Bank Loan Market-based Pricing and Bankruptcy Risk
Shuowei Song 10735909
University of Amsterdam Faculty of Business Economics
Master Thesis July 2015
Abstract:
In this paper I use CDS spread of firms as a measurement tool for bankruptcy risk, and test how the innovation of bank loan market-based-pricing changes a firms bankruptcy risk. My research is based on a difference-in-difference model, and I found that the average CDS spread is decreased by 122bps of firms that have been issued with market-based-priced loans. The bankruptcy risk of those firms is significantly decreased by market-based-priced loan. My result contributes to the theory of how to lower the bankruptcy risk and improve the stability of the financial system.
Keywords: CDS spread, bankruptcy risk, market-based-priced loan, interest rate
MSc Business Economics, Finance track Thesis supervisor: Rafael Matta
Statement of Originality:
This document is written by Shuowei Song who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements:
I wish to express my sincere thanks to my thesis supervisor, Rafael Matta, who always provided me with fresh and useful advice for my research. Mr. Matta gave me feedbacks and encouragement all the time. Thanks to Mr. Ivan and Mr. Vo for their kind feedbacks about this topic. I would also like to take this opportunity to express gratitude to all professors who have learn me knowledge about finance in the past year.
1. Introduction ... 5
2. Literature review ... 7
2.1 Theory of credit default swap ... 8
2.2 Credit default swap and firms’ bankruptcy risk ... 10
2.3 Corporate loan and market-based-pricing ... 11
3. Methodology ... 13
3.1 Hypothesis ... 13
3.2 Regression model ... 14
4. Data and descriptive statistics ... 16
4.1 Data processing ... 17
4.2 Sample summary ... 18
5. Results ... 21
5.1 The effect of treatment ... 21
5.2 Baseline regression on CDS spread ... 24
6. Robustness check and additional results ... 27
6.1 Industry random effect ... 28
6.2 Effect of market-based-priced loan under different levels of credit ratings ... 30
7. Conclusion and discussion ... 31
Reference ... 34
1. Introduction
Credit default swap (CDS) is widely used in financial market today to prevent the loss in case of default. The investigation of CDS is quite sufficient and interesting because it is about an innovation in the derivatives market. Credit default swap was devised by JP Morgan & Co. to hedge the risks of loan in 1994. In the past two decades, CDS had been growing exponentially. Though credit default swap lacked a central exchange system, the trading volume of CDS soared to $62.2 trillion in the end of 2007, and then dropped rapidly to $38 trillion in 2008. Till the end of 2014, the notional outstanding is $15.7 trillion1. Publicly reported CDS information can reflect the market situation. For instance, the CDS bid-ask spread is a sign of trading liquidity. The notional outstanding or net notional implies how invigorate the market is. CDS spread is the measure of the periodical cost of CDS contract. More importantly, CDS spread is a measure of the default probability of borrowing firms. The higher the CDS spread, the more the insurance cost, thus, the higher the default probability and bankruptcy risk of borrowing firms.
The quite sufficient literature about credit default swap in the past two decades can mainly sorted into several directions: arbitrage-free pricing model for CDS (Duffie, 1999); price of the counterparty credit risk (Arora, Gandhi and Longstaff, 2012); and the link of corporate bonds market and equity market to CDS market (Blanco Brennan and Marsh, 2005; Acharya and Johnson, 2007; Garleanu and Pederson, 2011). Besides, a very important research by Longstaff et al. (2005) uses information in CDS to obtain default and non-default components, and they confirm that the majority of the corporate spread is due to default risk. Longstaff et al. (2005) offers a perfect insight of the relevance of CDS spread and default risk.
As a tool against financial loss in case that the borrowing firms fail to repay the loan, CDS works as a kind of insurance for borrowing firms’ bankruptcy. However, some scholars argue that instead of decreasing the default risk, CDS does increase the probability of default (Campello and Matta, 2013; Subrahmanyam, Tang and Wang, 2014a). This point of view provides new insights to the chaotic situation of CDS market during the financial crisis. The expected function of credit default swap should be re-examined.
Credit default swap is an imported branch of credit derivatives, and it is traded in the over-the-counter market. Due to lacking of regulations and a central trade platform, the CDS market has deficits that related to insider trading. Hence therefore it is necessary to check what kind of information is contained in the CDS market and how insider traders make use of such information. Financial institutions, such as banks, are closely related to firms and they can easily obtain private information. This means the information of a firm’s default probability is not public. Market-based-priced loan is an innovation tying loan price to borrowing firms’ CDS spreads or CDX index. And when banks issue such loans, they will also leak private information they know such as the default probability and bankruptcy risk of the borrowing firms. Hence, the public is likely to get access to such information and change their behavior correspondingly in the equity market or the other markets.
Market-based-priced loans can provide extra information to the public. After 2008, there were more loans issued with interest rates tied to CDS spreads or CDX index. Appendix 1 shows the volume of market-based-priced loans after 2008. The volume of such loan is extremely high in 2011.
Ivanov, Stantos and Vo (2014) first brought up the research related to the market-based-priced loans. They prove that the market-market-based-priced loans are associated with lower interest rates both at origination and during the lifespan of the loans, and this kind of loan enables banks to simplify the covenant as well as reduce the cost of bank monitoring. In this paper, I investigate how firms’ bankruptcy risk change with market-based-pricing. This paper develops the theory in Santos et al. (2014), and it focuses on the role of market-based-priced loan in a macro environment instead of focusing on the cost of borrowing.
In my paper, section 2 introduces the literature I look into about theoretical framework of CDS, bankruptcy risk and market-based-priced loan. Section 3 presents the methodology and hypothesis of this paper. The model I use is a difference-in-difference model, and this model can efficiently avoid omitted variable bias. The hypothesis reveals a negative relevance of market-based-priced loan and the risk of bankruptcy. Section 4 summarizes the sample in my experiment, and section 5 shows the result diagrammatically and uses baseline regression to check the estimators empirically. The conclusion of my research is shown in section 6. My result is mainly two points. First, there are more market-based-priced loans issued to A, A-, B+ and B level firms than A+ rating firms. When banks choose firms to issue market-based-priced loans, they don’t
only consider the best rating firms. Instead, banks prefer firms with a quite high CDS spread in order to maintain the level of interest rate. The average CDS spread of those target firms is 61bps higher than the average CDS spread of firms with traditional loans. So banks may avoid the best rating firms on purpose, and they prefer firms with a higher bankruptcy risk rather than the firms with the lowest bankruptcy risk. Second, the bankruptcy risk of firms that have been issued with market-based-priced loans is significantly decreased, even in the period of financial crisis. In contrast, when the economy goes under, the bankruptcy risk of firms with traditional loans increased a lot. The average CDS spread of those firms decreased by 122bps compared to firms with traditional loans. This result reveals a strategic way to lower the bankruptcy risk.
