The Journal of Hospitality Financial Management

Fo

Improving airline bankruptcy prediction

rP

Journal: The Journal of Hospitality Financial Management

Manuscript ID Draft
ee

Manuscript Type: Refereed Paper

bankruptcy prediction, airline industry, Altman Z-Score model, multiple

Keywords:
rR

discriminant analysis, ratio analysis

ev
ie
w
On
ly

URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu

Page 1 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 1

1
2
3
4 Improving airline bankruptcy prediction
5
6 Abstract
7
8
9 The airline industry plays an important role in the global economy, but faces financial
10
11 challenges. Numerous firms have filed for bankruptcy protection or have liquidated completely,
12
13 each time having a devastating effect on the company’s stakeholders. The objective of this
14
Fo
15
16 study was to compare a traditional bankruptcy prediction model with a proposed alternative
17
18 model, with the goal of identifying a means of predicting the combinations of characteristics that
rP
19
20
are present when an airline is likely to fail. The alternate model proved to be more accurate than
21
22
the traditional model in predicting bankruptcy, providing improved forecasting up to four years
ee
23
24
25 prior to the bankruptcy filing date. Airlines can use this model to deploy corrective measures to
26
rR

27
28 alter the firm’s underlying problems, redefine strategies, and avoid bankruptcy, while investors
29
30 can use this model to avoid or reduce investments in questionable firms that cannot be salvaged.
ev

31
32 Keywords: bankruptcy prediction, airline industry, Altman Z-Score model, multiple
33
34
discriminant analysis, ratio analysis
ie

35
36
37
w

38
39 Introduction
40
On

41
42 Between 2001 and 2011, the U.S. airline business suffered $10 billion in losses (Neuman,
43
44 2011). The industry in aggregate lost over $60 billion in the thirty-two years following
45
ly

46
47
deregulation in 1978. Legacy airlines have had a particularly hard time. Since 1998, ten large
48
49 North American airlines have filed for bankruptcy: Air Canada; American; ATA [American
50
51 Trans Air]; Delta; Frontier; Hawaiian; Northwest; TWA [Trans World Airlines]; US Airways;
52
53
54 and United. TWA is no longer operating; it flew its last official flight on December 1, 2001.
55
56 ATA ceased operations in 2008. America West acquired US Airways, which then merged with
57
58
59
60
1
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 2 of 24

Improving airline bankruptcy prediction 2

1
2
3
4 American Airlines; Continental merged with United; Northwest merged with Delta; Southwest
5
6 acquired AirTran; and Alaska Airlines acquired Virgin America (Denning, 2011). By 2015, four
7
8
9 mega-carriers made up a combined 80% of all U.S. flights (Groden, 2015).
10
11 Warren Buffett, Chairman of Berkshire Hathaway and probably one of the most successful
12
13 investors in the world, called the airline industry a “death trap for investors” in 2013 (La Monica,
14
Fo
15
16 2017). What happened? Volatile fuel prices, overcapacity, the economic recessions, terrorism,
17
18 war, increased security costs, rising insurance premiums, high competition, and poor
rP
19
20
management have all contributed to the adverse financial impact on the airline industry (Federal
21
22
Aviation Administration, 2014). The airline industry was officially deregulated in October 1978,
ee
23
24
25 which brought about many changes including the strengthening of hub and spoke operations, fare
26
rR

27
28 cutting, and the entry of new competitors into the industry. Before airline deregulation, no major
29
30 airline filed for Chapter 11 bankruptcy (Cheng & McDonald, 1996). Following deregulation, the
ev

31
32 airline industry has suffered financially from various problems: the economic recessions of the
33
34
early 1980s and 2008-2013; rising jet fuel prices; rising labor costs; maintenance and interest
ie

35
36
37 costs; foreign exchange risk; rising insurance costs; and intensified competition.
w

38
39 The transition from a regulated to a deregulated environment increased the instability of the
40
On

41
42 carriers’ operating profits. Total risk, defined as the volatility of net profits and cash flows over
43
44 time, has increased dramatically in the airline industry. Oil prices have been, at times, over
45
ly

46
47
$100 per barrel. Consequently, jet fuel prices have skyrocketed, devastating the airlines’ bottom
48
49 lines.
50
51 If there were a means of predicting the combinations of characteristics of an airline that is
52
53
54 likely to fail, corrective measures could be taken to alter the firm’s underlying problems, redefine
55
56 strategies and procedures. This could also be useful to investors by helping them avoid or reduce
57
58
59
60
2
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 3 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 3

1
2
3
4 investments in questionable firms that cannot be salvaged (Patterson, 2001). One method of
5
6 predicting financial distress that has been widely used for nearly 50 years is the statistical
7
8
9 bankruptcy prediction model, first presented by Altman (1968). This model, the Altman Z-Score
10
11 model, is a popular approach for not only forecasting bankruptcy in advance of the event, but
12
13 also for gauging the overall financial condition of a firm.
14
Fo
15
16 The objective of this study was to analyze financially distressed and non-financially
17
18 distressed airlines using a traditional bankruptcy prediction model, in order to evaluate its ability
rP
19
20
to predict bankruptcy in the airline industry. This study will add to the bankruptcy research,
21
22
addressing Altman and Hotchkiss’ (2006) recommendations to enhance previous models by
ee
23
24
25 introducing an alternative statistical model that may better predict which airlines are likely to fail
26
rR

27
28 and which are not likely to fail. The new model’s classification rate is compared to the rate
29
30 generated using the Altman Z”-Score model, a variant of the original Altman Z-Score model.
ev

