Bankruptcy & Reorganization – Professor Altman 12/15/2014
1
STAR Model: Predicting ST* Companies in China Pushkar Bodas, Shyam Lingam, Shuyu Wang, Wendy Yang
Introduction
China’s securities market is an attractive investment opportunity. According to World
Federation of Exchanges, Shanghai and Shenzhen stock exchanges are ranked among the top 10
in terms of market cap and trade volume. Previously, foreign investors had minimal access due
to China’s regulations limiting inflows of foreign capital. Institutional investors were subject to
strict government-set quotas, and retail investors were excluded. Shanghai-Hong Kong Stock
Connect was established on 17 November 2014 and allows institutional and retail investors to
directly trade China-listed stocks, which caps at $2.11 billion per day. This represents a much
more open trading platform, available to international investors all over the world including
investors investing in Chinese distress stocks.
Companies are more likely to be distressed due to GDP deceleration and a decrease in
China’s economic growth. Companies in various industries, such as four solar-cell companies
between 2013 and 2014, have been listed in other countries as filing for bankruptcy due to
anemic domestic demands, and the Chinese government had to bail out several public companies
as a result of bond defaults. State-owned banks have increasingly high debt ratios at levels
reminiscent of the country’s last major debt crisis. As of 5 December 2014, China has 35
companies designated as Special Treatment, or ST, an identification of financial distress. ST
companies represent 1.38% of all the listed companies in the Shanghai and Shenzhen stock
exchanges and change as current ST companies recover and new companies become financially
distressed.
Every major stock exchange has its own set of delisting rules, which is a fundamental
component of a well-functioning securities market. When a public firm becomes financially
distressed, delisting rules ensure that the market recycles and redirects capital into more
promising companies and improve operating efficiencies. Corporate governance of public firms
better protect investors, as regulators are able to weed out poor performing companies and limit
systematic risks. The Chinese government is embracing long-overdue economic reforms; it
recently announced commitment to establish a deposit insurance system and frequently
emphasized the significance of easing regulations to file for bankruptcy in the nation’s economic
agenda.
Regulations of distressed firms and delisting standards have changed since trading began
on the Shanghai and Shenzhen Stock Exchanges in 1990. ST designation was established in
March 1998, the first law regulating distressed firms. Companies are classified as ST stock if (1)
the firm had negative earnings over the past two accounting years, or if (2) the firm’s
shareholders equity is lower than its registered capital in the most recent accounting year1. In
1999 and 2001, the government defined PT designation, a transfer warning, for ST companies
that did not have financial improvement in year one. A formal delisting process will begin for
ST companies that do not have financial improvement in year two2. In 2001, China had its first
delisted company. China Securities Regulatory Commission (CSRC), the governmental agency
responsible for regulating the exchanges, publically solicited advice to reform current delisting
rules in July 2014.
1 http://www.szse.cn/szseWeb/FrontController.szse?ACTIONID=15&ARTICLEID=883&TYPE=0 2 Altman
Bankruptcy & Reorganization – Professor Altman 12/15/2014
2
Chinese regulations changed in 2003, and the previous ST and PT standards were
eliminated. The publication of “Delisting Risk Warning Rules” stipulates that companies at risk
of being delisted from the Shanghai and Shenzhen Stock Exchanges will be designated as ST*,
or delisting risk warning special treatment, as a warning to potential investors. According to the
Shanghai Stock Exchanges, companies are designated as Special Treatment, or ST, if (1) the firm
had negative shareholder’s equity in its latest financial year or if (2) the firm’s accountant
expresses negative opinion in its auditing report, and other non-financial events3. ST designation
signifies primary level of financial distress. Companies are designated as ST* if (1) the firm had
negative earnings over the past two years or if (2) the firm has significant accounting fraud or
failed to disclose annual or half-yearly financial statements, as well as other non-financial
events4. ST* designation implies that firms are in a worse financial position and at higher risk of
delisting than ST firms are. It is important to note that a firm may skip ST and have ST*
designation. Table 1 provides historic ST* data, which indicates that, since 2003, ST*
companies increased over the subsequent years. From 2007 onward, we noticed a significant
drop in ST* companies until 2014. As a result of stricter enforcement and execution standard, no
public company were delisted5.
