DOES THE FINANCIAL DISTRESS FACTOR DRIVE THE

MOMENTUM ANOMALY?

Vineet Agarwal

and

Richard Taffler*

Cranfield School of Management

Version 9.2: February 28, 2005

Paper presented in the Finance Faculty Seminar Series at City University

(Cass) Business School, London
Wednesday 9th February 2005

*Corresponding author:
Professor Richard Taffler
Cranfield School of Management
Cranfield
Bedford MK43 0AL
U.K.

Tel: 00 44 (0) 1234 754543

Fax: 00 44 (0) 1234 752554
E-mail: R.J.Taffler@cranfield.ac.uk
 DOES THE FINANCIAL DISTRESS FACTOR DRIVE THE
MOMENTUM ANOMALY?

Abstract

This paper brings together the evidence on two asset pricing anomalies –
continuation of prior returns (momentum) and the market pricing of distressed firms.
Our empirical analysis demonstrates both these effects are driven by market
underreaction to bad news, and that momentum is largely subsumed by our distress risk
factor. We also extend the extant literature on the market pricing of distress risk by
considering this conditional on market state and GDP growth rate, and find little
evidence that financial distress risk is a priced risk factor.
 DOES THE FINANCIAL DISTRESS FACTOR DRIVE THE

MOMENTUM ANOMALY?

The existence of medium term continuation of stock returns (momentum) is widely accepted

in both the academic literature and in the professional investment community. There are several

competing explanations for this most challenging of all anomalies (Fama (1998)). The first is

risk based (e.g., Chan (1988), Ball and Kothari (1989), Zarowin (1990), Conrad and Kaul (1998)

and Chordia and Shivakumar (2002). Jegadeesh and Titman (1993, 2001) and other studies (e.g.,

Daniel and Titman (1999) and Hong, Lim and Stein (2000)) argue momentum is driven by

underreaction to information. Lesmond, Schill and Zhou (2004) provide a market microstructure

argument that apparent momentum profits are explained by small firm transaction costs. The

usual criticism of data mining also applies (Lo and MacKinlay (1990)).

Several behavioral models based on inherent biases in the way investors process information

have also been proposed to explain medium term continuation of returns. In Barberis, Shleifer

and Vishny (1998), investors are slow to update their beliefs in response to new public

information leading to underreaction, and this generates positive autocorrelation in stock returns.

In Hong and Stein (1999), not all investors receive the same information at the same time and, in

addition, investors are unable to extract fully other investors’ private information from market

prices. As news slowly diffuses among the wider investing public, stock prices also adjust slowly

leading to underreaction to new information and positive autocorrelations. The empirical results

of Hong, Lim and Stein (2000) provide evidence supportive of this. Daniel and Titman (1999)

find a significant momentum effect for growth stocks and no effect for value stocks. They

1
explain this in terms of overconfidence - growth stocks are more difficult to value and we are

more overconfident when faced with more demanding tasks. The behavioral literature thus

suggests that firms where there is likely to be greater information asymmetry (e.g., those not

followed by many analysts or difficult to value) should exhibit a greater momentum effect.

One such category is that of financially distressed firms. Judging the solvency position of a

firm is not any easy task for an investor. As such, there is greater likelihood of heterogeneous

beliefs with underlying information diffusing only slowly in the market following the trades of

more informed investors. Thus, following the gradual information diffusion model of Hong and

Stein (1999), distressed firms would have low prior returns with the market only slowly

adjusting to their true financial position, and such low returns would continue for some time into

the future. Griffin and Lemmon (2002) find that high Ohlson (1980) O-score (high bankruptcy

risk) firms are smaller firms with low analyst coverage and weak current earnings. They argue

that such firms suffer from a high degree of information asymmetry and, as such, are mispriced.

By this argument, we would expect the momentum effect to be stronger for distressed firms

(which are generally smaller and have lower analyst coverage) than for non-distressed firms.

The existing literature has attempted to explain returns on small size stocks and high B/M

stocks by linking size and B/M effects to financial distress (see for instance Chan and Chen

(1991) and Fama and French (1992)).1 However, Dichev (1998) finds firms with high probability

of bankruptcy on average underperform low risk firms by 1.2% per month over the period 1980-

95. He concludes such evidence is hard to reconcile with the pricing of risk in efficient markets

and mispricing is a more likely explanation for such anomalous results. Similarly, Lamont, Polk

and San-Requejo (2001) use the Kaplan and Zingales (1997) financial constraints index and find

1
Following Fama and French (1993, 1995) we define the term distress factor as representing individual firm
financial distress. As such, we use the terms financial distress and bankruptcy risk interchangeably.

2
that even though financially constrained firms have characteristics associated with higher returns

(high leverage, high B/M, high prior year returns), they earn lower returns than non-constrained

firms. Though their index does not directly measure financial distress, financially constrained

firms are more likely to face financial distress than non-constrained firms. Ferguson and

Shockley (2003) also find distressed firms earn lower returns. However, Vassalou and Xing

(2004) find contrary results using a problematic option-based model for assessing probability of

default (see Bharath and Shumway, 2004).

Whereas there is an extensive literature on momentum and a developing literature on the

distress factor, we are not aware of any study that has directly studied the important question of a

potential link between the two issues. However, there are several studies with highly suggestive

evidence. Beaver (1968) first demonstrated subsequently bankrupt firms underperform the

market for up to four years prior to bankruptcy, and particularly during the last year, when they

underperform by 32%. This suggests that the market is anticipating, but not fully incorporating,

the deteriorating financial health of a firm and distressed firms, therefore, experience lower past

realized returns. Hong, Lim and Stein (2000) and Lesmond, Schill and Zhou (2004) find most of

momentum profits come from the returns continuation of poor performers. If the market is

underreacting to the poor solvency position of firms, we would expect financially distressed

firms to be poor past performers and be driving the momentum effect.

Given the high risk of failure of such financially distressed firms, we may expect differential

market pricing effects in up and down markets. There is a substantial body of literature showing

factor risk premia are not stationary but vary over time (e.g., Jagannathan and Wang (1996), Ang

and Chen (2002)). Ang and Chen demonstrate asymmetric correlations in good and bad states of

the world. Specifically, they find correlations between portfolio returns and market returns are

3
much higher in downmarkets than they are in upmarkets both for smaller stocks and for high

B/M stocks thereby exposing investors to greater losses in bad states of the world. Similarly,

Zhang and Petkova (2004) find conditional betas of smaller stocks and high B/M stocks co-vary

positively with the expected risk premium. Bhardwaj and Brooks (1993), Ahmed and Lockwood

(1998) and Howton and Peterson (1998), among others, find that there are significant differences

in systematic risks in bull and bear markets. We explicitly consider the potential impact of such

factors on our analysis.

This paper explores directly the potential relationship between momentum and distress risk in

the London market with bankruptcy risk measured by a widely-used UK-based z-score model

(Taffler (1983, 1984) akin to Altman (1968)). We also extend our analysis to consider the time

varying nature of distress risk premia and momentum returns. Our main results are that (i) the

momentum effect in stock returns can be explained by distress risk, (ii) this relationship is

particularly strong in bear market conditions, and (iii) in contrast to the arguments of Fama and

French (1993, 1995), among others, and consistent with Dichev (1998), there is no evidence to

suggest size and B/M are capturing bankruptcy risk.

The paper is organized as follows: section I describes our data and methodology, section II

describes our results assuming constant risk premia, section III describes our results under

different states of the world and section IV provides robustness checks. Concluding section V

provides a summary and discussion of our findings.

4
I. Data and Methodology

A. Sample Selection

This study covers all non-finance industry UK companies listed on the London Stock

Exchange (LSE) at any time during the period 1979-2002. 2 If a company changes industry or

exchange of listing, it enters the respective portfolio only after it has been listed on the (main)

London Stock Exchange and/or is classified as non-financial for twenty-four months. If the

exchange and/or industry change during the holding period, returns after the change are deleted.

To be included in the sample, securities are required to meet three additional conditions:

1. they should have positive book value because interpretation of negative book-to-market

ratios is problematic. This does not impose any significant bias till 1990 as the number of

negative book value firms until then is small (between 1 and 14 a year). However, during

the 1990s, the number of such firms increases ranging from 28 to 53 a year. Almost all

negative book value firms have bankrupt z-scores.

