Financial Analysts Journal | A Publication of CFA Institute Research

Are Cash Flows Better

Stock Return Predictors
Than Profits?
Stephen Foerster, CFA, John Tsagarelis, CFA, and Grant Wang, CFA
Stephen Foerster, CFA, is professor of finance at Ivey Business School, Western University, London, Ontario. John Tsagarelis, CFA, is senior
research analyst at Highstreet Asset Management and adjunct lecturer in the Department of Economics, Western University, London,
Ontario. Grant Wang, CFA, is senior vice president and co–chief investment officer at Highstreet Asset Management, London, Ontario.

I
Although various income nvestors rely on financial information, such as profitability and
statement–based measures cash-related measures, to assess a company’s intrinsic equity
predict the cross section of stock value and predict the cross section of average returns. Fama
returns, direct method cash flow and French (2006) found that more profitable companies have
measures have even stronger higher expected returns. Novy-Marx (2013) showed that profit-
predictive power. We transform
ability, measured by the ratio of gross profits to assets, predicts
indirect method cash flow state-
the cross section of average returns just as well as the book-
ments into disaggregated and
more direct estimates of cash to-market ratio does. Fama and French (2015) incorporated a
flows from operations and other profitability measure as a new factor, extending their well-known
sources and form portfolios on the three-factor model. Hou, Xue, and Zhang (2015, 2016) created
basis of these measures. Stocks a q-factor model that incorporates a profitability factor, which
in the highest-cash-flow decile performs well in explaining anomalies. Ball, Gerakos, Linnainmaa,
outperform those in the lowest by and Nikolaev (BGLN 2015) showed that an operating profitability
over 10% annually (risk adjusted). measure that better matches current expenses and revenues is
Our results are robust to invest- an even better predictor of returns and also showed that results
ment horizons and across risk fac- depend on whether the denominator is total assets or the market
tors and sector controls. We also value of equity. Thus, the search continues for financial informa-
show that, in addition to operating
tion that better predicts stock returns.
cash flow information, cash taxes
and capital expenditures provide Although the income statement has long been at the center of
incremental predictive power. financial statement analysis, well-documented shortcomings1 call
into question the efficacy of relying on its components to value
stocks and predict stock returns. Notorious bankruptcies, including
Disclosures: John Tsagarelis and Grant Enron and WorldCom, graphically illustrate that profitable GAAP
Wang are employed by Highstreet Asset income statements can coexist with negative operating or free
Management, an investment management
firm that uses empirically based research
and the combination of quantitative and We thank Feng Chen, Rossana Di Lieto, Melanie Foerster, Keith Godfrey, Matthias
fundamental analysis to capture alpha
Hanauer, CFA, Darren Henderson, Mark Huson, Andrew Karolyi, Aditya Kaul, Stephannie
drivers—growth, value, and quality.
Proprietary models are based on numer- Larocque, CFA, Juhani Linnainma, Rick Robertson, Lu Zhang, and seminar participants
ous factors, only a small portion of which at the University of Alberta Frontiers in Finance Conference, the Northern Finance
are related to the cash flow measure Association Conference, and Western University for helpful comments and suggestions.
variables and findings in this article. We thank Executive Editor Stephen Brown and Co-Editor Luis Garcia-Feijóo as well as
Andrew L. Berkin and Heiko Jacobs for their comments—all of which helped strengthen
CE Credits: 1 our paper. The research assistance of Yufeng Wang is greatly appreciated.

Volume 73 Number 1 © 2017 CFA Institute. All rights reserved. 73

Financial Analysts Journal | A Publication of CFA Institute

cash flows for the same company for long periods changes in accounts receivable/payable or taxes;
(see Appendix A for a typical representation). and discontinued operations, restructuring, or
More specifically, we believe that existing GAAP special charges. Thus, the details presented in the
requirements permit too many alternative types of indirect cash flow statement consist of noncash
financial statement presentations; such informa- operating items included in net income (or loss)
tion is too aggregated and can be inconsistently rather than operating cash receipts and payments.
presented, making it difficult for users to under- Moreover, it is impossible to model the magnitude,
stand the relationship between how accounting frequency, timing, and volatility of prospective
information is presented and the underlying eco- cash flows because inflows and outflows are not
nomic results of the company. Novy-Marx’s (2013) grouped according to common economic character-
intuition in this area certainly rings true: The istics.3 Consider the simple example of a sale made
farther down the income statement one goes, the wholly on credit. An increase in accounts receivable
more “polluted” profitability measures become and resulting from a customer’s delayed payment leaves
the less related to “true” or economic profitability. both revenues and net income unchanged as if full
Yet it is not obvious that any accrual accounting payment had been received. Only by examining a
profit measure, regardless of where on the income separate operating cash flow statement can users
statement it appears, should be superior to cash- learn that cash inflows were reduced and accruals
based measures of performance and value. were increased by the same amount. Furthermore,
changes in accounts receivable are not grouped
In theory, if financial information is sourced from with their intuitive economic companion “revenues”
the same underlying economic data, there should but, rather, with “working capital changes.” The
be no difference between using information from shortcoming is evident: The indirect cash flow
the income statement and using information from statement derives net cash flow from operating
the cash flow statement when making investment activities without separately presenting any of the
decisions. If, however, this assumption does not operating cash receipts and payments. Although
hold (as was the case with Enron and WorldCom), the Financial Accounting Standards Board (FASB)
the income statement may depict one state of sees accruals as improving the ability of earnings
affairs and the cash flow statement another. We to measure company performance by smoothing
believe that the lack of uniformity among reported out “temporary” fluctuations in cash flows, we see
statements and their disjointed presentations the lack of cash inflows from the tardy customer
make it extremely difficult for investors to test the and the resulting inconsistent presentation across
quality of a corporation’s historical earnings and several financial statements as fertile ground for
compare the results within and across industries. security mispricing.
Our study shows that by using a standardized
“direct cash flow” template, investors can better Investors often adjust for some of these issues by
understand a company’s historical, contemporane- using approximations, including “free cash flow to
ous, and forecasted return potential. equity” (FCFE).4 FCFE fits well with our direct cash
flow template computation (D), adjusted for capital
Both International Financial Reporting Standards expenditures, as shown in Table 1. In our study,
(IFRS) and US GAAP encourage companies to use we were most interested in understanding how
the direct method of financial statement presenta- investors price companies’ ongoing competitive
tion for reporting operating cash receipts and pay- advantage as measured by their ability to gener-
ments. The vast majority of companies, however, ate free cash flow vis-à-vis net income. In other
elect to present operating cash flows using the words, we believe that companies that generate
indirect method. Unfortunately, under the indirect cash flow from irregular “nonoperating or extraor-
method, no operating cash receipts or payments are dinary activities” or from capital raising (e.g., debt,
presented in the statement of cash flows.2 Instead, preferred shares, or equity) are not as prized as
net income is merely reconciled to “operating cash companies that generate cash flow from recurring
flow” by adjusting for accounting entries that do skill-based activities.5 As shown in Table 2, there
not generate or use cash, including depreciation;

74 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

is a material difference between our direct cash

Table 1. Direct Cash Flow Template flow measures (e.g., CFODM, our measure of direct
method cash flow from operations) and the FCFE
Operating Activities
metric (CFONM in Table 2). When deflated by total
Cash Inflows: assets, the correlation between these two variables
Sales is only 45% and drops to –17% when deflated by
+/– Change in accounts receivable the market value of equity. Although standard
FCFE and indirect cash flow measures can be cured
+/– Change in deferred revenues
by making similar computation adjustments, we
+/– Change in other cash inflows from operations
favor the wide-scale adoption of direct cash flow
= Cash Inflows to the Company statements as a superior method and as potentially
opening the door to further discovery of unique
Cash Outflows:
fundamental factors.
Less: Cost of goods sold
Less: Selling, general, and administrative expenses Because the direct method cash flow (DMCF)
+/– Change in accounts payable from operations aggregates cash flows with similar economic char-
acteristics and disaggregates those with dissimilar
+/– Change in inventories
characteristics,6 investors can, for example, isolate
= Net Cash Flows from Operations (A)
the growth rate associated with net cash inflows
Financing Activities stemming from operating activities as distinct
from gross cash inflows stemming from asset sales,
Less: Interest expense
tax reimbursements, or foreign exchange gains.
+/– Other financing income/expenses
This calculation is practically impossible using
= Net Cash Flows from Operations after Financing existing indirect cash flow statements. Investors
Activities (B)
can also model the percentage change in cash
Tax Activities flows associated with operating, financing, tax, and
nonrecurring activities. In this article, we highlight
Less: Taxes on income statement
how investors can potentially achieve superior
+/– Change in account-payable taxes
incremental risk-adjusted returns by replacing
+/– Change in deferred taxes commonly used profitability ratios—including
= Net Cash Flows from Operations after Financing and return on equity and price–earnings multiples—
Tax Activities (C) with their cash flow equivalents. This line of rea-
soning is consistent with research suggesting that
Other Nonoperating Activities (Extraordinary
Activities) earnings targets influence accounting decisions,
thereby creating strong incentives to bias accruals
+/– Discontinued operations/special charges
either upward or downward.7 We believe that the
+/– Foreign exchange gains/losses
incremental alpha (discussed later in the article)
+/– Pension gains/losses/contributions would not have been possible without reference
= Net Cash Flows from Operations after Financing, to the structural logic and internal efficiency of the
Tax, and Extraordinary Activities (D) “direct cash flow” statement.
Investing Activities Our main finding is that direct cash flow measures
Less: Capital expenditures (capex) are generally better stock return predictors than
= Free Cash Flows to Equityholders (FCF) indirect cash flow measures, which in turn tend
to be better than various income statement
Notes: This table presents a template for the estimation of profitability measures that focus on gross profits,
various free cash flow direct measures presented in the article.
CFAFAT is measured as (C) – capex, CFAF is measured as (B) – operating profits, or net income. In our study, we
capex, and CFO is measured as (A) – capex. The indirect cash first outlined a systematic process for combin-
flow measure, CFIM, is similar to (D) – capex. ing existing income and indirect method cash
flow statements to generate a direct cash flow

Volume 73 Number 1 cfapubs.org 75

76 cfapubs.org
Table 2. Correlations

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

(1) CFODM/TA 1.00 — — — — — — — — — — — — —

(2) CFOIM/TA 0.81 1.00 — — — — — — — — — — — —
(3) CFONM/TA 0.45 0.66 1.00 — — — — — — — — — — —
Financial Analysts Journal | A Publication of CFA Institute

(4) FinAct/TA –0.13 –0.22 –0.21 1.00 — — — — — — — — — —

(5) TaxAct/TA 0.49 0.50 0.62 –0.27 1.00 — — — — — — — — —
(6) OthAct/TA –0.23 0.12 0.55 –0.02 0.24 1.00 — — — — — — — —
(7) Capex/TA 0.22 0.29 0.28 0.04 0.14 0.07 1.00 — — — — — — —
(8) CFODM/MVE 0.28 0.08 –0.21 0.27 –0.17 –0.31 0.04 1.00 — — — — — —
(9) CFOIM/MVE 0.22 0.32 0.02 0.21 –0.11 –0.01 0.14 0.74 1.00 — — — — —
(10) CFONM/MVE –0.07 0.12 0.47 0.12 0.10 0.53 0.14 –0.17 0.20 1.00 — — — —
(11) FinAct/MVE –0.17 –0.25 –0.28 0.60 –0.29 –0.06 –0.03 0.59 0.43 0.03 1.00 — — —
(12) TaxAct/MVE 0.05 0.09 0.25 0.01 0.52 0.28 0.05 –0.02 0.12 0.43 –0.03 1.00 — —
(13) OthAct/MVE –0.12 0.12 0.45 –0.08 0.23 0.70 0.07 –0.61 –0.20 0.67 –0.30 0.31 1.00 —
(14) Capex/MVE –0.08 –0.07 –0.11 0.32 –0.22 –0.01 0.52 0.50 0.52 0.20 0.57 0.04 –0.21 1.00

Notes: This table presents correlations of the time series of averages across firms of various cash flow measures and components: our direct method cash flow from operations
measure, CFODM; our indirect method cash flow from operations measure, CFOIM; Novy-Marx’s (2013) free cash flow measure of net income plus depreciation/amortization
less working capital change less capital expenditures, CFONM; financing activities, FinAct; tax activities, TaxAct; other activities, OthAct; and capital expenditures, Capex. Each
of the variables is deflated by either total assets, TA, or market value of equity, MVE.

