Latent Accounting Growth, Corporate Finance Policies, and Return Predictability Suresh Kumar Oad-Rajput a,* , Jianguo Chen a and Udomsak (Jeff) Wongchoti a a School of Economics and Finance, Massey University, Private Bag 11-222, Palmerston North, New Zealand Abstract We examine the interactions of the balance sheet growth information using factor analysis. We find that optimal corporate financing decisions embedded in multiple balance sheet accounts are five latent factors that are fundamental to the business value. The identified decisions are Financial Flexibility, Short-term Credit, Long-term Capital Investment, Convertible Debt Usage, and Preferred Stock Usage. Our evidence suggests that the other observed proxies of the corporate financing decision types suffer from missing variable bias. During the 1985-2009 research period, the new factors are robust predictors of future stock returns, firm profitability, and firm value. Predictability is complementary to the market controls but conflicting with the accounting controls and hold when controlling for global financial crises falling in our sample period. This draft: March, 2013 JEL classification: C38; G12; G32 Keywords: Asset Pricing Models, Multifactor Models, Accounting growth, Factor analysis, Cross-section stock returns, Corporate Finance Policies _____________ * Corresponding author is at: School of Economics and Finance, Massey University, Private Bag 11-222, Palmerston North, New Zealand. Tel.: +64 6 3569099, ext.:7368. E-mail: [email protected]
Transcript
Latent Accounting Growth, Corporate Finance Policies, and
Return Predictability
Suresh Kumar Oad-Rajputa,*
, Jianguo Chena and Udomsak (Jeff) Wongchoti
a
a School of Economics and Finance, Massey University, Private Bag 11-222, Palmerston North, New Zealand
Abstract
We examine the interactions of the balance sheet growth information using factor analysis. We find that
optimal corporate financing decisions embedded in multiple balance sheet accounts are five latent factors
that are fundamental to the business value. The identified decisions are Financial Flexibility, Short-term
Credit, Long-term Capital Investment, Convertible Debt Usage, and Preferred Stock Usage. Our evidence
suggests that the other observed proxies of the corporate financing decision types suffer from missing
variable bias. During the 1985-2009 research period, the new factors are robust predictors of future stock
returns, firm profitability, and firm value. Predictability is complementary to the market controls but
conflicting with the accounting controls and hold when controlling for global financial crises falling in
What are the optimal corporate financing decisions? The investigation of this question is directed to the
large body of literature on investment and financing decision that depart from a perfect capital markets
setting of Modigliani and Miller (1958). A large part of this literature consists of static models that study
the corporate financing decision in a single point in time and dynamic models that give indirect inferences
by studying only outcome of the decision process. Besides, there is some effort targeted at behavioral
aspects (use of survey or clinical methods) and interaction among decision types.1 Overall these models
suggest many forms of the corporate financing decisions, such as operational investments, long-term
investments, debt-to-equity ratio, equity structure issues, and financial flexibility. Of note, these
theoretical frameworks and their supporting empirical evidences are conflicting and even differ from real
world settings. Thus, our understanding remains incomplete and conflicting, and it requires more
attention.
The objective of this paper is to offer new insights into firms’ financing and investment decisions by
recognizing that the outcomes of these decisions are reflected in the shifts of their financial positions (the
balance sheet).2 The real challenge lies in the identification of the interactive relations of the shifts in the
accounting numbers to decipher the optimal number and type of corporate financing decision they
represent. In order to understand more accurately the corporate financing decisions, we seek to address
certain questions. How the balance sheet information content interact? Does this interaction help to
identify some common movements? How to decompose the information content based on observed
1 See, for example, Myers (1974), Jensen and Meckling (1976), Myers (1977), Myers and Majluf (1984), Dotan and
Ravid (1985), Dammon and Senbet (1988), Parrino and Weisbach (1999), Barclay and Smith (1995b), Mehran et al.
(1999), Rajan and Zingles (1995), Mello and Parsons (1992), Mello et al. (1995), Fries et al. (1997), Ericsson
(2000), Mauer and Ott (2000), Goldstein et al. (2001), Parrino et al. (2005), Baker and Wurgler (2002), Ju et al.
(2003), Welch (2004), Morellec (2004), Leary and Roberts (2005), Mauer and Sarkar (2005), Flannery and Rangan
(2006), Brennan and Schwartz (1984), Mauer and Triantis (1994), Leland (1998), Moyen (2007), Titman and
Tsyplakov (2007), La Porta et al. (1998; Porta et al., 1997), Graham et al. (2001), and Tuffano (2001), among
others. 2 refer the Statement of Financial Accounting Concept No. 1 that defines objectives of the financial reporting and
identifies the balance sheet statement being the important feature of the financial reporting.
2
commonalities into basic corporate financing decision types? Finally, how do these decision types align
with the theoretical models?
We depart from the literature that seek to find the determinants of the corporate financing decisions and
provide a way forward to overcome the problem of limited attention that arise with sub-optimal use of
observed accounting information as proxy for certain corporate financing decisions (Hirshleifer et al.,
2004). In fact, we assume that a single or pre-determined lump of accounting items cannot be
representative of certain decision type. Because, when we make a decision and the result of
implementation could be recorded in many accounts of an accounting book. The double entry
bookkeeping makes the information about a certain dimension buried in multiple accounting balances. On
the other hand, one account could be used to keep track of many different business activities. For
example, a cash account balance is influenced by the net business profits, it may also be caused by the net
investment activities. We have also noticed that the double entry accounting rule may produce some
additional information by separating the cash balance into different components. If the component of cash
account change is always moving together with the operating business incomes (Owner’s Equity), it
might be an indication of business profits. And, if another part of the cash balance change is always
moving together with the change of working capital or fixed assets that may be a good signal of
investment expenditure. These types of moving together regularly or most of times may be a good
indication of the business decision types we discussed above. If that is the case then the individual
decision type be represented by multiple accounting items and all items are expected to follow the similar
movements and represent same decision across the time.
In order to find these basic types of business activities we have used an ideal tool which is called factor
analysis. The factor analysis model takes into account the common variance and has been widely used in
finance literature (Abarbanell et al., 2003; Bushee, 1998; Pinches and Mingo, 1973; Sorensen, 2000).3 To
provide further guidance on the use of factor analysis, we know that all of the business activities are
3 See., section 2 for detailed discussion on factor analysis.
3
recorded in a business accounting book, and the accounting balance changes contain all of the business
information. If we found that a group of accounting ratios are stably (in a reasonably long period) moving
together, we may categorize this group as the result of a certain type of business decision. Factor analysis
is such a mathematical technique to identify the “most” important common movements among a long list
of variables.
We see no data mining concern in this research due to manifold reasons: First, we only apply factor
analysis on the balance sheet growth variables (X terms) not on market values or potential dependent
variables (Y terms). Second, the factor analysis is used to produce annual factors instead of whole period
factors that remove any data mining concerns in subsequent analysis that use the identified decision types
as independent variables. Third, subsequent return predictability test, the models are built on publicly
available information about all common stocks. Finally, there is no selection bias in selecting the balance
sheet growth items as inputs to the model as we follow strictly the basic assumptions of the model.4
One of the key findings to use of factor analysis on all the accounting ratios shown in the balance sheet
and in supplementary balance sheet statements for a period of 1985 to 2009 is that five factors (common
movements) contain about 70% of all of the accounting ratio changes.5 In another word, the information
shown by the five resulting factors is equivalent to the information shown in whole balance sheet ratios
(plus some supplementary statements). Making the result more interesting is the resulting five factors are
well aligned with the major corporate financing decisions shown in prior theoretical and empirical
literature. The identified decision types are Financial Flexibility (FIN_FLEX), Short-term Credit
(ST_CREDIT), Long-term Capital Investment (LT_INV), Convertible Debt Usage (CVT_DEBT) and
Preferred Stock Usage (PSK_USE).6
4 See., section 2 for detailed discussion on research method.
5 We convert the level balance sheet items to growth variables for our research following Chen et al. (2011) and
Lyandres et al. (2008). 6 For simplicity, in subsequent discussion the identified coporate decision types are written as the ‘new factors’.
