Restructuring Coal: A Natural Experiment using the

Alpha Natural Resources’ 2015 Bankruptcy

______

A THESIS

Presented to

The Faculty of the Department of Economics and Business

The Colorado College

In Partial Fulfillment of the Requirements for the Degree

Bachelor of Arts

By Sophie Leamon

May 2018

Restructuring Coal: A Natural Experiment using the

Alpha Natural Resources’ 2015 Bankruptcy

Sophie Leamon

May 2018

Mathematical Economics

Abstract

This study identifies the problem that coal companies are not responding effectively to declining demand for coal. The resultant bankruptcies have been numerous and costly, and the restructuring undergone has proven inadequate. This paper seeks to aid successful restructuring through estimating mine profitability in the coal industry. These estimates are done by proxying withheld financial information using the Alpha Natural Resources’ bankruptcy as a natural experiment. A probit model is used to estimate the profitability of the 4,160 U.S. coal mines that have been in production since 2000. The results can be used to identify mines that are likely unprofitable.

KEYWORDS: (Coal, Energy, Bankruptcy, Restructuring)

JEL CODE: A22, E27, H1

ON MY HONOR, I HAVE NEITHER GIVEN NOR RECEIVED UNAUTHORIZED AID ON

THIS THESIS

______

Signature TABLE OF CONTENTS

ABSTRACT

1 INTRODUCTION…………………………………………………………….……….….1

2 BACKGROUND……………………………………………………………………….....3 2.1 Coal Demand……………………………………...…………...………….3 2.2 Liabilities………………………………………………..………………...3 3 LITERATURE REVIEW………………………………………………………..………..6 3.1 Mine Cost Estimations…………………………………………………….6 3.2 Labor and Unions………………………………...………………………..6 3.3 Effects on Productivity………………………………………………...…..7 3.4 Coal Rank and Distribution……………………………………………..…7 3.5 Alpha Natural Resources Bankruptcy……………………………………..8 4 DATA AND METHODOLOGY...…………………………………………….….…..…13 4.1 Mine Data and Health Administration………………………………...….13 4.2 Energy Information Agency………………………………………………14 4.2.1 Reserves……………………………………………………………14 4.2.2 Projected Revenue…………………………………………………15 4.3 Variable Omissions…………………………….………………….………16 5 RESULTS ………………………………………………………………………..….……18 5.1 Goodness of Fit…………………………………………………………....18 5.2 Regression Analysis…………………………………………………….....18 5.3 Marginal Effects………………………………………………………...... 19 5.4 Profitability Tables………………………………………………………...20 5.4.1 James C. Justice II………………………………………………….21 5.4.2 Coronado Coal LLC………………………………………..………22 5.4.3 Edward A. Asbury………………………………………………….22 5.4.4 NACCO Inc. ……………………………………………………….23 5.4.5 Patriot Coal Corporation………………………………………...…24 5.4.6 Peabody Energy……………………………..…………………..…25 6 CONCLUSION……………………………………………………………………...….....26 7 REFERENCES………………………………………………………………………...…..27

Introduction

Demand for coal in both thermal and metallurgical sectors has decreased since 2011 and the coal industry has not responded well. Fifty U.S. coal companies have filed for bankruptcy since 2012 with combined debt of over $30 billion (McGinley, 2018).

These bankruptcies are a product of the industry’s inability to restructure effectively in response to negative shifts in demand. The issue is that closing coal mines is expensive: companies must first pay reclamation costs to return the mines to a natural or economically useable state. However, allowing a mine to continue producing, even if unprofitable, allows companies to continue returning profits to bond holders in the short-term. While the long-term health of a company requires profit, it has proven to be in coal executives’ best interest to continue producing at a marginal loss rather than face liabilities head on (Rehbach, 2016). Coal companies are looking at $70 billion dollars in liabilities; without rapid restructuring, this debt will not be servicable by the industry and fall to taxpayers instead (Rehbach, 2016).

The problem that this paper identifies is that coal companies are not restructuring effectively on their own, and the financial information that would allow regulation of this problem is not available.

Financial information is not disclosed until a company files for bankruptcy, which means that oversight can only happen once the company seeks government aid in managing debt. This study seeks to answer a simple question: based on publicly available information, is it possible to identify which mines are profitable and those which are unprofitable? This paper attempts to proxy profitability through the use of a natural experiment in which a bankruptcy settlement gave indirect insight into mine profitability.

