Journal of Banking & Finance 37 (2013) 3318–3333

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier .com/locate / jbf

The wisdom of crowds: Mutual fund investors’ aggregate asset allocationdecisions

0378-4266/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.jbankfin.2013.05.004

⇑ Corresponding author. Tel.: +1 519 888 4567; fax: +1 519 888 7562.E-mail addresses: jchalmer@uoregon.edu (J. Chalmers), akaul@business.ualber-

ta.ca (A. Kaul), brphilli@uwaterloo.ca (B. Phillips).1 From the Investment Company Institute (ICI) Fact Book, 2010.2 See Sirri and Tufano (1998) and Lynch and Musto (2003) for evidence of return

chasing. Jain and Wu (2000) and Cooper et al. (2005) find that investors direct flowtowards funds that advertise more and that undergo name changes to reflect currentmarket trends. Bailey et al. (2011) provide evidence that behavioral factors play a partin suboptimal mutual fund selection.

3 Individuals, rather than institutions, are the main holders of mutual funds. Forexample, using table 56 of the 2010 ICI Mutual Fund Fact Book, we calculate that in2008, 92% of equity fund assets and 67% of money market fund assets were held inindividual accounts.

John Chalmers a, Aditya Kaul b, Blake Phillips c,⇑a Lundquist College of Business, University of Oregon, Eugene, OR 97403, USAb School of Business, University of Alberta, Edmonton, AB, Canada T6G 2R6c School of Accounting and Finance, University of Waterloo, Waterloo, ON, Canada N2L 3G1

a r t i c l e i n f o

Article history:Received 19 July 2012Accepted 4 May 2013Available online 16 May 2013

JEL classification:G11G14G23G32

Keywords:Mutual fundMutual fund flowAsset allocation

a b s t r a c t

We find that the aggregate asset allocation decisions of US mutual fund investors depend on economicconditions. Both anticipated economic downturns and periods of turmoil lead investors to direct flowaway from risky equity funds and towards lower-risk money market funds. These patterns are markedlystronger for investors in low cost and low turnover funds relative to investors in high cost and high turn-over funds, consistent with sophisticated investors being more sensitive to changing conditions. Bench-marked against a buy-and-hold strategy, these asset allocation strategies reduce risk without degradingthe risk-return trade-off. Our evidence suggests that individual investors, often dismissed as noise trad-ers, collectively react to economic signals in a sensible manner when determining asset allocations.

� 2013 Elsevier B.V. All rights reserved.

1. Introduction

Commanding worldwide assets of 23 trillion USD at year-end2009, mutual fund investors collectively are major players in cap-ital markets.1 Academic research at the individual fund level ques-tions the rationality of these investors. For example, fund investorschase returns and react to non-informative name changes and adver-tising campaigns.2 In this paper, we study the aggregate asset alloca-tion decisions of US mutual fund investors. Our goal is to understandthe behavior of the amalgam of mostly small, retail investors.3 Muchas diversification minimizes the effects of idiosyncratic factors on

portfolio returns, aggregate allocation decisions may differ substan-tially from fund-level flows.

Our contribution is to provide evidence on the relation betweeneconomic conditions and the aggregate flows to mutual fund clas-ses with differing risks. More specifically, using monthly data be-tween February 1991 and March 2008, we address the followingthree questions. First, when making asset allocation decisions, domutual fund investors react to changing economic conditions?We compute aggregate monthly allocations to four major assetclasses: domestic equities, money market, bonds, and foreign equi-ties. We then relate these allocations to proxies for economic con-ditions: the Chicago Fed National Activity Index (CFNAI), the termspread (TERM), the default spread (DEF), the change in the short-term interest rate (DTB), the Treasury-Eurodollar spread (TED),and volatility in the stock and bond markets (SPV and TBV). Wefind that fund investors alter the riskiness of their portfolios in re-sponse to shifting economic conditions, increasing risk as the econ-omy is expected to improve and reducing risk in anticipation ofeconomic downturns. Thus, when the economy is expected to per-form favorably (i.e. TERM is high, DEF is low, DTB is low, or TED islow), investors direct flow away from money market funds and to-wards equity funds. This flow reaction is rapid and appears perma-nent, as we find no evidence either of a sluggish response or ofover-reaction to the explanatory variables. Consistent with rational

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3319

forecasting or, perhaps more plausibly, with fund investorsresponding to media commentary which incorporates the informa-tion in these variables, forward-looking financial market variablesdominate the predictive power of contemporaneous signals fromthe real economy.

Second, we consider whether mutual fund investors flee riskyinvestments during periods of crisis. The term ‘‘flight-to-safety’’is ubiquitous during turbulent times, the belief being that investorsgravitate to safer investments during such periods. Beber et al.(2009) study flight-to-quality and flight-to-liquidity in the Euro-area government bond market; otherwise, there is little evidenceof the pervasiveness or implications of flight-to-safety reactionsby investors. Consistent with pervasive safe-haven flows, thereare significant incremental shifts from higher-risk equity funds tolower-risk money market funds during crisis periods. The for-ward-looking market variables retain their explanatory powerwhen crisis effects are included, suggesting that the relation be-tween flow and economic conditions is not due to such episodes.

Third, we study whether the strength of the relation betweenasset allocation and economic conditions varies across mutual fundinvestor profiles. We separate high and low cost equity indexfunds, high and low cost equity funds and high and low portfolioturnover funds to distinguish unsophisticated from sophisticatedinvestors. We find that the allocations to funds which we wouldexpect to be held by more sophisticated investors are more sensi-tive to the economic forecasting variables and drive the relationbetween economy-wide allocations and changing economic condi-tions. In contrast, the allocations to funds likely held by more unso-phisticated investors show minimal sensitivity to these variables.This is broadly consistent with evidence in Bailey et al. (2011),who find that more sophisticated investors choose mutual fundswith lower fees and realize better performance.

These results have significant implications. Understanding andpredicting investor allocations is critical to the efficient manage-ment of mutual fund portfolios. Managers caught off guard byshifts in investor preferences can face asset-eroding trading costs.Perhaps more importantly, our results showing that the aggregateasset allocation decisions of fund investors are influenced by busi-ness cycle factors that appear to have rational underpinnings,stand in contrast to the weight of the fund-level evidence. Finally,the relation between expected returns and forecasting variablessuch as TERM and DEF is academically well established (e.g. Famaand French, 1989; Chen et al., 1986), but less is known about themechanism by which this relation comes to hold. Our analysis sug-gests that, as fund investors collectively rebalance their portfoliosin response to these variables, the resulting price effects contributeto the documented relation between expected returns and theforecasting variables.

While not the focus of our analysis, we briefly examine the per-formance implications of such time-varying asset allocation deci-sions. An extensive literature shows that expected returns andvolatility are high when business conditions are poor and low ingood economic times (e.g. Breeden, 1979; Fama and French,1989; Schwert, 1989). Thus, the business cycle-based allocationstrategies we document are likely to lower equity exposure whenthe premium for holding equities and return volatility are bothhigh, and increase exposure when expected returns and volatilityare low. We confirm that this is the case with aggregate mutualfund allocations. That is, inflows to equity funds precede periodsof low volatility and expected returns, and outflows foreshadowhigh volatility and expected returns. As a result, business-cyclebased allocation strategies lower both returns and volatility, andthe net benefit hinges on the risk-return tradeoff investors are ableto achieve.

To gain a sense of this potential tradeoff, we implement eightdynamic asset allocation strategies with trading triggers that are

based on the economic forecasting variables and realized equityflow. We find that returns are lower for each strategy relative toa buy-and-hold equity benchmark. Accounting for risk, the dy-namic portfolios have, on average, 16% higher Sharpe ratios and48% lower systematic risk. Following Ferson and Mo (2012), wealso estimate aggregate market and volatility timing ability andfind positive alphas that are in several cases marginally significant.This analysis suggests that investors are no worse off in a risk-ad-justed performance sense. In fact, for a risk-averse investor, such arisk-reducing strategy is potentially utility enhancing.

Our results contribute to research that examines investor fund-picking ability which yields mixed conclusions. Gruber (1996) andZheng (1999) provide the first evidence of the ‘smart money’ effect,reporting that funds experiencing money inflows have higher sub-sequent short-term performance than funds experiencing out-flows. However, Sapp and Tiwari (2004) argue that this effect isexplained by momentum in stock returns, and that investors haveno fund picking ability. Using dollar weighted returns, Friesen andSapp (2007) examine the market timing ability of investors in indi-vidual funds and find that timing decisions reduce performance by1.56% annually. This underperformance is greater for funds withhigh loads and large risk-adjusted returns. Keswani and Stolin(2008) separately examine UK fund-level inflows and outflowsand find that the smart money effect exists for both institutionaland individual investors. Huang et al. (2012) show that moresophisticated investors (identified using loads, or by separatinginstitutional and retail funds and funds with star managers versusno star managers) learn from past fund performance and becomeless performance-sensitive.

Our results also relate to a small literature examining aggregateflow. Edwards and Zhang (1998) and Santini and Aber (1998) re-port that US equity fund flows are positively related to stock mar-ket returns and contemporaneous personal disposable income andnegatively related to the lagged long-term interest rate. Cohen(1999) examines quarterly Federal Reserve flow-of-funds dataand finds associations between individual and institutional flowsand TERM, the dividend yield and TB. Goetzmann et al. (1999) findthat US equity flow is negatively correlated with flow to moneymarket and precious metals funds. Ben-Rephael et al. (2012) aggre-gate within-fund family bond to equity transfers in the US, and finda negative association between transfers and future market re-turns. Also using flow-of-funds data, Edelen et al. (2010) find thatthe ratio of individual investor equity allocation to market-wideequity allocation is positively associated with contemporaneousmarket returns and negatively predicts future market returns.Ben-Rephael et al. and Edelen et al. interpret their variable as asentiment indicator. We contribute to this literature by relatingaggregate mutual fund allocations to business conditions andexamining the performance implications of such decisions. Our re-sults indicate that aggregate allocations have rational drivers.

Finally, Dichev (2007) shows that aggregate dollar-weighted re-turns, which more heavily weight performance when greater cap-ital is invested, are lower than buy-and-hold returns for the US and19 international equity markets suggesting that investors are poormarket timers. The dollar-weighted return method does not incor-porate other asset classes or risk considerations. We extend Di-chev’s analysis by modeling aggregate mutual fund investorcapital allocations across asset classes and the business cycle andby studying investor sophistication. We show that investors movecapital across asset classes and that portfolio risk consequentlyvaries over time. Two implications of our results are that single as-set class based dollar-weighted returns provide an incomplete pic-ture of portfolio performance, and it is more meaningful tocompare risk-adjusted returns. Consistent with Dichev’s resultswe find that, on a raw return basis, asset re-allocations by fundinvestors are performance reducing. However, on a risk-adjusted

3320 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

basis, asset allocations can be performance neutral or enhancing,suggesting that the ‘smart money’ effect exists at the aggregate le-vel for groups of investors.

The rest of the paper is organized as follows. Section 2 outlinesour methods, data, and hypotheses. Section 3 describes the results.Section 4 discusses the implications of our results. Section 5 con-cludes. Additional details and analysis are presented in AppendixesA and B.

2. Data and methods

2.1. Sample

We obtain mutual fund data from the Center for Research inSecurity Prices (CRSP) Mutual Fund Database, which providesmonthly net asset value (NAV) and returns by fund share class aswell as quarterly or annual disclosures of management fees, port-folio turnover and fund objectives. Our sample period extends fromFebruary 1991, at which point CRSP reports monthly NAV for allfunds in the database, through March 2008.