31
32 Literature Review
33
34
Numerous models for predicting bankruptcy have been proposed. This paper will first
ie

35
36
37 examine general-purpose models, focusing on the Altman Z-Score model. Next, an overview of
w

38
39 bankruptcy prediction research in the hospitality industry is presented, followed by a focus on
40
On

41
42 past work predicting airline bankruptcy.
43
44 Common Bankruptcy Prediction Models
45
ly

46
47
Numerous studies have used the Altman Z-Score (Z-Score) model and its variants to predict
48
49 firm bankruptcies and financial distress for manufacturing firms. One method of predicting
50
51 financial distress that has been widely used for over 50 years is the statistical bankruptcy
52
53
54 prediction model, first presented by Altman (1968). The Altman Z-Score model uses five
55
56 financial ratios to represent the elements that predict failure. These elements are liquidity,
57
58
59
60
3
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 4 of 24

Improving airline bankruptcy prediction 4

1
2
3
4 cumulative profitability, productivity, solvency, and activity. Multiple discriminant analysis
5
6 (MDA) is applied to the financial ratios to understand group differences and to predict the
7
8
9 likelihood than an entity (individual or object) will belong to a particular class or group based on
10
11 several metric variables (Hair, Black, Babin, & Anderson, 2009).
12
13 In response to requests for a measure to predict the likelihood of bankruptcy for non-
14
Fo
15
16 manufacturing firms, Altman developed the Z” model (i.e., the ‘Z-prime-prime’), a four-variable
17
18 multiple discriminant model (Altman & Hotchkiss, 2006). The four financial ratios used in the
rP
19
20
Z”-Score model represent liquidity, cumulative profitability, productivity, and solvency.
21
22
The Altman Z”-Score model is shown below :
ee
23
24
25 Z” = 6.56 X1 + 3.26 X2 + 6.72 X3 + 1.05X4 + έ
26
rR

27
28 Where:
29
30 X1 = working capital/total assets;
ev

31
32 X2 = retained earnings/total assets;
33
34
X3 = earnings before interest and taxes/total assets;
ie

35
36
37 X4 = book value of equity/book value of total liabilities;
w

38
39 έ = error term; and,
40
On

41
42 Z” = overall index
43
44 While the newer Z”-Score model provides greater accuracy for non-manufacturing and emerging
45
ly

46
47
markets (non-US) firms, Altman and Hotchkiss (2006) do suggest that further developing of
48
49 bankruptcy prediction models for specific industries would be desirable.
50
51 Bankruptcy prediction models for the hospitality industry. There have been few studies
52
53
54 devoted to bankruptcy prediction in hospitality and travel. Gu and Gao (2000) used a multiple
55
56 discriminate analysis (MDA) model to predict bankruptcy for hospitality firms including
57
58
59
60
4
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 5 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 5

1
2
3
4 restaurant, hotel, and casino firms. Their model used five variables (total liabilities to total assets,
5
6 EBIT to current liabilities, gross profit margin, long-term debt to total assets, and sales to fixed
7
8
9 assets). This model could correctly predict bankruptcy in 93% of the sample firms one year in
10
11 advance of the actual bankruptcy. Gu (2002) focused on bankruptcy prediction in the restaurant
12
13 industry using the same MDA methodology. His MDA model predicted bankruptcy in 92% of
14
Fo
15
16 the sample firms one year in advance of the actual bankruptcy. A limitation of these two studies
17
18 was the short-term predictive ability of their MDA models. Both models predicted bankruptcy
rP
19
20
only one year in advance.
21
22
Kim and Gu (2006a) first developed a logistic regression (logit) model to improve the
ee
23
24
25 prediction accuracy over the traditional MDA model. In this instance, they developed the model
26
rR

27
28 for use in the restaurant industry and found that while both logit and MDA performed about the
29
30 same, the underlying soundness of the results was stronger when using logit. These models
ev

31
32 predicted restaurant bankruptcy approximately one year prior to the bankruptcy.
33
34
Kim and Gu (2006b) next revisited the logit approach to predict hospitality firm bankruptcy
ie

35
36
37 two years in advance. This logit model could correctly predict 91% of hospitality bankruptcy one
w

38
39 year in advance of the actual event, and 84% of bankruptcy two years in advance of the event.
40
On

41
42 The sample in this study included hotel, restaurant, and gaming companies. This improved
43
44 model was proposed to help provide an “early distant warning” to hospitality industry operators
45
ly

46
47
to allow adequate time to correct deficiencies and head off the bankruptcy.
48
49 Youn and Gu (2010) used a combination of logistic regression and an artificial neural
50
51 networking (ANN) model to predict Korean lodging firm failures. The addition of the ANN
52
53
54 process improved the accuracy of the prediction rate over using only logistic regression.
55
56 However, the use of an ANN requires a great deal of interpretation of “black box” processes.
57
58
59
60
5
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 6 of 24

Improving airline bankruptcy prediction 6

1
2
3
4 The ANN approach also does not provide prescriptive results, which has the disadvantage of not
5
6 providing actionable date for the firm to use to improve its areas of weak performance.
7
8
9 Kim (2011) compared four different methods – MDA, logit, ANN, and support vector
10
11 machine (SVM) models to determine which provided the greatest accuracy and which minimized
12
13 the incidences of Type 1 and Type 2 error. The author concluded that the ANN technique
14
Fo
15
16 provided the best accuracy of the four models.
17
18 Airline bankruptcy prediction models. Several different statistical techniques have been
rP
19
20
used in the past to assess airline financial performance, including MDA, logit, and ANN.
21
22
MDA. The first approach, MDA, attempts to classify airlines as financially distressed or not
ee
23
24
25 distressed (Gritta, 1974, 1982; Scaggs & Crawford, 1986; Golaszewski & Sanders, 1992; Chung
26
rR