2. they should have been listed for at least twenty-four months before the portfolio

formation date due to the data requirement for beta estimation. This constraint also

ensures that only post-listing accounting information is used; and

2
A security that belongs to any of the following categories in any month is excluded from the population for that
month: secondary stocks of existing firms, foreign stocks, or firms traded on the Unlisted Securities Market,
Alternative Investment Market (AIM), third market, or over-the-counter. Additionally, a firm that is classified
under Financials or Mining Finance by the London Stock Exchange during any month is also excluded for that
month.

5
3. they should have been traded in at least nine of the twelve months subsequent to portfolio

formation to circumvent any potential thin trading problem. This rule does not apply to

firms that do not survive the holding period.3

The last month return for firms that enter into bankruptcy (administration, receivership or

creditors’ voluntary liquidation etc.) is set to –100% in all but one case.4 To ensure the required

accounting information is available at the time of portfolio formation, a five-month lag between

the fiscal year-end date and the reporting date is assumed. So, for the portfolio formed on 30th

September,5 book value of equity and z-score are derived from the latest available financial

statements with fiscal year-end on or before April 30th. The final sample consists of 2,459 firms

and a total of 22,774 firm years. The yearly number of stocks in the sample ranges from a

minimum of 810 in 1992 to a maximum of 1,258 in 1981.

3
The number of firms excluded is high in the first two years of our sample period and thereafter ranges between
11 and 42 a year. However, the number of financially distressed firms excluded on this criterion is not
disproportionately high.
4
The UK bankruptcy regime differs significantly to Chapter XI in the US and it is very rare indeed for
stockholders to receive any terminal distribution. In fact there was only one case in our sample period, Railtrack,
in which equity holders were promised (by the government) any payout after all creditor claims were met.
5
We choose September 30th rather than June 30th as the portfolio formation date because unlike in the US, in the
UK year-ends are more diffuse. While 37% of the companies in our sample have December year-ends, about the
same number of companies have year-ends between January and April with approximately 22% of the
companies having March year-ends.

6
B. Data

We use z-score as a proxy for distress risk. Following Altman (1968), the z-score of a

firm is derived as a weighted sum of a set of pre-defined accounting ratios. Firms with z-scores

above a pre-determined cut-off rarely fail while the incidence of failure is high in firms with z-

scores below this cut-off. The UK-based z-score model of Taffler (1983, 1984) employed in this

study is derived in a similar way to Altman (1968) using a discriminant modeling approach.

Using this bankruptcy model, a firm with a computed z<0 has a financial profile more similar to

previous bankrupt firms and thus itself is at a risk of bankruptcy, whereas z>0 indicates a firm

not at such risk. The model was developed in 1977, hence derived z-scores are completely out-

of-sample. The proportion of firms at high risk of bankruptcy changes over time from a

minimum of 14% in 1979 to a maximum of 37% in 2001. Figure 1 plots the time-series

proportion of negative z-score stocks in the sample.

Figure 1 here

The accounting data required for z-score and B/M ratio computations is primarily collected

from the Thomson Financial Company Analysis, EXSTAT, MICROEXSTAT and

DATASTREAM databases in that order. For a small number of remaining cases the data is hand

collected from the actual annual reports. This procedure enables us to have complete coverage of

all eligible firms and, as such, the study is free of survivorship bias.

Monthly stock returns, exchange of listing and firm stock exchange industrial

classifications are collected from the London Business School London Share Price Database

7
(LSPD).6 The risk free rates (1-month and 3-month Treasury bill (T-Bill) rates), dividend yield

on the FTSE All Share index and 3-month LIBOR are collected from DATASTREAM. GDP

growth rates and national income are downloaded from the Office of National Statistics website

(www.statistics.gov.uk) and the government spot rates are from the Bank of England website

(www.BankofEngland.co.uk).

The list of firm failures is compiled from LSPD (codes 7, 16 and 20), the Stock Exchange

Official Yearbook, published by Waterlow Specialist Information Publishing, and CGT Capital

Losses published by FT Interactive.

C. Methodology

To study the link between momentum and bankruptcy risk we need to control for the

potential impact of size and book-to-market on the pricing of distressed firms. Therefore, we

form twenty-four portfolios as follows:7

At the end of September of each year we first rank firms on their market capitalization

and group them into two portfolios using the median as the break point. The stocks are then

independently ranked on B/M and grouped into three portfolios – one with the lowest 30%, one

with the middle 40% and one with the highest 30% B/M ratios. Securities are then separately

ranked on momentum and grouped into two portfolios using the median as the break point and,

finally, the stocks are independently ranked on z-score and grouped into two portfolios – one

with negative z-score stocks and the other with positive z-score stocks. Twenty-four size, B/M,

6
LSPD provides monthly returns as natural logarithms of returns adjusted for capital changes and dividends. The
returns are converted to simple arithmetic returns using the transformation: exp(ln(Rt))-1.
7
We do not form portfolios by ranking the stocks on z-scores, as with Dichev (1998), because z-score is, in effect,
a binary pattern recognition device that classifies firms into two categories: those with high risk of bankruptcy
and those with low (negligible) risk. The actual value of the z-score, per se, does not convey much information
about the financial state of a firm and the only meaningful distinction is between those with z-scores below and
above the cut-off point (for the model used here, 0).

8
momentum and z-score portfolios are then formed at the intersections of the two market

capitalization, three B/M portfolios, two momentum and two z-score portfolios.8

To test whether momentum is capturing bankruptcy risk, we employ Fama-MacBeth

(1973) cross-section methodology. Only publicly available information is used. Following Fama

and French (1992) and Jegadeesh and Titman (1993), our analysis focuses on a three-factor

model augmented by z-score and momentum. The following Fama-MacBeth (1973) cross-

section regressions are run each month from October 1979 to September 2002 where the binary

variable z(0/1) = 0 if z-score is >0, and = 1 if z-score≤ 0:9

Rit - RFt = αit + γ1t βit-1 + γ2t ln(sizeit-1) + γ3t ln(B/Mit-1) + γ4t Momit-1 + γ5t z(0/1)it-1 + εit (1)

Where:

Rit is the equally-weighted return on portfolio i during month t

RFt is the 1-month Treasury bill rate at the beginning of month t

βit-1 is the beta of portfolio i estimated at the portfolio formation date10

ln(sizeit-1) is the natural logarithm of average market capitalization of common equity of stocks

in portfolio i at the portfolio formation date

ln(B/Mit-1) is the natural logarithm of the average of B/M ratios of stocks in portfolio i at the

8
Fama (1998) points out that the results of many return predictability studies are sensitive to the trading rules
employed. To test the robustness of our results we repeat all our analyses using an alternative portfolio
formation method which avoids potential data-snooping bias (Lo and MacKinlay (1990)). Twenty-four size,
B/M and z-score portfolios are formed at the intersections of the independently sorted four market
capitalization, three B/M and two z-score portfolios. Results are essentially identical to our main findings and
thus not reported here to save space.
9
Throughout this paper, subscript ‘t’ represents time of portfolio formation and subscript ‘t-1’ shows that the
information is available at the time of portfolio formation.
10
We estimate portfolio beta by regressing monthly excess returns over the previous twenty-four months (before
portfolio formation) on each portfolio against monthly excess returns on an equally-weighted market index. We
use Dimson’s (1979) method with one lead and one lag to reduce problems of thin trading.

9
portfolio formation date. The B/M ratio of each stock is computed as book value of equity

(excluding preference capital) plus deferred taxes less minority interests divided by the market

capitalization at the time of portfolio formation. To avoid undue influence of outliers on the

regressions, the smallest and largest 1% of observations are set equal to 0.01 and 0.99 fractiles

respectively.

Momit-1 is the average monthly raw return over the eleven months from October year t-1 to

August year t for all the stocks in portfolio i

z(0/1)it-1 = 1 if the latest available z-score is negative, 0 otherwise, and

εit is a mean-zero stochastic error term.