We classify funds into one of five asset categories: DomesticEquity, Domestic Money Market, Domestic Bond and Foreign Equi-ty, as well as Other, which includes all remaining funds (seeAppendix A for details). We focus on the first four categories, whichmeet three conditions. First, these are the asset classes popularwith mutual fund investors. Second, they expose investors to dif-ferent risks. For example, the performance of equity funds will behighly sensitive to economic conditions, as the cash flows fromthe stocks they hold rise or fall with the economy. In contrast,the performance of money market funds, which hold short-termfixed income securities, is generally smoother, though tied directlyto the level of short-term interest rates. Finally, as discussed below,each of these asset categories is large, accounting for approxi-mately 10% or more of aggregate mutual fund assets over our sam-ple period.

Fig. 1, Panel (a) plots total assets for the five asset classes overthe sample period. Each category experiences rapid growth. Equityfunds are the largest category at the end of the period, with moneymarket funds a close second: both have over $3 trillion in assets.The other categories are smaller, with assets of approximately$1.5 trillion each. Panel (b) shows the proportion of industry netassets in each asset class. The four asset classes account for 82–90% of total mutual fund assets, though the asset class weightsvary significantly over time. For example, equity funds accountfor a low of 17% of industry net assets in early 1991 and a highof 40% in late 2000. Panel (c) shows the number of unique funds.4

At the end of the period, the sample includes approximately 4000equity funds, 3000 bond funds and 1000 money market funds. Thenumber of equity funds increases most rapidly over the sampleperiod.

2.2. Asset allocation measures

The objective of our analysis is to capture the extent to whichfund investors alter allocations across asset classes in response toeconomic signals. To that end, we calculate changes to aggregateasset allocations using a measure motivated by Frazzini and La-mont (2008). For each fund, we calculate monthly net flow as thechange in net assets resulting from purchases and redemptions,after accounting for returns:

NFi;t ¼ Ai;t � Ai;t�1 � Ai;t�1Ri;t ð1Þ

4 Funds with multiple share classes are counted as one unique fund.

where t denotes the month, and NF, A and R are net flow, net assets,and the monthly return for fund i. Next, we aggregate net flowacross the five broad asset classes. Then, as in Frazzini and Lamont(2008), benchmark flow is net asset-weighted flow (AWF), definedas:

AWFj;t ¼Aj;t�1P5

k¼1Ak;t�1

X5

k¼1

NFk;t ð2Þ

Here, subscript j denotes asset category and t denotes month. AWFj,t,Aj,t�1 and NFj,t are month t asset-weighted flow, month t � 1 assetsand month t net flow for asset class j. Thus, asset-weighted flow forasset class j in month t is the new flow that would go to class j ifallocated in proportion to the relative net assets of class j in montht � 1. The excess flow for category j, EFj,t, is constructed as actualflow for category j less asset-weighted flow from Eq. (2):

EFj;t ¼NFj;t � AWFj;t

At�1ð3Þ

To illustrate, suppose equity funds account for 40% of total assets atthe end of month t � 1. We then expect 40% of total flow in month tto be allocated to equity funds. Suppose total flow to all mutualfunds in month t is $10b, of which $5b goes to equity funds. In thisexample, raw excess flow to equity funds will be $1b (=$5b–10b � 0.40).

The flow measure in (3) scales raw excess flow by total net as-sets in the mutual fund industry (At�1). Our measure focuses onhow mutual fund wealth is allocated across classes; therefore itis logical to express this allocation as a fraction of overall mutualfund wealth. This standardization also provides a control for floweffects resulting from price appreciation and market growth. Weprefer the excess flow measure to other measures for several rea-sons. First, excess flow captures allocation adjustments betweenasset classes while imposing scale invariance. It can be thoughtof as measuring changes in the size of the wedges in a pie chartthrough time. Growth in overall fund assets that maintains the pre-vious month’s asset weights thus implies zero excess flow. Addi-tionally, to an extent, the excess flow measure captures thesimultaneity of the asset allocation decision. Directing assets toone asset class necessitates forgoing investment in other classes.This ebb and flow between asset classes is captured by excess flow,or even more precisely, by the difference in excess flow betweenthe asset classes. Importantly, however, we reach similar conclu-sions when we use other flow measures, e.g. percent flow (as dis-cussed below) or raw flow standardized by net assets to theindividual mutual fund asset classes.

One potential concern is that, by definition, excess flow for thefive asset categories must sum to zero. Equity and money marketfunds are the largest asset categories and the two that we will of-ten focus on. If the other three categories have static weights, theweights and thus excess flows for the equity and money marketclasses will show a mechanically-induced negative correlation.This, in turn, could lead to the conclusion that allocations to theequity and money market asset classes have the opposite relationto the predictor variables.

For two reasons, it is unlikely that this adding-up issue affectsour conclusions. First, as shown in Panel (b) of Fig. 1, there is sig-nificant time-series variation in all five asset class weights. Conse-quently, the excess flows for any two asset classes will notmechanically sum to zero. In particular, the sum of the equityand money market weights is far from constant, varying in thewide range of 50–70% over the sample period. For instance, bothweights increase in 1995 and 1996; the money market weight isflat while the equity weight increases between 1998 and 2001;and the two weights move in opposite directions between 2003and 2008. Second, a mechanical correlation between equity and

The evolution of key asset class variables over the sample period.

(a)

(b)

Tot

al N

et A

sset

s (T

rilli

on U

SD)

Ass

et C

lass

Wei

ght

Num

ber

of F

unds

(c)

Fig. 1. The evolution of key asset class variables over the sample period. This figure shows the time series variation in total net assets, asset class weights and the number offunds for five US mutual fund asset categories between February 1991 and March 2008. The five categories are: Domestic equity (Equity), Domestic Bond (Bond), DomesticMoney Market (Money Market), Foreign Equity and Other (which includes, for instance, foreign bond funds, mortgage backed security funds and international money marketfunds). Panel (a) shows total net assets for each asset category, Panel (b) shows the proportion of total net assets accounted for by each asset category, and Panel (c) shows thenumber of funds in each category. In Panel (c), multiple share classes of the same fund are consolidated as one fund.

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3321

money market excess flows, even if it exists, does not mechanicallyinduce a relation between the excess flow for either category andthe predictor variables, e.g. a positive coefficient on TERM in theequity flow regressions.

A commonly-used flow measure at the fund level is percentflow, defined as net flow from Eq. (1) standardized by lagged netassets (e.g. Sirri and Tufano, 1998). By calculating aggregate per-cent flow for each asset class, we can capture the level of flow tothe classes. However, this measure might not effectively captureallocations between classes. For example, general growth in thefund industry would appear as high percent flow to all asset classes

but provide little indication of investor preferences across assetclasses. Given our interest in allocation decisions, we focus onthe excess flow measure. However, as a robustness check, Appen-dix B shows that analysis using percent flow yields similarconclusions.

2.3. Explanatory variables

In order to understand how economic factors motivate mutualfund investor allocations, we relate excess flow for the four majorasset classes to proxies for economic conditions drawn from the

Table 1Predicted relation between investor flow and determinant variables.

Equities Bonds Money Market Foreign Equities

CFNAI + � � +TERM + � � +DEF � + + �DTB � + + �CONFID + � � +TED � ? + �SPV � + + ?TBV ? ? ? ?

This table summarizes the predicted relation between mutual fund investor flow,partitioned by asset category, and the determinant variables considered in thepaper. The symbols +, � and ? signify positive, negative and unclear predictedrelations, respectively. The principal flow measure is excess flow, calculated asaggregate net flow for each asset category in excess of the net flow that would haveresulted on an asset-weighted basis, standardized by the previous month’s mutualfund industry total net assets (Eqs. (1)–(3) in the text). CFNAI is the Chicago FedNational Activity Index. TERM, the term spread, is the difference in yields on the10 year U.S. Government Bond and the 3 month T-Bill. DEF, the default spread, is thedifference in yields on medium term corporate bonds and 3–5 year U.S. Govern-ment Bonds. DTB is the change in the yield on the 3 month T-Bill. CONFID is theUniversity of Michigan consumer confidence index value, derived from consumersurvey responses related to economic conditions perceptions. TED, the TreasuryEurodollar spread, is the difference between 3 month LIBOR and the 3 month T-BILLrate. RET is the equally-weighted monthly return to the US equity fund asset cat-egory. SPV, S&P 500 volatility, is the monthly sum of squared daily returns for theS&P 500 index and TBV, T-Bill volatility, is the monthly sum of squared daily yieldchanges for the 3 month T-Bill.

3322 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

real economy and financial markets. A large number of contempo-raneous real economy variables are available. Rather than makejudgments about which variables to include, we use the compre-hensive Chicago Fed National Activity Index (CFNAI). CFNAI is thefirst principal component of 85 monthly series that come fromthe broad categories of production and income, employment andhours, consumption and housing, and sales, orders and inventories.It includes most of the series we would think of as being importantindicators of current economic conditions (e.g. industrial produc-tion and unemployment).5

As financial market proxies we include the term spread (TERM),the default spread (DEF), the change in the short-term rate (DTB),the Treasury-Eurodollar spread (TED), and volatility in the stockand bond markets (SPV and TBV).6 Prior research suggests that ananticipated improvement in economic conditions is reflected in anincrease in TERM and/or a decline in DEF, DTB and TED. The volatil-ity measures capture the effects of stock and bond marketuncertainty.

The forecast power of variables such as TERM and TB for returnsis the subject of some debate. Campbell and Thompson (2010) sug-gest that the forecast power of these variables, while small, is eco-nomically meaningful. Henkel et al. (2011) find that the predictiveability of the term structure variables is largely confined to reces-sions. Ang and Bekaert (2007) suggest that the short-term interestrate predicts returns at shorter frequencies. In contrast, we inves-tigate the ability of such return forecasting variables to forecastmutual fund asset allocations. If variables such as TERM, DEF andTB predict reallocations between risky and safer assets, we havea potential channel through which these variables come to predictequilibrium returns.

To the list of variables, we add the University of Michigan Con-sumer Confidence Index (CONFID) as a measure of consumer sen-timent and the equally-weighted return to each asset class (RET) tocontrol for flow effects that might result from return chasing. Final-ly, we include CRISIS, a variable that is equal to one (and otherwisezero) during eight major shocks observed over the 1991–2008sample period:

(1) the subprime mortgage credit crisis (August 2007 to the endof the sample)

(2) the 9/11 terrorist attacks (September–December 2001)(3) the ‘‘Crash of 2000’’ when NASDAQ dropped 45.9% between

September and December 2000(4) Y2K concerns at the end of 1999 (October–December 1999)(5) the failure of the Long Term Capital Management Hedge

Fund (August–October 1998)(6) the Asian currency crisis (July–September 1997)(7) the Mexican currency crisis (December 1994–February

1995)(8) the Pound exiting the European Exchange Rate Mechanism

(September–November 1992)

5 A commonly-used real economy variable is industrial production growth (see forexample, Chen et al., 1986). We reach the same conclusions if industrial productiongrowth is used in place of CFNAI.

6 TERM is the difference in yields on the 10-year Treasury bond and the 3-monthTreasury bill. DEF is the difference between the yields on portfolios of medium-termcorporate bonds and medium maturity (3- to 5-year) government bonds. TED is thedifference between 3-month LIBOR and the 3-month T-Bill rate. DTB is the change inthe 3-month T-Bill yield. SPV is calculated as the sum of squared daily S&P 500returns in each month. An alternate volatility estimator that factors in twice the sumof the product of adjacent day returns in each month, following French et al. (1987),yields similar conclusions. TBV is calculated as the sum of squared daily changes inthe 3-month T-Bill yield in each month. Data on S&P 500 returns, consumerconfidence, LIBOR and U.S. bond yields come from DataStream. U.S. T-Bill data comefrom the St. Louis FED database.

