You are on page 1of 32

Bankruptcy Risk and the Value Premium in Australia†

Charles E. Hyde

MIR Investment Management

Abstract
We compare four bankruptcy prediction models using data from large listed companies in

Australia. The best performing model gets 70-90% of bankruptcy predictions correct. Using

this model we show that high bankruptcy risk stocks do not generate higher returns than low

risk stocks. The penalty for holding risk is highest in portfolios of small stocks and deep

value stocks. Value does not consistently increase with bankruptcy risk and bankruptcy risk

is shown to be higher in low value than high value portfolios. Adding an aggregate distress

factor to a 4-factor model, the value factor loading remains statistically significant, implying

that the value factor contains substantially more than just distress-related information.

JEL Code: G11, G33

Keywords: Bankruptcy, distress, value premium

† The author would like to thank Andrew Poppenbeek for assistance with data and John Beggs, Stephen Brown,

Rob Trevor and Michael Triguboff for helpful discussions. Correspondence: MIR Investment Management,

Level 40, 50 Bridge St, Sydney, NSW, 2000, Australia; Tel +612 8222 0824; Fax +612 9230 0543; Email

chyde@mir.com.au.

Electronic
Electroniccopy
copyavailable
availableat:
at:https://ssrn.com/abstract=1293913
http://ssrn.com/abstract=1293913
Bankruptcy Risk and the Value Premium in Australia

Abstract
We compare four bankruptcy prediction models using data from large listed companies in

Australia. The best performing model gets 70-90% of bankruptcy predictions correct. Using

this model we show that high bankruptcy risk stocks do not generate higher returns than low

risk stocks. The penalty for holding risk is highest in portfolios of small stocks and deep

value stocks. Value does not consistently increase with bankruptcy risk and bankruptcy risk

is shown to be higher in low value than high value portfolios. Adding an aggregate distress

factor to a 4-factor model, the value factor loading remains statistically significant, implying

that the value factor contains substantially more than just distress-related information.

Electronic
Electroniccopy
copyavailable
availableat:
at:https://ssrn.com/abstract=1293913
http://ssrn.com/abstract=1293913
1. Introduction

The substantial literature on bankruptcy prediction stems from the fact that a wide range

of economic agents have a vested interest in understanding bankruptcy, among them

creditors, shareholders and regulators. We approach the topic from the perspective of

investment analysis, using a bankruptcy prediction model for the Australian equities market

to determine whether bankruptcy risk is the source of the widely-documented value premium.

While Fama and French (1995) and Chen and Zhang (1998) have shown that value stocks

exhibit a range of characteristics one would expect to be associated with financial distress

(e.g., greater earnings uncertainty, higher leverage, higher likelihood of cutting dividends), as

yet there is no consensus that the value premium is in fact attributable to value stocks

experiencing higher levels of financial distress. Ferguson and Shockley (2003) and Vassalou

and Xing (2004) present evidence in support of this hypothesis while a number of other

studies take the opposite view.

Specifically, we inquire as to whether high bankruptcy risk stocks are associated with

higher returns and deeper value characteristics than low bankruptcy stocks. While in the

same vein as previous US studies such as Vassalou and Xing (2004), Zaretzky and Zumwalt

(2007) and Campbell, Hilscher and Szilagyi (2008), ours is the first study to bring data from

the Asia Pacific region to bear on this topic.1 The bankruptcy prediction literature is large,

beginning with the seminal contributions of Beaver (1966) and Altman (1968) who pioneered

the use of accounting ratios in multivariate models to predict bankruptcy. Other key studies

include Merton (1974) who developed the option pricing approach to default prediction,

Ohlson (1980) who introduced logit models to bankruptcy modeling, and Shumway (2001)

who augmented accounting ratios with market data. Previous studies of bankruptcy in

1
The only other non-US study we are aware of is Agarwal and Poshakwale (2008), who present evidence on the
UK market. They find that the value premium is not related to distress risk.

Electronic copy available at: https://ssrn.com/abstract=1293913


Australia include Castagna and Matolcsy (1981), Lincoln (1984), Izan (1984), Ghargori,

Chan and Faff (2007) and Tanthanongsakkun, Pitt and Treepongkaruna (2008).

We compare four different bankruptcy prediction models, of which three are from the

published Altman (1968), Ohlson (1980) and Shumway (2001) studies. The fourth

„Alternate‟ model employs both accounting ratios and market variables specifically selected

for this study. Stocks that are so small as to be uninvestible are excluded from the sample –

while this is crucial to ensuring the model is relevant to practitioners, many studies (including

the Australian studies above) have not screened these stocks out.2 We show that the

goodness of fit of the Alternate model is better than that achieved using the model Altman,

Ohlson and Shumway models. The Alternate model also has strong predictive accuracy. Of

the stocks in the highest decile of predicted probabilities generated by the model, 66% turn

out to be correct predictions, a higher hit rate than that achieved by the other three models.

Using a probability cut-off of 0.80 or higher to identify likely bankruptcy candidates from the

Alternate model, 90% of the stocks predicted to be bankrupt are correct predictions.

The Alternate model is then applied to historical data and stocks are then sorted by their

level of predicted bankruptcy risk into value-weighted portfolios. These sorts in conjunction

with value sorts are then used to analyze the characteristics of bankruptcy risk. Our approach

is thus similar to that employed by Dichev (1998), Griffin and Lemmon (2002), Vassalou and

Xing (2004) and Campbell et al. (2008).

First we show that high bankruptcy risk stocks do not reliably outperform low

bankruptcy risk stocks either in a benchmark-adjusted sense or a risk-adjusted sense. A

portfolio that is long high bankruptcy risk stocks and short low bankruptcy risk stocks yields

a return that is negative, although not statistically significantly different from zero.

2
Since financial distress risk is concentrated in very small stocks, not screening these stocks means that the
model will be calibrated largely against stocks which cannot be purchased by institutional investors.