27
28 & Szenberg, 1996; Gritta, Adrangi, Adams, & Tatyanina, 2008). Gritta (1974) looked at “the
29
30 significance of the effects of capitalization on the ‘perceived’ amounts of long-term debt in an
ev

31
32 airline firm’s capital structure, and therefore, on the debt-to-equity measures” (p. 47). Gritta
33
34
(1982) applied a generally accepted model to air transportation in an effort to appraise air
ie

35
36
37 carriers' financial strength and to predict likely bankruptcy candidates. The analysis was
w

38
39 designed to aid in pin-pointing the causes of the air carriers' financial difficulties and was
40
On

41
42 intended to be of interest to airline management, creditors, and regulators. Scaggs and Crawford
43
44 (1986) revisited Altman's bankruptcy model to determine if airline bankruptcy could be
45
ly

46
47
predicted. Their model predicted bankruptcy well, but did not predict non-bankruptcy well.
48
49 Golaszewski and Sanders (1992) stated that their object was “not to predict bankruptcy
50
51 performance per se; rather, it is to examine the status of a carrier’s finances prior to bankruptcy
52
53
54 and then identify other carriers which, while not bankrupt, are under significant financial
55
56 distress” (p. 313). Chung and Szenberg (1996) employed Altman’s Z-Score as a measure of the
57
58
59
60
6
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 7 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 7

1
2
3
4 airline industry’s financial stability in the 1980s, following the industry’s deregulation. They
5
6 identified increased volatility throughout the industry, classifying the industry’s “health” as
7
8
9 either healthy or unhealthy. However, the authors did not use the model to predict bankruptcy.
10
11 Gritta et al. (2008) assessed the financial condition of the major U.S. air carriers from 2000
12
13 to 2006 and compared their financial strength to the 1995-1999 period when the carriers earned
14
Fo
15
16 record profits. They found an increasing trend in negative equity and increased leverage,
17
18 predicting future bankruptcies if the low interest rates available from 2000-2006 were to rise.
rP
19
20
Kroeze and Mayer (2006) developed a variant of the MDA approach and found improved
21
22
predictive capabilities, and recommended retesting the model in the future to include a larger
ee
23
24
25 sample of companies that had undergone either Chapter 7 or 11 bankruptcy.
26
rR

27
28 Logistic regression. Other researchers used a different approach by developing logistic
29
30 regression (logit) models, which predict the odds-ratio of an event occurring. For example,
ev

31
32 Gudmundsson (1999; 2002) constructed a logistic regression model of airline distress prediction
33
34
using three years of worldwide airline data [1996-1998] including non-financial operating data
ie

35
36
37 and proxy variables for governmental influence and quality of economic environment. The
w

38
39 findings demonstrated a fairly good model, having 90.3 percent overall prediction accuracy.
40
On

41
42 These findings in conjunction with other research in this field, supported that models based on
43
44 non-financial variables showed good prediction traits comparable to financial based models, yet
45
ly

46
47
provided more explanatory power.
48
49 Neural networks. Next, a few researchers have employed a neural network approach to
50
51 predict airline bankruptcy. Davalos, Gritta, and Chow (1999) and Gritta, Wang, Davalos, and
52
53
54 Chow (2000) developed airline bankruptcy prediction models using this approach. Studies were
55
56 conducted on both major air carriers and regional air carriers using the neural network. Twenty-
57
58
59
60
7
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 8 of 24

Improving airline bankruptcy prediction 8

1
2
3
4 one pieces of financial information from carrier balance sheets and income statements were
5
6 entered into the model. The study successfully classified all the major carriers that filed for
7
8
9 receivership and most of the regional carriers. The use of neural networks may provide an
10
11 interesting supplement to the analyst in appraising financial health.
12
13 Finally, some researchers are exploring bankruptcy from different perspectives. For
14
Fo
15
16 example, Jayanti and Jayanti (2011) examine the effect of one airline company’s bankruptcy on
17
18 its competitors, finding that when a major airline announces it is going into Chapter 11
rP
19
20
restructuring, its competitors realize abnormally high returns, although these returns drop
21
22
precipitously when the bankrupt carrier emerges from bankruptcy.
ee
23
24
25 Purpose of the Study
26
rR

27
28 The purpose of this study is to test Altman’s Z”-score model in the context of the airline
29
30 industry, and then to compare it against a new, proposed model. It seeks to update and extend
ev

31
32 these earlier studies of airline bankruptcy prediction. The goal to add to theory and practice by
33
34
identifying a model that is easy to use (unlike ANN modeling), accurate, and one that provide the
ie

35
36
37 greatest advance notice of the risk of bankruptcy proceedings, in order to help airline operators to
w

38
39 take corrective action as soon as possible, as well as to help investors with decision-making. To
40
On

41
42 do this, the study tests the following hypotheses:
43
44 H10: There is no relationship between the Altman Z”-Score model and the likelihood of
45
ly

46
47
bankruptcy for an airline firm.
48
49 H1A: There is a relationship between the Altman Z”-Score model and the likelihood of
50
51 bankruptcy for an airline firm.
52
53
54 H20: The proposed bankruptcy prediction model is no better than the Altman Z”-Score
55
56 model in predicting the likelihood of bankruptcy for an airline firm.
57
58
59
60
8
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 9 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 9