In 2015, Alpha Natural Resources declared bankruptcy and during restructuring became two companies. Alpha was left with the company’s debt and a majority of mines, while Contura received seven mines, their associated facilities, and all of Alpha’s executives. Three months later, Alpha was sued by the Department of Environmental Protection (DEP) for over-reporting their projected cash flow by over $100 million, suing on the grounds that the new Alpha’s projected

1 revenue could not cover its liabilities. The DEP said that the transaction would transfer the debtors’ most valuable assets to Contura Energy while leaving the less valuable assets and significant environmental liabilities behind (Roberts, 2016). In their official statement, the DEP argues that senior management, none of them liable for the health of the now gutted Alpha Natural Resources, never intended for Alpha to succeed financially(“Response of the WVDEP”, 2016).

Under the assumption that the Contura mines were selected because they are more profitable, this experiment is used to understand mine profitability in the industry as a whole. Using this information and predictions from the U.S. Energy Information Administration (EIA), this study attempts to capture the effects that distinguish the two companies. The resulting model is then applied to all currently operating mines in the U.S. to estimate the probability that those mines are turning a profit.

2 Background

2.1 Coal Demand

The combination of low-cost natural gas from shale extraction, decreased export demand, and international environmental regulation has led to dramatic decreases in demand for coal after peak production levels in 2008. In 2014, the biggest three producers of the U.S. coal industry,

Peabody Energy, Arch Coal, and Alpha Natural Resources saw their combined market cap fall from $6 billion to $350 million, a decrease of 94%(Kuykendall, 2016). This trend was never reversed; this is the fifth consecutive year of falling prices and volume for coal producers. The coal mining industry has undergone mine downsizing and closures equivalent to 200 million tons per year, or 19%, from peak output in 2008 (Kuykendall, 2016). However, production capacity has not been declining fast enough and this has pushed coal prices ever lower.

Fifty U.S. coal companies have filed for bankruptcy since 2012 (McGinley, 2018) with three of the four largest mining companies declaring bankruptcy since 2013. Over 44% of coal mined in the United States is now mined by companies that have filed as bankrupt (Kuykendall,

2016). These companies have sought to restructure $30 billion in debt in their filings, with the recent Peabody bankruptcy alone seeking to restructure $8.4 billion (Kuykendall, 2016).

2.2 Liabilities

A large portion of this debt is defined as ‘reclamation costs.’ Under the Surface Mining

Control and Reclamation Act of 1977 (SMCRA), coal mines are required to provide financial assurances, called reclamation bonds, to guarantee that strip mines will be cleaned up or reclaimed, even if their owners declare bankruptcy. To ensure that this happens, coal companies can either back up these bonds with sureties or set aside cash or other assets to pay for the cleanup before beginning the project. However, in the case that companies are deemed to be

“financially healthy” (Williams-Derry, 2018), they are allowed to self-bond. The standards for financial health rely on the company’s balance sheet, which often does not reflect true market value. In Arch Coal’s case, in 2014, the total market value of Arch Coal stock sat below $70

3 million, but the company’s most recent balance sheet showed stockholder equity of over $1.6 billion (Williams-Derry, 2018). Loopholes such as this have allowed many of the biggest coal mining companies to sign off on their ability to pay for reclamation and many to reach bankruptcy before the agreement was ever reevaluated.

There is fear that coal companies will be able to transfer their liabilities for reclamation to taxpayers. Though the U.S. Congress passed the SMCRA for just such a scenario, EPA Head

Scott Pruitt ended this practice in 2017, saying that, “Additional financial assurance requirements are unnecessary and would impose an undue burden” (Preston, 2017). However, as it turns out, reclamation costs have shifted to taxpayers in bankruptcy settlements. Between Arch Coal and

Alpha Natural Resources, the two underwent bankruptcy settlements with $896 million in outstanding reclamation liabilities and bankruptcy courts have reduced their liability to around

$136 million, leaving roughly $760 million in cleanup costs that could fall to taxpayers

(Alexander, 2016).

For many companies, fronting the reclamation costs for closing an individual mine has proven a larger obstacle than continuing to operate below cost. Consequently, the industry has preferred to operate mines below average total cost and continue incurring debt (Rehbach and

Samek, 2017). In addition to reclamation costs, Rehbach and Samek (2017) outlined the extent of the industry’s financial liabilities, saying that, even after restructuring, the U.S. coal industry would have remaining liabilities of about $70 billion, a sum that far outweighs the industry’s profit-making ability. Based on their analysis, it costs the industry about $9 to $10 billion a year just to service these liabilities, which comes out to $10 a ton of coal produced. As these current bankruptcies show, the industry no longer has the cash-generation capability to pay these obligations, and the shortfall is only getting bigger (Rehbach and Samek, 2017). This need for cash has resulted in the decision by many coal companies to continue unprofitable production in order to return investments to bond holders. The result, the authors say, “is that the United States

4 is home to a collection of ‘zombie mines’ that cannot turn a profit but are too costly to close”

(Rehbach and Samek, 2017).

This is the effect that this paper attempts to capture. If these ‘zombie mines’ can be identified, effective restructuring can be implemented, and a healthy industry can attempt to regain appropriate demand capacity.