Electronic copy available at: https://ssrn.com/abstract=1293913


Conventional wisdom suggests that bankruptcy risk will be most heavily concentrated in deep

value stocks. It follows that if there is a reward for holding bankruptcy risk, it should be most

pronounced in portfolios of high (i.e., deep) value stocks than in portfolios of low value (i.e.,

growth) stocks. We show that this is not the case, indicating that reward does not follow risk

in relation to financial distress. This same relationship has been documented in the US data

by Dichev (1998), Griffin and Lemmon (2002), Avramov, Chordia, Jostova and Philipov

(2007), Garlappi, Shu and Yan (2008) and Campbell et al. (2008).

Fama and French (1996) suggest that the value premium represents compensation for

bearing financial distress risk. Since the Australian market – like most other countries –

displays a value premium, our results above suggest that the value premium in Australia is

not predominantly a reward for bearing financial distress risk.3 We undertake a detailed

analysis of the relationship between the value and distress characteristics of stocks,

considering value in terms of both book-to-price and earnings-to-price ratios. First we

examine the relationship by sorting stocks according to both value and bankruptcy risk. The

book-to-price ratio is shown not to increase monotonically with bankruptcy risk while the

earnings-to-price ratio is shown to be essentially independent of bankruptcy risk. In addition,

the average level of bankruptcy risk in high value portfolios is no higher, and sometimes

substantially lower, than that in low value portfolios. We also show that although the spread

of bankruptcy risk is much higher in portfolios of high bankruptcy risk than in portfolios of

low bankruptcy risk, the value premium does not vary in any consistent fashion with the level

of risk.

Second, we examine the relationship between value and distress through the lens of a

factor model. Following Vassalou and Xing (2004) we add a fifth factor, aggregate distress,

3
The S&P/Citigroup BMI Australia Value – Growth Index shows (geometric) average returns of 1.8% pa over
the period 1992 to 2007. Excluding the bull market period spanning 2004-2007, the return increases to 3.4% pa.

Electronic copy available at: https://ssrn.com/abstract=1293913


to the Carhart (1997) four-factor model and show that even in the presence of the distress

factor the loading on the value factor remains positive and statistically significant. This

strongly suggests that the value factor contains a significant amount of information that is

uncorrelated with distress. This result is robust to whether the aggregate distress factor is

constructed from the model described in this paper or from the Merton option-pricing based

approach to measuring financial distress.

On balance, our results suggest that the value premium is at best only weakly related to

financial distress. Our findings can thus be viewed as out-of-sample support for the results of

Dichev (1998), Griffin and Lemmon (2002), Campbell et al. (2008) using US data. The latter

two studies present evidence showing that the distress anomaly is strongest amongst stocks

likely to experience informational and/or trading inefficiencies, suggesting the value effect is

not compensation for distress risk. In contrast, Vassalou and Xing (2004) find that the value

premium (as well as the size effect) is largely a default risk effect.4 Specifically, they find

that the size and value effects exist only in portfolios of stocks with high default risk. They

also show that the Fama-French size and value factors contain information about default risk.

We describe the data and model specification in section 2. In section 3 we report the

regression results and statistics illustrating the prediction accuracy of the model. Section 4

examines the relationship between distress risk (as defined by the model) and returns, while

the relationship between distress risk and value characteristics is analyzed in section 5.

Conclusions are drawn in section 6.

4
Because size is one of the explanatory variables in our bankruptcy prediction model, it follows that the size
effect is strongly related to financial distress in our model.

Electronic copy available at: https://ssrn.com/abstract=1293913


2. Data and Model Specification

2.1. Data

Our data sample spans the period 1988-2008 and is restricted to listed Australian

companies included in the ASX All Ordinaries universe. Smaller stocks not included in this

universe are too illiquid and thus not investible for most institutional fund managers. The

sample includes 49 bankrupt stocks and 424 non-bankrupt stocks, giving a total of 473

observations. Each stock is included in the sample only once, the bankrupt stocks being

sampled just prior to the bankruptcy event and the non-bankrupt stocks at the end of the

sample period. The relatively low number of observed bankruptcies during the sample period

is a function of the low natural rate of bankruptcy across all companies, the fact that we

include only larger companies (for which the natural rate of bankruptcy is even lower) and

the exclusion of Financials sector companies from our sample.5

The financial data for bankrupt companies is measured using the last reported annual

financial statement available at least two months prior to the earlier of delisting or entering

administration. The two month window ensures that the financial data had time to be

communicated to the market prior to either entering administration or delisting. The market

data is also measured two months prior to the earlier of the date of delisting or the date of

entering administration. Bankrupt companies are only included in the sample if the period

between delisting and subsequently entering administration was not more than three years.

The financial data for non-bankrupt companies is measured using the last financial statement

5
According to Dun and Bradstreet, the rate of business failures in the US remained consistently below 0.75%
over the period 1934-81. Bickerdyke, Lattimore and Madge (2001) find that the rate of failure in
unincorporated businesses in Australia over the period 1929 to 1999 was never greater than 0.5% per year.
Assuming an average rate of 0.6% for the All Ordinaries population, a total of approximately 50 bankruptcies
would be expected over the period 1988-2008.

Electronic copy available at: https://ssrn.com/abstract=1293913


prior to the end of the sample period (May 2008). Market data for non-liquidated companies

is measured at the end of the sample period.

The bankrupt companies were identified using data files supplied by both the Australian

Stock Exchange (ASX) and the Australian Securities and Investments Commission (ASIC).

Table A1 in the Appendix documents the process by which the final list of bankrupt stocks

was constructed. All data is winsorized to lie within three standard deviations of the mean.

Companies with missing financial data were deleted from the sample.6

Table 1 below gives summary statistics for the explanatory variables used in four model

specifications we examine. See Table A2 in the Appendix for a description of these

variables.