The inclusion of CRISIS allows us to study whether turbulentepisodes are characterized by safe-haven flows. While the crisesare defined ex post, beyond anecdotal descriptions in the financialpress, we are unaware of empirical analysis documenting the exis-tence and magnitude of flight-to-safety flows during these epi-sodes. A contribution of this paper is to examine the extent towhich crisis periods are, in fact, associated with flight-to-safetyflows among mutual fund investors. Further, CRISIS serves as acontrol for turbulent episodes, thereby allowing us to assess thegeneral relation between flow and economic conditions.7

2.4. Predictions

Fama and French (1989) show that TERM and DEF track eco-nomic conditions. Specifically, TERM is wide near business cycletroughs, when conditions are expected to improve, and narrownear peaks, when conditions are expected to worsen. DEF is widewhen business conditions are poor and narrow when conditionsare favorable. Chen (1991) shows that DEF is negatively associatedwith GDP growth over the following two quarters while TERM ispositively associated with GDP growth over the following fivequarters. Merton (1973) and Shanken (1990) suggest that theshort-term interest rate is a natural candidate for a state variablethat captures variations in investment opportunities, while Chen(1991), among others, shows that high short-term rates signal fu-ture downturns. The TED spread compares 3-month LIBOR, whichincludes a premium for credit risk in the interbank loan market,and the yield on the 3-month T-Bill, which is free of default risk.Higher values of TED signify tightness in credit conditions andunsettled markets.8

Consumer confidence (CONFID) reflects consumer perceptionsregarding economic conditions and has been used by Qui and

7 We are not aware of a comprehensive list of events commonly accepted as crises.Our list of crises is constructed from various sources, e.g. Allen and Gale (2007), aswell as a review of the financial press.

8 For instance, TED reached a historical high in excess of 4% in October 2008, duringthe subprime crisis.

Table 2Summary statistics for excess flow by asset category.

Mean Median Q1 Q3 STD

Panel (a)Equity 3.44 1.37 �145.87 158.89 238.78Bond �31.01 �28.32 �109.74 43.09 181.36Money Market �4.74 0.13 �319.88 296.02 476.27Foreign Equity 36.35 30.57 �24.15 100.59 105.32

Equity Bond Money Market Foreign Equity

Panel (b)Equity 1Bond 0.01 1Money Market �0.66 �0.25 1Foreign Equity 0.49 �0.14 �0.48 1

Panel (a) of this table reports descriptive statistics for monthly excess flow, whereexcess flow is calculated as aggregate net flow for each asset category in excess ofthe net flow that would have resulted on an asset-weighted basis, standardized bythe previous month’s mutual fund industry total net assets (�10�6). Excess flow isreported separately for four mutual fund types: equity, bond, money market andforeign equity. Panel (b) reports the correlation matrix of excess flow across fundtypes. Here, values significant at the 5% level appear in bold face.

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3323

Welch (2006) as a measure of investor sentiment.9 Both interpreta-tions of this variable imply a positive relation between CONFID andallocations to riskier asset classes. To study the effects of stock andbond market uncertainty on allocations, we include stock marketand short-term interest rate volatility, SPV and TBV.

Table 1 summarizes the predicted coefficients on our explana-tory variables when we estimate regressions with excess flow forthe equity, bond, money market and foreign equity asset categoriesas the dependent variables. The predictions are clearest at theopposite ends of the risk spectrum, i.e. for equity funds and moneymarket funds. Investors should increase exposure to equity fundsand reduce exposure to fixed-income funds when they expect busi-ness conditions to remain strong or improve, and do the oppositewhen they expect conditions to deteriorate. Thus, allocations toequities should increase as TERM, CFNAI and CONFID increaseand as DEF, DTB and TED drop. Allocations to money market fundsare predicted to react in the opposite direction. Investors might beexpected to reduce their equity allocation and increase theirmoney market allocation at times of elevated stock or bond marketuncertainty (higher SPV or TBV).

To the extent that both foreign and domestic stocks are moresensitive to US business cycle fluctuations than are money marketfunds, investors will adjust allocations to foreign funds and domes-tic funds in the same direction. However, since international busi-ness cycles are imperfectly correlated, it is possible that investorswill seek the relative safety of foreign equities when US economicconditions deteriorate and will invest at home when conditions aremore favorable. If the payoffs to bond funds are largely safe, we ex-pect the allocation to bonds to behave similarly to the moneymarket allocation.10

A possible objection is that we are estimating allocations to theasset classes independently of each other, when the allocationdecisions are in fact simultaneous. We address this issue by also

9 As alternative proxies for investor sentiment, we consider the Baker and Wurgler(2006) sentiment index as well as the components of this index, such as the numberof IPOs, IPO underpricing and trading volume. Their inclusion does not change ourconclusions regarding the other variables of interest. We use consumer confidencebecause, at the time of data collection, the Baker–Wurgler index ended in 2005.

10 The bond asset category includes not just government bond funds but also munifunds and corporate bond funds, which are likely to have business cycle exposures.On balance, therefore, we expect the net effect of economic conditions to be less sharpfor bond funds than for money market funds. We do not create finer bondclassifications (e.g. government versus corporate bonds) owing to the relativelysmall size of each category—each comprises between 3% and 6% of overall mutualfund industry net assets.

modeling the difference between the equity and money mar-ket allocations. This captures the joint decision to allocate to equi-ties versus money market funds, and is similar to a reduced formequation.

3. Results

3.1. Descriptive statistics

Table 2, Panel (a) presents statistics on excess flow for the fourasset classes. Mean and median excess flow are close to zero forequity and money market funds, suggesting that net monthly flowsfor the equity and money market classes are roughly in line withtheir asset weights over the 1991–2008 sample period. The stan-dard deviation of excess flow is higher for domestic than foreignequities and highest for domestic equity and money market funds,the classes at the two extremes of the risk spectrum.

Fig. 2 plots excess flow for equity and money market funds overthe sample period. There are no obvious trends in the series thoughthey appear heteroskedastic. The greater volatility of money mar-ket flow is apparent. Also, there is a tendency for equity and moneymarket flow to move in opposite directions. However, there areseveral periods when they move together, e.g. 1993–1995.

Panel (b) of Table 2 provides the time-series correlations amongthe excess flows for the four asset categories. Consistent with therisk ordering of the series, domestic equity flow is positively corre-lated with foreign equity flow (0.49) and negatively correlatedwith money market flow (�0.66).11 Thus, investors appear to putmoney into, or pull money out of, domestic and foreign equity fundsat the same time. The strong negative correlation between equityand money market flow confirms the visual evidence in Fig. 2. Bondflow is not significantly correlated with equity flow and is negativelycorrelated with money market flow (�0.25).

Panel (a) of Table 3 provides descriptive statistics for the inde-pendent variables in our tests. The Chicago Fed National ActivityIndex (CFNAI) has a mean near zero, by construction, but the Q1and Q3 values of �0.35 and 0.43 imply substantial time-series var-iation. TERM, the difference between long-term and short-termbond yields, averages 2% per year. DEF, the spread between riskyand safe bond yields, averages 1.7% per year. The mean change inthe annualized T-Bill rate is �0.025% (median zero), while theaverage TED spread is 0.5% (annualized). The mean values of SPVand TBV, S&P 500 and T-Bill volatility, are 0.21% and 1% per month.The return to the US equity fund category has a mean of 11% and astandard deviation of 14% (annualized), while the values for theother three classes (not reported) are somewhat lower. By con-struction, CONFID has a mean close to 100.

Table 3, Panel (b) reports the time-series correlations amongthese variables. TERM and CONFID are highly negatively correlated(�0.71), consistent with the forward-looking nature of TERM. Thecorrelation of 0.58 between DEF and SPV is consistent with periodsof poor economic conditions also seeing high stock marketvolatility (e.g. Schwert, 1989). CFNAI is correlated negatively withDEF (-0.48) and positively with DTB (0.42), in line with the ideathat periods of economic prosperity see low corporate defaultsand rising interest rates.

3.2. Economic conditions, crises and asset allocation

The first two questions we address are whether economic con-ditions and crises influence asset allocation decisions. Table 4 pre-sents the results of regressions of excess flow on the proxies for

11 Goetzmann et al. (1999) report a correlation of �0.44 between aggregate USequity and money market flow.

Excess flow for the equity and money market asset categories.

Exc

ess

Flow

Fig. 2. Excess flow for the equity and money market asset categories. This figure shows monthly excess flow for the Equity and Money Market asset categories betweenFebruary 1991 and March 2008. Excess flow is calculated as aggregate net flow for the asset category in excess of the net flow that would have resulted on an asset weightedbasis, standardized by the previous month’s fund industry total net assets.

Table 3Summary statistics for independent variables.

Mean Median Q1 Q3 STD

Panel (a)CFNAI 0.007 0.060 �0.350 0.430 0.580TERM 2.047 1.846 0.918 3.469 1.434DEF 1.653 1.524 1.279 1.910 0.557DTB �0.025 0.000 �0.110 0.110 0.244CONFID 100.3 101.4 83.6 116.1 23.835TED 0.468 0.386 0.270 0.566 0.310RET 0.904 1.352 �1.824 3.780 4.105SPV 0.214 0.141 0.075 0.252 0.221TBV 1.045 0.168 0.086 0.395 8.212

TERM DEF DTB CONFID TED RET SPV TBV CFNAI

Panel (b)TERM 1DEF �0.154 1DTB �0.072 �0.24 1CONFID �0.712 0.211 0.139 1TED �0.4 0.109 �0.335 0.246 1RET �0.035 �0.105 0.063 �0.026 �0.008 1SPV �0.069 0.578 �0.317 0.226 0.189 �0.266 1TBV 0.064 0.14 �0.182 �0.112 0.272 �0.045 0.157 1CFNAI �0.017 �0.478 0.419 0.201 �0.05 �0.034 �0.212 �0.145 1

Panel (a) of this table reports descriptive statistics for the independent variables used as proxies for economic conditions. The correlation matrix for these variables appears inPanel (b), with values significant at the 5% level in bold face. The variables are as defined in Table 1. All variables are presented in percentage terms, with the exception ofCONFID and CFNAI, which are reported as index levels relative to neutral values of 100 and 0, respectively.

3324 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

economic conditions and controls. We scale the dependent andindependent variables by their standard deviations; this standard-ization allows us to directly assess economic significance. We alsocalculate Newey–West t-statistics that account for conditional het-eroskedasticity and autocorrelation.

To get a sense of the importance of variables from the real econ-omy, we first estimate the effects of the previous month’s CFNAI, acomposite proxy for the state of the economy, on mutual fund allo-cations. Panel (a) shows that excess flow for domestic and foreignequities is positively and significantly related to CFNAI with t-sta-tistics of 2.70 and 3.70, respectively. Excess bond and money mar-ket flow is negatively related to CFNAI, although only thecoefficient for bond flow is significant (t-statistics of 2.48 and0.95). The difference in excess flow for equity and money marketfunds is positively related to CFNAI, with a t-statistic of 1.92. Theseunivariate results are consistent with the hypothesis that investors

adjust allocations towards riskier investments as real economicconditions improve and towards safer investments as conditionsdeteriorate.

Panel (b) presents models that use financial market proxies foreconomic conditions. Our analysis of crises is subject to three crit-icisms: the list of crises is defined ex post, is subjective, and mightnot be comprehensive. To address these concerns, we report twospecifications of each model, one with the CRISIS variable andthe other without CRISIS.

When CRISIS is excluded from the model, we find that equityflow is positively and significantly associated with lagged TERMand negatively and significantly associated with both lagged DEFand lagged TED. Thus, an expected improvement in economic con-ditions (high values of TERM) causes investors to increase theirallocation to equity funds. A deterioration in economic conditions(high DEF) or tightness in financial markets (high TED) leads to

14 Gil-Bazo and Ruiz-Verdu (2009) show that funds with worse before-fee perfor-

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3325

reduced equity allocations. The coefficients on DTB and CONFIDare not significantly different from zero. Interestingly, equity ex-cess flow is unrelated to equity market volatility (SPV) but posi-tively associated with bond market volatility (TBV). Moneymarket excess flow is negatively associated with TERM and TBVand positively associated with TED. Thus, money market allocationsincrease during periods of financial tightness, and decline wheneconomic conditions are expected to improve or when interest rateuncertainty rises.