5 Literature Review

This review will focus on four important concepts in the mining industry: labor, cost, productivity, and coal type. This review finds these concepts essential to identifying profitable production. Finally, as this study rests on the assumption that Alpha Natural Resources’ bankruptcy holds as a good proxy for profitability, that will be further discussed in this section.

3.1 Mine Cost Estimations

Estimating costs for coal mines is most typically in the form of a predictive model for newly proposed mines. Budeba et Al. (2015) models the costs for proposed mines. This paper proposes an approach for positioning a mine to enter into the market at the lowest possible marginal cost curve. This research posits that proposed mines need to be both efficient and low- cost. The authors propose that this is achieved by locating a mine on easily extracted and high quality coal. Similarly, Schneider and Torries (1991) state that the cost of producing coal of a specific quality depends on a combination of geological conditions, that is, the quality of the unprocessed coal and the cost of extraction. Dehghani and Ataee-pour (2012) state that costs should be estimated using a method that will incorporate the effect of deposit-specific variables

(such as quality and geography) as well as other external variables (such as policy and inflation) that affect costs during operation. Shafiee et Al. (2009) uses deposit average thickness, stripping ratio, capital cost and daily production rate as the independent variables to explain average operating costs.

3.2 Labor and Unions

Stoker (2005) found that mine output is pre-determined relative to labor. This is a contracting practice in mining, where labor is set endogenously to produce necessary output at minimum cost to fulfill a contract with a specific buyer. Boal and Pencavel (1994) modeled the number of miners employed to understand the impact of unionization on wage and employment.

Boal and Pencavel(1994) measured negligible differences in union-nonunion employment in coal

6 mining but found that operating days in unionized mines were about 25 percent below those in nonunion mines.

3.3 Effects on Productivity

Lakhani (2003) determined that the most significant contributor to productivity in the coal industry is seam thickness. These deposits vary in size, shape, and accessibility (Stoker, 2005), and the larger the seam is and closer it lies to the surface, the easier it is to extract. Seam thickness is largely geographical, with wider seams occurring in Western states. Non-geographic productivity rests on competitive ownership and utilization of capacity (Lakhani 2003). The variables explaining decreases in productivity are increased use of semi-skilled labor, low strip ratios (in surface mines), mine accidents, and coal mine regulations. Kuby (2001), found that coal mining closures have boosted national productivity by getting rid of less productive mines, but also that additional productivity is gained when some of their market share is taken up by higher productivity mines, and gained again as those mines achieve higher economies of scale.

3.4 Coal Rank and Distribution

Coal is categorized by its rank. Different coals have various heating potential as measured in British thermal units (Btu). Rank refers to the steps in “coalification,” during which buried plant matter changes into a more dense, more carbon rich, and harder material.

“Table 1. Coal Rank”

% weight Anthracite Bituminous Sub-Bituminous Lignite Heat Content 13,000-15,000 11,000-15,000 85,00-13,000 4,000-8,300 Fixed carbon <15% 2-15% 10-45% 30-60% (“Alaska Center for Energy”, 2011)

The distribution of coal rank depends greatly on region, with the producing more bituminous coal, while northeast Pennsylvania is one of the only places where Anthracite can be found.

7 Surface coal is largely produced in the Western states and is lower in energy content than

Appalachian coal. Despite lower energy content, Western surface-mined coal is easier to extract and considered cleaner because of its low sulfur content. To put the productivity effects of mining

Western coal seams in perspective, by 1994, labor productivity in Western surface mines was 3.5 times as great as nationwide coal mine productivity (Darmstadter, 1997). Though productivity is higher, lower energy content is reflected in short ton price. For example, in 1995, coal averaged

$10.15 per short ton in Western States, $18.81 in the Interior, and $27.45 in

(Darmstadter, 1997).

“Figure 1. Geographical Coal Distribution”

(Brady, 2018) 3.5 Alpha Natural Resources Bankruptcy

This model rests on the assumption that the mines that remain under Alpha Natural

Resources are both statistically different from, and less profitable than, those that were transferred to Contura Energy upon restructuring. This assumption will now be examined.

8 In Alpha’s restructuring, the United States Bankruptcy Court for the Eastern District of

Virginia approved the sale of coal assets to Contura Energy, Inc., a new company formed by a group of Alpha's first lien lenders. What was uncommon about this transaction is that Alpha’s top executives signed themselves, in addition to those assets, over to the new company. The DEP said that, Alpha’s bankruptcy restructuring “would transfer the debtors’ most valuable assets to

[Contura Energy], while leaving the less valuable assets—and the very significant environmental liabilities—behind in what presently is proposed to be an inadequately funded and infeasible reorganized entity” (“Response of the WVDEP”, 2016).