1
2
3
4 H2A: The proposed bankruptcy prediction model is better than the Altman Z”-Score
5
6 model in predicting the likelihood of bankruptcy for an airline firm.
7
8
9 Methods
10
11 Altman (1993) states that, ideally, one would like to develop a bankruptcy prediction model
12
13 utilizing a homogeneous group of bankrupt companies and data as near to the present as possible.
14
Fo
15
16 Following Altman’s guidelines, this study used bankrupt and non-bankrupt airline firms’ 1998-
17
18 2005 financial statements. The financial statements were retrieved from the Securities Exchange
rP
19
20
Commission (SEC) website, http://www.sec.gov/, using the EDGAR database, and from airline
21
22
firms’ annual reports, which are available on-line from the individual companies’ websites. All
ee
23
24
25 publicly held U.S. companies are required to file their financial statements with the SEC.
26
rR

27
28 Bankrupt companies, for the purposes of this study, were defined as those meeting one of the
29
30 following conditions: (1) in Chapter 11 bankruptcy protection; or, (2) in Chapter 7 liquidation.
ev

31
32 Thus, those airline firms that were in one or more of these states at any time during the 1998-
33
34
2005 time period were considered to be bankrupt, leading to a dichotomous variable representing
ie

35
36
37 bankruptcy filing between 1998 and 2005.
w

38
39 This study used a census approach, rather than a random sample. Only major and national
40
On

41
42 airlines were selected for this study. Major airlines, or majors, are a group of large, certified air
43
44 carriers that have annual operating revenues over $1 billion. National airlines, or nationals, are a
45
ly

46
47
group of large, certified air carriers that have annual operating revenues of $100 million to $1
48
49 billion.
50
51 The financial data of publicly-held major passenger airlines, including Alaska, America
52
53
54 West, American, Continental, Delta, Northwest, Southwest, TWA, United, and USAir. Air
55
56 Canada was also included in this study, as it has sufficiently large revenues, and uses Generally
57
58
59
60
9
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 10 of 24

Improving airline bankruptcy prediction 10

1
2
3
4 Accepted Accounting Principles (GAAP) in preparing its financial statements, similar to
5
6 publicly-owned U.S. firms. Air Canada stock is traded on the American Stock Exchange. The
7
8
9 publicly-owned national passenger airlines (not including regional airlines), included AirTran
10
11 (formerly ValuJet), ATA (formerly Amtran), Frontier, Hawaiian, and JetBlue. Non-passenger
12
13 airlines (DHL, Federal Express, and United Parcel Service) were not included in this study.
14
Fo
15
16 During the period 1998-2005, the following nine airline firms were liquidated or filed for
17
18 Chapter 11 bankruptcy protection: Air Canada; ATA; America West; Delta; Hawaiian;
rP
19
20
Northwest; TWA; United; and US Airways. During the same time period, the following seven
21
22
firms were not liquidated, nor did they file for Chapter 11 bankruptcy protection: Air Tran;
ee
23
24
25 Alaska; American; Continental; Frontier; JetBlue; and Southwest. This study utilized a holdout
26
rR

27
28 sample to evaluate the model. Hold-out samples are useful where the period of the model is
29
30 different from the period of evaluation. With this technique of model evaluation, the in-sample
ev

31
32 data ends at a point in time, and the remainder of the data are held out as a non-overlapping
33
34
period of evaluation. The hold-out sample is used to compare the forecasting accuracy of models
ie

35
36
37 fit to past data (SAS Institute, 2017). The in-sample size of 62 data points and the holdout
w

38
39 sample size of 22 data points were adequate, given the statistical method chosen and the two
40
On

41
42 classification groups (Hair et al, 2009).
43
44
45
ly

46
47
Analysis and Results
48
49 The data were analyzed using multiple discriminant analysis (MDA) to predict group
50
51 membership (Tabachnick & Fidell, 2012). MDA is principally used to classify and to make
52
53
54 predictions in situations where the criterion variable is in categorical form (Hair et al. 2009), as
55
56 was the case in this study (e.g., bankrupt versus non-bankrupt). A significant difference between
57
58
59
60
10
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 11 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 11

1
2
3
4 the two groups, bankrupt or non-bankrupt, implies that one can predict whether a firm will be
5
6 bankrupt in one, two, or even three years, depending upon the score that the firm receives from
7
8
9 the application of MDA. The predictors were a set of financial ratios that measured a firm’s
10
11 liquidity, cumulative profitability, productivity, solvency, and cash flow (Beaver, 1966).
12
13 Assumptions of MDA. The key assumptions for deriving the discriminant function are
14
Fo
15
16 multivariate normality of the independent variable and equal covariances (Hair et al., 2009).
17
18 MDA is relatively robust to failures of normality if skewness, rather than outliers, causes the
rP
19
20
violation. Robustness is expected with 20 cases in the smallest group if there are only five or
21
22
fewer predictors (Tabachnick & Fidell, 2012). Therefore, in this case, robustness to any failures
ee
23
24
25 in normality of the residuals should be expected.
26
rR

27
28 Results
29
30 Altman’s Z”-Score model. Using the Altman Z”-Score model, a firm with a score under 1.1
ev

31
32 is classified as bankrupt. A firm with a score over 2.6 is classified as non-bankrupt. A firm
33
34
whose score is between 1.1 and 2.6 is classified as belonging to the “grey area.” Using 1998-
ie

