications using ABV

as the denominator. Here, we find even stronger evidence supporting the relevance of OBS intangibles

and intangible intensity. Coefficients for IH are negative and significant at the 5% level across all

specifications. F -tests strongly reject both equality and joint insignificance in every case. Finally, the

coefficients on OBS intangible are positive, large, and highly significant in forecasting ∆P AT /ABV .

These findings reinforce the view that intangible asset are highly informative about future innovative

outcomes.

We then examine whether intangible assets also help predict future stock returns, with results

presented in Panel C of Table 11. Consistent with Bai et al. (2016), we anticipate negative coefficients on

LM ABV , in line with the value effect, where firms with high market-to-book ratios (growth firms) tend

to underperform. The first row confirms the expected results: across all specifications, the coefficient

on LM ABV is negative and statistically significant at the 1% level. Our contribution shows that

incorporating OBS intangible measures enhances return predictability. In three of the four specifications,

OBS intangible proxies exhibit positive and statistically significant coefficients, and their economic

25
magnitudes generally exceed those of LM ABV . This suggests that markets may underreact to intangible

capital, and incorporating OBS measures offers incremental predictive content. However, the evidence

on interactions with high vs. low intangible intensity is more mixed in the return context

To compare model performance more formally, we conduct forecast encompassing tests across the

R&D, patent, and return models. Table 12 presents the results. In eleven of twelve cases (Columns

(1)–(4)), the competing specifications that include OBS intangible assets produce statistically significant

forecasts, implying that the baseline models using only the book values of assets fails to encompass

the information in intangibles. Furthermore, in Columns (5) of Panels A and C, we find that multiple

principal components extracted from OBS are also statistically significant, further supporting prediction

2; no single intangible metric contains all price-informative information for forecasting R&D, patents,

and returns.

6 Robustness

6.1 Assessing Incremental Intangible Information

Section 5.2 provides a model (Equation (8)) that uses the difference between LMA terms as a proxy for

OBS intangible assets, and Table 5 presents the corresponding estimation results. To ensure robustness,

we define and test alternative proxies for OBS intangible assets. These measures are designed to capture

the incremental informational content of OBS intangibles while excluding the book value of intangibles

(Compustat item intan):

Knowledge CapitalP T + Organizational CapitalP T

IP T =
AP T
Intangible Assets − gdwl
IEKP =
AEKP
(15)
Patent Capital
IW =
AW
Knowledge CapitalEP W + Organizational CapitalEP W
IEP W =
AEP W

We use Equation (15) to define the construction of LM AIJ , which represents the log of market capital-
ization over the incremental (I) OBS intangible metric, J. To assess the incremental predictive content
of OBS intangible assets, we test this relationship using two model specifications. The first model is:
       
CFi,t+1 Mi,t Mi,t Mi,t Mi,t
= α+β1 ln +β2 ln +βH IH ×ln +βL IL ×ln +γZi,t +ϵi,t+1 . (16)
ABV,i,t ABV,i,t AIJ,i,t ABV,i,t ABV,i,t

26
LM AIJ measures the price informativeness of the incremental information in OBS intangibles; significance implies
 
M
that the conventional measure using BV to deflates ignores relative information. Additionally, ln AIP i,t
T ,i,t
, like
LM AJ , includes market capitalization in the numerator, we also estimate a variation of Equation (7), replacing
AJ with AIJ to isolate the incremental information content in PT, EKP, W and EPW measures. The second
model specification is:

       
CFi,t+1 Mi,t AIJ,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
ABV,i,t ABV,i,t ABV,i,t ABV,i,t ABV,i,t
(17)

where AIJ represents the incremental component as defined in Equation (15). Significance implies the

different in information between BV and OBS intangibles has relevant price informative information for

forecasting future cash flow.

Table 13 presents the results for Equation (16), which tests the incremental predictive content of the

intangible measures. In the top panel, the incremental information contained in PT, EKP, and EPW

significantly predicts CFBV . Moreover, the first two principal components (PCs) derived from all four

incremental intangible metrics are also significant, implying that information from multiple measures is

collectively important.

The bottom panel further validates Prediction 2, confirming that no single intangible metric is

informationally complete. This panel evaluates the differential information in CFj using P CIJ , which

represents the principal components of the incremental information metrics. The significance of multiple

PCs in this analysis provides additional evidence that a single intangible measure cannot summarize all

relevant information, reinforcing the finding that information from multiple metrics is important.

Equation (17) provides an alternative approach to measuring incremental intangible assets, and

Panel B of Table 13 reports the estimation results. We find that βH < 0 and βL > βH across all five

specifications; β2 is both large and statistically significant for PT, EKP, and EPW. Furthermore, the

significance of the first two principal components is consistent with Prediction 2. Taken together, the

evidence in Table 13 indicates that models of PI (Equation 5) using only the book values of assets

are misspecified. We document that the incremental information in OBS assets is economically large,

statistically significant, and provides substantial predictive power for future cash flows.

27
6.2 Recent Time Period and Forecast Horizon

To examine the temporal stability of our findings, Table 14 reports price informativeness (PI) esti-

mates for the 2002–2022 period, which is approximately half the full sample. We select this subsample

to mitigate the confounding effects of the dot-com bubble and to align with the post-SFAS 142 reporting

regime regarding goodwill impairment.12

Table 14, Panel A reports price informativeness estimates for cash flow for the subsample and can

be compared with the results for the full sample in Table 5, Panel D. The coefficients of LM Aj and

(LM ABV −LM Aj ) are higher for the recent subsample. For LM Aj , all five metrics exceed 0.71, and their

average is 0.76 compared to an average of 0.52 in Table 14. The observed increase in price informativeness

is consistent with advancements in financial reporting quality and information technology (Gao and

Huang, 2020; Gu et al., 2020; Dessaint et al., 2024). Another difference is that the coefficients for (IL )

are all positive and significant for recent years, but only one coefficient is positive and significant for

the full sample. In summary, our results are robust for the recent time period and infer OBS intangible

assets and intangible intensity matter for the PI of cash flow.

Panel B of Table 14 focuses on R&D price informativeness over the most recent two decades. We find

that the results within this subsample are largely consistent with the baseline evidence shown in Panel

A of Table 11. We observe a similar stability for patents in Panel C; these estimates are qualitatively

comparable to the full-sample analysis in Panel A of Table 11.

Finally, Panel D presents the estimates for returns. When compared to the full sample in Table

11, the coefficients on LM Aj (row 1) retain similar sign and significance. However, the results for the

difference term, LM ABV −LM Aj (row 2), are sensitive to the sample period; neither the PT nor EKP

estimates maintain consistency in sign or significance.

Lastly, we consider a three-year horizon for cash-flow in Table 15. Results are relatively similar to

Panel A of the prior table. The coefficient for LM ABV −LM Aj is significant for PT, EKP and EPW.

Further, there are significant intangible intensity differences for four of the five metrics. Additional

results for patents, R&D and returns are available upon request.
12
Appendix 1 details the accounting treatment of goodwill and the 2001 regulatory change to goodwill recognition and
impairment.

28
7 Conclusion

We show that the book values of assets are an incomplete metric for deflating assets for measuring

price informativeness. Over the past three decades, OBS (off-balance sheet) intangibles have surged

alongside the expansion of the knowledge-based economy. Using four distinct measurement methods,

we document that OBS intangibles are statistically significant and economically substantial relative to

the book values of intangibles. However, prior literature typically scales price informativeness measures

by the book values of assets (total assets), effectively overlooking significant corporate investment in

internally generated OBS intangibles.

This study applies a q-theory conceptual framework incorporating OBS intangibles to derive four

predictions, which are tested using cash flow, R&D, patent activity, and returns. We find that OBS

intangibles provide incremental information beyond the book value of intangibles for forecasting firm

fundamentals. Forecast encompassing tests and principal components analysis show that no single proxy

for OBS intangibles fully captures the information content of this capital. We show that firms with higher

intangible intensity exhibit lower price informativeness, suggesting greater opacity and more challenges

for investors in interpreting market signals compared to tangible-heavy firms. In contrast, firms with

higher physical-to-intangible capital ratios display greater price informativeness, implying that tangible

assets remain easier to evaluate, and thus more reliably reflected in stock prices. Consequently, asset

composition, specifically the balance between tangible and intangible capital, significantly affects the

predictive power of prices for firm fundamentals.

Our findings have direct implications for empirical finance and financial analysis. We caution against

the conventional use of book value deflators, given the growing economic importance of OBS intangibles

and their distinct predictive content. Furthermore, the heterogeneity of internally generated intangible

capital and the lack of a consensus measure indicate that researchers should consider multiple proxies

for OBS intangibles in their analyses.

29
References

Aboody, D. and B. Lev (2000). Information asymmetry, R&D, and insider gains. The Journal of

Finance 55 (6), 2747–2766.