Under normal circumstances, bankruptcy settlements are intended to allow a company to relieve themselves of some debt in order to mitigate much larger debt taken on my the government in the future. For this reason, most bankruptcies settlements are a good example of healthy restructuring, as both sides of the table stand to lose if the company goes under. In this case however, Alpha executives had no reason to divide up assets to create two healthy companies, because in the restructuring they were released of any liabilities remaining at Alpha.

The settlement was meant to restructure Alpha into a viable company that would have the cash flow sufficient to cover its remaining obligations, but three months later, executives revealed that they had overstated Alpha’s cashflow by $100 million. The company as a whole has true estimated cash flows of less than 60% of what was stated upon restructuring. The DEP said that they would never have made the deal if the shortfall had been disclosed beforehand (“Response of the WVDEP”, 2016). In a process meant to ensure financial stabilitly, Alpha Natural

Resources is being sued by the DEP for fraud. To put it plainly, the DEP has little confidence, based on true financial projections, that the restructured Alpha Natural Resources has the ability to be a profiting company.

To understand the details of the settlement, we look to Alpha’s bankruptcy filings. Because we are hoping to look at individual mines’ productive capacity across companies for improved

9 restructuring, we are unconcerned with outstanding debt left in either of the two companies, but instead focus on projected productive capacity.

According to a citation taken from Alpha Natural Resource’s bankrupcty Form 8-K, the mines transferred to Contura Energy were the two mines from Pennsylvania, of the segment

“Northern Appalachia”, the two mines from Wyoming, of the segment “Powder River Basin”, two mines from Virginia, and one from West Virginia. That information is further summarized

Table 2, reformatted from information in Form 10-K.

“Table 2. Alpha Natural Resources Mine Classification”

COAL BASIN STATE TOTAL MINES 2014 PRODUCTION , Virginia, Central Appalachia West Virginia 56 33,897 Northern Appalachia Pennsylvania 2 13,454 (PW River Basin Wyoming 2 36,487 TOTALS: 60 83,538

(“Alpha Natural Resources”, 2016(b))

The mines transferred to Contura Energy were the two mines from Pennsylvania, from

Northern Appalachia, the two mines from Wyoming, from the Powder River Basin, two mines from Virginia, and one from West Virginia. According to Alpha’s Form 8-K, cited as originally sourced from the EIA 2014 Annual Energy Outlook, projected growth in the segments sold to

Contura, the mines in the Powder River Basin and Northern Appalachia, have increases in production of between 0.1 to 1.0% respectively over the next twenty years. The only segment remaining in Alpha Natural Resources, highlighted in red, has projected growth between -3.7 to -

1.0% over the next twenty years ((“Alpha Natural Resources”, 2016(b)).

10 “Table 3. Projected Coal Production by Region”

(Tons in millions) Preliminary Projected Projected Growth Rates Production by Region 2014 2019 2034 2019-2034 Powder River Basin 412 417 436 0.30% Central Appalachia 132 104 90 -1.00% Northern Appalachia 134 14 147 0.10% Illinois Basin 147 176 213 1.40% Other 207 216 221 0.40% Total 1032 1057 1107 (“U.S. Energy Deparment of Energy”, 2017(a

T-tests were used to compare Alpha Natural Resources to Contura Energy to see if there is a statistically significant difference between the projected revenue, production capacity, and remaining reserves of the two companies.

“Table 4. Two-Sample T Test (Projected Revenue)”

GROUP OBS MEAN STD. ERR. ALPHA 155 5.12738 0.02798 CONTURA 15 6.81277 0.56098

HA: DIFF < 0 0.0000 HA: DIFF > 0 1.0000

“Table 5. Two-Sample T Test (Productive Capacity)”

GROUP OBS MEAN STD. ERR. ALPHA 155 0.09669 0.00316 CONTURA 15 0.18925 0.10257

HA: DIFF < 0 0.0026 HA: DIFF > 0 0.9974

“Table 6. Two-Sample T Test (Reserves)”

GROUP OBS MEAN STD. ERR. ALPHA 155 6.2913 0.03439 CONTURA 15 6.7448 0.29515

HA: DIFF < 0 0.0009 HA: DIFF > 0 0.9991

11 The t-tests found that in comparisons of projected revenue (Table 4), production capacity

(Table 5), and mine reserves remaining in 2015 (Table 6), that the two groups: Alpha, and

Contura, are significantly different. Contura has statistically higher projected revenue and productive capacity, while mines with higher levels of coal reserves are more likely to fall under

Alpha Natural Resources. As a whole, this is sufficient evidence that the natural experiment can be used to further analyze profitability in the coal industry.

12 Data and Methodology

The purpose of this regression is to quantify the difference between the mines from Alpha and those from Contura. To do this, we use a probit regression test to model the binary response variable: whether a mine is under the control of Contura Energy or Alpha Natural Resources.

