faj.v69.n1.1

7/29/2019 faj.v69.n1.1

1/18

Januay/Febuay 2013 www.cfapubs.og 53

Financial Analysts JournalVolume 69 Number 1

2013 CFA Institute

Change in Cash-Holding Policies and StockReturn Predictability in the Cross Section

William r. sodjahin

The author ound that stocks with a positive change in company cash holdings have signifcantly higher risk-adjusted returns than stocks with a negative change in cash holdings (CCH). Moreover, the return predictive

power o CCH is (1) distinct rom the eect o cash holdings (CH), (2) absent among cash-rich companies, (3)stronger among small-cap stocks, and (4) limited to non-January months. The CCH anomaly appears to bemore contaminated than the CH eect by mispricing.

S

everal studies have ocused extensively onthe determinants o a companys cash hold-

ings and the holdings time-series proper-ties over time.1 Only recently, however, haveexcess cash holdings been linked to stock returns(Simutin 2010) and a precautionary savings policyto expected returns (Palazzo 2012). Simutin docu-mented a positive relationship between excess cashand subsequent stock returns. Palazzo showed apositive correlation between cash-holding poli-cies and average realized returns. Intuitively, bothauthors argued that riskier companies acingcostly external unding maintain higher levelso cash to nance their growth, and thus, a com-panys cash holdings can be linked to risk andexpected returns. Because risk is time varying anda companys investment opportunities are everchanging, a change in cash-holding policy canalso be viewed as a proxy or the arrival o newinvestment opportunities (or the disappearanceo existing opportunities). This dynamic impliesa positive relationship between a change in cash-holding policy and subsequent stock returns ascompensation or the risk that accompanies a newinvestment opportunity. Note that because o thetime to build that characterizes new investments,there is a lag between the arrival o the new invest-

ment opportunity and its undertaking or realiza-tion (see, e.g., Bar-Ilan and Strange 1996; Pindyck1991; MacRae 1989; Wheaton 1987).

Changes in cash holdings are, by denition,dierent rom the level o cash holdings (Palazzo2012) or excess cash holdings (Simutin 2010) in thesense that a company can increase its cash hold-ings and still have a low cash-holding level or acompany can decrease its cash holdings and still

maintain an excess cash-holding level. In my study,I ocused on the variation in the cash holdings o

companies as a signal or the emergence or disap-pearance o risky investment opportunities.

Discussion o fndings. The objective o mystudy was threeold. First, I investigated howchanges in companies cash holdings are related tosubsequent returns. In other words, can changesin cash holdings (CCH) predict stock returns?Using both FamaMacBeth (1973) cross-sectionalregressions (standard and risk adjusted) and port-olio analysis, I explored the return predictabilityo CCH, which is the cash-to-assets ratiolevelo cash holdings (CH)minus the lagged cash-to-

assets ratio (see, e.g., Bates, Kahle, and Stulz 2009). Iound that CCH has a strong, unconditional returnpredictive power.2 This power remains strongater controlling or such standard stock returnpredictors as size, book-to-market, momentum,asset growth, illiquidity ratio, and idiosyncraticvolatility in both cross-sectional regressions andportolio analysis. For example, the FamaFrench(1993) three-actor alpha (the Carhart [1997] our-actor alpha) o the strategy that goes long the topCCH quintile and short the bottom CCH quintileis 0.27% (0.20%) a month; this spread in alphas isstrongly signicant and substantially higher or

non-January months, at 0.34% (0.27%). Importantly,a CCH-based actor-mimicking portolio has aSharpe ratio o 0.15, higher than that o the mar-ket actor (0.09) or the SMB and HML actors (0.09and 0.13, respectively) o Fama and French (1993)and only slightly lower than that o the momen-tum actor (0.16). For the non-January months, aCCH-based actor-mimicking portolio has a muchhigher Sharpe ratio (0.21), which is still higher thanthat o the market, SMB, and HML actors (0.08,0.04, and 0.10, respectively) and very close to thato the momentum actor (0.22).

William R. Sodjahin is a research scientist at the SternSchool o Business, New York University.

7/29/2019 faj.v69.n1.1

2/18

Financial Analysts Journal

54 www.cfapubs.og 2013 CFA Intitute

Second, how distinct is the CCH eect rom theCH eect? Does the CCH eect exist or both cash-constrained and cash-rich companies? Palazzo(2012) showed that high-CH companies have moregrowth opportunities. From a rational viewpoint, Ihypothesized that CCH refects changes in utureinvestment opportunities. I conrmed this hypoth-

esis by showing empirically that companies thatincrease their cash holdings have signicantlyhigher (vis--vis companies that decrease their cashholdings) subsequent investment growth or up totwo years. Moreover, the cash holdings o cash-richcompanies may be less elastic to variations in uturerisky investment opportunities than the cash hold-ings o cash-constrained companies. Intuitively, atraditionally cash-rich company will not indenitelyincrease its cash holdings in response to the arrivalo new investment opportunities (because the levelo its cash holdings is already high), as opposed to

a traditionally low-cash company, which is morelikely to adjust its cash-holding policy to changesin its risky investment opportunities. Thereore, theCCH eect is expected to be stronger or exclusivelyconcentrated among cash-constrained (low-cash)companies. I ound evidence consistent with thesepredictions. In particular, I observed that CCHremains signicant ater controlling or CH (andvice versa) in cross-sectional regressions and port-olio analysis. Moreover, the spreads in subsequentreturns rom a longshort strategy based on CCHare concentrated exclusively in the universe o cash-constrained (low-CH) companies. Furthermore, myresults show that the CCH eect is stronger amongsmaller-cap companies.

Third, the CCH and CH eects may not becompletely rational. Hence, I gauged the extent towhich CCH and CH can, at least in part, be attrib-uted to mispricing caused by various investors

behavioral biases.3 In other words, I studied theimplications o arbitrage costs or both the CCHand the CH eects. Shleier and Vishny (1997) sug-gested that arbitrage is risky, costly, and limited.Because arbitrageurs are generally poorly diversi-ed, any nonsystematic risk would add consid-

erably to the total risk o their overall portoliosand hinder their arbitrage activities. And becausethe costs o arbitrage may outweigh its benets,it would be dicult or investors to ully arbi-trage away any gain associated with a companysCCH or CH eect so long as the eect is driven

by mispricing. Thus, i the CCH or CH eectis due to mispricing, it should be concentratedamong stocks that are more dicult to arbitrage(that have a higher arbitrage risk). Arbitrage costscomprise transaction costs and holding costs (seePonti 2006). Following Li and Sullivan (2011),

among others, I ocused on idiosyncratic volatility(IVol) as a proxy or the holding-cost componento arbitrage costs.4 Indeed, transaction costs wereunlikely to be weighted signicantly as stronglimits to arbitrage because my CCH and CH port-olios were rebalanced only annually. In short, theextent to which the CCH and CH eects are con-

centrated in high-IVol stocks refects the degreeto which mispricing is important in explainingthese eects. My ndings reveal the pervasive-ness o both CCH and CH anomalies across IVolgroups, though the CCH eect is substantiallyhigher among high-IVol stocks. I ound that theCCH eect is more contaminated (though onlypartly) by mispricing than is the CH eect. Finally,I documented a surprisingly negative CCH eector January months and showed that this eect isentirely driven by mispricing.