4
We label an individual factor based upon the first several accounting ratios included in the list of the
factor, it is highly possible that the other less important (less loads) accounting ratios have the function
finely tuning the measure of the particular factor. The argument is supported by the observation that most
accounting ratios included in a factor could easily be explained by the factor decision. It is also supported
by the factor analysis that the annual factors produced are very stable, the top five components of first and
second factor are always same and top three components of other factors are always same in nature
throughout the whole sample period. The only difference is the ranking position of the second to fifth
factor could be different, the first factor is always the same.7
Specifically, FIN_FLEX comprises mainly of shifts in common equity financing along with cash and
equivalents and invested capital. The components of FIN_FLEX proxy for both investment and financing
flexibility, the close substitutes (Gamba and Triantis, 2008) and offer an integrated measure of financial
flexibility.8 The ST_CREDIT is comprised of shifts in short-term financing (current liabilities and debt in
current liabilities) that proxy for the firm’s operating business growth. The LT_INV factor consists of
changes in the long-term capital investment (e.g. Fixed assets, long-term debt, and deferred taxes and
investment credit) and are considered to be the strongest investment growth measures (Cooper et al.,
2008; Titman et al., 2010). The next factor, CVT_DEBT represents the shifts in the convertibility of bond
and equity and may proxy for firms’ operational risk. The firms with high agency cost of debt due to the
presence of long-term debt and lower credit quality use these securities (Bodie and Taggart, 1978; Doukas
and Pantzalis, 2003). Finally, PSK_USE refers to the shifts in issuance of preferred stock, the measure
may relate to firms with poor profitability and high debt ratios (Howe and Lee, 2006). It is possible that
firms issue preferred stocks to avoid common equity issuance or breach their debt capacity.
7 See, section 3 for detailed discussion on the new five factors.
8 Denis (2011) argues that “financial flexibility refers to the ability to respond in a timely and value-maximizing
manner to unexpected changes in the firm’s cash flows or investment opportunity set.” Marchica and Mura (2010)
argue “since there is no well-defined measure of flexibility in the literature, this is an unobservable factor that
depends largely on managers’ assessment of future growth options.”
5
To provide further guidance on the role of the new factors, we examine separately the relation between
these factors with future stock returns, firm value, and firm profitability. We extend the particular line of
research that document that accounting variable can predict the stock returns and offer a number of
interesting results.9 This large body of literature documents that on aggregate the accounting variables
negatively predict stock returns and the firms that invest more earn lower future stock returns (Baker et
al., 2003; Polk and Sapienza, 2009; Titman et al., 2004). The focus of our empirical analysis is not about
the potential explanations of negative relation or earning lower stock returns, but mainly on the ability of
our new factors for return predictability. It is critical to note that the above factors are not governed by
stock returns in their construction. We also make sure that the information about the new factors is
available prior the return predictability tests.
This article uses data from listed common stocks in the U.S. from 1985 to 2009 to run Fama and MacBeth
(1973) cross-sectional regressions and finds that for the whole sample except PSK_USE all new factors
show the significant negative relation with the future stock returns. The preferred stock growth
(PSK_USE) factor shows the positive but insignificant relation with future stock returns. We further find
that the factors’ return predictability remains strong even after controlling for risk (size and book-to-
market variables), return momentum, profitability, and other corporate capital investment measures (the
total asset growth, the cumulative accruals, accounting accruals, and investment-to-assets). These results
hold even after controlling for all the financial crisis falling in our research period.
For potential size effect, following Fama and French (2008) we run Fama and MacBeth (1973) cross-
sectional regressions for large, small, and micro stock portfolios. Consistent with Fama and French (2008)
and Titman et al., (2010) we find weak return predictability by total asset growth and our latent growth
factors for large stock as compared to small and micro stocks. The PSK_USE appears to be highly
9 Ou and Penman (1989), Holthausen and Larcker (1992), Lev and Thiagarajan (1993), Abarbanell and Bushee
(1997), Piotroski (2000), Sloan (1996), Spiess and Affleck-Graves (1999), Richardson and Sloan (2003), Fairfield,
Whisenant, and Yohn (2003), Hirshleifer, Hou, Teoh, and Zhang (2004), Titman, Wei, and Xie (2004), Daniel and
Titman (2006), Bradshaw, Richardson, and Sloan (2006), Lyandres, Sun, and Zhang (2008), Fama and French
(2008), Pontiff and Woodgate (2008), Xing (2008), Cooper, Gulen, and Schill (2008), Chen, Novy-Marx, and Zhang
(2011).
6
significantly positive related to subsequent future returns for large and micro firms but significant
negative relation to small stocks. The size effect tests suggest that whole period the negative relationship
between these new factors and the future stock returns is size specific. Among five factors only
FIN_FLEX and LT_INV are able to maintain the relationship. Collectively, our return predictability
evidence for different size firms implies that the value impact of managers’ choice of corporate financing
policies differs with firm size.
In our trading strategy based upon buying the lowest Latent Factors decile and selling short the highest
Latent Factors decile, the FIN_FLEX is profitable in 20 years, the ST_CREDIT is profitable in 17 years,
the LT_INV is profitable in 19 years, the CVT_DEBT is profitable in 14 years, and PSK_USE is
profitable in 13 out of the 25 years in the sample. The mean annual or monthly abnormal returns on
CVT_DEBT and PSK_USE ranked extreme decile is quite similar so their hedge (low-minus-high)
almost converges to zero. This is probably due to the similar level of positive profitability and firm size
across two extreme deciles. The mean equal-weighted annual buy-and-hold returns spread for
FIN_FLEX, ST_CREDIT, LT_INV, CVT_DEBT, and PSK_USE for whole sample are 15.2%, 6.1%,
10.1%, -1.5%, and -1.4% respectively.10
We also show the mean equal (value) weighted monthly
abnormal returns for two extreme decile portfolios and their hedge (low minus high) portfolio for first,
second, and third year for three capitalization (large, small, and micro) portfolios and a portfolio that does
not contain micro stocks (All but Micro).
In our further robustness test, we find that the industry returns are better explained by latent growth
factors than other compatible factors. The latent factors are important in certain industries; most of the
other accounting factors can be replaced by the latent factors. Similar to capitalization based regression
results the negative relationship between the latent factors (including other accounting measures) and
future returns is not stable across industries and even vary across annual cross-sections. Finally, in other
robustness tests we find the new latent factors as robust determinants of firm profitability and firm value.
10
In table 5 we present results for capitalization based portfolios but not for whole sample.
7
Overall, our empirical results suggest that the newly identified factors are optimal proxies of corporate
finance decisions that represent the most of the variation in the complete balance sheet information
content. These factor decisions can better capture the variations in the firms’ stock returns than the
predetermined accounting growth measures. Conventional accounting growth measures such as total asset
growth, cumulative accruals and accounting accruals suffer from missing variable bias. For example, the
ASSETG measure is the aggregate of the major items in the balance sheet but miss out other associated
items. This results in the missing variable bias.11
To this we argue that each individual latent growth
factor in our analysis is an aggregation of all associated aspects relating to specific types of corporate
financing policies. For the same reason, our first three factors (FIN_FLEX, ST_CREDIT, and LT_INV)
present true decomposition of the so called overall asset investment growth measure (ASSETG). Besides,
we find that financial flexibility measure (FIN_FLEX) appears to be the most important component of the
ASSETG in its return generating effects.
The rest of this paper is organized as follows. In section 2, we discuss the data and methodology, where
the present account of data samples, descriptive statistics, and research method. Section 3 provide
discussion on the new factors. Section 4 presents the results for growth rates, and returns to growth
portfolios in event time, annual buy-and-hold returns by year, and cross-sectional tests of return
predictability. Section 5 reports the further robustness tests including industry effect, time effect, firm
profitability and firm value. Section 6 concludes.