First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

approximation; that is, we created a new state- We constructed new profitability measures on the
ment that disaggregates operating, financing, tax, basis of various estimates of cash flows from opera-
and nonoperating cash flows to isolate recur- tions less capital expenditures, adjusted for financing
ring value-creation activities. We then created a and taxes. Consistent with BGLN (2015), we used
series of cash-based financial ratios and compared both total assets and market value of equity in the
them with BGLN’s (2015) operating profitability denominator of these ratios. We compared these
measure, Novy-Marx’s (2013) gross profitability ratios with Novy-Marx’s (2013) gross profitability
measure, and the traditional return on assets measure, BGLN’s (2015) operating profitability mea-
(ROA) measure—all of which have total assets sure, and return on assets as well as the earnings-
in the denominator. (We repeated our analysis to-price ratio (E/P) and related measures based on
with return on capital measures as well as the the market value of equity. We found a significant
traditional return on equity measure and found positive relationship between the derived cash-
that the results [unreported] are qualitatively the based ratios and future stock returns as measured by
same.) Consistent with BGLN (2015), we repeated high–low portfolio returns, information ratios, and
our analysis using market value of equity as the risk-adjusted alphas. Stocks in the highest-cash-flow
denominator. In this article, we show that our new decile outperformed those in the lowest-cash-flow
measures generate greater risk-adjusted returns decile by over 10% annually, after controlling for
than those based on standard income statement well-known risk factors. In contrast, other profitabil-
information. Our results are robust across invest- ity and earnings-based ratios generally had relatively
ment horizons and risk factors, including control- weak relationships with future returns in our sample
ling for sector differences. of S&P 1500 stocks over 1994–2013. We also show
that, in addition to operating cash flow information,
Our study is most closely related to and extends information on cash, taxes, and capital expenditures
Novy-Marx (2013) and BGLN (2015), as well as provides incremental predictive power.
Hou, Karolyi, and Kho (2011). Although Novy-
Marx (2013) argued that gross profitability
is the “cleanest” accounting measure of true The Direct Cash Flow Template
economic profitability and BGLN (2015) derived As an alternative to the traditional income and
an improved operating profitability measure, cash flow statements, we offer a direct cash flow
we argue that even “cleaner” measures can be template and an alternative set of cash-based
derived by focusing on the cash flow statement capital efficiency and valuation ratios. Unlike the
and cash that is available to equityholders.8 Like indirect approach that begins with net income and
BGLN (2015), we examined disaggregated infor- reconciles to operating cash flows by reversing
mation but with a focus on cash flows rather than noncash activities, the direct cash flow template
on accounting measures. Like Hou et al. (2011), adheres to the culinary principle of mise en place.
we found that cash-based measures capture That is, it organizes and sorts financial information
significant time-series variation in stock returns; into clusters of homogeneous business activities,
but our focus was on the US market rather than thereby permitting the user to understand how
the global market, and we examined much more value is being created and what can be extrapo-
extensive types of cash measures in addition to lated. In this way, operating activities associated
the measure of cash flow to price (cash earnings with a competitive advantage are value enhancing
before noncash charges, such as depreciation and and are likely to repeat. Although obtaining an
amortization). We empirically tested the relation- optimal capital structure can enhance value (given
ship between DMCF-based cash flow estimates the benefit of debt and interest deductibility
and future stock returns in the US equity market for tax purposes but balanced with the associ-
in order to test our conjecture that disaggregated ated risk), financing activities per se, as reflected
direct cash flows permit investors to separate in interest expenses, are not value enhancing.
recurring from nonrecurring value-adding activi- Potential benefits of certain tax structures have
ties, leading to superior return predictions. received considerable media attention recently

Volume 73 Number 1 cfapubs.org 77

Financial Analysts Journal | A Publication of CFA Institute

(i.e., large, profitable companies paying little in account-payable taxes and deferred taxes. Finally,
taxes) and may be value enhancing because the we estimate net cash flows from operations
lower the amount of cash taxes paid, the better; after financing, tax, and extraordinary activities
but their sustainability might be questionable. (D) by accounting for other nonoperating activi-
Other nonoperating activities may also have ties, including discontinued operations, foreign
some value-enhancing attributes but are probably exchange, and pension-related items. For com-
unsustainable. This overview is consistent with pleteness and to reconcile this cash flow estimate
earlier studies that demonstrated that cash-based with the traditional free cash flows to equityhold-
components of earnings are more persistent and ers, we account for cash flows from investing
thus of “higher quality” than accrual-based compo- activities by simply subtracting the cash outflows
nents (Sloan 1996). associated with capital expenditures. We con-
jecture that measures related to either net cash
We recognize that this template does not gener- flows from operations after financing activities
ate actual direct cash flows as envisioned by or net cash flows from operations after financing
the FASB (2008) because corporations are not and tax activities should be best at predicting the
required to publish segregated operating and non- cross section of future stock returns, depend-
operating cash flows. We believe, however, that ing on the sustainability of value-enhancing tax
it provides a better method than nontransformed structures.
accounting data for estimating intrinsic values.
Although all the individual components of the
cash flows are publicly available, we conjectured Data and Methodology
in our study that their availability alone does not We obtained the fundamental and pricing data
ensure their timely integration and use by inves- from the Standard & Poor’s Xpressfeed North
tors to estimate intrinsic values. Patterns must be American database (Xpressfeed). Our sample’s
isolated, revealed, and applied before stock prices investment universe is the S&P 1500, consisting
can reflect this information. Our analysis also of the largest 500 stocks by market capitaliza-
highlights what types of information are most tion, the mid-cap 400, and the small-cap 600.
informative. We chose the S&P 1500 to ensure that the
The direct cash flow template (Table 1) works stocks would be truly investable and that the
as follows. First, we directly estimate net cash tested investment strategies would be relevant to
flows from operating activities. We begin with the practitioners. The historical S&P 1500 constitu-
main cash inflow of sales, adjusted for changes in ents are also from Xpressfeed. For each month,
accounts receivable, deferred revenues, and other we obtained a list of the stocks in the S&P 1500
cash inflows from operations. We subtract cost of at that time to avoid any survivorship bias. As
goods sold as well as selling, general, and admin- is common in studies that rely on accounting
istrative expenses and then adjust for changes in metrics—and given major structural differences
accounts payable from operations and changes in between financial and nonfinancial companies
inventories. The result is our estimate of net cash (e.g., leverage)—for most of our reported analysis
flows from operations (A), similar to Ball, Gerakos, we excluded financial companies (banks, insur-
Linnainmaa, and Nikolaev (2016).9 ance companies, and REITs), about 15% of the
overall sample.10 We included companies with
Next, we estimate net cash flows from opera- negative earnings (about 13% of the sample) and
tions after financing activities (B) by subtracting companies with negative operating cash flows
interest expenses and adjusting for other financ- (about 6% of the sample). We used data for
ing income and expenses. We then estimate October 1994–December 2013; cash flow state-
net cash flows from operations after financing ments have been mandatory in the United States
and tax activities (C) by subtracting cash flows only since 1987, and the actual index constituents
related to tax activities—including taxes on of the S&P 1500 have been available in readily
the income statement, adjusted for changes in usable database form only since October 1994.

78 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

The major fundamental variables that we used in includes current assets; investments; intangibles;
our study are as follows (Xpressfeed mnemonics): and property, plant, and equipment, as reported on
each company’s balance sheet, and (2) market value
Operating activities net cash flow: OANCF of equity is measured as common shares outstand-
Net sales: SALE ing multiplied by the month-end prices.
Accounts receivable – Decrease/increase: On the basis of our template definitions in Table 1,
RECCH we created various measures of direct and indirect
Cost of goods sold: COGS free cash flow. CFAFAT represents net cash flows
from operations after financing and tax activities
Selling, general, and administrative expense:
(C) – Capex; CFAF represents net cash flows from
XSGA
operations after financing activities (B) – Capex;
Depreciation and amortization – Income state- and CFO represents net cash flows from opera-
ment: DP tions (A) – Capex. We compared these direct
Depreciation and amortization – Cash flow method cash flow (DMCF) measures with the
statement: DPC indirect method cash flow measure CFIM, defined
as operating activities net cash flow (OANCF) –
Funds from operations – Other: FOPO
Capex. Note that this indirect measure, OANCF,
Accounts payable and accrued liabilities – results in the same number as net cash flows from
Increase/decrease: APALCH operations after financing, tax, and extraordinary
Inventory – Decrease/increase: INVCH activities (D) in Table 1.
Assets and liabilities – Other – Net change: We then used these definitions of free cash flow to
AOLOCH construct various metrics for cash returns on assets
Sale of property, plant, and equipment and and free cash flow yield. The variable CFAFAT/TA
investments – Gain/loss: SPPIV is the cash return on assets measured as the direct
free cash flow metric CFAFAT divided by total
Interest and related expense – Total: XINT
assets. The variables CFAF/TA, CFO/TA, and CFIM/
Income taxes – Total: TXT TA are alternative measures of return on assets,
Income taxes – Accrued – Increase/decrease: with the numerators CFAF, CFO, and CFIM, respec-
TXACH tively. We compared these measures with a number
of accounting-based measures. The numerator of
Deferred taxes – Cash flow: TXDC
the operating profit measure, OP/TA, is operating
Special items: SPI profits (as estimated by BGLN [2015] by subtracting
Discontinued operations: DO from sales the cost of goods sold and selling, gen-
eral, and administrative expenses excluding R&D)
Extraordinary items: XI
and the denominator is total assets. The numerator
Capital expenditure: CAPX of Novy-Marx’s (2013) gross profit to total assets,
Income before extraordinary items – Available GP/TA, is gross profit measured as sales less cost of
for common: IBCOM goods sold. The traditional return-on-assets ratio is
measured as IBCOM (income before extraordinary
Following industry practice, we created trail- items available for common shareholders) divided
ing 12-month (TTM) values for the fundamental by total assets. The price yield measures are similar
variables. We updated these TTM values with each to the measures just described except that the
quarter’s new information and combined them with market value of equity (MVE) replaces total assets
monthly return data to conduct the tests. Despite as the denominator. Note that the traditional earn-
the quarterly updates and to avoid any look-ahead ings yield, E/P (the inverse of the price-to-earnings
bias, we used only accounting data that had been ratio), is measured as IBCOM, or net income,
lagged by four months.11 We used two variables divided by the market value of equity and is noted
to normalize various measures: (1) Total assets simply as NI/MVE. On the basis of our discussion

