International Review of Financial Analysis 35 (2014) 178–189

Contents lists available at ScienceDirect

International Review of Financial Analysis

Liquidity risk and the performance of UK mutual funds☆

Jason Foran a, Niall O'Sullivan a,b,⁎
a
Centre for Investment Research, University College Cork, Ireland
b
School of Economics, University College Cork, Ireland

a r t i c l e i n f o a b s t r a c t

Article history: We examine the role of liquidity risk, both as a stock characteristic as well as systematic liquidity risk, in UK
Received 20 March 2014 mutual fund performance for the first time. We find that on average UK mutual funds are tilted towards liquid
Received in revised form 11 July 2014 stocks (except for small stock funds as might be expected) but that, counter-intuitively, liquidity rather than
Accepted 1 September 2014
illiquidity, as a stock characteristic is positively priced in the cross-section of fund performance. We find that
Available online 16 September 2014
systematic liquidity risk is positively priced in the cross-section of fund performance although controlling for
JEL classification:
momentum effects weakens the robustness of this finding somewhat. Overall, our results reveal a strong role
C15 for stock liquidity level and systematic liquidity risk in fund performance evaluation models.
G11 © 2014 Elsevier Inc. All rights reserved.

Keywords:
Mutual fund performance
Liquidity risk
Liquidity characteristics

1. Introduction as a risk factor refers to systematic liquidity risk, i.e., the sensitivity of
returns to changes in market liquidity that may not be diversifiable. A
During the recent financial crisis fund managers witnessed a severe number of papers demonstrate commonality in liquidity across stocks
drop in liquidity across global financial markets. This led to a large in- (Chordia, Roll, & Subrahmanyam, 2000; Hasbrouck & Seppi, 2001)
crease in trading costs and greater price impact and has heightened while Pastor and Stambaugh (2003), Acharya and Pedersen (2005),
awareness of the importance of liquidity risk. We examine the role of Chen (2005), Korajczyk and Sadka (2008) and Sadka (2006) provide
liquidity risk in mutual fund performance in the UK. The pricing of li- evidence of a premium for this systematic liquidity risk. There is also
quidity risk has attracted some attention in US studies but almost no strong evidence indicating that liquidity plays a role in asset pricing in
work has been done on the UK market. The US and UK operate under UK equities. Lu and Hwang (2007) report counter-intuitive findings
different market structures. Unlike the US where trading is fragmented, around the pricing of liquidity as a stock characteristic in the UK where
in the UK all trading takes place on a single exchange. In the US, trading liquid stocks are found to outperform illiquid stocks, Foran, Hutchinson,
on Nasdaq is order book driven while the NYSE has a hybrid system and O'Sullivan (2014b) confirm this result. Foran, Hutchinson, and
whereas in the UK, London Stock Exchange (LSE) trading is a mix of O'Sullivan (2014a) report evidence of a premium for systematic liquidity
order book driven (the Stock Exchange Electronic Trading Service risk in the UK equity market.
(SETS)) and a hybrid quote/order book driven system (SETSmm). We examine the role of liquidity risk in UK mutual fund perfor-
The differing market structures of UK and US exchanges lead to large mance. To our knowledge, in the case of the UK mutual fund industry
differences in liquidity characteristics (Huang & Stoll, 2001). Liquidity there have been no past studies of performance which control for
may be priced in two ways. Liquidity as a priced characteristic considers stocks' liquidity characteristics and systematic liquidity risk in perfor-
a stock's own liquidity as a determinant of its return. Amihud and mance. We address this gap in the literature. Using a high frequency
Mendelson (1986) argue that illiquid stocks should earn a premium tick data set, which covers much of the financial crisis period, we first
over liquid stocks to compensate investors for the trading costs incurred construct several measures of stock liquidity, some of which are not
which reduce realisable returns, e.g., wider bid–offer spreads. Liquidity possible with lower frequency daily data. We construct risk mimicking
factor portfolios for both liquidity as a stock characteristic and systemat-
ic liquidity risk. We then examine the exposure of UK mutual funds to
these liquidity risks as well as their pricing in the cross-section of fund
☆ We are grateful for financial support from the Irish Research Council and the Strategic
Research Fund, University College Cork.
performance. In particular, for the first time in the UK mutual fund in-
⁎ Corresponding author. dustry, we examine the impact on performance alphas of the inclusion
E-mail address: [email protected] (N. O'Sullivan). of both these liquidity factors.

http://dx.doi.org/10.1016/j.irfa.2014.09.001
1057-5219/© 2014 Elsevier Inc. All rights reserved.
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 179