The contribution of this paper is two-folded. First of all, my study examines the influence of market-based-priced loan. Till now, limited literature has discussed this type of loan, and my findings will contribute to the theory of market-based-priced loan. Secondly, this paper gathers the theoretical and empirical evidence on information problems in the CDS market, and develops into a deeper view of bankruptcy risk. Since CDS increase the probability of loan default, market-based-pricing can significantly lower the bankruptcy risk, thus stabilizing the market. My study provides a fresh view to look into the insight of bankruptcy risk.
2. Literature review
My study integrates three strands of literature: theory of credit default swap, discussion on the bankruptcy risk, and the loan pricing method based on CDS spread. The first part shows the progress of theory development of credit default swap. More importantly, it gathers the ideas about the information-based problem in the CDS market. This part will deeply look into the relevance of information and bankruptcy risk, which is closely related to my research question. The second part zooms into the literature of bankruptcy risk. From a general definition and determinants of bankruptcy, this part provides me with sufficient clues of how to analyze my research question. Finally, the third part introduces the mechanism of market-based-priced loan and its potential influence on bankruptcy risk.
2.1 Theory of credit default swap
As a kind of credit insurance contract against corporate default, credit default swap is widely used in the credit derivatives market. In the trading of credit default swap, the party who buys protection pays the CDS seller a fixed premium basically in basis point in each period, until either default occurs or the swap contract matures. In return, if the underlying firm defaults on its debt, the protection seller is obligated to buy back from the protection buyer the defaulted asset at its par value (Longstaff, Mithal and Neis, 2005). And the default firm can either claim bankruptcy or conduct renegotiation. The nature of credit default swap has decided the market reaction to positive and negative information of firms’ performance. That’s to say, if the trading of CDS contains positive news, the default risk of the borrowing firm should be low, and vice versa. In my research, the way that banks issue loans to a certain firm can reflect the attitude of banks as well as the quality of the firm. In the trading of CDS, the periodic rate the buyer pays to the seller is called CDS spread. Such spread reflects market predictions about the insured entities’ financial health situation and credit worthiness (Annaert, Ceuster, Roy and Vespro, 2013). CDS spread is calculated in basis points (bps), and one basis point is equal to 0.01% (0.0001). The change of CDS spread is a measure of change in bankruptcy risk in this paper.
The CDS market has been developed rapidly, however due to the lack of supervision in the over-the-counter market, the CDS market also raised a number of policy concerns such as market instability and the risk of adverse selection. Tang and Yan (2007) pointed out that those problems would prohibit investors from trading actively, and bring a negative effect to CDS liquidity.
One problematic issue is the information asymmetry in CDS market because the public has limited resource to receive useful information. Bloomfield, Robert, and O'Hara (1999) demonstrate that the level of market transparency importantly affects market equilibrium. Bloomfield, Robert, and O'Hara (1999) point out that transparency issue arises in derivatives market where the vary viability depends on the transparency of the underlying equity market. They study the effects of transparency on information efficiency, bid-ask spreads and trader welfare based on experimental economics, and find that trade disclosure has a dramatic impact on market makers’ welfare - the average trading gains is much more higher in a full transparent setting than in a semi-opaque setting. The CDS market is an over-the-counter market with limited disclosure of public
information. As John Hull (2009) stated, unlike the CDS market, the market of other over-the-counter derivatives is based on public announced information such as interest rates, exchange rates and equity index to prohibit traders from having better information than other participants. In the CDS market, the trade of contract is based on the probability of credit default, allowing some market participants exploit private information to estimate default probability more accurately. Parlour and Plantin (2005) point out that banks are comparative protected buyers in the credit derivatives market and it can generate two effects: it introduces more information asymmetry because banks are likely to have better access to the private information; it may lower information quality because banks may have a chance to monitor the firms that they use CDS contracts while the public could not.
The theories of Acharya and Johnson (2007) and John Hull (2009) point out the information asymmetry problem. Acharya and Johnson (2007) claim that since a large amount of CDS transactions are made on the desks of loan managers with financially associated firms, such insider traders can purchase credit insurance based on superior information they obtain from their prerogative relationships with clients. Acharya and Johnson (2007) show that the insider trading reflected in CDS prices is revealed before the credit level reflected by the borrower’s stock price, and this evidence is stronger if the borrower has a higher number of bank relationships. In 2010, they provide further evidence by showing that banks’ CDS insider trading grows stronger as the number of parties involved in the transactions increases.
Though financial institutions tend to command more information about the probability of default, there is still a limitation of using CDS by banks. Minton, Stulz and Wiliamson (2009) argue that this limitation is due to the lack of liquidity of CDS contracts. In the research of Minton et al. (2009), only 23 out of the 395 large banks in the sample used credit derivatives in 2005, and the hedging positions are rather small compared to the notional loan amounts. Their findings imply that banks may not effectively use CDS contracts on client firms. However, banks can make use of CDS information in another way. Since 2008, banks have increasingly extended loans with interest rate spreads tied to the borrower’s CDS spread or CDS index over the life of the loan (Ivanov, Stantos and Vo, 2014). Santos et al. (2014) find that banks issue loans at a loan spread 40bps lower if a loan is market-based-priced. They also find that the reduction in interest rate derives from the savings on bank monitoring costs. When there is a negative shock in the CDS market, market-based-pricing will create liquidity spirals
in the financial system, and in turn, increasing borrowers’ CDS spreads and leading to additional spikes in loan rates.
From the theories above, it is clear that banks get private information on borrowing firms’ default risk, and the market of CDS is lacking of information efficiency. When banks change the traditional loan into market-based-priced loan, it means less incentive to monitor the borrowing companies. And when such change is detected, the public receives a kind of positive information on the borrowing firms and then reacts actively in the other markets. So the bankruptcy risk of the borrowing companies should be decreased.
2.2 Credit default swap and firms’ bankruptcy risk
Credit default swap can protect the CDS buyer to some degree. A financially stressed borrowing firm can declare bankruptcy or to conduct renegotiation. Under the situation of bankruptcy, lenders that have bought CDS can still receive repayment from CDS sellers. Lenders that haven’t bought such insurance have to bear the losses. CDS spread, as a fee that protection buyers need to pay to protection sellers, is a numeric measure of default risk and bankruptcy risk. Longstaff et al. (2005) use CDS spread directly as a measurement of default component in corporate spreads.
An important study by Edward, Haldeman and Narayanan (1977) investigate the development of a ZETA bankruptcy model by incorporating the sample with comprehensive inputs from 1969 to 1975. They select a 7-variable model which proved to be the most reliable to identify bankruptcy risk of corporations. Those variables include the financial scale quality indicators such as information about size, assets and debts of firms. This paper provides a standard to choose the determinants of bankruptcy risk.
In recent years, concerns have been raised about whether CDS trading itself can affect the credit risk of the relevant entities. For credit default swap and bankruptcy risk, the number one concern is about whether CDS increases or decreases the bankruptcy risk. Recently, Campello and Matta (2013) bring out their estimation on this question. Campello and Matta (2013) develop the first model to check the optimal demand for credit default swap, and they show that CDS could lead to risk-shifting, thus increasing the probability of default. Look into the period of financial crisis, the booming CDS notional is persuasive to explain the high rate of firms’ bankruptcy and the chaotic
financial situation. Previous study by Hu and Black (2008), and Bolton and Oehmke (2011) share the same opinion on empty credit problem. They think that between bankruptcy and renegotiation, banks that with CDS insurance stand to lose less in the case of default. So, with the bargaining power advantage and less forgiving attitude in renegotiation, banks will force more financial distressed firms to declare bankruptcy.