Volume 73 Number 1 cfapubs.org 79

Financial Analysts Journal | A Publication of CFA Institute

in the previous section, we would expect those CFONM/TA measures: 0.81 for the first two
measures with CFAFAT and CFAF in the numerator measures, dropping to 0.45 for the first and third.
to be superior. Consistent with BGLN (2015), the correlations
across variables with the same numerator but
For each month, we ranked the S&P 1500 with MVE or TA in the denominator are quite low
companies by a particular measure of return in many instances: for CFODM, 0.28; for CFOIM,
on assets or yield and divided the universe into 0.32; and for CFONM, 0.47. These results show
deciles from lowest (P1) to highest (P10). We the impact of the book value of total assets, which
calculated one-month-ahead portfolio returns on changes slowly over time, compared with the
a value-weighted basis (we also considered other market value of equity, which changes frequently
horizons). In standard fashion, we estimated a and potentially by larger magnitudes.
long–short portfolio return as the spread return
difference between the highest (P10) and low- Across the three cash flow measures deflated by
est (P1) portfolio returns. We then tested for the TA, we see a consistent pattern of correlations
significance of the return differences. for FinAct (negative), TaxAct (positive), and Capex
(positive). The absolute values of the correlations
for FinAct are generally low, below 0.23, whereas
Information in Cash Flow those for Capex are generally slightly higher and
Components those for TaxAct higher still (above 0.48). The
We began our analysis by investigating the impact absolute values of the correlations between OthAct
of segregating various cash flow components in and the CFODM and CFOIM variables are low,
order to see how those components might provide but the absolute value of the correlation between
incremental information about predicting stock OthAct and CFONM is surprisingly high (0.55). Not
returns. For example, in addition to information surprisingly, the correlations tend to be lower with
regarding a company’s operations, is there incre- MVE in the denominator. These preliminary results
mental information in cash flows about financing, prompted us to conduct a more rigorous investiga-
taxes, investments, or other nonoperating activi- tion of the components of cash flows.
ties? We were also able to compare operating cash Following BGLN (2015, Table 7), we performed
flow information with accounting measures of Fama and MacBeth (1973) cross-sectional regres-
gross profit and operating profit. sions.12 For each month of our sample, we
As shown in Table 2, we first examined the cor- regressed the one-month-ahead returns for each
relations of the time series (winsorized at 1% and stock on our measure of net cash flows from
99%) of equal-weighted averages across compa- operations, net CF ops (“A” in the Table 1 tem-
nies of various cash flow measures and compo- plate). In additional regressions, we included a
nents: our measure of direct method cash flow number of other independent variables: financing
from operations, CFODM; our measure of indirect activities as measured by interest expenses ±
method cash flow from operations, CFOIM; other financing income/expenses; tax activities as
Novy-Marx’s (2013) free cash flow measure of measured by taxes on the income statement ±
net income plus depreciation/amortization minus changes in account-payable taxes ± changes in
working capital change minus capital expendi- deferred taxes; other nonoperating activities
tures, CFONM; financing activities, FinAct (see measured as discontinued operations/special
Table 1 for definitions of this and other “activities” charges ± foreign exchange gains/losses ± pension
measures); tax activities, TaxAct; other activities, gains/losses/contributions; and capital expendi-
OthAct; and capital expenditures, Capex. We tures (capex). We deflated all these variables by
deflated each of the variables by either total assets the book value of total assets. Similar to BGLN
(TA) or market value of equity (MVE). (2015), we included a number of control variables:
Log(BVE/MVE) is the natural logarithm of the ratio
Not surprisingly, there are high correlations of book value of equity to market value of equity;
between the CFODM/TA, CFOIM/TA, and log(ME) is a size variable measured as the natural

80 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

logarithm of the market value of equity; r1,1 is the Given our earlier results suggesting that (for our
stock’s one-month-prior return; and r12,2 is the sample) the gross profit measure is dominated by
stock’s prior-year return (skipping a month). We other measures, this result is not surprising. In
winsorized all independent variables on the basis column 7, the operating profit measure remains
of the 1st and 99th percentiles. We also per- significant, with a t-statistic of 3.85. However,
formed separate regressions, replacing the net CF both the tax activities coefficient estimate and
ops variable with (1) Novy-Marx’s (2013) gross the capex coefficient estimate are marginally
profit measure, (2) the BGLN (2015) operating significant and negative. The column 8 net CF ops
profit measure, (3) our measure of cash flows from measure results are similar. In addition to the net
operations based on the indirect method cash flow CF ops variable—which remains significant, with
(CFOIM), or (4) Novy-Marx’s (2013) free cash flow a t-statistic of 3.60—the tax activities coefficient
measure of net income plus depreciation/amorti- estimate is negative and marginally significant,
zation minus working capital change minus capital with a t-statistic of –1.91; the capex coefficient
expenditures (CFONM). estimate is also negative and significant, with a
t-statistic of –2.03.
Table 3 reports our results, including averages of
the Fama–MacBeth (1973) slope coefficients and Our finding no significance in the financing activi-
their t-values. Column 1 presents the regression ties coefficient estimate is consistent with our
of returns on gross profit alone and the control initial conjecture and also with BGLN (2015), who
variables; column 2 does the same with operating found no significance in their interest coefficient
profit and the control variables, whereas column estimate for their “all-but-microcaps” sample.
3 does the same with net CF ops and the control Unlike BGLN (2015), however, we found our tax
variables. We can see in column 1 that the gross activities coefficient estimate to be negative and
profit measure is only marginally significant, with a significant.14 We would expect a negative correla-
t-statistic of 1.81. In column 2, the operating profit tion with the intrinsic value of equityholders as
measure is significant, with a t-statistic of 3.38; in cash taxes increase.15
column 3, the net CF ops coefficient estimate is
The negative coefficient estimate for the capex
also significant and of a similar order of magnitude,
variable, which is consistent with prior literature, is
with a t-statistic of 3.22.
known as the investment effect or the asset growth
The regression results reported in columns 4 and 5 effect, whereby companies that increase their
allow us to compare the results in columns 1 and 2 investments (e.g., as measured by capital expendi-
with the effect of adding the net CF ops measure. tures as a percentage of total assets) subsequently
In column 4, we see that the cash flow measure experience lower risk-adjusted returns.16 Two
subsumes the gross profit measure, which is no possible explanations are related to the q-theory
longer significant. Column 5 shows that neither of investment and overinvestment. In the first
the operating profit measure nor the cash flow explanation, companies invest more when stock
measure is significant (perhaps owing to multicol- returns are lower than expected. In the second
linearity), although the cash flow measure does explanation, companies that increase investments
have a slightly higher t-statistic.13 may overinvest. Markets may interpret some capi-
tal investments as inefficient; for example, capital
The regression results presented in columns 6, 7, expenditures in large, risky projects may lead to
and 8 repeat the analysis in columns 1, 2, and 3 unfavorable outcomes.
but with additional segregated cash flow compo-
nents related to financing activities, tax activities, Column 9 presents the regression of returns on
nonoperating activities, and capex (investing CFOIM and the segregated cash flow components
activities). In all cases, the adjusted R2 increases. related to financing activities, tax activities, non-
In column 6, the gross profit measure drops in operating activities, and capex (investing activi-
significance, with a t-statistic of 1.64, and none of ties), as well as the control variables. As expected,
the cash flow component measures are significant. the CFOIM coefficient estimate is significant, with

Volume 73 Number 1 cfapubs.org 81

 Table 3. Fama–MacBeth Regressions for Profitability and Cash Flow Components (t-statistics in parentheses)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Gross profit 0.0046 0.0018 0.0045
(1.81) (0.70) (1.64)
Operating profit 0.0159 0.0090 0.0205
(3.38) (1.11) (3.85)

82 cfapubs.org
Net CF ops 0.0153 0.0140 0.0094 0.0194
(3.22) (2.92) (1.18) (3.60)
CFOIM 0.0257
(3.71)
CFONM 0.0077
(0.93)
FinAct –0.0360 –0.0450 –0.0577 –0.0403 –0.0512
Financial Analysts Journal | A Publication of CFA Institute

(–0.66) (–0.84) (–1.08) (–0.72) (–0.92)

TaxAct –0.0051 –0.026 –0.0286 –0.0223 –0.0052
(–0.41) (–1.70) (–1.91) (–1.49) (–0.36)
Nonop. activities –0.0146 –0.0107 0.0025 –0.0157 –0.0233
(–1.63) (–1.17) (0.25) (–1.69) (–2.17)
Capex –0.0118 –0.0171 –0.0200 –0.0215 –0.0065
(–1.23) (–1.70) (–2.03) (–2.18) (–0.55)
Log(BVE/MVE) 0.0020 0.0026 0.0020 0.0022 0.0025 0.0016 0.0021 0.0013 0.0013 0.0013
(1.67) (2.49) (1.71) (1.89) (2.44) (1.41) (1.93) (1.19) (1.10) (1.13)
Log(ME) –0.0006 –0.0007 –0.0008 –0.0008 –0.0007 –0.0005 –0.0006 –0.0008 –0.0009 –0.0007
(–1.03) (–1.17) (–1.36) (–1.30) (–1.27) (–0.94) (–1.11) (–1.44) (–1.52) (–1.21)
r1,1 –0.0240 –0.0240 –0.0239 –0.0243 –0.0247 –0.0250 –0.0250 –0.0248 –0.0250 –0.0245
(–3.11) (–3.13) (–3.09) (–3.16) (–3.31) (–3.26) (–3.36) (–3.28) (–3.30) (–3.24)
r12,2 –0.0007 –0.0001 –0.0009 –0.0008 –0.0004 –0.0010 –0.0006 –0.0014 –0.0014 –0.0010
(–0.20) (–0.03) (–0.24) (–0.23) (–0.11) (–0.28) (–0.17) (–0.38) (–0.41) (–0.28)
Adjusted R2 5.78% 5.52% 5.64% 6.10% 6.05% 7.44% 7.22% 7.29% 7.31% 7.31%

Notes: This table presents average Fama and MacBeth (1973) slope coefficients and their t-values from cross-sectional regressions. The dependent variables are the one-month-ahead returns for
each stock. The independent variables are gross profit as measured by Novy-Marx (2013), Gross profit; operating profit as measured by BGLN (2015), Operating profit; net cash flow from operations
(“A” in Table 1), Net CF ops; net cash flow from operations measured by the indirect cash flow method, CFOIM; free cash flows measured as net income plus depreciation less change in working
capital less capital expenditures, CFONM; FinAct, financing activities as measured by interest expenses +/– other financing income/expenses; TaxAct, tax activities as measured by taxes on the income
statement +/– changes in accounts-payable taxes +/– changes in deferred taxes; other nonoperating activities, Nonop. activities, measured as discontinued operations/special charges +/– for-eign
exchange gains/losses +/– pension gains/losses/contributions; and Capex, capital expenditures. We deflate all these variables by the book value of total assets. Other control variables include

First Quarter 2017

log(BVE/MVE), the natural logarithm of the ratio of the book value of equity to the market value of equity; log(ME), the natural logarithm of the market value of equity; r1,1, the stock’s one-month-
prior return; and r12,2, the stock’s prior-year return skipping a month. We winsorize all independent variables on the basis of the 1st and 99th percentiles.
 Are Cash Flows Better Stock Return Predictors Than Profits?