Controller_ID = α+ β1 Status + β2 High_Ouput##Reserves + β3 Low_Output##Reserves +

β4 Projected_Revenue + β4 Sum_Labor+ε

(3.1)

Controller_ID = {0 = Alpha Natural Resources, 1 = Contura Energy}

Status = {Abandoned and Sealed = 1, Abandoned = 2, Non-Producing = 3, Temporarily Idle = 4,

Active = 5, New Mine = 6}

Sum_Labor = Sum of labor hours (in thousands) from 2000-2016

High_Output = {0,1}, 1 if SUMLAB > 20,000

Low_Output = {0,1}, 1 if 5,000 > SUMLAB

Reserves = Percentage of remaining recoverable coal

Projected_Revenue = (EIA projected price in 2050)*(EIA projected regional demand)

4.1 Mine Safety and Health Administration

Mine data for this study comes from the Mine Safety and Health Administration (MSHA), an agency of the United States Department of Labor which serves as a regulatory body under the

Federal Mine Safety and Health Act of 1977. Since 2000, data on production, employment, labor hours, changes in mine ownership, and control was compiled. This data is stored under a unique mine identifier which has been used across Department of Labor mine datasets. In the model,

MSHA data was used to obtain the following variables:

Controller_ID = {0 = Alpha Natural Resources, 1 = Contura Energy}

13 Status = {Abandoned and Sealed = 1, Abandoned = 2, Non-Producing = 3, Temporarily Idle = 4,

Active = 5, New Mine = 6}

Sum_Labor = Sum of labor hours from 2000-2016

To account for the effect that unionizing has on operating days as identified by Boal and

Pencavel (1994), the author uses labor hours instead of short ton output in the model. The efficacy of this variable, as opposed to the standard use of output, is corroborated by Stoker

(2005).

4.2 Energy Information Agency

Data for projected price and remaining reserves comes from the 2014 EIA Short-Term

Energy Outlook (“U.S. Energy Department of Energy, 2017(d)), and data for projected production comes from the EIA Analysis and Projections (“U.S. Energy Deparment of Energy,

2017(c)). To account for differences in production associated with the previous discussion of coal rank and geographical productivity levels, price is used. Price does a good job balancing differences in coal across region. Price allows for differentiation between the low price of

Western coal and its relative ease of extraction, and more expensive and more difficult to extract, but higher energy content Appalachian coal. Because rank is reflected in price and output is reflected in productivity as suggested by Darmstadter (1997), this discrepancy can be accounted for.

Reserves. The reserve variable is a measure of recoverable coal reserves percentage. This variable is defined as the percentage of recoverable coal reserves at producing mines that represent the quantity of coal that can be mined from existing coal reserves at reporting mines, weighted for all mines in the reported geographic area. This data excludes mines producing less than 25,000 short tons, which are not required to provide data on refuse recovery to the EIA

(“U.S. Energy Deparment of Energy, 2017(e)). The lack of information from mines producing under 25,000 short tons means that there may be inaccuracies in reserve estimation for mines

14 producing under that amount yearly. This bias should serve as an overestimator for a mine’s worth, as larger mines tend to span greater area and could hold greater levels of reserves.

The interaction variables, High_Ouput##Reserves and Low_Output ##Reserves are used to understand the relationship between reserves and productivity. In regressions where the reserves was regressed alone, the model found that higher levels of recoverable reserves made a mine more likely to fall under Alpha Natural Resources. This proved problematic because reserves are of inherent value, however they are useful to production only when a mine is actually producing.

The interaction term is used to parcel the distinct effects associated with mines that have reserves and are highly productive opposed to those that have reserves and are producing very little.

Projected Revenue. This variable is an estimation mine’s projected revenue in 2050. Data for this variable comes from the EIA. The EIA produces an estimation for mine-mouth coal demand as in million short tons in 2050 (“U.S. Energy Deparment of Energy, 2017(c)). This data is sorted into three categories: Appalachia, Interior, and West, (see Table 7). Price is estimated by the EIA in 2015 dollars per short ton, and categorized into eleven geographical regions and the associated coal type.

It should be noted that this is a rough estimation for each individual mine. Because data is only available by geographical region, and at best state, it proves difficult to get a precise estimation for this value for each mine. It is important to note, however, that shifts in production are dramatic across geographic region, and this effect is worth capturing. Additionally, these projections were cited by Alpha in Form 8-K, indicating that Alpha could have been using the

EIA’s projected demand for coal to determine asset sales as well.