Table 1: Summary Statistics

Variable Mean Median Min Max Std Dev


2.988 2.988 2.661 3.238 0.092
0.021 0.011 -0.066 0.407 0.038
-7.808 -7.957 -12.437 -3.181 1.523
0.039 0.071 -2.491 1.512 0.308
0.012 0.050 -2.932 1.456 0.329
-0.248 0.049 -23.006 0.647 2.031
0.472 0.476 0.002 4.378 0.411
0.124 0.098 -1.650 0.984 0.271
3.206 0.635 0.000 532.922 34.597
-0.013 0.120 -5.947 5.979 1.055
19.389 2.501 0.022 264.098 51.220
0.747 0.416 0.000 20.493 1.544
0.015 0 0 1 0.121
0.214 0 0 1 0.410
0.054 0.067 -1 1 0.508
-0.801 -11.881 -98.934 214.610 57.504
12.091 10.317 1.964 44.315 6.971

6
For example, a number of non-liquidated companies had listed within 12 months of the end of sample period
cut-off date – these companies were eliminated from the sample.

Electronic copy available at: https://ssrn.com/abstract=1293913


2.2. Model Specifications

The three previously published models we examine are that of Altman (1968), Ohlson

(1980) and Shumway (2001). The first two employ only accounting ratios, while the third

additionally employs market information – see Table 2 for detail on the explanatory variables

used in each model. The remainder of this section outlines the specification of the Alternate

model, which we subsequently compare to the above three models

The specification of the Alternate conditional logit model is described by Equations (1)

and (2), where the dependent variable takes on a value of zero for non-bankrupt stocks and

one for bankrupt stocks:

While most of the explanatory variables above have been considered in previous

bankruptcy prediction models, none have used these variables in precisely the same

combination as used here. The variable does not appear to have previously been used

in the published literature. Hillegeist, Keating, Cram and Lundstedt (2004) assert that the

Merton option pricing-based default likelihood indicator (DLI) is more informative than

accounting ratio-based models. Like Campbell et al. (2008), we find that adding the DLI

does not significantly improve the statistical fit or predictive power of the Alternate model.

This is likely due to the Alternate model, like Campbell et al. (2008), employing market

information (in particular volatility) in addition to accounting ratios.

Electronic copy available at: https://ssrn.com/abstract=1293913


Because we use the financial and market data up to two months prior to the earlier of

delisting or entering administration, it is natural to think of the four models having a

prediction horizon in the range of 2-14 months.7 While this is a shorter prediction time

horizon than has been used in some other studies (12 months or more), we argue that it is

closer to the horizon that investors actually use.

3. Model Estimation and Prediction Accuracy

3.1. Estimation

The estimated coefficients, associated t-statistics and goodness of fit of all four models

are reported in Table 2. Both the Psuedo-R2 and Bayesian Information Criterion (BIC) show

that the Alternate model has a better goodness of fit than the Altman, Ohlson and Shumway

specifications.8 The signs of all parameter estimates are consistent with that predicted by

theory. While a large proportion of the parameter estimates in the Altman and Shumway

models are statistically significant (at the 5% probability level), all estimates are statistically

significant in the Alternate model with all but one being also significant at the 1% probability

level. Thus, the Alternate is the preferred model on the basis of statistical properties.

Sensitivity analysis shows that for the Alternate model, a 1% increase in the and

variables has the largest positive impact on the predicted probability, while a 1%

increase in the has the largest negative impact on the predicted probability.9

7
The market data is always measured two months prior to bankruptcy, while the financial data can be measured
up to 14 months prior to bankruptcy due to being annual in frequency.
8
A better fit is associated with a higher Psuedo-R2 and a lower BIC. While not reported, we also found that the
Alternate model specification has a better goodness of fit than Izan (1984) which was also estimated on
Australian data.
9
These sensitivities are calculated at the mean of the dependent and explanatory variables across the entire
sample.

Electronic copy available at: https://ssrn.com/abstract=1293913


Table 2: Comparison of Statistical Fit of Models

Alternate Altman Ohlson Shumway


44.43** -2.11** 11.02** -11.04**
(4.30) (9.05) (3.75) (6.99)
-16.32** -0.71**
(4.60) (4.66)
-32.61**
(3.12)
-4.54** -3.68**
(3.85) (2.96)
-0.03** -0.03**
(2.72) (2.83)
2.26* 2.14**
(2.17) (2.77)
-0.01** -0.01
(2.77) (1.92)
0.11** 0.08**
(3.46) (3.42)
-1.20 -2.60*
(1.21) (2.56)
-1.76**
(4.09)
0.06
(0.37)
1.96*
(2.03)
-0.03
(1.15)
-2.33* -2.41*
(2.35) (2.80)
0.18
(0.88)
0.44
(1.16)
0.97
(0.63)
-0.09
(0.16)
-0.77**
(4.57)
Psuedo-R2 0.470 0.304 0.318 0.398
BIC 0.457 0.541 0.584 0.479
* Denotes significant at the 5% probability level
** Denotes significant at the 1% probability level

10

Electronic copy available at: https://ssrn.com/abstract=1293913


3.2. Prediction Accuracy

The prediction accuracy of four models is compared in Table 3, where for each model we

measure the proportion of stocks in the top one and two deciles of predicted probabilities that

are actually bankrupted. Clearly, regardless of whether one focuses on the top one or two

deciles of probability predictions, the Alternate model generates a higher hit rate than the

other three models. Thus, on the basis of both statistical fit and prediction power, the

Alternate model dominates the other three models.