a t-statistic of 3.71. Consistent with our column The returns generally increase somewhat mono-
7 and 8 results, the capex coefficient estimate tonically from the low to the high portfolios for the
is significantly negative. The tax activities coef- various measures. We see from Table 4 that the
ficient estimate is still negative but not significant, Novy-Marx (2013) GP/TA variable has the largest
whereas the nonoperating activities coefficient P10 return among the ROA variables but not the
estimate is negative and marginally significant. biggest P10–P1 spread because many of the cash
Finally, column 10 presents a regression similar to flow measures have more cross-sectional variation.
column 9’s but with the CFOIM measure replaced For the portfolios sorted on the basis of our three
with free cash flows (CFONM) measured as net DMCF efficiency measures (CFAFAT/TA, CFAF/
income plus depreciation minus change in working TA, and CFO/TA), the high–low (P10–P1) return
capital minus capital expenditures (Novy-Marx differences are significantly positive (t-statistics
2013). Only the nonoperating activities variables range from 2.14 to 2.73), with monthly return
are significant. differences ranging from 0.64% to 0.82% (or 8.0%
to 10.3% annually). Information ratios range from
Overall, our results suggest that there is additional 0.489 to 0.623. For the indirect method cash flow
information in segregating cash flow components return measure (CFIM/TA), the high–low monthly
for predicting the cross section of returns. In return difference drops to 0.49% (6.0% annually)
particular, in addition to the information contained but is not significant, and the information ratio
in cash flows from operations, there is incremental drops to 0.356. The return difference for the
information in disaggregating tax activities and portfolios sorted on the basis of operating profits
capital expenditures. Companies that generate to total assets (OP/TA) is very similar to that of
strong cash flows from operations while paying the CFIM/TA variable (0.49%; 6.0% annually) and
relatively low cash taxes and having relatively less is not significant, with the same information ratio
capital expenditure tend to perform best. (0.356). The high–low portfolio return based on
the measure of gross profit to total assets (GP/
TA) is also similar (0.49%; 6.1% annually) but, given
Cash Flow, Profitability Return and the lower standard deviation, is significant, with a
Yield Measures, and One-Month- t-statistic of 2.10; the information ratio is higher, at
Ahead Returns 0.480. Finally, for the measure of return on assets
Next, we analyzed our results with the aggregated (NI/TA), the monthly high–low return difference is
cash flow measures. Table 4 reports our results only 0.06% (0.7% annually) and is not significantly
from comparing the one-month-ahead returns for different from zero, with the information ratio
the various value-weighted portfolios (P1 through dropping to 0.040.18 Consistent with our conjec-
P10), as well as the P10–P1 (high–low) portfolio ture, among these cash flow return measures, the
return spread, sorted on the basis of the previous best-performing high–low portfolio is based on
month’s return-on-assets or yield measure over sorting by the CFAF measure (net cash flows from
October 1994–December 2013. Table 4 also shows operations after financing activities [“B” in Table
the standard deviations of the P10–P1 returns, 1] less capex). This measure also has the highest
the t-statistics of the significance of the P10–P1 information ratio. These initial results suggest that
returns, the minimum and maximum monthly cash flow measures are superior to profitability
P10–P1 observations, and the information ratio measures in long–short positions because (1) the
(IR) measured as the annualized P10–P1 return cash-related information may be more informative
divided by the annualized standard deviation of the and (2) the cash flow measures capture a wider
return difference.17 As expected, there is a general dispersion across the portfolios.
(but not perfect) monotonic relationship across all The bottom part of Table 4 presents measures with
measures, with higher measures tending to exhibit market value of equity in the denominator. We can
higher one-month-ahead returns than lower mea- see that all the return differences for the portfolios
sures. The top half of the table presents measures sorted on the basis of our three direct method
with total assets in the denominator.

Volume 73 Number 1 cfapubs.org 83

84 cfapubs.org
Table 4. Return and Yield Measures and One-Month-Ahead Returns, October 1994–December 2013

P1 P10 P10– Std.

Measure (low) P2 P3 P4 P5 P6 P7 P8 P9 (high) P1 Dev. t-Stat. Min. Max. IR

CFAFAT/TA 0.517% 0.547% 0.688% 0.745% 0.931% 0.759% 1.079% 1.135% 1.058% 1.159% 0.642% 4.555 2.14 –11.740% 14.952% 0.489
CFAF/TA 0.413 0.679 0.417 1.040 0.824 0.859 1.063 1.207 0.890 1.232 0.819 4.554 2.73 –12.906 15.300 0.623
Financial Analysts Journal | A Publication of CFA Institute

CFO/TA 0.435 0.633 0.694 0.872 0.787 0.859 1.031 1.168 0.947 1.221 0.786 4.572 2.61 –12.089 16.011 0.595
CFIM/TA 0.672 0.494 0.672 0.939 0.801 0.865 0.932 1.159 1.034 1.161 0.489 4.757 1.56 –15.334 13.541 0.356
OP/TA 0.630 0.816 0.681 1.058 0.770 0.786 0.885 0.878 0.995 1.118 0.488 4.743 1.56 –18.139 14.284 0.356
GP/TA 0.785 0.781 0.796 0.690 0.958 0.847 1.009 0.733 1.057 1.276 0.491 3.547 2.10 –8.589 11.545 0.480
NI/TA 1.055 0.710 0.918 0.638 0.976 0.940 0.942 0.858 0.781 1.110 0.056 4.847 0.17 –23.592 15.464 0.040
CFAFAT/
MVE 0.760 0.556 0.497 0.560 0.691 1.223 1.061 1.122 1.311 1.476 0.716 3.935 2.77 –11.336 20.771 0.630
CFAF/MVE 0.728 0.354 0.532 0.437 0.909 1.129 1.170 1.090 1.437 1.494 0.766 3.837 3.03 –11.875 13.083 0.692
CFO/MVE 0.740 0.376 0.341 0.704 1.150 1.056 1.019 1.257 1.348 1.294 0.554 4.401 1.91 –13.778 20.451 0.436
CFIM/MVE 0.762 0.606 0.410 0.640 0.597 0.976 1.146 1.407 1.213 1.490 0.729 4.086 2.71 –12.829 12.088 0.618
OP/MVE 0.611 0.833 0.807 0.891 0.816 0.931 1.127 1.118 1.474 1.310 0.699 6.381 1.67 –22.587 38.757 0.380
GP/MVE 0.792 0.646 0.829 0.974 1.068 1.023 1.151 1.296 1.268 1.373 0.581 6.427 1.37 –20.541 43.622 0.313
NI/MVE 0.916 0.663 0.804 0.870 0.846 0.870 0.909 1.015 1.248 1.341 0.425 5.437 1.19 –26.958 24.932 0.271

First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

cash flow yield measures (CFAFAT/MVE, CFAF/ returns of portfolios P1 through P10 as well as the
MVE, and CFO/MVE) and the indirect method cash P10–P1 return spread (i.e., the portfolios sorted
flow return measure (CFIM/MVE) are significantly on either the return-on-assets variable or the yield
positive (t-statistics range from 1.91 to 3.03), with variable) to run regressions against well-known
monthly return differences ranging from 0.55% to risk factors. We used monthly data over the entire
0.77% (6.9% to 9.6% annually). Information ratios sample period, October 1994–December 2013.
range from 0.436 to 0.692. The return difference Our regression model is
for the portfolios sorted on operating profit yield
(OP/MVE) is 0.70% (8.7% annually) and margin- Pi ,t + 1 = α i + β1iMKTRFt + β2i SMBt + β3iHML t + εi ,t + 1 ,
(1)
ally significant, but the information ratio is lower t = 1,...,N.
(0.380). For the gross profit yield (GP/MVE), the
high–low return difference is 0.58% (7.2% annu- The dependent variables are the individual decile
ally) but is not significant, and the information ratio portfolios, P1 (low) through P10 (high), as well
drops to 0.313. Finally, for the earnings yield mea- as the P10–P1 return difference for the various
sure (NI/MVE), the monthly high–low return differ- portfolios sorted on the return-on-assets and
ence is 0.43% (5.2% annually) and is not significant, yield measures (i) in Table 4 for each monthly
with an information ratio of only 0.271. Consistent time series. For example, we regressed returns
with our conjecture, among these cash flow return for November 1994 (t + 1) on the basis of infor-
measures, the best-performing long–short portfolio mation as of the end of October 1994 (t). The
is again based on sorting by CFAF/MVE (net cash independent variables are the Fama–French
flows from operations after financing [“B” in Table (1993) three-factor model, in which MKTRF is
1] less capex) and generates the highest informa- the market risk premium (market return in excess
tion ratio. The sortings on the AFAT and AF DMCF of the risk-free rate), SMB is the small minus
measures result in higher information ratios than big (size) factor, and HML is the high minus low
does the sorting on the indirect cash flow measure (book-to-market) factor (we discuss the Fama–
(CFIM/MVE). Our results (with market value of French five-factor model later in the article). 21
equity in the denominator of our measures) suggest After controlling for all these risk factors, the
a robustness of the superiority of cash flow mea- intercept, α, is interpreted as the alpha or abnor-
sures compared with profitability measures. Our mal return. If alpha is positive and significant,
results are also robust across subperiods.19 the implication is that the other factors cannot
be combined to form a mean–variance-efficient
Our overall interpretation of the results in Table 4
portfolio, suggesting how an investor could tilt a
is as follows. Direct cash flow measures tend to be
portfolio toward a profitability strategy, such as
superior to indirect cash flow measures because of
investing in portfolios that have recently gener-
the additional information provided by segregating
ated high cash flows relative to their total assets
operating, financing, and tax activities. Cash flow
or market value of equity. We used White’s
measures tend to be better stock return predic-
(1980) heteroskedasticity-consistent standard
tors than profitability measures (operating profit,
errors. For comparison, we present the aver-
gross profit, and net income) because cash-related
age monthly returns in excess of the monthly
measures are economically “cleaner” than profit-
Treasury bill (from Ken French’s website).
ability measures and are more closely related to
how companies are valued. Thus, inconsistencies Table 5 summarizes our regression results. Panel
in how accrual financial statements are presented A reports the regression results for the P1 (low)
can confuse investors who are trying to under- portfolio sorted on the various return and yield
stand what to extrapolate and how sustainable the measures. Panel B presents the regression results
value-creating activities are.20 for the P10 (high) portfolio sorted on the various
return and yield measures. Panel C shows the
Portfolio-Sorted Regressions. We adjusted regression results for the P10–P1 return differ-
for risk in a more systematic manner by using the ence sorted on the same measures.

Volume 73 Number 1 cfapubs.org 85

Financial Analysts Journal | A Publication of CFA Institute

Table 5. Regression Results for Return and Yield Measures, October 1994–December 2013
(t-statistics in parentheses)

Average Excess
Dependent Variable Return (%) α (%) MKTRF (%) SMB (%) HML (%) Adjusted R2

A. P1
CFAFAT/TA 0.285 –0.23 1.250 0.053 –0.212 0.766
(0.63) (–1.06) (22.26) (0.62) (–2.16)
CFAF/TA 0.181 –0.31 1.221 0.068 –0.243 0.758
(0.41) (–1.40) (22.56) (0.77) (–2.36)
CFO/TA 0.203 –0.29 1.240 0.049 –0.270 0.764
(0.45) (–1.28) (22.23) (0.56) (–2.67)
CFIM/TA 0.440 –0.16 1.315 0.175 –0.165 0.772
(0.93) (–0.69) (22.59) (2.18) (–2.01)
OP/TA 0.398 –0.25 1.328 –0.030 0.193 0.806
(0.91) (–1.32) (26.62) (–0.47) (2.96)
GP/TA 0.553 0.19 0.957 –0.177 0.145 0.805
(1.79) (1.34) (26.90) (–4.58) (2.65)
NI/TA 0.823 0.22 1.363 0.234 –0.334 0.811
(1.65) (0.99) (20.33) (2.86) (–3.88)
CFAFAT/MVE 0.528 –0.01 1.132 0.024 0.213 0.761
(1.38) (–0.04) (23.31) (0.27) (2.61)
CFAF/MVE 0.496 –0.03 1.137 0.040 0.169 0.760
(1.28) (–0.17) (21.98) (0.46) (2.15)
CFO/MVE 0.508 0.00 1.166 0.023 0.026 0.748
(1.25) (–0.02) (21.86) (0.26) (0.30)
CFIM/MVE 0.530 –0.12 1.270 0.061 0.254 0.742
(1.21) (–0.49) (19.68) (0.62) (2.87)
OP/MVE 0.379 0.05 1.169 –0.032 –0.650 0.842
(0.85) (0.26) (24.41) (–0.54) (–9.28)
GP/MVE 0.560 0.28 1.101 –0.127 –0.609 0.858
(1.37) (1.77) (25.93) (–2.04) (–10.07)
NI/MVE 0.684 0.00 1.405 0.236 –0.134 0.780
(1.35) (0.02) (20.27) (2.42) (–1.28)