When CRISIS is included in the equity flow regression, the coef-ficient is negative and significant, confirming anecdotal evidencethat equity flow declines during major shocks. In the money mar-ket regression, the coefficient on CRISIS is positive and significant,indicating that disruptions see increased allocations to moneymarket funds. The remaining coefficients, with the exception ofthat on TED, retain their significance for the equity and moneymarket allocations. In the final column of Table 4, the differencebetween the equity and money market allocations is the depen-dent variable. Here, the coefficients on TERM, DEF, TED (in themodel without CRISIS), TBV and CRISIS are significant. Thus, asexpectations regarding economic conditions change, investors re-allocate between the two asset classes with extreme business cycleexposures.

Taken together, these results provide strong evidence that mu-tual fund investors actively shift their portfolios to less risky assetsduring crisis periods. Furthermore, after controlling for the safe-haven effects of crises, the proxies for economic conditions remainsignificant predictors of allocations to the risky equity and rela-tively safe money market asset classes. To provide a sense of eco-nomic significance, we estimate that a one standard deviationincrease in TERM results in a $790 million monthly increase inthe excess allocation to equity funds at the end of our sampleperiod.12 This represents roughly 10% of the average monthly flowto equity funds in the final year of our sample period.

The coefficient on the lagged return (included as a control var-iable) is not statistically significant. This appears to be at odds withthe robust evidence of return chasing in the literature (e.g. Brownet al., 1996; Sirri and Tufano, 1998). However, as discussed in Sec-tion 2, the excess flow variable that we study is not the same as theflow measure used in the literature that documents return chasing.When we separately consider the two components of excess flow,net flow and asset-weighted flow (NF and AWF in Eq. (3), each stan-dardized by aggregate fund assets), we find significant return chas-ing in both components. We conclude that high aggregate equityfund returns lead to larger flows to the equity asset class, but donot prompt reallocations across asset classes relative to an assetweighted benchmark. Separately, we add lags of the category re-turns (going back up to 1 year) and the lagged returns to other cat-egories (e.g. lagged bond returns to the equity allocation model).The additional return terms are usually not significant, nor do theyalter our inferences about the remaining coefficients. Hence, we donot report these models.13

The allocation to bond funds is related positively and signifi-cantly to DEF, TBV and SPV, while the remaining coefficients arenot significant. The positive coefficient on TBV suggests that inves-

12 This is based on the equity allocation model that includes crisis. In thiscalculation, we use net assets of $10.7 trillion and $3.2 trillion for the entire industryand the equity asset class, respectively.

13 This absence of return chasing is not entirely unexpected. For instance, if the sizesof all asset classes other than equity funds are held constant, a positive equity returnin month t � 1 results in an increase in the equity class weight at the end of montht � 1 and thus in AWF in month t. Thus, the return chasing seen in raw flow will onlyappear in excess flow if this return chasing is strong enough to exceed the increase inAWF. As one case where this condition is met, Frazzini and Lamont (2008) find returnchasing in a portfolio long stocks with the highest flow and short stocks with thelowest flow.

tors treat bond funds in the same way as equity funds. However,the positive coefficients on DEF and SPV indicate that investors in-crease their allocations to bond funds in the face of weak condi-tions and heightened stock market uncertainty.

For the foreign equity allocation, the coefficients are similar tothose for domestic equities. The coefficients on DEF, TED and CRI-SIS are negative and significant, and the coefficient on TBV is posi-tive and significant. Thus, deteriorating US conditions causeinvestors to reduce their allocations to foreign equities, not justdomestic equities; it does not appear that US investors opt for in-creased international diversification at such times. This resultcould also reflect the strong influence of the US economy on globaleconomic conditions.

In unreported analysis, we jointly consider the effects of CFNAIand the financial market variables and find that CFNAI loses signif-icance. This is consistent with forward-looking financial marketvariables subsuming the information in current measures of realeconomic conditions. Given this result, we focus on the financialmarket variables in the remainder of our tests.

We carry out several robustness checks. For example, we reachsimilar conclusions when we standardize excess flow by lagged netassets for the asset class or by the market capitalization of theNYSE, AMEX and NASDAQ, as in Warther (1995). We also find sim-ilar results when we include lagged excess flow as an additionalindependent variable. Appendix B shows that similar results followfrom an alternative allocation measure based on percent flow. Fi-nally, we add lags of the key independent variables and find thatthe coefficients on these variables are insignificantly different fromzero. The fact that these coefficients are not significant indicatesthat the allocation response is rapid and permanent. Further, thefact that the coefficients are not negative suggests that flow doesnot over-react to the economic variables.

We conclude from the results in Table 4 that flight-to-safetyconsiderations appear to influence mutual fund investors’ collec-tive asset allocation decisions during crises. Furthermore, aboveand beyond such crisis-driven reallocations, fund investors’asset allocations react to forward-looking financial variables in amanner one would expect of sophisticated investors.

3.3. Investor sophistication and asset allocation

The third question we address is the extent to which the sensi-tivity of asset allocations to economic conditions varies with inves-tor sophistication. We use fund fees and fund portfolio turnover asproxies for investor sophistication. Given evidence that both fac-tors are a drag on performance, we expect more sophisticatedinvestors to hold funds with low fees and low turnover.14 We firstexamine equity index funds.15 Sophisticated investors should buylow cost index funds, which provide them with cheap market-wideexposure. Investors who buy higher fee index funds are likely to beespecially unsophisticated, given the high level of homogeneity inindex fund portfolios. We then repeat the fee analysis for all

mance charge higher fees, and Carhart (1997) and Gompers and Metrick (2001) findthat net fund return is negatively related to fund turnover. Houge and Wellman(2006) show that mutual funds segment customers by their level of sophisticationand then charge less-knowledgeable investors higher fees. We do not separatelyconsider 12b-1 fees (which are a component of expense ratios) or loads assophistication proxies for the following reasons. 12b-1 fees are constrained to amaximum value of 1% and many funds charge the maximum. Depending on investorhorizon, front or back end load share classes may have lower overall costs than a no-load share class charging 12b-1 fees.

15 The index fund sample includes all equity funds for which the word ‘‘index’’appears in the objective name. In an average month, there are approximately 120funds in the sample. The vast majority of these funds are funds tracking the S&P 500index

Table 4Asset category excess flow and economic conditions.

Equity Money Market Bond Foreign Equity Equity–Money Market

Panel (a)CFNAIt�1 0.20 �0.07 �0.18 0.29 0.27

(2.70) (0.95) (2.48) (3.70) (1.92)R2 0.04 0.00 0.01 0.09 0.02

Equity Money Market Bond Foreign Equity Equity – MoneyMarket

Panel (b)TERMt�1 0.25 0.31 �0.19 �0.23 0.08 0.08 0.00 0.05 0.45 0.55

(2.29) (2.81) (1.80) (2.31) (0.71) (0.71) (0.01) (0.46) (2.90) (3.58)DEFt�1 �0.17 �0.16 0.06 0.06 0.13 0.13 �0.22 �0.20 �0.23 �0.21

(2.36) (2.27) (1.09) (1.07) (2.11) (2.11) (2.50) (2.39) (1.91) (1.81)DTBt�1 0.12 0.14 �0.11 �0.12 0.09 0.09 0.13 0.14 0.24 0.27

(1.14) (1.46) (1.00) (1.26) (0.95) (0.95) (1.39) (1.72) (1.14) (1.46)CONFIDt�1 0.13 0.14 0.11 0.09 �0.18 �0.18 �0.21 �0.20 0.02 0.06

(1.09) (1.26) (1.16) (1.04) (1.63) (1.63) (1.86) (1.69) (0.12) (0.28)TEDt�1 �0.19 �0.06 0.23 0.12 �0.11 �0.11 �0.23 �0.13 �0.42 �0.17

(2.20) (0.62) (3.01) (1.35) (1.20) (1.12) (3.19) (1.48) (2.72) (0.93)RETt�1 0.02 �0.03 0.02 0.01 0.12 0.12 0.08 0.03 0.03 �0.13

(0.27) (0.40) (0.21) (0.12) (1.53) (1.52) (1.36) (0.60) (0.19) (0.87)SPVt�1 �0.01 0.04 0.00 �0.05 0.13 0.13 �0.10 �0.08 �0.02 0.06

(0.10) (0.33) (0.03) (0.52) (1.83) (1.67) (1.14) (0.74) (0.13) (0.30)TBVt�1 0.09 0.10 �0.11 �0.11 0.05 0.05 0.14 0.15 0.20 0.21

(3.75) (4.45) (5.86) (6.81) (2.07) (2.07) (7.83) (8.11) (4.73) (5.61)CRISIS �0.26 0.20 0.00 �0.21 �0.50

(2.47) (2.15) (0.01) (2.77) (2.54)R2 0.17 0.22 0.24 0.25 0.12 0.12 0.32 0.35 0.20 0.24

This table reports coefficients and Newey–West t-statistics from time-series regressions of monthly excess flow for four asset categories (equities, bonds, money market, andforeign equities) on lagged asset category return and proxies for economic conditions. The dependent variable in the final column is the difference between equity and moneymarket excess flow. Excess flow is calculated as aggregate net flow for each asset category in excess of the net flow that would have resulted on an asset-weighted basis,standardized by the previous month’s fund industry total net assets. The independent variables are lagged by one period and are as described in Table 1, except for CRISIS,which is a dummy variable equal to one for the following financial market shocks: the subprime mortgage credit crisis (August 2007 to the end of the sample), the 9/11terrorist attacks (September–December 2001), the ‘‘Crash of 2000’’ when the NASDAQ dropped 45.9% between September and December 2000, Y2 K concerns at the end of1999 (October–December 1999), the failure of the Long Term Capital Management Hedge Fund (August–October 1998), the Asian currency crisis (July–September 1997), theMexican currency crisis (December 1994–February 1995) and the Pound exiting the European Exchange Rate Mechanism (September–November 1992). Panel (a) presentsresults related to real economy variables and Panel (b) focuses on financial market variables. In Panel (b), two models are reported for each dependent variable, including andexcluding CRISIS. The dependent and independent variables are standardized and intercepts are not reported for brevity. Coefficients significant at the 10% level appear in boldface.

3326 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

domestic equity funds and, for robustness, also examine the effectsof the portfolio turnover proxy.16

We form the fee groups as follows. At the start of each month,we examine the most recent cross-sectional distribution of feesfor index or all domestic equity funds and classify a fund as belong-ing to: (i) the low fee group if fees are in the first quartile, (ii) thehigh fee group if fees are in the fourth quartile, and (iii) the inter-mediate fee group if fees are in the second or third quartiles. UsingEqs. (1)–(3), we then compute aggregate excess flow for the threefee groups in the month. The turnover sort proceeds identically.The monthly spread in fees and turnover is large: on average, thefourth quartile values for fees and turnover are six times, twiceand thrice the first quartile values, i.e. 1.2% versus 0.18% for indexfund fees, 1.8% versus 0.9% for equity fund fees, and 1.15 versus0.34 for turnover. Thus, the variations in fees and turnover are eco-nomically material.

Table 5, Panel (a) reports the results when we sort the indexfund sample based on fees. Allocation to low fee index funds ishighly responsive to variations in economic conditions. Excess flowfor these funds increases with increases in TERM, CONFID and TBVand drops with increases in DEF. Furthermore, low fee index fundsalso see significant outflows during crises. In contrast, excess flowfor high fee funds responds only to DEF. Excess flow for intermedi-ate fee funds displays intermediate sensitivity to the economic

16 The average cross-sectional correlation between fund fees and fund turnover is0.03 and is insignificant, suggesting that fees and turnover provide independentevidence on allocations.

variables, with the coefficients on DEF and SPV being significantor marginally significant.