2. Data and methodology
2.1. Data Sample
Our research is solely US based and includes all firms listed on AMEX, NYSE and NASDAQ. There
are two data sources for the empirical analysis of this paper. The fiscal year balance sheet and income
11
Fama-French (2008) argue that the total asset growth effect is of secondary importance as it cannot be observed in
large cap firms. At the same time, Chan et al., (2007) show that asset growth takes a variety of forms like growth in
cash, current assets, or long-term assets. And, Cao (2011) shows that, cash and operating liabilities do not contribute
to the total asset growth effect.
8
statement data is obtained from the COMPUSTAT database for all firms. The market data for prices,
returns, trading volume, permno numbers, sic codes, and shares outstanding is obtained from the CRSP
database for listed stocks. Our COMPUSTAT data cover the twenty five year period from 1985 to 2009.
In case of balance sheet data, firms having no total-asset data are excluded from the analysis. Information
embedded within database codes is re-estimated using accounting formulas for calculation of particular
items and replaced accordingly. Similarly, income statement data are subject to availability for Sales
(Net). In case of CRSP data, we download monthly data for the 1984 to 2011 period.
The datasets obtained from both sources are merged by using PERMNO and CUSIP. In order to
construct the growth of individual asset/liability item of balance sheet, we require the firms to have at
least two years of observations. The growth rates are estimated for fiscal years “t-2” to “t-1”. Following
the general practice in investment literature, we exclude the financial and utilities firms from our data
sample (e.g. Stocks with four digits SIC Code 6000-6999 and 4900-4999).12
We calculate the market
equity at June of “t-1” for the firm size. The market equity at December is estimated for calculating book-
to-market ratio at the end of year “t-1”. For estimation of book value we follow Fama and French (1992).
Thus, the final data sample contains 821736 firm-month observations that are quite large for our research
period. Before factor analysis all variables are winsorized at 1st and 99th percentiles to mitigate the
potential outliers affects.13
2.2. Research method
2.2.1. Data construction
Suppose there are N=f {I1, I2, I3, …, In} business decision types and each business decision type In=f
{b1, b2, b3, b3, …, bn} is a function of bn the balance sheet elements. The every bn item of In decision types
is expected to have significant correlation with every other candidate item to be representative of
particular business decision type. Alternatively, the elements of a balance sheet that proxy certain
12
The cases where shares outstanding are negative or zero are also excluded. In order to calculate the book equity
for book-to-market ratio, we delete the firms with negative and zero book equity. 13
For winsorization of 1% at both tails, we follow Butler, Cornaggia, Grullon, and Weston (2011), Baxamusa (2011), Sullivan and Zhang (2011), and Teoh, Welch and Wang (1998) and many other in literature.
9
decision type are expected to follow the same movements or movements in systematic shifts across the
time. Thus, the consistency of membership makes each individual business decision type independent or
orthogonal.
The individual bn item is further converted into growth variable Xi, t that is the ratio of change in bn items
from time ‘t’ and time ‘t-1’ and the lagged total assets (Total Assett-1).
Xi, t = (bi, t - bi, t-1) /Total Assett-1. (1)
Where, bi, t stands for the balance sheet item i at time t and bi, t-1 is the lagged balance sheet item
for the same firm. We winsorized the Xi, t at 99 percentile and 1 percentile to remove outliers and to make
the data smooth. These growth variables are then standardized by fixing their mean at zero and standard
deviation as one. The purpose of standardization is to align all the variables to same scale. This removes
any doubt of mistake that may arise of different units of the variables (Jobson, 1992). In further regression
analysis, standardized variables help to compare the relation of each independent variable on the
dependent variable (Joseph F. Hair et al., 1998).
2.2.2. Factor analysis model
2.2.2.1. Model characteristics and relevant literature
Subsequent to standardization we employ factor analysis to growth variables that helps to reduce
the dimensional space by identifying the common movements into few representative factors. Factor
analysis determines groupings of objects of interest without prior associations and brings high internal
homogeneity (within the group) and high external heterogeneity (between groups) (Pinches et al., 1975).
Tsay (2005) observes that the number of factors is more well-defined in statistical factor models than the
way three factors obtained using Fama-French (1993) approach. According to Fabozzi and Markowitz
(2011) the latent (unobserved) factors are preferred because the “… observed factors may be measured
with errors or have been already anticipated by investors… factor analysis… explain complex
phenomena through a small number of basic factors”. They further mention that under arbitrage pricing
10
theory (APT) framework given by Ross (1976), “ the return covariance matrix formula hold regardless of
whether the factors are observable or latent”, whereas, for orthogonally rotated new factors, we have to
have identity covariance matrix.
The motivation for using factor analysis stems from our assumption cum objective that there are
certain optimal corporate financing decisions that are embedded in several balance sheet entries (due to
the double entry bookkeeping) who share the common variance and if their movement identified correctly
can produce the basic decision types that are fundamental to the business. For the similar objective the
researchers have applied the factor analysis in a variety of financial research settings. It has been applied
to to predict the industrial bond ratings (Pinches and Mingo, 1973). To investigate the institutional
investor’s influence on the R&D investment behavior of corporate managers (Bushee, 1998). Author use
factor analysis for grouping up the nine variables into three factors, to be used for subsequent analysis.
Similarly, for investigating the institutional investor preferences and price pressures around spin-offs
(Abarbanell et al., 2003). They employ factor analysis to aggregate the 15 variables that describe the
investment preferences of institutional investors into 4 factors. Then, for investigating the characteristics
of firms that involve in mergers and acquisition (Sorensen, 2000). He uses factor analysis to group up
twenty two ratio variables into three factors. Additionally, the factor analysis processes employ the
correlational structure of the potential input information that coincides with the Subrahmanyam (2010)
argument that researchers needs to employ correlational structure amongst the variables in predicting the
stock returns.
2.2.2.2. The data mining concerns
Moreover, this technique may not pose any data mining concern as the research does not fall
under the definitions of data mining given by Black (1993). According to him, the researcher is doing
data mining, if he reports the t-statistics of only few significant variables, when majority of others do not
have the same significance. Then, the researcher is also doing data mining, if he chooses not to report all
results of analysis. Similarly, if researcher starts research by choosing topic to work on and follows the
11
others in similar areas for the method. Finally, researcher finds any pattern in past security returns and
then designs his research. We see no data mining concern in this research due to manifold reasons: First,
we only apply factor analysis on the balance sheet growth variables (X terms) not on market values or
potential dependent variables (Y terms). Second, the factor analysis is used to produce annual factors
instead of whole period factors that remove any data mining concerns in subsequent analysis that use the
identified decision types as independent variables. Third, subsequent return predictability test, the models
are built on publicly available information about all common stocks. Finally, there is no selection bias in
selecting the balance sheet growth items as inputs to the model as we follow strictly the basic
assumptions of the model.
The factor analysis assumptions include small partial correlations among the input variables;
sampling adequacy of around 0.60 and a correlation cutoff of 0.45 (Joseph F. Hair et al., 1998). The
sampling adequacy of 0.50 is suggested for both the overall test and individual variable and 0.60 for
successful factor analysis. Even, correlations below 0.30 make factor analysis inappropriate (Joseph F.
Hair et al., 1998). Following above literature’s recommendation, variables which do not meet above
criteria are discarded from the analysis. Thus, final set of growth variables includes thirty seven variables.
This removal of variables is followed by the identification of a number of factors.
2.2.2.3. The basic model assumptions
The number of factors is generally obtained by looking at EIGENVALUES equal to one and
above one or by looking at the SCREE plot. But, this is not certain that we get an accurate number of
factors following above practice so it also depends on the researcher skills to get correct structure. The
correct factor structure requires at least three variables group together with distinct loadings for individual
factor. Then, selected factors are given orthogonal rotation with a principle component factor analysis
method to create final factor structure. Finally, we label factors based on the highest loadings of factor
constituent growth rates.