Table 3: Comparison of Prediction Accuracy of Models

Bankrupt Stocks as % of Top n Deciles


n Alternate Altman Ohlson Shumway
1 66.0 55.3 59.6 63.8
2 87.3 65.9 63.9 76.6

We now examine the prediction accuracy of the Alternate model in more detail, looking

at how the hit rate varies as a function of the chosen probability cut-off. The analysis is

presented in Table 4. A probability cut-off of 0.5 means that stocks with predicted

probabilities greater than 0.5 are identified as bankruptcy predictions. The hit rate refers to

the proportion of these predictions that were in fact correct (i.e., stocks that did experience

bankruptcy). The # Correct/Total # ratio gives the component numbers that make up the hit

rate. Again using a cut-off of 0.5, of the 29 stocks predicted to be high risk, 23 of these

turned out to be liquidated stocks. Type I errors (false negatives) refer to stocks that were not

bankruptcy predictions but which in fact did experience bankruptcy, while Type II errors

(false positives) refer to stocks that were bankruptcy predictions but which in fact did not

experience bankruptcy. Type I errors can be expected to be more costly than Type II errors

since investing in a stock which is about to enter bankruptcy is likely to have a large negative

impact on performance, whereas not investing in a stock which in fact does not enter

11

Electronic copy available at: https://ssrn.com/abstract=1293913


bankruptcy is unlikely to have much impact on portfolio performance. The prediction

accuracy of the model is the weighted sum of the proportion of correct predictions for

bankrupt and non-bankrupt stocks.

Table 4: Alternate Model Prediction Accuracy

Probability Cut-off
(%) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Hit rate 43 62 71 73 79 85 88 90 94
# Correct /Total # 45/104 41/66 37/52 29/40 23/29 22/26 21/24 18/20 16/17
Prediction accuracy 87 93 94 93 93 93 93 93 93
Type I errors 8 16 24 41 53 55 57 63 67
Type II errors 14 6 4 3 1 1 1 0 0

For any probability cut-off above 0.3, over 70% of the stocks identified by the Alternate

model as bankruptcy predictions are correct. Using the probability cut-off that maximizes the

hit rate (0.9), the model accurately identifies 16 of the 17 bankruptcies predicted, implying a

hit rate of 94%.

4. Is there a Distress Effect?

We now examine whether there is any compensation for holding bankruptcy risk.

Specifically, we analyze both the excess returns and risk-adjusted alpha associated with

portfolios that are long high bankruptcy risk stocks and short low bankruptcy risk stocks.

We now apply the Alternate model to monthly data for all stocks in the S&P/ASX 300

universe over the period January 1993 to December 2007 (excluding Financials sector

stocks).10 Each month the predicted probabilities of bankruptcy are ranked from highest to

10
The S&P/ASX 300 is a subset of the ASX All Ordinaries universe. By focusing here on this smaller universe
we maximize the relevance of the results to institutional investors since the largest fund managers often find
stocks outside of the S&P/ASX 300 universe to be uninvestible. Our analysis, not reported here, showed that

12

Electronic copy available at: https://ssrn.com/abstract=1293913


lowest and value-weighted portfolios are formed on the basis of these rankings. Denote by

P1 the highest risk quintile portfolio (80-100) and P5 the lowest risk quintile portfolio (0-20).

We focus on the returns generated by a zero investment portfolio that is long P1 and short P5.

In some instances we also present the results for a portfolio that is long P1 in order to shed

light on whether the long or short side exposures are driving the reported results for the long

P1/short P5 portfolio.

Portfolio returns are measured over a 6-month holding period.11 For a portfolio which is

formed in January 1993, the bankruptcy risk of that portfolio is measured in January 1993 (at

the time of portfolio formation) and the associated returns are measured over the January

1993 to July 1993 period. In the case of stocks that were delisted (for reasons of bankruptcy

or otherwise) while being held in a portfolio, the missing months of return data were replaced

by the yield on cash (the 90-day Bank Bill rate) during that period. That is, the return from

liquidating the delisted stock is held in cash rather than reinvested in the remaining stocks in

the portfolio. Holding cash may often be preferable to reinvesting in the portfolio due to the

transactions associated with the latter.

We now measure the excess return and the risk-adjusted alpha associated with two

portfolios: the zero investment long P1/short P5 portfolio, and the long P1 portfolio. In the

case of the long P1 portfolio, excess return is measured relative to the S&P/ASX 300

benchmark. We estimate two different risk-adjusted alphas in the context of the Carhart

(1997) 4-factor model: one where the value factor is defined in terms of the book-to-price

(BP) ratio, the other where the value factor is defined in terms of the earnings-to-price (EP)

the bankruptcy model retained a high level of predictive accuracy over the S&P/ASX 300 universe, with the hit
rate being at least 50% for all probability cut-offs above 0.3.
11
Given our use of monthly data, this results an overlapping structure to the returns data. The risk-adjusted
alphas are estimated using GMM and employing the Newey-West correction for the induced autocorrelation.

13

Electronic copy available at: https://ssrn.com/abstract=1293913


ratio.12 The rationale for examining two formulations of the 4-factor model is that value is

widely characterized in terms of both the BP and (particularly by practitioners) EP ratios.

Table 5 shows that the long P1/short P5 portfolio generates a negative excess return and

also negative alpha, although none of these estimates are statistically significantly different

from zero. The same is true for the long P1 portfolio. We conclude that high bankruptcy risk

stocks fail to generate positive returns relative to either low bankruptcy risk stocks or the

benchmark.13

Table 5: Bankruptcy Risk Premium

Long P1/
Value Factor Long P1
Short P5
-1.96 -1.99
Excess return (%)
(-0.11) (-0.12)

BP -1.00 -1.62
4-factor alpha (%)
(-0.70) (-1.04)

EP -1.08 -0.35
4-factor alpha (%)
(-0.70) (-0.24)
* Statistically significant at the 5% probability level.
** Statistically significant at the 1% probability level.

Figure 1 provides more detail on how excess returns and risk-adjusted alpha vary with

bankruptcy risk. The 0-5 (95-100) category on the far left (right) of the horizontal axis is the

portfolio of stocks consisting of the lowest (highest) 5% of bankruptcy risk probabilities.14

The excess return is close to zero for all but the 80-90 portfolio, for which it is -3%. In all

cases these estimates are not statistically significantly different from zero. In contrast, the

risk-adjusted alphas show a clear downward trend as bankruptcy risk increases, except in the

12
See the Appendix for details on the construction of the factors in the 4-factor model.
13
While not reported here, we also examined equal-weighted portfolios and found the results to be similar to the
value-weighted results. All estimates of excess returns and risk-adjusted alpha were insignificantly different
from zero.
14
These two portfolios contain 5% of stocks by name.