Studies of UK mutual fund performance typically evaluate either ex- (LSPD) Archive file which records the constituents of the FTSE All Share
post risk adjusted performance or ex-ante performance persistence index historically. We cross-reference the LSE and LSPD data sets by com-
(Cuthbertson, Nitzsche, & O'Sullivan, 2012; Cuthbertson, Nitzsche, & paring SEDOL numbers.2 This leaves us with a comprehensive universe of
O'Sullivan, 2008; Fletcher, 1997; Otten & Reijnders, 2012; Quigley & stocks that UK equity mutual funds realistically choose from.
Sinquefield, 1999) Risk adjusted fund performance is typically taken Our mutual fund data set is obtained from Morningstar and contains
as the estimated alpha from a multi-factor model which attempts to monthly returns on 1141 actively managed UK equity unit trusts and
control for return attributable to various risk factors. Perhaps the most Open Ended Investment Companies. ‘UK Equity’ funds (by definition)
well established models here are the Fama and French (1996) and have at least 80% of the fund invested in UK equity. By restricting our
Carhart (1997) models which control for market, size, value and mo- analysis to funds investing in UK equities, more accurate performance
mentum risk factors. Cuthbertson, Nitzsche, and O'Sullivan (2010) pro- benchmarks may be used. This data set represents almost the entire
vide a comprehensive survey of both the theory and empirical findings set of UK equity funds which have existed at any point during the period
around mutual fund performance globally. Cuthbertson et al. (2008) January 1997–June 2009, including 672 nonsurviving funds. Funds are
specifically examine UK mutual fund performance, distinguishing skill also categorised by investment objectives: ‘Equity income’ funds (221
from luck in performance using a nonparametric bootstrap procedure funds), which aim to achieve a dividend yield greater than 110% of the
to construct a distribution of random sampling variation in performance market, ‘general equity’ funds (779), which invest in a broad range of
or luck against which a sample of actual funds' performance is compared. equity and small company funds (141), which are invested in stocks
The paper concludes that less than 2% of funds achieve a level perfor- which form the lowest 10% of the market by market capitalization.
mance beyond that which could be attributed to chance. Cuthbertson Fund returns are measured before taxes on dividends and capital
et al. (2012) apply a false discovery rate (FDR) procedure to UK mutual gains but net of management fees.
funds. This method determines the proportion of significant fund alphas Table 1 reports summary statistics of the mutual fund sample. Panel
that are not just type 1 errors or ‘false discoveries’. The authors find a A presents the number of funds in the sample by year which ranges
false discovery rate of around 30% among funds. from 447 in 2000 (total across all investment styles) to 792 in 2005.
However, the literature on mutual funds seldom accounts for liquid- The table also provides a yearly breakdown of the numbers of new
ity in estimating risk adjusted performance. Given the theoretical and funds entering the industry along with the numbers of nonsurvivors
empirical findings around the pricing of stock liquidity characteristics exiting which includes funds either closing down or merging. We see
and systematic liquidity risk, our objective here is to examine the role a particularly large number of funds exiting the industry around 1999
of both these risks in UK mutual fund performance for the first time. around the Asian and Russian financial crisis periods and again in
The paper is organised as follows: Section 2 describes our tick data 2007/8 following the more recent financial crisis period. In Panel B,
set of trades on the London Stock Exchange (LSE) as well as our mutual we present statistics describing the distribution of returns in the
fund data set. Section 3 outlines our testing methodology while in cross-section of funds over time, which we break down by fund invest-
Section 4 we describe our results. ment style. Equity income funds yield the highest average monthly re-
turn of 0.74% and the lowest standard deviation of 0.61% while at
2. Data 0.44% small company funds yield the lowest return but the highest stan-
dard deviation of 0.89% where, in results not shown, returns range from
We use two large data sets in our analysis. We obtain tick data and 6.69% to −5.14%. All fund styles exhibit sufficient variation in returns
best price data from the London Stock Exchange (LSE) information prod- which is helpful in identifying the potential impact of the various risk
ucts division.1 Our mutual fund data set is obtained from Morningstar. factors including liquidity. We return to discuss the normality charac-
The sample covers the period January 1997 to February 2009. teristics of the fund returns later and the need to calculate nonparamet-
The tick file contains all trades of which the LSE has a record. The ric bootstrap p-values in tests of statistical significance.
data for each trade includes the trade time, publication time, price at
which the trade occurs, the number of shares, the currency, the tradable
instrument code (TIC) and SEDOL of the stock, the market segment and 3. Methodology
sector through which the trade was routed as well as the trade type. The
tick data files contain 792,995,147 trades. In this section we develop factor models against which we evaluate
The best price files contain the best bid and ask prices available on mutual fund performance. Our baseline models are the Fama and
the LSE for all stocks for the same time period; this includes the tradable French (1996) three factor model and the Carhart (1997) four factor
instrument code (TIC), SEDOL, country of register, currency of trade and model with market, size, value and momentum risk factors. We
time stamp of best price. The files contain 1,956,681,874 best prices. augment these models with a liquidity factor mimicking portfolio —
In cleaning the data set some trades are excluded as follows: Trades firstly with an illiquidity characteristic risk mimicking portfolio and
outside the Mandatory Quote Period (SEAQ)/continuous auction (SETS) secondly with a systematic liquidity risk mimicking portfolio. In
are removed (i.e., only trades between 08:00:00 and 16:30:00 are each case, we measure liquidity by four alternative measures. We
included). Cancelled trades are excluded. We also exclude opening employ several alternative liquidity measures as the different
auctions as their liquidity dynamics may differ from that of continuous measures may capture different facets of liquidity. We employ quoted
auction trades. We exclude trades not in sterling. Best prices that only spread and effective spread as well the temporary fixed price impact
fill one side of the order book (e.g., where there is a best bid but no cor- measure and permanent fixed price impact measures of Sadka (2006).
responding ask price) are removed. We also remove a small number of We choose these liquidity measures as these are the measures found
trades with unrealistically large quoted spreads: for stocks with a price to have the strongest asset pricing effects in previous research on liquid-
greater than £50, spreads N10% are removed while for stocks with ity risk in UK equities (Foran et al., 2014a). We begin in this section by
prices less than £50, spreads N25% are removed. Only ordinary, auto- briefly describing our four liquidity measures.3
matic and block trades are used in this study. Following these filters,
673,421,155 trades and 594,647,452 best bid and ask prices remain.
We conduct our analysis on the historic constituents of the FTSE All 2
To control for the fact that the SEDOL numbers of certain stocks have changed multiple
Share index, i.e., we cross-reference with the London Share Price Database
times over the sample period we use the LSPD's SEDOL Master File.
3
As the liquidity measures have been previously presented in the literature (Foran
1
This data set is the same as that used in Foran et al. (2014a) which provides further et al., 2014a; Korajczyk & Sadka, 2008; Sadka, 2006) we provide only a brief description
data discussion. here.
180 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189

Table 1
Descriptive statistics of the mutual fund sample.

Panel A: The number of funds that exist at the start of each year is reported for the three investment styles. The second column under each investment objective reports the
numbers of funds that enter and exit the sample during each year.

Equity income funds General equity funds Small company funds

Year Start of year Entered/exit Start of year Entered/exit Start of year Entered/exit

1997 117 5/0 343 18/0 88 2/0

1998 122 1/0 361 38/0 90 5/0
1999 123 17/60 399 36/126 95 5/42
2000 80 9/0 309 39/0 58 11/0
2001 89 16/0 348 62/0 69 9/0
2002 105 19/0 410 54/0 78 7/0
2003 124 14/2 464 59/3 85 5/0
2004 136 5/0 520 37/0 90 4/0
2005 141 5/10 557 38/46 94 2/6
2006 136 9/7 549 34/27 90 2/3
2007 138 5/22 556 19/72 89 0/22
2008 121 0/38 503 3/182 67 1/3
2009 83 0/0 324 0/0 65 0/0

Panel B: Statistics describing the entire distribution of returns across funds are reported by investment objective. The total number of funds examined in the sample under each
objective I also reported.

Equity income General equity Small company

Mean 0.74 0.55 0.44

Standard dev. 0.61 0.67 0.89
Skewness −0.23 −1.35 0.80
Kurtosis 3.24 12.06 28.95
Max. 2.22 3.31 6.69
75th 1.01 0.94 0.63
Median 0.70 0.52 0.46
25th 0.44 0.23 0.21
Min. −1.48 −4.35 −5.14
Number 221 779 141

3.1. Liquidity measures effects where in turn each of these effects are also modelled as fixed
(independent of trade size) and variable (dependent on trade size).
3.1.1. Quoted spread The model is given by
The (average) quoted spread for stock s in month m is given as
Δpt ¼ Ψεψ;t þ λελ;t þ ΨΔDt þ λΔðDVt Þ þ yt ð3Þ
1 X Ps;t −PBs;t
qus;m A
Q s;m ¼  ð1Þ
qus;m t¼1 ms;t where Δpt is the change in price between trade t and trade t − 1. Dt is an
indicator variable equal to +1 (−1) for a buyer (seller) initiated trade.
where PAs,t is the ask price of quote t for stock s, PBs,t is the bid price of ΔDt is the change in order direction for trade t. ΔDVt is the change in
quote t for stock s, qus,m is the number of quotes in month m for stock total signed order size in trade t. εψ,t is the unexpected trade direction,
A B
s. ms,t = (Ps,t + Ps,t)/2 is the midpoint of the bid/ask prices. Higher levels ελ,t is the unexpected signed order flow. Ψs,t, λs,t, Ψs;t , and λs;t are the
of quoted spread are associated with lower levels of liquidity. permanent fixed, permanent variable, temporary fixed and temporary
variable price impact measures respectively for stock s in month t. All
3.1.2. Effective spread price impact measures are scaled by price to allow the coefficient to
We calculate the effective spread by comparing the price at which a be interpreted as the percentage impact on price rather than the abso-
trade occurs with the midpoint of the latest best bid/ask price that was lute impact. In this study we use the temporary fixed and permanent
in place at least five seconds previously. We express this as a percentage fixed price impact measures. Our liquidity measures are winsorised at
of the midpoint and as an average across all trades for stock s in month the 1% and 99% percentiles to reduce the effect of outliers (Korajczyk
m as follows: & Sadka, 2008).4

1 Xtr
s;m tr
Ps;t −ms;t−5 3.2. Constructing liquidity factors
Es;m ¼ 
trs;m t¼1 ms;t−5 ð2Þ
  3.2.1. Illiquidity characteristic mimicking portfolio
A B
ms;t−5 ¼ Ps;t−5 þ Ps;t−5 =2
Several studies such as Amihud and Mendelson (1986) and Lu
and Hwang (2007) argue that a stock's illiquidity level is priced as
A B
where Ps,t − 5 and Ps,t − 5 are the ask and bid prices in place five seconds a characteristic. In order to test this in the performance of mutual
before trade t for stock s, trs,m is the number of trades in month m for funds, we begin by constructing an illiquidity characteristic mimick-
tr
stock s. Ps,t is the price at which a trade occurs. Higher levels of effective ing portfolio for each liquidity measure as follows: each month all
spread are associated with lower levels of liquidity. stocks are sorted into decile portfolios based on their liquidity