B. P10
CFAFAT/TA 0.927 0.68 0.986 –0.215 –0.360 0.827
(2.68) (4.51) (26.74) (–3.54) (–6.22)
CFAF/TA 1.000 0.80 0.945 –0.246 –0.388 0.827
(2.98) (5.42) (27.00) (–4.29) (–7.19)
(continued)

86 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

Table 5. Regression Results for Return and Yield Measures, October 1994–December 2013
(t-statistics in parentheses) (continued)

Average Excess
Dependent Variable Return (%) α (%) MKTRF (%) SMB (%) HML (%) Adjusted R2

CFO/TA 0.989 0.78 0.948 –0.243 –0.360 0.820

(2.95) (5.18) (26.25) (–4.17) (–6.51)
CFIM/TA 0.929 0.73 0.952 –0.239 –0.453 0.855
(2.73) (5.42) (29.44) (–4.54) (–9.41)
OP/TA 0.886 0.69 0.975 –0.239 –0.531 0.825
(2.44) (4.40) (26.19) (–4.22) (–9.02)
GP/TA 1.044 0.79 0.887 –0.077 –0.190 0.758
(3.30) (4.87) (23.88) (–1.41) (–3.01)
NI/TA 0.878 0.71 0.924 –0.258 –0.456 0.852
(2.65) (5.32) (32.07) (–5.74) (–10.04)
CFAFAT/MVE 1.244 0.54 1.147 0.177 0.640 0.744
(3.07) (2.58) (16.77) (2.05) (6.05)
CFAF/MVE 1.262 0.57 1.163 0.136 0.625 0.753
(3.12) (2.65) (19.00) (1.75) (6.58)
CFO/MVE 1.062 0.30 1.220 0.148 0.755 0.795
(2.54) (1.49) (19.48) (2.09) (8.59)
CFIM/MVE 1.258 0.55 1.158 0.116 0.745 0.798
(3.18) (3.13) (24.53) (1.74) (9.37)
OP/MVE 1.078 0.27 1.323 0.142 0.629 0.690
(2.26) (1.11) (14.92) (1.06) (3.99)
GP/MVE 1.141 0.32 1.273 0.335 0.676 0.690
(2.40) (1.26) (13.23) (2.58) (4.39)
NI/MVE 1.109 0.52 1.110 0.034 0.458 0.741
(2.91) (2.67) (20.47) (0.46) (5.72)

C. P10–P1
CFAFAT/TA 0.642 0.91 –0.265 –0.268 –0.148 0.121
(2.14) (3.16) (–3.62) (–2.55) (–1.14)
CFAF/TA 0.819 1.11 –0.276 –0.313 –0.146 0.145
(2.73) (3.90) (–3.98) (–2.89) (–1.11)
CFO/TA 0.786 1.07 –0.292 –0.292 –0.090 0.147
(2.61) (3.69) (–4.08) (–2.74) (–0.70)
CFIM/TA 0.489 0.90 –0.363 –0.414 –0.289 0.234
(1.56) (3.12) (–5.11) (–4.41) (–2.76)
OP/TA 0.488 0.94 –0.354 –0.208 –0.724 0.305
(1.56) (3.64) (–5.71) (–2.31) (–7.92)
(continued)

Volume 73 Number 1 cfapubs.org 87

Financial Analysts Journal | A Publication of CFA Institute

Table 5. Regression Results for Return and Yield Measures, October 1994–December 2013
(t-statistics in parentheses) (continued)

Average Excess
Dependent Variable Return (%) α (%) MKTRF (%) SMB (%) HML (%) Adjusted R2

GP/TA 0.491 0.59 –0.072 0.101 –0.335 0.109

(2.10) (2.56) (–1.29) (1.49) (–3.74)
NI/TA 0.056 0.49 –0.438 –0.492 –0.122 0.325
(0.17) (1.78) (–5.67) (–5.72) (–1.19)
CFAFAT/MVE 0.716 0.56 0.018 0.152 0.426 0.124
(2.77) (2.27) (0.26) (1.69) (4.45)
CFAF/MVE 0.766 0.61 0.028 0.096 0.456 0.144
(3.03) (2.51) (0.43) (1.19) (5.00)
CFO/MVE 0.554 0.31 0.055 0.124 0.730 0.279
(1.91) (1.21) (0.76) (1.37) (7.69)
CFIM/MVE 0.729 0.67 –0.111 0.054 0.491 0.188
(2.71) (2.68) (–1.58) (0.61) (6.16)
OP/MVE 0.581 0.25 0.158 0.172 1.279 0.405
(1.37) (0.79) (1.53) (1.19) (7.25)
GP/MVE 0.699 0.04 0.173 0.462 1.285 0.419
(1.67) (0.13) (1.61) (3.15) (7.34)
NI/MVE 0.425 0.52 –0.294 –0.203 0.592 0.278
(1.19) (1.74) (–3.45) (–2.00) (4.92)

Across the panels, we see that the portfolio alphas of the DMCF variables result in larger alphas than
(and the average returns) increase as we move from the indirect method variables. Among the DMCF
P1 to P10 for the various return-on-assets and yield variables with TA in the denominator, the alpha for
measures. In Panel A (P1 results), most of the alphas CFAF is 1.11%, which translates into an annual return
are negative but not significant, whereas in Panel difference of 14.2%; for CFO, the alpha is 1.07%,
B (P10 results), all the alphas are positive and most which translates into an annual return difference of
are significant, with the exception of CFO/MVE, 13.6%, even stronger than the Table 4 results. With
OP/MVE, and GP/MVE. Consistent with our results MVE as the denominator, CFIM has an alpha of
in Table 4, in Panel C all three of the DMCF return 0.67%, for an annual difference of 8.3%. Thus, even
and yield measure P10–P1 return differences are controlling for well-known risk factors, we found sig-
significantly positive (with the exception of CFO/ nificant value in using information available from our
MVE), as are the indirect measure return differences. “direct cash flow” statements. In general, direct cash
The OP return differences are significantly positive flow measures tend to be superior to indirect cash
vis-à-vis total assets but not market value of equity. flow measures, which in turn tend to be better than
In most cases, the cash flow–based measures have profitability measures; among profitability measures,
higher alphas than do the corresponding OP mea- operating profit measures tend to be superior to
sures. The GP measure is significant with TA as the gross profit measures, which in turn are superior to
denominator but is not significant with MVE as the net income measures.
denominator, in both cases with a lower alpha than
the corresponding OP measure. For the net income We contrasted our P10–P1 results with those
measure, the results are marginally significant. Most of Novy-Marx (2013, Table 2). Despite some

88 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

differences in methodologies,22 we found that Fama–French Five-Factor Model

our alpha estimate for the measure of gross profit Regressions. Fama and French (2015) recently
to total assets, 0.59%, is close to his estimate of introduced a five-factor model, which builds on
0.52%. Thus, we conclude not that his measure per- their original three-factor model:
forms poorly given our different sample and design
but, rather, that cash-based measures perform even Pi ,t + 1 = α i + β1iMKTRFt + β2i SMBt + β3iHML t
better. + β4iRMWt + β5i CMA t + εi ,t + 1, (2)
t = 1,...,N.
Recently, Harvey, Liu, and Zhu (2016) presented a
critique of empirical studies purporting to uncover The two additional factors are RMW, which is the
new factors. They highlighted data-mining con- average return on the robust operating profit-
cerns, suggesting that the threshold of reported ability portfolios minus the average return on the
t-statistics should be raised far above the traditional weak operating profitability portfolios, and CMA,
2.0 level to around 3.0. They noted, however, that which is the average return on the conservative
factors developed from first principles rather than (low) investment portfolios minus the average
as a purely empirical exercise arguably have a lower return on the aggressive (high) investment port-
threshold. Note that in our Panel C results, all the folios. They argue that this model is better than
cash flow measures with total assets as the denomi- their three-factor model at capturing size, value,
nator comfortably exceed the 3.0 level, as does the profitability, and investment patterns. They also
operating profit measure but not the gross profit argue that if exposure to the five factors captures
and net income measures. For measures with MVE all the variation in expected returns, the intercept
in the denominator, the t-statistics of the cash flow (alpha) should be zero.
measures are generally above 2.0 but below 3.0,
whereas the t-statistics of all the profitability mea- The addition of the RMW factor is quite relevant to
sures are below 2.0. We argue that our cash flow our study because we argue that there is informa-
measures are derived from first principles based on tion in cash-based measures that is not captured
the direct cash flow measure. in accounting-based profitability measures. If our
results are robust and if we find positive and signifi-
As noted by Fama and French (1992) and BGLN cant alphas for our P10–P1 long–short portfolios,
(2015), the relationship between the various we can confidently conclude that cash-based
measures of return on assets and yield and future variables matter more than profitability measures.
stock returns can be explained by either rational In fact, that is what we found, in results reported in
or irrational asset pricing stories. Mispricing can be Table 6. In measures relative to total assets, all the
explained by the behaviors and biases of irrational cash-related measures of alpha remain positive and
investors or short-term trading frictions; alterna- significant, as in the case of the Fama–French three-
tively, there may be risk factors not captured by the factor model (Table 6 repeats the results from Table
three factors in our regression model. Persistence 5 for comparison [FF3 α]). The operating profit
of returns, discussed in the next section, may be measure is also significant but is smaller than all
consistent with a rational explanation if we assume the cash-related measures of alpha (although very
that trading frictions should not persist over longer close to the indirect method measure). Neither the
periods. In the next section, we also consider a five- gross profit measure of alpha nor the net income
factor model that captures more risk factors. measure of alpha is significant. In measures relative
to market value of equity, the alphas for the cash
measures (except CFO) are higher than the alphas
Robustness Checks for the profitability measures and are marginally
In presenting our robustness checks, we re-examine significant, whereas none of the alphas for the
P10–P1 return alphas using the Fama–French five- profitability measures are significant. Thus, on the
factor (instead of three-factor) model regressions, basis of our cash-sorted measures, exposure to the
consider extended-horizon returns, and repeat our five factors does not capture all the variation in
analysis but with sector-neutral P10–P1 returns. expected returns.23

Volume 73 Number 1 cfapubs.org 89

Financial Analysts Journal | A Publication of CFA Institute

Table 6. Fama–French Five-Factor Model Regression Results for Return and Yield Measures,
October 1994–December 2013 (t-statistics in parentheses)

MKTRF Adjusted
Dependent Variable FF3 α (%) α (%) (%) SMB (%) HML (%) RMW (%) CMA (%) R2