II. Results with constant risk premia

In this section, we first show z-score predicts bankruptcy and then present preliminary

evidence on the relationship between prior-year returns, z-scores, size, B/M and failure rates. We

also provide some summary statistics before reviewing the results of our analysis relating to

whether momentum is capturing distress risk.

A. Is z-score a valid measure of bankruptcy risk?

In our sample, the mortality rate (delisting for any reason) is much higher in firms with

negative z-scores than firms with positive z-scores. Approximately 9.4% of all negative z-score

firms are delisted within the next twelve months while the mortality rate for positive z-score

firms is almost half at 5.1%. The difference in proportions is highly significant (z = 11.5). Also,

out of 200 actual bankruptcies, only 9 firms are misclassified as solvent by their z-scores derived

on the basis of last available annual accounts. Our sample comprises of 5,466 firm years with

10
negative z-scores and 17,308 firm years with positive z-scores. The conditional probability of

failure given a negative z-score is 3.49% and is significantly different to the base failure rate of

0.88% (z = 20.7). Similarly, the conditional probability of non-failure given a positive z-score is

99.95% and is significantly different to the base rate of 99.12% (z = 11.7).11 As such, our z-score

variable constitutes a valid ex-ante measure of corporate bankruptcy risk.

B. Momentum, z-scores and failure rates

To unearth the potential relation between momentum, size, B/M and distress factor, we form the

following additional portfolios using two-way sorts:

(1) We rank firms on z-score and group them into two portfolios – one with negative z-score

stocks and the other with positive z-score stocks. Securities are then independently ranked

on their prior 11-month returns, i.e. from October year t-1 to August year t, and grouped into

ten portfolios with approximately equal numbers of securities. Twenty portfolios are then

formed at the intersections of z-score and momentum.

(2) We rank firms on z-score and group them into two portfolios – one with negative z-score

stocks and the other with positive z-score stocks. Securities are then independently ranked

on their market capitalization on the 30th September of each year and grouped into ten

portfolios with approximately equal numbers of securities. Twenty portfolios are then

formed at the intersections of z-score and market capitalization.

11
Including negative B/M stocks, our sample contains 214 cases firm bankruptcy of which, again, only 9 have z>0
on the basis of their last accounts. Our firm population now comprises of 5,896 firm years with negative z-scores
and 17,363 firm years with positive z-scores. The conditional probability of failure given a negative z-score is
3.48%, and is significantly different to the base failure rate of 0.92% (z = 20.6). Similarly, the conditional
probability of non-failure given a positive z-score is 99.95%, and is significantly different to the base rate of
99.14% (z = 11.5).

11
(3) We rank firms on z-score and group them into two portfolios – one with negative z-score

stocks, and the other with positive z-score stocks. Securities are then independently ranked

on their B/M ratio on the 30th September of each year and grouped into ten portfolios with

approximately equal numbers of securities. Twenty portfolios are then formed at the

intersections of z-score and B/M.

Table 1 presents the portfolio-wise distribution of the 200 stocks that failed. Panel A

shows that more than half of the failures are in the two lowest momentum deciles but there is no

obvious monotonic relation between momentum and failure rate. Panel B, similarly, clearly

shows that two-thirds of the failures in our sample are in the two smallest size deciles, and the

failure rate drops with increasing firm size. Finally, Panel C shows that approximately half of the

failed firms are in the two highest B/M deciles. However, in contrast to size, there is no

monotonic relationship between B/M and failure rates.

The table provides some preliminary evidence that low prior-year return firms, smaller

firms and high B/M firms are more likely to fail and, therefore, have higher financial distress

risk. However, it also raises the prospect of there being some interaction between these variables.

Table 1 here

Table 2 presents the distribution of failures and failure rates for our twenty-four

portfolios formed on size, B/M, momentum, and z-score. It shows that for negative z-score

stocks, controlling for size and B/M, low momentum stocks are more likely to fail than high

momentum stocks while controlling for size and momentum, high B/M stocks are more likely to

fail than low B/M stocks only for small size, low momentum portfolios. Interestingly, 43% of the

12
failures are negative z-score, low momentum small stocks with high B/M ratios, while such

stocks constitute just 5.3% of our sample.

Table 2 here

The preliminary evidence, then, in tables 1 and 2 suggests there could be some link between

momentum and bankruptcy risk, although this pattern may be being somewhat obscured by size

and B/M factors. As such, we need to conduct multivariate analysis to disentangle the underlying

relations.

C. Size, book-to-market, momentum and distress risk: preliminary evidence

Chan and Chen (1991) and Fama and French (1992), among others, argue that smaller

firms and high B/M firms are relatively distressed, and higher returns on such firms are a

compensation for this risk. Dichev (1998), on the other hand, finds distressed firms earn lower

returns than non-distressed firms, and size and B/M are orthogonal to distress risk. Griffin and

Lemmon (2002) find the results of Dichev (1998) are driven by low B/M stocks. Lesmond,

Schill and Zhou (2004) find momentum profits are driven by small stocks with high trading

costs, and such profits disappear once trading costs are taken into account. We therefore

investigate if bankruptcy risk, as measured by z-score, is being driven by small stocks, high B/M

stocks or high momentum stocks. Summary statistics for three different portfolio formation

methods are presented in table 3.

Panel A of table 3 shows that controlling for past year returns, distressed stocks are

smaller than non-distressed stocks and have higher B/M ratios. Consistent with the evidence of

Lesmond, Schill and Zhou (2004) both high momentum and low momentum stocks are smaller.

Importantly, though, the z-score effect is significant in four of the five low momentum deciles

13
but is not significant for higher momentum deciles. These results are consistent with momentum

and distress risk being related to each other. Positive z-score stocks significantly outperform

negative z-score stocks but only when prior-year returns are low.

Panel B shows, controlling for size, the average B/M of distressed stocks is higher than

for non-distressed stocks for all but the two smallest size deciles. Contrary to the evidence of

Vassalou and Xing (2004), the size effect is much stronger for non-distressed stocks (1.88% per

month, t = 4.58) than for distressed stocks (1.17% per month, t = 2.30). Also, the difference

between the returns on distressed (z<0) and non-distressed (z>0) stocks for the two smallest size

deciles is statistically insignificant (t = 1.59 and 1.36 respectively), showing that the z-score

effect is not being driven by very small stocks. Nonetheless, the three size deciles with strongest

z-score effect are still quite small with average market capitalization of under £50 million ($90

million) and trading costs could be substantial for these deciles. Consistent with the predictions

of Hong and Stein’s (1999) slow information diffusion hypothesis, distressed stocks have much

lower prior year returns than non-distressed stocks in smaller size portfolios (1 – 4) but have

higher prior year returns for decile size portfolios 5 – 10. Controlling for size, distressed stocks

do not appear to be less risky than non-distressed stocks as they also have higher betas than non-

distressed stocks for all size deciles.

Finally, Panel C shows controlling for B/M, non-distressed stocks are larger than

distressed stocks. Distressed stocks have higher betas than non-distressed stocks for every B/M

decile, and there is an inverse and almost monotonic relationship between B/M and past year

returns. Particularly interesting are the prior-year returns for the highest B/M decile – these are

almost zero for non-distressed stocks and negative for distressed stocks. Contrary to the distress

factor hypothesis of Chan and Chen (1991) and Fama and French (1992), the B/M effect is not

14
being driven by distressed stocks (0.30% per month, t = 0.75) while it is significant for non-

distressed stocks (1.13% per month, t = 3.63). Contrary to Griffin and Lemmon (2002), we find

our z-score effect is independent of B/M.

Table 3 here

To explore the relation between z-score, prior-year returns, size and B/M in more detail,

table 4 presents the characteristics of portfolios formed by four-way sorts. Panel A

provides evidence of medium term continuance of returns for all the distressed portfolios

(z<0) controlling for size and B/M. There is no evidence of momentum for high B/M non-

distressed stocks (z>0). It also shows, controlling for size and B/M, distressed stocks

underperform non-distressed stocks for low prior-year return portfolios (except for small

size, low B/M stocks). Panel B demonstrates distressed stocks have higher betas than non-

distressed stocks, controlling for size, B/M and prior-year returns, showing firms with

higher bankruptcy risk tend to have higher sensitivity to market movements.