14

Electronic copy available at: https://ssrn.com/abstract=1293913


extreme highest risk portfolio (95-100).15 The only portfolios for which the alpha estimates

are statistically significant are the 80-90 and 90-95 portfolios, for which the alphas are

strongly negative in both factor models. Thus, the lack of significance of the risk-adjusted

alphas in Table 5 is largely due to the 5% of stocks with the very highest bankruptcy risk –

the remaining stocks in P1 (80-95) generate returns that are negative and statistically

significant.

Figure 1: The Effect of Bankruptcy Risk on Excess Return and Alpha

0
0-5 5-10 10-20 20-40 40-60 60-80 80-90 90-95 95-100
Alpha (%)

-2

-4

-6

-8

-10
Excess return 4-factor alpha (BP) 4-factor alpha (EP)

Table 5 indicates that there is no distress effect in the S&P/ASX 300 universe. But are

there pockets of stocks for which there is a premium for holding high bankruptcy risk stocks?

We now sort the universe by size and value to determine whether certain segments of the

market display a distress effect. Since it is widely believed that the size and value effects in

part reflect a distress premium, it is natural to explore the size and value dimensions for such

pockets.

Table 6 shows the excess returns and risk-adjusted alpha for long P1/short P5 portfolios

constructed on the S&P/ASX 100 (“Big”) stocks and S&P/ASX 300 ex 100 (“Small”) stocks.

15
The large downward spike in the estimated alpha for the 90-95 portfolio is due to a sharp increase in the size
factor loading for this portfolio.

15

Electronic copy available at: https://ssrn.com/abstract=1293913


The excess returns are lower for the Small portfolio than the Big portfolio, although neither is

statistically significantly different from zero (or each other). In contrast, the risk-adjusted

alphas are much more negative in the Small portfolio than the Big portfolio and statistically

significantly different from each other. As conventional wisdom suggests that distress risk is

concentrated (and most extreme) in the small stocks, it seems plausible to conjecture that

long/short returns will be higher in the Small portfolio than the Big portfolio. Table 6 shows

that this is clearly not the case, indicating that either the initial premise is false (i.e., small

stocks are not more distressed) or that there is no compensation for bearing the risk

associated with the distress. Given that we find size to be an important predictor of

bankruptcy in section 2, our results favor the second explanation.

Table 6: Returns to Big versus Small Stocks – Long P1/Short P5

Value
Big Small
Factor
-0.60 -2.79
Excess return (%)
(-0.04) (-0.16)

BP -1.48 -5.03**
4-factor alpha (%)
(-1.53) (-2.70)

4-factor alpha (%) EP -0.75 -4.29*


(-0.61) (-2.43)
* Statistically significant at the 5% probability level.
** Statistically significant at the 1% probability level.

The excess returns and alpha obtained from portfolios formed on the 30% of stocks with

the highest book-to-price (BP) scores (“High”) and the 30% of stocks with the lowest BP

scores (“Low”) are shown in Table 7. Once again, the excess returns across both High and

Low portfolios are not statistically significantly different from zero (or each other). The risk-

adjusted alphas are negative for the High value portfolios and positive for the Low value

portfolios, with the estimates being statistically significantly different in the case of the EP

16

Electronic copy available at: https://ssrn.com/abstract=1293913


value factor. This result is striking as it shows that to the extent distress risk is concentrated

in high value stocks, there is no compensation for bearing this risk.

Table 7: Returns to High versus Low Value Stocks – Long P1/Short P5

Value
High Low
Factor
Excess return (%) -3.14 1.61
(-0.12) (0.07)

BP -2.90 0.33
4-factor alpha (%)
(-1.06) (0.19)

EP -4.65* 2.97
4-factor alpha (%)
(-2.16) (1.51)
* Statistically significant at the 5% probability level.
** Statistically significant at the 1% probability level.

Our analysis in the following section casts doubt on whether distress risk is in fact

concentrated in high (rather than low) value stocks, thus partially explaining the results in

Table 7. Our results indicate the absence of a distress effect in the S&P/ASX 300 universe.

In fact, the excess returns and risk-adjusted alphas are lowest in those portfolios where

distress would be expected to be highest – the Small size portfolios and the High value

portfolios.

5. Is Value a Proxy for Bankruptcy Risk?

It has long been suggested that the value factor is a proxy for financial distress risk (Fama

and French, 1995). Our finding that there is no systematic evidence of compensation for

bearing bankruptcy risk together with the existence of a (positive) value premium in the

17

Electronic copy available at: https://ssrn.com/abstract=1293913


Australian market suggests strongly at the outset that the value factor is not a proxy for

financial distress in Australia.16

5.1. Sorts Analysis

Using the same sorting structure as in the previous section, Figure 2 shows how average

value varies in moving from low to high bankruptcy risk portfolios. Measuring value in

terms of the BP ratio, we see that value increases with bankruptcy risk up to the highest risk

portfolio (95-100), whereupon it falls sharply. Turning to the EP ratio, no relationship exists

between value and bankruptcy risk. Thus, the correlation between value and bankruptcy risk

is weak, lacking consistency across value metrics and across the risk spectrum.

Figure 2: Relationship between Value and Bankruptcy Risk

0.7 0.07

Earnings-to-Price ratio
0.6 0.06
Book-to-Price ratio

0.5 0.05
0.4 0.04
0.3 0.03
0.2 0.02
0.1 0.01
0 0.00

Avg BP Avg EP

Another perspective on the relationship between bankruptcy risk and the value factor is

obtained from comparing the average bankruptcy risk in value sorted portfolios. Table 8

shows that for low bankruptcy risk stocks (0-20) the difference in risk between High and Low

value portfolios is small regardless of whether the BP or EP ratios are used to measure value.
16
While it is equally valid to ask whether the size factor is a proxy for bankruptcy risk, by construction we
know that size will be strongly related to bankruptcy risk in our model.