Anderson, S. B., K. I. Mendoza, and C. Mongold (2025). The effect of intangible asset classification on

professional financial statement users’ assessments. Journal of Accounting Research, forthcoming.

Arellano, M. and S. Bond (1991, April). Some Tests of Specification for Panel Data: Monte Carlo

Evidence and an Application to Employment Equations. The Review of Economic Studies 58 (2),

277.

Ayyagari, M., A. Demirgüç-Kunt, and V. Maksimovic (2024). The Rise of Star Firms: Intangible

Capital and Competition. The Review of Financial Studies 37 (3), 882–949.

Bai, J., T. Philippon, and A. Savov (2016, December). Have financial markets become more informative?

Journal of Financial Economics 122 (3), 625–654.

Barron, O. E., D. Byard, C. Kile, and E. J. Riedl (2002). High-technology intangibles and analysts’

forecasts. Journal of Accounting Research 40 (2), 289–312.

Belo, F., V. D. Gala, J. Salomao, and M. A. Vitorino (2022, February). Decomposing firm value.

Journal of Financial Economics 143 (2), 619–639.

Blundell, R. and S. Bond (1998). Initial conditions and moment restrictions in dynamic panel data

models. Journal of Econometrics 87 (1), 115–143.

Bond, P., A. Edmans, and I. Goldstein (2012). The Real Effects of Financial Markets. Annual Review

of Financial Economics 4, 339–360.

Bronnenberg, B. J., J.-P. H. Dubé, and M. Gentzkow (2012, October). The Evolution of Brand Prefer-

ences: Evidence from Consumer Migration. American Economic Review 102 (6), 2472–2508.

Carpenter, J. N., F. Lu, and R. F. Whitelaw (2021, March). The real value of China’s stock market.

Journal of Financial Economics 139 (3), 679–696.

Chan, L. K., J. Lakonishok, and T. Sougiannis (2001). The stock market valuation of research and

development expenditures. Journal of Finance 56 (6), 2431–2456.

30
Chong, Y. Y. and D. F. Hendry (1986). Econometric Evaluation of Linear Macro-Economic Models.

The Review of Economic Studies 53 (4), 671–690.

Cohen, L., K. Diether, and C. Malloy (2013). Misvaluing Innovation. The Review of Financial Stud-

ies 26 (3), 635–666.

Crouzet, N. and J. Eberly (2019, May). Understanding Weak Capital Investment: The Role of Market

Concentration and Intangibles. Technical Report w25869, National Bureau of Economic Research,

Cambridge, MA.

Crouzet, N., J. C. Eberly, A. L. Eisfeldt, and D. Papanikolaou (2022, August). The Economics of

Intangible Capital. Journal of Economic Perspectives 36 (3), 29–52.

Dessaint, O., T. Foucault, and L. Frésard (2024). Does alternative data improve financial forecasting?

The horizon effect. The Journal of Finance 79 (3), 2237–2287.

Eberly, J. (2022, September). The Value of Intangible Capital. NBER Reporter, No. 3, National Bureau

of Economic Research.

Eisfeldt, A. L., E. T. Kim, and D. Papanikolaou (2022). Intangible Value. Critical Finance Review 11 (2),

299–332.

Eisfeldt, A. L. and D. Papanikolaou (2013). Organization Capital and the Cross-Section of Expected

Returns. The Journal of Finance 68 (4), 1365–1406.

Eisfeldt, A. L. and D. Papanikolaou (2014). The Value and Ownership of Intangible Capital. The

American Economic Review 104 (5), 189–194.

Ewens, M., R. H. Peters, and S. Wang (2025). Measuring Intangible Capital with Market Prices.

Management Science 71 (1), 407–427.

Fair, R. C. and R. J. Shiller (1990). Comparing Information in Forecasts from Econometric Models.

The American Economic Review 80 (3), 375–389.

Falato, A., D. Kadyrzhanova, J. Sim, and R. Steri (2022, October). Rising Intangible Capital, Shrinking

Debt Capacity, and the U.S. Corporate Savings Glut. The Journal of Finance 77 (5), 2799–2852.

31
Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal

of Finance 25 (2), 233–241.

Farboodi, M., A. Matray, L. Veldkamp, and V. Venkateswaran (2022, June). Where Has All the Data

Gone? The Review of Financial Studies 35 (7), 3101–3138.

Gao, M. and J. Huang (2020). Informing the market: The effect of modern information technologies on

information production. The Review of Financial Studies 33 (4), 1367–1411.

Gu, F. and W. Wang (2005). Intangible assets, information complexity, and analysts’ earnings forecasts.

Journal of Business Finance & Accounting 32 (9-10), 1673–1702.

Gu, S., B. Kelly, and D. Xiu (2020). Empirical asset pricing via machine learning. The Review of

Financial Studies 33 (5), 2223–2273.

Gulen, H., D. Li, R. H. Peters, and M. Zekhnini (2025, February). Intangible Capital in Factor Models.

Management Science 71 (2), 1756–1778.

Hou, K., P.-H. Hsu, S. Wang, A. Watanabe, and Y. Xu (2022). Corporate R&D and Stock Returns:

International Evidence. Journal of Financial and Quantitative Analysis 57 (4), 1377–1408.

Hulten, C. and X. Hao (2008, December). What is a Company Really Worth? Intangible Capital

and the “Market to Book Value” Puzzle. Technical Report w14548, National Bureau of Economic

Research, Cambridge, MA.

Kacperczyk, M., S. Sundaresan, and T. Wang (2021, February). Do Foreign Institutional Investors

Improve Price Efficiency? The Review of Financial Studies 34 (3), 1317–1367.

Koijen, R. S. J., R. J. Richmond, and M. Yogo (2024, July). Which Investors Matter for Equity

Valuations and Expected Returns? Review of Economic Studies 91 (4), 2387–2424.

Lev, B. (2001). Intangibles: Management, Measurement, and Reporting. Brookings Institution Press.

Lev, B. (2018). Intangibles. Working Paper, [Link] New York

University.

Lev, B. (2019, August). Ending the Accounting-for-Intangibles Status Quo. European Accounting

Review 28 (4), 713–736.

32
Li, L. (2025a, May). The Role of Intangible Investment in Predicting Stock Returns: Six Decades of

Evidence. Financial Management, e12505.

Li, W. C. Y. and B. H. Hall (2020, March). Depreciation of Business R&D Capital. Review of Income

and Wealth 66 (1), 161–180.

Li, Y. (2025b, April). Fragile New Economy: Intangible Capital, Corporate Savings Glut, and Financial

Instability. American Economic Review 115 (4), 1100–1141.

Liang, C. (2024). Advertising rivalry and discretionary disclosure. Journal of Accounting and Eco-

nomics 77 (1), 101611.

Mincer, J. A. and V. Zarnowitz (1969). The evaluation of economic forecasts. In J. A. Mincer and

V. Zarnowitz (Eds.), Economic Forecasts and Expectations: Analysis of Forecasting Behavior and

Performance, pp. 3–46. National Bureau of Economic Research.

Nickell, S. (1981). Biases in dynamic models with fixed effects. Econometrica : Journal of the Econo-

metric Society 49 (6), 1417–1426.

Park, H. (2022). An intangible-adjusted book-to-market ratio still predicts stock returns. Critical

Finance Review 11 (2), 265–297.

Peters, R. H. and L. A. Taylor (2017). Intangible capital and the investment-q relation. Journal of

Financial Economics 123 (2), 251–272.

Peters, S. J. and M. P. Winters (2025, February). Intangible Assets: Before Recognition, Improved

Disclosures and Disaggregation Are Needed. CFA Institute Research & Policy Center, Investor Per-

spectives Series.

Woeppel, M. (2022, December). Using Patent Capital to Estimate Tobin’s Q. Journal of Financial and

Quantitative Analysis 57 (8), 2929–2967.

Wu, K. and S. Lai (2020). Intangible intensity and stock price crash risk. Journal of Corporate

Finance 64, 101682.

33
Figure 1: Five Metrics of Intangibles over Book Values of Assets
The figure presents box plots for five different measures of intangible assets, each scaled by total assets.
BV is based on the Compustat intangible variable (intan), while P T BV , EKP BV , W BV , and
EP W BV correspond to the approaches proposed by Peters and Taylor (2017), Eisfeldt et al. (2022),
Woeppel (2022), and Ewens et al. (2025), respectively.

Figure 2: Five Metrics of Intangibles over Their Corresponding Total Assets

The figure presents box plots for five different measures of intangible assets, each scaled by the corre-
sponding definition of total assets. BV is based on the Compustat intangible variable (intan) divided
by book value of assets (AT ). The remaining measures, P T , EKP , W , and EP W , reflect intangible
assets estimates following the approaches of Peters and Taylor (2017), Eisfeldt et al. (2022), Woeppel
(2022), and Ewens et al. (2025), respectively, each scaled by their corresponding definitions of total
assets that incorporate OBS intangibles.