The empirical result on this question agrees with the theories above. Subrahmanyam, Tang and Wang (2014a) test on the sample of CDS trading of 901 North American corporate issuers between June 1997 and April 2009. And they find that both the probability of credit rating downgrade and probability of bankruptcy increase after CDS trading. Their finding is also robust after controlling for the endogeneity of CDS trading. More specifically, firms with larger amounts of CDS contracts outstanding and more “no restructuring” contracts than other types of contracts covering restructuring are more adversely affected.
In addition, in the view of market efficiency (Fama Eugene, 1970), the CDS market meets the standards of semi-strong-form market efficiency for two reasons. First, the CDS market provides a part of information on CDS trading, and those information could be detect by the public. Second, according to the literature about information problems in the CDS market, insider trading can certainly obtain the most profitable return. The liquidity of CDS market is frequently discussed under an assumption of informed trading. From an information-related theoretical point of view, the asymmetric information increasingly deepen the degree of semi-strong-form market efficiency, and the nature of such a market will adversely influence market participants’ behavior, and indirectly effect on the financial development.
2.3 Corporate loan and market-based-pricing
Issuing a corporate loan requires a decision to be made simultaneously: choosing a spread over base rate, time to maturity and whether to secure the loan. Gottesman and Roberts (2004) try to solve the puzzle between loan spread and maturity under the tradeoff hypothesis (which implies a positive link between loan spread and maturity) and the credit quality hypothesis (which implies a negative link between loan spread and maturity). They find evidence supporting both the hypotheses. This research shows that the loan market is quite flexible and difficult to predict.
Loan pricing is the key to determine the performance of a bank. Banks need to carefully settle the loan price. If the margin is too low, the risk of assets will be under priced; if the margin is too high, the bank will loss competition and be forced out of the market since the borrowers are always looking for cheap loans.
The literature on the market-based-priced loan is very limited. Michael (2013) highlights that using CDS spreads in loan pricing is widely accepted by portfolio managers. CDS spreads may influence the cost of refinancing and the access to funding. Via this pricing mechanism, the loan price should cover both the expected loss and unexpected loss especially when the information of CDS spread is generated in a less regulated OTC market. To get a more detailed view, Michael (2013) breaks the loan margin into four components: CDS spread, cost of equity, cost of funding and settlement margin.
Market-based-priced loan is an innovation on loan pricing and it was first applied in 2008. According to Ivan, Santos and Vo (2014), market-based-priced loan ties its interest rate spread to the borrower’s CDS spread or CDX index. This kind of loan becomes popular because it can simplify the lending covenant structure and it associates with 40bps lower interest rate comparing to traditional loans. The information about market-based-priced loan is very limited in prior literature. The available way to figure it out is to check the information in company fillings2, or to check the IFR comments under the content of facility documents from Thomson One database. In Santos et al. (2014), they first describe and analyze the market-based-priced loans. They then test the sample including 145 separate facilities from 51 firms. From the collected data, few characteristics of market-based-priced loan can be summarized: the loans are usually investment grade loans3; it tends to belong to revolving loan type; it is issued with a floor and cap in basis points to control the minimum and maximum value of the CDS spread or CDX index added to the base rate.
Santos et al. (2014) gives me some directions to study the influences of information inflow on the CDS market, since it can solve the following problem: If I want to test the private information effect on the CDS market, how I can distinguish insider trading? Though there are many researches that can indicate the characteristics of insider trading activities, for example, Acharya and Johnson (2007) claim that insider
2 See U.S. Securities and Exchange Commission website.
trading is more likely to happen when a firm has more connections with banks. Christophe, Ferri, and Hsieh (2010) suggest to suspect those “just in time” reaction before the announcement downgrades. However, it is too complex to list and study those behaviors to get a comparatively precise result, because the standard of insider trading is not completely defined and the data is too chaotic. However, with the market-based-priced loan, the public can detect such private information and make use of it. Thus the effect of private information is enlarged. And it is possible to conduct a natural experiment to test how the shock of information influences the CDS spread. This method avoids the complex criteria to distinguish private information, and it allows quite a comprehensive observation.
3. Methodology
This section describes the hypothesis and the model I use. The first part describes the hypothesis I want to test based on the theories and empirical evidence from literature. The second part contains the difference-in-difference regression model and detailed descriptions for the variables that are included in the regression model.
3.1 Hypothesis
Regarding to the theories and empirical evidence shown in the literature, my hypothesis is mainly based on the previous literature of Bloomfield, Robert, and O'Hara (1999), Campello and Matta (2013), and Ivanov, Stantos and Vo (2014). The theories in their paper present a clear view and a logical analysis of my research question - First of all, the defect of contact with unequal information gives the chance of informed trading in the CDS market, and CDS lead to the raise of the bankruptcy risk. Secondly, issuing market-based-priced loans is a kind of information import to the public, and the information is positive since banks know company better than the public.
I use the market-based-priced loan as a natural experiment to test how the firms’ bankruptcy risk reacts to the enriched information in the CDS market. Different from the previous literature, I base my method upon natural experiment to catch the private information flow rather than depending on the external factors such as Acharya and Johnson (2007), in which they use the number of bank relationships to distinguish insider
trading. The natural experiment can avoid the vague theory of how to distinguish insider traders. My research is also different from Santos et al. (2014). Santos et al. (2014) investigates how loan spreads react to market-based-pricing behavior, and they mainly test the change in loan cost of banks. This paper concentrates not on banks, but on the bankruptcy risk of borrowing firms. To this point, this paper can contribute to the larger topics of information quality and bankruptcy probability. Since the bankruptcy risk changes according to how the market participants react to the information change. Under the positive information referring to the company, the bankruptcy risk should be lowered. The traditional way of banks to price loans is to settle a fixed loan spread or tie the loan spreads to base rates such as LIBOR and prime rate. The traditional way of loan pricing does not offer any new information about the default possibility of a firm. So, comparing to those firms, the firms with market-based-priced loans should have a lower bankruptcy risk. Moreover, I test the factors that can strengthen the effect on bankruptcy risk. Since credit rating is a principle estimator in deciding bankruptcy risk (Altman et al., 1977), I compare the bankruptcy risks under different S&P credit ratings, and it will offer a more accurate and comprehensive understanding to this issue.
Hypothesis 1: The bankruptcy risk should decrease for the firms with market-based-priced loans comparing to that of firms with loans issued in a traditional way.