Table 1 Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Total # of
Listed
Companies
1086 1160 1235 1296 1384 1387 1440 1530 1604 1700 2063 2342
Total # of
Delisted
Companies
3 7 4 8 12 4 6 1 0 0 0
% of Total
Companies 0.26% 0.57% 0.31% 0.58% 0.87% 0.28% 0.39% 0.06% 0.00% 0.00% 0.00%
Total # of
ST
Companies
15 12 38 44 34 31 50 46 19 24 37 15
% of Total
Companies 1.38% 1.03% 3.08% 3.40% 2.46% 2.24% 3.47% 3.01% 1.18% 1.41% 1.79% 0.64%
Total # of
ST*
Companies
34 28 26 48 43 19 20 35 12
% of Total
Companies 2.62% 2.02% 1.87% 3.33% 2.81% 1.18% 1.18% 1.70% 0.51%
% of Total
ST
Companies
77.27% 82.35% 83.87% 96.00% 93.48% 100.00% 83.33% 94.59% 80.00%
For our research paper, we created the STAR model that predicts the probability of
Chinese companies that will be in financial distress. Of the two designations, we decided to
focus on the predictability of ST* companies because the financial analysis and accounting ratios
3 http://www.htsc.com.cn/browser/view/investorsReadingNews.jsp?nodeName=tzzjy2&docId=4583704&cur=default 4 http://www.htsc.com.cn/browser/view/investorsReadingNews.jsp?nodeName=tzzjy2&docId=4583704&cur=default 5 http://www.csrc.gov.cn/pub/zjhpublic/zjh/201410/P020141017502705934937.pdf
Bankruptcy & Reorganization – Professor Altman 12/15/2014
3
can better predict the scenarios that result in ST* designation. Our research is important for
investors investing in Chinese distressed stocks, regulators, and researchers. We noticed that the
stock price fall a month ahead of the delisting which will allow investors to adjust their strategies
and limit their investment exposure. Regulators are able to weed out poor performing companies
and limit systematic risks. And for academics, the STAR model provides the foundation for
further research.
Literature Review
Numerous scholars have researched the relationship between corporate governance and
financial distress. Based on data on 96 financially distressed companies and 96 healthy
companies, Wang and Deng (2006) 6 analyzed the influence of governance structure and
managerial agency cost factors on a company’s probability of financial distress. They found that
a higher percentage of independent directors siting in a company’s board, large shareholder and
state ownership, and lower managerial agency costs, as measured by administrative expense ratio,
tend to decrease the likelihood that a company will be in financial distress. Meanwhile,
managerial ownership, board size, and CEO longevity have no significant influence on a
company’s financial performance.
Kam, Citron, and Muradoglu’s research (2007)7 focused on the survival of distressed
firms designated as ST or ST* and found that merger & acquisitions might not be ideal. Their
research concluded that private companies, rather than state-owned companies, are more likely to
gain financial strength by doing M&A. However, state-owned firms tend to lose value when
controlling ownership is transferred. Furthermore, markets perceive asset sales as a negative
sign of a company’s financial health for both competitive and state-owned firms.
Zhang, Wu, and Yan (2012) 8 used key ratios of a company’s financial structure,
specifically assets, to predict a company’s stock price based on a sample of ST companies from
2007 and 2009. Based on those companies’ financial performance after their ST designation,
their research concluded that higher current asset to total asset ratio and collateralizable assets to
total asset ratio predict better stock price performance, while equity per share of the company
does not affect the predictability.
Z-China Model
The Z-Score China model uses a multivariate discriminant analysis to classify whether a
company is ST* or not. Despite volatile market dynamics and various new techniques such as
neural networks,, this model has been very popular as it is easy to understand and replicate and
has consistent success of predictability. The Z-score model used to identify distressed companies
in China uses four independent variables in a linear combination as follows:
Z = 0.0517 + 0.460X1+9.320X2+0.388X3+1.158X4
where,
X1= asset liability ratio (total liabilities/total assets);
6 Wang, Zong-Jun, and Xiao-Lan Deng. "Corporate governance and financial distress: Evidence from Chinese listed
companies." Chinese Economy 39, no. 5 (2006): 5-27. 7 Kam, Amy, David Citron, and Gulnur Muradoglu. "Distress and restructuring in China: Does ownership
matter?." China Economic Review 19, no. 4 (2008): 567-579. 8 DONG, Nan-yan, Zong-wu JIA, and Jun-rui ZHANG. "Research on the Value Relevance of Asset Structure for
Financially Distressed Listed Companies on China Security Market." In Statistics & Information Forum, vol. 5, p. 010. 2012.
Bankruptcy & Reorganization – Professor Altman 12/15/2014
4
X2 = rate of return on total assets (net profit/average total assets);
X3= working capital to total asset ratio (working capital/total assets), where working
capital equals current assets minus current liabilities;
X4= retained earnings to total assets ratio (retained earnings/total assets).