35
36
37 2003 data, this model did an unsatisfactory job of predicting airline bankruptcy. Overall accuracy
w

38
39 was only 57.5 percent over the entire time period, which is little better than a coin toss,
40
On

41
42 Alternate model. An alternate three-variable model for predicting airline firm bankruptcy
43
44 was created, using 1998-2003 data for an in-sample analysis. Of the financial ratios tested, it was
45
ly

46
47
found that only three variables were statistically significant as predictors and did not have a
48
49 collinearity issue. Data from 2004-2005 was used as a hold-out sample and to verify the
50
51 prediction accuracy of the original model.
52
53
54 The alternate model is as follows:
55
56 Ya = .268X1 + .838X2 +.111X3 + έ
57
58
59
60
11
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 12 of 24

Improving airline bankruptcy prediction 12

1
2
3
4 Where:
5
6 X1 = working capital/total assets;
7
8
9 X2 = retained earnings/total assets;
10
11 X3 = book value of equity/total liabilities;
12
13 έ = error term; and,
14
Fo
15
16 Ya = overall index.
17
18 Using 1998-2003 in-sample data, 82.5 percent of the bankrupt airline firms were correctly
rP
19
20
classified as bankrupt, and 73.9 percent of the non-bankrupt firms were correctly classified as
21
22
non-bankrupt. Overall, the accuracy was 80.9 percent. This result represented a considerable
ee
23
24
25 improvement over the Z”-Score model’s 57.5 percent. The alternate model was then tested using
26
rR

27
28 the holdout sample of 2004-2005 data. The model correctly predicted bankruptcy 93% of the
29
30 time one year ahead of the event, 87% of the time two years ahead of the event, and 73% of the
ev

31
32 time for three years ahead. The prediction accuracy, using the holdout sample, was even better
33
34
than that of the in-sample data. This implies that the alternate model could do a respectable job in
ie

35
36
37 predicting future bankruptcy and could be a useful management tool. Table 1 displays the
w

38
39 number of years, in advance, that each airline received a bankruptcy classification. A negative
40
On

41
42 score indicates that the airline was classified as bankrupt.
43
44 Insert Table 1
45
ly

46
47
Next, the scores for each non-bankrupt airline during the model development and testing
48
49 period are displayed in Table 2. A positive score indicates that the airline was classified as non-
50
51 bankrupt, according to the alternate model. Please note that although American Airlines
52
53
54 received negative scores for 2004-5, it did not file for bankruptcy until 2011.
55
56 Insert Table 2
57
58
59
60
12
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 13 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 13

1
2
3
4 Testing the alternate model with recent data. The financial data for four mega-carriers
5
6 and two large national carriers were entered into the model to test its robustness over time. Table
7
8
9 3 displays the 2015 and 2016 scores using the most recent financial data available for American,
10
11 Delta, Southwest, United, Alaska, and JetBlue. A positive score indicates that the airline was
12
13 classified as non-bankrupt, according to the alternate model.
14
Fo
15
16 Insert Table 3
17
18 Please note that American Airlines a received negative score in 2015. It had large net losses
rP
19
20
each year from 2009 – 2013. It declared bankruptcy in 2011 and merged with US Airways in
21
22
2013. Although the company turned a profit in 2014 and 2015, its retained earnings were still
ee
23
24
25 negative in 2015. This resulted in a negative alternate model score for 2015. Its retained earnings
26
rR

27
28 became positive in 2016, and it achieved a positive score.
29
30 Frontier has been owned by an affiliate of Indigo Partners, LLC, a private equity firm, since
ev

31
32 December 2, 2013 (Drum, 2013); its financial statements are no longer publicly available.
33
34
Frontier has a small market share, less than 3% (Bureau of Transportation Statistics, 2017). Air
ie

35
36
37 Canada is now ACE Aviation. The company’s financial statements are no longer available on
w

38
39 www.sec.gov and the stock is no longer publicly available for sale in the US. Therefore, Air
40
On

41
42 Canada and Frontier no longer meet the study’s requirements and have been dropped from the
43
44 analysis.
45
ly

46
47
Chi Square Analysis and Hypothesis Testing
48
49 Next, the two models were tested for statistical significance. The data results were tested
50
51 using chi-square analysis to see how unlikely the observed value is if the null hypothesis is true.
52
53
54 The chi-square test’s assumptions are: 1) the categories of a variable do not overlap, 2) most of
55
56
57
58
59
60
13
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 14 of 24

Improving airline bankruptcy prediction 14

1
2
3
4 the expected counts must be greater than five; and, 3) none of the expected counts can be fewer
5
6 than one. These assumptions were met.
7
8
9 First, the Altman Z”-Score model was tested to determine if it classified the companies’
10
11 bankruptcy status better than a naïve prediction. Hypothesis H10 states that there is no
12
13 relationship between the Altman Z”-Score model and the likelihood of bankruptcy for an airline
14
Fo
15
16 firm. The critical chi-square value (p <.05, df 1) was not met. The Altman Z”-Score model
17
18 failed, and hypothesis H10 is not rejected. There is no relationship between the Altman Z”-Score
rP
19
20
and the likelihood of bankruptcy for an airline firm, therefore the alternate hypothesis is rejected.
21
22
Next, the alternate model was tested. The second hypothesis states that a revised bankruptcy
ee
23
24
25 prediction model is no better than the Altman Z”-Score model in predicting the likelihood of
26
rR

27
28 bankruptcy for an airline firm. The critical chi-square value of 7.87944 was achieved (p < .005,
29
30 df 1), thus rejecting H20. Therefore, the alternate hypothesis is accepted, since the alternate
ev