12
3. Preliminary results on the factor analysis
The factor analysis results are given in Table 1, the model produces the five latent factors that
proxy for five optimal corporate financing decision types that are labelled as, FIN_FLEX (financial
flexibility), ST_CREDIT (short-term credit), LT_INV (long-term capital investment), CVT_DEBT
(convertible debt usage), and PSK_USE (preferred stock usage). These five latent factors together can
capture 70% of the common variation in the aggregate balance sheet (including supplementary items)
growth variable. The factors are extracted annually instead of the whole sample period for their suitability
for subsequent financial analysis. Any increase in the number of factors is not suitable for our data sample
because first it is adding no more than 3 to 5% in the explanatory power of each additional factor
annually. Second, the additional factor does not meet the basic assumptions of the model discussed above.
Finally, the scree plot indicates the five factors are adequate. The details of these new factors are
discussed in the following sub-section.
[Insert Table 1]
3.1. Discussion on the latent growth factors
The five factors and their component loads are shown in Table 1. In this section, we discuss the
construction of each factor to understand why the components have high loadings on that particular factor
and to get economic meaning out of that factor. Understanding of the components would help us to label
the factors. The labels should be able to convey the underlying economic meaning of the each factor. In
fact, we check the factor component values across each factor sorted deciles for strengthening the point
that components have a linear association with their factor and nonlinear to other factors. In turn we may
be able to relate each factor to known growth measures and other determinants of stock returns.
Subsequently, we also check the stability of each of the factors. For stability the component membership
needs to be stable across the time period for each factor.
13
3.1.1. Financial Flexibility (FIN_FLEX)
From table 1, first factor FIN_FLEX has the highest weight on common equity-total, common
equity-liquidation, cash and equivalents, common equity-tangible (with weighting loads above 0.84), the
other important elements are current assets-total, invested capital, total assets, cash, and capital surplus
(with weighting loads above 0.45). The factor analysis helps to identify and aggregate all the associated
components of the financial flexibility so that there may be no missing variable bias. The dominant
components of common equity, current liquid assets, and invested capital growth indicates that they
represent the aggregate financial flexibility of a firm. The measure can offer important insights into the
corporate capital investment literature. It appears that managers seek the both equity and cash balances to
meet liquidity requirements to finance growth options with positive NPV. This possible fact may explain
the why cash holdings and common equity are highly correlated.
We think that finding of the FIN_FLEX makes it the important contribution to literature on
financial flexibility and capital structure. This measure has all possible attributes required to fit the ad-hoc
definition of financial flexibility. Denis (2011) argues that “financial flexibility refers to the ability to
respond in a timely and value-maximizing manner to unexpected changes in the firm’s cash flows or
investment opportunity set.” Marchica and Mura (2010) argue “since there is no well-defined measure of
flexibility in the literature, this is an unobservable factor that depends largely on managers’ assessment of
future growth options.” In fact, FIN_FLEX’s associated attributes are the financial slack or cash holding,
positive relation with the future investment opportunities, and it is unobserved in nature.
Further, FIN_FLEX is the always first factor almost in all the annual factor analysis models and
also explains the large part of the common variations of the balance sheet information content as
compared to other four factors. This finding is consistent with the Graham and Harvey (2001) that
financial flexibility is the most important determinant of capital structure. Although, it is not our quest but
FIN_FLEX may help address the DeAngelo and DeAngelo (2007) argument that financial flexibility is
missing link that is important to connect the capital structure theory with observed firm behavior observed
14
by Graham and Harvey (2001). In fact, this factor offers the integrated measure of both financial and
investment flexibility, that are reported to be close substitutes (Faulkender and Wang, 2006; Gamba and
Triantis, 2008)).
Besides, FIN_FLEX may be linked to stock financing sub-component of the asset growth given
in Cooper, et al. (2008). But, it differs from stock financing that it does not take into account the preferred
stock’s growth, because preferred stocks and common equity have insignificant correlation (-0.05). The
preferred stock is represented by separate factor. Similarly, FIN_FLEX include liquid assets as major
components due to their strong association to common equity. Conversely, Cooper, et al. (2008) considers
the financial assets as separate components of the total asset growth.
3.1.2. Short-term credit (ST_CREDIT)
The short-term credit (ST_CREDIT) makes our second factor that has the highest weight on the
current liabilities-total, followed by liabilities-total and debt in current liabilities (with loads above 0.70).
The remaining elements are accounts payable, accounts receivable, total debt, notes payable, total assets,
current liabilities-others, inventory-total and current assets-total (all have weighting loads above 0.45).
The second factor is linked to the level of firms’ operating business growth thus consists of all
associated aspects of the short-term financing making it well-defined. According to Diamond (1991),
Chung (1993), and Doukas and Pantzalis (2003) large firms have higher short-term debt borrowing than
small firms, because higher-credit ratings of the larger firms have easier access to short-term borrowing
than smaller firms. Similarly, Easterwood and Kadapakkam (1994), Barclay and Smith (1995a), and
Doukas and Pantzalis (2003) find that larger firms have high information asymmetries that cause them to
issue more short-term debt. In further financial research this factor may help us to identify credible firms
to be part of investment portfolios.
Next, we find that ST_CREDIT factor differs from Cooper, et al. (2008)’s debt financing growth
component in construction. ST_CREDIT encompasses the Noncash current asset growth (∆CurAsset),
Debt financing growth (∆Debt), and operating liabilities growth (∆OpLiab), thus one factor represents the
15
overall operating business growth of the firm. Moreover, Cooper, et al. (2008) combines the growth both
in debt in current liabilities and long-term debt, but we find an insignificant correlation (-0.03) among
them. This is why factor analysis puts them in separate factors. This shows the fundamental difference
between Cooper and our result. Cooper tried to explain the asset growth variable by analyzing the link of
the total to the subcomponents. Among other constituents of ST_CREDIT, there are also significant loads
on accounts payable, notes payable and accounts receivables growth. Hence, makes this factor possibly
well-defined to reflect on the business credibility.
3.1.3. Long-term capital investment (LT_INV)
The long-term capital investment (LT_INV) makes third factors and its most important
components are deferred taxes and investment credit, plant, property, & equipment-gross and net (with
weighting loads above 0.71). The remaining influential elements are long-term debt, invested capital and
total assets (all have a load greater than 0.45). This factor represents is one of the major components of
the overall corporate capital investment (Titman et al., 2010) but it offers a more comprehensive measure
than other similar measures. The other observed measure of long-term capital investment does not
consider the associated components of the investment in fixed assets that are hidden in different balance
sheet items may be helpful in fine tuning the measure.
Among the components the deferred taxes show timing differences in financial reporting for tax
purposes. Likewise, investment tax credit is generated due to new capital investment (include long-term
debt, preferred stock, minority interest and common equity). Deferred taxes in most companies are
generated by the depreciation of fixed assets, and thus, creates deferred tax liability. And it is because of
timing difference that there is more book income than tax income (Young, 1997). Generally, depreciation
methods for financial reporting and for tax purposes are different due to timing differences between
taxable income and accounting income that can be inferred from changes in deferred tax balance.
Moreover, this factor may help infer issues relating to investment tax shields and debt tax shields
(Trezevant, 1992). Literature suggests that long-term debt borrowings are mainly done by small firms that
16
have low credit rating and high agency cost of debt (Chung, 1993; Diamond, 1991; Doukas and Pantzalis,
2003). LT_INV is comparable but it appears to be more comprehensive to Property, plant, and equipment
growth (∆PPE) sub-component given in Cooper, et al. (2008) as considering all associated aspects of
long-term investment.