3.2 Regression model
The literature above involves some regressions related to CDS spread. Acharya and Johnson (2007) take credit rating as independent variables together with company qualities to test the information effect on CDS price. But their result of information on credit rating is not significant. Their problem is that the CDS market is very complex and there is a large chance of omitted variable bias. Furthermore, the control of randomization is required. In this paper, I capture the information flow by observing the companies with market-based-priced loans, and conduct a difference-in-difference experiment. This can help to solve the following concerns.
First, it is difficult to distinguish informed trading of CDS. With the result in Ivanov, Stantos and Vo (2014), pricing a loan according to CDS spread generates a kind of positive information to the public. Instead of searching and screening informed trading, this is a much more simple and reliable way to capture the enriched information.
Second, the sample tends to be out control of randomization. CDS trading is not highly published, and the over-the-counter market is lacking of a central trading platform, so the data can hardly be random. Using a difference-in-difference model avoids this problem because it allows the treatment “as if ” randomly assigned. In the difference-in-difference regression, I defined companies that have been issued with market-based-priced loans as treatment group, and the companies with traditional loans as control group. The “before and after time threshold” is the second quarter of 2008, when the first market-based-priced loan was applied. The control variables 𝑍 are firm and loan specifics. By comparing the change in the outcomes of pre- and post-treatment, the effect of market-based-priced loan can be easily detected. Though the exact dates of bank issuing market-based-priced loan to treatment group firms are not the same, I choose this time threshold because it can represent the information gap and the before-after effect of such loan.
My regression is:
𝐶𝐷𝑆𝑆𝑝𝑟𝑒𝑎𝑑!,! = 𝛽!+ 𝛽!∙ 𝑀𝐵𝑃! + 𝛽!∙ 𝑃𝑒𝑟𝑖𝑜𝑑!+ 𝛽!∙ 𝑀𝐵𝑃!∙ 𝑃𝑒𝑟𝑖𝑜𝑑!
+𝛽!∙ 𝑍!,!+𝑢!
In this model, 𝐶𝐷𝑆𝑆𝑝𝑟𝑒𝑎𝑑!,! is the CDS spread of firm i at date t, 𝑀𝐵𝑃! is a binary variable which takes a value of 1 if a firm has loan spread linked to its CDS spread, and 0 otherwise. 𝑃𝑒𝑟𝑖𝑜𝑑! is a binary variable that takes a value of 1 if the time is later than the second quarter of 2008, and 0 if the time is earlier than the second quarter of 2008. The interaction terms of two binary variables also included: the joint effect of treatment and time 𝑀𝐵𝑃! ∙ 𝑃𝑒𝑟𝑖𝑜𝑑!. While attempting to test the difference, I need to control a set of firm and loan specific factors 𝑍!,! of firm i at date t.
The first structural model about credit risk was introduced by Merton (1974). Merton analyses the factors that influence the credit risk. Recently, Cesare and Guazzarotti (2010) provide an empirical result of the determinants of CDS spread changes before and during the subprime financial turmoil. The explanatory power of their fundamental variables is more than 53% on average, which is higher than the explanatory power of any previous studies on CDS spread. Ivanov, Stantos and Vo (2014) include the firm control variables, loan control variables and a set of macroeconomic factors. Here I employ some of the variables in their models that are commonly used and
comparatively significant in their regressions.
The firm control variables are: risk-free interest rate (𝑅𝐹𝐼𝑅): It is a parameter regarding to the market situation. A higher risk-free interest rate means the higher value of future assets, thus increasing the credit spread; Nominal debt outstanding (𝐿𝑜𝑔𝐷𝑒𝑏𝑡): This should be positively related to spread, since the more debt a firm has, the larger chance it cannot meet its obligation; Asset (𝐿𝑜𝑔𝐴𝑠𝑠𝑒𝑡): Asset represents the firm value and its capability to repay the loan. A negative sign should be expected; Stock return (𝑆𝑡𝑜𝑐𝑘𝑅𝑒𝑡𝑢𝑟𝑛): A negative sign is expected, the higher the stock return, the higher the future profitability. Thus, the default probability is small; Stock return on S&P index (𝑅𝑒𝑡𝑢𝑟𝑛𝑆&𝑃): It is defined as net growth rate of S&P index from previous day to today, and it reflects the grouwth of the whole stock market. When the market is in a good situation, the CDS spread should be narrowed; Leverage (𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒): Calculated by total liabilities divided by total assets, leverage ratio is positively connected to the spread; Sales (𝐿𝑜𝑔𝑆𝑎𝑙𝑒𝑠): Sales is a signal of firm size. Large sales amount means the better diversification in supply, customers and industries, so the default chance should be lowered; Tangibles (𝐿𝑜𝑔𝑇𝑎𝑛𝑔𝑖𝑏𝑙𝑒𝑠): The tangible assets are inventories plus plant, property, and equipment. When the firm default, tangible assets loss less value than intangible assets, so it’s a standard to check the assets stability of a firm, and a negative sign is expected; S&P quality rating (𝑆&𝑃𝑄𝑢𝑎𝑙𝑖𝑡𝑦𝑅𝑎𝑡𝑖𝑛𝑔): Measured by the Standard & Poor's current credit rating, it is a opinion of the overall creditworthiness. The range is from A+ to D. The credit rating has an inverse relationship with the possibility of debt default.
The loan control variables are: facility size (𝐿𝑜𝑔𝐹𝑎𝑐𝑖𝑙𝑖𝑡𝑦𝑆𝑖𝑧𝑒). This can control the size of loan. And a large amount of loan implies a larger chance to default; Maturity (𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦): The effect is two sides: a longer maturity generates higher credit risk, or the firms who are obligated to the long-term loan tend to be well rated.
4. Data and descriptive statistics
This part present the data used in this paper and the source of the data. The first part shows how the data are collected and processed. The second part describes the final sample I get and summarizes some features of the sample.
4.1 Data processing
The data I use in this research including data of CDS spread, information of the firms and loans. The most important information is which loans are market-based-priced. The data for this research comes from Thomson Reuters DataStream, Thomson Reuters DealScan, Thomson One, Thomson Reuters Loan Connector, Compustat and Center for Research in Security Prices (CRSP) from Wharton Research Data Services.
I limit the target time period from the first quarter of 2004 to the third quarter of 2012. The first market-based-priced loan appeared in the second quarter of 2008, which marks the middle point of the panel data time range I chose. I choose to use quarterly data because daily data can hardly perform the significant changes in CDS spreads, and the available firm information is reported quarterly. For CDS spread, I use a quarterly mean. For loan size and maturity, I use a quarterly sum number.
Thomson Reuters DataStream provides the daily CDS spreads. From the first quarter of 2004 to the third quarter of 2012, there are 11,690 quarterly numbers of CDS spreads from 334 firms.
Thomson Reuters DealScan contains information on loan pricing and contract details, terms, and conditions. Under DealScan, I use the datasets “Package”, “Facility” and “Current facility pricing” to collect the basic data about the loan information. “Package” lists the package of one or few facilities, and it shows the loan identification, package amount and deal active dates. More importantly, “Package” includes comments about loans, which gives me the information whether the loan is based on CDS spread or CDX index. The only problem is that the comment is not necessary for every package, and two thirds of the comments are missing in this dataset. “Facility” and “Current facility pricing” provides more detailed information of each loan transaction. The information includes the loan type (revolving loan or term loan), facility amount, information about whether the facility is secured or not, and the information about base rates (LIBOR, prime rate or fix rate).