34
Figure 3: Violin Plots of Intangibles over Assets
The figure displays violin plots for five different measures of intangible assets, each scaled by total
assets. The BV measure is based on the Compustat intangible variable (intan), while P T , EKP , W ,
and EP W correspond to the approaches proposed by Peters and Taylor (2017), Eisfeldt et al. (2022),
Woeppel (2022), and Ewens et al. (2025), respectively.

Figure 4: Book Values (BV) and Incremental Off-Balance-Sheet (OBS) Intangibles

The figure presents box plots comparing the book value of intangibles (BV), based on the Compu-
stat intangible variable (intan), with incremental intangibles measures. Incremental intangibles are
defined as the difference between each OBS measure and BV , and are deflated by their respective to-
tal asset measure as specified in Equation (15). The incremental measures—IP T , IEKP , IW , and
IEP W —correspond to the approaches proposed by Peters and Taylor (2017), Eisfeldt et al. (2022),
Woeppel (2022), and Ewens et al. (2025), respectively. For example, IP T = (P T − BV )/AP T .

35
Figure 5: Histograms of Intangibles over Assets
This figure presents overlapping histograms comparing the distribution of intangible assets scaled by
total assets across different measurement approaches. The top panel compares the distributions of book
value (BV) and PT intangibles. The middle panel compares EKP and EPW, while the bottom panel
contrasts PT and W.
36
 Table 1: Summary Statistics for Assets
This table presents summary statistics for intangible and tangible assets in Panel A, and for total assets in
Panel B, over the period 1980-2022. Intangible assets (IN j ) and total assets (Aj ) are estimated using five
different approaches, denoted by j. The BV method uses the book values of assets (Compustat item at). The
PT approach includes OBS intangibles as measured by Peters and Taylor (2017). The EKP method incorporates
OBS intangibles following the method of Eisfeldt et al. (2022). The W approach is based on the estimates from
Woeppel (2022), and the EPW method follows Ewens et al. (2025). Tangible assets are measured as Net Property,
Plant, & Equipment (Compustat item ppent). Reported statistics include the mean (µ), median (p50), standard
deviation (σ), skewness (γ), and change in mean (∆µ), all expressed in millions of U.S. dollars. N denotes the
number of observations. t-tests compare the values of µ, ∆µ, and σ in column (1) against those in columns (2)
through (5). ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Summary Statistics of Intangible (INj ) and Tangible (NPPE ) Assets

IN BV IN P T IN EKP IN W IN EP W NPPE
(1) (2) (3) (4) (5) (6)
µ 2,463.73 5,356.98*** 8,482.97*** 8,763.36*** 4,653.05*** 2,962.92
∆µ 226.99 452.05*** 544.00*** 887.84*** 406.58*** 209.28
p50 299.16 1,449.94 2,725.71 894.10 1,170.55 711.60
σ 7,639.85 14,222.87*** 18,666.36*** 30,616.22*** 12,796.49*** 7,435.54
γ 7.02 6.62 4.95 7.00 6.70 6.46
Panel B: Summary Statistics of Total Assets (Aj )
ABV AP T AEKP AW AEP W
(1) (2) (3) (4) (5)
µ 10,035.07 12,778.33*** 18,100.12*** 16,251.66*** 12,096.12***
∆µ 756.12 962.61*** 1,026.86*** 1,391.39*** 914.91***
p50 3,041.67 4,063.01*** 6,426.14*** 3,744.59*** 3,815.02***
σ 23,966.93 30,241.77*** 37,716.11*** 46,604.52*** 28,717.36***
γ 6.53 6.41 5.21 67.86 6.45
N 27,213 27,183 19,324 27,213 26,967

37
 Table 2: Intangible Asset Ratios, Means and Standard Deviations

This table presents t-statistics for pairwise ratio tests comparing five different measures of intangible assets.
The measures include the book value of intangibles (INTAN) and four OBS intangible asset estimates based
on Peters and Taylor (2017) (PT), Eisfeldt et al. (2022) (EKP), Woeppel (2022) (W), and Ewens et al. (2025)
(EPW). Panel A reports t-statistics testing the equality of means between each pair of intangible measures
(e.g., H0 : µP T = µIN T AN ). Panel B shows the ratio of standard deviations (σrow /σcolumn ) for each pair,
along with tests of whether the ratio differs significantly from one (e.g., H0 : σP T /σIN T AN = 1). For instance,
σP T /σIN T AN = 2.61 and σEP W /σW = 0.16, and both ratios are statistically different from one at the 1%
significance level (***). Panel C reports t-statistics testing for equality in the change in means across pairs of
measures (e.g., H0 : ∆µP T = ∆µIN T AN ). ***, **, and * denote statistical significance at the 1%, 5%, and 10%
levels, respectively.

Panel A: t-tests for mean (µ)

INTAN PT EKP W
PT 50.78***
EKP 40.11*** 25.26***
W 32.91*** 23.47*** 11.82***
EPW 49.23*** 42.12*** 30.76*** 26.73***
Panel B: t-tests for standard deviation (σ)
INTAN PT EKP W
PT 2.61***
EKP 5.81*** 2.22***
W 13.12*** 5.02*** 2.25***
EPW 2.05*** 0.79*** 0.35*** 0.16***
Panel C: t-tests for change in mean (∆µ)
INTAN PT EKP W
PT 21.49***
EKP 14.14*** 0.30
W 20.90*** 14.70*** 12.59***
EPW 18.85*** 8.28*** 2.76*** 16.00***

38
 Table 3: Summary Statistics for Alternative Asset Deflators
This table presents summary statistics for various ratios constructed using five different asset measures as deflators.
The asset measures are all based on the book value of assets (Total Assets) but vary depending on their approach
to valuing intangible assets. BV uses the book values of intangibles. PT includes OBS intangibles measured
following Peters and Taylor (2017). EKP includes OBS intangibles measured following Eisfeldt et al. (2022). W
includes OBS intangibles measured following Woeppel (2022). EPW includes OBS intangibles measured following
Ewens et al. (2025). The table reports the mean (µ), median (p50), standard deviation (σ), skewness (γ), and
change in mean (∆µ) for ratios involving intangible assets, market capitalization, and cash flow. t-tests compare
the µ, ∆µ, and σ in the first column to each of the values in columns (2) through (5). N denotes the number of
observations. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Intangible Assetsj /AssetsBV

IN BV /ABV IN P T /ABV IN EKP /ABV IN W /ABV IN EP W /ABV
(1) (2) (3) (4) (5)
µ 0.178 0.519*** 0.864*** 0.540*** 0.434***
∆µ 0.006 0.007** −0.005*** 0.025*** 0.007***
p50 0.100 0.502 0.707 0.279 0.401
σ 0.200 0.332*** 0.740*** 0.821*** 0.295***
γ 1.184 0.665 1.872 4.166 0.660
N 24,427 24,404 18,385 24,427 24,197
Panel B: Intangible Assetsj /Assetsj
IN BV /ABV IN P T /AP T IN EKP /AEKP IN W /AW IN EP W /AEP W
(1) (2) (3) (4) (5)
µ 0.178 0.366*** 0.449*** 0.306*** 0.328***
∆µ 0.006 0.005*** 0.001*** 0.008*** 0.005***
p50 0.100 0.490 0.253 0.316 0.198
σ 0.200 0.197 0.226*** 0.257*** 0.197
γ 1.184 0.060 −0.417 0.487 0.312
N 24,427 24,404 18,385 24,427 24,197
Panel C: Log(Market Cap/Assetsj )
log(M/ABV ) log(M/AP T ) log(M/AEKP ) log(M/AW ) log(M/AEP W )
(1) (2) (3) (4) (5)
µ 0.239 −0.036*** −0.251*** 0.020*** 0.019***
∆µ −0.021 −0.022** −0.154*** −0.030*** −0.021**
p50 0.207 −0.060 −0.262 −0.029 −0.004
σ 0.908 0.872*** 0.915* 0.888 0.860**
γ 0.202 0.212 0.145 0.323 0.194
N 27,196 27,196 27,196 27,196 26,950
Panel D: Cash Flow/Assetsj
CF /ABV CF /AP T CF /AEKP CF /AW CF /AEP W
(1) (2) (3) (4) (5)
µ 0.115 0.087*** 0.072*** 0.095*** 0.093***
∆µ −0.005 −0.004*** −0.003*** −0.005 −0.004***
p50 0.106 0.084 0.069 0.083 0.089
σ 0.130 0.103*** 0.082*** 0.119*** 0.106***
γ −0.295 −0.186 −0.585 0.302 −0.092
N 18,147 18,147 18,147 18,147 17,997
39
 Table 4: Price Informativeness
This table presents regressions estimating one-year-ahead cash flow using the following specification:
   
CFi,t+1 Mi,t CFi,t
= α + β1 ln +ρ + γXi,t + ds 1s + ϵi,t+1 .
Aj,i,t Aj,i,t Aj,i,t

where Mi,t is market capitalization, CFi,t is cash flow (EBIT), and Aj,i,t denotes total assets measured using
alternative definitions indexed by j. The term ln(Mi,t /Aj,i,t ) is referred to as the log market-to-asset ratio
(LM A). Panel A includes industry fixed effects, Panel B adds year fixed effects, and Panel C incorporates the
lagged dependent variable (CFi,t /Aj,i,t ), and firm-level control variables including institutional ownership ratio
(IOR), idiosyncratic volatility (IV OL), industry concentration (HHI), and analyst coverage (COV ERAGE).
All regressions include robust standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and
10% levels, respectively.