A comparison of the coefficients shows that economic signifi-cance for the variables tends to be largest for low fee funds andsmallest for high fee funds. Further, the difference in allocationsfor low versus high fee funds (reported in the final column) is sig-nificantly related to TERM, DEF, CONFID, TBV and CRISIS. Finally,the regression R2 drops dramatically for high fee relative to lowerfee index funds, e.g. from 0.20 to 0.12 in the full specification.

Turning to the fee partitions for the universe of domestic equityfunds in Panel (b), we find similar results. Flow for low fee funds istypically more responsive to variations in economic conditions. Forexample, the R2 in the full specification is 0.29 versus 0.16 for lowand high fee funds, respectively. We interpret these results to im-ply that more sophisticated mutual fund investors are sensitive tochanging economic conditions while naïve fund investors tend tobe relatively insensitive. In fact, the general lack of significanceof the economic variables outside the low fee group suggests thatsophisticated fund investors are driving the results for aggregateallocations we have documented in the previous sub-section.

Panel (c) provides the results of the turnover sort. The resultsare stronger than those for fees. For low turnover funds, the coef-ficients on TERM, CONFID and TBV are positive and significant,while those on DEF, TED and CRISIS are negative and significant.In the case of high turnover funds, only DTB is significant at con-ventional levels. Excess flow for funds with intermediate turnoveris significantly related to the same variables that drive flow for thelow turnover funds, though with smaller coefficients. As with fees,

Table 5Equity excess flow and economic conditions: Fee and turnover sub-samples.

Low Fee Intermediate Fee High Fee Low–High

Panel (a)TERMt�1 0.27 0.34 0.13 0.16 0.10 0.12 0.17 0.22

(2.11) (2.68) (1.09) (1.28) (0.96) (1.15) (2.21) (2.76)DEFt�1 �0.20 �0.20 �0.34 �0.33 �0.16 �0.16 �0.04 �0.04

(2.05) (2.34) (4.24) (4.23) (2.21) (2.18) (2.14) (2.36)DTBt�1 �0.08 �0.06 �0.03 �0.02 �0.07 �0.07 �0.01 �0.05

(1.09) (0.95) (0.38) (0.31) (0.88) (0.85) (1.10) (0.93)CONFIDt�1 0.28 0.31 0.13 0.13 0.22 0.23 0.06 0.08

(2.17) (2.56) (0.88) (0.94) (1.33) (1.38) (2.31) (2.65)TEDt�1 �0.11 0.06 �0.01 0.04 0.11 0.16 0.22 �0.10

(1.40) (0.74) (0.15) (0.59) (1.05) (1.44) (1.56) (0.58)RETt�1 0.15 0.08 0.11 0.09 0.08 0.06 0.07 0.02

(2.23) (1.26) (1.25) (0.94) (1.21) (0.95) (2.02) (1.20)SPVt�1 �0.05 0.02 0.19 0.21 0.09 0.11 �0.14 �0.09

(0.61) (0.24) (1.66) (1.95) (1.56) (1.73) (0.59) (0.21)TBVt�1 0.08 0.09 �0.01 �0.01 �0.05 �0.05 0.13 0.14

(3.67) (4.68) (0.24) (0.14) (1.11) (1.07) (4.26) (5.32)CRISIS �0.34 �0.11 �0.10 �0.24

(3.40) (1.33) (1.00) (3.36)R2 0.13 0.20 0.10 0.11 0.07 0.08 0.06 0.12

Panel (b)TERMt�1 0.29 0.37 0.14 0.16 0.15 0.17 0.14 0.20

(2.89) (3.92) (1.27) (1.52) (0.81) (0.91) (2.91) (3.93)DEFt�1 �0.16 �0.14 �0.16 �0.15 �0.28 �0.28 0.12 0.14

(2.03) (2.09) (2.08) (1.99) (2.75) (2.69) (1.36) (1.44)DTBt�1 0.08 0.11 0.16 0.16 0.05 0.06 0.03 0.05

(0.79) (1.20) (1.41) (1.54) (0.53) (0.60) (0.14) (1.16)CONFIDt�1 0.14 0.16 �0.01 0.01 0.20 0.20 �0.06 �0.04

(1.15) (1.47) (0.12) (0.05) (1.39) (1.43) (1.50) (1.21)TEDt�1 �0.24 �0.05 �0.13 �0.07 �0.10 �0.04 �0.14 �0.01

(2.30) (0.53) (1.54) (0.67) (1.01) (0.42) (2.80) (0.48)RETt�1 0.05 �0.03 0.03 �0.06 0.06 0.03 �0.01 0.00

(0.58) (0.37) (0.29) (0.56) (0.79) (0.46) (0.34) (0.48)SPVt�1 �0.02 0.04 0.01 0.03 0.02 0.04 0.00 0.00

(0.23) (0.31) (0.12) (0.39) (0.25) (0.43) (0.97) (0.28)TBVt�1 0.14 0.15 0.04 0.05 0.00 0.00 0.14 0.15

(5.66) (7.03) (1.76) (1.98) (0.07) (0.01) (7.55) (7.72)CRISIS �0.36 �0.13 �0.11 �0.25

(3.34) (1.40) (1.22) (3.49)R2 0.21 0.29 0.13 0.14 0.12 0.13 0.09 0.16

Low turnover Intermediate turnover High turnover

Panel (c)TERMt�1 0.35 0.42 0.21 0.27 0.07 0.09 0.28 0.33

(3.35) (4.29) (1.75) (2.18) (0.58) (0.76) (2.95) (3.70)DEFt�1 �0.19 �0.18 �0.17 �0.16 �0.02 �0.02 �0.17 �0.16

(2.55) (2.76) (2.02) (1.92) (0.18) (0.22) (2.19) (2.30)DTBt�1 �0.02 0.00 0.09 0.11 0.27 0.28 �0.29 �0.29

(0.25) (0.05) (0.85) (1.11) (2.59) (2.69) (1.53) (1.53)CONFIDt�1 0.26 0.28 0.15 0.16 �0.05 �0.04 0.31 0.32

(1.96) (2.24) (1.28) (1.45) (0.33) (0.28) (1.78) (1.94)TEDt�1 �0.32 �0.16 �0.20 �0.07 0.01 0.07 �0.33 �0.23

(3.40) (1.69) (2.11) (0.63) (0.18) (0.76) (3.60) (2.28)RETt�1 0.06 0.01 0.07 0.01 0.07 0.04 �0.01 0.02

(0.72) (0.17) (0.80) (0.18) (0.85) (0.59) (0.36) (0.46)SPVt�1 �0.04 0.02 0.00 0.05 0.02 0.04 �0.02 0.04

(0.44) (0.17) (0.00) (0.38) (0.23) (0.36) (0.64) (0.02)TBVt�1 0.16 0.17 0.08 0.09 0.03 0.03 0.13 0.14

(6.71) (8.05) (3.15) (3.77) (1.13) (1.25) (6.19) (7.03)CRISIS �0.34 �0.26 �0.11 �0.23

(3.63) (2.32) (1.15) (3.78)R2 0.27 0.34 0.14 0.19 0.07 0.08 0.20 0.26

This table reports coefficients and Newey–West t-statistics from time-series regressions of monthly excess flow for equity fund sub-samples on the lagged sub-sample returnand proxies for economic conditions. Excess flow for the sub-sample of interest is calculated as aggregate net flow for that sub-sample in excess of the net flow that wouldhave resulted on an asset-weighted basis, standardized by the previous month’s fund industry total net assets. The dependent variable in the final column is the differencebetween excess flow for the Low and High portfolios. The variables are as defined in Table 1, and are standardized with intercepts not reported for brevity. Two models arereported for each dependent variable, including and excluding CRISIS. Coefficients significant at the 10% level appear in bold face. Panel (a) reports the results for a sort basedon domestic equity index fund fees, Panel (b) for a sort based on fees for all domestic equity funds and Panel (c) for a sort based on portfolio turnover for all domestic equityfunds. Funds are sorted into quartiles based on reported fee or turnover values at the end of the prior month and the middle two quartiles combined into the Intermediatecategory (details in Section 3.3).

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3327

the R2 is sharply higher for low turnover funds than higher turn-over funds. To the extent that sophisticated investors hold fundswith low turnover, these results buttress the conclusion from Panel

(a) and Panel (b) that sophisticated investors reallocate funds moreaggressively than more naïve investors do as economic conditionschange.

3328 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

An alternative explanation for our results is that high fee fundsmight also have high back-end loads. Further, actively managedfunds with greater portfolio turnover charge higher fees (Christof-fersen and Sarkissian, 2011) and may also be more likely to imposeback-end loads. Thus, our results for the fee and turnover sortscould reflect rational investor responses to switching costs ratherthan differential sensitivities to changing economic conditions.However, Nanda et al. (2009) find that back-end load funds tendto have flows that are more sensitive to performance. Bergstresseret al. (2009) find little difference in performance chasing betweenhigher fee (broker-sold) and lower fee (directly sold) funds. Thesestudies suggest that loads are unlikely to deter investors frommoving between funds. If anything, the greater sensitivity of inves-tor flow for high load funds shown in Nanda et al. (2009) suggeststhat high load fund flow should be more (not less) sensitive to eco-nomic conditions.

4. Implications of time-varying asset allocation

This section examines the implications of economic conditions-based reallocation strategies for portfolio performance. As aprecursor to this analysis, we consider the predictive power of allo-cations for volatility and the equity risk premium. By examiningthe relation between allocations and returns, we add to a largeliterature that studies flow and returns (e.g. Warther, 1995; Edelenand Warner, 2001).

Several papers explore the optimality of investment strategiesthat condition on variables shown to forecast the economic cycle.For example, Brandt (1999) theoretically models asset allocationbetween the NYSE index and a 30-day Treasury bill for a constantrelative risk aversion investor. He finds that investor utility is opti-mized when portfolio choice varies with the dividend yield, defaultpremium, term premium and lagged excess return. Ang and Beka-ert (2004) implement a regime-switching asset allocation modelconditioning on international equity return volatility and highercorrelations coinciding with bear markets. They find that portfolioperformance can be significantly improved if investors dynami-cally alter allocations to cash, bonds and equities, e.g. moving intocash when persistent bear markets hit. Mutual funds provide anarena where such regime-dependent asset allocation strategiescan be implemented conveniently and at low cost, particularlyfor funds with no back or front-end fees.

It is important to note that the optimality of business condi-tions-driven allocation strategies must stem largely from riskavoidance. This conclusion follows from the large theoretical andempirical literature linking expected returns and volatility to busi-ness conditions. The consumption smoothing models of Lucas(1978) and Breeden (1979), among others, suggest that expectedreturns rise during economic downturns, when consumption islow and thus the marginal utility of consumption is high. A sepa-rate argument is that expected returns are high when economicconditions are poor because equity investments are riskier.17

Empirically, Keim and Stambaugh (1986) and Fama and French(1989), among many others, suggest that expected returns are highwhen economic conditions are poor and low when conditions arefavorable. Our results suggest that switching investors are out ofthe market during downturns, when expected returns and volatilityare likely high, and in the market during good times, when expectedreturns and volatility are low. Consequently, such a strategy would

17 For example, Schwert (1989) and Whitelaw (1994) show that economicdownturns are associated with high volatility. A possible source of the highervolatility during bear markets is that the correlations among equities increase indown markets (e.g. Ang and Chen, 2002). French et al. (1987), Campbell andHentschel (1992) and Bekaert and Wu (2000) provide evidence of a positive relationbetween the expected market risk premium and market volatility.

be expected to underperform strategies with continuous equityexposure on a raw return basis and may only be optimal in a risk-ad-justed sense.