CFAFAT/TA 0.91 0.70 –0.179 –0.187 –0.352 0.313 0.251 0.141

(3.16) (2.28) (–2.22) (–1.64) (–2.08) (1.72) (1.14)
CFAF/TA 1.11 0.78 –0.145 –0.178 –0.450 0.503 0.348 0.194
(3.90) (2.63) (–1.78) (–1.56) (–2.82) (2.88) (1.68)
CFO/TA 1.07 0.74 –0.163 –0.161 –0.394 0.490 0.355 0.194
(3.69) (2.48) (–1.97) (–1.43) (–2.44) (2.87) (1.74)
CFIM/TA 0.90 0.62 –0.257 –0.292 –0.527 0.433 0.241 0.265
(3.12) (2.08) (–3.15) (–2.83) (–3.41) (2.79) (1.20)
OP/TA 0.94 0.62 –0.231 0.038 –0.925 0.716 –0.071 0.398
(3.64) (2.48) (–3.41) (0.38) (–7.40) (4.72) (–0.41)
GP/TA 0.59 0.31 0.038 0.255 –0.561 0.506 0.154 0.185
(2.56) (1.29) (0.62) (3.28) (–4.70) (3.96) (0.89)
NI/TA 0.49 0.07 –0.279 –0.136 –0.352 1.006 –0.227 0.514
(1.78) (0.31) (–3.85) (–1.36) (–2.57) (7.03) (–1.13)
CFAFAT/MVE 0.56 0.41 0.080 0.161 0.243 0.123 0.351 0.143
(2.27) (1.56) (1.12) (1.46) (1.62) (0.88) (1.79)
CFAF/MVE 0.61 0.45 0.094 0.109 0.262 0.141 0.360 0.165
(2.51) (1.73) (1.46) (1.10) (1.93) (1.03) (1.78)
CFO/MVE 0.31 0.16 0.116 0.141 0.553 0.139 0.321 0.292
(1.21) (0.57) (1.52) (1.26) (4.02) (0.93) (1.54)
CFIM/MVE 0.67 0.42 –0.015 0.142 0.260 0.344 0.292 0.219
(2.68) (1.63) (–0.20) (1.47) (2.00) (2.62) (1.57)
OP/MVE 0.25 0.10 0.214 0.269 1.180 0.295 0.010 0.413
(0.79) (0.31) (1.97) (1.69) (5.63) (1.44) (0.03)
GP/MVE 0.04 –0.16 0.251 0.549 1.108 0.311 0.185 0.428
(0.13) (–0.46) (2.29) (3.45) (5.24) (1.83) (0.69)
NI/MVE 0.52 0.19 –0.171 0.114 0.441 0.864 –0.311 0.399
(1.74) (0.65) (–2.00) (0.89) (2.93) (4.36) (–1.24)

Extended-Horizon Returns. We re-examined significantly positive for all the cash flow return
our one-month-ahead value-weighted return measures as well as the operating profit (OP/TA)
results by considering a variety of extended- and gross profit (GP/TA) measures, whereas the
horizon holding periods, including 3 months, 6 traditional measure of return on assets (NI/TA)
months, and 12 months. The results are gener- is not significant. Return differences are highest
ally consistent with the one-month-ahead return for the DMCF measures, followed by the GP/TA
results. The three-month return differences are measure and then the indirect method cash flow

90 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

measures. The three-month return difference for that longer holding periods also reduce turnover
the CFAF/TA measure is 2.1% (8.7% annualized). and thus transaction costs. We attempted to
For the yield measures, all the return differences quantify the order of magnitude of these costs
are significantly positive, with a similar order of by examining the turnover related to the various
magnitude for the CFAF/MVE, CFIM/MVE, OP/ measures and horizons. For one-month holding
MVE, and GP/TA measures, whereas the earn- periods, the average two-way turnover across the
ings yield, NI/MVE, is only marginally significant. four cash flow measures is 19% with total assets
The three-month return difference for the CFAF/ in the denominator and 25% with market value of
MVE measure is 2.1% (8.6% annualized). Among equity in the denominator. Across the three profit-
the return-on-assets measures, the information ability measures, the average monthly turnover is
ratio is highest for the GP/TA measure given the 10% with total assets in the denominator and 24%
much lower standard deviation of returns. Among with market value of equity in the denominator.
the yield measures, the information ratio is much For annual holding periods, the average two-way
higher for the cash flow measures than for the turnover across the four cash flow measures is
profitability measures. 113% with either total assets or market value of
equity in the denominator. Across the three profit-
For the six-month horizon, the return differences ability measures, the average annual turnover
for the return-on-assets measures are again is 78% with total assets in the denominator and
significantly positive in all cases except the NI/TA 100% with market value of equity in the denomi-
measure. The highest return difference is for the nator. Following Leclerc, L’Her, Mouakhar, and
CFAF/TA variable, with a six-month return of 3.4% Savaria (2013), we assumed two-way rebalancing
(6.9% annualized). Among the yield measures, the costs of 0.882 bps per 1% of turnover (recall that
CFIM/MVE and GP/MVE six-month returns are our sample highlights the largest and generally
almost identical, at around 3.8% (7.7% annual- most liquid stocks). Across the cash flow measures,
ized). For the return-on-assets measures, the net monthly returns would be reduced by approxi-
information ratio is highest for the GP/TA mea- mately 0.17% with total assets in the denomina-
sure, followed by the various cash flow measures; tor and by 0.22% with market value of equity in
among the yield measures, the cash flow measures the denominator. Across the three profitability
dominate on the basis of the information ratio. measures, net monthly returns would be reduced
by 0.09% with total assets in the denominator
The results for the 12-month returns are similar
and by 0.21% with market value of equity in the
to the results for the 6-month returns. Among
denominator. For the cash flow measures, net
both the return-on-assets measures and the yield
measures, the DMCF measures exhibit superior annual returns would be reduced by approximately
0.99% with total assets in the denominator and by
information ratios compared with the indirect
1.00% with market value of equity in the denomi-
measures. Most of the return differences of the
nator. Across the three profitability measures, net
DMCF and indirect method return-on-assets
annual returns would be reduced by 0.69% with
measures are generally superior to those of the
total assets in the denominator and by 0.88% with
profitability measures, which in turn are superior
market value of equity in the denominator. Even
to the return differences of the NI/TA and NI/MVE
after rebalancing costs, our results are robust,
measures. Even for the 12-month horizon, the
with cash flow–based measures providing higher
high–low CFAF/MVE portfolio offers returns of
returns than profitability measures. We also found
almost 7.5%. Overall, our extended-horizon results
that monthly versus annual rebalancing leads to
are consistent with our earlier results and also
superior net returns.
suggest, though still positive and significant, only
a small tapering of return differences for longer
Sector Analysis. To check whether our results
horizons.
are sector specific, we performed a number of
Although longer holding periods result in slightly tests. First, we segregated our sample among the
lower returns across all measures, the trade-off is 10 Global Industry Classification Standard (GICS)

Volume 73 Number 1 cfapubs.org 91

Financial Analysts Journal | A Publication of CFA Institute

sectors: energy, materials, industrials, consumer exceptions. For the return-on-assets results, we see
discretionary, consumer staples, health care, that cash flow measures tend to have higher return
financials, information technology, telecommunica- differences and information ratios than profitability
tion services, and utilities (recall that financials are measures, except that the NI/TA measure does very
not included in our previous results but are included well. Interestingly, neither the OP/TA return differ-
here because results are segmented by industry). ence nor the GP/TA return difference is significant,
We estimated information ratios for each measure suggesting that results highlighting these variables
within each sector. On the basis of untabulated in other studies may be benefiting from a long posi-
results, we found no consistent pattern across all tion in stocks in one industry and a short position
sectors but can summarize our results as follows. in another industry. Among the yield measures, the
For the measures with total assets as the denomi- only marginally significant results are for the GP/
nator, cash flow measures tend to do better than MVE measure, whereas all the other return differ-
the various profitability and earnings measures, ences are significant.
particularly in the energy and industrials sectors
but also in materials, consumer discretionary, health For our final sector robustness check, Panel B of
care, telecom, and financials. Information ratios are Table 7 reports sector-neutral portfolio results for
strong across all measures for telecom companies, the regression analysis. These results are similar to
with the operating profitability measure doing the those in Table 5, with exceptions similar to those
best. Information ratios are low across all measures uncovered in Panel A. The OP/TA and GP/TA
in consumer staples (except NI/TA) and utilities. For alphas are lower than those of the corresponding
the measures with market value of equity as the cash-based measures. For the yield measures, the
denominator, cash flow measures tend to do better OP/MVE and GP/MVE alphas are not significant.
than the various profitability and earnings mea- Thus, it appears that our results are not being
sures, particularly in the industrials, consumer dis- driven by sector-specific performance. Moreover,
cretionary, and information technology sectors but for our sample, sorting on operating profit and
also in consumer staples and telecom. Profitability gross profit in a sector-neutral setting is not as
measures tend to do better in the energy, materials, strong as cash flow sorting.
and health care sectors. Information ratios are low
across all measures in utilities, and only NI/MVE
does well in the financials sector.
Conclusion
Inspired by Novy-Marx (2013), much recent atten-
Table 7 reports another set of robustness checks tion has been focused on the importance of income
regarding the sectors in which stocks in the long– statement–related measures, such as gross profit
short portfolio are invested. It is possible that the to total assets, as predictors of the cross section of
composition of the lowest-return portfolio based returns. In this article, we showed that commonly
on sorting, P1, might be very different from the used income statement–related metrics, including
composition of the highest-return portfolio based return on assets and earnings yield, have some pre-
on sorting, P10, in terms of the sectors in which the dictive power, but in general, cash-based measures
stocks are invested. To control for this possibility, are superior to measures of operating profitability
we created sector-neutral portfolios. We repeated and gross profitability to total assets. We examined
our sorting procedure within each of 20 of the 24 strategies that suggest superior risk-adjusted returns
GICS industry groups (recall that financials—banks, from using transformed financial statements with
insurance companies, diversified financials, and our DMCF to estimate free cash flows. We created
REITs—are omitted from our sample). P1 (P10) is a direct cash flow template by segregating operating
populated by the lowest-performing (highest-per- cash flows from financing, tax, nonoperating, and
forming) stocks by decile within each sector. investing activites. We argue that this segregation
helps investors understand how value is created
Panel A of Table 7 presents the one-month-ahead and thus better estimate the true intrinsic value
value-weighted return results, which are generally of a company’s equity. We found that both DMCF
consistent with those in Table 2, with some notable

92 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

Table 7. Sector-Neutral Results for Return and Yield Measures, October 1994–December
2013

Measure P10–P1 (%) Std. Dev. t-Stat. Min. (%) Max. (%) IR

A. One-month-ahead return
CFAFAT/TA 0.606 3.985 2.31 –16.943 14.375 0.527
CFAF/TA 0.661 4.233 2.37 –17.556 14.866 0.541
CFO/TA 0.693 4.104 2.57 –17.547 13.238 0.585
CFIM/TA 0.525 4.210 1.90 –18.022 10.631 0.432
OP/TA 0.340 4.272 1.21 –15.177 18.118 0.276
GP/TA 0.203 3.436 0.90 –12.193 11.459 0.204
NI/TA 0.644 4.033 2.43 –10.512 12.615 0.553
CFAFAT/MVE 0.523 3.411 2.33 –10.482 13.318 0.532
CFAF/MVE 0.619 3.490 2.70 –11.953 13.887 0.615
CFO/MVE 0.774 3.618 3.25 –10.206 11.792 0.741
CFIM/MVE 0.504 3.461 2.21 –9.476 13.487 0.504
OP/MVE 0.868 5.448 2.42 –15.708 30.744 0.552
GP/MVE 0.592 5.025 1.79 –15.73 17.600 0.408
NI/MVE 0.567 3.768 2.29 –9.457 15.645 0.521

Average
Dependent Variable Return (%) α (%) MKTRF (%) SMB (%) HML (%) Adjusted R2

B. Regression analysis (t-statistics in parentheses)

CFAFAT/TA 0.606 0.90 –0.185 –0.248 –0.465 0.185
(2.31) (3.69) (–2.90) (–1.94) (–4.18)
CFAF/TA 0.661 0.95 –0.184 –0.246 –0.471 0.166
(2.37) (3.69) (–2.64) (–1.86) (–3.83)
CFO/TA 0.693 0.99 –0.186 –0.252 –0.456 0.172
(2.57) (3.91) (–2.72) (–2.05) (–4.03)
CFIM/TA 0.525 0.89 –0.153 –0.462 –0.618 0.312
(1.90) (3.69) (–2.49) (–5.74) (–6.09)
OP/TA 0.340 0.64 –0.324 0.112 –0.535 0.234
(1.21) (2.57) (–5.47) (1.05) (–5.18)
GP/TA 0.203 0.41 –0.228 0.067 –0.338 0.156
(0.90) (1.98) (–4.62) (0.98) (–4.24)
NI/TA 0.644 0.97 –0.256 –0.427 –0.238 0.236
(2.43) (3.98) (–4.12) (–5.59) (–2.27)
CFAFAT/MVE 0.523 0.66 –0.150 –0.231 0.097 0.137
(2.33) (3.02) (–2.33) (–2.80) (1.13)
CFAF/MVE 0.619 0.72 –0.161 –0.202 0.209 0.172
(2.70) (3.31) (–2.68) (–2.82) (2.82)
(continued)