Table 4 here

Thus, table 4 provides preliminary evidence of the momentum effect being driven by

distressed stocks once the effects of size and B/M on stock returns are controlled for. Panels C

and D show we are largely successful in controlling for size and B/M effects in our portfolio

sorts, although not in every case. Panel E shows higher stock return variability for distressed

stocks: such stocks earn lower prior-year returns than non-distressed stocks for low momentum

portfolios and higher prior-year returns for high momentum portfolios. Distressed stocks also

have higher betas than non-distressed stocks suggesting they are riskier, even though they

subsequently earn lower returns.

15
D. The relation between momentum and distress risk: regression evidence

Table 5 presents the results for our Fama-MacBeth (1973) cross-section regressions for

the twenty-four portfolios formed on four-way sorts using equation (1). It shows that negative z-

score stocks earn lower returns than positive z-score stocks and the coefficient on the z-score

binary measure becomes stronger when size, B/M and momentum are present in the pricing

equation. 12

Table 5 here

The results show that, conditional on beta, size, B/M and prior-year returns, negative z-

score portfolios underperform positive z-score portfolios by 31 basis points per month, a

difference that is statistically significant at the 5% level (t = 2.55). There is no evidence of any

size effect, while high B/M stocks outperform low B/M stocks by 38 basis points per month, a

difference that is statistically significant at the 1% level (t = 3.44). The presence of z-score in the

pricing equation has no influence on either of the size or B/M coefficients indicating there is no

common variation between financial distress risk, size or B/M that is linked to stock returns. The

coefficient on momentum is 13 basis points per month (t = 3.39) when z-score is excluded from

the pricing equation. Importantly, the coefficient on momentum becomes statistically

insignificant (6 basis points; t = 1.43) when z-score is included in the pricing equation. These

results strongly suggest momentum is proxying for distress risk, when the distress factor is not

12
This result is consistent with the findings of Dichev (1998) with respect to size and B/M. Ferguson and Shockley
(2003) also find high z-score firms earn higher returns than low z-score firms.

16
included in the asset pricing model. When it is added, the momentum factor is no longer

significant.

III. Is the relation between momentum and financial distress risk conditional?

Our results so far suggest that bankruptcy risk is driving the stock return momentum

effect. Chordia and Shivakumar (2002) show that momentum is only manifest in up-market

conditions and, based on this, they argue that it is a priced risk factor. In this section we explore

whether the relation between prior-year returns and distress risk holds across different market

conditions and different states of the business cycle.

A. Momentum, financial distress risk and expected market returns

If momentum and z-score are truly related factors, their regression coefficients will co-

vary as the state of the market changes. To test whether the unconditional findings of section II

hold conditionally we run separate cross-section regressions for up- and down-markets using

equation (1).

Authors typically use ex-post realized market returns as conditioning variables for state

of the market.13 However, Zhang and Petkova (2004) show that this approach is theoretically

and methodologically incorrect. We therefore employ expected market returns generated by a

macroeconomic forecasting model as conditioning variables. Jagannathan and Wang (1996) use

change in labor income and default risk premium in their conditional asset pricing model. Since

there is no index for yields on corporate bonds, we use the difference between 3-month T-Bill

yield and 3-month LIBOR to proxy for default premium. Chen, Roll and Ross (1986) use a proxy

13
See, for example, Lakonishok and Shapiro (1986), Bhardwaj and Brooks (1993), Lakonishok, Shleifer and
Vishny (1994), Ang and Chen (2002), and Griffin, Ji and Martin (2003).

17
for term structure of interest rates defined as the difference between yields on 10-year and 1-year

government bonds. Since, in the UK, these yields are not available before 1984, we use the

difference between 10-year UK government nominal spot rate and 1-year UK government

nominal spot rate to proxy for the term structure. Following Chordia and Shivakumar (2002), we

also use dividend yield on the FTSE All Share index and 3-month T-Bill yield.

Our expected returns macroeconomic forecasting model is thus given by:

RMt+1 = α + β1 labort + β2 yieldt + β3 divyldt + β4 termt + β5 premt + εt+1 (2)

Where:

RMt+1 is the month t+1 realized return on an equally-weighted market index

labort is the monthly change in economy-wide average income (seasonally adjusted) for month t

yieldt is the yield on the 3-month T-Bill at the end of month t

divyldt is the dividend yield on the FTSE All Share index at the end of month t

termt is the difference between government securities’ 10-year zero rate and 1-year zero rate at

the end of month t

premt is the difference between 3-month T-Bill yield and 3-month LIBOR at the end of month t,

and

εit+1 is a mean-zero stochastic error term.

Using the derived coefficients from fitting equation (2) we estimate the next month

expected market return and divide our 276 months into those with lowest 25% expected market

return (down-market conditions) and those with highest 25% expected market return (up-market

conditions). Table 6 provides parallel results to table 5 for our Fama-Macbeth (1973) cross-

18
section regressions of equation (1) but now bifurcated into lowest quartile expected market

return and highest quartile expected market return portfolios.

Table 6 here

The table shows that negative z-score stocks underperform positive z-score stocks in both

states of the market. Negative z-score stocks underperform positive z-score stocks by 84 basis

points (t = 3.38) in down-market months (Panel A), while in up-market months (Panel B), they

underperform by 49 basis points per month (t = 2.58). During down-markets, smaller stocks

underperform larger stocks by 24 basis points per month (t = 2.36), while high B/M stocks out-

perform low B/M stocks by 61 basis points per month (t = 2.89). There is again little association

between size, B/M and z-scores as they have little impact on each other conditionally.

When z-score is excluded from the pricing equation, momentum is manifested during

both states of the market. During down-market months, it is 15 basis points (t = 2.23) and is

reduced to zero by introduction of z-score into the pricing equation. In up-market conditions,

introduction of z-score reduces momentum from 19 basis points per month (t = 2.18) to a

statistically insignificant 12 basis points per month (t = 1.47).14 Momentum is “driven out” by z-

score in both states of the market. Our results reinforce the evidence presented earlier, size and

B/M are distinct from distress risk while momentum is no longer significant, both

unconditionally and conditionally, when z-score is included in the cross-section regressions.

14
There is no significant z-score effect (5 basis points per month, t=0.39) in the middle two expected market
return quartiles.

19
B. Momentum, financial distress risk and the state of the economy

In this sub-section we investigate the relation between momentum and financial distress

risk in different economic states. Any bankruptcy risk premium is likely to vary with the state of

the economy because poorly performing or distressed firms should be especially sensitive to

economic conditions with their stock returns being affected by common macro-economic factors

such as credit squeezes, liquidity crunches or flight towards quality. Financially distressed firms

are likely to be able to prosper better when periods of high economic growth are expected.

However, they will be harder hit when economic conditions are poor.

We use next quarter GDP growth rate as an indicator of the state of the economy based

on evidence that the stock market seems to lead GDP growth rate by at least a quarter (Fama

(1981), Ogden (2003)). 15 We compute the quarterly long-run average GDP growth rate from

1955 to 2003 and divide the quarters into those with best 25% GDP growth rate and those with

worst 25% GDP growth rate. We then divide our 276 sample months according to the state of the

economy (i.e., whether the next quarter GDP growth rate is among the best 25% or worst 25%)

and conduct Fama-MacBeth (1973) regressions separately for good and bad states of the

economy employing asset pricing equation (1). The results are presented in table 7.

Table 7 here

Panel A of table 7 highlights how distressed stock portfolios (z<0) underperform non-

distressed portfolios by 90 basis points per month conditional on beta, size, B/M and momentum

during expected economic down-turns with the mean regression coefficient more than three

15
For example, January, February and March returns in year t are associated with the GDP growth rate for Q2
year t.

20
standard errors from zero. On the other hand, panel B shows that during expected up-turns, there

is no significant difference in returns between distressed stocks and non-distressed stocks (t =

1.03). Similarly, momentum is manifested only during expected economic down-turns (32 basis

points per month, t = 4.46) and the momentum effect is largely accounted for by z-score when

entered into the pricing equation (15 basis points per month, t = 1.91).16

Table 7 also shows firm size is priced only during periods of expected economic up-turn

(25 basis points per month; t = 2.44) while the B/M effect is manifested during economic down-

turns (but uninfluenced by z-score). These results are, once again, inconsistent with size and B/M

effects proxying for bankruptcy risk. Consistent with the results for different states of the market;

in different economic states of the world, momentum is subsumed by z-score and thus mainly

driven by the distress factor conditionally, as well as unconditionally.