4.1. Equity allocations, the equity risk premium and realized volatility

We start by studying whether the equity premium and the vol-atility of market returns are higher (lower) when equity fundinvestors are out of (in) the market. Our tests examine the predic-tive power of the aggregate equity allocation in period t for theequity risk premium and realized volatility in period t + 1. Thefee and turnover analysis in Table 5 suggests that allocations forlow fee and low turnover funds are most sensitive to economicconditions. Motivated by this finding, we separately study the pre-dictive power of excess flow to index funds in the low fee quartileand domestic equity funds in the low fee and portfolio turnoverquartiles (as defined in Section 3.3).

As before, realized volatility is measured as the sum of thesquared daily returns to the S&P 500 index in the month. As aproxy for the equity risk premium, we use the price-earnings(PE) ratio for the S&P 500 index, calculated as the ratio of aggregatemarket value divided by aggregate earnings to constituent firms(e.g. see Longstaff et al., 2011). The intuition behind using a broadmarket PE as an estimate of the equity risk premium is that, ceterisparibus, the PE ratio is inversely related to expected returns (in itssimplest form, P

E ¼ 1r�g, where r is the equity risk premium and g is

the growth rate of aggregate earnings). Campbell and Vuolteenaho(2004) recommend using a moving average of earnings over sev-eral years to smooth the effects of earnings spikes. Hence, we alsocalculate a second, trailing PE ratio using average earnings to theconstituent firms over the prior 2 years.

Three other effects could cloud the interpretation of the PE ratioin this context. First, there is the possibility of a mechanical posi-tive relation between PE and equity allocations, if aggregate mu-tual fund flows contemporaneously affect stock market prices. Tocontrol for this effect we include in the model the market returnthat is contemporaneous to flow. Second, Conrad et al. (2002) ar-gue that the PE ratio has an alternative, behavioral interpretation,e.g. the PE ratio will be high during periods of excessive optimism.We control for this effect by adding lagged consumer confidence asa proxy for consumer sentiment, and expect to find a positive rela-tion between PE and confidence. Finally, to this model we addearnings growth, calculated as the percent change in aggregateearnings, to isolate the effects of discount rates on PE.

Table 6 contains the results. As previously discussed, one impli-cation of altering asset class weights based on business cycle risk isthat investors overweight equities when the reward for holdingsuch assets is low (i.e. when the equity premium is low and thePE ratio is high). Consistent with this implication, we find a posi-tive and significant relation between the PE ratio in month t + 1and the four equity allocation measures in month t; the resultsare stronger with the trailing PE ratio. Thus, when fund investorsincrease their allocations to equity funds, the subsequent PE ratiois high and the equity premium is relatively low. In line with thearguments made by Conrad et al. (2002), we find a positive and sig-nificant relation between the PE ratio and consumer confidence.The relation between PE and the other control variables isinsignificant.

The second implication of a business cycle-motivated tradingstrategy is risk mitigation. Risk and reward go hand-in-hand inan efficient market; if risk is tied to business conditions, fundinvestors might be willing to surrender higher returns as the priceof avoiding periods of higher risk. Consistent with this motive, wefind a significant, negative relation between S&P 500 volatility andthe lagged equity allocation across the flow measures. This result

Table 6Excess flow, the equity risk premium and realized volatility.

AGG FLOWt�1 INDEX FLOWt�1 FEE FLOWt�1 TURN FLOWt�1 CONFIDt�1 EGt�1 SPRETt�1 R2

PE 0.12 0.22 �0.09 �0.01 0.06(1.77) (3.28) (1.34) (0.18)

PE 0.14 0.20 �0.08 �0.03 0.07(2.16) (2.99) (1.16) (0.40)

PE 0.17 0.24 �0.10 �0.02 0.08(2.46) (3.46) (1.48) (0.28)

PE 0.18 0.24 �0.10 �0.03 0.08(2.62) (3.45) (1.47) (0.41)

Trailing PE 0.17 0.03(2.52)

Trailing PE 0.19 0.04(2.76)

Trailing PE 0.21 0.05(3.12)

Trailing PE 0.20 0.04(3.01)

SPV �0.15 0.02(2.11)

SPV �0.12 0.02(1.74)

SPV �0.16 0.02(2.27)

SPV �0.19 0.04(2.81)

This table reports coefficients and Newey–West t-statistics from time-series regressions of PE, the equity risk premium (proxied by the S&P 500 price-earnings ratio) and SPV,S&P 500 volatility, on equity fund excess flow and controls. PE is the log of the price-earnings ratio for the S&P 500 index using the most recent earnings. Trailing PE is the logof the price-earnings ratio calculated using a moving average of earnings over the prior 2 years. SPV is the sum of squared daily within-month returns to the S&P 500 index.AGG FLOW, INDEX FLOW, FEE FLOW and TURN FLOW are excess flow to domestic equity funds (AGG) and domestic equity funds in the bottom index fee (INDEX), fee (FEE) andportfolio turnover (TURN) quartiles. The fee and portfolio quartiles are defined in Section 3.3. CONFID is the consumer confidence index. EG (earnings growth) is the percentchange in aggregate earnings to firms in the S&P 500 index. SPRET is the monthly S&P 500 index return. All the independent variables are lagged by one period. The dependentand independent variables are standardized and intercepts are not reported for brevity. Coefficients significant at the 10% level appear in bold face.

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3329

suggests that high (low) equity allocations foreshadow low (high)volatility.

To summarize, when mutual fund investors reduce their equityallocation, subsequent portfolio return volatility is lower. This riskmitigation comes at a cost, as investors are out of the market pre-cisely when the premium for bearing equity risk is highest. In thenext section, we investigate the performance implications of eco-nomic conditions-based allocation strategies.

18 Note that the time the investor spends in equities versus money market funds,and thus the dynamic portfolio returns, are determined by the values of the signals.

19 An alternative is to use the fitted excess flow from the regression models in Tables3 and 4 to more directly capture business conditions-driven excess flow. A concernwith this approach is that fitted flow may spuriously capture the relation between theforecasting variables and equity returns. We implement this approach as a robustnesstest and find similar results to those reported in Table 7.

20 Quartile values are based on the entire equity fund excess flow monthly seriesfrom 1991 to 2008.

4.2. Dynamic asset allocation portfolio performance

In order to evaluate the relative performance of such dynamicasset allocation strategies, we examine the returns for two setsof portfolios. The first is a buy-and-hold equity fund portfolio,which we assume receives the mean monthly return across allUS domestic equity funds. This is essentially the market portfolio(it has a beta of 1.01). The second, dynamic portfolio holds eitherequity or money market funds depending on the values of forecast-ing variables. We assume that the investor allocates 100% of hisinvestible wealth to equity funds, but switches his wealth entirelyto money market funds in anticipation of poor economic condi-tions. These trading rules are motivated by the previously dis-cussed results in Ang and Bekaert (2004), and by the strongnegative relation between equity and money market flows, dis-cussed in Section 3.1.

An infinite number of potential asset allocation strategies exist.Our goal is simply to get a sense of the implications of adoptingtime-varying asset allocation strategies and not to definitively de-fine the performance outcome. Accordingly, we employ eight sep-arate switching signals. The first two use TERM and DEF, whichfeature prominently in the academic literature on business condi-tions (e.g. Fama and French, 1989). The investor is assumed to allo-cate 100% of wealth to money market funds when TERM drops intoits bottom quartile or DEF climbs into its top quartile, otherwise

allocating 100% of wealth to equity funds. These cutoffs are com-puted each month using the five most recent years of monthlydata.18 The portfolio is assumed to receive the mean return for theequity or money market asset class in months when wealth is allo-cated to that class.

The six remaining switching signals are based on the lagged ex-cess flow for equity funds. Here, investors use the actual allocationdecisions of other investors. A key advantage is that actual alloca-tions reflect the influence of any switching signals used by inves-tors. A drawback is that allocations reflect not only economicsignals but also other trade motives. To minimize the latter influ-ence, we focus on the excess flow for funds in the bottom indexfund fee, equity fund fee and turnover quartiles, which is most sen-sitive to economic signals.19 We assume that an investor invests100% of wealth in equity funds unless excess equity flow (computedusing Eq. (3)) is in its bottom quartile, at which point the investorswitches entirely into money market funds. The investor remainsin money market funds until excess flow moves either above itsmedian (Flow Signal 1) or into its top quartile (Flow Signal 2), atwhich point he switches entirely into equity funds.20 The median ex-cess flow value is zero, so we rely on the upper quartiles (where ex-cess flow is positive) as a signal to switch back into equities.

As risk-adjusted performance measures, we examine the Sharpeand Treynor–Black ratios. We also address the possibility of marketand volatility timing using the approach introduced by Ferson and

Table 7Portfolio analysis.

Buy-and-hold portfolio Dynamic portfolios

TERM DEF Flow Signal 1 Flow Signal 2

INDEX FEE TURN INDEX FEE TURN

Mean Return 10.42 7.77 8.61 10.25 8.63 9.24 9.57 8.21 10.08STD Return 14.15 10.64 10.01 9.66 10.10 10.92 10.28 9.39 9.98Sharpe Ratio 0.631 0.589 0.711 0.906 0.706 0.709 0.785 0.715 0.859Market Beta 1.01 0.58 0.49 0.47 0.60 0.51 0.53 0.50 0.44Treynor–Black ratio 8.83 10.81 14.52 18.72 11.89 15.18 15.31 13.42 19.50

am 0.08 0.43 2.49 1.48 1.62 1.15 1.30 2.79(0.05) (0.25) (1.55) (0.96) (0.95) (0.73) (0.79) (1.65)

av 0.48 1.49 0.94 0.94 0.77 1.02 0.66 0.85(0.97) (1.84) (1.81) (1.67) (1.30) (1.88) (1.29) (1.50)

This table reports portfolio analysis for eight portfolios. The buy-and-hold portfolio allocates 100% of wealth to equity funds and receives the mean return to equity funds ineach month. The dynamic portfolios allocate 100% of wealth to equity funds but transfer 100% of wealth to money market funds when the term spread (TERM) or excess flowto equity funds (Flow Signal) is in its bottom quartile, or the default spread (DEF) is in its top quartile. For the TERM and DEF portfolios, 100% of wealth reverts to equity fundswhen TERM (DEF) exits the bottom (top) quartile. For the Flow Signal portfolios, 100% of wealth reverts to equity funds when flow enters the third or the fourth quartiles (FlowSignal 1) or the fourth quartile (Flow Signal 2). Here, three excess flow signals are used: excess flow to domestic equity index funds in the lowest fee quartile (INDEX), excessflow to all domestic equity funds in the lowest fee quartile (FEE) and lowest portfolio turnover quartile (TURN). The index, fee and turnover quartiles are constructed asdescribed in Section 3.3 and TERM and DEF are as described in Table 1. Mean Return is the annualized mean monthly return and STD Return is the annualized standarddeviation of monthly returns for each portfolio over the sample period, March 1991–April 2008 (206 observations). The Sharpe Ratio is Mean Return less the mean risk-freerate (1.5%) divided by STD Return. The Market Beta is from an asset pricing model that includes the Fama–French excess market return as the only independent variable. TheTreynor–Black ratio is calculated as the Mean Return less the mean risk-free rate divided by the Market Beta. am and av are annualized market timing and volatility timing alphaestimates based on the model in Ferson and Mo (2012). The model is estimated using the Generalized Method of Moments, with t-statistics in parentheses.

22 Our focus being asset allocation, we do not consider security selection. Thus, weestimate Eqs. (11a)–(11e) in Ferson and Mo (2012). Also, given our interest in astrategy that switches between essentially the market portfolio and the risk freesecurity, we use a one factor stochastic discount factor model that includes the excessmarket return.

23

3330 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

Mo (2012). Table 7 reports the mean return, standard deviation,Sharpe and Treynor–Black ratios, and Ferson–Mo (2012) alphasfor the portfolios. These statistics are computed using the time-ser-ies of monthly asset class returns over the entire sample period.The mean annualized returns to the eight reallocation portfolios,which range between 7.8% and 10.3%, are without exception lowerthan the mean equity buy-and-hold portfolio return (10.4%). This isas we expect, since the dynamic portfolios are likely to be out ofthe market when expected returns are relatively high.