Data Description and AnalysisI obtained stock return data rom the CRSPMonthly Stock File and accounting data rom theCRSP/Compustat Merged Industrial Annual File,via Wharton Research Data Services (WRDS).I included in my sample (July 1965December2010) all ordinary common shares (share codes 10and 11 in CRSP) traded on the NYSE, Amex, andNASDAQ with available accounting and returndata. I excluded regulated utilities (SIC codes49004999) and nancial companies (SIC codes60006999) and also excluded observations con-

cerning suspended, halted, or nonlisted shares (i.e.,exchange codes lower than 1 and higher than 3).Following prior research (see, e.g., Li and Sullivan2011), I ocused on companies with a Decemberscal year end. I obtained one-month T-bill ratesrom Kenneth Frenchs website.5

The main variable o interest in my study wasthe change in cash holdings (CCH) o a company,measured as the companys cash-to-asset ratio(Compustats item CHE divided by Compustatsitem AT) minus the lagged cash-to-asset ratio(see, e.g., Bates, Kahle, and Stulz 2009; Palazzo2012). LogAT is the natural log o the companys

total assets. BM is the natural log book-to-marketequity as the log book equity or the scal year end-ing in t minus the log market equity at the end oDecember in year t. Book equity is the Compustat

book value o stockholders equity (Compustatannual item SEQ) plus balance-sheet deerred taxes(item TXDB) and investment tax credit (item ITCI, iavailable) minus the book value o preerred stock.To estimate the book value o preerred stock, Iused redemption (item PSTKRV), liquidation (itemPSTKL), or par value (item PSTK), in that orderand depending on availability. Market equity is the

7/29/2019 faj.v69.n1.1

3/18

Change in Cash-Holding Policies and Stock Return Predictability in the Cross Section


CRSP price per share times the number o sharesoutstanding (SHROUT). I measured IVol (idiosyn-cratic volatility) as the standard deviation o theresidual values rom the FamaFrench three-actormodel by regressing daily excess stock returns onrelevant actors. Excess returns are in excess o theone-month T-bill rate. I used daily stock and ac-

tor returns in the prior JulyJune period to estimateIVol or each month.

I winsorized all accounting variables at the 1stand 99th percentiles to mitigate the eect o outli-ers. Panel A oTable 1 reports summary statisticso the sample. The mean o the key variable (CCH)is negative (0.4%), with a standard deviation o8.40%, which indicates a relatively important varia-tion in CCH across companies and over time. TheCCH distribution is skewed toward companiesthat reduce their cash holdings. The mean o CCHis our times larger (in absolute value) than the

median. CH has a larger mean in absolute value(16%), with a larger standard deviation, than CCH.CH is skewed toward companies with high cashholdings but is less skewed than CCH. The mean o

CH is about two times the median. LogAT, BM, andIVol display less-skewed patterns.

Panel B presents the samples paired, cross-sectional Spearmans rank correlations betweenthe key variables in Panel A. The change in cashholdings is not strongly correlated with the level ocash holdings. The correlation coecient between

CCH and CH is not very close to zero but is lessthan 25% (0.21). The correlation between CCH andLogAT is positive but very weak (0.02). The corre-lation between CCH and BM is also positive andvery weak (0.01). In comparison, the correlations

between CH and LogAT and between CH and BMare negative and greater (0.23 and 0.19, respec-tively). As expected rom the literature (see, e.g.,Li and Zhang 2010), the greatest correlation is theone between IVol and LogAT (0.64). In short, thecorrelation matrix reveals that large-asset compa-nies tend to have a low cash-holding ratio, a posi-

tive change in cash holdings, a low book-to-marketratio, and low idiosyncratic volatility.Panel C reports uture investment ratios and

investment growth or years t, t + 1, t + 2, and t +

Table 1. Summary Statistics, July 1965December 2010

Mean Standard Deviation Median 10th 90th

A. Descriptive statistics

CCH 0.004 0.084 0.001 0.086 0.073

CH 0.157 0.201 0.072 0.010 0.450

LogAT 5.003 2.025 4.855 2.458 7.779

BM 0.615 0.953 0.569 1.776 0.514IVol 3.369 2.332 2.725 1.360 6.153

CCH CH LogAT BM IVol

B. Correlation matrix (Spearman)

CCH 1.00

CH 0.21 1.00

LogAT 0.02 0.23 1.00

BM 0.01 0.19 0.04 1.00

IVol 0.02 0.12 0.64 0.09 1.00

Investment Ratio Investment GrowthYear

tYeart + 1

Yeart + 2

Yeart + 3

Yeart

Yeart + 1

Yeart + 2

Yeart + 3

C. Change in cash holdings, uture investment ratio, and investment growth

Low CCHt (< 0) 0.29 0.25 0.23 0.22 0.31 0.11 0.08 0.13

Middle CCHt ( 0) 0.21 0.20 0.20 0.20 0.12 0.09 0.08 0.09

High CCHt (> 0) 0.23 0.23 0.24 0.22 0.10 0.19 0.16 0.12

High low 0.06 0.02 0.01 0.00 0.21 0.08 0.08 0.01

t-Statistic 8.82 2.44 1.06 0.09 10.35 4.00 4.98 0.57

Note: The t-statistics are derived rom the NeweyWest (1987) procedure with one lag.

7/29/2019 faj.v69.n1.1

4/18



3 by change in cash holdings o year t (CCHt) orcompanies that survived or the entire three subse-quent years. Following Xing (2008), I measured theinvestment ratio as the ratio o capital expenditures(Compustat item CAPX) to the net book value oxed assets (Compustat item PPENT) at the begin-ning o each scal year. Investment growth is the

growth rate or a companys capital expenditures.The summary statistics in Panel C show that com-panies that increase their cash holdings have sig-nicantly lower contemporaneous investmentgrowth (compared with companies that decreasetheir cash holdings), but they have signicantlyhigher subsequent investment growth or up totwo years. Moreover, these companies have verysignicantly lower contemporaneous investmentratios and less signicantly lower investment ratiosin the rst subsequent year. In the second and thirdsubsequent years, the dierences between the

investment ratios o companies that increase theircash holdings and the investment ratios o compa-nies that reduce their cash holdings become insig-nicantly dierent rom zero. Overall, these resultssuggest that CCH does indeed signal uture invest-ment growth and can proxy or changes in uturerisky investment opportunities.

Change in Cash Holdings and

Subsequent ReturnsI then examined FamaMacBeth cross-sectionalregressions and conducted a portolio analysis in

order to test the CCH eect on returns.

FamaMacBeth Cross-Sectional Regressions.I investigated the cross-sectional relation betweenthe change in a companys cash holdings and sub-sequent stock returns. CCH or a scal year end t 1 becomes available or monthly FamaMacBeth(1973) regressions the seventh month (July o yeart) ater the scal year end. To begin my analysis,I conducted standard FamaMacBeth regressionso excess returns (risk unadjusted) on CCH aloneand then on CCH while controlling or such pre-dictors o stock returns as market capitalization

(ME), book-to-market (BM), momentum (MOM),asset growth (AG; see Cooper, Gulen, and Schill2008), illiquidity (Illiq; see Amihud 2002), andmonthly idiosyncratic volatility (IVolm; see Ang,Hodrick, Xing, and Zhang 2006). ME is the logmarket capitalization at the end o June in yeart, and BM is the log book-to-market equity as thelog book equity or the scal year end t 1 minusthe log market equity at the end o December inyear t 1. Book equity is the Compustat bookvalue o stockholders equity (Compustat annualitem SEQ) plus balance-sheet deerred taxes (item

TXDB) and investment tax credit (item ITCI, iavailable) minus the book value o preerredstock. To estimate the book value o preerredstock, I used redemption (item PSTKRV), liquida-tion (item PSTKL), or par value (item PSTK), inthat order and depending on availability. Marketequity is the CRSP price per share times the num-

ber o shares outstanding (SHROUT). MOM isthe log prior-12-month returns (with a one-monthgap between the holding period and the currentmonth). AG is the percentage change in totalassets (Compustats item AT) rom scal year t 2to scal year t 1. Illiq is the average ratio o abso-lute daily return to trading value over the month,computed as ollows:

IlliqD

r

iced volumeti

ti

d ti

d ti

d

Dti

=

=

1

1

,

,

,Pr

(1)

where

rd ti

, = the return or stock i on day d omonth t

Priced volume = the priced daily trading volumeD = the number o days in the

month or which data areavailable

IVolm is the monthly standard deviation o theresidual values rom the FamaFrench three-actormodel, obtained by regressing daily excess stockreturns on relevant actors and dened as

var

t

i

( )in the ollowing equation:6

r r MKT SMB

HML

ti

fi

MKTi

t SMBi

t

HMLi

t ti

= + +

+ +

,

(2)

where

rt

i = the daily return o stock i

r = the risk-ree rate

MKT= the excess return on the market portolio

SMB = the FamaFrench (1993) size actor

HML = the FamaFrench (1993) value actor

To check or robustness, I also conducted therisk-adjusted FamaMacBeth (FM) regressions byreplacing excess returns with risk-adjusted returnsas dependent variables. I based the risk adjustmenton the FamaFrench (1993) three-actor model andthe Carhart (1997) our-actor model (an extensiono the FamaFrench model that contains an addi-tional momentum actor). I used the Dimson (1979)procedure with one lag to adjust or thin trading(see, e.g., Brennan, Chordia, and Subrahmanyam1998). I estimated the actor loadings by using a60-month rolling window.