V. Concluding remarks

The primary contribution of this paper is to provide a potential distress factor explanation

for the momentum anomaly. We also explore the impact of different states of the world on factor

risk premia and extend existing work on the market pricing of the distress risk factor using U.K.

capital markets-based data.

The distress factor hypothesis states that small size firms and high B/M firms are

relatively distressed and these factors capture the distress risk missed by the market factor. We

find that, contrary to the distress factor hypothesis, financially distressed stocks earn lower

returns than non-distressed stocks and size, B/M and z-scores are unrelated to each other.

Dichev (1998) and Ferguson and Shockley (2003) reach the same conclusion with their data, but

16
As with expected market returns, there is also no significant z-score effect (-26 basis points per month, t=1.51)
in the middle two next quarter GDP growth rate quartiles.

21
see Vassalou and Xing (2004) for contrary evidence. However, unlike Griffin and Lemmon

(2002), our U.K.-based results are not driven by small firms or low B/M stocks.

The extant literature is still debating whether continuation of prior returns is due to risk or

underreaction to new information. We argue that under Hong and Stein’s (1999) gradual

information diffusion hypothesis there would be an underreaction to the poor solvency position

of a firm leading to medium-term continuation of stock returns. Consistent with this hypothesis,

we find more than half of the 200 bankruptcies in our sample fall in the lowest momentum

quintile. Since such firms have seen their stock prices collapse for some time before failure, they

tend to be of small size and with high B/M ratios. Hong, Lim and Stein (2000) and Lesmond,

Schill and Zhou (2004) find that the momentum effect is driven by return continuation of poor

performers and, consistent with their results, we find distressed stocks (poor past performers)

earn lower subsequent returns. These results are hard to reconcile with rational asset pricing as

distressed stocks are riskier on conventional measures (have higher betas, higher B/Ms, and are

smaller). Further, contrary to any risk story, distressed stocks do not outperform in good states of

world. Importantly, we find momentum is subsumed by z-score showing the momentum effect is

driven by the distress factor effect. As such, we conclude that a financial distress factor should

replace momentum in the standard Carhart (1997) four-factor asset pricing equation.

22
 References

Ahmed, P. and L.J. Lockwood, 1998, Changes in Factor Betas and Risk Premiums Over Varying
Market Conditions, The Financial Review 33, 149-168.
Altman, E.I., 1968, Financial Ratios, Discriminant Analysis and the Prediction of Corporate
Bankruptcy, Journal of Finance 23, 589-609.
Altman, E.I., 1997, An International Survey of Business Failure Classification Models, Financial
Markets, Institutions & Instruments, Vol.6 No.2, 1-57.
Ang, A. and J. Chen, 2002, Asymmetric Correlations of Equity Portfolios, Journal of Fiancial
Economics 63, 443-494.
Ball, R. and S.P. Kothari, 1989, Nonstationary Expected Returns: Implications for Market
Efficiency and Serial Correlations in Returns, Journal of Financial Economics, 25, 51-74.
Barberis, N., A. Shleifer and R. Vishny, 1998, A Model of Investor Sentiment, Journal of
Financial Economics, 49, 307-343.
Beaver, W.H., 1968, Market Prices, Financial Ratios, and the Prediction of Failure, Journal of
Accounting Research, 179-192.
Bharath, S. T. and T. Shumway, 2004, Forecasting Default with the KMV-Merton Model,
Working paper, University of Michigan Business School.
Bhardwaj, R.K. and L.D. Brooks, 1993, Dual Betas from Bull and Bear Markets: Reversal of the
Size Effect, The Journal of Financial Research 16, 269-283.
Chan, K., 1988, On The Contrarian Investment Strategy, Journal of Business, 61(2), 147-164.
Chan, K. and N. Chen, 1991, Structural and Return Characteristics of Small and Large Firms,
Journal of Finance, 46(4), 1467-1484.
Chen, N-F., R. Roll and A. Ross, 1986, Economic Forces and the Stock Market, Journal of
Business, 59(3), 383-404.
Chordia, T. and L. Shivakumar, 2002, Momentum, Business Cycle and Time-varying Expected
Returns, Journal of Finance 57, 985-1019.
Conrad, J. and G. Kaul, 1998, An Anatomy of Trading Strategies, Review of Financial Studies,
11, 489-519.
Daniel, K. and S. Titman, 1999, Market Efficiency in an Irrational World, Financial Analyst
Journal, 55, 28-40.
Dichev, I.D., 1998, Is the Risk of Bankruptcy a Systematic Risk?, Journal of Finance 53, 1131-
1147.
Dimson, E., 1979, Risk Measurement When Shares Are Subject To Infrequent Trading, Journal
of Financial Economics 7, 197-226.

23
Fama, E.F., 1981, Stock Returns, Real Activity, Inflation And Money, American Economic
Review 71, 545-565.
Fama, E.F., 1998, Market Efficiency, Long-term Returns and Behavioural Finance, Journal of
Financial Economics 49, 283-306.
Fama, E., and J.D. MacBeth, 1973, Risk, Return, and Equilibrium: Empirical Tests, Journal of
Political Economy 81, 607-636.
Fama E.F. and K.R. French, 1992, The Cross-section of Expected Stock Returns, Journal of
Finance 47, 427-465.
Fama E.F. and K.R. French, 1993, Common Risk Factors in the Returns on Stocks and Bonds,
Journal of Financial Economics 33, 3-56.
Fama E.F. and K.R. French, 1995, Size and Book-to-Market Factors in Earnings and Returns,
Journal of Finance 50, 131-155.
Ferguson, M.F. and R.L. Shockley, 2003, Equilibrium “Anomalies”, Journal of Finance 58,
2549-2580.
Griffin, J., X. Ji and J.S. Martin, 2003, Momentum Investing and Business Cycle Risk: Evidence
from Pole to Pole, Journal of Finance 58, 2515-2547.
Griffin, J. and L. Lemmon, 2002, Book-to-Market Equity, Distress Risk, and Stock Returns,
Journal of Finance 57, 2317-2336.
Hong, H. and J. Stein, 1999, A Unified Theory of Underreaction, Momentum Trading and
Overreaction in Asset Markets, Journal of Finance, 54, 2143-2184.
Hong, H., T. Lim and J. Stein, 2000, Bad News Travels Slowly: Size Analyst Coverage, and the
Profitability of Momentum Strategies, Journal of Finance, 55, 265-295.
Howton, S. and D. Peterson, 1998, An Examination of Cross-Sectional Realized Stock Returns
using a Varying-Risk Beta Model, The Financial Review 33, 199-212.
Jagannathan, R. and Z. Wang, 1996, The Conditional CAPM and the Cross-Section of Expected
Returns, Journal of Finance 51, 3-53.
Jegadeesh, N. and S. Titman, 1993, Returns to Buying Winners and Selling Losers: Implications
for Stock Market Efficiency, Journal of Finance, 48, 65-91.
Jegadeesh, N. and S. Titman, 2001, Profitability of Momentum Strategies: An Evaluation f
Alternative Explanations, Journal of Finance 56 (2), 699-720.
Kaplan, S. and L. Zingales, 1997, Do Financing Constraints Explain Why Investment is
Correlated with Cash Flow, Quarterly Journal of Economics, 112, 168-216.
Lakonishok, J. and A. Shapiro, 1986, Systematic Risk, Total Risk and Size as Determinants of
Stock Market Returns, Journal of Banking and Finance 10, 115-132.