Also without exception, the annualized standard deviation islower for the dynamic portfolios, varying between 9.4% and10.9%, than for the buy-and-hold portfolio (14.2%). To allow arisk-adjusted comparison of portfolio performance, we calculateeach portfolio’s Sharpe ratio (SR), defined as the mean portfolio re-turn less the average risk-free rate divided by the standard devia-tion of the portfolio return. The portfolio that uses TERM as thetrading signal performs slightly worse (SR of 0.59) than the buy-and-hold portfolio (SR = 0.63) on a risk-adjusted basis. However,the Sharpe ratio for the DEF portfolio is 0.71 while those for theflow signals are, on average, 19% higher than the buy-and-holdportfolio’s ratio. The portfolio based on allocations to low fee indexfunds and Flow Signal 1 has the highest Sharpe ratio of 0.91.21

In order to get a sense of the effects on market exposure, Table 7also presents the market beta and Treynor–Black ratio for the port-folios. The single-factor market betas for the eight reallocationportfolios, which range between 0.44 and 0.60, are approximatelyhalf as large as the beta for the buy-and-hold portfolio, which is1.01. The reduction in market beta occurs because the switchingstrategy invests either in equities, with a market beta of 1, or inthe money market portfolio, with a beta of essentially 0. Thus, acombination of the two portfolios will have a beta below 1 andthe actual beta will reflect the time the portfolio is in equities ver-sus money market funds. While the reallocation portfolio betamust be lower than the buy-and-hold portfolio beta, the questionof interest is how the switching strategy affects the risk-returntradeoff. As a result of the appreciably lower betas, the Treynor–

21 Formal tests of the null hypothesis that the Sharpe ratios are equal, followingJobson and Korkie (1981), never reject the null. Jobson and Korkie suggest, however,that this test often suffers from low power.

Black ratio of mean excess return to beta is, on average, 60% higherfor the switching portfolios relative to the buy-and-hold portfolio.

The market and volatility timing statistics are presented in thefinal rows of the table. As in Ferson and Mo (2012), we estimate themodel using portfolio weights and the Generalized Method of Mo-ments, with robust t-statistics in parentheses.22 The volatility tim-ing alpha is always positive and has marginal significance for threesignals, consistent with the switching investor gaining value byavoiding high volatility periods. The market timing alpha is, some-what surprisingly, also positive. However, neither is significant atconventional levels. The point estimates are modest, ranging be-tween 0.08% and 2.80% (annualized), and the market-timing termtends to be larger, although less precisely estimated.

To summarize, while the effect on risk-adjusted performance isnot clear-cut, an unambiguous benefit is the reduced level of vola-tility and systematic risk. Depending on the utility function consid-ered, risk-averse investors could plausibly realize greater utilityfrom this strategy which minimizes portfolio volatility at the costof expected returns. Thus, a switching strategy motivated by eco-nomic conditions yields potential benefits to risk-averseinvestors.23

4.3. Expected returns and economic conditions: The role of flow

In closing, we turn to the issue of the link between economicconditions and expected returns. Our results shed light on thechannels by which macro-economic signals affect prices. A largeliterature (e.g. Fama and French, 1989; Ferson and Harvey, 1991)suggests that variations in, for instance, DEF affect expected re-turns. The traditional view of this process is that investors see a

Note that this analysis has ignored the transaction costs (e.g. loads) associatedwith the dynamic strategies. We incorporate costs using the fees and holding periodsreported by Khorana et al. (2009). Due to more frequent trading, the reallocationportfolios face extra trading costs of 0.20% per annum relative to the buy-and-holdportfolio. The small size of these costs means that our conclusions will not change.Investors who hold no-load funds can avoid transaction costs altogether.

Table 8Asset category percent flow and economic conditions.

Equity Money Market Bond Foreign Equity Equity – Money Market

TERMt�1 0.03 0.07 0.12 0.09 �0.09 �0.10 �0.07 �0.04 �0.09 �0.02(0.22) (0.54) (0.66) (0.47) (0.58) (0.60) (0.66) (0.36) (1.77) (2.35)

DEFt�1 �0.33 �0.32 0.14 0.14 �0.15 �0.15 �0.36 �0.36 �0.47 �0.46(3.24) (3.07) (1.57) (1.63) (1.39) (1.38) (3.78) (3.66) (2.56) (2.34)

DTBt�1 �0.15 �0.14 �0.17 �0.19 �0.19 �0.19 �0.03 �0.02 0.02 0.05(2.00) (2.01) (1.70) (2.00) (2.39) (2.42) (0.39) (0.31) (0.10) (0.31)

CONFIDt�1 0.50 0.53 �0.15 �0.16 �0.48 �0.48 �0.38 �0.37 0.65 0.69(3.20) (3.47) (1.57) (1.73) (3.11) (3.11) (2.86) (2.76) (2.08) (2.43)

TEDt�1 �0.21 �0.12 0.11 0.03 �0.16 �0.16 �0.11 �0.05 �0.32 �0.15(2.70) (1.45) (1.39) (0.27) (2.10) (2.09) (1.42) (0.68) (3.61) (1.75)

RETt�1 0.21 0.17 0.34 0.33 0.29 0.29 0.15 0.12 �0.13 �0.16(4.21) (3.95) (2.05) (2.11) (4.05) (4.05) (2.54) (2.08) (1.11) (0.38)

SPVt�1 0.05 0.08 0.06 0.02 0.20 0.19 �0.01 0.00 �0.01 0.06(0.73) (1.21) (0.70) (0.18) (2.84) (2.62) (0.19) (0.03) (0.10) (0.59)

TBVt�1 0.02 0.03 �0.15 �0.15 �0.02 �0.02 0.01 0.01 0.17 0.18(0.97) (1.23) (7.22) (7.63) (0.80) (0.81) (0.65) (0.77) (4.73) (5.28)

CRISIS �0.19 0.18 0.01 �0.11 �0.37(2.50) (1.95) (0.17) (1.88) (2.90)

WEIGHTt�1 �0.73 �0.74 �0.39 �0.40 �0.17 �0.17 �0.03 �0.02 �0.34 �0.34(6.36) (6.57) (2.52) (2.70) (0.96) (0.95) (0.32) (0.23) (4.05) (4.85)

R2 0.46 0.48 0.22 0.24 0.31 0.31 0.39 0.40 0.34 0.39

This table reports coefficients and Newey–West t-statistics from time-series regressions of monthly percent flow for four US asset categories (equities, bonds, money market,and foreign equities) on the lagged asset category return and proxies for economic conditions. The dependent variable in the final column is the difference between equity andmoney market percent flow. Percent flow is calculated as aggregate net flow for each asset category, standardized by the previous month’s category total net assets. Theindependent variables are lagged by one period and are as described in Table 1, except for WEIGHT, which is the lagged value of the category’s total net assets divided by totalnet assets for all mutual funds. Two models are reported for each dependent variable, excluding and including CRISIS. The dependent and independent variables arestandardized and the intercepts not reported for brevity. Coefficients significant at the 10% level appear in bold face.

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3331

higher DEF in period t � 1 and immediately adjust prices down int � 1, thereby raising expected returns in periods t, t + 1, etc. Thischannel does not require any trading.

However, we provide evidence of a second, complementarychannel, where DEF affects allocations and thereby expected re-turns. As we show, higher DEF in t � 1 leads to outflows from equi-ty funds in period t. These outflows contemporaneously drivedown prices and thereby contribute to higher expected returns.Thus, our results contribute to the asset pricing literature ontime-varying expected returns: we document a role for the mutualfund investor in generating the empirically observed relation be-tween expected returns and economic variables such as DEF.

5. Concluding comments

This paper examines the aggregate asset allocation decisions ofUS mutual fund investors, focusing on the effects of economicconditions. Our results suggest that an expected deterioration ineconomic conditions leads mutual fund investors to allocate lessto equity funds and more to money market funds, while an antic-ipated improvement in conditions induces rebalancing in theopposite direction. In addition, we document significant shiftsfrom risky to less risky assets during crises. These patterns arestronger in funds with relatively low fees and turnover, consistentwith sophisticated investors being more sensitive to changingconditions.

Our research indicates that anticipated changes in economicconditions cause investors to adjust the riskiness of their assetholdings in sensible ways. Reinforcing this conclusion, investorswho follow such strategies do not face a poorer risk-return trade-off. Thus, our results imply that, in the aggregate, fund flows haveat least partly rational motivations. In contrast, fund-level researchdocuments a set of possibly irrational factors that motivate fundflows.

Our results also suggest that, taken together, fund investorsplay a meaningful role in price formation. Among others, Famaand French (1989) show that variables such as DEF and TERM havepredictive power for asset prices. We show that these predictive

variables serve as cues for reallocations across asset classes by fundinvestors. The relation between aggregate allocations and the prox-ies for economic conditions (e.g. the positive relation betweenequity flow and TERM) suggests that fund investors contribute tothe relation between the predictive variables and asset prices.

Our evidence suggests that mutual fund investors collectivelyrespond to the information in forward-looking financial variables.A reasonable question to ask is whether the financial market prox-ies for economic conditions, while well-established in academic re-search, are the types of signals that mutual fund investors arelikely to use in their asset allocation decisions. Exactly how inves-tors make these decisions reaches beyond the granularity of ourdata. However, plausible information transmission channels mayinclude financial analysts, journalists or advisors who rely onthese variables in writing their reports with asset allocationrecommendations.

Although our analysis indicates that time-varying asset alloca-tion decisions might, by lowering portfolio risk, benefit switchinginvestors, it is worth noting that the implementation of such strat-egies imposes trading costs on the affected funds and disrupts theirinvestment strategies. These costs are borne predominantly bybuy-and-hold investors who remain with the fund over the cycle.Thus, our results point to wealth transfers from buy-and-holdinvestors to more transient investors seeking to time the businesscycle. Our research also highlights a set of aggregate-level variablesthat fund managers can use to anticipate flow variations and re-duce their effects on performance.

Acknowledgements

The authors are grateful for financial support from a NationalResearch Program in Financial Services & Public Policy grant fromthe Schulich School of Business and the Social Sciences andHumanities Research Council of Canada. We thank an anonymousreferee, the editor Ike Mathur, Greg Bauer, Dick Beason, WolfgangBessler, Susan Christoffersen, Wayne Ferson, Ro Gutierrez, MarkHuson, Joachim Inkmann, Marty Luckert, Vikas Mehrotra, MatthewSpiegel, and session participants at the 2008 European Financial

3332 J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333

Management Association, 2008 Financial Management Association,2009 Northern Finance Association, 2011 European Retail InvestorConference and 2011 Asian Financial Management Associationmeetings, and seminar participants at the Bank of Canada, andthe Universities of Hitotsubashi, Kobe and Lugano for helpfulcomments.

Appendix A. Fund classifications

CRSP reports the percentage holdings of each fund across threebroad categories, equities, bonds and cash. Using these holdingsfigures, we classify any fund with more than 80% of its assets inequities, bonds or cash as an equity, bond or money market fund.Funds with less than 80% of their assets in any one category areclassified as balanced funds. We then use the fund objective codeto differentiate between domestic and foreign funds as well as toclassify funds that are missing percentage holdings data. We focuson asset categories with, on average, at least 10% of total mutualfund industry assets, and assign the remaining, smaller asset clas-ses to the ‘‘Other’’ category. Included in the ‘‘Other’’ category are,for example, mortgage backed security funds, international bondfunds and international money market funds. In the paper, we referto Domestic Equity, Domestic Money Market and Domestic Bondmore simply as Equity, Money Market and Bond.