We collected data on 204 companies, half were ST* and half were not. Table 2 sums up the
accuracy in predicting ST* companies. The Z-China model resulted in (4+3)/102, or 7% Type 1
error, which refers to the actual ST* companies that the model had predicted as non ST*. It also
resulted in (18+26)/102, or 18% Type 2 error, which refers to the actual non ST* companies that
the model had predicted as ST*. Type 2 error is higher than we expected. Graph A is a plot
showing the distribution of our analysis using the Z-China model. The x-axis lists the
companies, and the y-axis is the z score. If Z<0.5, the firm is distressed. If 0.5<Z<0.9, the firm
is potentially distressed. If Z>0.9, the firm is financially healthy.
Table 2
Actual ST Actual Non ST*
Predicted Good 3 58
Predicted Maybe 4 26
Predicted Bad 95 18
Total 102 102
Graph A
O-Score model
The Ohlson’s model builds on existing MDA models but uses probabilistic model to
interpret the results, easily differentiating companies in financial distress from companies that
Bankruptcy & Reorganization – Professor Altman 12/15/2014
5
are financially healthy. To improve accuracy, Ohlson uses nine independent variables including
two binary ones. Most of these variables have appeared in prior research and were used based on
their frequency of appearance.
O-score = -1.32 – 0.407*log (TA) + 6.03*(TL/TA) – 1.43*(WC/TA) + 0.757*(CL/CA) –
2.37*(NI/TA) – 1.83*(FFO/TL) -1.72*X + 0.285*Y – 0.521*(NIt-NI t-1/
|NIt|+|NIt-1|)
where,
TA = total assets;
TL = total liabilities;
WC = working capital;
CL = current liabilities;
CA = current assets;
X = 1 if TL > TA, 0 otherwise;
NI = net income;
FFO = funds from operations;
Y = 1 if a net loss for the last two years, 0 otherwise
Using logistic regression, the O-score is converted to predict the probability that a
Chinese company will be designated as ST* as follows:
Probability of Financial distress (P) = exp (O-score)/1+exp (O-score)
Using the same data on 204 companies, Table 3 sums up the accuracy in predicting ST*
companies. The O-Score model resulted in 35/102, or 35% Type 1 error and 30/102, or 30%
Type 2 error. Graph B is a plot showing the distribution of our analysis using the O-Score
model. The x-axis lists the companies, and the y-axis is the predicted probability of ST*. If the
probability is greater than 50%, the firm is distressed. If the probability is less than 50%, the
firm is financially healthy.
Table 3
Actual ST* Actual Non ST*
Predicted ST* 67 30
Predicted non ST* 35 72
Total 102 102
Bankruptcy & Reorganization – Professor Altman 12/15/2014
6
Graph B
STAR Model
Using the Z-China and O-score models as benchmark, we came up with the STAR model.
In addition to the four variables from the Z-China model, we added the percentage of cash in
total assets, an indicator variable for change in cash level, and the net profit over average total
assets. The following are detailed descriptions of each variable analyzed, including the ones that
we eliminated for the STAR model.
Following previous studies linking ownership structure and number of independent
directors to the predicted probability that a company will be designated as ST*, we analyzed and
concluded that management ownership, employee ownership, and state ownership are not
relevant. There was no significant information on the managerial and employee ownership one
year before a company was designated as ST*, and there were no reliable data for the number of
independent directors for our sample set of companies. Reduction in state ownership provided
some meaningful information and was a leading indicator in most cases. However, there was no
pattern in the reduction of state ownership.
We analyzed and concluded that higher inventory to total assets ratio is not relevant to
predicting financial distress. A company might be facing financial issues because not everything
that they produce might be consumed and hence they might have a higher inventory than the
companies that are not facing financial problems. However the correlation of net inventory to
total assets ratio the year before with the designation of ST* the following year is only -0.0544
for our sample set.
We analyzed and concluded that higher available-for-sale financial assets to total assets
ratio is not relevant. If a company encounters financial problems, it might sell assets to generate
cash flow to continue operation resulting in higher available-for-sale financial assets to total
Bankruptcy & Reorganization – Professor Altman 12/15/2014
7
assets ratio. However, the correlation was only -0.0997 for companies the year before their ST*
designation with the following year.
We analyzed and concluded that long-term investments are not relevant. We tested
whether long-term investments of firms will reduce one year from two years before their ST*
designation. However, the correlation is -0.0174.
We analyzed and concluded that higher DE ratio is not relevant. Financial distress could
result from poor management decisions in allocating capital structure and debt overhang problem.
Higher interest payments resulting from more debt might cause a company to be designated as
ST*. We compared DE ratio with the industry average DE ratio in the year before the ST*
designation as DE ratios varied by industries. (For example, the railroad industry capital
intensive.) We assigned an indicator variable of 1 if the ratio is higher and 0 if the ratio is lower.