Thomson One provides corporate filings containing the information about detailed loan pricing. By using this database, I collect a part of the market-based-priced loans in my sample. An example of such filings is showed in Appendix 2.
Thomson Reuters Loan Connector provides more information of the market-based-priced loans and the pricing grid. Loan Connector records the North American market-based-pricing grid since September 2009, and it tracks the latest deals that
contain market-based-pricing and the applicable pricing terms till March of 2015. From the second quarter of 2008 to the third quarter of 2012, I get 236 loans with interest rate tied to CDS spread or CDX index.
CRSP and Compustat summarize the periodical information about the stock market and company specifics. Those data are reliable and comprehensive since it comes from financial reports. I gather the quarterly data from Compustat to get the features on firm specifics including asset, debt, sales, leverage ratio, tangible assets, risk-free rate and EPS. CRSP provides me the daily market information such as stock price, stock return and stock performance related to market S&P index. The number of observations I got from the first quarter of 2004 to the third quarter of 2012 is 53,889, and the data comes from 1,288 firms.
By identifying symbol “gvkey” and using DealScan – Compustat link table4, I got the merged file of CDS spread, firm performance and market-based-priced loan with 2,800 observations. Then I merge in the loan information of each firm in different periods. The missing values are filled with the most recent available data. The final file includes 490 observations from 14 firms in the treatment group, and 2,310 observations from 66 firms in the control group. Then I winsorize the CDS spread and replace the outliers on 99th percentile.
4.2 Sample summary
The summary of variables is presented in Table 1. Table I presents an overview of the characteristics of the sample. It provides an overview of the complete sample I get. In the table, I summarize the mean and difference of dependent variable and independent variables by groups. There is a media in the brackets under each mean of first three columns. The differences between the means of treatment group and control group are tested with a t-test, and the t-statistics are presented in the brackets in the last column. The significance level can provide an insight of the differences.
Table 1 Summary statistics of the sample. The sample contains all the data derived from the selection described in section 4.1. This table shows the characteristics of the complete sample, treatment group and control group. Treatment group contains the firms had market-based-priced loans for at least one time. Control group contains firms that never had market-based-priced loan. All
4 DealScan-Compustat link table is provided by Chava, Sudheer, and Michael R. Roberts, 2008, How Does Financing
US dollar amounts is considered as in millions, except for specific explanation.
Complete Sample Treatment group Control group Difference
CDS Spread (bps) 123.0688 83.9634 131.3638 -47.4004*** [70.16859] [52.396] [80.0271] [8.3023] Firm Specifics Asset (US$) 35292.13 56060.24 30886.77 25173.47*** [12412.55] [33333] [10786.9] [-5.6853] Debt (US$) 8319.651 8969.124 8181.884 787.24 [2940.791] [4684.356] [2493.292] [-0.4804] Revenue (US$) 5753.872 8853.71 5096.331 3757.379*** [2200.263] [6713.684] [1864.903] [-6.6571] Tangible (US$) 28322.33 38888.16 26081.09 12807.07*** [10720.5] [26332] [9040.399] [-3.308] Leverage 0.2515 0.17689 0.2674 -0.0905*** [0.2274] [0.1716] [0.2412] [12.4064] EBIT (US$) 3523.531 7160.641 2752.023 4408.618*** [1162.738] [3968] [901.335] [-11.7838] EPS 0.6849 0.906 0.638 0.2681*** [0.63] [0.82] [0.58] [-4.4635] Risk-free Rate 3.0228 2.8491 3.0593 -0.2102*** [2.87] [2.71] [2.87] [3.134] Stock Price (dollar) 46.301 55.472 44.3556 11.1163***
[42.3617] [54.125] [40.0617] [-7.3955]
Stock Return 0.0059 0.0059 0.0058 0
[0.0081] [0.0055] [0.0081] [-0.0319] Stock Return on S&P 0.0035 0.0035 0.0034 0
[0.0055] [0.0055] [0.0055] [-0.0349] S&P Quality Rating
(numeric) 4.775 6.2857 4.4545 1.8311***
[4] [6] [4] [-24.5895]
Loan Specifics
Facility Size (US $) 1730 2660 1530 1130*** [1000] [2000] [850] [-7.7209] Facility Maturity (month) 59.6 42.6735 63.1905 -20.517*** [60] [48] [60] [8.1099] Sample Number of Observation 2800 490 2310 Number of Firms [80] [14] [66]
S&P quality rating is numerically transferred from rating levels: 8 for A+, 7 for A, 6 for A-, 5 for B+, 4 for B, 3 for B-, 2 for C, 1 for D. Difference of the means of treatment group and control group are statistically significant at the ***1% significance level.
In the samples I get, there are 490 samples from 14 firms had been issued with market-based-priced loans, and 2,310 samples from 66 firms that have never been issued with market-based-priced loan. The features of treatment group are strongly matching the
prediction of previous theories. From the table, the mean CDS spread of treatment group firms is significantly 47bps lower than the control group, that is, the overall mean CDS spread of the firms with market-based-priced loans tends to be smaller through time. So, the bankruptcy risk is lower of firms in treatment group than in control group. This phenomenon is explainable since the theory of Santos et al. (2014) implies that banks issue market-based-priced loans to firms with a good credit record. The mean CDS spread is 56% higher of firms in control group than in treatment group.
Firms with market-based-priced loans are proved to have larger amounts of assets, revenue, tangible assets, and EBIT. Those qualities mean a good finance situation and potential growth power of a firm. Moreover, those firms have higher EPS and stock price, implying a good reputation and stable future development in the capital market. A lower leverage ratio of treatment group is corresponding to the banks’ consideration of the firms’ repayment ability. S&P quality rating is an important standard to set the interest rate for the lender banks. So the firms with market-based-priced loans are recognized with better rating conditions. A larger facility size helps to prove the trust of banks to the treatment group firms and their future growth potential. A shorter repayment period shows the firms with market-based-priced loans are trustable, and they can repay the loan in a quite short period.
The sample shows some similarities to relevant studies. Previously, the Merton (1974) model shows that the leverage ratio and asset volatility can effectively influence the default possibilities. Tang and Yan (2007) prove the effect of leverage and highlight the determinant role of credit rating. Iuliana et al. (2010) use event study to investigate how CDS spread reacts to the credit rating change, and they find the a negative rating change has more influence than a positive rating change event to the CDS spread. Iuliana et al. (2010) also prove that debt can significantly contribute to the change of CDS spread.
Santos et al. (2014) use the sample including 145 market-based-priced loans and 7,715 traditional loans, and test how the different types of issuing loans can affect the loan spread. Their sample shows the significance of the difference from two groups on loan amount, maturity, leverage ratio and tangible assets. So my sample in this paper is qualified and it can support my empirical experiment.