3.1.4. Convertible debt usage (CVT_DEBT)
The convertible debt usage (CVT_DEBT), the fourth factor has the highest weights of debt
convertible-total and debt convertible-preferred stock (with loads greater than 0.80). The others are debt
convertible and subordinated and long-term debt (with loads about 0.64). Firms that circumvent financial
risks by raising capital from long-term sources of capital like debt convertible and subordinated are
represented by our fourth factor. The factor possibly reflects on the financing and investment constraints
of firms like weak capital structure and financial distress. In such a situation firms avoid issuing stocks
and resort to use convertible bonds, notes, debentures, and subordinated debt and preferred stock to raise
equity. These are called as hybrid securities and can be helpful in minimizing the higher costs of external
financing.
The convertible debt is a fraction of long-term debt, firms use to avoid long-term debt in the
presence of agency costs (Bodie and Taggart, 1978; Doukas and Pantzalis, 2003). Firms that use this type
of debt are small and micro firms with lower credit ratings, low information asymmetries, and high
agency costs of debt. According to Hovakimian et al., (2001) convertible debt issuance is second largest
after equity issuance, on an average firms show poor performance subsequent to issuance of convertible
debt (Lewis et al., 2001; Stein, 1992). Further, we find that the convertible debt usage (CVT_DEBT) is
not part of the total asset growth decomposition done by Cooper, et al. (2008). Moreover, CVT_DEBT
constituent variables have insignificant low correlation with the total asset growth (ASSETG) that makes
them orthogonal. In fact, we find this factor as measure of the operational risk arising from the
convertibility of bond and equity.
17
3.1.5. Preferred stock usage (PSK_USE)
Finally, Fifth factor PSK_USE constitutes of purely preferred stock-total, liquidation and
nonredeemable (with loads greater than 0.62) (i.e., compustat item #A130, A209 & A10) and also lower
cross-loading of the total long-term debt convertible. According to Weston and Copeland (1992) and
Howe and Lee (2006) preferred stocks are hybrid security with features similar to bonds (as it has a par
value due in the event of liquidations and preferred dividends) and common stocks (equity to bondholders
and debt to stockholders and in the balance sheet reported as “preferred stock” or “preferred equity”). The
preferred stock is considered better than debt, as firms cannot be forced into bankruptcy on the failure of
dividend payments. At the same time, the common stockholders are not liable to share the success of the
firm with preferred stockholder due to their fixed preferred dividends. On the contrary, preferred stock
holders have priority claim than the common stockholders.
Houston and Carol Houston (1990) find that firm issuing preferred stock have a propensity for
lower tax rates and firms investing in preferred stock are likely to have higher tax rates. Further, they find
that industrial firms not utilities issue the majority of preferred stocks; and major use is in the merger
market by both target and acquiring firms. Target firms use it as antitakeover device and Acquirer for tax
benefits. Houston and Carol Houston (1990) also discuss the literature that integrates preferred stock into
the capital structure framework for example Fooladi and Roberts (1986), Miller (1977), Elmer (1988),
and DeAngelo and Masulis (1980). Hovakimian et al., (2001) document that preferred stock issuers
realize lower return compared to non-issuers. Howe and Lee (2006) find that in long-run preferred
issuers’ underperformance is transient and confined to small firms, such firms have poor profitability and
high debt ratios, thus they use preferred stock to offset the other expensive securities.
3.2. Summary statistics
In Table 2, we report the mean (median) values for some well-know accounting and market
characteristics for two extreme decile portfolios ranked on the five latent growth factors. For financial
flexibility ranked portfolios, the lowest decile stocks have lower mean (median) values for total asset
18
growth, cumulative accruals growth, accounting accrual growth, profitability, and are smaller in size; and
have relatively higher values for investment-to-asset growth, leverage, and the book-to-market ratio
characteristics as compared to the highest decile portfolios.
For short-term credit ranked portfolios, the lowest decile stocks have lower mean (median) values
for the total asset growth, cumulative accrual growth, accounting accruals, lower leverage and are smaller
in size; and have relatively higher profitability, higher investment-to-asset growth, and higher book-to-
market ratios as compared to the highest decile portfolios. For long-term capital investment ranked
portfolios, the lowest decile stocks have lower mean (median) values for all accounting characteristics,
have lower profitability, and are smaller in size but have relatively higher book-to-market ratios as
compared to the highest decile portfolios.
Then, for debt convertible usage, the lowest decile stocks have similar mean (median) values for
the total asset growth, cumulative accruals growth, investment-to-asset growth, size, and book-to-market
ratio as compared to highest decile stocks. However, we find differences across the other characteristics
for example leverage and the accounting accrual growth have lower values and profitability of these
stocks is higher than the highest decile stocks. Finally, the preferred stock usage ranked portfolios, the
except the total asset growth the lowest decile stocks have higher mean (median) values for all other
characteristics as compared to highest decile stocks.
Overall, we are able to find clear distinctive patterns observing these characteristics when ranked
on individual new factor. In table 3, report the Pearson and Spearman correlations for five latent growth
factors and the other characteristics discussed above. The the latent growth factors appear to be having no
significant correlations because of the orthogonality induced during the factor analysis process. However,
there exist significant correlation between the total asset growth and our first latent factors showing that
the true decomposition of the total asset growth.
[Insert Table 2 and 3]
19
3.3. Latent growth stability check
Finally, we also make check for factor stability across time. Factor components are checked in 25
annual factor model and found consistently appearing in the same factor. The top five components of the
financial flexibility and the short-term credit are always the same in the list of their components. The top
four components of the other three factors are always same in the list of their components. However, we
find that except financial flexibility the ranking of the other four factors does change across the 25 years.
Based upon this evidence we may say that the factor structure is stable. In other stability check we find all
the components that are linearly related to its factor and non-linearly to other factors.
The stability of the identified factors suggests some real economic forces that explain the
movement of the firms. A natural prediction of the understanding is that these factors have predictive
power for the equity return rates, the results of testing the prediction could be an important evidence for
the existence of these latent factors. In the next section, we provide the testing results.
4. Results
4.1. Portfolio analysis
4.1.1. Time trends in profitability and returns for extreme latent growth factor ranked deciles
Following Cooper, et al. (2008) we report in the figure 1a to 1e the long-run profitability and
return effects for two extreme decile portfolios. We first sort and rank the data sample on the lagged latent
growth factors into decile growth portfolios every t year at the end of June over 1985 t0 2009. Using the
June t growth cutoffs, we form portfolios that are held for one year from July of t year to June of t+1 and
then rebalanced. When sorted on latent growth factors. In this figure we plot the mean monthly
profitability values and mean monthly raw returns at equal-weighted basis to decile latent growth
portfolios at event time (5-year prior and 5-years after portfolio formation). The profitability is measured
as net income divided by the total assets. The returns are accumulated monthly raw returns starting four
months after the fiscal year end.
20
Fig. 1a plots the time series means of firm profitability and mean monthly raw returns for extreme
(deciles 1 and 10) FIN_FLEX deciles. The profitability of lowest financial flexibility decile firms is
always lower and negative as compared to the highest financial flexibility decile firms in 5-years prior
and 5-years post ranking period. The high FIN_FLEX decile firms appear to earn higher mean monthly
returns in 5-years prior to ranking year, but we see a crossover after ranking period, when high
FIN_FLEX earns lower mean monthly returns. The trend shows that despite the firms in post ranking
period are more profitable than prior years but are unable to sustain abnormal returns. Alternatively, we
may say that firms that have lower financial flexibility and have negative profitability produce a
significant mean monthly return at least for next three years as compared to firms that are large with high
level of financial flexibility and positive profitability.