18

Electronic copy available at: https://ssrn.com/abstract=1293913


For high bankruptcy risk stocks, there is little difference in the distress risk between High and

Low value portfolios when using the BP ratio, but a large difference when using the EP ratio.

In both cases, however, the risk is higher in the Low value portfolio than the High value

portfolio. First, these results indicate that value and bankruptcy risk are not strongly related.

Second, they go some way to explaining the results in Table 7 which show that portfolios of

Low value stocks yield substantially higher returns than portfolios of High value stocks,

particularly when value is measured in terms of the EP ratio. The High value portfolio

(unexpectedly) has lower bankruptcy risk than the Low value portfolio.

Table 8: Average Bankruptcy Risk in Value Sorted Portfolios

0-20 80-100
BP High 0.01 0.38
Low 0.00 0.44
EP High 0.00 0.19
Low 0.00 0.70

Another approach to understanding whether high bankruptcy risk stocks tend to also be

higher value stocks involves comparing the value effect (the difference between returns to

high and low value portfolios) in high versus low bankruptcy risk portfolios. The range of

bankruptcy probabilities is much larger in high risk portfolios than low risk portfolios – see

Figure 3 which shows for each level of bankruptcy risk the average probability of bankruptcy

and the standard deviation. Thus, if the value effect is largely due to bankruptcy risk then it

should be much stronger in high bankruptcy risk portfolios.

19

Electronic copy available at: https://ssrn.com/abstract=1293913


Figure 3: Average Bankruptcy Risk and its Dispersion within Portfolios

0.9 0.20
Average Probability of

Standard Deviation of
0.8 0.18
Liquidation 0.7 0.16
0.14

Probability
0.6
0.12
0.5
0.10
0.4
0.08
0.3 0.06
0.2 0.04
0.1 0.02
0.0 0.00

Avg Prob. Std Dev

Figure 4 shows the difference between the returns to High and Low value portfolios at

various levels of bankruptcy risk. Under both BP and EP measures of value, the High - Low

return is no higher for the highest bankruptcy risk portfolio (80-100) than for the lowest risk

portfolio (0-20). More generally, there appears to be no systematic relationship between the

size of the value effect and the level of bankruptcy risk. Thus, taken together Figures 3 and 4

suggest that value is not highly correlated with bankruptcy risk.

Figure 4: The Value Effect as a Function of Bankruptcy Risk

5
High -Low Return (%)

4
3
2
1
0
-1 0-20 20-40 40-60 60-80 80-100

-2
BP EP

20

Electronic copy available at: https://ssrn.com/abstract=1293913


5.2. Factor Model Analysis

An alternative approach to understanding the relationship between bankruptcy risk and

value involves examining how the loadings in a factor model respond to changes in

specification of that model. Specifically, we now examine how the returns to portfolios

screened on bankruptcy risk correlate with returns to portfolios screened on value within the

context of the 4-factor model. In characterizing value, henceforth we focus only on the BP

ratio given the observed similarity of our findings above using the BP and EP ratios.

Table 9 shows that the returns from the long P1/short P5 portfolio load most heavily on

the market and size factors. Not only are the coefficients on the market and size factors much

higher in magnitude than for the other two factors, but they are statistically significant from

zero at a much lower probability level (i.e., the t-statistics are much higher). This analysis

thus shows that bankruptcy risk displays a low correlation with the value factor (compared to

the market and size factors).

Table 9: Factor Loadings in the 4-Factor Model – Long P1/Short P5

Loading
Market 1.19**
(12.54)
Size 1.73**
(13.33)
Value (BP) 0.33**
(3.75)
Momentum -0.34**
(-3.32)
* Statistically significant at the 5% probability level.
** Statistically significant at the 1% probability level.

Figure 5 below provides more detail as to how the factor loadings vary from very low to

very high bankruptcy risk portfolios.17 The market and momentum factor loadings are

17
The results for equal-weighted portfolios are very similar.

21

Electronic copy available at: https://ssrn.com/abstract=1293913


essentially constant across the risk spectrum, while the size factor trends up strongly. The

value factor loading trends up gently, suggesting value and bankruptcy risk are somewhat

positively correlated. Figure 5 reaffirms the insight from Table 9 that bankruptcy risk is most

strongly correlated with the size factor loading.

Figure 5: The Effect of Bankruptcy Risk on Factor Loadings

3.0
2.5
2.0
Alpha (%)

1.5
1.0
0.5
0.0
-0.5 0-5 5-10 10-20 20-40 40-60 60-80 80-90 90-95 95-100
-1.0
Market Size Value Momentum

Following Vassalou and Xing (2004), denote by the aggregate probability of

bankruptcy in the S&P/ASX 300 population at time as determined by the model described

in sections 2 and 3. Let , so is the (scaled) six-month change

in the aggregate probability of bankruptcy over the portfolio holding period. We now add

as a fifth factor to the 4-factor model in order to directly capture changes in distress. If

the value factor is acting as a proxy risk for bankruptcy risk, then adding as an

additional factor should have a discernible impact on the loading on the value factor.

Table 10 shows that the loading on is negative but not statistically significant.18

The column labeled Δ Loading compares the estimate in Table 10 to that in Table 9 in order

to show the impact on the 4-factor loadings of adding the distress factor. Adding is

shown to have a statistically significant negative impact on the value factor loading, implying

18
Vassalou and Xing (2004) also find distress risk to be negatively priced.

22

Electronic copy available at: https://ssrn.com/abstract=1293913


that the value factor does in part reflect distress risk. But note that the value factor loading

remains positive and statistically significantly different from zero even in the presence of

. We thus conclude that there is significant information in the value factor that is not

reflected in the factor. In other words, the value factor cannot be viewed as essentially

a proxy for distress risk.