We also computed the DE ratio for all the companies in the sector from CSMAR [R] and took an
average for the industry average DE ratio. However, the correlation is 0.1815.
We analyzed and concluded that the percentage of cash in total assets and an indicator variable
for change in cash level are relevant. If a company is close to financial problems, its Cash/Total
Assets ratio will fall as it burns through cash. A correlation of -0.26679 validated our hypothesis.
Also, if the Cash/Total Assets increased from two years to one year before the designation of ST,
there is a lower chance that the company will face financial distress, again validating our
hypothesis.
The last variable that we added is the Net Profit over Average Total Assets, which is 6-
level categorical variable, in which a value is assigned to each of the six scenarios based on
NP/ATA at 2 years and 1 year before the ST* designation and the change in NP/ATA between
year 2 and year 1. This flow information is significant since it can determine if a company is
doing well or not. In Table 4, if the company had a positive NP/ATA two years before and if it
increased one year before ST*, we denoted the scenario as direction 1.
Table 4
NP/ATA
(2 years before ST*)
NP/ATA
(1 year before ST*) NP/ATA Change Direction
Positive Positive Increased 1
Positive Positive Decreased 2
Positive Negative Decreased 3
Negative Positive Increased 4
Negative Negative Increased 5
Negative Negative Decreased 6
Table 5 sums up the correlations of the independent and dependent variables. The NP/ATA
independent variable one year before the ST* designation had a very high correlation of
0.6382914.
Bankruptcy & Reorganization – Professor Altman 12/15/2014
8
Table 5
X1-X2 X1-X3 X1-X4 X1-X5 X1-X6 X1-X7 X1-ST*
-0.122753 -0.77376 -0.670092 -0.213532 0.0232682 0.1859236 0.2527391
X2-X3 X2-X4 X2-X5 X2-X6 X2-X7 X2-ST*
0.1791181 -0.123929 0.3244134 0.1365146 -0.611659 -0.591269
X3-X4 X3-X5 X3-X6 X3-X7 X3-ST*
0.5554479 0.3929681 0.0622433 -0.212798 -0.318324
X4-X5 X4-X6 X4-X7 X4-ST*
0.0140018 -0.058825 -0.14593 -0.174492
X5-X6 X5-X7 X5-ST*
0.2794833 -0.171598 -0.26679
X6-X7 X6-ST*
-0.015604 -0.11767
X7-ST*
0.6382914
From our sample set, we decided to keep the 25 ST* and 25 non ST* companies from
2014 as a test set. We used the remaining 77 ST* and 77 non ST* companies as a training set to
come up with the STAR model. We used logistic regression to come up with a STAR score and
ran our data using WEKA to come up with suitable coefficients. (Please note that logistic
regression was best as we also tested using Naïve Bayes and Neural Networks.) Based on the
models generated, we tweaked the model’s co-efficients for accuracy and to ensure that the
model is not constrained to the training set. The STAR score is as follows and Table 6 lists
explains the STAR variables:
STAR Score = -3.7937 -1.0651 * X1 + 15.7617 * X2 + 2.9307 * X3 + 5.6007 * X4 -
4.4805 * X5 + 1.5383 * X6 + X7
Bankruptcy & Reorganization – Professor Altman 12/15/2014
9
Table 6
Variable Value
X1 Total Liabilities/Total Assets
X2 Net Profit/Avg total Assets [NP/ATA]t
X3 Working Capital/Total Assets
X4 Retained Earnings/Total Assets
X5 Cash/Total Assets
X6 If Cash/Total Assets increased from last year, 1 else 0
X7 Value Condition
6.50 [NP/ATA]t > 0; [NP/ATA]t-1 > 0; [NP/ATA]t > [NP/ATA]t-1
4.00 [NP/ATA]t > 0; [NP/ATA]t-1 > 0; [NP/ATA]t < [NP/ATA]t-1
2.25 [NP/ATA]t < 0; [NP/ATA]t-1 > 0;
4.50 [NP/ATA]t > 0; [NP/ATA]t-1 < 0;
1.00 [NP/ATA]t < 0; [NP/ATA]t-1 < 0; [NP/ATA]t > [NP/ATA]t-1
-20.00 [NP/ATA]t < 0; [NP/ATA]t-1 < 0; [NP/ATA]t < [NP/ATA]t-1
Based on the STAR score, we came up with the probability for predicting the probability
that a company will be designated as ST* in the following year:
P (ST*) = 1/ (1+EXP (STAR Score))
Using the training set, Table 7 shows that the STAR model had a 92% accuracy rate and
resulted in 5/77, or 6% Type 1 error and 8/77, or 10% Type 2 error.