Dep. Vars. CF /ABV CF /AP T CF /AEKP CF /AW CF /AEP W

(1) (2) (3) (4) (5)
Panel A: Model with industry fixed effects
LM A 0.389*** 0.394*** 0.465*** 0.333*** 0.388***
(0.052) (0.051) (0.044) (0.053) (0.050)
Panel B: Model with industry and year fixed effects
LM A 0.429*** 0.426*** 0.491*** 0.362*** 0.423***
(0.069) (0.067) (0.057) (0.067) (0.065)

Panel C: Full model with controls and fixed effects

LM A 0.468*** 0.442*** 0.514*** 0.383*** 0.448***
(0.122) (0.127) (0.115) (0.116) (0.123)
CFt /At 0.225** 0.223* 0.392** 0.195* 0.218*
(0.100) (0.115) (0.165) (0.101) (0.112)
IOR 0.128 0.116 −0.101 0.153 0.160
(0.210) (0.204) (0.225) (0.178) (0.207)
IV OL −2.004 −2.267 −2.006 −1.375 −2.201
(1.837) (1.964) (1.983) (1.733) (1.960)
HHI 0.153 0.145 −0.030 0.169 0.177
(0.210) (0.204) (0.221) (0.178) (0.207)
COV ERAGE 0.123** 0.102 0.111* 0.112** 0.112
(0.060) (0.069) (0.068) (0.056) (0.070)
N 13,701 13,701 13,701 13,701 13,698

40
 Table 5: Price Informativeness of CF/ABV with Intangible Intensity

This table tests price informativeness for CF/ABV with the following equation:
   
CFi,t+1 Mi,t Aj,i,t
= α + β1 ln + β2 ln + γZi,t + ϵi,t+1 ,
ABV,i,t ABV,i,t ABV,i,t
 
Aj,i,t
where LM ABV −LM Aj = ln ABV,i,t , and Zi,t includes the lagged dependent variable, firm-level controls, and
year and industry fixed effects are included in Zi,t . Panel A reports results for this baseline model. Panel B
extends the analysis by interacting the benchmark LM A with firm-level intangible intensity indicators. Firms
are classified into high (IH ) and low (IL ) intangible intensity groups based on the top and bottom 30% of the
distribution within each sector. The following model is estimated:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 .
ABV,i,t ABV,i,t ABV,i,t ABV,i,t ABV,i,t

All regressions include industry and year fixed effects, as well as the firm-level control variables. Robust standard
errors are used in all specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.

Panel A: Including LM ABV −LM Aj

Dep. Var. CF/ABV
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM ABV 0.468*** 0.446*** 0.366*** 0.468*** 0.446***
(0.122) (0.122) (0.119) (0.122) (0.122)
LM ABV −LM Aj 1.145*** 0.876*** 0.026 1.145***
(0.174) (0.097) (0.104) (0.174)
Panel B: Including interaction term with intangible intensity dummies
Dep. Var. CF/ABV
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM ABV 0.534*** 0.459*** 0.477*** 0.506*** 0.508***
(0.107) (0.108) (0.043) (0.102) (0.106)
LM ABV −LM Aj 1.735*** 1.798*** 0.595*** 2.003***
  (0.227) (0.156) (0.102) (0.251)
Mi,t
IH × ln ABV,i,t -0.129*** -0.065** 0.043 -0.063** -0.070**
  (0.030) (0.028) (0.038) (0.027) (0.029)
Mi,t
IL × ln ABV,i,t -0.020 0.066 0.126*** 0.017 0.026
(0.081) (0.071) (0.042) (0.106) (0.074)
F : βH = βL 2.42 2.96 0.59 2.04 2.56
0.1199 0.0853 0.4429 0.1537 0.0983
F : βH = βL = 0 23.37 9.17 5.75 8.14 8.96
0.0000 0.0102 0.0565 0.0171 0.0092

41
 Table 6: Price Informativeness of CF/Aj with Intangible Intensity

This table tests price informativeness for CF/Aj with the following equation:
   
CFi,t+1 Mi,t Aj,i,t
= α + β1 ln + β2 ln + γZi,t + ϵi,t+1 ,
Aj,i,t Aj,i,t ABV,i,t

where the lagged dependent variable, firm-level controls, and year and industry fixed effects are included in Zi,t
for brevity. Panel A reports results for this baseline model. Panel B extends the analysis by interacting the
benchmark LM A with firm-level intangible intensity indicators. Firms are classified into high (IH ) and low (IL )
intangible intensity groups based on the top and bottom 30% of the distribution within each sector. The following
model is estimated:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 .
Aj,i,t Aj,i,t ABV,i,t Aj,i,t Aj,i,t

All regressions include industry and year fixed effects, as well as the firm-level control variables. Robust standard
errors are used in all specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.

Panel A: Including LM ABV −LM Aj

Dep. Var. CF/ABV CF /AP T CF /AEKP CF /AW CF /AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.468*** 0.458*** 0.514*** 0.382*** 0.480***
(0.122) (0.130) (0.115) (0.125) (0.125)
LM ABV − LM Aj 0.842*** 0.480*** -0.003 1.064***
(0.188) (0.149) (0.103) (0.169)
Panel B: Including interaction term with intangible intensity dummies
Dep. Var. CF/ABV CF /AP T CF /AEKP CF /AW CF /AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.534*** 0.459*** 0.667*** 0.421*** 0.500***
(0.107) (0.116) (0.055) (0.105) (0.113)
LM ABV −LM Aj 1.036*** 0.980*** 0.025 1.247***
  (0.200) (0.136) (0.091) (0.237)
Mi,t
IH × ln Aj,i,t -0.129*** -0.095*** -0.023 -0.133*** -0.084***
  (0.030) (0.027) (0.036) (0.025) (0.028)
Mi,t
IL × ln Aj,i,t -0.020 0.106 0.199*** 0.066 0.053
(0.081) (0.071) (0.051) (0.109) (0.075)
F : β H = βL 2.42 9.33 14.92 3.72 4.26
0.1199 0.0023 0.0001 0.0538 0.0389
F : β H = βL = 0 23.37 21.86 16.86 34.74 14.17
0.0000 0.0000 0.0002 0.0000 0.0008

42
 Table 7: Price Informativeness of Cash-Flows deflated by Sales

This table tests price informativeness for Cash-Flow divided by Sales with the following equation:
   
CFi,t+1 Mi,t Aj,i,t
= α + β1 ln + β2 ln + γZi,t + ϵi,t+1 ,
SLj,i,t Aj,i,t ABV,i,t

where SL denotes Sales, the lagged dependent variable, firm-level controls, and year and industry fixed effects
are included in Zi,t for brevity. Panel A reports results for this baseline model. Panel B extends the analysis
by interacting the benchmark LM A with firm-level intangible intensity indicators. Firms are classified into high
(IH ) and low (IL ) intangible intensity groups based on the top and bottom 30% of the distribution within each
sector. The following model is estimated:
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 .
SLi,t Aj,i,t ABV,i,t Aj,i,t Aj,i,t

All regressions include industry and year fixed effects, as well as the firm-level control variables. Robust standard
errors are used in all specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.