7/29/2019 faj.v69.n1.1

5/18



Specically, I estimated the ollowing equation:

r CCH ME BM

MOM AG I

t t t t t t t t

t t t t t

+= + + +

+ + +

1 0 1 2 3

4 5 6

, , , ,

, , , llliq

IVolm

t

t t t+ + + 7 1, ,

(3)

where rt+1 is the excess returns, risk-adjusted returns

based on the FamaFrench three-actor model(FF3RA), or risk-adjusted returns based on the

Carhart our-actor model (FF4RA). The results o thecross-sectional regressions are reported in Table 2.

The reported coecients are the time-seriesaverages rom monthly cross-sectional regressions,and the t-statistics are based on the time series omonthly coecient estimates. I used the NeweyWest (1987) procedure to correct or potential serial

correlation. Model 1 presents the results o the FMregression based on excess returns (ER), which

Table 2. FamaMacBeth Cross-Sectional Regressions of Monthly Percentage Excess Returnsand Risk-Adjusted Returns on Changes in Cash Holdings, July 1965December 2010(t-tatitic in paenthee)

Model INT CCH CH ME BM MOM AG Illiq IVolm

A. Dependent variable = ER

1 0.92 1.25***

(3.06) (3.21)

2 0.83 0.47

(2.76) (1.22)

3 3.05 0.93*** 0.15*** 0.16*** 0.30* 0.56*** 0.23*** 0.20***

(4.87) (2.80) (3.72) (2.08) (1.75) (6.39) (2.80) (5.15)

4 2.74 0.94*** 0.14*** 0.22*** 0.27 0.46*** 0.27*** 0.20***

(4.36) (3.00) (3.39) (3.32) (1.55) (5.83) (3.27) (5.39)

B. Dependent variable = FF3RA

5 0.15 1.12***

(2.32) (3.12)

6 0.04 0.90***(0.59) (3.36)

7 2.61 0.93*** 0.16*** 0.12** 0.09 0.40*** 0.28*** 0.35***

(7.15) (2.77) (5.78) (1.99) (0.62) (4.30) (3.52) (9.73)

8 2.48 0.74*** 0.15*** 0.07 0.07 0.29*** 0.29*** 0.35***

(6.95) (3.00) (5.78) (1.36) (0.45) (3.43) (3.38) (9.93)

C. Dependent variable = FF4RA

9 0.20 1.24***

(3.17) (3.40)

10 0.12 0.63**

(1.82) (2.41)

11 2.40 0.96*** 0.12*** 0.03 0.20 0.40*** 0.29*** 0.40***

(6.62) (2.86) (4.36) (0.40) (1.36) (4.65) (3.32) (10.61)

12 2.31 0.59** 0.11*** 0.01 0.18 0.30*** 0.28*** 0.41***

(6.63) (2.37) (4.45) (0.21) (1.23) (3.79) (3.16) (10.84)

*Signicant at the 10% level.**Signicant at the 5% level.

***Signicant at the 1% level.

7/29/2019 faj.v69.n1.1

6/18



includes only the CCH as an explanatory variable.For comparison, Model 2 regresses ER on CH (thelevel o cash holdings) alone. Models 3 and 4 areversions o Models 1 and 2, respectively, that controlor the various stock return predictors in Equation3. The dependent variables or Models 5 through 12are risk-adjusted returns (FF3RA and FF4RA).

Model 1 shows that CCH is a signicant pre-dictor o uture stock returns. As expected, the coe-cient on CCH is positive and statistically signi-cant (at the 1% level), with a t-statistic o 3.21. CCHremains a strong stock return predictor (still sig-nicant at the 1% level) ater controlling or stockreturn predictors in Equation 3. In contrast, theunconditional eect o CH is insignicant. Indeed,the coecient on CH is insignicant in Model 2.However, the coecient on CH becomes stronglysignicant when the control variables are included.In line with prior literature, the control variables

display the expected signs and are all strongly sig-nicant in Models 3 and 4, except or MOM, whichis signicant only at the 10% level in Model 3 andinsignicant in Model 4.

The results rom the risk-adjusted FamaMacBeth regressions conrm the strong predictivepower o CCH. The coecient on CCH is signicantat the 1% level in all specications (see Models 5,7, 9, and 11). Contrary to Model 2 (risk-unadjustedregressions), the coecient on CH in Models 6 and10 (risk-adjusted regressions) is signicant. The CHeect becomes weaker when the risk adjustment is

based on the Carhart our-actor model. The coe-

cient on CH is signicant at the 5% level in bothspecications (Models 10 and 12).

To make sure that my previous results werenot driven by any biases arising rom the Januaryeect, I repeated the analysis in Table 2 without

January returns. Indeed, Keim (1983) documentedthat much o the abnormal returns to small com-panies occur in January; moreover, Li and Zhang(2010) showed that the abnormal corporate invest-ment eect o Titman, Wei, and Xie (2004) is drivenentirely by the January eect. The results o the FMregressions without the January returns are pre-

sented inTable 3

. The CCH eect becomes evenstronger in all specications when January returnsare excluded. In contrast, the CH eect slightlyweakens, especially when the risk adjustmentis based on the Carhart our-actor model. (SeeAppendix A or the results o standard and risk-adjusted FamaMacBeth regressions or Januarymonths only.)

Portfolio Analysis. I then investigated whethersorting stocks by CCH alone or by CCH while con-trolling or the various company characteristicsin Equation 3 generates signicant variation in

portolio returns. The purpose o controlling orthese characteristics was to examine the robust-ness o my results with the CCH portolio orma-tion strategy to various cross-sectional risk actors.For comparison, I also report the results o myCH-based portolio analysis.

Panel A o Table 4 reports various summary

statistics or value-weighted quintile portoliossorted on the basis o CCH or all months. Therst two columns are the mean and Sharpe ratioo monthly excess returns. The excess returns aremeasured in percentages and are calculated as theraw stock return less the risk-ree rate. The columnslabeled -CAPM, -FF3, and -FF4 reportthe time-series alphas o these portolios relativeto the capital asset pricing model (CAPM), theFamaFrench three-actor model, and the Carhartour-actor model (the FamaFrench model aug-mented with the momentum actor), respectively.