24
Lakonishok, J., A. Shleifer and R. Vishny (1994), Contrarian Investment, Extrapolation and
Risk, Journal of Finance 49, 1541-1578.
Lamont, O., C. Polk and J. Saa-Requejo, 2001, Financial Constraints and Stock Returns, The
Review of Financial Studies, 14(2), 529-554.
Lesmond, D., M. Schill and C. Zhou, 2004, The Illusory Nature of Momentum Profits, Journal
of Financial Economics, 71, 349-380.
Lo, A.W. and C.A. MacKinlay, 1990, Data Snooping in Tests of Financial Asset Pricing Models,
Review of Financial Studies 10, 431-467.
Ogden, J.P., 2003, The Calendar Structure of Risk and Expected Returns on Stocks and Bonds,
Journal of Financial Economics 70, 29-67.
Ohlson, J.A., 1980, Financial Ratios and the Probabilistic Prediction of Bankruptcy, Journal of
Accounting Research 19, 109-131.
Taffler, R.J., 1983, The Assessment of Company Solvency and Performance Using a Statistical
Model, Accounting and Business Research 15, 295-307.
Taffler, R.J., 1984, Empirical Models for the Monitoring of UK Corporations, Journal of
Banking and Finance 8, 199-227.
Vassalou, M. and Y. Xing, 2004, Default Risk in Equity Returns, Journal of Finance, 59 (2),
831-868.
Zarowin, P., 1990, Size, Seasonality, And Stock Market Overreaction, Journal of Financial and
Quantitative Analysis, 25(1), 113-126.
Zhang, L. and R. Petkova, 2004, Is Value Riskier than Growth?, Working paper, Simon School
of Business, Rochester.

25
Figure 1: Year-wise proportion of firms with negative z-scores (1979-2001)
At the end of September of each year from 1979 to 2001, all the stocks in our sample are
allocated to two groups based on whether their latest available z-score is positive or negative. Z-scores are
computed using the discriminant model of Taffler (1983, 1984) and firms with negative z-scores have a
higher risk of financial distress. To minimize the look ahead bias, a five month lag between the balance
sheet date and reporting date is assumed. Portfolios are rebalanced at the end of September each year.
Negative B/M stocks are excluded.

40

35

30
% of negative z-score firms

25

20

15

10

0
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001

26
Table 1: Failure rates in two-way portfolios
Portfolios in panel A are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on momentum and grouped into
ten portfolios. Twenty portfolios are then formed at the intersections of momentum and z-score.
Momentum is defined as the 11-month return from October year t-1 to August year t.
Portfolios in panel B are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on market capitalization and
grouped into ten portfolios. Twenty portfolios are then formed at the intersections of size and z-score.
Portfolios in panel C are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on B/M and grouped into ten
portfolios. Twenty portfolios are then formed at the intersections of B/M and z-score. B/M is computed as
the ratio of book value of equity (excluding preference capital) plus deferred taxes less minority interests
from the latest available financial statements, divided by the market value of equity on September 30th.
Portfolios are rebalanced at the end of September each year. Negative B/M stocks are excluded.
The list of failures is compiled from LSPD, The Stock Exchange Official Yearbook and CGT Capital
Losses.

27
 % of firms Distribution of failures (%) Failure rate (%)
with Z<0 Z<0 Z>0 Total Z<0 Z>0 Total
A. Momentum and z-score portfolios
Low 41.3 39.0 0.5 39.5 8.3 0.1 3.5
2 27.6 11.0 1.0 12.0 3.5 0.1 1.0
3 22.3 10.0 0.5 10.5 3.9 0.1 0.9
4 19.8 4.5 0.5 5.0 2.0 0.1 0.4
5 18.9 7.5 0.5 8.0 3.5 0.1 0.7
6 19.0 4.0 0.0 4.0 1.9 0.0 0.4
7 19.7 4.0 0.5 4.5 1.8 0.1 0.4
8 19.5 6.0 0.0 6.0 2.7 0.0 0.5
9 22.0 2.5 0.5 3.0 1.0 0.1 0.3
High 29.7 7.0 0.5 7.5 2.1 0.1 0.7
Total 24.0 95.5 4.5 100.0 3.5 0.1 0.9
B. Size and z-score portfolios
Small 43.1 40.0 1.0 41.0 8.6 0.2 3.8
2 35.3 24.0 1.5 25.5 5.2 0.2 1.9
3 29.7 7.5 1.0 8.5 2.3 0.1 0.8
4 27.2 9.5 0.0 9.5 2.9 0.0 0.8
5 22.8 6.0 0.0 6.0 2.5 0.0 0.6
6 20.2 4.0 0.0 4.0 1.8 0.0 0.4
7 17.1 2.5 0.0 2.5 1.3 0.0 0.2
8 15.6 0.5 0.5 1.0 0.3 0.1 0.1
9 12.2 1.0 0.0 1.0 0.7 0.0 0.1
Big 15.8 0.5 0.5 1.0 0.3 0.1 0.1
Total 24.0 95.5 4.5 100.0 3.5 0.1 0.9
C. B/M and z-score portfolios
Low 27.9 8.5 0.0 8.5 2.7 0.0 0.7
2 21.7 5.5 0.0 5.5 2.2 0.0 0.5
3 19.6 10.0 1.5 11.5 4.5 0.2 1.0
4 18.4 2.5 0.0 2.5 1.2 0.0 0.2
5 18.4 5.0 1.0 6.0 2.4 0.1 0.5
6 19.8 5.5 0.0 5.5 2.5 0.0 0.5
7 22.8 6.0 0.0 6.0 2.3 0.0 0.5
8 24.8 5.5 0.0 5.5 1.9 0.0 0.5
9 30.8 15.5 0.0 15.5 4.4 0.0 1.4
High 35.6 31.5 2.0 33.5 7.7 0.3 2.9
Total 24.0 95.5 4.5 100.0 3.5 0.1 0.9

28
Table 2: Failure rates in four-way portfolios
At the end of September of each year from 1979 to 2001, all the stocks in our sample are ranked
on market capitalization and grouped into two portfolios, independently ranked on B/M and grouped into
three portfolios, independently ranked on prior-year return and grouped into two portfolios and, finally,
separately allocated to two groups based on whether their latest available z-score is negative or positive.
Twenty-four portfolios are then formed at the intersections of size, B/M, momentum and z-score.
Portfolios are rebalanced at the end of September each year. Negative B/M stocks are excluded. The list
of failures is compiled from LSPD, The Stock Exchange Official Yearbook and CGT Capital Losses.

Prior year Low B/M Medium B/M High B/M

Size
return Z<0 Z>0 Z<0 Z>0 Z<0 Z>0
A. Distribution of failures (%)
Small Low 12.0 0.0 12.0 0.0 43.0 2.0
Small High 8.5 0.5 4.5 1.0 7.0 0.0
Big Low 2.5 1.0 1.0 0.0 1.5 0.0
Big High 1.0 0.0 1.5 0.0 1.0 0.0
B. Failure rates (%)
Small Low 6.9 0.0 4.0 0.0 7.1 0.2
Small High 3.7 0.1 1.8 0.1 2.7 0.0
Big Low 2.0 0.2 0.6 0.0 1.4 0.0
Big High 0.4 0.0 0.8 0.0 1.6 0.0