15 “Table 7. Coal Supply by Region”

COAL SUPPLY (million short tons) Production 2016 2050 2017 -2050 Appalachia 183 138 -1.1% Interior 147 240 1.4% West 411 368 -0.1% TOTAL 741 747 -0.1% (“U.S.” Department of Energy, 2017(c))

4.3 Omissions

Several variables that appear to be important were not included in the model because of a lack of available data. First, because it is difficult to ascertain financial information, the model stayed away from modeling profitability directly. The other method for estimating cost in previous models has been focused on proposed mines, where companies can position new operations geographically and technologically for profit. This study takes on a different task, which is to broadly describe profitability in pre-existing and currently operating mines. For this reason, this model focuses on demand and capacity and stays away from modelling costs directly.

Similarly, it proved next to impossible to model the impact of self-bonding on profitability.

The only data found for self-bonding comes from the 2014 Interstate Mining Compact

Commission. Self-bonding data is distinct to each state, and no database or record is kept by the overseeing body, the Office of Surface Mining Reclamation and Enforcement (OSMRE). The only public record of mine self-bonding was a survey taken in 2014 and it only includes self- bonding records from 3 states (Holmes, 2018). Even the data reported from those three states appears to be incomplete. Because this data was incomplete it was not used in the model, however, this is accounted for in several ways. One of the concerns with self-bonding is that mines that are temporarily inactive, or even abandoned, will stay in a status preceding

“Abandoned and Sealed,” regardless of production, profitability, or productivity. As a potential proxy for this information, mine status was used.

16 Bankruptcy data is equally difficult to obtain. Documents are cataloged in the Public

Access to Court Electronic Records (PACER), with documents filed under Employer

Identification Number (EIN). The EIN is a unique nine-digit number assigned by the Internal

Revenue Service to business entities, and can be found on company issued paychecks or W-2 forms. PACER charges $0.10 per view per page. These components insure that while this bankruptcy data is public information, it is difficult to obtain. Beyond that, it is not guaranteed that bankruptcy debt statements are correct in any case. For these reasons, neither bankruptcy data nor self-bonding data were used as estimator variables.

17 Results

This model was able to provide a description of coal companies by identifying their mines from most to least profitable. These results should be used to further investigate those companies with mines that have very small probabilities of profitability.

Probit is an econometric technique designed to deal with limited dependent variables. In this analysis, the dependent variable is {0,1} where 1 is a profitable mine and 0 is an unprofitable mine. The model captured the effects of all six predictor variables with significance and has goodness of fit measures that indicate that the model fits well.

6.1 Goodness of Fit

Three goodness of fit tests were assessed. The log likelihood is a measure of the probability that the data is observed given the parameter estimates. In the probit model, the log likelihood will always be negative and higher values, (i.e. closer to zero), indicating a better fitting model.

For this model, the log likelihood measure of -16.6326 indicates a model that fits well. The problem with the log likelihood measure is that it is dependent on sample size, (i.e. a log likelihood value will improve just by increasing the number of estimates, even if they do not better explain the data). Without further corroboration from other measures of fit, the log likelihood will not stand alone.

Similar to, the traditional R^2 value, McFadden's Rho squared value is used for models with restricted dependent variables. Rho squared is defined as the ratio of the maximized likelihood value from the fitted model and the maximized likelihood value from the intercept only model. If the model has no predictive ability, the ratio of the two log-likelihoods will be close to one, and the rho squared value will be close to zero. With a Rho-Squared value of 67% this model is a good fit for the data.

To test that the predictor variables are providing useful information, the Wald Chi test is used. The Wald test uses an F-Test with the null hypothesis that the coefficients are simultaneously equal to zero. If the test fails to reject the null hypothesis, this suggests that

18 removing the variables from the model will not disrupt the fit of that model, in other words, the estimators used to explain the data were not useful at explaining the data. In this model, with a

Wald Test with high coefficients and a low P-value (Prob>Chi^2 = 0.0000), this indicates that independent variables included are important to the fit of the model.

6.2 Regression Analysis

The tables below give the log likelihood and related marginal effects for the estimator variables:

“Table 8. Regression Output”

Regression: Log likelihood -16.632691 Wald Chi^2 55.15 Prob> Chi^2 0.0000 Psuedo R^2 0.6722 Num of Obs. 170

“Table 9. Marginal Effects”

variable dy/dx std. err z p>|z| status -0.0609 0.016 3.89 0.000 high_output 0.3459 0.162 2.14 0.003 low_output -0.2165 0.067 -3.22 0.003 projected_rev 0.1653 0.053 -2.67 0.000 sum_labor -0.0150 0.043 3.82 0.000 reserves -0.2165 0.617 -2.44 0.000

6.3 Marginal Effects

Each additional step in the process of mine retirement, i.e. “Abandoned” to “Abandoned and Sealed,” means that a mine is 6% more probable to fall under the dominion of Alpha Natural

Resources. Mines that have output of more than 10,000 labor hours per year (thousands) are 35% more likely to fall under Contura Natural Energy. A mine that utilizes less than 5000 labor hours per year (thousands) is 21% more likely to fall under Alpha Natural Resources. A mine with an additional percentage unit of recoverable reserves is 22% more likely to fall under Alpha Natural

19 Resources. A one unit increase in projected revenue indicates that a mine is 17% more likely to fall under Contura Energy. Finally, an increase in sum labor hours indicates that a mine is 1.5% more likely to fall under Alpha Natural Resources.