5. Results
This part shows the empirical results of my research. This section is divided into two main parts. For a difference-in-difference experiment, the comparison of treatment group and control group should be presented. So the first part focus on the comparison of CDS spreads from treatment group and control group with a threshold of the treatment time. The second part focuses on the baseline regression to test the hypothesis. In this part, I show the effect of market-based-pricing on borrower firms’ bankruptcy risk.
5.1 The effect of treatment
First I want to show the overall effect of market-based-pricing before running the regression. I calculate the mean of quarterly CDS spreads before winsorizing in terms of treatment group and control group covering the entire time rage from 2004 to 2012. This mean presents the average level of CDS spread for each quarter. I get Graph 1. From Graph 1, it is clear that the mean CDS spread of firms that with loan interest rates tied to CDS spreads or CDX index has a decreasing trend, while the mean CDS spread of firms with traditional loans has an increasing trend. The sharp change appeared around the treatment time, that is, the second quarter of 2008. So, the bankruptcy risk of firms with market-based-priced loans has been decreased considerably. While the bankruptcy risk of firms with traditional loans has been worsened. In the period of financial crisis, the mean CDS spread of firms in control group is much more higher than that of firms in treatment group. So it is reasonable to conclude that market-based-priced loan can improve the condition of a firm and lower its bankruptcy risk even when the whole economy goes down.
The information visualized in Graph 1 could be explained by the quality of treatment group firms. According to Santos et al. (2014), firms in treatment group require less bank attention of monitoring. And the default risk of those firms is lower. The rapid increase of CDS spread of firms in control group from 2007 to 2009 reflects the bad credit situation and a high bankruptcy risk during the financial crisis period. This finding is constant with the calculation in Shan et al. (2014), which shows an enlarged CDS spread during financial crisis period. Though the CDS spread of firms in treatment group also increases slightly during this period, there is still a significant gap between the two groups around 330bps maximum.
Graph 1 Quarterly CDS spreads of treatment and control group. This graph shows the means of quarterly CDS spread from the first quarter of 2004 to the third quart of 2012.The treatment group contains the firms that have been issued with market-based-priced loans, and the control group contains firms that issued only with traditional loans. The treatment time is the second quarter of 2008 which represents the time threshold. This graph is based on the sample that contains all the firms resulting from the selection procedure of section 4.1 before winsorizing.
Then I further look into the overall CDS spread change around the treatment time threshold. Since the first market-based-priced loan appears in the second quarter of 2008, it should be a positive sign that firms with market-based-priced loan are more trustable. Accordingly, the CDS spread of those firms should be comparatively low.
In Graph 2, I calculate the overall levels of CDS spreads before treatment and after treatment for the two groups. Before the treatment, the average level of CDS spreads of treatment firms is around 317bps, and it drops to 74bps after treatment. On the contrary, the control group has a mean CDS spread level around 102bps before treatment, and rises to 252bps after treatment. The gap between CDS spreads of firms in treatment group and control group changes from positive 243bps to negative 140bps. When the type of loan pricing brings good signals to the public, the default risk of firms is automatically reduced. This result agrees with Ismailescu et al. (2010). Ismailescu et al. (2010) get the result that that positive events has a powerful influence in improving the firms’ credit situation and lower the possibility of default.
Graph 2 Overall average CDS spreads by treatment group and control group around treatment. This graph shows change of CDS spread within four years among the time threshold, the second quarter of 2008. The treatment group contains the firms that have been issued with market-based-priced loans, and the control group contains firms that issued only with traditional loans. This graph is based on the sample that contains all the firms resulting from the selection procedure of section 4.1 before winsorizing.
Then I test the empirical result of the difference by using the model in Card and Krueger (1993). Table 2 shows the empirical result of the treatment effect before winsorizing. The results in the table are tested by the classic model shown in Card and Krueger (1993). In their regression, they check the minimum wage effect on employment of two states, and show a similar structured table as following.
Table 2 Average CDS spreads before and after treatment. The sample contains all the data derived from the selection described in section 4.1 before winsorizing. This table shows the effect of the dummy variable MBP on CDS spread of treatment group, control group, and their difference. Treatment group contains the firms had market-based-priced loans for at least one time. Control group contains firms that never had market-based-priced loan. The treatment time is the second quarter of 2008. The t-statistics are presented in the brackets under each coefficient.
Control group (i) Treatment group (ii) Difference (Treatment-Control) (iii) CDS spread, Before 95.3463*** 350.0132*** 254.6669*** [31.33] [5.71] [-4.15] CDS spread, After 218.4154*** 64.2541*** -154.161*** [35.67] [52.78] [-24.7] Change in CDS spread 123.0691*** -285.7591*** -408.8282*** [25.02] [-6.66] [-9.47] Coefficients are statistically significant at the ***1% significance level.
The regressions use heteroskedasticity-robust standard errors. Column (i) shows the effect of dummy variable MBP on CDS spread of control group, and the average CDS spread increase significantly by 123bps after market-based-pricing has been applied. This result shows that for companies with traditional loan, the overall level of CDS spread is widened with the economy situation going down. Explained by the theory in
50 100 150 200 250 300 2004q3 2006q3 2008q3 2010q3 2012q3 date
Treatment group Control group Overall CDS spread O ve ra ll C D S sp re a d _ me a n
Santos et al. (2014), such companies are less trustable, so that default probability and bankruptcy risk are high. Column (ii) shows the effect of the dummy variable MBP on CDS spread of treatment group, and the CDS spread decrease significantly by 286bps after market-based-pricing has been applied. On the contrary to the control group, firms with market-based-priced loans reach the requirement of banks in the aspects of credit record and repayment ability. Those firms could bear the side effect of financial crisis and maintain their CDS spread at a quite low level. Thus, the bankruptcy risk of those firms is still not high after financial crisis happened. Column (iii) presents the difference of treatment group and control group over time. The gap of the two groups changes from a positive number to a negative number by 409bps, and it reveals the strong power of MBP to CDS spread. That is, the bankruptcy risk get reduced significantly.
5.2 Baseline regression on CDS spread
Graph 1 and Graph 2 in part 5.1 give a direct impression on the difference of CDS spreads of treatment group and control group. The results come from calculating the means of CDS spreads. Now I test the coefficient of market-based-priced loan to CDS spread, and get a more detailed result under the condition of including the control variables. The regression model is stated in section 3.2 and the data processing follows the instruction in part 4.1 after winsorizing.
Table 3 shows the regression result of my hypothesis. CDS spread is a measure of default probability, as well as bankruptcy risk since it is the percentage that the CDS buyers need to pay for the credit insurance. Column (i) contains the pure effect of market-based-priced loans. Column (ii) is the original determinant power from control variables of the firm and loan specifics. Column (iii) is the complete regression with dummy variables and control variables.