In a similar way for ST_CREDIT ranked extreme deciles, the firm profitability is positive for
both deciles in before and after the event year. Overall, except year -5, -4, and -3 the high ST_CREDIT
decile firms are more profitable than low decile firms. In terms of mean monthly returns generating
capabilities, the high ST_CREDIT decile firms earn higher returns in 5-prior event years as compared to
post-ranking years. However, the in post ranking year the return spread is marginal. Then for LT_INV
ranked extreme decile portfolios mean profitability and mean monthly returns plotted in fig 1c, the high
LT_INV firms are always positively profitable as compared to low LT_INV firms. Where low decile
firms have actually negative profitability only in year -1. And, in case of returns there is a crossover in
year -1, the high LT_INV firms earn higher returns prior to -1 and lower return in later years until they
converge in the year 4. It appears that despite of high profitability levels the high LT_INV firms are
unable to sustain the higher return generation in post event years. Our evidence for LT_INV is quite
consistent with the NOA (net operating assets) evidence (Hirshleifer et al., 2004).
Next from fig. 1d, we can say that the low CVT_DEBT decile firms are way profitable than the
high CVT_DEBT decile firms. However, the low CVT_DEBT decile firms earn higher mean monthly
returns than high CVT_DEBT decile firms prior to the year 1, but later year the spread is meaningless
may be owing to the similar level of capital investment, firm size, and the book-to-market ratios. Finally,
21
for PSK_USE ranked extreme deciles (see, fig. 1e), the low PSK_USE decile firms have always higher
and positive profitability than the high PSK_USE decile firms. In case of high PSK_USE decile the
profitability is negative in year -2 and -1. However, we do not see any clear pattern of mean monthly
returns on the PSK_USE ranked deciles.
[Insert Figure 1a-1e]
4.1.2. Abnormal returns by Latent Growth Variable deciles
In table 4 we report the average monthly abnormal returns for five latent growth factors sorted
extreme decile portfolios in subsequent 1, 2, and 3 years after portfolio formation. For the period of July
1985 to October 2009, we formed monthly decile portfolios by ranking on the last year Latent Factor.
There is a 4-month lag between the fiscal year end and the portfolio formation month. The time-series
average of monthly equal (value) weighted abnormal returns (the abnormal returns are the difference
between the stock returns and size, book-to-market, and momentum matched benchmark portfolio
returns) are reported for each decile and hedge portfolio. The hedge portfolio is created by going long in
the lowest ranked Latent Factor portfolio and short in the highest ranked Latent Factor portfolio.
Comparatively, we also control for risk (size and book-to-market) and other important return anomalies
and report the alpha estimates (t-statistics) for CAPM, Fama and French (1993) three factor model, and
Carhart (1997) four factor model for both equal- and value-weighted hedge portfolios. We report the
results for four capitalization ranked portfolios following Fama and French (2008), the stocks under 20th
percentile of market cap at the end of the June each year are grouped as micro stock. The stocks that are
between 20th and 50
th percentile are grouped as small stocks and above median are grouped as large
stocks. In All but Micro firm portfolios we only exclude the micro stocks. Although not reported here but
we also obtain the similar results for the whole sample period.
The results for characteristic adjusted hedge returns and controlling for the pricing factors are
similar. In panel A of table 4 we report results at equal-weighted basis when firms are ranked by
FIN_FLEX, we indicate that for a large firm, the average monthly hedge returns for year t+1, t+2, and t+3
22
are -0.01%, -0.28%, and -0.57% with t-statistics of -0.17, -3.67,-6.77 respectively. For small stocks, the
average monthly hedge returns for year t+1, t+2, and t+3 are 0.20%, 0.19%, and 0.27% with t-statistics of
3.16, 2.24, and 2.80 respectively. For micro stocks, the average monthly hedge returns for year t+1, t+2,
and t+3 are 0.60%, 0.75%, and 0.86% with t-statistics of 8.2, 8.26, and 9.67 respectively.We find no
significant variation in results when controlling for pricing factors and also obtain similar results at a
value-weighted basis. For a whole period there is an obvious size effect but within size groups there exists
no size effect. It appears that firms that lacks financial flexibility earn greater average monthly abnormal
returns as compared to firms that has higher financial flexibility.
In case of ST_CREDIT ranked portfolios, the panel B results indicate that on an equal-weighted
basis the low-minus-high hedge portfolio the mean monthly abnormal returns for large stocks at year t+1,
t+2, and t+3 are -0.16%, -0.50%, and -0.71% with t-statistics of -4.21, -8.19,-10.02 respectively. For
small stocks, the average monthly hedge returns for the subsequent three years are -0.22%, -0.28%, and -
0.39% with t-statistics of -3.20, -2.88, and -3.87 respectively. Then, for micro stocks, the average monthly
hedge returns for the subsequent three years are 0.17%, 0.39%, and 0.50% with t-statistics of 2.77, 5.60,
and 7.17 respectively. Similar to financial flexibility, the results are similar when controlling for pricing
factors and when results are obtained at value-weighted basis.
Then for LT_INV ranked portfolios, the panel C results indicate that on an equal-weighted basis
the mean monthly abnormal returns for large stock hedge portfolios in three subsequent years are not
significant except year t+1 with returns of 0.11% and t-statistics of 2.69. However, at value-weighted
basis the first two years offer significant abnormal returns of 0.15% and 0.24% with t-statistics of 3.27
and 3.63. For small stocks, we also notice the negative returns spread in year t+2 and t+3 are significant
both equal- and value-weighted basis. Then, for micro stocks, the average monthly hedge returns for the
subsequent three years are 0.45%, 0.76%, and 0.82% with t-statistics of 8.21, 9.52, and 8.63 respectively.
Next, in case of CVT_DEBT ranked portfolios, in panel D, the results indicate that large firms
earn mean monthly positive abnormal returns significant at equal-weighted basis for year t+1, t+2, and
t+3 are 0.08%, 0.2%, and 0.27% with t-statistics of 1.83, 5.09, and 3.99 respectively. The results for large
23
stocks are more pronounce at value-weighted basis. However, for small and micro firms we find negative
and insignificant abnormal returns.
Finally, PSK_USE ranked portfolios, the panel E results indicate that on an equal-weighted basis
the low-minus-high hedge portfolio the mean monthly abnormal returns for large stocks at year t+1, t+2,
and t+3 are -0.13%, -0.30%, and -0.41% with t-statistics of -2.28, -4.74,-5.40 respectively. For small
stocks, the average monthly hedge returns for the subsequent three years are 0.13%, 0.15%, and 0.22%
with t-statistics of 2.33, 1.70, and 2.42respectively. Then, for micro stocks, the average monthly hedge
returns for the subsequent three years are -0.07%, -0.30%, and -0.36% with t-statistics of -1.29, -4.03, and
-4.96 respectively. The results are similar when controlling for pricing factors and when results are
obtained at value-weighted basis.
[Insert Table 4]
4.1.3. Annual buy-and-hold returns
In Table 5 we report the average annual buy-and-hold returns for Low and High latent growth
portfolios for the period of 1986 to 2010. We also report the low-minus spread of means for the same time
period. We obtain both equal- and value-weighted portfolios returns for three capitalization levels and all
but micro stocks (excludes the micro stocks) portfolios. The capitalization levels are constructed
following NYSE capitalization break points as defined in Fama and French (1992).
In our trading strategy based upon buying the lowest Latent Factors decile and selling short the
highest Latent Factors decile, the FIN_FLEX is profitable in 20 years, the ST_CREDIT is profitable in 17
years, the LT_INV is profitable in 19 years, the CVT_DEBT is profitable in 14 years, and PSK_USE is
profitable in 13 out of the 25 years in the sample. The mean annual abnormal returns on CVT_DEBT and
PSK_USE ranked extreme decile is quite similar so their hedge (low-minus-high) almost converges to
zero. This is probably due to the similar level of positive profitability and firm size across two extreme
deciles. The mean equal-weighted annual buy-and-hold returns spread for FIN_FLEX, ST_CREDIT,
24
LT_INV, CVT_DEBT, and PSK_USE for whole sample are 15.2%, 6.1%, 10.1%, -1.5%, and -1.4%
respectively.