In addition, Panel A reports the excess returns andalphas or a zero-cost longshort spread, or tradingprotability, o the CCH trading strategy. This port-olio represents the dierence between the highest-and lowest-ranked quintiles. Panel C replicatesPanel A or non-January months only. Consistentwith the results o FamaMacBeth cross-sectionalregressions, the monthly excess returns increase,nearly monotonically, rom 0.68% (0.30% or non-

January months) or Portolio 1 to 0.90% (0.62%or non-January months) or Portolio 5. The di-erences between Portolios 5 and 1 are stronglysignicant at the 1% level. Moreover, higher-CCHportolios exhibit signicantly higher CAPM, FF3,and FF4 alphas. The Gibbons, Ross, and Shanken(GRS 1989) test, which indicates the null hypoth-esis that all alphas o each model are jointly zero, isstrongly rejected or the three models. Controllingor the market actor and the FamaFrench actorsexacerbates the 51 spread (rom 0.22% to 0.24%and 0.27% a month, respectively), whereas control-ling or the FamaFrench and momentum actorsdecreases the 51 spread to 0.20%. These alphaspreads are strongly signicant and substantiallyhigher or non-January months0.34%, 0.34%, and

0.27% a month, respectively.Panel B o Table 4 reports the results o a single

sort on the CH (Panel D reports the same results ornon-January months only). Even though the excessreturns increase monotonically across the quintileportolios, the dierences between Portolios 5 and1 are lower (especially or non-January months)and insignicant. These results are consistent withthe results o the unconditional FM regressions inTables 2 and 3. The spread in CAPM alphas is alsovery small and insignicant. The GRS test is notrejected or the CAPM. Moreover, in line with the

7/29/2019 faj.v69.n1.1

7/18



results o the conditional FM regressions in Tables2 and 3, the spreads in FF3 and FF4 alphas betweenPortolios 5 and 1 are all signicant at the 1% level.In addition, the GRS test is strongly rejected or

both models.To make sure that the spread in average

returns on the CCH portolios observed in Table4 was not due to the eects o various character-istics controlled or in Tables 2 and 3, I examined

the robustness o the results with the CCH porto-lio ormation strategy to the eects o the ollow-ing characteristics: ME, BM, MOM, AG, Illiq, andIVolm. I rst sorted stocks into ve quintiles on the

basis o each characteristic. Then, within each quin-tile, I sorted stocks into ve quintiles on the basiso their CCH. Ater orming the 5 5 portolios,I averaged the returns o each CCH quintile overthe ve characteristic portolios. Thereore, these

Table 3. FamaMacBeth Cross-Sectional Regressions of Monthly Percentage Excess Returnsand Risk-Adjusted Returns on Changes in Cash Holdings (No January Returns),July 1965December 2010(t-tatitic in paenthee)



1 0.55 1.86***

(1.80) (4.77)

2 0.46 0.41

(1.54) (1.03)

3 1.95 1.31*** 0.07* 0.13 0.57*** 0.50*** 0.16** 0.27***

(3.10) (3.87) (1.73) (1.58) (3.20) (5.60) (1.97) (6.80)

4 1.65 0.80** 0.05 0.19*** 0.54*** 0.41*** 0.18** 0.27***

(2.61) (2.42) (1.32) (2.70) (3.05) (5.21) (2.42) (7.00)


5 0.01 1.68***

(0.09) (4.27)

6 0.09 0.77***

(1.27) (2.73)

7 2.00 1.30*** 0.11*** 0.12* 0.33** 0.30*** 0.21** 0.38***

(5.29) (3.73) (3.89) (1.80) (2.25) (3.33) (2.57) (10.73)

8 1.89 0.65** 0.10*** 0.06 0.30** 0.22*** 0.21** 0.39***

(5.02) (2.51) (3.74) (1.12) (2.03) (2.66) (2.49) (10.91)


9 0.08 1.83***

(1.36) (4.50)

10 0.03 0.42

(0.43) (1.52)

11 1.77 1.40*** 0.06** 0.00 0.39** 0.33*** 0.22** 0.43***

(4.36) (3.85) (2.04) (0.00) (2.53) (3.84) (2.55) (11.36)

12 1.71 0.47* 0.06** 0.04 0.36** 0.25*** 0.20** 0.44***

(4.34) (1.80) (2.03) (0.66) (2.38) (3.10) (2.26) (11.60)



7/29/2019 faj.v69.n1.1

8/18



Table 4. Portfolios Sorted on Capital Changes in Cash Holdings, July 1965December 2010

Mean Sharpe Ratio -CAPM -FF3 -FF4

A. Portolios sorted on CCH

1 0.68 0.10 0.14 0.06 0.16

2 0.86 0.13 0.37 0.09 0.24

3 0.73 0.17 0.26 0.06 0.08

4 0.93 0.15 0.45 0.15 0.27

5 0.90 0.14 0.38 0.21 0.36

5 1 0.22 0.15 0.24 0.27 0.20

t-Statistic 3.43 3.82 4.26 3.23

GRS 8.15 5.91 7.56

p(GRS) 0.00 0.00 0.00

B. Portolios sorted on CH

1 0.71 0.12 0.23 0.14 0.02

2 0.75 0.13 0.26 0.10 0.05

3 0.82 0.14 0.31 0.02 0.184 0.88 0.14 0.36 0.19 0.40

5 0.93 0.13 0.38 0.37 0.58

5 1 0.22 0.05 0.15 0.52 0.56

t-Statistic 0.99 0.71 2.85 3.20

GRS 1.95 2.89 6.53

p(GRS) 0.08 0.01 0.00

C. Portolios sorted on CCH (non-January months)

1 0.30 0.05 0.17 0.24 0.01

2 0.56 0.10 0.12 0.02 0.12

3 0.49 0.09 0.08 0.12 0.02

4 0.68 0.12 0.26 0.07 0.21

5 0.62 0.10 0.17 0.10 0.26

5 1 0.32 0.21 0.34 0.34 0.27

t-Statistic 7.19 7.84 8.13 6.36

GRS 7.35 6.52 5.66

p(GRS) 0.00 0.00 0.00

D. Portolios sorted on CH (non-January months)

1 0.41 0.07 0.01 0.24 0.08

2 0.45 0.08 0.02 0.19 0.033 0.53 0.09 0.09 0.07 0.11

4 0.58 0.10 0.12 0.05 0.27

5 0.57 0.08 0.08 0.16 0.36

5 1 0.16 0.04 0.09 0.40 0.44

t-Statistic 0.98 0.55 3.06 3.44

GRS 0.70 2.36 3.70

p(GRS) 0.62 0.04 0.00

Notes: GRS denotes the Gibbons, Ross, and Shanken (1989) test, andp(GRS) is the associatedp-value. Panels Cand D replicate Panels A and B, respectively, or non-January months.

7/29/2019 faj.v69.n1.1

9/18



quintile CCH portolios control or dierences ineach characteristic.

Panel A o Table 5 reports the FF3 and FF4alphas, the dierences in FF3 and FF4 alphas

between the quintile portolios with the highestand lowest CCH, and t-statistics indicating theirstatistical signicance. All the portolios sorted on

CCH are value weighted. The results show thatcontrolling or characteristics reduces (or in thecase o BM, increases) only slightly the magnitudeo the 51 spread in portolio FF3 and FF4 alphas.Moreover, the characteristic-controlled dierences(the 51 CCH spread) are still highly signicant.Thus, these characteristics cannot account or thespread in returns resulting rom dierences inCCH. Panel B replicates the same analysis in PanelA or CH. The results indicate that controlling orcharacteristics reduces or increases only slightlythe magnitude o the 51 CH spread in portolio

FF3 and FF4 alphas, which all remain strongly sig-nicant. Hence, the CH eect is not explained bythese characteristics either.

How Distinct Are the CCH and CH Effects? Isthere any source o risk related to CCH that is notcaptured by CH, or vice versa? Palazzo (2012)showed that high-CH companies have more growthopportunities than low-CH companies. I arguedearlier that CCH signals a change in growth oppor-tunities and provided empirical evidence to backthis argument. Indeed, I showed that companiesthat increase their cash holdings have signicantly

higher (compared with companies that decreasetheir cash holdings) subsequent investment growthor up to two years. I then went urther by explicitlytesting whether the CCH eect remains ater con-trolling or CH, and vice versa. I rst ran a FamaMacBeth cross-sectional regression o subsequentreturns on both CCH and CH, controlling or othercharacteristics. Next, using CCH and CH as charac-teristics, I repeated the explicit double-sorted char-acteristic controls (a sort on CCH while controllingor CH, and vice versa). Finally, I explored whetherthe CCH is pervasive across dierent levels o cashholdings (cash-constrained versus cash-rich com-

panies). Table 6 reports the results o the rst twostages o my analysis.