Table 10: Factor Loadings in the Augmented 4-Factor Model

Loading Δ Loading
Market 1.20** 0.01
(12.27)
Size 1.63** -0.09
(11.97)
Value (BP) 0.17* -0.16*
(2.17)
Momentum -0.39** -0.05
(-3.75)
Distress -5.42
(-1.33)
* Statistically significant at the 5% probability level.
** Statistically significant at the 1% probability level.

We have repeated the above analysis using a default likelihood indicator based on

Merton‟s (1974) option pricing model to construct the factor – this is the same approach

that Vassalou and Xing (2004) used. The results provide even stronger evidence in favour of

rejecting the hypothesis that the value factor is simply a proxy for the compensation due to

bearing distress risk. We find that the value factor loading in the augmented 4-factor model

actually increases and is more strongly statistically significant than reported in Table 10.19

As a final test of whether the value effect is mostly compensation for distress risk, we

now regress the residuals from the 4-factor model against the factor (calculated using

our bankruptcy model). The rationale here is that the residuals are in principal orthogonal to

19
Specifically, the value factor loading increases from 0.33 to 0.49 with a t-statistic of 5.49.

23

Electronic copy available at: https://ssrn.com/abstract=1293913


the four factors. Thus, if the value factor primarily represents compensation for distress risk,

the residuals should exhibit a statistically weak relationship with the distress factor, .

The results provide further evidence that the value factor captures information that is

materially different from distress risk. The estimated loading on is not statistically

significant, indicating that there is little additional information in the distress factor that is not

already captured in the value factor.20

In summary, the evidence presented in this section argues strongly against the

proposition that the value factor is simply a proxy for distress risk. There is considerable

additional information embedded in the value factor that is independent of distress risk.

6. Conclusions

We have constructed a bankruptcy prediction model for the Australian market and shown

that it has strong statistical and predictive properties. Our analysis establishes two key results

for the Australian market. First, high bankruptcy risk stocks do not outperform low

bankruptcy risk stocks in value-weighted portfolios. In fact, in those parts of the market

commonly presumed to be most exposed to bankruptcy risk – small cap and high value stocks

– high bankruptcy risk stocks significantly underperform low risk stocks. Second, the value

factor is not essentially a proxy for distress risk. This implies that either the value premium

reflects compensation for other kinds of risks, or that it is rooted in behavioral inefficiencies.

Our findings provide out-of-sample support for the conclusions of Campbell et al. (2008) and

others based on US data.

One area for further research is to test more recent risk-based theories of the source of the

value premium, such as those put forward by Zhang (2005) and Novy-Marx (2006). If these

20
The estimated loading is -3.68 with t-statistic -0.97.

24

Electronic copy available at: https://ssrn.com/abstract=1293913


explanations can also be rejected, then behavioral theories such as those proposed by

Lakonishok, Schleifer and Vishny (1994) will need to be given more serious consideration as

the source of the value premium.

25

Electronic copy available at: https://ssrn.com/abstract=1293913


Appendix

A.1. Construction of Bankruptcy Dataset

The screening process we followed in identifying the set of bankrupt stocks is described

in Table A1 below.

Table A1: Screens Employed in Constructing Bankruptcy Dataset

Action Screen Source


Initial All stocks in the ASX All Ordinaries that were delisted over the
Universe period 1988-2008, ASX
Delistings due to takeovers, mergers, schemes of arrangement and
Exclude non-distress reasons. ASIC
Stocks satisfying ASIC Administration Type “Members Voluntary
Exclude Liquidation”. ASIC
Stocks which lodged a “Declaration of Solvency” with ASIC at the
Exclude time of entering administration. ASIC
Stocks for which the date of entering administration was more than
Exclude three years after delisting. delisted.com.au
Exclude Stocks in GICS 4010, 4020, 4030
Exclude Stocks with missing financial data Aspect

All stocks remaining at the end of the screening process had ASIC Status of “Under

External Administration and/or Controller Appointed” or “Deregistered”. The screening

process described in Table A1 ensured that only bankruptcies that are associated with

genuine financial distress were included in the sample. For example, the class of “Members

Voluntary” liquidations is not usually associated with financial distress, but rather the

discretionary winding up of a company structure that is no longer needed. The Banking,

Diversified Financials and Insurance sectors (GICs 4010, 4020, 4030) were excluded from

the analysis for the usual reason that financial ratios for companies in these sectors are

difficult to interpret and compare to non-Financial companies.

26

Electronic copy available at: https://ssrn.com/abstract=1293913


A.2. Definition of Explanatory Variables

Each of the explanatory variables listed in Table 1 are defined in Table A2.

Table A2: Description of Explanatory Variables

Symbol Financial Item


21
, where total assets is deflated using Australian GDP deflator
Change in over the past 1 year
Market Capitalization / Total Market Cap of All Ordinaries benchmark
Earnings Before Interest & Tax / Total Assets
Net Profit After Tax / Total Assets
Retained Earnings / Total Assets
Total Liabilities / Total Assets
Working Capital / Total Assets
Current Liabilities / Current Assets
Operating Cash Flow / Total Liabilities
Market Value of Equity / Total Liabilities
Sales / Total Assets
Equal to 1 if TL>TA, otherwise 0
Equal to 1 if NPAT>0, otherwise 0
Change in over the past 1 year22
Excess Return relative to All Ordinaries benchmark over past 1 year
Volatility of over past 1 year

A.3. Carhart 4-Factor Model

The 4-factor model of Carhart (1997) involves regressing portfolio returns against the

market return, the size factor, the value factor and the momentum factor. Returns are

measured net of the risk-free return. The market return is the S&P/ASX 300 benchmark

accumulation return. The risk-free rate is the Australian Government 90-day Bank bill rate.

The estimated equation is

21
Size is often measured as , though we find the log of this quantity gives a better model fit.
22
, where the subscript refers to the 1 year prior
observation.