29
Table 3: Summary statistics: two-way portfolios
Portfolios in panel A are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on market capitalization and
grouped into ten portfolios. Twenty portfolios are then formed at the intersections of size and z-score.
Portfolios in panel B are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on B/M and grouped into ten
portfolios. Twenty portfolios are then formed at the intersections of B/M and z-score.
Portfolios in panel C are formed as follows: at the end of September of each year from 1979 to
2001, all the stocks in our sample are allocated to two groups based on whether their latest available z-
score is negative or positive. The stocks are also independently ranked on momentum and grouped into
ten portfolios. Twenty portfolios are then formed at the intersections of momentum and z-score.
Momentum is defined as the monthly return averaged over the 11-month period from October year t-1 to
August year t. Portfolios are rebalanced at the end of September each year.
B/M is computed as the ratio of book value of equity (excluding preference capital) plus deferred
taxes less minority interests from the latest available accounts divided by the market value of equity on
September 30th. Average monthly excess return is the time series average of the difference between
monthly stock returns and one-month Treasury bill rate observed at the beginning of the month. Portfolio
betas are the sum of slopes in the regression of the return on a portfolio on the current, prior and next
month’s market returns. Average size, average B/M and average momentum are the time-series averages
of monthly averages of market capitalizations, B/M and prior twelve month returns (excluding
September) respectively for stocks in the portfolio at the end of September of each year. Negative B/M
stocks are excluded. The last monthly return for failed stocks is set equal to –100%.
Average market Average prior
Average excess monthly
capitalization Average B/M year return Average beta
returns (%)
(£m) (%)
Z<0 Z>0 Z<0 Z>0 Z<0 Z>0 Z<0 Z>0 Z<0 Z>0 t-diff
A. Momentum and z-score portfolios
Low 76.6 163.3 1.90 1.53 -5.05 -4.26 1.06 1.04 0.09 0.86 2.58
2 274.2 387.9 1.41 1.24 -1.65 -1.63 1.02 0.92 -0.15 0.63 3.04
3 357.7 345.0 1.31 1.15 -0.44 -0.44 1.05 0.91 -0.19 0.51 2.76
4 347.2 562.0 1.20 1.05 0.43 0.42 1.04 0.91 0.69 0.58 -0.41
5 534.0 796.1 1.12 1.00 1.17 1.18 1.05 0.93 -0.10 0.48 2.36
6 470.3 660.2 1.01 0.95 1.89 1.89 1.03 0.95 1.03 0.67 -1.50
7 644.9 647.5 1.03 0.90 2.66 2.62 1.03 1.00 0.89 0.52 -1.76
8 458.9 604.0 0.88 0.82 3.51 3.52 1.21 1.02 0.46 0.59 0.52
9 448.9 590.9 0.82 0.76 4.81 4.79 1.19 1.08 0.98 0.82 -0.71
High 339.2 308.2 0.72 0.68 9.25 8.43 1.48 1.30 0.54 0.69 0.61
t (High-Low) 1.09 -0.53
B. Size and z-score portfolios
Small 2.9 3.0 1.69 1.87 -0.16 1.19 0.96 0.70 1.50 2.24 1.59
2 7.2 7.5 1.50 1.51 0.56 1.29 1.06 0.92 0.74 1.07 1.36
3 13.2 13.2 1.26 1.25 1.54 1.57 1.12 0.93 0.04 0.74 2.80
4 21.4 21.5 1.18 1.09 1.32 1.57 1.20 0.97 -0.02 0.44 2.02
5 34.9 35.2 1.09 0.98 1.61 1.53 1.26 1.00 -0.02 0.73 3.09
6 57.5 58.1 1.00 0.81 2.06 1.90 1.39 1.00 0.15 0.40 1.02
7 99.2 98.4 0.86 0.77 2.10 1.87 1.31 1.05 -0.06 0.39 1.81
8 181.2 186.3 0.76 0.71 2.44 2.14 1.28 1.10 0.18 0.43 1.10
9 470.4 467.0 0.84 0.67 2.02 2.07 1.20 1.10 0.36 0.42 0.28
Big 3451.0 4175.0 0.69 0.67 2.05 2.04 0.96 0.94 0.33 0.36 0.16
t (Small-Big) 2.30 4.58

30
C. B/M and z-score portfolios
Low 580.0 1191.7 0.16 0.18 4.24 3.86 1.24 1.11 0.45 0.11 -1.10
2 412.2 553.7 0.33 0.34 3.07 2.84 1.20 1.08 0.23 0.20 -0.12
3 769.0 601.6 0.47 0.47 2.22 2.39 1.13 1.01 -0.46 0.25 2.66
4 491.0 801.0 0.60 0.60 1.59 2.06 1.14 1.02 0.28 0.47 0.67
5 449.3 466.3 0.73 0.73 1.71 1.70 1.18 0.97 0.44 0.50 0.19
6 443.3 434.3 0.88 0.88 1.32 1.49 0.99 0.95 0.31 0.76 1.70
7 239.0 438.3 1.06 1.06 0.95 1.20 1.11 0.95 0.33 0.93 2.75
8 118.3 376.1 1.31 1.31 0.86 1.08 1.14 0.92 0.72 0.96 1.00
9 136.6 143.8 1.72 1.71 0.08 0.60 1.11 0.97 0.45 0.96 2.18
High 96.8 262.7 3.21 3.20 -1.44 0.02 1.14 0.92 0.75 1.24 1.55
t (High-Low) 0.75 3.63

31
 Table 4: Summary statistics – size, B/M, momentum and z-score portfolios
At the end of September of each year from 1979 to 2001, all the stocks in our sample are ranked
on market capitalization and grouped into two portfolios, independently ranked on B/M and grouped into
three portfolios and independently ranked on prior year return and grouped into two portfolios. The stocks
are also independently allocated to two groups based on whether their latest available z-score is negative
or positive. Twenty-four portfolios are then formed at the intersections of size, B/M, momentum and z-
score. Portfolios are rebalanced at the end of September each year. B/M is computed as the ratio of book
value of equity (excluding preference capital) plus deferred taxes less minority interests from the latest
available accounts divided by the market value of equity on September 30th. Average monthly excess
return is the time series average of the difference between monthly stock returns and one-month Treasury
bill rate observed at the beginning of the month. Portfolio betas are the sum of slopes in the regression of
the return on a portfolio on the current, prior and next month’s market returns. Average size, average B/M
and average momentum are the time-series averages of monthly averages of market capitalizations, B/M
and prior 11-month average monthly returns (October year t-1 to August year t) respectively for stocks in
the portfolio at the end of September of each year. Negative B/M stocks are excluded. The last monthly
return for failed stocks is set equal to –100%.

Low B/M Medium B/M High B/M

Size Momentum
z<0 z>0 z<0 z>0 z<0 z>0
A. Average excess monthly returns (%)
Small Low 0.35 0.25 0.14 0.89 0.28 1.10
High 0.44 0.95 0.87 0.84 1.51 1.12
Large Low -0.56 -0.11 -0.26 0.30 0.33 0.90
High 0.25 0.27 0.57 0.65 0.83 0.92
B. Average beta
Small Low 0.97 1.04 1.00 0.92 1.05 0.87
High 1.37 1.04 1.26 1.01 1.09 1.00
Large Low 0.93 0.90 1.01 0.86 1.15 0.92
High 1.35 1.00 1.09 0.95 1.46 0.96
C. Average market capitalization (£m)
Small Low 14.1 25.1 14.6 19.8 10.4 13.1
High 18.6 23.7 15.3 18.9 11.4 13.0
Large Low 1018.1 997.8 935.5 977.5 521.5 975.2
High 1027.2 930.7 886.0 923.9 940.2 1430.5
D. Average B/M
Small Low 0.32 0.37 0.86 0.85 2.38 2.16
High 0.29 0.35 0.84 0.83 1.90 2.07
Large Low 0.31 0.35 0.82 0.81 1.93 1.73
High 0.29 0.31 0.79 0.77 1.58 1.71
E. Average prior year returns (%)
Small Low -1.63 -0.79 -1.78 -0.81 -2.45 -1.26
High 6.52 5.46 4.74 4.18 4.01 3.78
Large Low -0.44 -0.11 -0.76 -0.34 -1.72 -0.84
High 5.33 4.32 3.75 3.40 4.23 3.14

32
Table 5: Cross-section regression results
At the end of September of each year from 1979 to 2001, all the stocks in our sample are ranked
on market capitalization and grouped into two portfolios, independently ranked on B/M and grouped into
three portfolios and independently ranked on prior year return and grouped into two portfolios. The stocks
are also independently allocated to two groups based on whether their latest available z-score is negative
or positive. Twenty-four portfolios are then formed at the intersections of size, B/M, momentum and z-
score. Portfolios are rebalanced at the end of September each year.
B/M is computed as the ratio of book value of equity (excluding preference capital) plus deferred
taxes less minority interests from the latest available accounts divided by the market value of equity on
September 30th. To avoid undue influence of outliers on the regressions, the smallest and largest 1% of
the observations on B/M are set equal to 0.01 and 0.99 fractiles respectively. Portfolio betas are the sum
of slopes in the regression of the return on a portfolio on the current, prior and following month’s market
returns.
Rit is the equally-weighted return on portfolio i during month t and RFt is the one-month Treasury
bill rate at the beginning of month t. βit-1 is the beta of portfolio i estimated at the portfolio formation date.
ln(sizeit-1) and ln(B/Mit-1) are the natural logarithms of average of market capitalizations and average of
B/M ratios respectively of stocks in portfolio i at the portfolio formation date. Momit-1 is the average
monthly return over the 11 months from October year t-1 to August year t prior to the month of portfolio
formation of all the stocks in portfolio i. z(0/1)it-1 is equal to 1 if the latest available z-score is negative, 0
otherwise. The slopes are estimated by Fama-MacBeth cross-section regressions for each of the 276
months from October 1979 to September 2002. Figures in brackets are the respective t-statistics. Negative
B/M stocks are excluded. The last period return for failed stocks is set equal to –100%.