Panel A: Including LM ABV −LM Aj

Dep. Var. CF/SL CF /SL CF /SL CF /SL CF /SL
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.264*** 0.252*** 0.241*** 0.268*** 0.256***
(0.050) (0.052) (0.055) (0.052) (0.049)
LM ABV − LM Aj 0.609*** 0.339*** 0.194*** 0.647***
(0.131) (0.071) (0.066) (0.161)
Panel B: Including interaction term with intangible intensity dummies
Dep. Var. CF/SL CF /SL CF /SL CF /SL CF /SL
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.283*** 0.230*** 0.250*** 0.258*** 0.232***
(0.048) (0.053) (0.051) (0.053) (0.047)
LM ABV −LM Aj 0.705*** 0.357*** 0.229*** 0.724***
  (0.158) (0.082) (0.075) (0.181)
Mi,t
IH × ln Aj,i,t -0.075*** -0.062*** -0.031 -0.034* -0.047***
  (0.020) (0.020) (0.022) (0.019) (0.018)
Mi,t
IL × ln Aj,i,t 0.004 0.091** 0.120** 0.063 0.093**
(0.042) (0.039) (0.048) (0.050) (0.043)
F : βH = βL 2.5 9.5 6.46 2.85 7.46
0.1138 0.0021 0.011 0.0913 0.0063
F : βH = βL = 0 15.24 11.15 6.6 3.94 9.03
0.0005 0.0038 0.0369 0.1398 0.011

43
 Table 8: Forecast Encompassing Test of CF/ABV

This table presents results for the one-year-ahead forecast encompassing tests of cash flows, CF/ABV . Panel A
contains Fair-Shiller regressions. The last column includes the principal component forecast, constructed from
the four OBS intangibles. Panel B reports the Chong-Hendry tests, which regress the forecast residual (eBV )
on the differences between competing forecasts (CF d BV,j − CF
d BV,BV ). All regressions include industry and year
fixed effects, as well as the firm-level control variables. Robust standard errors are used in all specifications. ***,
**, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Fair-Shiller encompassing test

Dep. Var. CF/ABV
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W LM AP C
(1) (2) (3) (4) (5) (6)
ŷBV 1.024*** 0.520*** 1.189*** 0.599*** 0.450*** 0.159
(0.012) (0.141) (0.057) (0.104) (0.145) (0.170)
ŷj 0.567*** -0.080* 0.490*** 0.633***
(0.167) (0.043) (0.144) (0.158)
P C1 0.265***
(0.047)
P C2 -0.370***
(0.048)
P C3 -0.093
(0.118)
R2 0.4187 0.4579 0.429 0.4582 0.4558 0.4533
Panel B: Chong-Hendry encompassing test
Dep. Var. eBV
LM Aj Metric LM AP T LM AEKP LM AW LM AEP W LM AP C
(1) (2) (3) (4) (5)
d BV,j − CF
CF d BV,BV -0.407*** 0.037 -0.407*** -0.469***
(0.134) (0.044) (0.101) (0.151)
P C1 0.028***
(0.010)
P C2 0.070***
(0.023)
P C3 -0.004
(0.033)
R2 0.0704 0.0692 0.0731 0.0683 0.0894

44
 Table 9: Forecast Encompassing Test of CF/Aj

This table presents the one-year ahead forecast encompassing tests for CF/Aj . Unlike the previous table, the
dependent variable is cash flow deflated by the alternative j measures of assets. Panel A compares the forecasts
based on LM Aj and LM ABV . Panel B then reports results using principal components of the forecasts, excluding
information from that asset. Panel C performs the Chong-Hendry test. All regressions include robust standard
errors, industry and year fixed effects, and firm-level control variables. Robust standard errors are used in all
specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Fair-Shiller encompassing test

Dep. Var. CF /AP T CF /AEKP CF /AW CF /AEP W
(1) (2) (3) (4)
ŷj 1.123*** 0.085 0.664*** 0.890***
(0.142) (0.057) (0.225) (0.248)
ŷBV -0.063 0.910*** 0.430** 0.169
(0.097) (0.067) (0.171) (0.199)
R2 0.4083 0.3171 0.5027 0.4166
Panel B: Fair-Shiller encompassing test using PC
Dep. Var. CF /AP T CF /AEKP CF /AW CF /AEP W
(1) (2) (3) (4)
ŷj 0.837*** 0.190*** 1.049*** 1.740***
(0.153) (0.057) (0.144) (0.282)
P Cj1 0.072* 0.223*** -0.007 -0.121**
(0.081) (0.016) (0.028) (0.058)
P Cj2 -0.181* -0.181* -0.206*** 0.161***
(0.081) (0.078) (0.078) (0.031)
R2 0.399 0.3271 0.4892 0.506
Panel C: Chong-Hendry encompassing test
Dep. Var. eP T eEKP eW eEP W
(1) (2) (3) (4)
d BV,BV − CF
CF d BV,j 0.047 -0.919*** -0.334 -0.189
(0.115) (0.064) (0.227) (0.184)
R2 0.0594 0.2219 0.0354 0.0597

45
 Table 10: Price Informativeness with Physical Capital Intensity

This table presents estimates for the following model of one-year-ahead forecasts of CF/Aj :
       
CFi,t+1 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βHK IHK × ln + βLK ILK × ln + γZi,t + ϵi,t+1 ,
Aj,i,t Aj,i,t ABV,i,t Aj,i,t Aj,i,t

where IHK and ILK are indicator variables denoting firms with high (top 30%) and low (bottom 30%) relative
physical capital intensity. Intensity is calculated as the difference between physical capital (NPPE ) and intangible
assets (INT ), scaled by total assets Aj . All regressions include industry and year fixed effects, as well as the
firm-level control variables. Robust standard errors are used in all specifications. ***, **, and * denote statistical
significance at the 1%, 5%, and 10% levels, respectively.

Dep. Var. CF/ABV CF /AP T CF /AEKP CF /AW CF /AEP W

LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.450*** 0.479*** 0.655*** 0.404*** 0.511***
(0.154) (0.111) (0.059) (0.110) (0.099)
LM ABV − LM Aj 0.996*** 0.957*** 0.021 1.205***
  (0.192) (0.138) (0.085) (0.231)
Mi,t
IHK × ln Aj,i,t 0.192*** 0.056 0.210*** 0.076 0.016
  (0.073) (0.085) (0.057) (0.091) (0.112)
Mi,t
ILK × ln Aj,i,t -0.065** -0.102*** -0.006 -0.098*** -0.088**
(0.031) (0.032) (0.036) (0.027) (0.035)
F : βHK = βLK 18.53 4.35 12.37 3.26 1.21
0.0000 0.037 0.0004 0.0711 0.2719
F : βHK = βLK = 0 23.96 15.32 14.48 14.63 11.27
0.0000 0.0005 0.0007 0.0007 0.0036

46
 Table 11: Price Informativeness of R&D, Patents, and Returns

This table reports regression estimates of three forward-looking outcomes using alternative asset definitions (Aj,i,t )
as scaling variables. Panel A presents regressions of one-year-ahead R&D expenditures scaled by assets, specified
R&Di,t+1 ∆P ATi,t+1
as Aj,i,t . Panel B displays regressions for the one-year-ahead change in patent, scaled similarly as Aj,i,t .
Panel C reports regressions of one-year-ahead stock returns, measured as ln(RETi,t+1 ). All panels employ the
following regression specification:
       
Mi,t Aj,i,t Mi,t Mi,t
Y = α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
Aj,i,t ABV,i,t Aj,i,t Aj,i,t

where Y corresponds to the outcome variable of interest in each respective panel. All regressions include industry
and year fixed effects, as well as the firm-level control variables. Robust standard errors are used in all specifica-
tions. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Forecasts of R&D/Aj

Dep. Var R&D/ABV R&D/AP T R&D/AEKP R&D/AW R&D/AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.396*** 0.406*** 0.553*** 0.299*** 0.432***
(0.053) (0.050) (0.077) (0.041) (0.052)
LM ABV − LM Aj 1.310*** -0.012 0.080 1.166***
  (0.227) (0.124) (0.081) (0.352)
Mi,t
IH × ln Aj,i,t -0.105*** -0.032 -0.033 -0.031 -0.023
  (0.031) (0.030) (0.032) (0.022) (0.026)
Mi,t
IL × ln Aj,i,t 0.162*** 0.029 0.085* 0.225*** 0.089
(0.054) (0.052) (0.051) (0.073) (0.057)
F : βH = βL 16.56 1.19 3.79 9.63 3.08
0.0001 0.2758 0.0515 0.0019 0.0793
F : βH = βL = 0 18.18 1.67 3.8 9.7 3.1
0.0001 0.4334 0.1499 0.0078 0.2124
Panel B: Forecasts of ∆P AT /Aj
Dep. Var ∆P AT /ABV ∆P AT /AP T ∆P AT /AEKP ∆P AT /AW ∆P AT /AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.446*** 0.386*** 0.374*** 0.408*** 0.415***
(0.040) (0.037) (0.040) (0.041) (0.039)
LM ABV − LM Aj 1.438*** 0.391*** 0.878*** 1.973***
  (0.110) (0.071) (0.067) (0.169)
Mi,t
IH × ln Aj,i,t -0.092*** -0.056*** -0.026 0.015 -0.068***
  (0.022) (0.021) (0.021) (0.029) (0.022)
Mi,t
IL × ln Aj,i,t 0.018 0.032 -0.041* -0.097*** 0.040
(0.033) (0.026) (0.024) (0.037) (0.029)
F : βH = βL 8.05 6.33 0.28 6.52 8.63
0.0046 0.0119 0.5965 0.0107 0.0033
F : βH = βL = 0 17.3 8.12 3.76 7.71 11.27
0.0002 0.0173 0.1523 0.0212 0.0036