27

Electronic copy available at: https://ssrn.com/abstract=1293913


denotes the total return from the portfolio, denotes the risk-free return, denotes

the market total return, denotes the size factor, denotes the value factor, and

denotes the momentum factor. Consistent with our approach to model construction,

we exclude the GICS 4010, 4020 and 4030 sectors in constructing the factor returns. Factor

returns are based on portfolios which are formed each month and have a 6-month holding

period (weights are rebalanced on the same 6-month frequency).

The factors are constructed using an independent double-sort methodology similar to that

used by Fama and French. Stocks are sorted into two size groups: the S&P/ASX 100 stocks

and the S&P/ASX 300 ex 100 stocks, denoted B(ig) and S(mall) respectively. Independently,

stocks are sorted into three value groups based on the book-to-price ratio, using the 30th and

70th percentile cut-offs. Stocks above the 70th percentile are denoted H(igh) and stocks below

the 30th percentile are denoted L(ow). Stocks between the 30th and 70th percentiles are

denoted M(edium). Together, this generates six different portfolios: BH, BM, BL, SH, SM

and SL. The factor is constructed as

, where is the return on the stocks in the BH portfolio. The factor is constructed

as .

The momentum factor is constructed in an analogous way to the value factor above,

using an independent double-sort methodology over 12-month momentum and size, where

size is sorted as above and the 30th and 70th percentile cut-offs are used to sort stocks into

three momentum categories. Stocks above the 70th percentile are denoted U(p) and stocks

below the 30th percentile are denoted D(own). The factor is constructed as

28

Electronic copy available at: https://ssrn.com/abstract=1293913


Appendix

Agarwal, V., Poshakwale, S., 2008. Does distress risk explain size and book-to-market

effects? Working paper, Cranfield School of Management.

Altman, E.I., 1968. Financial ratios, discriminant analysis, and the prediction of corporate

bankruptcy, Journal of Finance 23, 589.609.

Avramov, D., Chordia, T., Jostova, G., Philipov, A., 2007. Momentum and credit rating,

Journal of Finance 62, 5, 2503.2520.

Beaver, W.H., 1966. Financial ratios as predictors of failure, Journal of Accounting Research

4, 71.111.

Bickerdyke, I., Lattimore, R., Madge, A., 2000. Business failure and change: An Australian

perspective, Productivity Commission Staff Research Paper, AusInfo, Canberra.

Campbell, J.Y., Hilscher, J., Szilagyi, J., 2008. In search of distress risk, forthcoming in

Journal of Finance, December issue.

Carhart, M., 1997. On persistence in mutual fund performance, Journal of Finance 52, 57.82.

Castagna, A.D., Matolcsy, Z.P., 1981. The prediction of corporate failure: Testing the

Australian experience, Australian Journal of Management 6, 23.50.

Chen, N.F., Zhang, F., 1998. Risk and return of value stocks, Journal of Business 71,

501.535.

Dichev, I.D., 1998. Is the risk of bankruptcy a systemic risk? Journal of Finance 53,

1131.1147.

29

Electronic copy available at: https://ssrn.com/abstract=1293913


Fama, E.F., French, K.R., 1993. Common risk factors in returns on stocks and bonds, Journal

of Financial Economics 33, 3.56.

Fama, E.F., French, K.R., 1995. Size and book-to-market factors in earnings and returns,

Journal of Finance 50, 131.155.

Fama, E.F., French, K.R., 1996. Multifactor explanations of asset pricing anomalies, Journal

of Finance 51, 55.84.

Ferguson, M.F., Shockley, R.L., 2003. Equilibrium “anomalies”, Journal of Finance 58,

2549.2580.

Garlappi, L., Shu, T., Yan, H., 2008. Default risk, shareholder advantage and stock returns,

forthcoming in The Review of Financial Studies.

Ghargori, P., Chan, H., Faff, R., 2007. Investigating the performance of alternative default

risk models: Option-based versus accounting-based approaches, Australian Journal of

Management 31, 2, 207.234.

Griffin, J., Lemmon, M., 2002. Does book-to-market equity proxy for distress risk? Journal

of Finance 57, 2317.2336.

Hillegeist, S.A., Keating, E.K., Cram, D.P., Lundstedt, K.G., 2004. Assessing the probability

of bankruptcy, Review of Accounting Studies 9, 5.34.

Izan, H.Y., 1984. Corporate distress in Australia, Journal of Banking and Finance 8, 303.320.

Lakonishok, J, Shleifer, A., Vishny, R.W., 1994. Contrarian investment, extrapolation, and

risk, Journal of Finance 49, 1541.1578.

Lincoln, M., 1984. An empirical study of the usefulness of accounting ratios to describe

levels of insolvency risk, Journal of Banking and Finance 8, 321.340.

30

Electronic copy available at: https://ssrn.com/abstract=1293913


Merton, R.C., 1974. On the pricing of corporate debt: The risk structure of interest rates,

Journal of Finance 28, 449.470.

Novy-Marx, R., 2006. Investment-cash flow sensitivity and the value premium, Working

paper, University of Chicago Graduate School of Business.

Ohlson, J., 1980. Financial ratios and the probabilistic prediction of bankruptcy, Journal of

Accounting Research 19, 109.131.

Shumway, T., 2001. Forecasting bankruptcy more accurately, Journal of Business 74, 1,

101.124.

Tanthanongsakkun, S., Pitt, D., Treepongkaruna, S., 2008. A comparison of corporate models

in Australia: the Merton vs accounting-based models, forthcoming in Asia Pacific

Journal of Risk and Insurance.

Vassalou, M., Xing, Y., 2004. Default risk in equity returns, Journal of Finance 59, 831.868.

Zaretzky, K., Zumwalt, J.K., 2007. The relation between distress risk, B/M, and return

premium, Managerial Finance 33, 10, 788.797.

Zhang, L., 2005. The value premium, Journal of Finance 60, 67.103.

31

Electronic copy available at: https://ssrn.com/abstract=1293913

You might also like