31
32 model is better than the Altman Z”-Score model in predicting the likelihood of bankruptcy for an
33
34
airline firm.
ie

35
36
37 Key Variables. The alternate model calculates a score, which, if negative, indicates a
w

38
39 classification of bankruptcy and, if positive, the classification is non-bankruptcy. In the alternate
40
On

41
42 model, the most important predictor of bankruptcy is the variable that represents retained
43
44 earnings divided by total assets. This variable corresponds with the largest coefficient, 0.838,
45
ly

46
47
making it the most important predictor of bankruptcy. It makes intuitive sense that negative
48
49 retained earnings would spell financial distress for an airline. A firm cannot sustain net losses
50
51 for an extended amount of time without failing.
52
53
54 As shown in Table 1, the alternate model predicted that Air Canada, Hawaiian, Northwest,
55
56 and US Airways would go bankrupt four years in advance, that America West and TWA would
57
58
59
60
14
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 15 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 15

1
2
3
4 go bankrupt three years before they did, that Delta and ATA would go bankrupt two years prior
5
6 to the actual event, and that United would go bankrupt one year prior to its actual bankruptcy
7
8
9 filing. These results constitute good prediction accuracy for up to four years before the actual
10
11 event occurs.
12
13 Discussion
14
Fo
15
16 The transportation industry is critical to the economy to the United States. Hospitality and
17
18 tourism, for example, rely on the movement of consumers as a crucial part of their business.
rP
19
20
Reliable, affordable air transport allows people to conduct business effectively and enjoy leisure
21
22
activities away from home. The airline industry employs millions of people, directly and
ee
23
24
25 indirectly. There are many stakeholders in the future of the airline industry including travelers,
26
rR

27
28 employees, and investors.
29
30 Some airlines have responded to their financial troubles by merging. America West acquired
ev

31
32 US Airways, which then merged with American Airlines; Continental merged with United;
33
34
Northwest merged with Delta; Southwest acquired AirTran; and Alaska Airlines acquired Virgin
ie

35
36
37 America (Denning, 2011). By 2015, four mega-carriers made up a combined 80% of all U.S.
w

38
39 flights (Groden, 2015). The industry’s financial health has improved to the point where, in 2016,
40
On

41
42 Warren Buffett (who called the airline industry a financial investment death trap in 2013)
43
44 invested in the four mega-carriers: American, United, Delta, and Southwest (La Monica, 2017).
45
ly

46
47
There is also a not-so-subtle irony about bankruptcy: it costs money. And the bigger the
48
49 company, the more it costs. Filing for Chapter 11 cost AMR hundreds of millions of dollars in
50
51 attorney and other professional fees. There is also the stigma that is attached to bankruptcy.
52
53
54 Nobody wants to work for a bankrupt company, and few CEOs want to be in charge of a
55
56 bankrupt company (Neuman, 2011)
57
58
59
60
15
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 16 of 24

Improving airline bankruptcy prediction 16

1
2
3
4 This study tested two corporate bankruptcy prediction models. The Altman Z”-Score model
5
6 was tested for its capacity to predict airline firm bankruptcy, using financial statements from the
7
8
9 period 1998-2003. A new model was created, using three of the four Altman Z”-Score model’s
10
11 predictor variables. Both of these models were compared against the results of a naïve
12
13 prediction. The Altman Z”-Score model performed no better than a naïve prediction in
14
Fo
15
16 predicting airline firm bankruptcy.
17
18 The new three-variable model, however, could predict airline firm bankruptcy up to four
rP
19
20
years before the actual event. The alternate model not only performed better than the Z”-score
21
22
model, but also was simpler, since it used fewer variables. Further, it predicted membership in
ee
23
24
25 one of only two groups (rather than three under the Z”-Score model), and used a single score (of
26
rR

27
28 zero) as a cut-off to distinguish whether a firm belonged to the bankrupt group or the non-
29
30 bankrupt group. Therefore, the alternate model emerges as a more sophisticated and practical
ev

31
32 model at the same time, having major implications as a bankruptcy alert tool for the airline
33
34
industry and any other industry that wishes to adapt it. Furthermore, this study response to
ie

35
36
37 Altman and Hotchkiss (2006) suggestion that there is a need for further development of
w

38
39 bankruptcy prediction models for specific industries.
40
On

41
42 Limitations of the Study and Suggestions for Future Research
43
44 There are several limitations to this study. First, this analysis was limited by the availability
45
ly

46
47
of financial data on airlines. Only publicly-traded corporations are required to make their
48
49 financial statements available to everyone, via filings with the Securities Exchange Commission.
50
51 Therefore, only publicly-traded airlines were part of this study. The data used in this study was
52
53
54 limited to financial statements that are available in filings with the SEC.
55
56
57
58
59
60
16
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 17 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 17

1
2
3
4 A second limitation is the variation in operating models among the sample’s airlines. For
5
6 example, some airlines lease their jets, which are the most important assets for an airline. Other
7
8
9 airlines purchase their jets, using long-term debt financing. This leasing versus ownership
10
11 difference may have an impact on the presentation of an airline’s balance sheet accounts.
12
13 A third limitation was the use of ratio analysis. Ratios are extremely useful to owners,
14
Fo
15
16 creditors, and management in evaluating the financial condition of airlines. Ratios, however, are
17
18 only indicators. They do not reveal a problem’s source, just indicate that indicate that there may
rP
19
20
be a problem (Beaver, 1966). Future models should develop a solution to not only detect a
21
22
problem, but also pinpoint the source of the problem to permit companies to take corrective
ee
23
24
25 action. Financial ratios appear to work well to help predict bankruptcy up to four years ahead of
26
rR