3.1.3. Price impact mode — Sadka (2006)

Sadka (2006) suggests that trades affect prices in four ways — 4
We refer the reader to Sadka (2006) and Korajczyk and Sadka (2008) for fuller discus-
through permanent informational effects and temporary inventory sion of the price impact model.
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 181

Fig. 1. Liquidity mimicking factor portfolios. Time series plots of the illiquidity level factor and the liquidity risk factor by liquidity measure as indicated.

where decile 1 represents high liquidity stocks while decile 10 repre- mean of liquidity measure i for stock s up to time t − 1, σ
i
^ s;t is the esti-
sents low liquidity stocks. Equal weighted decile portfolio returns mated standard deviation of liquidity measure i for stock s up to
are calculated over the following one month holding period and time t − 1 and NLis,t is the normalised liquidity observation. Our liquidity
the process is repeated over a one month rolling window. The illi- measures are measures of illiquidity. In keeping with approaches in the
quidity characteristic mimicking portfolio is the difference between literature, we sign all extracted factors so as to represent liquidity. Here,
the returns of the top decile (decile 10) and bottom decile (decile factors are signed to be negatively correlated with the time series of the
1) portfolios, or illiquid minus liquid stocks. monthly cross-sectional average of the relevant measure. In order to ex-
amine the risk around market liquidity shocks rather than anticipated
changes in market liquidity, in the case of each liquidity factor we use
3.2.2. Systematic liquidity risk mimicking portfolio the residuals of an AR(2) process applied to the factor.
Korajczyk and Sadka (2008), Sadka (2006) and Foran et al. (2014a) In order to capture systematic liquidity risk in a mimicking portfolio,
all provide evidence of a premium for systematic liquidity risk. In we do the following: for each market liquidity factor, i.e., the first ex-
order to test this in mutual fund performance we need to construct a tracted principal component, pre-whitened to measure market liquidity
systematic liquidity risk mimicking portfolio. For each liquidity measure shocks, each month individual stock (excess) returns are regressed
we have a (T × n) matrix of liquidity observations where T = number of on the market liquidity factor as well as factors for market, size, value
months and n = number of stocks. In a procedure similar to Korajczyk and momentum risk. We estimate this regression over the previous
and Sadka (2008), from this matrix we extract the first principal compo- 36 months (minimum 24 month requirement for stock inclusion).
nent, which captures systematic variation or commonality in liquidity Stocks are then sorted into deciles according to their liquidity risk,
across stocks. We refer to this as a systematic liquidity risk factor. We i.e., their estimated beta (sensitivity) relative to the market liquidity
first normalise all liquidity measures before extracting the principal factor as follows:
i
Lis;t −μ^ s;t
components as follows5: NLis;t ¼ ^ is;t
where Lis,t is the liquidity obser-
σ
L O
i ri;t ¼ θi þ βi  Ft þ γi  Ft þ εi;t ð4Þ
vation of liquidity measure i for stock s at time t, μ^ s;t is the estimated

5
This is to avoid issues of scale in the different liquidity measures affecting the extract- where FLt is the relevant (pre-whitened) market liquidity factor, L = 1,
ed factors. 2 … 4. FOt is a matrix of the other risk factors, ri,t is the excess return on
182 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189

Table 2
UK mutual fund industry performance: liquidity factor augmented models.

Baseline Illiquidity level Liquidity risk Illiquidity level + Risk

CAPM FF Carhart CAPM FF Carhart CAPM FF Carhart CAPM FF Carhart

Α 0.04 −0.06 −0.14⁎⁎ 0.18⁎⁎ −0.12⁎⁎ −0.16⁎⁎⁎ 0.06 −0.06 −0.14⁎⁎ 0.17⁎ −0.12⁎⁎ −0.16⁎⁎⁎
p-Value 0.64 0.20 0.02 0.04 0.02 0.01 0.62 0.25 0.01 0.07 0.03 0.00
Illiquidity level 0.09⁎⁎⁎ −0.03⁎⁎⁎ −0.02 0.10⁎⁎⁎ −0.04⁎⁎ −0.02
p-Value 0.00 0.01 0.16 0.00 0.02 0.11
Liquidity risk −0.01 0.00 −0.02 0.01 −0.01 −0.02
p-Value 0.71 0.86 0.20 0.69 0.47 0.14
Market 0.97⁎⁎⁎ 0.94⁎⁎⁎ 0.96⁎⁎⁎ 0.93⁎⁎⁎ 0.95⁎⁎⁎ 0.96⁎⁎⁎ 0.96⁎⁎⁎ 0.94⁎⁎⁎ 0.96⁎⁎⁎ 0.94⁎⁎⁎ 0.95⁎⁎⁎ 0.96⁎⁎⁎
p-Value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Size 0.34⁎⁎⁎ 0.36⁎⁎⁎ 0.37⁎⁎⁎ 0.37⁎⁎⁎ 0.34⁎⁎⁎ 0.36⁎⁎⁎ 0.37⁎⁎⁎ 0.38⁎⁎⁎
p-Value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Value −0.02 −0.01 −0.01 −0.01 −0.02 0.00 −0.01 0.00
p-Value 0.32 0.62 0.54 0.16 0.35 0.92 0.70 0.91
Momentum 0.03⁎⁎ 0.02 0.04⁎⁎ 0.03⁎⁎
p-Value 0.03 0.16 0.01 0.05
Non-normality 0.94 0.92 0.92 0.72 0.68 0.65 0.69 0.65 0.62 0.71 0.64 0.62

Each month fund returns are averaged across funds and the resultant time series is regressed on the CAPM, Fama and French (1996) and Carhart (1997) models. Each model is then aug-
mented with the illiquidity characteristic (level) mimicking portfolio and/or the liquidity risk mimicking portfolio. The illiquidity level mimicking factor is formed by each month ranking
stocks based on average effective spread over the previous 11 months and calculating the return on a long position in the most illiquid decile of stocks and a short position in the most
liquid decile. The liquidity risk factor is formed each month by measuring the sensitivity of stock returns to an extracted market liquidity factor over the previous 36 months, sorting stocks
into deciles based on sensitivity and calculating the return on a long position in the most sensitive decile and a short position in the least sensitive decile. This table reports model alphas
and loadings as well as bootstrap p-values. ‘Illiquidity level + Risk’ denotes both liquidity factors specified simultaneously. The last row denoted “normality” presents the percentage
of funds where we reject the null hypothesis of normally distributed residuals at the 5% significance. ⁎ Indicates significance at 10%. ⁎⁎ Indicates significance at 5%. ⁎⁎⁎ Indicates significance
at 1%.