Table 7
Actual ST* Actual non ST* Total
Predicted ST* 72 8 80
Predicted non ST* 5 69 74
Total 77 77
Using the test set from 2014, Table 8 shows that the STAR model had an accuracy of 88%
with 2/25, or 8% Type 1 error and 4/25, or 16% Type 2 error.
Table 8
Actual ST* Actual non ST* Total
Predicted ST* 23 4 27
Predicted non ST* 2 21 23
Total 25 25
For the overall data, the STAR model had 7/102, or 7% Type 1 error and 12/102 Type 2
error. We plotted the results to show the distribution for the training set in Graph C. The x-axis
lists the companies, and the y-axis is the predicted probability of ST*. If the probability is
Bankruptcy & Reorganization – Professor Altman 12/15/2014
10
greater than 50%, the firm is distressed. If the probability is less than 50%, the firm is
financially healthy.
Graph C
Comparison of Results
We compared the probability of ST* for all three models using our 2014 test data. For
meaningful comparison, we used the same training set as the STAR model for uniformity and
created a logistic regression model for the Z-China model using the four variables from the
original paper as follows:
Z = 1.3741 -1.2491X1 + 30.1693X2 + 1.9696X3+ 7.4372X4
Where,
X1= asset liability ratio (total liabilities/total assets)
X2 = rate of return on total assets (net profit/average total assets)
X3= working capital to total asset ratio (working capital/total assets), where working
capital equals current assets minus current liabilities
X4= retained earnings to total assets ratio (retained earnings/total assets)
Bankruptcy & Reorganization – Professor Altman 12/15/2014
11
We used the same probability formula from STAR model as follows:
P (ST*) = 1/ (1+EXP (STAR Score))
Table 9 sums up the accuracy in predicting ST* companies. For Z-China Logistic model,
Type 1 error is 1/25, or 4%, and Type 2 error is 11/25, or 44%. For Ohlson’s model, Type 1
error is 6/25, or 24%, and Type 2 error is 11/25, or 44%. For STAR model, Type 1 error is 2/25,
or 8%, and Type 2 error is 4/25, or 16%. Graph D is a plot showing the distribution of our
analysis using the Z-China Logistic model for 2014 test data. Graph E is a plot showing the
distribution of our analysis using the Ohlson’s model for 2014 test data. Graph F is a plot
showing the distribution of our analysis using the STAR model for 2014 test data. The x-axis
lists the companies, and the y-axis is the predicted probability of ST*. If the probability is
greater than 50%, the firm is distressed. If the probability is less than 50%, the firm is
financially healthy.
Table 9
Z-China (Logistic)
model
Ohlson’s model STAR model
Actual
ST*
Actual
Non ST*
Actual
ST*
Actual
Non ST*
Actual
ST*
Actual
Non ST*
Predicted
ST* 24 11
19 11 23 4
Predicted
Non ST* 1 14
6 14 2 21
Total 25 25 25 25 25 25
Bankruptcy & Reorganization – Professor Altman 12/15/2014
12
Graph D
Graph E
Pro
ba
bil
ity
of
ST
*
Company
Z-China Logistic Model for 2014 Data SetP
rob
ab
ilit
y o
f S
T*
Company
O-Score Model for 2014 Data Set
ST*
Non ST*
Y = 0.5
Bankruptcy & Reorganization – Professor Altman 12/15/2014
13
Graph F
Conclusion
Of the two designations, we decided to focus on the predictability of ST* companies in
China because the financial analysis and accounting ratios can better predict the scenarios that
result in ST* designation. Our research is important for investors investing in Chinese distressed
stocks, regulators, and researchers. China has a 5% gain or loss limit per day. It caps the move
at intraday prices. We noticed that over a period of a month, the companies that are designated
as ST* started losing a significant portion of their market value, which will allow investors to
adjust their strategies and limit their investment exposure. Regulators are able to weed out poor
performing companies and limit systematic risks. And for academics, the STAR model provides
the foundation for further research.
Our conclusion is that the STAR model reduced Type 2 error significantly while
maintaining Type 1 error. This improved predictability allows investors to better invest in or
steer clear of distressed Chinese companies and allows for regulators to identify distressed
companies early on to properly allocate public capital. Of our data set, only 1 company was
delisted from the stock exchange; therefore, it is not a predictor of whether a company will be
delisted if it transitions from ST* to ST.
Pro
ba
bil
ity
of
ST
*
Company
STAR Model for 2014 Data Set
ST*
Non ST*
Y = 0.5