The explanatory variables with the largest marginal effects are the mines that are utilizing above 10,000 labor hours and those using less than 5,000. As expected, it appears that the greater the productive capacity, the more likely it is that a mine would fall under Contura Energy.

Additionally, the greater the labor hours utilized, the more likely the that the mine fall under

Alpha Natural Resources. This could be because mines moved to Contura are more productive per labor hour, or because in restructuring, the company is seeking to reduce costs by shifting reducing labor.

Those mines that were generated as having low percentage chance of true profitability but very high levels of reserves ought to be assessed further. This effect was accounted for by using interaction variables to parcel the effects on profitability of mines with high production and high levels of reserves, from those with low production and high reserves. However, this model may still underestimate profitability for mines with very high levels of reserves. Nonetheless, because this analysis is intended for company restructuring, and reserves are owned at a company level, this may not be so significant. For example, if high levels of reserves are present but the company is unprofitable elsewise, the mine should be retired and reserve assets sold, just as in the Alpha example. To be sure however, mines that retain high levels of reserves should be reassessed.

6.4 Profitability Tables

The following tables are samples from the full data set, sorted by mine controller. The tables contain the following information: {MSHA Mine ID, Mine Name, Controller Name, Estimated

Percent Chance of Profitability}. The blue graphic is a visual aid to percent profitability. The author has chosen six controllers as a sample and reproduced tables that contain every mine that the controller currently owns (as of March 13, 2017). These tables should be used to identify

20 those mines that are least profitable amongst a company’s roster of mines. The full dataset and results are available on request from the author.

6.4.1 James C Jusice II

21 6.4.2 Coronado Coal LLC

6.4.3 Edward A. Asbury

22 6.4.4 NACCO Industries Inc

23 6.4.5 Patriot Coal Corporation

24 6.4.6 Peabody Energy

25 Conclusion

The coal industry has already weathered more than fifty bankruptcies and continues to struggle to shift capacity to meet ever lower demand for coal. With $70 billion dollars in liabilities looming in

2020 (Rehbach and Samek, 2017), coal companies must eliminate mines that are not profitable to ensure that those liabilities will not fall to taxpayers. This process is difficult to enforce because financial information is not available until companies declare bankruptcy and seek government aid in managing debt. In order to predict which companies are profitable without needing any company information, this paper proxies profitability through the use of the Alpha Natural Resources’ 2015 bankruptcy in which the settlement allowed indirect profitability insight.

In 2015, Alpha Natural Resources declared bankruptcy. In their official statement, the DEP argues that senior management never intended for the restructured Alpha to succeed financially. Under the assumption that the Contura mines were selected because they are more profitable, this data was used to fit a regression predicting the profitability of a mine.

This model fitted a probit model to estimate this natural experiment using labor hours, remaining reserves, mine status, and projected revenue. This model was then used to estimate profitability of the 4,160 existing mines in the U.S. These results provide a profitability guide to coal companies for regulatory use. This data can be used to identify which mines in a company are most profitable, and those that have a very small likelihood of being profitable. This information should be used in restructuring to identify those mines that should to be closed and reclaimed first.