stock i and time t. Stocks are assigned to a portfolio based on β ^ , which the sample by each month forming a portfolio that is long the decile of
i
measures sensitivity to market liquidity shocks, in ascending order, the smallest stocks and short the decile of the biggest stocks based on
e.g., portfolio 1 contains low liquidity risk (low beta) stocks while port- market capitalisation and holding for one month before reforming.
folio 10 contains high liquidity risk (high beta) stocks. Each portfolio re- The value factor, high book to market minus low book to market stocks
turn is the equal weighted average return of its constituent stocks for (HML), is the return on the Morgan Stanley Capital International (MSCI)
the following month. Portfolios are reformed monthly. The liquidity UK Value Index minus the return on the MSCI UK growth index. The mo-
risk mimicking portfolio is taken to be the difference between the mentum factor (MOM) is formed by ranking stocks each month based
high minus low portfolios, i.e., 10 − 1. on performance over the previous 11 months. A factor mimicking port-
Fig. 1 shows time series charts of both the illiquidity characteristic folio is formed by going long the top performing 1/3 of stocks and taking
risk mimicking portfolio (factor) and the systematic liquidity risk mim- a short position in the worst performing 1/3 of stocks over the following
icking portfolio (factor) for each liquidity measure. Consistent with the
findings of Lu and Hwang (2007), the chart reveals that for most of the
period the illiquidity characteristic risk factor (returns on illiquid stocks Table 3
minus returns on liquid stocks) is negative indicating that illiquid stocks Liquidity factor augmented models — the Schwarz Bayesian Information Criterion.
underperformed liquid stocks. We investigate its pricing in mutual fund Baseline Illiquidity level Liquidity risk Level and risk
performance below. The systematic liquidity risk factor is generally
Quoted spread
positive, more pronounced in the early part of the sample period, indi- CAPM 1.266 1.238 1.296 2.863
cating that market liquidity sensitive stocks offered a premium. FF 1.022 1.010 1.052 1.016
Carhart 1.018 1.018 1.053 1.052
3.3. Mutual fund performance Effective spread
CAPM 1.266 1.217 1.282 2.814
FF 1.022 1.018 1.037 1.002
Having constructed risk mimicking portfolios for characteristic illi- Carhart 1.018 1.024 1.036 1.041
quidity risk and systematic liquidity risk, we first examine the exposure Temporary fixed
of UK mutual funds to these liquidity risks and then estimate the liquid- CAPM 1.266 1.218 1.291 2.824
FF 1.022 1.013 1.042 1.005
ity risk adjusted performance, alpha, of the UK mutual fund industry. In
Carhart 1.018 1.019 1.042 1.044
particular, for the first time in the UK mutual fund industry, we compare Permanent fixed
the Fama and French three factor and Carhart four factor alpha with CAPM 1.266 1.217 1.293 2.828
alpha that controls for characteristic and systematic liquidity risk. Our FF 1.022 1.007 1.045 1.002
mutual fund performance evaluation model is of the form Carhart 1.018 1.014 1.043 1.038

For each fund, returns are regressed on the CAPM, Fama and French (1996) and Carhart
ri;t ¼ αi þ βM  rm;t þ βS  SMBt þ βV  HMLt þ βMOM  MOMt þ βL  LIQ t þ εt ð5Þ (1997) models. Each model then is augmented with the illiquidity characteristic (level)
mimicking portfolio and/or the liquidity risk mimicking portfolio. This is done for each li-
quidity measure separately. The illiquidity level mimicking factor is formed by each month
where ri,t is the excess return of fund i in month t, rm,t is the excess FTSE ranking stocks based on average liquidity over the previous 11 months and calculating the
All Share return in month t, SMBt, HMLt, and MOMt are the size, value return on a long position in the most illiquid decile of stocks and a short position in the
and momentum risk mimicking portfolios or benchmark factors in most liquid decile. The liquidity risk factor is formed each month by measuring the sensi-
month t. LIQt is either the illiquidity characteristic risk or systematic li- tivity of stock returns to an extracted market liquidity factor over the previous 36 months,
sorting stocks into deciles based on sensitivity and calculating the return on a long position
quidity risk mimicking portfolio (or both may be specified in some in the most sensitive decile and a short position in the least sensitive decile. Table 3 reports
model estimations). FTSE All Share returns are used to represent market the Schwarz Information Criterion, averaged across fund regressions. The lowest values in
returns. The size risk factor, small minus big (SMB), is calculated from each group are bolded and underlined.
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 183

Table 4
Cross-sectional regressions of alpha on liquidity factor loadings.

Quoted spread Effective spread Temporary fixed Permanent fixed

Illiquidity level Liquidity risk Illiquidity level Liquidity risk Illiquidity level Liquidity risk Illiquidity level Liquidity risk

All funds
3FF −0.82 0.01 −0.56 1.15 −0.56 0.69 −0.76 1.07
(0.00) (0.93) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
4F −0.22 −0.14 −0.21 0.66 −0.18 −0.08 −0.20 0.94
(0.02) (0.36) (0.02) (0.00) (0.05) (0.55) (0.03) (0.00)

Income
3FF −0.89 −0.68 −0.53 0.86 −1.31 −1.02 −0.78 1.22
(0.00) (0.08) (0.09) (0.04) (0.00) (0.01) (0.00) (0.00)
4F −0.35 −0.65 −0.19 0.81 −0.45 −0.72 −0.07 1.73
(0.20) (0.09) (0.50) (0.04) (0.14) (0.10) (0.79) (0.00)

General equity
3FF −0.64 0.09 −0.38 0.93 −0.36 0.59 −0.54 0.95
(0.00) (0.56) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
4F −0.31 −0.09 −0.36 0.62 −0.27 −0.14 −0.31 0.78
(0.01) (0.62) (0.00) (0.00) (0.03) (0.36) (0.01) (0.00)

Small stock
3FF −1.16 −0.15 −0.74 1.73 −0.48 1.72 −1.02 1.04
0.00 0.78 0.01 0.00 0.12 0.00 0.00 0.02
4F −0.16 −0.85 −0.14 0.38 0.04 0.54 −0.07 0.76
(0.54) (0.09) (0.63) (0.37) (0.89) (0.22) (0.80) (0.06)

For each fund, returns are regressed on the (i) Fama and French (1996) three factor and (ii) Carhart (1997) four factor models and performance alphas are estimated in each
case. These two models are then augmented with the illiquidity characteristic (level) mimicking portfolio or the liquidity risk mimicking portfolio and the two liquidity
factor loadings are estimated. The illiquidity level mimicking factor is formed by each month ranking stocks based on average liquidity over the previous 11 months and
calculating the return on a long position in the most illiquid decile of stocks and a short position in the most liquid decile. The liquidity risk factor is formed each month
by measuring the sensitivity of stock returns to a market liquidity factor over the previous 36 months, sorting stocks into deciles based on sensitivity and calculating
the return on a long position in the most sensitive decile and a short position in the least sensitive decile. This is done for each liquidity measure separately. Table 4 presents
results of cross-sectional (across funds) regressions of (i) the estimated three factor alpha and (ii) the estimated four factor alpha on the estimated illiquidity level loading
and liquidity risk loading. Specifically, we report the coefficients and their p-values (in parentheses) on the illiquidity level loading and liquidity risk loading. We report
results for all funds taken together as well by investment style.

Table 5
The cross-sectional distribution of mutual fund alpha pre and post liquidity factor adjustment.

3 Factor Max Max 99% Max 95% Max 90% Max 75% Median Min 25% Min 10% Min 5% Min 1% Min

Panel A: Quoted spread

Alpha 0.97 0.72 0.46 0.29 0.09 −0.07 −0.23 −0.36 −0.45 −0.81 −2.36
t-Stat 2.29 2.17 1.31 1.66 0.58 −0.76 −1.54 −1.50 −4.08 −1.98 −4.41
Bootstrap p-value 0.04 0.10 0.28 0.12 0.60 0.50 0.15 0.20 0.00 0.05 0.00

3 Factor + Illiquidity level Kolmogorov–Smirnov p-value: 0.00

Alpha 1.27 0.79 0.40 0.28 0.06 −0.14 −0.27 −0.40 −0.49 −0.75 −1.63
t-Stat 2.07 1.32 1.66 0.74 0.27 −0.88 −2.58 −1.08 −4.32 −4.75 −3.18
Bootstrap p-value 0.12 0.30 0.12 0.49 0.84 0.44 0.02 0.27 0.00 0.00 0.00