The results rom the FM regressions in PanelA show that the coecient on CCH is stronglysignicant at the 1% level or non-January monthsand marginally signicant at the 10% level orall months. Conversely, the coecient on CH ismarginally signicant at the 10% level or non-

January months and signicant at the 5% levelor all months. Panel B reveals that controlling orCH decreases the magnitude o the CCH spreadin the portolio FF3 alpha, which does, however,

remain signicant at the 1% level. The magnitudeo the CCH spread in the portolio FF4 alpha alsodecreases and becomes marginally signicantat the 10% level. However, the results or non-

January months show that ater controlling or CH,the magnitude o the CCH spread in portolio FF3and FF4 alphas is higher and remains strongly sig-

nicant at the 1% level. These results suggest thatthe CCH eect is not explained by CH, especiallywhen the January returns are excluded. Panel Balso shows that controlling or CCH decreases themagnitude o the CH spread in portolio FF3 andFF4 alphas, which remain signicant at the 1%level. These alphas decrease urther in magnitudeand become signicant at the 5% level or non-

January months.Table 7 reports the results o the third step

o my analysis. Using median splits (low andhigh) based on the CH, I divided the sample each

month. Panel A shows that the spread in alphasrom a longshort strategy based on the CCH isexclusively concentrated in the universe o cash-constrained (low-CH) companies. Indeed, the CCHspread in portolio FF4 alphas or high-CH stocks isvery close to zero (0.02% or all months and 0.08%or non-January months) and insignicant. In con-trast, the same alphas are much larger (0.19% orall months and 0.25% or non-January months)and strongly signicant (at the 1% level or non-

January months and the 5% level or all months)or low-CH stocks. Thus, the CCH anomaly exists

predominantly among those stocks with relativelylow cash-holding levels.Panel B o Table 7 shows that the CCH eect

is stronger among smaller-cap companies (thosewith an ME below the median). Consistent withPalazzo (2012), I ound (in unreported results) thatthe CH eect is also stronger among smaller-capcompanies.

Overall, in line with a rational viewpoint, (1)both the CCH and the CH eects exist and are dis-tinct, (2) the CCH eect is present only among cash-constrained companies, and (3) the CCH eect isstronger in the universe o smaller-cap companies

more likely to ace nancial constraints.

Implications of Arbitrage Costs for

Both CCH and CH EffectsFinally, I looked or any evidence in support omispricing. Indeed, the CCH and CH eects maynot be completely rational; recently, many papershave stressed the importance o limits to arbi-trage (i.e., mispricing) in explaining investment,asset growth, and accrual anomalies (see, e.g., Liand Zhang 2010; Li and Sullivan 2011). Does the

7/29/2019 faj.v69.n1.1

10/18



Table 5. Alphas of Portfolios Sorted on Changes in Cash Holdings: Controlling forCharacteristics (Robustness Checks), July 1965December 2010(t-tatitic in paenthee)

1 2 3 4 5 5 1

A. Alphas o portolios sorted on CCH

ME

-FF3 0.01 0.08 0.04 0.18 0.27 0.26

(0.13) (1.03) (0.50) (2.16) (3.14) (3.93)

-FF4 0.24 0.22 0.19 0.32 0.44 0.20

(2.26) (2.59) (2.19) (3.78) (3.34) (3.04)

BM

-FF3 0.08 0.09 0.07 0.11 0.26 0.34

(0.88) (1.33) (0.88) (1.51) (3.26) (5.66)

-FF4 0.15 0.13 0.07 0.23 0.42 0.27

(1.51) (1.33) (0.92) (3.22) (4.42) (4.33)

MOM

-FF3 0.01 0.03 0.05 0.09 0.18 0.19

(0.14) (0.44) (0.63) (1.22) (2.20) (3.40)

-FF4 0.18 0.20 0.11 0.26 0.38 0.20

(1.97) (2.42) (1.47) (3.35) (4.00) (3.29)

AG

-FF3 0.02 0.03 0.02 0.04 0.27 0.25

(0.21) (0.47) (0.30) (0.63) (3.31) (3.60)

-FF4 0.24 0.19 0.12 0.20 0.41 0.17

(2.37) (2.38) (1.53) (4.40) (4.15) (2.56)

Illiq-FF3 0.02 0.07 0.04 0.13 0.24 0.26

(0.21) (0.90) (0.44) (1.59) (2.76) (4.01)

-FF4 0.21 0.24 0.13 0.31 0.42 0.21

(1.99) (2.75) (1.40) (3.48) (4.07) (3.14)

IVol

-FF3 0.05 0.03 0.07 0.15 0.19 0.24

(0.52) (0.36) (0.84) (1.88) (2.04) (3.37)

-FF4 0.17 0.20 0.10 0.33 0.36 0.19

(1.71) (2.30) (1.06) (3.86) (2.59) (4.33)

B. Alphas o portolios sorted on CH

ME

-FF3 0.11 0.09 0.08 0.21 0.46 0.56

(1.24) (0.91) (0.94) (2.15) (3.16) (3.16)

-FF4 0.05 0.09 0.24 0.45 0.66 0.61

(0.58) (0.85) (2.88) (3.74) (4.18) (3.59)

BM

-FF3 0.16 0.10 0.01 0.21 0.40 0.56

(1.93) (1.16) (0.18) (2.36) (2.77) (2.91)

(continued)

7/29/2019 faj.v69.n1.1

11/18



predictive power o CCH and CH vary with arbi-trage costs? To address this question, I divided thesample each month by using median splits (lowand high) based on the idiosyncratic volatility(IVol, my proxy or arbitrage costs). Table 8 reportsthe alphas or the quintile portolios and the zero-cost longshort spreads (trading protability) othe CCH and CH trading strategies.

The results in Table 8 show that both the CCHand the CH eects are pervasive across the IVolgroups. However, the CCH eect is substantiallyhigher among high-IVol stocks. In particular, theabnormal return FF3 alpha (FF4 alpha) or thelongshort portolioormed as the dierence

between the high and low quintileso CCH is0.33% (0.25%) a month or the high-IVol stocks. Incomparison, the monthly alphas or the low-IVolstocks are 21% and 15%, respectively.7 The monthlyalphas o the CH spread portolios across low-IVoland high-IVol stocks are all signicant and veryclose to each other. For example, the FF3 alphas are

0.46% or low-IVol stocks and 0.45% or the subseto high-IVol stocks.

How can the signicant negative CCH eector January months reported in Appendix A beexplained? Can this negative eect be attributedentirely to mispricing? To answer these questions, Ireport in Table A2 the FF3 and FF4 CCH alphas or

January months by IVol group. For comparison, Ialso report the FF3 and FF4 CH alphas or Januarymonths by IVol group. As these results show, thenegative CCH eect is concentrated solely amongstocks with high arbitrage risk (high IVol). Indeed,the spread portolio alphas (FF3 and FF4) are signi-icantly dierent rom zero in the high-IVol group,whereas they are not statistically dierent rom zeroin the low-IVol group. Moreover, the dierences inspread portolio alphas between high and low IVolare statistically signicant. The CH alphas on thelongshort portolio are all positive, insignicantor low-IVol stocks, and signicant or high-IVolstocks. However, the dierences in spread portolio

1 2 3 4 5 5 1

-FF4 0.01 0.07 0.17 0.41 0.58 0.59

(0.14) (0.77) (2.02) (4.09) (3.89) (3.43)

MOM

-FF3 0.15 0.07 0.07 0.19 0.33 0.49

(1.84) (0.80) (0.31) (2.02) (2.48) (2.96)

-FF4 0.00 0.11 0.17 0.39 0.52 0.52

(0.03) (1.32) (2.12) (3.91) (3.85) (3.45)