47
Table 11 continued
Panel C: Forecasts of ln (RET )
Dep. Var ln (RET )
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj -0.644*** -0.701*** -0.786*** -0.621*** -0.659***
(0.049) (0.051) (0.062) (0.051) (0.051)
LM ABV − LM Aj 0.915*** 0.398*** -1.118*** 1.321***
  (0.210) (0.120) (0.123) (0.270)
Mi,t
IH × ln Aj,i,t -0.050 -0.077 0.062 -0.106** -0.053
  (0.051) (0.051) (0.054) (0.054) (0.051)
Mi,t
IL × ln Aj,i,t 0.032 0.112** 0.042 0.150*** 0.087
(0.061) (0.050) (0.055) (0.056) (0.055)
F : βH = βL 0.92 6.15 0.08 9.57 3.13
0.3378 0.0132 0.7822 0.002 0.0767
F : βH = βL = 0 1.11 6.51 1.78 9.8 3.26
0.5745 0.0385 0.4116 0.0075 0.1961

48
 Table 12: Forecasting Encompassing Tests: R&D, Patents, and Returns

This table reports the one-year-ahead forecast encompassing tests for R&D (Panel A), Patents (Panel B), and
Returns (Panel C) using Fair-Shiller regressions. The last column includes the principal component forecast,
constructed from the four OBS intangibles. All regressions include industry and year fixed effects, as well as the
firm-level control variables. Robust standard errors are used in all specifications. ***, **, and * denote statistical
significance at the 1%, 5%, and 10% levels, respectively.

Panel A: R&D/ABV
Dep. Var. R&D/ABV
(1) (2) (3) (4) (5)
ŷBV 0.372*** 0.608*** 0.151* 0.426*** 0.671***
(0.121) (0.121) (0.088) (0.123) (0.126)
ŷj 0.567*** 0.388*** 0.873*** 0.495***
(0.127) (0.127) (0.102) (0.124)
P C1 0.166***
(0.054)
P C2 0.197
(0.168)
P C3 1.043***
(0.168)
R2 0.7126 0.7286 0.7281 0.7109 0.7434
Panel B: ∆P AT /ABV
Dep. Var. ∆P AT /ABV
(1) (2) (3) (4) (5)
ŷBV 0.453*** 0.564*** 0.876*** 0.321* 0.575**
(0.155) (0.082) (0.040) (0.178) (0.245)
ŷj 0.524*** 0.424*** 0.141*** 0.652***
(0.160) (0.084) (0.051) (0.183)
P C1 0.249
(0.155)
P C2 0.042
(0.072)
P C3 0.863***
(0.234)
R2 0.7439 0.7554 0.7313 0.7463 0.7308
Panel C: ln (RET )
Dep. Var. ln (RET )
(1) (2) (3) (4) (5)
ŷBV -0.040 -0.011 0.159*** -0.063** -0.115
(0.029) (0.026) (0.046) (0.029) (0.127)
ŷj 0.166*** 0.153*** -0.040 0.187***
(0.029) (0.026) (0.043) (0.029)
P C1 0.038***
(0.003)
P C2 -0.044***
(0.014)
P C3 0.045
(0.039)
R2 0.2851 0.3026 0.2841 0.2856 0.3033

49
 Table 13: Robustness: Price informativeness of CF with Incremental LM A

This table examines the price informativeness of incremental intangible assets using two model specifications.
 
Mi,t
Panel A estimates the following specification, and directly tests the LMA of incremental assets, ln AIJ,i,t :
       
CFi,t+1 Mi,t Mi,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
ABV,i,t ABV,i,t AIJ,i,t ABV,i,t ABV,i,t

where IJ denotes the incremental information defined in Equation (15). Panel B uses analternative
 specification,
AIJ,i,t
where it tests the difference in LMA between BV and incremental OBS intangibles, ln ABV,i,t :
       
CFi,t+1 Mi,t AIJ,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
ABV,i,t ABV,i,t ABV,i,t ABV,i,t ABV,i,t

where IJ represents the incremental information in Equation (15). Panel C estimates the first specification All
regressions include industry and year fixed effects, as well as the firm-level control variables. Robust standard
errors are used in all specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.

Panel A: Forecasts of CF/ABV

Dep. Var. CF/ABV
LM AIJ Metric LM AIP T LM AIEKP LM AIW LM AIEP W LM AIP C
(1) (2) (3) (4) (5)
LM ABV 0.549*** 0.687*** 0.627*** 0.545*** 0.662***
(0.111) (0.052) (0.044) (0.109) (0.057)
LM AIJ -0.140** -0.200*** -0.061*** -0.128**
  (0.065) (0.029) (0.020) (0.057)
Mi,t
IH × ln ABV,i,t -0.120*** -0.066** -0.069** -0.120*** -0.033
  (0.032) (0.029) (0.032) (0.032) (0.032)
Mi,t
IL × ln ABV,i,t -0.018 0.073 0.035 -0.018 0.057
(0.089) (0.049) (0.046) (0.089) (0.053)
F : βH = βL 1.84 5.50 3.08 1.85 1.78
0.1746 0.019 0.0792 0.1742 0.1827
F : βH = βL = 0 19.46 6.92 4.91 19.47 1.83
0.0001 0.0314 0.0858 0.0001 0.4005
IP C1 -0.267***
(0.033)
IP C2 0.216***
(0.064)
IP C3 0.187**
(0.074)

50
Table 13 continued
Panel B: Forecasts of CF/ABV
Dep. Var. CF/ABV
LM AIJ Metric LM AIP T LM AIEKP LM AIW LM AIEP W LM AIP C
(1) (2) (3) (4) (5)
LM ABV 0.447*** 0.566*** 0.531*** 0.448*** 0.539***
(0.103) (0.049) (0.107) (0.101) (0.049)
LM ABV − LM AIJ 2.641*** 1.840*** 0.409 2.718***
  (0.286) (0.225) (0.299) (0.365)
Mi,t
IH × ln ABV,i,t -0.118*** -0.099*** -0.124*** -0.114*** -0.090***
  (0.031) (0.028) (0.031) (0.031) (0.028)
Mi,t
IL × ln ABV,i,t -0.007 0.089* -0.028 -0.009 0.094*
(0.084) (0.052) (0.085) (0.084) (0.052)
P CJ1 0.497***
(0.065)
P CJ2 -0.275***
(0.068)
P CJ3 0.013
(0.042)
F : βH = βL 2.32 10.69 1.74 2.08 9.90
0.1276 0.0011 0.1876 0.1491 0.0017
F : βH = βL = 0 20 16.24 21.24 18.74 14.23
0.0000 0.0003 0.0000 0.0001 0.0008
Panel C: Forecasts of CF/Aj
Dep. Var. CF/AP T CF/AEKP CF/AW CF/AEP W
LM AIJ Metric LM AIP T LM AIEKP LM AIW LM AIEP W
(1) (2) (3) (4)
LM AIJ 0.654*** 0.649*** 0.559*** 0.662***
  (0.063) (0.067) (0.053) (0.057)
Mi,t
IH × ln ABV,i,t -0.079** -0.046 -0.095*** -0.033
  (0.032) (0.033) (0.031) (0.032)
Mi,t
IL × ln ABV,i,t 0.142*** 0.180*** 0.180*** 0.057
(0.045) (0.060) (0.044) (0.053)
P CIJ1 -0.286*** -0.250*** -0.230*** -0.267***
(0.034) (0.037) (0.028) (0.033)
P CIJ2 0.136** 0.139** 0.162*** 0.216***
(0.057) (0.070) (0.050) (0.064)
P CIJ3 0.136** 0.139** 0.162*** 0.216***
(0.057) (0.070) (0.050) (0.064)

51
 Table 14: Robustness: Recent Time Period

This table presents regression estimates for four forward-looking firm outcomes, using alternative asset definitions
(Aj,i,t ) as scaling variables, over the sample period from 2002 to 2022. Panel A reports estimates for one-year-
CFi,t+1
ahead cash flow scaled by assets, specified as Aj,i,t . Panel B presents regressions of one-year-ahead R&D
R&Di,t+1
expenditures scaled by assets, specified as Aj,i,t . Panel C displays regressions for the one-year-ahead change
∆P ATi,t+1
in patent, scaled similarly as Aj,i,t .
Panel D reports regressions of one-year-ahead stock returns, measured
as ln(RETi,t+1 ). All panels employ the following regression specification:
       
Mi,t Aj,i,t Mi,t Mi,t
Y = α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
Aj,i,t ABV,i,t Aj,i,t Aj,i,t

where Y corresponds to the outcome variable of interest in each respective panel. All regressions include industry
and year fixed effects, as well as the firm-level control variables. Robust standard errors are used in all specifica-
tions. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Price Informativeness of CF 2002–2022