3 Factor + Liquidity risk Kolmogorov–Smirnov p-value: 1.00

Alpha 0.96 0.72 0.46 0.28 0.09 −0.07 −0.22 −0.36 −0.47 −0.78 −1.80
t-Stat 1.91 2.31 1.33 2.20 0.33 −0.37 −1.21 −2.70 −1.90 −4.76 −3.62
Bootstrap p-value 0.08 0.05 0.32 0.05 0.75 0.67 0.25 0.01 0.07 0.00 0.00

3 Factor + Illiquidity level + Liquidity risk Kolmogorov–Smirnov p-value: 0.00

Alpha 1.28 0.81 0.40 0.28 0.07 −0.12 −0.27 −0.41 −0.50 −0.74 −1.34
t-Stat 2.12 1.31 1.68 2.33 0.52 −0.56 −2.97 −2.42 −3.57 −4.29 −2.75
Bootstrap p-value 0.12 0.32 0.11 0.01 0.60 0.58 0.01 0.02 0.00 0.00 0.02

Panel B: Effective spread

Alpha 0.97 0.72 0.46 0.29 0.09 −0.07 −0.23 −0.36 −0.45 −0.81 −2.36
t-Stat 2.29 2.17 1.31 1.66 0.58 −0.76 −1.54 −1.50 −4.08 −1.98 −4.41
Bootstrap p-value 0.04 0.10 0.28 0.12 0.60 0.50 0.15 0.20 0.00 0.05 0.00

3 Factor + Illiquidity level Kolmogorov–Smirnov p-value: 0.00

Alpha 1.45 0.87 0.43 0.29 0.06 −0.12 −0.26 −0.39 −0.50 −0.79 −1.92
t-stat 2.40 2.20 1.06 0.99 0.33 −1.48 −7.75 −0.92 −3.25 −4.78 −3.35
Bootstrap p-value 0.09 0.08 0.28 0.37 0.73 0.20 0.00 0.37 0.00 0.00 0.01

3 Factor + Liquidity risk Kolmogorov–Smirnov p-value: 0.02

Alpha 0.87 0.71 0.44 0.28 0.10 −0.04 −0.18 −0.31 −0.41 −0.71 −1.30
t-stat 5.53 2.35 1.18 1.55 0.82 −0.26 −0.84 −1.84 −2.58 −1.82 −3.82
Bootstrap p-value 0.00 0.07 0.33 0.13 0.44 0.80 0.37 0.06 0.03 0.08 0.00

(continued on next page)

184 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189

Table 5 (continued)

3 Factor Max Max 99% Max 95% Max 90% Max 75% Median Min 25% Min 10% Min 5% Min 1% Min

Panel B: Effective spread

3 Factor + Illiquidity level + Liquidity risk Kolmogorov–Smirnov p-value: 0.02
Alpha 1.54 0.85 0.43 0.30 0.08 −0.10 −0.24 −0.37 −0.45 −0.76 −1.01
t-stat 2.09 2.80 1.09 0.44 0.51 −0.47 −1.16 −3.23 −4.01 −2.08 −2.28
Bootstrap p-value 0.04 0.01 0.30 0.69 0.62 0.65 0.33 0.00 0.00 0.10 0.06

Panel C: Temporary fixed price impact

Alpha 0.97 0.72 0.46 0.29 0.09 −0.07 −0.23 −0.36 −0.45 −0.81 −2.36
t-Stat 2.29 2.17 1.31 1.66 0.58 −0.76 −1.54 −1.50 −4.08 −1.98 −4.41
Bootstrap p-value 0.04 0.10 0.28 0.12 0.60 0.50 0.15 0.20 0.00 0.05 0.00

3 Factor + Illiquidity level Kolmogorov–Smirnov p-value: 0.00

Alpha 1.54 0.86 0.42 0.26 0.03 −0.16 −0.31 −0.43 −0.52 −0.83 −2.06
t-Stat 2.41 2.47 1.58 2.09 0.13 −0.69 −1.37 −1.98 −3.07 −2.27 −3.37
Bootstrap p-value 0.08 0.03 0.11 0.04 0.89 0.57 0.18 0.07 0.00 0.01 0.00

3 Factor + Liquidity risk Kolmogorov–Smirnov p-value: 0.23

Alpha 0.92 0.70 0.46 0.28 0.10 −0.05 −0.19 −0.31 −0.43 −0.71 −1.50
t-Stat 2.86 1.44 2.14 0.62 0.76 −0.51 −1.71 −1.86 −2.78 −2.43 −3.33
Bootstrap p-value 0.01 0.25 0.03 0.57 0.43 0.66 0.14 0.07 0.02 0.00 0.01

3 Factor + Illiquidity level + Liquidity risk Kolmogorov–Smirnov p-value: 0.00

Alpha 1.60 0.87 0.42 0.27 0.05 −0.13 −0.29 −0.43 −0.51 −0.79 −1.29
t-Stat 2.00 2.53 1.76 0.68 0.29 −1.53 −2.64 −3.70 −2.65 −5.00 −2.44
Bootstrap p-value 0.05 0.05 0.13 0.54 0.83 0.19 0.02 0.00 0.02 0.00 0.04

Panel D: Permanent fixed price impact

Alpha 0.97 0.72 0.46 0.29 0.09 −0.07 −0.23 −0.36 −0.45 −0.81 −2.36
t-Stat 2.29 2.17 1.31 1.66 0.58 −0.76 −1.54 −1.50 −4.08 −1.98 −4.41
Bootstrap p-value 0.04 0.10 0.28 0.12 0.60 0.50 0.15 0.20 0.00 0.05 0.00

3 Factor + Illiquidity level Kolmogorov–Smirnov p-value: 0.00

Alpha 1.40 0.82 0.39 0.27 0.06 −0.12 −0.27 −0.40 −0.49 −0.78 −1.48
t-Stat 1.84 2.11 0.96 0.73 0.44 −0.96 −2.31 −3.72 −2.35 −5.07 −2.74
Bootstrap p-value 0.09 0.07 0.47 0.51 0.73 0.40 0.02 0.00 0.02 0.00 0.02

3 Factor + Liquidity risk Kolmogorov–Smirnov p-value: 0.34

Alpha 0.87 0.73 0.44 0.25 0.09 −0.05 −0.20 −0.32 −0.43 −0.75 −1.46
t-Stat 5.58 1.42 2.66 1.54 0.62 −0.40 −1.49 −2.27 −2.53 −4.49 −3.35
Bootstrap p-value 0.00 0.26 0.01 0.16 0.58 0.70 0.13 0.03 0.03 0.00 0.00

3 Factor + Illiquidity level + Liquidity risk Kolmogorov–Smirnov p-value: 0.00

Alpha 1.60 0.79 0.39 0.27 0.06 −0.12 −0.27 −0.39 −0.49 −0.78 −1.22
t-Stat 2.16 2.57 1.61 2.16 0.35 −1.12 −2.44 −2.02 −3.10 −5.12 −2.82
Bootstrap p-value 0.04 0.02 0.12 0.05 0.75 0.31 0.02 0.04 0.01 0.00 0.03

For each fund, returns are regressed on the Fama and French (1996) three factor models. This model is then augmented with the illiquidity characteristic (level) mimicking portfolio and/or
the liquidity risk mimicking portfolio. The illiquidity level mimicking factor is formed by each month ranking stocks based on average liquidity over the previous 11 months and calculating
the return on a long position in the most illiquid decile of stocks and a short position in the most liquid decile. The liquidity risk factor is formed each month by measuring the sensitivity of
stock returns to a market liquidity factor over the previous 36 months, sorting stocks into deciles based on sensitivity and calculating the return on a long position in the most sensitive
decile and a short position in the least sensitive decile. This table presents alpha, its t-statistic and the bootstrap p-value of alpha at various points in the cross-sectional distribution (t-stats
are Newey–West adjusted for lag order 2). Panels A to D present results for the alternative liquidity measures as indicated. The Kolmogorov–Smirnov statistic tests the significance of the
difference between the distributions of alpha from the baseline three factor model and the liquidity augmented models.