CRSP has used three fund objective classification systems be-tween 1991 and 2008: Wiesenberger (1991–1992), Strategic In-sight (1992–1998) and Lipper (1999–2008). For most funds,multiple classification sources are available. We rely predomi-nantly on Lipper, given its significantly greater level of detail. Wie-senberger usually uses three character detailed codes (e.g. CBD forcorporate bond) but on occasion uses generic single character clas-sifications (e.g. S for stability). Funds lacking both percentage hold-ings data and non-ambiguous objective codes are excluded fromthe sample (approximately 10% of the funds in the CRSP databaseare lost).

Appendix B. An alternative asset allocation measure

As argued in Section 2, we believe that excess flow is ideal forthe purposes of studying asset allocation because it has a naturaland clear interpretation, reflecting the extent to which flow devi-ates from a value-weighted benchmark. However, to address po-tential concerns with excess flow, we calculate percent flow foreach asset category as an alternative allocation measure (e.g. seeSirri and Tufano, 1998).

Percent flow is calculated as net flow (defined in Eq. (1)) foreach asset class scaled by lagged total net assets for the asset class.We relate percent flow to the lagged economic indicators (TERM,DEF, DTB, CONFID, TED, SPV and TBV), CRISIS, lagged return(RET), as well as the lagged asset class weight (WEIGHT).24 As pre-viously discussed, percent flow avoids the potential adding-up prob-lem with excess flow, but does not capture asset allocation ascleanly. For instance, if economic conditions improve investors arelikely to increase flow to both equity and money market funds,and the precise shift towards or away from equities is more difficultto identify.

Table 8 reports the results. Paralleling our earlier results, aggre-gate percent flow for domestic equity funds is related negatively toDEF and TED, and positively to CONFID. For percent flow, unlike ex-cess flow, the coefficients on TERM and TBV are not significant, butthe coefficient on DTB now is negative and significant. Since

24 Sirri and Tufano (1998) include lagged total assets in their fund-level study. Sinceour focus is asset allocation, the relative size of each class is more relevant. We findsimilar results using lagged assets.

increases in T-Bill rates tend to foreshadow downturns, it is rea-sonable that investors move away from risky assets as short ratesincrease. As with excess flow, the negative coefficient on CRISISindicates that percent flow to equities drops during crises. Thecoefficients on the control variables RET and WEIGHT are signifi-cant. The positive coefficient on lagged returns indicates returnchasing at the asset class level. The negative coefficient on WEIGHTis consistent with Sirri and Tufano’s evidence at the fund level.

Aggregate percent flow to money market funds is associatedpositively with CRISIS and lagged DEF, and negatively with CON-FID, TBV and DTB (the last coefficient is significant at the 10% levelin one specification). Therefore, periods of crises, deteriorating eco-nomic conditions (high values of DEF) or low interest rate uncer-tainty see increased allocations to money market funds. Thecoefficient on lagged returns is significantly above zero and thaton the asset class weight significantly below zero. Thus, moneymarket funds see inflows following high money market returns.25

For bonds, percent flow is associated negatively with DTB, TEDand CONFID and positively with SPV. The first two coefficients sug-gest that bond flow behaves like equity flow, while the coefficientson confidence and equity volatility indicate that bond flow behaveslike money market flow. Foreign equity flow is negatively relatedto DEF and CONFID. High values of DEF indicate that the US econ-omy is not performing well, and US investors tend to avoid foreignstocks at such times. When confidence is high and US investors feelgood about the home economy, they tend to invest less overseas.The positive coefficient on RET for both categories is again consis-tent with return chasing.

The analysis of percent flow shows that investors increase theirallocations to riskier asset classes when the US economy is ex-pected to perform well and move to safer assets in the face of dete-riorating conditions. Thus, these results are consistent with theresults in Section 3.2 showing that asset allocation is sensitive toeconomic conditions.

References

Allen, F., Gale, D., 2007. Understanding Financial Crises. Oxford University Press,Oxford.

Ang, A., Bekaert, G., 2004. How regimes affect asset allocation. Financial AnalystsJournal 60, 86–99.

Ang, A., Bekaert, G., 2007. Stock return predictability: is it there? Review of FinancialStudies 20, 651–707.

Ang, A., Chen, J., 2002. Asymmetric correlations of equity portfolios. Journal ofFinancial Economics 63, 443–494.

Bailey, W., Kumar, A., Ng, D., 2011. Behavioral biases of mutual fund investors.Journal of Financial Economics 102, 1–27.

Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of stockreturns. Journal of Finance 61, 1645–1680.

Beber, A., Brandt, M., Kavajecz, K., 2009. Flight-to-quality or flight-to-liquidity?Evidence from the Euro-area bond market. Review of Financial Studies 22, 925–957.

Bekaert, G., Wu, G., 2000. Asymmetric volatility and risk in equity markets. Reviewof Financial Studies 13, 1–42.

Ben-Rephael, A., Kandel, S., Wohl, A., 2012. Measuring investor sentiment withmutual fund flows. Journal of Financial Economics 104, 363–382.

Bergstresser, D., Chalmers, J., Tufano, P., 2009. Assessing the costs and benefits ofbrokers in the mutual fund industry. Review of Financial Studies 22, 4129–4156.

Brandt, M., 1999. Estimating portfolio and consumption choice: a conditional Eulerequations approach. Journal of Finance 54, 1609–1645.

Breeden, D., 1979. An intertemporal asset pricing model with stochasticconsumption and investment opportunities. Journal of Financial Economics 7,265–296.

Brown, K., Harlow, W., Starks, L., 1996. Of tournaments and temptations: an analysisof managerial incentives in the mutual fund industry. Journal of Finance 51, 85–110.

Campbell, J., Hentschel, L., 1992. No news is good news: an asymmetric model ofchanging volatility in stock returns. Journal of Financial Economics 31, 281–331.

25 Shown in the final column, the difference between equity and money marketflows is negatively related to DEF and TED and somewhat surprisingly to TERM, andpositively related to CONFID and TBV.

J. Chalmers et al. / Journal of Banking & Finance 37 (2013) 3318–3333 3333

Campbell, J., Thompson, S., 2010. Predicting excess stock returns out of sample: cananything beat the historical average? Review of Financial Studies 21, 1509–1531.

Campbell, J., Vuolteenaho, T., 2004. Bad beta, good beta. American Economic Review94, 1249–1275.

Carhart, M., 1997. On persistence in mutual fund performance. Journal of Finance52, 57–82.

Chen, N., 1991. Financial investment opportunities and the macroeconomy. Journalof Finance 46, 1467–1484.

Chen, N., Roll, R., Ross, S., 1986. Economic forces and the stock market. Journal ofBusiness 59, 383–403.

Christoffersen, S., Sarkissian, S., 2011. The demographics of fund turnover. Journal ofFinancial Intermediation 20, 414–440.

Cohen, R., 1999. Asset Allocation Decisions of Individuals and Institutions. WorkingPaper, Harvard Business School.

Conrad, J., Cornell, B., Landsman, W., 2002. When is bad news really bad news?Journal of Finance 57, 2507–2532.

Cooper, M., Gulen, H., Rau, R., 2005. Changing names with style: mutual fundname changes and their effects on fund flows. Journal of Finance 60, 2825–2858.

Dichev, I., 2007. What are stock investors’ actual historical returns? Evidence fromdollar-weighted returns. American Economic Review 97, 386–401.

Edelen, R., Warner, J., 2001. Aggregate price effects of institutional trading: a studyof mutual fund flow and market returns. Journal of Financial Economics 59,195–220.

Edelen, R., Marcus, A., Tehranian, H., 2010. The relative sentiment indicator: on theasset allocation decisions of institutional vs. retail investors. Financial AnalystsJournal 66, 20–32.

Edwards, F., Zhang, X., 1998. Mutual funds and stock and bond market stability.Journal of Financial Services Research 13, 257–282.

Fama, E., French, K., 1989. Business conditions and expected returns on stocks andbonds. Journal of Financial Economics 25, 23–49.

Ferson, W., Harvey, C., 1991. The variation of economic risk premiums. Journal ofPolitical Economy 99, 385–415.

Ferson, W., Mo, H., 2012. Performance Measurement with Market and VolatilityTiming Selectivity. Working Paper, University of Southern California.

Frazzini, A., Lamont, O., 2008. Dumb money: mutual fund flow and the cross-sectionof stock returns. Journal of Financial Economics 88, 299–322.

French, K., Schwert, W., Stambaugh, R., 1987. Expected stock returns and volatility.Journal of Financial Economics 19, 3–30.

Friesen, G., Sapp, T., 2007. Mutual fund flows and investor returns: An empiricalexamination of fund investor timing ability. Journal of Banking and Finance 31,2796–2816.

Gil-Bazo, J., Ruiz-Verdu, P., 2009. The relation between price and performance in themutual fund industry. Journal of Finance 64, 2153–2183.

Goetzmann, W., Massa, M., Rouwenhorst, K., 1999. Behavioral Factors in MutualFund Flow. Working Paper, Yale University.

Gompers, T., Metrick, A., 2001. Institutional investors and equity prices. QuarterlyJournal of Economics 116, 229–259.

Gruber, M., 1996. Another puzzle: the growth in actively managed mutual funds.Journal of Finance 51, 783–810.

Henkel, S., Martin, J., Nardari, F., 2011. Time-varying short-horizon predictability.Journal of Financial Economics 99, 560–580.

Houge, T., Wellman, J., 2006. The use and abuse of mutual fund expenses. Journal ofBusiness Ethics 70, 23–32.

Huang, J., Wei, K., Yan, H., 2012. Investor Learning and Mutual Fund Flows. WorkingPaper, University of Texas.

Jain, P., Wu, J., 2000. Truth in mutual fund advertising: evidence on futureperformance and fund flows. Journal of Finance 55, 937–958.

Jobson, D., Korkie, B., 1981. Performance hypothesis testing with the Sharpe andTreynor measures. Journal of Finance 36, 889–908.

Keim, D., Stambaugh, R., 1986. Predicting returns in the stock and bond markets.Journal of Financial Economics 17, 357–390.

Keswani, A., Stolin, D., 2008. Which money is smart? Mutual fund buys and sells ofindividual and institutional investors. Journal of Finance 63, 85–118.

Khorana, A., Servaes, H., Tufano, P., 2009. Mutual fund fees around the world.Review of Financial Studies 22, 1279–1310.

Longstaff, F., Pan, J., Pedersen, L., Singleton, K., 2011. How sovereign is sovereigncredit risk? American Economic Journal: Macroeconomics 3, 75–103.

Lucas, R., 1978. Asset prices in an exchange economy. Econometrica 46, 1429–1445.Lynch, A., Musto, D., 2003. How investors interpret past fund returns. Journal of

Finance 58, 2033–2058.Merton, R., 1973. An intertemporal capital asset pricing model. Econometrica 41,

867–887.Nanda, V., Wang, Z., Zheng, L., 2009. The ABCs of mutual funds: on the introduction

of multiple share classes. Journal of Financial Intermediation 18, 329–361.Qui, L., Welch, I., 2006. Investor Sentiment Measures. Working Paper, Brown

University.Santini, D., Aber, J., 1998. Determinants of net new money flow to the equity mutual

fund industry. Journal of Economics and Business 50, 419–429.Sapp, T., Tiwari, A., 2004. Does stock return momentum explain the ‘‘smart money’’

effect? Journal of Finance 59, 2605–2622.Schwert, G., 1989. Why does stock market volatility change over time? Journal of

Finance 44, 1115–1153.Shanken, J., 1990. Intertemporal asset pricing: an empirical investigation. Journal of

Econometrics 45, 99–120.Sirri, E., Tufano, P., 1998. Costly search and mutual fund flow. Journal of Finance 53,

1589–1622.Warther, V., 1995. Aggregate mutual fund flow and security returns. Journal of

Financial Economics 39, 209–235.Whitelaw, R., 1994. Time variations and covariations in the expectation and

volatility of stock market returns. Journal of Finance 49, 515–541.Zheng, L., 1999. Is money smart? A study of mutual fund investors’ fund selection

ability. Journal of Finance 54, 901–933.