Table 3 Baseline regressions on CDS spread. The sample contains all the data derived from the selection described in section 4.1 after winsorizing. This table shows the baseline regression of only dummy variables, and the regression with control variables. MBP takes a value of 1 if a firm belongs to treatment group and a value of 0 if a firm belongs to control group. Treatment group contains the firms that have been issued with market-based-priced loans. Control group contains firms that never had market-based-priced loan. Period marks the time before and after treatment. It takes a value of 1 if the time is after treatment, and a value of 0 if the time is before treatment. The time threshold is the second quarter of 2008. The t-statistics are presented in the brackets under each coefficient.
(i) (ii) (iii) MBP 16.2609** 61.3114*** [1.98] [6.92] Period 82.9456*** 81.3824*** [17.86] [16.17] MBP*Period -123.7859*** -122.3073*** [-13.34] [-13.03] LogAsset 14.2847 -5.6658 [1.28] [-0.47] LogDebt 5.5365 5.5928 [1.49] [1.45] LogRevenue -2.1761 -0.4636 [-0.76] [-0.18] LogTangible -4.6766 4.3 [-0.58] [0.5] Leverage 123.5135*** 116.8784*** [4.52] [4.88] EBIT -0.0018*** -0.0008** [-5.28] [-2.54] EPS -4.554* -3.8167** [-1.93] [-1.97] Risk-free Rate -4.8991*** 3.2602* [-2.74] [1.86] Stock Price -0.422*** -0.3897*** [-5.41] [-5.61] Stock Return -104.8309 1134.732* [-0.15] [1.73] Stock Return on S&P -112.6604 -1366.452**
[-0.16] [-1.98] S&P Quality Rating -11.8885*** -11.7162***
[-8.74] [-8.26] Facility Size -12.9124*** -7.0601*** [-4.69] [-2.79] Facility Maturity 0.0895* -0.017 [1.91] [-0.39] Constant 88.70611*** 336.3172*** 240.6488*** [29.41] [6.49] [4.93] Observations 2800 2800 2800 R-square 0.1345 0.1604 0.2519
Coefficients are statistically significant at the ***1%, **5%, and *10% significance level.
From regression (i) and (iii), I can see that the positive effect of the dummy variable MBP on CDS spread, which implies the CDS spread of firms in treatment group is originally higher than those in the control group. And the effect is stronger when I
include control variables. This result is interesting because it is opposite to my expectations. If a firm with market-based-priced loan is more trustable than the firms with traditional loans, the default probability, which reflected by CDS spread, should be lower. One possible explanation is that banks prefer to issue market-based-priced loans to high-risk firms on purpose. If a loan is market-based-priced, then its interest rate will be tied to the firm’ CDS spread or CDS index. A higher default risk leads to a higher interest rate, and it is possible to raise the interest rate in such an indirect way for banks’ benefit. With control variables, the CDS spread of firms in treatment group is originally 61bps higher than the CDS spread of firms in the control group. That is, when banks choose a firm to issue market-based-priced loan, the base rate of this firm is 61bps higher. If I relate this to Santos et al. (2014) in which they find that market-based-priced loan can reduce the loan spread by 40bps, the net effect of market-based-priced loan should be offset for a large part.
After the second quarter of 2008, the overall level of CDS spread in the market has been raised by more than 80bps under the condition with or without control variables. This result means the default probability is worsened, and the bankruptcy risk is adversely affected by this time period. This phenomenon is relevant to the financial crisis, and my result is constant with Shan et al. (2014), which thinks CDS market helped to trigger the chaos in the whole financial market; and Annaert et al. (2013), which states that the steeply rising CDS spreads are due to the increased credit risk for the recent crisis.
The joint effect of MBP and period generates a negative effect on CDS spread. It tests my hypothesis that the bankruptcy risk of firms gets reduced after applying market-based-priced loans. The results of regression with or without control variables are quite the same. Overall, the CDS spread has been decreased for more than 122bps. This result is reasonable as the public receives the positive signals, and automatically reacts actively to the relevant markets, leading a better development of the firms and lowering their bankruptcy risk. This result can further prove the conclusion in Santos et al. (2014) that the decreased interest base rate contributes to the decreased average level of loan spread, and it is profitable for the underlying firms. Another interesting thing to mention is that after a firm has applied market-based-priced-loan for one time, it is very likely that its next loan is traditional priced rather than market-based-priced again. This phenomenon could be detected in the sample, and it explains why my sample in the treatment group are not applying market-based-priced loan all the time – in case of the interest rate
continuously decreasing, banks need to consider their long run profitability. If a bank issues market-based-priced loans to a certain firm all the time, though this bank can save expenses of monitoring, the interest level the bank gets will be lower and lower. By stop issuing market-based-priced loan, the bank can maintain the interest rates at a certain level. This finding reveals the banks’ incentive that they need to prevent the interest rate fall to a quite low level even after they issued market-based-priced-loans.
About the control variables, though all the variables I choose were proved to be significant in deciding CDS spread in the previous literature, in my regression some of them lose their significance. Asset, debt, revenue and tangible asset are not efficiently influence CDS spread anymore. On the contrary, the financial health of a firm such as leverage ratio and the stock market potential such as stock return together with dummy variables are still significantly matter to the CDS spread of a firm. For loan specifics, the size of facility affects the CDS spread rather than the maturity of a facility. In regression (iii), leverage plays an important role in deciding CDS spread. Stock return on S&P, which reflects the market situation, is quite influential determinant for bankruptcy risk.
6. Robustness check and additional results
This part includes extra regressions with industry random effect to conduct a robustness check on my results. Since the methodology used in this research is a difference-in-difference model, it is unnecessary to do regressions with time or entity fixed effect, because the time and entity effects are already automatically controlled in this model. If I still involve time and entity fixed effect, it will lead to an omitted variable problem. When I choose between industry fixed effect and industry random effect, I first use Hausman test to check the p-value of each effect, and it comes out that I should use random effect instead of fixed effect. This confirms the consistency of my estimates, since random effect should be used only when the covariance of the dummy variables and control variables equals to zero.
As stated above, I conduct regressions to include the effect from industry variety. The industries are divided by North American Industrial Classification System (NAICS), and each industry is represented by an industry code. In my sample, the firms are categorized into 59 industries.
regression on CDS spread using the baseline model but separately check the effect of market-based-priced loan under different levels of credit ratings. The credit rating distributes from A+ to D, that is, 8 levels. Since the firms with level D, C and B- do not have market-based-priced loans, here exists a collinearity problem. So the estimates of those firms are not available.
6.1 Industry random effect
Table 4 shows the results of baseline regressions with industry random effect. Comparing to the baseline regressions in Table 3, I find a slight change of the coefficients and their significance level.
Table 4 Regressions on CDS spread with industry random effect. The sample contains all the data derived from the selection described in section 4.1 after winsorizing. This table shows the regression with industry random effect. The effect is tested by Hausman test. MBP marks treatment group and control group. MBP takes a value of 1 if a firm belongs to treatment group and a value of 0 if a firm belongs to control group. Treatment group contains the firms that have been issued with market-based-priced loans. Control group contains firms that never had market-based-priced loan. Period marks the time before and after treatment. It takes a value of 1 if the time is after treatment, and a value of 0 if the time is before treatment. The time threshold is the second quarter of 2008. The z-statistics are presented in the brackets under each coefficient.