month. All portfolios are equal weighted. The risk free rate is the yield the tails of the cross-sectional performance distribution. To allow for
on 3 month sterling denominated gilts. this, we calculate and report bootstrap p-values of alpha.
In addition to the above unconditional model, several conditional
models have also appeared in the mutual fund performance literature
that allow for time varying factor loadings based on public information 4. Empirical results
(Christopherson, Ferson, & Glassman, 1998; Ferson & Schadt, 1996). We
also tested conditional models here but they were found to have no We begin our analysis by examining the performance of the UK mu-
additional explanatory power and were consistently strongly rejected tual fund industry in a portfolio of funds approach. We construct a time
by the Schwarz Bayesian Information Criterion in favour of the more series of the monthly cross-sectional (equally weighted) average fund
parsimonious unconditional model.6 This was also a robust finding in return and estimate various forms of Eq. (5). Results are presented in
Cuthbertson et al. (2008). Table 2. We begin with a baseline model, i.e., either the CAPM, Fama
We estimate various forms of Eq. (5) and examine the pricing of our and French (1996), denoted ‘FF’, or Carhart (1997) model. We then
two liquidity factors as well as their impact on alpha in the cross-section augment this baseline model with the illiquidity characteristic risk
of fund performance. We conduct separate analyses for alternative fund mimicking factor (henceforth ‘illiquidity level’ factor) or the systematic
investment styles including income funds, general equity funds and liquidity risk mimicking factor (henceforth ‘liquidity risk’ factor) or
small stock funds. We find that the majority of funds exhibit non- both. Results in Table 2 are based on the effective spread liquidity
normally distributed residuals in the estimation of Eq. (5). Cuthbertson measure.7
et al. (2008) find that this non-normality significantly alters the inter-
pretation of performance findings for many funds, particularly those in 7
In Table 2 in order to conserve space we present only the results for the effective
spread measure of liquidity. The same tests for our other liquidity measures yield qualita-
6
To conserve space we do not present these results in the paper. tively similar results, available on request.
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 185

Consistent with previous findings in the literature, our baseline When we augment the baseline models with the liquidity risk
model results (first column) indicate a statistically significant role (by factor (column 3), the initial results indicate that systematic liquid-
the bootstrap p-values) for market, size and momentum risk in ity risk does not explain mutual fund returns where the liquidity
explaining mutual fund returns but an insignificant role for value risk risk loadings in all augmented models are not statistically signifi-
(Cuthbertson et al., 2008). The last row denoted “Non-normality” cant. However, this result conceals positive and negative loadings
presents the percentage of funds where the null hypothesis of normally on the liquidity risk factor across individual funds which cancel
distributed residuals is rejected at 5% significance — the high percent- out in this portfolio of funds approach. In results not shown, the
ages motivate our use of bootstrap p-values. On average the industry number of funds with positive and negative loadings on the liquid-
yields a negative and statistically significant alpha by the Carhart four ity risk factor is approximately equal. This is a consistent finding
factor model. In column 2 when we augment the baseline models across all our liquidity measures and prompts us to carry out further
with the illiquidity level factor (illiquid stock returns minus liquid cross-sectional tests of liquidity risk pricing below. These findings
stock returns) we see that it has a negative loading in the augmented around the illiquidity level factor and the liquidity risk factor are
Fama and French model – statistically significant at the 1% significance unchanged when we augment the baseline model with both liquid-
level – indicating that on average mutual funds are tilted towards liquid ity factors at the same time (denoted ‘Illiquidity level + Risk’ in col-
stocks. In Fig. 1 previously, counter-intuitively, liquid stocks outperform umn 4), indicating that illiquidity as a stock characteristic and
illiquid stocks, or liquidity level as a stock characteristic is positively systematic liquidity risk measure distinct effects.
priced over time. There is evidence of a possible interaction between The role of an illiquidity level factor as well as a liquidity risk factor in
illiquidity level and momentum where, again in column 2, when a mutual fund performance models is further supported by the results
momentum factor is specified in the Carhart model the illiquidity presented in Table 3. Here, we report the average Schwartz Information
level factor becomes statistically insignificant. A similar pattern Criterion (SIC) model selection metric for our baseline CAPM, Fama and
can be seen in column 4 where the illiquidity level factor and liquid- French and Carhart models as well as for each baseline model augment-
ity risk factor are both added to the baseline models. We return to ed by the illiquidity level factor, liquidity risk factor and both factors
this later. specified together. We present these results for liquidity factors derived

Fig. 2. Kernel density estimate of fund alphas. For each fund, returns are regressed on the Fama and French (1996) factors and alpha is estimated. This model is then augmented
with the illiquidity characteristic (level) mimicking portfolio and/or the liquidity risk mimicking portfolio and alpha in the liquidity augmented model is estimated. Fig. 1 plots
the Kernel density estimates of the cross-sectional distributions of alphas from the three factor versus the augmented models as indicated. The charts relate to the effective
spread liquidity measure. Panels A, B and C show results for equity income, general equity and small stock funds respectively. Effective spread is used here, additional liquidity
measures available on request.
186 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189

from all four liquidity measures. In the case of all four liquidity measures In the case of the illiquidity level factor (returns on illiquid stocks
the (lowest) SIC indicates that a Fama and French three factor model minus returns on liquid stocks) we find a significant negative relation
augmented by the illiquidity level factor and/or the liquidity risk factor between the Fama and French three factor alpha and the illiquidity
is a better fit than a Carhart four factor model augmented by the liquid- level loading. This is a consistent finding across all four liquidity mea-
ity factors. The Fama and French three factor model augmented by the sures indicating, counter-intuitively, that holding more liquid stocks is
illiquidity level and liquidity risk factors is generally the most parsimo- positively priced in the cross-section of fund performance. This finding
nious best fit model of all. is robust across all fund investment styles when examined separately
To further investigate the role of liquidity exposure both as a stock except in the case of the temporary fixed priced impact measure for
characteristic and as a systematic risk factor in fund performance, we small stock funds (it is significant at the 10% significance level by the ef-
conduct cross-sectional pricing tests. For each fund, returns are regressed fective spread liquidity measure in the case of income funds). One pos-
on the (i) Fama and French (1996) three factor models and (ii) Carhart sible explanation for this counter-intuitive finding is a possible overlap
(1997) four factor models and performance alphas are estimated in between momentum (winning) stocks and liquid stocks. Hence the
each case. These two models are then augmented with the illiquidity positive pricing of liquidity may reflect momentum risk. Returning to
level factor or the liquidity risk factor and the two liquidity factor load- Table 2 there is evidence of an interaction between illiquidity level
ings are estimated in each model. This is done separately for all four li- and momentum where when a momentum factor is specified in the
quidity measures. Table 4 presents the slope coefficients and their p- Carhart model the illiquidity level factor becomes statistically insignifi-
values from cross-sectional (across funds) regressions of the estimated cant. However, our results in Table 4 for ‘All funds’ indicate that this
Fama and French three factor alpha and the Carhart four factor alpha positive pricing of illiquidity level is in fact robust to controlling for mo-
on (i) the estimated illiquidity level factor loading and (ii) the estimated mentum where we find a significant relation between the Carhart four
liquidity risk factor loading. We report results for all funds taken together factor alpha and the illiquidity level loading across all four liquidity
as well as for income funds, general equity funds and small stock funds measures. On the whole then, illiquidity level and momentum are dis-
separately. If the liquidity factors are not priced independently of the tinct effects. When we look across investment styles, however, we see
Fama and French and Carhart factors there should be no relation be- that this is only the case for general equity funds, whose large num-
tween alpha and the liquidity loadings. bers dominate the sample, but that in the case of income funds and