AG

-FF3 0.15 0.11 0.02 0.23 0.38 0.53

(1.79) (1.26) (0.21) (2.64) (2.89) (3.30)

-FF4 0.01 0.08 0.21 0.43 0.53 0.52

(0.16) (0.89) (2.68) (4.12) (3.99) (3.41)

Illiq

-FF3 0.16 0.09 0.02 0.14 0.43 0.59

(1.74) (0.91) (0.23) (1.37) (2.87) (3.37)

-FF4 0.02 0.10 0.20 0.37 0.65 0.63

(0.25) (0.98) (2.45) (3.13) (3.92) (3.63)

IVol

-FF3 0.17 0.11 0.02 0.16 0.33 0.50

(1.87) (1.18) (0.19) (1.59) (2.59) (3.18)

-FF4 0.01 0.09 0.22 0.38 0.50 0.49

(0.07) (0.90) (2.46) (3.24) (3.90) (3.43)

Table 5. Alphas of Portfolios Sorted on Changes in Cash Holdings: Controlling forCharacteristics (Robustness Checks), July 1965December 2010 (continued)(t-tatitic in paenthee)

7/29/2019 faj.v69.n1.1

12/18



alphas between high and low IVol are not statisti-

cally signicant.Taken together, the results or the ull sampleand the non-January months suggest that only parto the CCH anomaly results rom high barriers toarbitrage as proxied by high idiosyncratic volatili-ties. Moreover, the CCH eect is more contaminated(though only partly) by mispricing than is the CHeect. The dierence in the FF4 alpha o the CCHspread portolio between high and low IVol is twicethe dierence in the FF4 alpha o the CH spreadportolio between high and low IVol. The resultsor January months indicate that the surprisingly

negative CCH eect is driven completely by mis-

pricing (see Appendix A).

ConclusionWith this study, my contribution to the literaturecan be characterized as threeold. First, I providedevidence that a change in a companys cash-holdingpolicy predicts subsequent stock returns. The abnor-mal returns associated with a change in cash hold-ings remain signicant in the presence o size, book-to-market, momentum, asset growth, illiquidity, andidiosyncratic volatility. Importantly, the Sharpe ratioo a CCH-based actor-mimicking portolio is higher

Table 6. Is the CCH Effect Distinct from the CH Effect? July 1965December 2010(t-tatitic in paenthee)

Sample INT CCH CH ME BM MOM AG Illiq IVol

A. FamaMacBeth regressions

All months 2.81 0.63* 0.77** 0.14*** 0.19*** 0.29* 0.54*** 0.23*** 0.20***

(4.57) (1.68) (2.32) (3.53) (2.95) (1.69) (6.25) (2.77) (5.23)Non-Januarymonths 1.77 1.08*** 0.59* 0.06 0.16** 0.56*** 0.48*** 0.15* 0.27***

(2.85) (2.94) (1.75) (1.52) (2.22) (3.17) (5.49) (1.92) (6.90)

Ranking on CCH (CH)

Control Alpha 1 2 3 4 5 5 1

B. Portolio analysis

All months CH -FF3 0.11 0.14 0.16 0.16 0.08 0.19

(1.21) (1.91) (2.47) (2.17) (1.10) (2.88)

-FF4 0.11 0.30 0.30 0.29 0.23 0.12

(1.23) (3.83) (4.36) (3.80) (2.80) (1.81)

CCH -FF3 0.11 0.07 0.04 0.20 0.33 0.44

(1.33) (0.88) (0.59) (2.53) (2.69) (2.90)

-FF4 0.04 0.08 0.20 0.37 0.47 0.43

(0.49) (1.07) (2.67) (4.34) (3.97) (3.14)

Non-Januarymonths CH -FF3 0.28 0.02 0.07 0.08 0.02 0.26

(3.69) (0.29) (1.23) (1.25) (0.27) (4.19)

-FF4 0.06 0.19 0.21 0.22 0.13 0.19

(0.75) (2.64) (3.33) (3.27) (1.67) (3.10)

CCH -FF3 0.21 0.15 0.04 0.07 0.17 0.38

(2.35) (2.02) (0.61) (1.08) (1.49) (2.37)

-FF4 0.05 0.01 0.12 0.25 0.31 0.36

(0.58) (0.12) (1.80) (3.41) (2.81) (2.48)



7/29/2019 faj.v69.n1.1

13/18



than that o the market actor or the SMB and HMLactors o Fama and French (1993) and is very closeto that o the momentum actor.

Second, I compared the CCH eect with theCH eect documented by Palazzo (2012). I oundthat (1) both eects are distinct and coexist (i.e.,

Table 7. Alphas of Portfolios Sorted on Changes in Cash Holdings and Grouped by Level ofCash Holdings and Market Capitalization, July 1965December 2010(t-tatitic in paenthee)

Alpha 1 2 3 4 5 5 1

A. Ranking on CCH by CH

All months Low CH -FF3 0.22 0.03 0.11 0.04 0.04 0.26

(2.46) (0.34) (1.14) (0.46) (0.47) (3.30)

-FF4 0.03 0.12 0.04 0.09 0.16 0.19

(0.29) (1.29) (0.43) (1.23) (1.76) (2.24)

High CH -FF3 0.07 0.26 0.32 0.34 0.16 0.09

(0.56) (2.53) (3.67) (3.79) (1.49) (1.00)

-FF4 0.33 0.40 0.43 0.47 0.35 0.02

(2.49) (3.70) (4.76) (4.70) (2.80) (0.23)

Non-Januarymonths Low CH -FF3 0.35 0.13 0.18 0.10 0.02 0.32

(3.87) (1.57) (1.86) (1.15) (0.37) (4.05)

-FF4 0.15 0.00 0.02 0.05 0.10 0.25

(1.63) (0.05) (0.24) (0.60) (1.00) (3.08)

High CH -FF3 0.14 0.12 0.22 0.26 0.03 0.17

(1.15) (1.31) (2.72) (3.12) (0.31) (1.93)

-FF4 0.14 0.26 0.34 0.40 0.22 0.08

(1.06) (2.60) (3.77) (4.54) (1.85) (0.93)

B. Ranking on CCH by ME

All months Low ME -FF3 0.05 0.17 0.10 0.38 0.40 0.35

(0.35) (1.39) (0.87) (3.22) (3.28) (3.28)

-FF4 0.31 0.34 0.27 0.58 0.60 0.29

(1.77) (2.47) (1.97) (4.53) (4.17) (2.78)

High ME -FF3 0.11 0.01 0.13 0.02 0.11 0.22

(1.43) (0.11) (1.77) (0.26) (1.35) (3.60)

-FF4 0.09 0.13 0.02 0.11 0.24 0.15

(1.23) (1.90) (0.22) (1.64) (2.90) (2.41)

Non-Januarymonths Low ME -FF3 0.33 0.14 0.15 0.07 0.13 0.46

(2.27) (1.28) (1.27) (0.63) (1.06) (4.49)

-FF4 0.10 0.01 0.04 0.25 0.30 0.40

(0.61) (0.09) (0.30) (1.99) (2.19) (3.78)

High ME -FF3 0.13 0.00 0.11 0.05 0.12 0.25

(1.70) (0.01) (1.40) (0.64) (1.45) (3.97)

-FF4 0.09 0.14 0.03 0.17 0.28 0.19

(1.27) (1.93) (0.40) (2.47) (3.38) (2.84)

7/29/2019 faj.v69.n1.1

14/18



Table 8. Alphas of Portfolios Sorted on Changes in Cash Holdings and Grouped by Level ofIdiosyncratic Volatility, July 1965December 2010(t-tatitic in paenthee)

IVol Alpha 1 2 3 4 5 5 1

A. Ranking on CCH

All

months Low IVol -FF3 0.00 0.06 0.07 0.11 0.21 0.21(0.06) (0.85) (0.87) (1.63) (3.05) (3.74)

-FF4 0.14 0.17 0.05 0.20 0.29 0.15

(2.26) (2.50) (0.62) (2.97) (4.10) (2.67)