Rit - RFt = αit + γ1t βit-1 + γ2t ln(sizeit-1) + γ3t ln(B/Mit-1) + γ4t Momit-1+ γ5t z(0/1)it-1
α γ1 γ2 γ3 γ4 γ5
2.02 -0.10 -0.07 0.26
(2.10) (-0.55) (-1.37) (2.65)
1.93 -0.01 -0.06 0.30 -0.33
(2.00) (-0.04) (-1.20) (2.95) (-2.81)
1.84 -0.07 -0.06 0.38 0.06 -0.31
(1.83) (-0.22) (-1.22) (3.44) (1.43) (-2.55)
2.49 -0.31 -0.10 0.37 0.13
(2.55) (-1.06) (-1.88) (3.46) (3.39)

33
Table 6: Regression results – bifurcation into up- and down-markets
At the end of September of each year from 1979 to 2001, all the stocks in our sample are ranked
on market capitalization and grouped into two portfolios, independently ranked on B/M and grouped into
three portfolios and independently ranked on prior year return and grouped into two portfolios. The stocks
are also independently allocated to two groups based on whether their latest available z-score is negative
or positive. Twenty-four portfolios are then formed at the intersections of size, B/M, momentum and z-
score. Portfolios are rebalanced at the end of September each year.
B/M is computed as the ratio of book value of equity (excluding preference capital) plus deferred
taxes less minority interests from the latest available accounts divided by the market value of equity on
September 30th. To avoid undue influence of outliers on the regressions, the smallest and largest 1% of
the observations on B/M are set equal to 0.01 and 0.99 fractiles respectively. Portfolio betas are the sum
of slopes in the regression of the return on a portfolio on the current, prior and following month’s market
returns.
Rit is the equally-weighted return on portfolio i during month t and RFt is the one-month Treasury
bill rate at the beginning of month t. βit-1 is the beta of portfolio i estimated at the portfolio formation date.
ln(sizeit-1) and ln(B/Mit-1) are the natural logarithms of average of market capitalizations and average of
B/M ratios respectively of stocks in portfolio i at the portfolio formation date. Momit-1 is the average
monthly return over the 11 months from October year t-1 to August year t prior to the month of portfolio
formation of all the stocks in portfolio i. z(0/1)it-1 is equal to 1 if the latest available z-score is negative, 0
otherwise. The slopes are estimated by Fama-MacBeth cross-section regressions for each of the 276
months from October 1979 to September 2002. Figures in brackets are the respective t-statistics. Negative
B/M stocks are excluded. The last period return for failed stocks is set equal to –100%.
The months when the expected market return is in the lowest quartile are classified down-market
and the months when the expected market return is in the highest quartile are classified as up-market.

Rit - RFt = αit + γ1t βit-1 + γ2t ln(sizeit-1) + γ3t ln(B/Mit-1) + γ4t Momit-1 + γ5t z(0/1)it-1
α γ1 γ2 γ3 γ4 γ5
A. Down-market conditions (worst 25% months)
-3.88 -0.68 0.20 0.51
(-2.13) (-2.36) (2.04) (2.66)
-4.35 -0.07 0.21 0.55 -0.80
(-2.34) (-0.19) (2.19) (2.86) (-3.28)
-2.75 -1.15 0.15 0.54 0.15
(-1.49) (-2.41) (1.53) (2.77) (2.23)
-4.82 -0.08 0.24 0.61 0.00 -0.84
(-2.45) (-0.17) (2.36) (2.89) (0.00) (-3.38)
B. Up-market conditions (best 25% months)
3.77 0.25 -0.16 0.12
(2.00) (0.63) (-1.53) (0.53)
4.18 0.19 -0.16 0.15 -0.58
(2.24) (0.43) (-1.55) (0.61) (-2.85)
3.86 -0.42 -0.16 0.40 0.19
(2.04) (-0.61) (-1.55) (1.53) (2.18)
4.49 -0.03 -0.19 0.32 0.12 -0.49
(2.44) (-0.04) (-1.80) (1.18) (1.47) (-2.58)

34
Table 7: Regression results - bifurcation into up and down states of the economy
At the end of September of each year from 1979 to 2001, all the stocks in our sample are ranked
on market capitalization and grouped into two portfolios, independently ranked on B/M and grouped into
three portfolios and independently ranked on prior year return and grouped into two portfolios. The stocks
are also independently allocated to two groups based on whether their latest available z-score is negative
or positive. Twenty-four portfolios are then formed at the intersections of size, B/M, momentum and z-
score. Portfolios are rebalanced at the end of September each year.
B/M is computed as the ratio of book value of equity (excluding preference capital) plus deferred
taxes less minority interests from the latest available accounts divided by the market value of equity on
September 30th. To avoid undue influence of outliers on the regressions, the smallest and largest 1% of
the observations on B/M are set equal to 0.01 and 0.99 fractiles respectively. Portfolio betas are the sum
of slopes in the regression of the return on a portfolio on the current, prior and following month’s market
returns.
Rit is the equally-weighted return on portfolio i during month t and RFt is the one-month Treasury
bill rate at the beginning of month t. βit-1 is the beta of portfolio i estimated at the portfolio formation date.
ln(sizeit-1) and ln(B/Mit-1) are the natural logarithms of average of market capitalizations and average of
B/M ratios respectively of stocks in portfolio i at the portfolio formation date. Momit-1 is the average
monthly return over the 11-months from October of year t-1 to August of year t prior to the month of
portfolio formation of all the stocks in portfolio i. z(0/1)it-1 is equal to 1 if the latest available z-score is
negative, 0 otherwise. The slopes are estimated by Fama-MacBeth cross-section regressions for each of
the 276 months from October 1979 to September 2002. Figures in brackets are the respective t-statistics.
Negative B/M stocks are excluded. The last period return for failed stocks is set equal to –100%.
The quarters when next quarter GDP growth rate is in the lowest 25% for our sample period are
classified as down-turns and the quarters when next quarter GDP growth rate is in the highest 25% for our
sample period are classified as up-turns. Negative B/M stocks are excluded for the portfolios. The last
month return for failed stocks is set equal to –100%.

Rit - RFt = αit + γ1t βit-1 + γ2t ln(sizeit-1) + γ3t ln(B/Mit-1) + γ4t Momit-1 + γ5t z(0/1)it-1
α γ1 γ2 γ3 γ4 γ5
A. Economic down-turns (next quarter GDP growth rate in lowest quartile)
-3.31 -0.31 0.19 0.15
(-2.17) (-0.72) (2.19) (0.78)
-3.15 0.14 0.18 0.16 -1.04
(-1.99) (0.29) (2.07) (0.81) (-3.98)
-2.14 -0.52 0.13 0.48 0.32
(-1.25) (-0.86) (1.32) (2.39) (4.46)
-3.07 0.03 0.17 0.43 0.15 -0.90
(-1.70) (0.04) (1.82) (2.14) (1.91) (-3.37)
B. Economic up-turns (next quarter GDP growth rate in highest quartile)
6.38 0.31 -0.27 0.01
(3.22) (0.77) (-2.61) (0.06)
6.15 0.08 -0.25 0.08 0.19
(3.13) (0.17) (-2.37) (0.35) (1.00)
6.25 0.02 -0.26 0.07 0.07
(3.18) (0.04) (-2.61) (0.31) (1.00)
6.09 -0.26 -0.25 0.13 0.09 0.19
(3.15) (-0.45) (-2.44) (0.53) (1.14) (1.03)

35