Fig. 2 (continued).
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 187

Fig. 2 (continued).

small stock funds the positive pricing of the illiquidity level factor in window. From Section 3.2.1 when constructing the illiquidity level
the three factor model is explained by momentum in a four factor mimicking portfolio we rebalance the portfolio monthly. In robustness
model. tests here we rebalance it annually. Also, from Section 3.2.2 when con-
From Table 2 previously initial results indicated that systematic li- structing the systematic liquidity risk mimicking portfolio, each month
quidity risk does not explain mutual fund returns where the liquidity individual stock (excess) returns are regressed on the market liquidity
risk loadings were not statistically significant but this concealed positive factor as well as factors for market, size, value and momentum risk.
and negative loadings across individual funds which cancelled out in the We estimate this regression over the previous 36 months. In robustness
portfolio of funds approach. In Table 4, our cross-sectional tests again tests here, we also examine (i) a backward looking window of 24 months
examine this further. In the case of the liquidity risk factor for all (instead of 36 months) and (ii) a holding period of 12 months (instead of
funds we find a significant positive relation between the Fama and 1 month). While we do not tabulate these voluminous results, we can re-
French three factor alpha and the liquidity risk loading, with the ex- port that none of these robustness tests change the overall conclusions
ception of the quoted spread liquidity measure where the relation is presented in Table 4. These results are available on request.
insignificant. This finding is consistent across all investment styles Our results provide evidence that both liquidity (rather than il-
and indicates that systematic liquidity risk is positively priced. liquidity) as a stock characteristic and systematic liquidity risk are
Funds which are tilted towards high (low) liquidity risk stocks positively priced in the cross-section of fund performance. We ex-
have higher (lower) Fama and French three factor alphas. On con- amine the impact on mutual fund performance alphas of adjusting
trolling for momentum in the cross-sectional regressions of the for liquidity exposure both as a stock characteristic and as a system-
Carhart four factor alpha on the liquidity risk loadings, the results atic risk factor. Table 5 reports fund alpha at various points in the
are somewhat more mixed but generally continue to support the cross-sectional distribution pre- and post-adjusting for our illiquid-
positive pricing of liquidity risk particularly in the case of the effec- ity level factor and liquidity risk factor. Notwithstanding a possible
tive spread and permanent fixed price impact liquidity measures interaction between illiquidity level and momentum for income
though not in the case of the temporary fixed price impact measure funds and small stock funds, the Schwartz Information Criterion
and less so in the case of quoted spread. values in Table 3 consistently point to the Fama and French three
In order to test the robustness of our findings in Table 4, we repeat factor models augmented with the illiquidity level and liquidity
the analysis for all funds while varying the lengths of the backward risk factors as the most parsimonious best fit model. Hence these
looking ranking time window and forward looking holding period are the models we focus on here in Table 5.
188 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189

Table 5 first shows the baseline three factor alpha and its (Newey– negative loading on the illiquidity level factor, i.e., are tilted towards liq-
West) adjusted t-statistic at various points in the cross-sectional distri- uid stocks while small stock funds have a positive loading on the illi-
bution, e.g., ‘Max’ denotes the highest alpha, ‘max 99%’ is the alpha at quidity level factor, i.e., are tilted towards illiquid stocks. Graphically,
the 99th percentile etc. Owing to a significant degree of non-normality by eye the impact of adjusting for the liquidity risk factor, the lower
in the fund regression residuals we also report nonparametric bootstrap right graphs in Panels A, B and C, show slight shifts to the right in the
p-values to test the statistical significance of alpha. The table then shows case of general equity funds and small stock funds while in the case of
the alpha, t-statistic of alpha and bootstrap p-value of the corresponding income funds the right tail shifts slightly leftward. However, these shifts
fund from the same baseline model augmented with the illiquidity level are not as pronounced as in the case of adjusting for the illiquidity level
factor, liquidity risk factor and both factors as indicated. Panels A, B, C factor and by the Kolmogorov–Smirnov statistic the shift is only signif-
and D present results for the quoted spread, effective spread, temporary icant for general equity funds at the 5% significance.
fixed price impact and permanent fixed price impact liquidity measures Overall, our results reveal a strong role for liquidity as a stock char-
respectively. For example, by the quoted spread measure of liquidity in acteristic in UK mutual fund performance evaluation. Unexpectedly,
Panel A, the median Fama and French three factor alpha is −0.07% per we find that exposure to liquid stocks is positively priced in the cross-
month but falls to −0.14% per month after adjusting for the illiquidity section of fund performance. While, a priori, a possible interaction be-
level factor. Scanning the data in Table 5 generally indicates that adjusting tween liquidity and momentum in stocks may explain why liquidity,
for illiquidity level causes an increase in alpha at both the extreme high rather than illiquidity, is positively priced in fund performance, our
and low ends of the distribution while in the middle of the distribution, cross-sectional tests show that for the fund sample as a whole liquidity
performance disimproves (around the median and Min 25% areas). and momentum represent distinct effects. However, in the smaller
Adjusting for liquidity risk generally points to no notable change in number of income funds and small stock funds examined separately
alpha. More formally, however, we also report the Kolmogorov–Smirnov the positive pricing of the illiquidity level factor in the three factor
statistic in each case to test the significance of the difference between the model is explained away by momentum in a four factor model. Overall,
distributions of alpha from the baseline three factor model and the liquid- the Schwartz Information Criterion robustly points to a Fama and
ity factor augmented models. The null hypothesis that the cross-sectional French three factor model augmented by the illiquidity level factor
distributions of alpha pre- and post-liquidity factor adjustment are from and the liquidity risk factor as the most parsimonious model of best
the same population distribution is firmly rejected in the case of illiquid- fit. Exposure to the systematic liquidity risk factor varies from positive
ity level for all four liquidity measures. However, in the case of the liquid- to negative across funds. However, the cross-sectional tests generally
ity risk factor we fail to reject this hypothesis at the 5% significance in find that systematic liquidity risk is positively priced in the cross-
the case of all liquidity measures, except the effective spread measure section of fund performance although the robustness of this finding is
(Panel B).8 weakened somewhat on controlling for momentum.
While the Schwartz Information Criterion values in Table 3 indicate Amihud and Mendelson (1986) argue that illiquid stocks should
that the Fama and French three factor models augmented with the illi- earn a premium over liquid stocks to compensate investors for the
quidity level and liquidity risk factors is the most parsimonious best fit costs incurred by illiquidity. Our findings are at variance with this ex-
model, in order to test the robustness of our findings in Table 5, we re- pectation where liquidity, rather than illiquidity, as a stock characteris-
peat the analysis presented therein for the Carhart four factor alpha in- tic earns a premium for UK equity mutual funds. This unusual finding in
stead of the Fama and French three factor alpha. The result (not shown) the UK market is consistent with past findings (Lu & Hwang, 2007;
is that the Kolmogorov–Smirnov statistics in Table 3 prove strongly ro- Foran et al., 2014b). An obvious question is what risk factors are respon-
bust: adjusting the Carhart model for the illiquidity level factor leads to sible for this liquidity premium? Lu and Hwang (2007) ask whether
a statistically significant shift in the cross-sectional distribution of alpha there is any connection between liquidity and (market) beta. They re-
(for the group of all funds) for all liquidity measures though, as in port that the beta of the most liquid (illiquid) decile portfolio is 1.36
Table 5, this is not the case for the systematic liquidity risk factor. (0.90) and the Wald test highly rejects the equality of these two betas.
Fig. 2 presents graphical illustrations of the impact on the cross- However, Foran et al. (2014b) find that while cross-sectional differences
sectional distribution of Fama and French three factor alphas of in returns exist across portfolios sorted by liquidity level, these are
adjusting for the illiquidity level and liquidity risk factors. We present strongly robust to market, size, value and momentum risks. Our findings
Kernel density estimates of the cross-sectional distributions of alpha here also indicate that the liquidity premium earned by most general
pre- and post-liquidity factor adjustment. To conserve space Fig. 2 pre- equity mutual funds is also robust to these risk factors but for income
sents results for the effective spread liquidity measure, which are repre- funds and small stock funds momentum appears to explain the liquidity
sentative of all measures (other results are available on request). Panels premium. Lu and Hwang (2007) ask whether liquidity is a systematic
A, B and C relate to income funds, general equity funds and small com- risk and (by inference) whether this might explain the observed premi-
pany funds respectively. In each panel the upper graph shows the um. Foran et al. (2014a) find a strong commonality in liquidity and re-
change in the three factor alpha after adjusting for both the illiquidity port a premium to stocks which exhibit high systematic liquidity risk
level and liquidity risk factors together, the lower left graph shows the but also report that controlling for liquidity level as a stock characteristic
change in alpha after adjusting for the illiquidity level factor while the does not alter that conclusion. Our results here support this finding
lower right graph shows the change in alpha after adjusting for the li- where our discussion around the results in Table 2 indicates that illi-
quidity risk factor. In the case of income funds and general equity quidity as a stock characteristic and systematic liquidity risk measure
funds, after adjusting for (i) the illiquidity level factor and (ii) the illi- distinct effects. In short, the unexpected finding that liquidity, rather
quidity level and liquidity risk factors together, the cross-sectional dis- than illiquidity, offers a return premium is consistent with past research
tribution of alpha clearly shifts to the left while in the case of small on the UK equity market, is robust to other commonly tested risk factors
stock funds the distribution clearly shifts to the right. In results not and is distinct from systematic liquidity risk. It remains a puzzle which
shown, the Kolmogorov–Smirnov tests conducted separately for each warrants further investigation. One possible avenue of investigation is
fund style indicate that these shifts are statistically significant. These re- that of Dong, Feng, and Sadka (2013). Although examining systematic
sults are consistent with income and general equity funds having a liquidity risk, rather than liquidity as a stock characteristic, the authors
show that fund liquidity–risk exposures provide valuable information
8
about future performance. However, the authors then show that only
Again, this is likely to be because the positive and negative loadings on liquidity risk
across funds causes alpha to decrease and increase respectively but in aggregate the distri-
a small portion of the liquidity–risk-exposure premium is explained
bution is unchanged from that of the Fama and French three factor alpha according to the by the liquidity–beta premium of funds' underlying assets. The
Kolmogorov–Smirnoff test. remainder is most likely explained by the fund manager's ability to
 J. Foran, N. O'Sullivan / International Review of Financial Analysis 35 (2014) 178–189 189