Volume 73 Number 1 cfapubs.org 93

Financial Analysts Journal | A Publication of CFA Institute

Table 7. Sector-Neutral Results for Return and Yield Measures, October 1994–December
2013 (continued)

Average
Dependent Variable Return (%) α (%) MKTRF (%) SMB (%) HML (%) Adjusted R2

CFO/MVE 0.774 0.78 –0.065 –0.118 0.283 0.110

(3.25) (3.26) (–0.92) (–1.54) (3.30)
CFIM/MVE 0.504 0.71 –0.197 –0.380 0.068 0.262
(2.21) (3.42) (–3.19) (–5.42) (0.88)
OP/MVE 0.868 0.39 0.261 0.339 0.913 0.315
(2.42) (1.34) (3.22) (2.83) (6.26)
GP/MVE 0.592 0.14 0.163 0.440 0.982 0.421
(1.79) (0.53) (2.45) (5.27) (9.17)
NI/MVE 0.567 0.66 –0.102 –0.285 0.182 0.142
(2.29) (2.76) (–1.57) (–3.68) (2.10)

return-on-assets measures and DMCF yield mea- income statement profitability measures. Although
sures are generally better at predicting stock returns “quality investing” strategies based on various
than indirect method cash flow measures, which fundamental criteria have recently gained notori-
in turn are generally superior to income statement ety, our study suggests that analysts and investors
measures based on profitability. After controlling for would be better off following the cash.
well-known risk factors, we found similar results. We
also tested various holding periods and performed
robustness checks that controlled for sector perfor- Appendix A. Comparing Income
mance, finding similar results. Our Fama–MacBeth
regressions suggest the incremental importance Statements with Both Indirect and
of information on taxes, capital expenditures, and Direct Cash Flow Statements
operating cash flows. Future research may provide Corporations can present their net cash flow from
further explanations for why those variables are operating activities by using either the direct or the
important in forecasting stock returns. indirect method. Under each method, net cash flow
from operating activities will be the same. However,
Our conclusion—that there is incrementally better
the direct method presents the specific amounts of
information in segregating cash flows to account
cash received and paid for each significant busi-
for both investing activities and taxes, in addi-
ness activity and the resulting net cash flow from
tion to operating cash flow information—lends
operating activities. We argue that by using the
support to the initiatives of the International
structure of the income statement and the informa-
Accounting Standards Board (IASB) and the US
tional content of the cash flow statement, investors
Financial Accounting Standards Board (FASB) to
are better able to isolate recurring cash flows that
require reconstructed financial statements. From
add to net present value, segregate financing activi-
an investment perspective, investors may be able
ties that do not add to net present value, and flag
to obtain better information about investment
one-off gains/losses.
prospects—and thus future stock returns—by rely-
ing on cash flows that disaggregate operating cash The indirect method derives net cash from operat-
flows from financing, tax, investing, nonoperating, ing activities without separately presenting any
and nonrecurring activities rather than relying on of the operating cash receipts and payments. The

94 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

indirect approach merely reconciles information between net income and working capital changes,
between the income statement and actual cash as shown in Table A3.
flows. In this way, the indirect statement is not as
intuitive as the direct statement. In addition, it is Under both methods, net cash from operating
difficult to forecast corporate results that are pre- activities is –$102. The direct method, however,
sented “net” of operating, financing, and tax effects. shows specific amounts of cash receipts and
payments segregated by major business activ-
To highlight some of the key concepts and illus- ity. The indirect method merely reconciles net
trate how operating cash flows are reported using income to the operating cash flows. Users of the
both the direct and the indirect methods, we begin indirect statement may not fully appreciate that
with XYZ Corporation’s income statement and business activities (prefinancing, pretax) generated
balance sheet amounts, shown in Table A1. –$150, whereas a tax benefit of $90 minimized
the overall cash outflows. Moreover, although the
Under the direct method, net operating cash flows direct statement links working capital gains/losses
are calculated by using specific cash inflows and to their respective source accounts (accounts
outflows related to the income statement and receivable to sales, accounts payable to cost of
balance sheet working capital accounts, as shown goods sold), the indirect statement relates these
in Table A2. working capital changes to net income, leaving
the user with the cumbersome task of sorting and
The information presented here depicts gross cash
rearranging. Thus, investors have three different
inflows and outflows in a straightforward manner,
measures to extrapolate: net income ($148), direct
along with an undistorted view of accrual revenues
business cash flows (–$150), and net operating
and costs versus cash inflows and outflows. For
cash flows (–$102).
example, although XYZ booked $1,000 in sales,
only $800 was received in cash over the period
and the balance was sold on credit. Cost of goods Editor’s Notes
sold, operating expenses, and working capital needs This article was externally reviewed using our double-blind
totaled $850. On an operating cash flow basis, peer-review process. When the article was accepted for
publication, the authors thanked the reviewers in their
the company suffered a –$150 loss for the period.
acknowledgments. Andrew L. Berkin and Heiko Jacobs
Recall that net income was reported as +$148. As a were the reviewers for this article.
result of interest expense and a favorable tax effect,
Submitted 2 October 2015
net operating cash flow was only –$102.
Accepted 31 May 2016 by Stephen J. Brown
Under the indirect method, net operating cash
flows are calculated by using the differences

Table A1. XYZ Corporation Income Statement and Balance Sheet Amounts

Income Statement Amounts Balance Sheet Amounts

Sales $1,000 Change in accounts receivable (A/R) +$200

– Cost of goods sold 700 Change in inventory +100
– Operating expenses 100 Change in accounts payable (A/P) operations –50
– Depreciation 10 Change in accounts payable (A/P) taxes +90
– Interest expense 10
+ Interest income 3
– Corporate taxes 35
     Net income $148

Volume 73 Number 1 cfapubs.org 95

Financial Analysts Journal | A Publication of CFA Institute

Table A2. Calculation of XYZ Corporation’s Net Operating Cash

Flow (Direct Method)

Cash inflows
Sales $1,000
Change in A/Ra –200
Gross cash inflows $800

Cash outflows
Cost of goods sold –$700
Operating expenses –100
Change in inventory –100
Change in A/P operationsb –50
Gross cash outflows –950
Direct business cash flows –$150

Financing
Interest expense –$10
Interest income +3
Financing outflows –$7

Corporate taxes
Income statement taxes –$35
Change in A/P taxes +90
Net tax cash flows +$55

Net cash from operating activities –$102

aWhere A/R increases, cash has yet to be received.

bWhere A/P decreases, cash was paid.

Table A3. Calculation of XYZ Corporation’s Net Operating Cash

Flow (Indirect Method)

Net income $148

Add/subtract noncash adjustments:
Depreciation +10
Change in A/R –200
Change in inventory –100
Change in A/P operations –50
Change in A/P taxes +90
Net cash from operating activities –$102

96 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

Notes
1. See Sloan (1996) and Markham (2006). See also, more gen- Anecdotally, Apple had zero net borrowing from Q1 2005
erally, American Institute of Certified Public Accountants to Q1 2013 and then incurred $17 billion in net borrowing
(1973); a committee chaired by Robert Trueblood urged in Q2 2013.
the accounting profession to be more responsive to
investors and creditors in providing information about 6. In practice, companies often group items that are different
the companies they audit and to show less deference to in nature and that respond differently to the same eco-
corporate management. nomic events. For example, fixed-cost lease payments ver-
sus variable labor expenses are aggregated and presented
2. The accounting standard for presenting the statement as part of selling, general, and administrative expenses
of cash flows is documented in FASB (1987, p. 6) as FAS (SG&A). If lease payments are not functionally related to
No. 95, which states that “the information provided in revenues but labor costs are, investors are better informed
a statement of cash flows . . . should help investors . . . by presenting the information separately.
assess the enterprise’s ability to generate positive future
net cash flows.” In an exposure draft (FASB 2008), the 7. See Graham, Harvey, and Rajgopal (2005); for abnormal
International Accounting Standards Board (IASB) and the accruals, see Healy and Wahlen (1999).
Financial Accounting Standards Board (FASB) examined
the use of the direct cash flow method for measuring a 8. Our study also complements previous studies that investi-
company’s operating performance as opposed to the com- gated cash flows and investments (e.g., Hackel and Livnat
monly used indirect method that starts with net income 1992; Lakonishok, Shleifer, and Vishny 1994; Hackel,
and makes adjustments. According to FASB (2008, p. Livnat, and Rai 1994), accruals and investments (e.g., Sloan
45), the indirect approach includes a major deficiency: “It 1996; Xie 2001; DeFond and Park 2001; Thomas and
derives the net cash flow from operating activities without Zhang 2002; Richardson, Sloan, Soliman, and Tuna 2005;
separately presenting any of the operating cash receipts Livnat and Santicchia 2006), and cash flows, accruals, and
and payments.” It merely reconciles information and is not investments (e.g., Houge and Loughran 2000; Livnat and
a valid substitute for reporting operating cash inflows and López-Espinosa 2008).
outflows. For another overview of the proposed changes,
see Reilly (2007). 9. Subsequent to earlier drafts of this article, a first draft of
Ball et al. (2016) appeared on SSRN (http://papers.ssrn.
3. For example, net income can include results from primary com/sol3/papers.cfm?abstract_id=2587199). Similar to
business operations, different products or jurisdictions, what we show in this article, they showed that a cash-
proceeds from the sale of assets, discontinued operations, based operating profitability measure is a strong predictor
or investment income, among others. Each earnings source of the cross section of returns. Their measure is similar to
can respond differently to the same economic factors. our measure of net cash flows from operations. They did
Likewise, expenses associated with raw materials, direct not adjust their cash flow measures for capital expendi-
labor, transportation, and energy costs are not segregated tures, and they focused on accruals.
in the indirect cash flow statement, preventing inves-
tors from understanding how economic circumstances 10. In addition, financial companies face unique accounting
might affect forecasted results and thus the relationship regulations, including how investments are classified,
between cash inflows and outflows. More generally, the which can materially affect operating cash flows.
difference between net income and cash earnings has
11. The four-month lag is conservative. In our untabulated
spawned considerable academic interest in the area of
analysis of S&P 1500 companies between 1993 and 2013,
“quality of earnings” and has contributed to investors’
99% of them filed 10-Q statements within three months
renewed interest in cash flow statements and cash flow–
after the quarter-end.
related measures. See O’Glove (1987); Kellogg and Kellogg
(1991); Siegel (1991).
12. For a discussion of the advantages and disadvantages
of the Fama–MacBeth approach compared with sorting
4. FCFE = Cash flow from operations + Depreciation –
returns, see Fama and French (2006, 2008).
Capital expenditures, or Earnings before interest and
taxes, × (1 – Tax rate) + Depreciation – Interest expense ×
13. In their Fama–MacBeth analysis over a longer period
(1 – Tax rate) +/– Working capital changes – Capital expen-
(1963–2014), BGLN (2016) found that their cash-based
ditures. Some believe that free cash flow to equity should
operating profitability measure is significant whereas their
include debt, both preferred and equity. Our primary
accounting-based operating profitability measure is not.
purpose in measuring free cash flows, as distinct from
owner sources of financing, was to measure competitive 14. Thomas and Zhang (2014) offered a variety of explana-
advantage and assess the probability of recurrence. The tions for why some studies have found significant positive
same cannot be said of external financing. relationships and others have found significant negative
relationships.
5. In an unreported analysis, we found that our direct cash
flow measures are superior to FCFE measures, particularly 15. In untabulated results, we observed that corporate cash
those FCFE measures that include any new debt, which taxes paid were consistently lower than the income
can add a nonoperational and sporadic component. statement tax account. Through tax planning and timing