The above buy-and-hold returns based upon each latent factor suggest that these factors could be
important in explaining the market performance of the stocks; a better test is to put these factors in a
controlled model to control their unique contribution to the market returns. For this purpose we have
implemented Fama-Macbeth models in the following sub-section.
[Insert Table 5]
4.2. Cross-sectional regression results
At individual stock level analysis, we employ Fama-MacBeth (1973) cross-sectional regression
framework to test the future return predictability of the latent growth factors for a period of nineteen years
(July, 1993 to June 2011). The latent growth factors’ return predictive power is further tested by
introducing the other well-known return determinants in regression models. Prior literature documents the
Parrino, R., Weisbach, M. S., 1999. Measuring investment distortions arising from stockholder–
bondholder conflicts. Journal of Financial Economics 53, 3-42.
Pinches, G. E., et al., 1975. The hierarchical classification of financial ratios. Journal of Business
Research 3, 295-310.
Pinches, G. E., Mingo, K. A., 1973. A multivariate analysis of industrial bond ratings. The Journal of
Finance 28, 1-18.
Piotroski, J. D., 2000. Value investing: The use of historical financial statement information to separate
winners from losers. Journal of Accounting Research 38, 1-41.
Polk, C., Sapienza, P., 2009. The stock market and corporate investment: A test of catering theory. The
Review of Financial Studies 22, 187-217.
Pontiff, J., Woodgate, A., 2008. Share issuance and cross-sectional returns. The Journal of Finance 63,
921-945.
Porta, R., et al., 1997. Legal determinants of external finance. The Journal of Finance 52, 1131-1150.
Rajan, R. G., Zingales, L., 1995. What do we know about capital structure? Some evidence from
international data. The Journal of Finance 50, 1421-1460.
Richardson, S., Sloan, R., 2003. External financing and future stock returns. Rodney L. White Center for
Financial Research Working Paper.
Ross, S. A., 1976. The arbitrage theory of capital asset pricing. Journal of Economic Theory 13, 341-360.
Sloan, R. G., 1996. Do stock prices fully reflect information in accruals and cash flows about future
earnings? The Accounting Review 71, 289-315.
Sorensen, D. E., 2000. Characteristics of merging firms. Journal of Economics and Business 52, 423-433.
Spiess, D. K., Affleck-Graves, J., 1999. The long-run performance of stock returns following debt
offerings. Journal of Financial Economics 54, 45-73.
41
Stein, J. C., 1992. Convertible bonds as backdoor equity financing. Journal of Financial Economics 32, 3-
21.
Subrahmanyam, A., 2010. The cross-section of expected stock returns: What have we learnt from the past
twenty-five years of research? European Financial Management 16, 27-42.
Sullivan, M., Zhang, A., 2011. Are investment and financing anomalies two sides of the same coin?
Journal of Empirical Finance 18, 616-633.
Teoh, S. H., et al., 1998. Earnings management and the underperformance of seasoned equity offerings.
Journal of Financial Economics 50, 63-99.
Titman, S., Tsyplakov, S., 2007. A dynamic model of optimal capital structure. Review of Finance 11,
401-451.
Titman, S., et al., 2004. Capital investments and stock returns. Journal of Financial and Quantitative
Analysis 39, 677-700.
Titman, S., et al., 2010. Unexpected investment, overinvestment, and stock returns: Working paper,
University of Texas at Austin.
Trezevant, R., 1992. Debt financing and tax status: Tests of the substitution effect and the tax exhaustion
hypothesis using firms' responses to the economic recovery tax act of 1981. The Journal of Finance 47,
1557-1568.
Tsay, R. S., 2005. Analysis of financial time series: Wiley-Interscience.
Tufano, P., 2001. Hbs-jfe conference volume: Complementary research methods. Journal of Financial
Economics 60, 179-185.
Welch, I., 2004. Capital structure and stock returns. Journal of Political Economy 112, 106-132.
Weston, J. F., Copeland, T. E., 1992. Financial theory and corporate policy: Addison Wesley.
Xing, Y., 2008. Interpreting the value effect through the q-theory: An empirical investigation. The
Review of Financial Studies 21, 1767-1795.
Young, D., 1997. Economic value added: A primer for european managers. European Management
Journal 15, 335-343.
Zhang, Y., 2005. Net operating assets as a predictor for future stock returns–an industry analysis: Ohio
State University.
42
Table 1: Latent Growth Factor Pattern
This table presents the factor structure as the outcome of the application of the annual factor analysis model on the balance sheet growth rates for the period of 1985-2009.
We estimate 37 balance sheet growth rates using the following formula Xi, t = (bi, t - bi, t-1) /Total Assett-1. Where, bi, t stands for the balance sheet item i at time t and
bi, t-1 is the lagged balance sheet item for the same firm. For each of new latent growth factors we present constituent variable names and their codes. The five factors
are also given a label that reflects the constituent items that have higher loading on that particular factor.
Fig. 1a. Mean Profitability (A) and Mean monthly returns (B).
The figures show the mean profitability measured as net income divided by the total assets and the average
monthly raw returns for FIN_FLEX sorted portfolios 5 years prior the ranking year and 5 year post ranking period.
Returns are accumulated monthly raw returns starting 4 months after fiscal year end. We first sort and rank the data
sample on the lagged financial flexibility growth (FIN_FLEX) rate into decile growth portfolios every t year at the end
of June over 1985 to 2009. Using the June t asset growth cutoffs, we form portfolios that are held for one year from
July of t year to June of t+1 and then rebalanced.
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an R
OA
Sorting Year
A Lowest FIN_FLEX Decile Highest FIN_FLEX Decile
0.00
0.01
0.02
0.03
0.04
0.05
0.06
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an M
on
thly
Re
turn
Sorting Year
B Lowest FIN_FLEX Decile Highest FIN_FLEX Decile
46
Fig. 1b. Mean Profitability (A) and Mean monthly returns (B).
The figures show the mean profitability measured as net income divided by the total assets and the average
monthly raw returns for ST_CREDIT sorted portfolios 5 years prior the ranking year and 5 year post ranking period.
Returns are accumulated monthly raw returns starting 4 months after fiscal year end. We first sort and rank the data
sample on the lagged financial flexibility growth (ST_CREDIT) rate into decile growth portfolios every t year at the
end of June over 1985 to 2009. Using the June t asset growth cutoffs, we form portfolios that are held for one year
from July of t year to June of t+1 and then rebalanced.
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an R
OA
Sorting Year
A Lowest ST_CREDIT Decile Highest ST_CREDIT Decile
0.00
0.01
0.01
0.02
0.02
0.03
0.03
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an M
on
thly
Re
turn
s
Sorting Year
B Lowest ST_CREDIT Decile Highest ST_CREDIT Decile
47
Fig. 1c. Mean Profitability (A) and Mean monthly returns (B).
The figures show the mean profitability measured as net income divided by the total assets and the average
monthly raw returns for LT_INV sorted portfolios 5 years prior the ranking year and 5 year post ranking period.
Returns are accumulated monthly raw returns starting 4 months after fiscal year end. We first sort and rank the data
sample on the lagged financial flexibility growth (LT_INV) rate into decile growth portfolios every t year at the end of
June over 1985 to 2009. Using the June t asset growth cutoffs, we form portfolios that are held for one year from July
of t year to June of t+1 and then rebalanced.
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an R
OA
Sorting Year
A Lowest LT_INV Decile Highest LT_INV Decile
0.00
0.01
0.01
0.02
0.02
0.03
0.03
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an M
on
thly
Re
turn
s
Sorting Year
B Lowest LT_INV Decile Highest LT_INV Decile
48
Fig. 1d. Mean Profitability (A) and Mean monthly returns (B).
The figures show the mean profitability measured as net income divided by the total assets and the average
monthly raw returns for CVT_DEBT sorted portfolios 5 years prior the ranking year and 5 year post ranking period.