generate abnormal performance. We leave a similar analysis of the UK Chen, J. (2005). Pervasive liquidity risk in asset pricing. Working paper. Columbia
University.
market to future research. Chordia, T., Roll, R., & Subrahmanyam, A. (2000). Commonality in liquidity. Journal of
Financial Economics, 56, 3–28.
5. Conclusion Christopherson, J., Ferson, E., & Glassman, D. (1998). Conditioning manager alphas on eco-
nomic information: Another look at the persistence of performance. Review of
Financial Studies, 11, 111–142.
We find that in the UK mutual fund industry income funds and gen- Cuthbertson, K., Nitzsche, D., & O'Sullivan, N. (2010). Mutual fund performance: Measure-
eral equity funds are tilted towards liquid stocks while small stock funds ment and evidence. Journal of Financial Markets, Instruments and Institutions, 19(2),
95–187.
are tilted towards illiquid stocks. However, counter-intuitively, liquidity, Cuthbertson, K., Nitzsche, D., & O'Sullivan, N. (2012). False discoveries in UK mutual fund
rather than illiquidity, as a stock characteristic is positively priced in the performance. European Financial Management, 18(3), 444–463.
cross-section of fund performance. On controlling for stock holdings' li- Cuthbertson, K., Nitzsche, D., & O'Sullivan, N. (2008). UK mutual fund performance: Skill
or luck? Journal of Empirical Finance, 15, 613–634.
quidity, there is a statistically significant shift leftward (reduction) in
Dong, X., Feng, S., & Sadka, R. (2013). Does liquidity beta predict mutual-fund alpha.
the cross-sectional distribution of the Fama and French three factor al- Working paper (available at SSRN: http://ssrn.com/abstract=1785561).
phas. This is a robust finding across all fund investment styles examined Fama, E., & French, K. (1996). Multifactor explanations of asset pricing anomalies. Journal
and is also robust across alternative liquidity measures. This finding is of Finance, 51, 55–84.
Ferson, W., & Schadt, R. (1996). Measuring fund strategy and performance in changing
also robust to a momentum factor for the majority of our sample of economic conditions. Journal of Finance, 52, 425–462.
funds (general equity funds) but for the smaller set of income funds Fletcher, J. (1997). An examination of UK unit trust performance within the arbitrage
and small stock funds, there is evidence that momentum largely explains pricing framework. Review of Quantitative Finance and Accounting, 8, 91–107.
Foran, J., Hutchinson, M., & O'Sullivan, N. (2014a). The asset pricing effects of UK market
the pricing of this liquidity risk. Exposure to systematic liquidity risk liquidity shocks: Evidence from tick data. International Review of Financial Analysis, 32,
varies from positive to negative across funds. However, the cross- 85–94.
sectional tests generally find that systematic liquidity risk is positively Foran, J., Hutchinson, M., & O'Sullivan, N. (2014b). Liquidity commonality and pricing in
UK equities. Working paper. Centre for Investment Research, University College
priced in the cross-section of fund performance, although the robustness Cork (available at www.ucc.ie/en/cir).
of this finding is weakened somewhat on controlling for momentum ef- Hasbrouck, J., & Seppi, D. (2001). Common factors in prices, order flows and liquidity.
fects. The Schwartz Information criterion indicates that a Fama and Journal of Financial Economics, 59, 383–411.
Huang, R., & Stoll, H. (2001). Tick-size, bid–ask spreads and market structure. Journal of
French (1996) three factor performance model augmented by factors Financial and Quantitative Analysis, 36, 503–522.
for illiquidity as a stock characteristic (illiquidity level) and systematic li- Korajczyk, R., & Sadka, R. (2008). Pricing the commonality across alternative measures of
quidity risk is the most parsimonious best fit model. Overall, our results liquidity. Journal of Financial Economics, 87, 45–72.
Lu, C., & Hwang, S. (2007). Cross-sectional stock returns in the UK market: The role of li-
reveal a strong role for liquidity as a stock characteristic and systematic
quidity risk. Forecasting expected returns. Butterworth-Heinemann.
liquidity risk in fund performance evaluation models. Otten, R., & Reijnders, M. (2012). The performance of small cap mutual funds: Evidence
for the UK. available at. http://ssrn.com/abstract=2056932
References Pastor, L., & Stambaugh, R. (2003). Liquidity risk and expected stock returns. Journal of
Political Economy, 111, 642–685.
Acharya, V., & Pedersen, L. (2005). Asset pricing with liquidity risk. Journal of Financial Quigley, G., & Sinquefield, R. (1999). The performance of UK equity unit trusts. Report by
Economics, 77, 375–410. Dimensional Fund Advisors for Institute for Fiduciary Education.
Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid–ask spread. Journal of Sadka, R. (2006). Momentum and post-earnings announcement drift anomalies: The role
Financial Economics, 17, 223–249. of liquidity risk. Journal of Financial Economics, 80, 309–349.
Carhart, M. (1997). On persistence in mutual fund performance. Journal of Finance, 52,
57–82.