Volume 73 Number 1 cfapubs.org 97

Financial Analysts Journal | A Publication of CFA Institute

differences between capital expenditures and deprecia- period (1994–2013) compared with his 1963–2010
tion, deferred tax liabilities are typically generated but are study; it is possible that the profitability measure has
paid at a later date. weakened more recently. Moreover, our sample of stocks
is different. He used the universe of Compustat stocks
16. See Baker, Stein, and Wurgler (2003); Titman, Wei, and Xie (excluding financials), whereas we focused on the largest
(2004, 2013); Anderson and Garcia-Feijóo (2006). 1,500 stocks, which is a more liquid and tradable sample.
Although Novy-Marx considered an estimate of free
17. We also repeated our analysis with returns measured using cash flows in part of his analysis, he defined free cash
equal weights and separately with financial companies in flow as net income plus depreciation minus working
the sample. The results are substantively similar. capital change minus capital expenditures. Interestingly,
in Appendix Table A2 in Novy-Marx’s paper, his EBITDA-
18. Although the P10–P1 return difference is small for NI/ to-assets variable—most similar to some of our DMCF
TA, it is worth noting that the P1 return is quite high (as is measures except that it does not account for capital
the P1 return for the NI/MVE variable). This result may be expenditures—performs well, even better than his free
driven by negative-earnings companies with strong one- cash flow measure.
month-ahead returns.
21. Data are available from Ken French’s website (http://
19. In unreported results, we divided our sample into four mba.tuck.dartmouth.edu/pages/faculty/ken.french/
subperiods. Subperiod 1 (October 1994–February 2000; data_library.html). We also examined a model with two
65 months) coincides with the “technology bubble” additional factors: UMD is the up minus down (price
period, with a cumulative stock market return, as momentum) factor (also available from Ken French’s
measured by the S&P 500 Composite Index total return, website), and LIQ is Pástor and Stambaugh’s (2003) traded
of 109.5% (1.1% a month, compounded); subperiod 2 liquidity factor, derived from dividing US common stocks
(March 2000–July 2007; 89 months) coincides with into 10 portfolios on the basis of each stock’s sensitivity
the post–technology bubble/pre–financial crisis period, to the nontraded liquidity innovation factor—the return on
with a cumulative stock market return of 20.3% (0.2% the portfolio of high-liquidity betas less the return on the
a month); subperiod 3 (August 2007–May 2009; 22 portfolio of low-liquidity betas. For these two factors, the
months) coincides with the financial crisis period, with results are qualitatively the same.
a cumulative stock market return of –34.0% (–1.9% a
month); and subperiod 4 (June 2009–December 2013; 55 22. In addition to some sampling differences in Novy-Marx’s
months) coincides with the post–financial crisis period, (2013) Table 2, he reported quintile differences rather than
with a cumulative stock market return of 121.6% (1.5% a decile differences.
month). Cash flow–based measures tend to dominate the
profitability measures in all but the third subperiod. 23. As a further robustness check, we replaced the Fama–
French five-factor model with the Hou et al. (2016)
20. In comparing our results with those of Novy-Marx (2013), q-model variables and found that the results are qualita-
we note some sampling differences. Given our data tively the same.
limitations, we focused on a shorter and more recent

References
American Institute of Certified Public Accountants. 1973. Fama, Eugene F., and Kenneth R. French. 1992. “The Cross-
“Objectives of Financial Statements.” Report of the Study Section of Expected Returns.” Journal of Finance, vol. 47, no. 2
Group on the Objectives of Financial Statements. (June): 427–465.
Anderson, Christopher, and Luis Garcia-Feijóo. 2006. “Empirical ———. 1993. “Common Risk Factors in the Returns on Stocks
Evidence on Capital Investment, Growth Options, and Security and Bonds.” Journal of Financial Economics, vol. 33, no. 1
Returns.” Journal of Finance, vol. 61, no. 1 (February): 171–194. (February): 3–56.
Baker, Malcolm, Jeremy Stein, and Jeffrey Wurgler. 2003. ———. 2006. “Profitability, Investment and Average Returns.”
“When Does the Market Matter? Stock Prices and the Journal of Financial Economics, vol. 82, no. 3 (December): 491–518.
Investment of Equity-Dependent Firms.” Quarterly Journal of
Economics, vol. 118, no. 3 (August): 969–1005. ———. 2008. “Dissecting Anomalies.” Journal of Finance, vol.
63, no. 4 (August): 1653–1678.
Ball, Ray, Joseph Gerakos, Juhani Linnainmaa, and Valeri
Nikolaev. 2015. “Deflating Profitability.” Journal of Financial ———. 2015. “A Five-Factor Asset Pricing Model.” Journal of
Economics, vol. 117, no. 2 (August): 225–248. Financial Economics, vol. 116, no. 1 (April): 1–22.

———. 2016. “Accruals, Cash Flows, and Operating Fama, Eugene F., and James D. MacBeth. 1973. “Risk, Return,
Profitability in the Cross Section of Stock Returns.” Journal of and Equilibrium: Empirical Tests.” Journal of Political Economy,
Financial Economics, vol. 121, no. 1 (July): 28–45. vol. 81, no. 3 (May–June): 607–636.

DeFond, Mark, and Chul Park. 2001. “The Reversal of FASB. 1987. “Statement of Financial Accounting Standards No.
Abnormal Accruals and the Market Valuation of Earnings 95: Statement of Cash Flows.” Financial Accounting Standards
Surprises.” Accounting Review, vol. 76, no. 3 (July): 375–404. Board (November).

98 cfapubs.org First Quarter 2017

 Are Cash Flows Better Stock Return Predictors Than Profits?

———. 2008. “Preliminary Views on Financial Statement Markham, Jerry. 2006. A Financial History of Modern U.S.
Presentation.” Financial Accounting Series Discussion Paper No. Corporate Scandals: From Enron to Reform. Armonk, NY: M.E.
1630-100, Financial Accounting Standards Board (16 October): Sharpe.
http://fasb.org/draft/DP_Financial_Statement_Presentation.pdf.
Novy-Marx, Robert. 2013. “The Other Side of Value: The
Graham, John R., Campbell R. Harvey, and Shiva Rajgopal. Gross Profitability Premium.” Journal of Financial Economics,
2005. “The Economic Implications of Corporate Financial vol. 108, no. 1 (April): 1–28.
Reporting.” Journal of Accounting and Economics, vol. 40, no.
1–3 (December): 3–73. O’Glove, Thornton L. 1987. Quality of Earnings: The Investor’s
Guide to How Much Money a Company Is Really Making. New
Hackel, Kenneth S., and Joshua Livnat. 1992. Cash Flow and York: Free Press.
Security Analysis. Homewood, IL: Business One Irwin.
Pástor, Luboš, and Robert F. Stambaugh. 2003. “Liquidity Risk
Hackel, Kenneth S., Joshua Livnat, and Atul Rai. 1994. “The and Expected Stock Returns.” Journal of Political Economy, vol.
Free Cash Flow/Small-Cap Anomaly.” Financial Analysts 111, no. 3 (June): 642–685.
Journal, vol. 50, no. 5 (September/October): 33–42.
Reilly, David. 2007. “Profit as We Know It Could Be Lost with
Harvey, Campbell R., Yan Liu, and Heqing Zhu. 2016. “… and New Accounting Statements.” Wall Street Journal (12 May):
the Cross-Section of Expected Returns.” Review of Financial http://online.wsj.com/article/SB117893520139500814.html.
Studies, vol. 29, no. 1 (January): 5–68.
Richardson, Scott A., Richard G. Sloan, Mark T. Soliman, and
Healy, Paul M., and James M. Wahlen. 1999. “A Review of İrem Tuna. 2005. “Accrual Reliability, Earnings Persistence and
the Earnings Management Literature and Its Implications Stock Prices.” Journal of Accounting and Economics, vol. 39, no.
for Standard Setting.” Accounting Horizons, vol. 13, no. 4 3 (September): 437–485.
(December): 365–383.
Siegel, Joel. 1991. How to Analyze Businesses, Financial
Hou, Kewei, G. Andrew Karolyi, and Bong-Chan Kho. 2011. Statements, and the Quality of Earnings, 2nd ed. Englewood
“What Factors Drive Global Stock Returns?” Review of Financial Cliffs, NJ: Prentice Hall.
Studies, vol. 24, no. 8 (August): 2527–2574.
Sloan, Richard G. 1996. “Do Stock Prices Fully Reflect
Hou, Kewei, Chen Xue, and Lu Zhang. 2015. “Digesting Information in Accruals and Cash Flows about Future
Anomalies: An Investment Approach.” Review of Financial Earnings?” Accounting Review, vol. 71, no. 3 (July): 289–315.
Studies, vol. 28, no. 3 (March): 650–705.
Thomas, Jacob, and Frank Zhang. 2014. “Valuation of
———. 2016. “A Comparison of New Factor Models.” Charles Tax Expense.” Review of Accounting Studies, vol. 19, no. 4
A. Dice Center Working Paper No. 2015-05 (June): http:// (December): 1436–1467.
papers.ssrn.com/sol3/papers.cfm?abstract_id=2520929.
Thomas, Jacob K., and Huai Zhang. 2002. “Inventory Changes
Houge, Todd, and Tim Loughran. 2000. “Cash Flow Is King? and Future Returns.” Review of Accounting Studies, vol. 7, no. 2
Cognitive Errors by Investors.” Journal of Psychology and (June): 163–187.
Financial Markets, vol. 1, no. 3–4: 161–175.
Titman, Sheridan, K.C. John Wei, and Feixue Xie. 2004.
Kellogg, Irving, and Loren Kellogg. 1991. Fraud, Window Dressing, “Capital Investments and Stock Returns.” Journal of Financial
and Negligence in Financial Statements. New York: McGraw-Hill. and Quantitative Analysis, vol. 39, no. 4 (December): 677–700.
Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny. ———. 2013. “Market Development and the Asset Growth
1994. “Contrarian Investment, Extrapolation, and Risk.” Journal Effect: International Evidence.” Journal of Financial and
of Finance, vol. 49, no. 5 (December): 1541–1578. Quantitative Analysis, vol. 48, no. 5 (October): 1405–1432.
Leclerc, Frank, Jean-François L’Her, Tammam Mouakhar, and White, Halbert. 1980. “A Heteroskedasticity-Consistent
Patrick Savaria. 2013. “Industry-Based Alternative Equity Covariance Matrix Estimator and a Direct Test for
Indices.” Financial Analysts Journal, vol. 69, no. 2 (March/April): Heteroskedasticity.” Econometrica, vol. 48, no. 4 (May): 817–838.
42–56.
Xie, Hong. 2001. “The Mispricing of Abnormal Accruals.”
Livnat, Joshua, and Germán López-Espinosa. 2008. “Quarterly Accounting Review, vol. 76, no. 3 (July): 357–373.
Accruals or Cash Flows in Portfolio Construction?” Financial
Analysts Journal, vol. 64, no. 3 (May/June): 67–79.
Livnat, Joshua, and Massimo Santicchia. 2006. “Cash Flows,
Accruals, and Future Returns.” Financial Analysts Journal, vol.
62, no. 4 (July/August): 48–61.

Volume 73 Number 1 cfapubs.org 99