27
28 the event. Future studies, however, could use these methods to predict bankruptcy in other types
29
30 of businesses.
ev

31
32 In summary, this study has shown that financial ratios can be used to predict airline firm
33
34
bankruptcy. The accuracy of a traditional model was tested. The traditional model did not
ie

35
36
37 predict airline firm bankruptcy accurately. A new, simpler alternate model was developed,
w

38
39 which could predict airline firm bankruptcy up to four years ahead of the actual event.
40
On

41
42
43
44 References
45
ly

46
47
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate
48
49 bankruptcy. Journal of Finance, 23, 589-609.
50
51 Altman, E. I., & Hotchkiss, E. (2006). Corporate Financial Distress and Bankruptcy: Predicting
52
53
54 & Avoiding Bankruptcy, Analyze and Invest in Distressed Debt. Hoboken, NJ: John
55
56 Wiley & Sons, Inc.
57
58
59
60
17
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 18 of 24

Improving airline bankruptcy prediction 18

1
2
3
4 Beaver, W. H. (1966). Financial ratios as predictors of failure. Journal of Accounting Research,
5
6 4 (Supplement), 71-111.
7
8
9 Bureau of Transportation Statistics. (2017). Retrieved from www.transtats.bts.gov
10
11 Cheng, L. T. W., & McDonald, J. E. (1996). Industry structure and ripple effects of bankruptcy
12
13 announcements. The Financial Review, 31, 783 – 807.
14
Fo
15
16 Chung, C.C. & Szenberg, M. (1996). The effects of deregulation on the U.S. airline industry.
17
18 Journal of Applied Business Research 12(3), 133-140.
rP
19
20
Davalos, S., Gritta, R. D., & Chow, G. (1999). The application of a neural network approach to
21
22
predicting bankruptcy risks facing the major US air carriers: 1979-1996. Journal of Air
ee
23
24
25 Transport Management, 5, 81-86.
26
rR

27
28 Denning, L. (2011, November 30). American Airlines takes evasive action. Wall Street
29
30 Journal. Retrieved from
ev

31
32 https://www.wsj.com/articles/SB10001424052970203441704577068422381677392
33
34
Drum, B. (2013, December 4). Frontier Airlines is now owned by Indigo Partners [World
ie

35
36
37 Airline News]. Retrieved from https://worldairlinenews.com/category/frontier-airlines-
w

38
39 2nd/page/8/
40
On

41
42 Federal Aviation Administration. (2014). FAA aerospace forecast: Fiscal years 2015 – 2035.
43
44 Retrieved from https://www.faa.gov/data_research/aviation/
45
ly

46
47
Golaszewski, R. & Saunders, M. (1992). Financial stress in the U.S. airline industry. Journal of
48
49 the Transportation Research Forum, 32, 313-319.
50
51 Gritta, R. D. (1974). The impact of capitalization of leases on financial analysis: A case study in
52
53
54 air transport. Financial Analysts Journal, 30(2), 47 – 52.
55
56
57
58
59
60
18
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 19 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 19

1
2
3
4 Gritta, R. D. (1982). Bankruptcy risks facing the major US airlines. Journal of Air Law &
5
6 Commerce, 48, 89- 108.
7
8
9 Gritta, R. D., Adrangi, B., Adams, B., & Tatyanina, N. (2008). An update on airline financial
10
11 condition and insolvency prospects using the Altman Z” score model. Transportation
12
13 Research Forum, 47, 133 – 138.
14
Fo
15
16 Gritta, R. D., Wang, M., Davalos, S., & Chow, G. (2000). Forecasting small air carrier
17
18 bankruptcies using a neural network approach. Journal of Financial Management and
rP
19
20
Analysis, 13, 44-49.
21
22
Groden, C. (2015, July 23). These cities have lost the most flights. Fortune. Retrieved June 24,
ee
23
24
25 2017, from http://fortune.com/2015/07/23/fewer-flights-airline-consolidation/
26
rR

27
28 Gudmundsson, S. (1999). Airline failure and distress prediction: a comparison of quantitative
29
30 and qualitative models. Transportation Research, Part E 35, 155-182.
ev

31
32 Gudmundsson, S. (2002). Airline distress prediction using non-financial indicators. Journal of
33
34
Air Transportation, 7, 3-24.
ie

35
36
37 Gu, Z. (2002). Analyzing bankruptcy in the restaurant industry: A multiple discriminant model.
w

38
39 Hospitality Management, 21, 25-42.
40
On

41
42 Gu, Z., & Gao, L. (2000). A multivariate model for predicting business failures of hospitality
43
44 firms. Tourism and Hospitality Research, 2, 37-49.
45
ly

46
47
Hair, J. F. Jr., Black, W. C., Babin, B. J., & Anderson, R. E. (2009). Multivariate Data Analysis
48
49 (7th Ed.). Englewood Cliffs, New Jersey: Pearson.
50
51 Jayanti, R. K., & Jayanti, S. V. (2011). Effects of airline bankruptcies: An event study. Journal
52
53
54 of Services Marketing, 25, 399-409.
55
56
57
58
59
60
19
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 20 of 24