Returns are accumulated monthly raw returns starting 4 months after fiscal year end. We first sort and rank the data
sample on the lagged financial flexibility growth (CVT_DEBT) rate into decile growth portfolios every t year at the
end of June over 1985 to 2009. Using the June t asset growth cutoffs, we form portfolios that are held for one year
from July of t year to June of t+1 and then rebalanced.
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an R
OA
Sorting Year
A Lowest CVT_DEBT Decile Highest CVT_DEBT Decile
0.00
0.01
0.01
0.02
0.02
0.03
0.03
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an M
on
thly
Re
turn
s
Sorting Year
B Lowest CVT_DEBT Decile Highest CVT_DEBT Decile
49
Fig. 1e. Mean Profitability (A) and Mean monthly returns (B).
The figures show the mean profitability measured as net income divided by the total assets and the average
monthly raw returns for PSK_USE sorted portfolios 5 years prior the ranking year and 5 year post ranking period.
Returns are accumulated monthly raw returns starting 4 months after fiscal year end. We first sort and rank the data
sample on the lagged financial flexibility growth (PSK_USE) rate into decile growth portfolios every t year at the end
of June over 1985 to 2009. Using the June t asset growth cutoffs, we form portfolios that are held for one year from
July of t year to June of t+1 and then rebalanced.
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an R
OA
Sorting Year
A Lowest PSK_USE Decile Highest PSK_USE Decile
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.04
0.04
-5 -4 -3 -2 -1 1 2 3 4 5
Me
an M
on
thly
Re
turn
s
Sorting Year
B Lowest PSK_USE Decile Highest PSK_USE Decile
50
Table 4
Average monthly abnormal returns for Latent Factors (FIN_FLEX, ST_CREDIT, LT_INV, CVT_DEBT, PSK_USE) ranked extreme decile portfolios and hedge portfolio in 1, 2 and 3 years after portfolio formation are given in this table. For period of July 1985 to October 2009, we form monthly decile portfolios by ranking on
last year Latent Factor. There is 4-month lag between fiscal year end and the portfolio formation month. The time-series average of monthly equal (value) weighted
abnormal returns (the abnormal returns are the difference between the stock returns and size, book-t-market, and momentum matched benchmark portfolio returns) are reported for each decile and hedge portfolio with respective t-statistics. Hedge portfolio is created by going long in the lowest ranked Latent Factor portfolio and short
in the highest ranked Latent Factor portfolio. Comparatively, we report the alpha estimates (t-statistics) for CAPM, Fama and French (1993) three factor model, and
Carhart (1997) four factor model for both equal and value weighted hedge portfolios. The bold numbers indicate significance at less than 5% level.
Panel A: FIN_FLEX ranked portfolios
Equal weighted
t-values (Equal weighted)
Value weighted
t-values (Value weighted)
Large Firms raw (t+1) Adj(t+1) Adj(t+2) Adj(t+3) Raw( t+1) Adj(t+1) Adj(t+2) Adj(t+3)
Table 6: Fama-MacBeth Regressions of Annual Stock Returns on Asset Growth Rates and Other Variables
In this table, the annual geometric compounding future returns are regressed on the observed Latent growth measures and the other growth measures. For every year for period of
1985 to 2009 we run Fama-MacBeth (1973) cross-sectional regressions at the individual stock level. We run regressions for two types of model.
(1)
(2)
The first model is labeled as a base regression model; and in second model the base regression model is augmented by control variables. FIN_FLEX growth measure is the latent
growth score obtained via factor analysis procedures and it proxy for financial flexibility. Similar is the construction of other latent growth factors. ST_CREDIT growth measure
proxy for short-term credit of firms. LT_INV growths proxy for long-term capital investment. CVT_DEBT growth measure proxy for convertible debt usage asset. PSK_USE
growth measure for preferred stock usage. For The five factors, original input variables are estimated by ratio of change in balance sheet variables to lagged total assets before
processing through the factor analysis model. LSIZE is the log market value of equity at June of year t-1. LBTM is the log book equity to market equity (Fama and French, 1992).
BHRET6 is the 6-month buy-and-hold returns over January (t) to June (t). BHRET36 is the 3-year buy-and-hold returns over July (t-3) to June (t). In case of controls we have
ASSETG as a measure of the total asset growth that the year-to-year change in total assets [(TASSETSt-TASSETSt-1)/TASSETSt-1] (Cooper, et al. 2008). ACCR (accruals)
equals the change in accounts receivables plus the change in inventories plus the change in the other current assets minus the change in account payable minus the change in other
current liabilities minus depreciation and scaled by lagged total assets as defined by Polk and Sapienza (2009). NOATA represents cumulative accruals as defined by Hirshleifer et
al. (2004). The measure of profitability (ROA) is net income (COMPUSTAT item A178) divided by the total assets (COMPUSTAT item A6). I_A is the sum of change in
inventories and change in gross property, plant, and equipment (PPE) scaled by lagged total assets. The table shows average slope coefficients and t-statistics with
Model 1 0.1926 -0.0876 -0.0099 -0.0181 -0.0122 -0.0029
0.011
6.49 -8.21 -1.21 -3.16 -5.43 -0.49
Model 2 0.2071 -0.0544 -0.0063 -0.0088 -0.0089 -0.0052 -0.0143
0.012
7.16 -3.49 -0.73 -1.06 -3.88 -0.85 -1.12
Model 3 0.2599 -0.0730 0.0034 0.0204 -0.0088 -0.0082
-0.0962
0.019
5.57 -6.46 0.34 2.00 -3.19 -1.46
-2.87
Model 4 0.1786 -0.0784 -0.0043 -0.0171 -0.0105 -0.0057
-0.0698
0.014
5.77 -7.56 -0.53 -2.95 -4.84 -0.98
-5.38
Model 5 0.1833 -0.0698 -0.0160 -0.0159 -0.0146 -0.0104
-0.1699 -0.0002
0.027
6.22 -9.49 -2.84 -3.59 -6.98 -2.03
-2.84 -2.19
Model 6 0.3717 -0.0662 -0.0099 -0.0027 -0.0061 -0.0051
0.0503 -0.0384
0.034
5.95 -7.89 -1.34 -0.50 -2.68 -0.90
1.79 -6.78
Model 7 0.1555 -0.0692 -0.0054 -0.0247 -0.0075 -0.0052
-0.0505 -0.0741 0.026
6.34 -7.58 -0.70 -4.78 -2.99 -0.54 -5.72 -2.36
57
Table 7: Asset Growth Rates and Industry Effect.
This table reports industry effect in the return predictability of the latent growth factors and other growth measures. The dependent annual compounding geometric future returns
are regressed on the base variables for the period of 1985 to 2009. The data set is partitioned into 11 industry sub-samples.
(1)
(2)
Model 1 includes only latent asset growth factors as independent variables and Model 2 includes other accounting growth variables (ASSETG, NOATA, ROA), log of market size,
and log book-to-market ratio (see, table 2 header for construction of these variables). The 10 Fama-French Industry groups are defined on the Ken French's website,
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html). NoDur represent consumer non-durables that include food, tobacco, textiles, apparel, leather, and toys
industries. Durbl represent consumer durables that include cars, TV’s, furniture, and household appliances. Manuf represents manufacturing that includes machinery, trucks,
planes, office furniture, paper, and commercial printing industries. Energy includes oil, gas, and coal extraction and products industries. Chems includes chemicals and allied
products industries. BusEq represents business equipment that includes computers, software, and electronic equipment industries. Telcm includes telephone and television
transmission industries. Utils represents utilities. Shops include wholesale, retail, and some services (laundries, repair shops) industries. Hlth include healthcare, medical
equipment, and drug industries. Others include mines, construction, building material, transportation, hotels, business service, and entertainment industries. The table shows
average slope coefficients and t-statistics with significance as boldface below 5%.