26 References: ______Alaska Center for Energy and Power. (2011, February 21). Overview of Alaska’s Coal Basins. Retrieved March 13, 2018, from http://energy-alaska.wikidot.com/overview-of-alaska-s- coal-basins Alpha Natural Resources Form 8-K (United States Bankruptcy Court for the Eastern District of Virginia July 27, 2016). Alpha Natural Resources Form 10-K (United States Bankruptcy Court for the Eastern District of Virginia July 27, 2016). Boal, W. M., & Pencavel, J. (1994, February 01). Effects of Labor Unions on Employment, Wages and Days of Operation: Coal Mining in West Virginia. Retrieved March 13, 2018. Brady, L. (2010, November 06). Kansas Coal Resources and Production. Retrieved March 13, 2018, from http://www.kgs.ku.edu/Publications/Bulletins/GB4/02_coal.html Budeba, M. D., Joubert, J. W., & Webber-Youngman, R. C. (2015, November). A proposed approach for modelling... Retrieved March 13, 2018, from http://www.scielo.org.za/pdf/jsaimm/v115n11/18.pdf Darmstadter, J. (1997). Productivity Change in U.S. Coal Mining. Resources for the Future. Retrieved March 12, 2018, from https://pdfs.semanticscholar.org/95a4/df9f97989d4d3c11290b8864e1d268002c26.pdf Dehghani, H. and Ataee-Pour, M. (2012). Determination of the effect of operating cost uncertainty on mining project evaluation. Resources Policy, Retrieved March 12, 2018. Holmes, C. (2018, January 24). Re: Self-Bonding Inquiry [E-mail to the author]. Lakhani, H. A. (2003, August 11). Impact of technological change on labor productivity in U.S. coal mines-Evidence from time series and cross sectional data. Retrieved March 13, 2018, from https://www.sciencedirect.com/science/article/pii/0360544282900263 Kuby, M. (2001, September 26). The effect of restructuring on US coal mining labor productivity, 1980–1995. Retrieved March 14, 2018, from https://www.sciencedirect.com/science/article/pii/S0360544201000512 Kuykendall, T. (2016, April 13). Companies recently filing bankruptcy produce more than 2/3 of PRB coal. Retrieved March 12, 2018, from https://www.snl.com/InteractiveX/articleabstract.aspx?id=36118340&KPLT=8 McGinley, P. (2018, March 08). Will taxpayers foot the cleanup bill for bankrupt coal companies? Retrieved March 12, 2018, from http://theconversation.com/will-taxpayers- foot-the-cleanup-bill-for-bankrupt-coal-companies-56415

27 Mine Safety and Health Administration, U.S. Department of Labor. (n.d.). Retrieved March 13, 2018, from https://www.msha.gov/data-reports Preston, B. (2017, December 17). Trump EPA rule change exploits taxpayers for mine cleanup, critics say. Retrieved March 12, 2018, from https://www.theguardian.com/environment/2017/dec/17/donald-trump-epa-mining- pollution-rules Rehbach, S., & Samek, R. (2017, November). Downsizing the US coal industry: Can a slow- motion train ... Retrieved March 12, 2018, from https://www.mckinsey.com/~/media/mckinsey/dotcom/client_service/metals%20and%20 mining/pdfs/downsizing-the-us-coal-industry.ashx Response of the WVDEP to the Reorganized Debtors' Motion to Approve Settlements with Contura Energy, Inc. and the First Lien-Lenders (United States Bankruptcy Court for the Eastern District of Virginia Richmond Division November 15, 2016). Ryan, A. (2016, June 08). Coal's Costly Compensation. Retrieved March 12, 2018. Schneider, R.J. and Torries, T.F. (1991). Competitive costs of foreign and U.S. coal in North Atlantic markets. Mining Science and Technology, Retrieved March 12, 2018. Shafiee, S. and Topal, E. 2012. New approach for estimating total mining costs in surface coal mines. Mining Technology, Retrieved March 12, 2018. Stoker, T. M. (2005, August). Panel Data Analysis of U.S. Coal Productivity. Retrieved March 13, 2018, from https://www.sciencedirect.com/science/article/pii/S0304407604001411 Storrow, B. (2016, June 22). Feds, questioning reclamation plan, play trump card in Alpha Natural Resources bankruptcy. Retrieved March 13, 2018, from http://trib.com/business/energy/feds-questioning-reclamation-plan-play-trump-card-in- alpha-natural/article_09ee8ce0-8357-5346-951f-c1fbf8f4fe0a.html U.S. Department of Energy, Energy Information Administration, Independent Statistics & Analysis. (November 17, 2017 (a)). Open Data, Retrieved from https://www.eia.gov/tools/faqs/faq.php?id=69&t=2 U.S. Department of Energy, Energy Information Administration, Independent Statistics & Analysis. (November 17, 2017 (b)). Which States Produce the Most Coal?. Retrieved from https://www.eia.gov/tools/faqs/faq.php?id=69&t=2 U.S. Department of Energy, Energy Information Administration, Analysis and Projections. (November 17, 2017 (c)). Minemouth Coal Demand, Retrieved from www.eia.gov/analysis/projection-data.php#annualproj.

28 U.S. Department of Energy, Energy Information Administration, Short Term Energy Outlook. (November 17, 2017 (d)). Short Term Energy Outlook, Retrieved from https://www.eia.gov/outlooks/steo/ U.S. Department of Energy, Energy Information Administration, Projection Data. (November 17, 2017(e)). Recoverable Coal Reserves and Average Recovery Percentage at Producing Underground Coal Mines by State and Mining Method, Retrieved from https://www.eia.gov/coal/annual/pdf/table16.pdf. Williams-Derry, C. (n.d.). How Coal "Self-Bonding" puts the Public at Risk. Retrieved March 12, 2018, from http://www.sightline.org/2015/07/06/how-self-bonding-puts-the-public-at- risk/

29