Improving airline bankruptcy prediction 20

1
2
3
4 Kim, H., & Gu, Z. (2006a). Predicting restaurant bankruptcy: A logit model in comparison with
5
6 a discriminant model. Journal of Hospitality & Tourism Research, 30, 474 – 493.
7
8
9 Kim, H., & Gu, Z. (2006b). A logistic regression analysis for predicting bankruptcy in the
10
11 hospitality industry. Journal of Hospitality Financial Management, 14, 17–34.
12
13 Kim, S. Y. (2011). Prediction of hotel bankruptcy using support vector machine, artificial
14
Fo
15
16 neural network, logistic regression, and multivariate discriminant analysis. The Service
17
18 Industries Journal, 31, 441-468.
rP
19
20
Kroeze, C., & Mayer, K. (2006, October). Predicting airline corporate bankruptcies using a
21
22
modified Altman z-score model. Paper presented at the EuroCHRIE Conference,
ee
23
24
25 Thessaloniki, Greece.
26
rR

27
28 La Monica, P. (2017). Warren Buffett’s big bet on airlines flies high. CNN. Retrieved June 26,
29
30 2017, from http://money.cnn.com/2017/01/05/investing/airlines-warren-
ev

31
32 buffett/index.html
33
34
Neuman, S. (2011, November 29). Is bankruptcy ‘business as usual’ for airlines? NPR. Retrieved
ie

35
36
37 June 27, 2017, from http://www.npr.org/2011/11/29/142909836/is-bankruptcy-business-
w

38
39 as-usual-for-airlines
40
On

41
42 Patterson, D. W. (2001). Bankruptcy prediction: A model for the casino industry [Doctoral
43
44 dissertation], University of Nevada, Las Vegas.
45
ly

46
47
SAS Institute Inc. (2017). Using Holdout Samples. Retrieved June 28, 2017 from
48
49 http://support.sas.com.
50
51 Scaggs, M. & Crawford, P. (1986). Altman’s corporate bankruptcy model revisited: Can airline
52
53
54 bankruptcy be predicted? Review of Regional Economics and Business, 11, 11-17.
55
56
57
58
59
60
20
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 21 of 24 The Journal of Hospitality Financial Management

Improving airline bankruptcy prediction 21

1
2
3
4 Tabachnick, B. G., & Fidell. L. S. (2012). Using Multivariate Statistics, 6th Edition. Upper
5
6 Saddle River, NJ: Pearson.
7
8
9 United Airlines 2002 Annual Report (2003). Retrieved from www.united.com/ual/en/us/.
10
11 Youn, H., & Gu, Z. (2010). Predicting Korean lodging failures: An artificial neural network
12
13 model along with a logistic regression model. International Journal of Hospitality
14
Fo
15
16 Management, 29, 120-127.
17
18
rP
19
20
21
22
ee
23
24
25
26
rR

27
28
29
30
ev

31
32
33
34
ie

35
36
37
w

38
39
40
On

41
42
43
44
45
ly

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
21
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 22 of 24

1
2
3
Table 1: Alternate Model Results for Bankrupt Firms
4
5
6 Airline Bankruptcy Number of Score, year Score, one Score, two Score, three Score, four
7 declared in years in that year before years before years before years before
8 this year advance that bankruptcy bankruptcy bankruptcy bankruptcy bankruptcy
9 bankruptcy was
10 was declared
11 correctly
12 predicted
13 ATA 2004 2 -1.3 -.16 -.16 .02 .05
14 Air Canada 2003 4 -.07 -.45 -.03 -.09 -.08
Fo
15 America West 2005 3 -.01 -.01 -.05 -.12 .05
16 Delta 2005 2 -.43 -.23 -.07 .03 .09
17 Hawaiian 2003 4 -.36 -.5 -.2 -.27 -.08
18 Northwest 2005 4 -.37 -.16 -.08 -.11 -.04
rP
19 TWA 2001 3 -.6 -.5 -.41 -.2 n/a
20 US Airways 2004 4 -.05 -.03 -.67 -.39 -.09
21 United 2002 1 -.14 -.02 .08 .09 .03
22
ee
23
24
25
26
rR

27
28
29
30
ev

31
32
33
34
ie

35
36
37
w

38
39
40
On

41
42
43
44
45
ly

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
Page 23 of 24 The Journal of Hospitality Financial Management

1
2
3
Table 2: Alternate Model Results for Non-Bankrupt Firms
4
5
6
7 Airline Score, 2005 Score, 2004
8 Air Tran 0.14 0.30
9
Alaska 0.20 0.20
10
11 American -0.04 -0.01
12 Continental 0.07 0.06
13 Frontier 0.20 0.23
14 JetBlue 0.06 0.09
Fo
15 Southwest 0.42 0.44
16
17
18
rP
19
20
21
22
ee
23
24
25
26
rR

27
28
29
30
ev

31
32
33
34
ie

35
36
37
w

38
39
40
On

41
42
43
44
45
ly

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu
 The Journal of Hospitality Financial Management Page 24 of 24

1
2
3
Table 3: Recent Alternate Model Results for the Largest Remaining Airlines
4
5
6
7 Airline Score, 2016 Score, 2015
8 Alaska 0.33 0.43
9
American 0.02 -0.03
10
11 Delta 0.12 0.07
12 Hawaiian 0.16 0.24
13 JetBlue 0.28 0.20
14 Southwest 0.20 0.23
Fo
15 United 0.06 0.09
16
17
18
rP
19
20
21
22
ee
23
24
25
26
rR

27
28
29
30
ev

31
32
33
34
ie

35
36
37
w

38
39
40
On

41
42
43
44
45
ly

46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
URL: http://mc.manuscriptcentral.com/jhfm Email: sheel@isenberg.umass.edu