High IVol -FF3 0.14 0.09 0.00 0.24 0.19 0.33

(0.87) (0.65) (0.02) (1.90) (1.25) (3.33)

-FF4 0.17 0.30 0.18 0.48 0.42 0.25

(0.99) (1.94) (1.21) (3.53) (2.40) (2.46)

Non-Januarymonths Low IVol -FF3 0.01 0.08 0.06 0.14 0.24 0.25

(0.16) (1.08) (0.71) (2.04) (3.32) (4.21)

-FF4 0.15 0.21 0.08 0.26 0.34 0.19

(2.47) (2.96) (1.59) (3.95) (4.70) (3.28)

High IVol -FF3 0.48 0.24 0.30 0.06 0.07 0.41

(3.34) (1.83) (2.47) (0.49) (0.47) (4.46)

-FF4 0.17 0.07 0.17 0.17 0.14 0.31

(1.09) (0.53) (1.26) (1.26) (0.85) (3.29)

B. Ranking on CH

All

months Low IVol

-FF3 0.15 0.03 0.03 0.18 0.31 0.46(1.80) (0.41) (0.39) (2.61) (4.28) (4.59)

-FF4 0.01 0.09 0.15 0.27 0.39 0.40

(0.07) (1.14) (1.93) (4.08) (5.14) (3.86)

High IVol -FF3 0.14 0.21 0.04 0.18 0.31 0.45

(1.07) (1.52) (0.29) (1.08) (1.59) (2.07)

-FF4 0.07 0.06 0.34 0.51 0.52 0.45

(0.48) (0.37) (1.96) (2.47) (2.72) (2.22)

Non-Januarymonths Low IVol -FF3 0.15 0.00 0.05 0.21 0.31 0.45

(1.73) (0.06) (0.60) (2.92) (4.53) (4.44)

-FF4 0.01 0.15 0.20 0.33 0.40 0.39

(0.11) (1.89) (2.41) (4.78) (5.32) (3.39)

High IVol -FF3 0.46 0.54 0.28 0.16 0.02 0.44

(3.26) (3.75) (2.05) (1.00) (0.13) (1.86)

-FF4 0.27 0.30 0.00 0.13 0.18 0.46

(1.87) (1.94) (0.01) (0.73) (1.01) (2.14)

7/29/2019 faj.v69.n1.1

15/18



the abnormal returns associated with CCH remainater controlling or CH, and vice versa); (2) a sig-nicant unconditional (and conditional) relationexists between CCH and uture stock returns,whereas the link between CH and uture stockreturns exists only when conditioning on size and

book-to-market; (3) the CCH eect is negative and

signicant or January months, whereas the CHeect or January months remains positive and sig-nicant; (4) both the CCH and the CH eects arestronger in the universe o smaller-cap companies(more likely to ace nancial constraints); and (5)the CCH anomaly exists predominantly amongcash-constrained companies (stocks with relativelylow cash-holding levels).

Third, I explored the implications o arbitragerisk or both the CCH and the CH eects and oundthat unlike accrual and asset growth anomalies (seeLi and Sullivan 2011), both eects are pervasive

across arbitrage risk groups (proxied by idiosyncraticrisk) but that the CCH anomaly appears to be moreinfuenced than the CH eect by mispricing. Thesurprisingly negative CCH eect ound or Januarymonths is, however, driven entirely by mispricing.

I am grateul or helpul comments rom Kose John,Edward Altman, and Anthony Saunders.

This article qualifes or 1 CE credit.

Appendix A. Regressions andAlphasTable A1 reports the results o standard and risk-adjusted FamaMacBeth regressions or January

months only. Contrary to the results or the ullsample and the non-January months, the sign o theCCHreturn relationship is negative, which is theopposite direction rom the rational prediction. Inall specications (both standard and risk adjusted),the coecient on CCH is signicant. Consistent withthe results or the ull sample and the non-Januarymonths, the CH eect remains positive. The coe-cient on CH is signicant in all specications exceptwhen the risk adjustment is based on the Carhartour-actor model. I also conrmed the pronouncednegative momentum eect or January months

reported by Chou, Ho, and Ko (2012). This eectactually weakens the signicance o the positivemomentum eect in the ull sample (see Table 2).

I also investigated whether the negative CCHeect observed or January months in standardand risk-adjusted FamaMacBeth regressions(Table A1) is conrmed in the portolio analysis.Table A2 reports the alphas o the CCH porto-lios or January months only. Consistent with

Table A1. FamaMacBeth Cross-Sectional Regressions for January Returns, July 1965December 2010(t-tatitic in paenthee)



1 15.23 3.22** 1.06*** 0.48 2.67*** 1.27*** 1.07*** 0.59***

(5.71) (2.13) (5.13) (0.95) (4.36) (4.27) (5.71) (4.06)

2 14.95 2.48** 1.06*** 0.59 2.28*** 0.96*** 1.21** 0.55***

(6.03) (2.01) (5.46) (1.34) (4.33) (3.70) (2.00) (3.64)


3 9.32 3.22* 0.70*** 0.27 2.56*** 1.52*** 1.14** 0.03

(4.33) (1.79) (4.32) (0.89) (4.17) (3.91) (2.20) (0.12)

4 9.08 1.74** 0.70*** 0.20 2.52*** 1.09*** 1.25** 0.01(4.11) (2.17) (4.14) (0.75) (3.96) (3.58) (2.01) (0.03)


5 9.42 3.85** 0.72*** 0.31 1.92*** 1.17*** 1.03** 0.01

(3.71) (2.43) (3.37) (1.15) (3.37) (3.03) (2.37) (0.07)

6 8.99 1.90 0.70*** 0.28 1.90*** 0.88*** 1.21** 0.05

(3.72) (1.52) (3.42) (1.09) (3.34) (2.90) (2.08) (0.26)



7/29/2019 faj.v69.n1.1

16/18



Table A2. Alphas of Portfolios Sorted on Changes in Cash Holdings for January Returns, July1965December 2010(t-tatitic in paenthee)

Sample IVol Alpha 1 2 3 4 5 5 1

A. Ranking on CCH

All January months -FF3 2.04 1.28 0.66 1.04 1.38 0.66

(4.90) (4.96) (1.77) (3.83) (5.20) (2.12)

-FF4 1.96 1.22 0.61 1.02 1.34 0.62

(4.77) (4.96) (1.83) (3.89) (5.02) (2.02)

Subsamples Low IVol -FF3 0.00 0.12 0.25 0.20 0.05 0.05

(0.00) (0.37) (0.78) (0.79) (0.21) (0.16)

-FF4 0.06 0.16 0.29 0.20 0.04 0.02

(0.25) (0.53) (0.96) (0.76) (0.20) (0.09)

High IVol -FF3 4.11 3.88 3.46 3.83 3.10 1.01

(5.10) (4.74) (4.94) (4.77) (5.18) (2.00)

-FF4 4.04 3.82 3.36 3.77 3.02 1.01(4.90) (4.54) (4.77) (4.70) (4.85) (1.93)

DierencesHigh IVol

Low IVol -FF3 0.96

(1.93)

High IVol Low IVol -FF4 1.03

(2.10)

B. Ranking on CH

All January months -FF3 0.90 0.96 1.00 1.60 3.08 2.17

(2.79) (3.09) (4.26) (3.02) (5.33) (3.04)-FF4 0.83 0.92 0.97 1.52 3.01 2.18

(3.12) (3.22) (4.50) (2.71) (5.02) (3.06)

Subsamples Low IVol -FF3 0.23 0.40 0.12 0.11 0.39 0.62

(0.64) (1.00) (0.44) (0.84) (1.33) (1.47)

-FF4 0.28 0.43 0.15 0.12 0.37 0.65

(0.85) (1.08) (0.57) (0.92) (1.28) (1.56)

High IVol -FF3 3.58 3.74 3.58 3.84 4.53 0.95

(5.06) (4.82) (4.38) (4.34) (6.04) (1.92)

-FF4 3.50 3.65 3.49 3.70 4.50 1.00

(4.93) (4.61) (4.14) (3.94) (5.90) (2.08)

DierencesHigh IVol Low IVol -FF3 0.33

(0.60)

High IVol Low IVol -FF4 0.35

(0.64)

7/29/2019 faj.v69.n1.1

17/18



the results o FamaMacBeth cross-sectionalregressions, the dierence portolio (i.e., the51 spread portolio) shows negative and sig-nicant FF3 and FF4 alphas (0.66% and 0.62%,

respectively) or January months. Thus, the port-olio analysis conrms that there is a CCH eector January months, which is at odds with therational explanation.