Dep. Var. CF/ABV CF /AP T CF /AEKP CF /AW CF /AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.766*** 0.767*** 0.784*** 0.712*** 0.760***
(0.059) (0.066) (0.071) (0.063) (0.063)
LM ABV −LM Aj 1.665*** 1.019*** 0.323*** 1.872***
  (0.274) (0.158) (0.092) (0.309)
Mi,t
IH × ln Aj,i,t -0.078** -0.104*** -0.008 -0.118*** -0.066*
  (0.031) (0.035) (0.042) (0.039) (0.035)
Mi,t
IL × ln Aj,i,t 0.110* 0.219*** 0.208*** 0.177*** 0.197***
(0.064) (0.055) (0.060) (0.053) (0.054)
F : βH = βL 6.93 25.28 10.09 24.26 19.22
0.0085 0.0000 0.0015 0.0000 0.0000
F : βH = βL = 0 9.32 25.52 12.52 24.82 19.24
0.0095 0.0000 0.0019 0.0000 0.0001
Panel B: Price Informativeness of R&D 2002–2022
Dep. Var. R&D/ABV R&D/AP T R&D/AEKP R&D/AW R&D/AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.469*** 0.479*** 0.604*** 0.307*** 0.472***
(0.066) (0.077) (0.096) (0.046) (0.064)
LM ABV −LM Aj 2.825*** 0.048 0.061 1.535***
  (0.353) (0.160) (0.087) (0.530)
Mi,t
IH × ln Aj,i,t -0.086*** 0.011 -0.028 -0.031 0.017
  (0.030) (0.044) (0.039) (0.026) (0.030)
Mi,t
IL × ln Aj,i,t 0.138*** -0.057 0.080 0.075* 0.032
(0.050) (0.066) (0.055) (0.039) (0.046)
F : βH = βL 19.11 1.58 2.34 6.63 0.12
0 0.2809 0.1265 0.01 0.7339
F : βH = βL = 0 21.39 1.6 2.44 6.65 0.6
0 0.5494 0.2953 0.0359 0.7423
52
Table 14 continued
Panel C: Price Informativeness of Patents 2002–2022
Dep. Var. ∆P AT /ABV ∆P AT /AP T ∆P AT /AEKP ∆P AT /AW ∆P AT /AEP W
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.517*** 0.409*** 0.438*** 0.340*** 0.446***
(0.046) (0.037) (0.047) (0.023) (0.039)
LM ABV −LM Aj 1.565*** 0.408*** 0.682*** 2.092***
  (0.146) (0.086) (0.070) (0.221)
Mi,t
IH × ln Aj,i,t -0.097*** -0.041* -0.011 -0.003 -0.061***
  (0.026) (0.022) (0.021) (0.018)
Mi,t
IL × ln Aj,i,t -0.013 0.014 -0.077*** -0.070*** 0.011
(0.037) (0.020) (0.025) (0.024) (0.027)
F : βH = βL 3.98 3.77 5.41 6.09 4.74
0.0461 0.0521 0.02 0.0136 0.0294
F : βH = βL = 0 13.61 4.32 9.23 8.76 9.02
0.0011 0.1155 0.0099 0.0125 0.011
Panel D: Price Informativeness of RET 2002–2022
Dep. Var. ln (RET )
LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj -0.491*** -0.493*** -0.515*** -0.502*** -0.483***
(0.034) (0.037) (0.036) (0.035) (0.033)
LM ABV −LM Aj 0.035 -0.177** -0.359*** 0.491***
  (0.103) (0.070) (0.068) (0.169)
Mi,t
IH × ln Aj,i,t 0.023 -0.009 0.045* 0.006 0.009
  (0.028) (0.026) (0.027) (0.028)
Mi,t
IL × ln Aj,i,t 0.058 0.033 -0.012 0.093** 0.037
(0.038) (0.032) (0.031) (0.038) (0.032)
F : βH = βL 1.53 2.31 3.73 0.51 1.25
0.2157 0.1285 0.0536 0.4759 0.2631
F : βH = βL = 0 1.77 3.29 6.12 1.32 1.36
0.4123 0.1926 0.047 0.5157 0.5065

53
 Table 15: Robustness to Alternative Horizon, CF 3 /Aj

This table presents regressions estimating three-year-ahead cash flow from 2002–2022. The model specification
is:
       
CFi,t+3 Mi,t Aj,i,t Mi,t Mi,t
= α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
Aj,i,t Aj,i,t ABV,i,t Aj,i,t Aj,i,t

where Mi,t is market capitalization, CFi,t is cash flow (EBIT), and Aj,i,t denotes total assets measured
using alternative definitions indexed by j. The term ln(Mi,t /Aj,i,t ) is referred to as the log market-
to-asset ratio (LM A). All regressions include industry and year fixed effects, as well as the firm-level
control variables. Robust standard errors are used in all specifications. ***, **, and * denote statistical
significance at the 1%, 5%, and 10% levels, respectively.

Dep. Var. CF/ABV CF /AP T CF /AEKP CF /AW CF /AEP W

LM Aj Metric LM ABV LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4) (5)
LM Aj 0.269*** 0.207*** 0.240*** 0.146*** 0.191***
(0.052) (0.048) (0.061) (0.050) (0.051)
LM ABV −LM Aj 1.005*** 0.521*** 0.047 1.230***
  (0.194) (0.136) (0.066) (0.240)
Mi,t
IH × ln Aj,i,t -0.065 0.044* -0.007 0.091*** 0.108***
  (0.046) (0.027) (0.039) (0.028) (0.030)
Mi,t
IL × ln Aj,i,t 0.171*** 0.006 -0.022 0.079 0.065
(0.047) (0.039) (0.040) (0.051) (0.041)
F : βH = βL 0.95 0.07 0.07 1.14 10.85
0.3308 0.7847 0.7949 0.2851 0.001
F : βH = βL = 0 2.89 0.33 0.49 12.98 13.95
0.2352 0.8500 0.0053 0.0015 0.0009

54
Appendix 1 Variable Definitions

Variable Description
TOTAL ASSETS Aj,i,t is total assets using approach j for firm i at time t. j ∈ {BV, PT, EKP, W, EPW}.
ABV,i,t = at, where Compustat item at is total assets on the balance sheet.
AP T,i,t = at + KnowledgeCapitalP T + OrganizationalCapitalP T , Peters and Taylor (2017).
AEKP,i,t = at + IN T − gdwl, Eisfeldt et al. (2022).
AW,i,t = at + P atentCapital, Woeppel (2022).
AEP W,i,t = at + KnowledgeCapitalEP W + OrganizationalCapitalEP W , Ewens et al. (2025).
LMA The log of market capitalization (Mi,t ) for firm i at time t, divided by total assets (Aj,i,t ), where
approach j is used to incorporate OBS intangible assets into total assets. Bai et al. (2016) calculate
LMA using the market capitalization measured on March 31 and Compustat item at for total assets.
The regression coefficient on LMA serves as the primary measure of price informativeness (PI).
CASH FLOW CFi,t+1 is next year’s firm i cash flow measured as EBIT following Bai et al. (2016). Compustat
item ebit equals Net Sales (sale) minus Cost of Goods Sold (cogs) minus Selling, General &
Administrative Expense (xsga) minus Depreciation/Amortization (dp).
SG&A Compustat variable xsga is the selling, general, and administrative expense. This generally includes
all operating expenses not directly assigned to COGS. While Compustat has a separate variable
for R&D and most companies report this expense separately, xsga includes R&D expense as well
as marketing, advertising, employee training and IT.
R&D Compustat item xrd is the research and development expense.
RET The return is calculated over the one-year period starting in April using CRSP data.
NPPE Compustat item ppent is Net Property, Plant and Equipment. It represents the cost of tangible
fixed assets used to generate revenue (ppegt) minus Accumulated Depreciation and Amortization
(dpact)
INTAN Compustat item intan is the value of intangible assets on the balance sheet. We refer to this as
the book value (BV) of intangible assets. Intan is the sum of Compustat items ggdwl and intano.
GDWL Compustat item gdwl reflects the excess cost of equity of an acquired company. Under Accounting
Principles Board (APB) Opinion No. 17, companies amortized goodwill on a straight-line basis
over a period not exceeding 40 years. This approach was superseded in 2001 when FASB issued
SFAS 141 and SFAS 142. SFAS 141 eliminated the pooling of interests method of accounting for
business combinations. SFAS 142, which applied to mergers and acquisitions completed after June
30, 2001, requires goodwill to be capitalized and subject to an annual impairment test.
INTANO Compustat item intano represents all intangible assets other than goodwill. These include assets
such as copyrights, patent costs, client lists, subscription lists, technology know-how intellectual
property, and trademarks/tradenames.
TURNOVER Share turnover is measured as the natural log of the number of shares traded divided by the number
of shares outstanding. Calculated over the 12-month period corresponding to the fiscal year.
IVOL Idiosyncratic volatility is the standard deviation of daily abnormal returns over the 252 calendar
days before (and including) March 31. At least 21 return observations (one-month trading days)
over that 252-day period for a permno are required to calculate its stock return volatility. This
approach follows Bai et al. (2016).
IOR Institutional ownership ratio calculated using 13-F filings.
HHI Institutional ownership concentration - Herfindahl-Hirschman Index.
COVERAGE The natural log of 1 plus the number of equity analysts covering the stock.