(i) (ii) (iii)
MBP 16.2609*** 14.6218*** [4.74] [3.97] Period 82.9456*** 80.2982*** [85.51] [76.47] MBP*Period -123.7859*** -122.2175*** [-53.38] [-53.97] LogAsset 42.5113*** -0.5144 [4.73] [-0.11] LogDebt 1.7520 1.6747 [0.58] [1.04] LogRevenue -7.9906*** -1.5726 [-4.22] [-1.54] LogTangible 5.2065 1.1248 [0.59] [0.25] Levaerage 45.1155*** 1.2045 [2.59] [0.13] EBIT -0.0032*** -0.0004** [-8.07] [-2.02] EPS -1.2966* -1.3913*** [-1.8] [-3.51]
Risk-free Rate -13.1774*** -3.0523*** [-19.98] [-8.32] Stock Price -0.2849*** -0.1557*** [-6.35] [-6.72] Stock Return 585.9071** 1443.156*** [2.3] [10.27]
Stock Return on S&P -886.1805*** -1720.612*** [-3.3] [-11.62]
S&P Quality Rating -2.0021 1.6805*
[-0.57] [1.78] Facility Size -9.0294*** 0.2599 [-7.42] [0.4] Facility Maturity 0.1517*** -0.0011 [7.35] [-0.09] Constant 88.70611*** -117.0721** 88.139*** [22.34] [-2.54] [4.74] Observations 2800 2800 2800 R-square (overall) 0.7737 0.305 0.8106
Coefficients are statistically significant at the ***1%, **5%, and *10% significance level.
Column (i) is the result of regression without control variables. These estimators and their significance level stay almost the same as the results in Table 3, which means the result of regression (i) in Table 3 is robust. In addition, the values of t-statistics are increased and the significance of MBP is improved after involving the industry random effect.
Column (ii) shows the effect of firm and loan specifics to CDS spread only. Comparing to regression (ii) in Table 3, assets and revenue have gained significance, while leverage ratio lose a part of influence power though it is still significant. The estimators of EPS and stock price increased, and estimators of EBIT and risk-free rate decreased. The estimator of return on S&P gains a significant negative influence power, while the estimator of credit rating losses a significant negative influence power. The estimator of stock return changes from negative to positive. And facility maturity becomes more influential. After including industry random effect, firm and loan specifics have obviously changed their influence to CDS spread, so the result of regression (ii) in Table 3 is lacking of robustness. The reason could be that the firm and loan specifics are related to industry distributions. This regression does not include any dummy variable indicating treatment effect, and the results do not automatically fix the time and entity effect. So this kind of results is explainable.
Column (iii) shows that the estimator of MBP decreased from 61bps to 15bps. However, this estimator maintains 1% significance level. The robustness of interaction effect of MBP and period is quite stable. The estimators of EPS, risk-free rate and stock return have improved their significant level, and the estimator of facility size and S&P quality rating loss its significance. Regression (iii) in Table 3 is combined of the previous two regressions. And I can conclude that the result from the overall baseline regression in Table 3 is relatively robust under the condition that the control variables are related to industry distribution to some degree.
6.2 Effect of market-based-priced loan under different levels of credit ratings
In Table 5 I show the effect of market-based-priced loan to CDS spread under different levels of credit ratings. The regression includes control variables. Here I mainly focus on only dummy variables.
Table 5 Regressions on CDS spread under different rating levels. The sample contains all the data derived from the selection described in section 4.1 after winsorizing. This table shows the regressions on CDS spread under each credit rating level. MBP marks treatment group and control group. MBP takes a value of 1 if a firm belongs to treatment group and a value of 0 if a firm belongs to control group. Treatment group contains the firms that have been issued with market-based-priced loans. Control group contains firms that never had market-based-priced loan. Period marks the time before and after treatment. It takes a value of 1 if the time is after treatment, and a value of 0 if the time is before treatment. The time threshold is the second quarter of 2008. The t-statistics are presented in the brackets under each coefficient.
A+ A A- B+ B MBP -23.6446* 22.466*** 24.8576*** 14.2341*** 11.112 [-1.95] [3.38] [6.21] [2.74] [0.84] Period 75.5027*** 81.4772*** 75.2897*** 80.9303*** 81.354*** [12.77] [21.18] [16.49] [25.92] [41.07] MBP*Period -106.9436*** -131.1606*** -116.2586*** -123.9812*** -120.9536*** [-13.5] [-22.98] [-23.36] [-19.44] [-15.33] Constant 758.2273*** 74.3811 194.1649*** 85.7831*** 91.9734*** [3.86] [1.46] [5.44] [3.22] [3.74]
Control variables Yes Yes Yes Yes Yes
Observations 210 315 350 490 805
R-square 0.8042 0.8282 0.8074 0.806 0.8036
Coefficients are statistically significant at the ***1%, **5%, and *10% significance level.
The coefficient of MBP under credit rating A+ is -23.6, which is significant at 10% significance level. Instead of the estimator of MBP 16.26 in Table 3, this result means that the firms with S&P credit rating A+ have a CDS spread 23.6bps lower than the other
firms. That’s to say, if a bank choose an A+ rated firm to issue market-based-priced loan, the interest base rate of this facility will be lowered since A+ firms have a much lower bankruptcy risk. The other estimators of MBP under credit level A, A-, B+ and B share the same quality as the results in Table 3. So I can get the conclusion that when banks choose the target to issue market-based-priced loan, though they prefer a higher CDS spread to avoid a low interest rate, they still take good credit rating into consideration even the interest rate they can settle is not very high. However, banks still will to issue such loans to firms with bankruptcy risk. From the number of observations under each credit rating, I can see that the number of the best rated firm is quite low comparing to the number of worse credit rated firms. A+ credit rating firms can reduce their CDS spread by 107bps, which is the lowest in all other firms’. The other firms can get a reduction even more. So, the bankruptcy risk get less improved for the best rating firms, and the bankruptcy risk of firms with credit rating A get reduced most by market-based-priced loan.
7. Conclusion and discussion
In this paper I test whether the market-based-priced loan can lower the bankruptcy risk of firms. I set the time threshold as the second quarter of 2008 when the first market-based-priced loan got applied, and the treatment group as the firms that have been issued loans with interest rates tied to CDS spreads or CDX index. I employ the difference-in-difference model to eliminate the effect from time trend and firm quality difference, and got quite robust regression results. I get the main findings in this research as following.
First, the bankruptcy risk of the firms with market-based-priced loans has been lowered significantly. The empirical result shows that the CDS spread of those firms decreased by 122bps compared to firms with traditional loans. This result fits in the theoretical framework provided in the literature part. This result is in line with the theories in John Hull (2009), Acharya and Johnson (2007) about information effect on CDS spread. It is also partly in line with the result in Iuliana et al. (2010) in their event study. My result implies that the way of market-based-pricing offers extra positive information to the public. This type of loan works efficiently for improving the information transparency in the CDS market, and its spillover effect to the other markets