Notes1. See, or example, Opler, Pinkowitz, Stulz, and Williamson

(1999); Dittmar and Mahrt-Smith (2007); Bates, Kahle, and

Stulz (2009).

2. Consistent with Palazzo (2012), I ound no uncondi-

tional relation between CH and uture stock returns but

observed a strong CH eect when conditioning on size and

book-to-market.

3. Investors may be too slow to incorporate inormation regard-

ing changes in a companys cash holdings into its stock

prices (or a similar argument on company investment, see

Titman, Wei, and Xie 2004; Cooper, Gulen, and Schill 2008).

Recently, many papers have stressed the importance o limits

to arbitrage (i.e., mispricing) in explaining investment, asset

growth, and accrual anomalies (see, e.g., Li and Zhang 2010;

Li and Sullivan 2011).

4. As stated in Li and Sullivan (2011, p. 51), When conrontedwith holding a position with high IVol, investors are less will-ing to engage in arbitrage because such a position is costly tohedge. Note that other studies have also used idiosyncraticstock return volatility to measure arbitrage risk (see, e.g.,Ponti 1996; Mashruwala, Rajgopal, and Shevlin 2006).

5. http://mba.tuck.dartmouth.edu/pages/aculty/ken.rench/.

6. Note that IVolm is a monthly idiosyncratic volatility and astock return predictor estimated as in Ang, Hodrick, Xing,and Zhang (2006), whereas IVol is a 12-month (JulyJune)idiosyncratic volatility that is used to split the sample intolow- and high-IVol groups (see, e.g., Li and Sullivan 2011).

7. For non-January months, these alphas are 41% and 31% (orhigh-IVol stocks) and 25% and 19% (or low-IVol stocks),

respectively.

ReferencesAmihud, Y. 2002. Illiquidity and Stock Returns: Cross-Sectionand Time-Series Eects.Journal o Financial Markets, vol. 5, no.1 (January):3156.

Ang, A., R.J. Hodrick, Y. Xing, and X. Zhang. 2006. TheCross-Section o Volatility and Expected Returns. Journal oFinance, vol. 61, no. 1 (February):259299.

Bar-Ilan, A., and W.C. Strange. 1996. Investment Lags.American Economic Review, vol. 86, no. 3 (June):610622.

Bates, T.W., K.M. Kahle, and R.M. Stulz. 2009. Why Do U.S.Firms Hold So Much More Cash Than They Used To? Journal oFinance, vol. 64, no. 5 (October):19852021.

Brennan, M.J., T. Chordia, and A. Subrahmanyam. 1998.Alternative Factor Specications, Security Characteristics, andthe Cross-Section o Expected Stock Returns.Journal o FinancialEconomics, vol. 49, no. 3 (September):345373.

Carhart, M. 1997. On Persistence in Mutual Fund Perormance.Journal o Finance, vol. 52, no. 1 (March):5782.

Chou, P.H., P.H. Ho, and K.C. Ko. 2012. Do Industries Matterin Explaining Stock Returns and Asset-Pricing Anomalies?Journal o Banking & Finance, vol. 36, no. 2 (February):355370.

Cooper, M.J., H. Gulen, and M.J. Schill. 2008. Asset Growth and

the Cross-Section o Stock Returns. Journal o Finance, vol. 63,no. 4 (August):16091651.

Dimson, E. 1979. Risk Measurement When Shares Are Subjectto Inrequent Trading.Journal o Financial Economics, vol. 7, no.2 (June):197226.

Dittmar, A., and J. Mahrt-Smith. 2007. Corporate Governanceand the Value o Cash Holdings. Journal o Financial Economics,vol. 83, no. 3 (March):599634.

Fama, E.F., and K. French. 1993. Common Risk Factors in theReturns on Stocks and Bonds. Journal o Financial Economics,vol. 33, no. 1 (February):356.

Fama, E.F., and J. MacBeth. 1973. Risk, Return, and Equilibrium:Empirical Tests. Journal o Political Economy, vol. 81, no. 3(MayJune):607636.

Gibbons, M., S.A. Ross, and J. Shanken. 1989. A Test or theEciency o a Given Portolio. Econometrica, vol. 57, no. 5(September):11211152.

Keim, D.B. 1983. Size-Related Anomalies and Stock ReturnSeasonality: Further Empirical Evidence. Journal o FinancialEconomics, vol. 12, no. 1 (June):1332.

Li, D., and L. Zhang. 2010. Does Q-Theory with InvestmentFrictions Explain Anomalies in the Cross Section o Returns?Journal o Financial Economics, vol. 98, no. 2 (November):297314.

Li, X., and R.N. Sullivan. 2011. The Limits to ArbitrageRevisited: The Accrual and Asset Growth Anomalies. FinancialAnalysts Journal, vol. 67, no. 4 (July/August):5066.

MacRae, K.M. 1989. Critical Issues in Electric Power Planning inthe 1990s. Calgary, Canada: Canadian Energy Research Institute.

Mashruwala, C., S. Rajgopal, and T. Shevlin. 2006. Why Isthe Accrual Anomaly Not Arbitraged Away? The Role oIdiosyncratic Risk and Transaction Costs.Journal o Accountingand Economics, vol. 42, no. 12 (October):333.

Newey, W.K., and K.D. West. 1987. A Simple, PositiveSemi-Denite, Heteroskedasticity and AutocorrelationConsistent Covariance Matrix. Econometrica, vol. 55, no. 3(May):703708.

Opler, T., L. Pinkowitz, R. Stulz, and R. Williamson. 1999. TheDeterminants and Implications o Corporate Cash Holdings.Journal o Financial Economics, vol. 52, no. 1 (April):346.

Palazzo, B. 2012. Cash Holdings, Risk, and Expected Returns.Journal o Financial Economics, vol. 104, no. 1 (April):162185.

Pindyck, Robert S. 1991. Irreversibility, Uncertainty, andInvestment. Journal o Economic Literature, vol. 29, no. 3(September):11101148.

7/29/2019 faj.v69.n1.1

18/18


Ponti, J. 1996. Costly Arbitrage: Evidence rom Closed-EndFunds. Quarterly Journal o Economics, vol. 111, no. 4(November):11351151.

. 2006. Costly Arbitrage and the Myth o IdiosyncraticRisk. Journal o Accounting and Economics, vol. 42, no. 12(October):3552.

Shleier, A., and R.W. Vishny. 1997. The Limits o Arbitrage.

Journal o Finance, vol. 52, no. 1 (March):3555.Simutin, M. 2010. Excess Cash and Stock Returns. FinancialManagement, vol. 39, no. 3 (Autumn):11971222.

Titman, S., K.C.J. Wei, and F. Xie. 2004. Capital Investments andStock Returns.Journal o Financial and Quantitative Analysis, vol.39, no. 4 (December):677700.

Wheaton, W. 1987. The Cyclic Behavior o the National OceMarket.Journal o the American Real Estate and Urban EconomicsAssociation, vol. 15, no. 4 (Winter):281299.

Xing, Y. 2008. Interpreting the Value Eect through the

Q-Theory: An Empirical Investigation. Review o FinancialStudies, vol. 21, no. 4 (July):17671795.

Documents

faj.v69.n1.1