55
Appendix 2 Intangible Intensity: Best Buy & Ross Stores

Note. The plots show the intangible intensity for Best Buy and Ross Stores from 2001 to 2023. Intangible
intensity is the ratio of intangible assets to total assets. Intangible assets are measured using the
following approaches: BV, PT, EKP, W, and EPW. The intangible intensity measured using BV and
W are the same for Best Buy during this time period.

56
Appendix 3 Relative Intangible Intensity: Best Buy & Ross Stores

Note. The plot shows the difference in intangible intensity for Best Buy and Ross Stores from 2001
to 2023. Best Buy and Ross Stores have distinct business models and competitive positioning in the
retail industry. This results in fundamentally different strategies with respect to intangible capital
investments.

Best Buy: Significant Investments in Brand and Employee Capital

• Geek Squad: distinctive service brand relying on employee expertise,

• Advertising: high spending to promote brand awareness and customer retention,

• Omnichannel Experiences: e-commerce and mobile apps.

Ross Stores: Operational Efficiency with Low Expenses

• Organic Growth: minimal acquisitions and word-of-mouth marketing,

• Advertising: low spending relative to other retailers,

• In-Store Experiences: no e-commerce platform.

The approach used to measure intangible capital significantly affects a comparison of these companies
over this time period. From the perspective of intangible assets on the balance sheet (BV) and patent
capital (W), there appears to be little change (2%) in their relative amounts of intangible capital.
However, measures that include knowledge and organizational capital (PT, EKP, and EPW) reveal that
Best Buy accumulated substantially more intangible capital. Over this period, the relative increase in
intangible intensity amounts to between 27% and 35% of total assets using these methods.

57
Appendix 4 Intangible Intensity: Astronics & Oshkosh

Note. The plot shows the difference in intangible intensity for Astronics (ATRO) and Oshkosh (OSK)
from 2001 to 2023. Intangible intensity is the ratio of intangible assets to total assets. Intangible assets
are measured using the following approaches: BV, PT, EKP, W, and EPW.

58
Appendix 5 Relative Intangible Intensity: Astronics & Oshkosh

Note. The plot shows the difference in intangible intensity for Astronics and Oshkosh from 2001 to
2023. Astronics and Oshkosh are industrial firms in the Capital Goods Industry, and both companies
have significant sales to government clients. However, their business operations use different levels of
intangible capital.

Astronics: Building Organizational Capital to Create Sticky Customer Relationships

• Highly engineered electronic systems and components,

• Skilled employees working closely with buyers to provide customized solutions,

• Builds capabilities internally or through smaller, strategic acquisitions.

Oshkosh: Portfolio of Leading Industrial Brands

• Leading market share in each of its businesses,

• Extensive dealer network driving sales,

• History of large-scale, transformative acquisitions.

From 2014–2023, the companies had similar intangible intensity when measured using the book values.
The measures including OBS intangible capital all indicate larger differences between the two firms,
and these are consistent with the differences in the business models.

59
Appendix 6 A q-theory framework

In this section, we present the economic rationale for our price informativeness measure, which

adapts the framework of Bai et al. (2016) to test the paper’s main predictions. In q-theory, a firm with

investment k in new capital and productivity shock z has an ex post fundamental value of

γ 2
v(z, k) = (1 + z)(k̄ + k) − k − k ,
2k̄

where k̄ is the firm’s current assets and γ is a parameter that governs adjustment costs. Assuming that

the discount rate is normalized to zero, the q-theory investment equation becomes

k∗
γ = E[z|Im ],
k̄

where Im is the information set of the firm’s manager and k ∗ is the value-maximizing level of capital,

that is k ∗ = argmaxk E[v(z, k)|Im ]. Therefore, the ex post maximized value of the firm is

v(z, k ∗ ) z 1
= 1 + z + E[z|Im ] − E[z|Im ]2 .
k̄ γ 2γ

The expected firm value conditional on the manager’s information set is given by

v(z, k ∗ )
 
1
E |Im = 1 + E[z|Im ] + E[z|Im ]2 .
k̄ 2γ

Assume that z has a mean of zero. Then, the aggregate efficiency is defined as the ex ante firm value

k̄
E[v(z, k ∗ )] = k̄ + V ar(E[z|Im ]).
2γ

As reflected in the above equation, the aggregate welfare depends on the value of current assets k̄ and

on the variance of the forecastable productivity under the manager’s information set.

Regarding the information environment, a firm’s manager has information about z produced from

the firm itself (inside information), which is denoted by

η = z + εη , εη ∼ N (0, ση2 ).

60
Similarly, informed traders also have information about z, denoted by

s = z + εs , εs ∼ N (0, σs2 ).

Bai et al. (2016) also assumes that traders observe an additional signal from the manager, denoted by

η ′ = η + ε′η , ε′η ∼ N (0, ση2′ ).

Firm managers use their information set to determine k, so their optimal decision is

k̄
k∗ = E[v(z, k ∗ )|η, η ′ , p].
γ

The stock market aggregates these signals into the stock price p. In equilibrium, the stock price depends

on the firm’s fundamental value conditional on the informed trader’s information set, IT = {η ′ , s}, that

is

p = αE[v(z, k ∗ )|η ′ , s] + βu,

where u is noise trading demand, and α and β are endogenous coefficients in the rational expectation

equilibrium. Bai et al. (2016) defines forecasting price efficiency (FPE) as the variance of the predictable

component of firm value v/k̄ given q, where q = p/k̄ is the firm’s market-to-book ratio. That is

VF P E = V ar(E[z|q]).

Similar to Bai et al. (2016), our analysis focuses on FPE, which measures the total amount of information

about future payoffs contained in market prices.

61
Appendix 7 Additional Tables

Table A.1: Price Informativeness of R&D and Patents

This table reports regression estimates of two forward-looking outcomes using book values of assets (ABV,i,t ) as
scaling variables. Panel A presents regressions of one-year-ahead R&D expenditures scaled by assets, specified as
R&Di,t+1 ∆P ATi,t+1
ABV,i,t and Panel B displays regressions for the one-year-ahead change in patent, scaled similarly as ABV,i,t .
Both panels employ the following regression specification:
       
Mi,t Aj,i,t Mi,t Mi,t
Y = α + β1 ln + β2 ln + βH IH × ln + βL IL × ln + γZi,t + ϵi,t+1 ,
Aj,i,t ABV,i,t Aj,i,t Aj,i,t

where Y corresponds to the outcome variable of interest in each respective panel. All regressions include industry
and year fixed effects, as well as the firm-level control variables. Robust standard errors are used in all specifica-
tions. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

Panel A: Forecasts of R&D/ABV

Dep. Var R&D/ABV
LM Aj Metric LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4)
LM ABV 0.380*** 0.351*** 0.382*** 0.409***
(0.053) (0.052) (0.053) (0.053)
LM ABV − LM Aj 1.265*** 0.391*** 0.528*** 1.006***
  (0.255) (0.115) (0.128) (0.335)
Mi,t
IH × ln Aj,i,t -0.110*** -0.106*** -0.103*** -0.107***
  (0.031) (0.031) (0.032) (0.031)
Mi,t
IL × ln Aj,i,t 0.174*** 0.178*** 0.173*** 0.179***
(0.058) (0.056) (0.058) (0.057)
F : βH = βL 16.38 17.19 14.93 17.06
0.0001 0.0000 0.0001 0.0000
F : βH = βL = 0 18.57 18.66 16.23 18.54
0.0001 0.0001 0.0003 0.0001
Panel B: Forecasts of ∆P AT /ABV
Dep. Var. ∆P AT /ABV
LM Aj Metric LM AP T LM AEKP LM AW LM AEP W
(1) (2) (3) (4)
LM Aj 0.380*** 0.343*** 0.401*** 0.422***
(0.037) (0.038) (0.033) (0.038)
LM ABV − LM Aj 1.772*** 0.704*** 1.799*** 2.236***
  (0.133) (0.079) (0.099) (0.166)
Mi,t
IH × ln Aj,i,t -0.097*** -0.090*** -0.092*** -0.091***
  (0.022) (0.022) (0.018) (0.022)
Mi,t
IL × ln Aj,i,t 0.026 0.033 0.017 0.033
(0.033) (0.032) (0.033) (0.032)
F : βH = βL 9.99 10.12 8.82 10.49
0.0016 0.0014 0.0008 0.0012
F : βH = βL = 0 20.85 17.74 27.79 18.80
0.0000 0.0001 0.0000 0.0001

62