The tradeoff between public, private and non-bank
funding: Evidence from REITs
Master thesis in Business Economics – Specialization Real Estate Finance March 2014
Student: Naut Biegel – 5672872 Supervisor: J.E. Ligterink
2
Content
1 Introduction ... 3
2 Literature, Theory and Existing Evidence ... 5
2.1 Capital Structure ... 5
2.2 The trade-off between public and private debt in the US market ... 6
2.2.1 The size hypothesis, public versus private debt ... 6
2.2.2. The credit quality hypothesis ... 9
2.3 The trade-off between public and private debt in the Real Estate sector ... 11
2.3.1 The size hypothesis ... 13
2.3.2 The credit quality hypothesis and REITs ... 16
2.4 Causality ... 17
2.5 Hypotheses ... 17
3 Data Collection and Sample Description ... 18
3.1 US market sample. ... 18
3.2 Real Estate sample ... 21
4 Methodology ... 26
4.1 regression approach ... 26
4.2 Models used ... 27
4.2.1 The US market model ... 27
4.2.2 The Real Estate model ... 28
4.3 Some concluding remarks ... 28
5 Results ... 29
5.1 Cross-section of the US market. ... 29
5.1.1 US market regression output ... 30
5.2 Real Estate sample ... 33
5.2.1 Regression output ... 35
5.3 US market sample versus Real Estate sample ... 38
6 Conclusion ... 39
Bibliography ... 41
Appendix 1 – The Altman Z-score ... 45
3
1 Introduction
Originally, Real Estate projects are financed through bank funding. However, the last decades there has been a shift towards more complex forms of financing (Hancock & Wilcox, 1997 and Davis & Zhu, 2009). The United States Private Placement or “USPP” is one of these newer forms of finance. Through an USPP a company enters the private market of the United States to attract funding for its intended project. Counter parties of an USPP can be mutual funds, insurance companies or pension funds. One of the key elements of an USPP is that because a private placement is offered to a few select “private parties”1, the placement does not have to be registered with the Securities and Exchange Commission. Borrowers do not require a public rating but do need to have a National Association of Insurance Commissioners rating (NAIC) by the Securities Valuation Office2.
Although the USPP is an important type of funding for both US corporations, as well as an increasing amount of non US companies, USPP as a use of Real Estate finance was granted relatively little attention in the 20th century academic literature (Carey, Prowse, Rea & Udell, 1993). Even though the USPP, now days, has found its place in more recent academic literature, Lee and Kocher (2001) provided research regarding characteristics of firms using private placement methods of issuing common stock with those using public offering methods and Denis and Mihov (2003) examine the trade-off between public and (non-bank) private debt by means of firm characteristics. The role of the United States Private Placement in the capital structure theory in the Real Estate industry lacks a solid academic foundation. Despite the lack of thorough coverage in the current academic literature, global Real Estate agencies label the Real Estate related USPP as “the new hot thing”3.
While USPPs can be considered to be private transactions, a lot of data concerning USPPs has been made public. The presence of data and the relatively low amount of published papers combined with the rapid increase of its usage make the USPP a viable research subject.
Throughout this research, it’s investigated why a company attracts financing through an USPP to fund its Real Estate project instead of another (more conventional) way of financing. Firm characteristics will be used to form an empirical model which determines the probability of a company’s choice to use an USPP to fund its Real Estate project. After constructing a
1 The definition of Private Parties can also refer to Qualified Institutional Buyers (QIBs) 2
This insight is obtained through a lecture of the CBRE Global Investors EMEA Treasury & Debt Financing department.
3
This statement was given during a lecture of the CBRE Global Investors EMEA Treasury & Debt Financing department.
4 theoretical framework it is investigated which company characteristics contribute to the type of funding. The empirical model that is used during this research will be a LOGIT model which will correct for fixed effects. First, a benchmark is constructed by only modeling the probability of non-Real Estate companies in the US. This benchmark is compared with modeled probability of solely Real Estate companies in the US, in order to investigate if companies active in the US Real Estate sector follow the same pattern as non-Real Estate US companies.
This research provides new insights in the different financing possibilities regarding Real Estate projects. The main contribution of this research is the answer to the question if the current academic theory regarding a firm’s size and risk levels and its type of funding also apply to companies active in the US Real Estate sector.
The findings of this research are relevant since they deliver a contribution to the existing literature due to the fact that there hasn’t been much research conducted regarding USPPs as Real Estate funding. As mentioned above, this relatively new form of financing in the field of Real Estate is regarded by professionals as “the new hot thing”. Therefore, any research that provides more insight in this way of financing can be defined as relevant. This research uses the findings of Lee and Kocher (2001) as well as the findings of Denis and Mihov (2003) regarding firm characteristics and their relation with corporate capital structures and projects these findings on a sample of companies active in the Real Estate sector. By doing so, this research expands the current academic literature covering the capital structure theory from a Real Estate perspective.
The outline of this thesis is as follows. The second paragraph discusses the current literature covering USPPs with a focus on Real Estate funding. By doing so, two hypotheses are formed regarding capital structure. The third paragraph elaborates on the data and samples used during this research. The fourth paragraph explains the methodology of this research. The fifth paragraph provides the results found during this research and tests the hypotheses from paragraph two. The sixth paragraph concludes.
5
2 Literature, Theory and Existing Evidence
The following paragraph covers the current literature and existing theory regarding the trade-off between public and private debt in the US market. By forming a general theoretical framework regarding the trade-off between public and private debt in the US market, a benchmark is formed which is used further on in this research to compare this trade-off with respect to companies active in US Real Estate sector.
2.1 Capital Structure
With their theory based on frictionless financial markets, Modigliani and Miller (1958) were the founders of the traditional capital structure theory (Leland, 1994). Modigliani and Miller (M&M) argued that under specific theoretical assumptions, firm value was unaffected by its capital structure. Although the M&M theorem was ground breaking and most definitely the foundation of the current capital structure literature and theories, it only provided a qualitative guidance (Leland, 1994). In an attempt to complete the works of Modigliani and Miller, numerous authors have tried to provide a quantitative as well as qualitative backbone for the M&M theorem. Although many of them succeeded, the theorem is still founded on severe theoretical assumptions. Therefore capital structure remains an interesting and relevant topic in the world of finance. One of the puzzles of capital structure is the trade-off between public and private debt (Myers, 1984; Denis and Mihov, 2003).
In this research the definitions of public and private debt as used by Denis and Mihov (2003) are adopted. These authors refer to public debt as capital raised through public offerings and private debt as capital raised through other means than public offerings (Denis and Mihov, 2003). Although the definition of public debt speaks for itself (simply a way of funding using public offerings), the definition of private debt may need some further explanation.
Private debt can roughly be separated into two categories, bank and non-bank debt. Non-bank debt is usually offered by so called Qualified Institutional Buyers (QIBs). QIBs can vary from insurance companies to pension funds. Both categories of private debt have their own specific characteristics regarding risk aversion, adverse selection and moral hazard (these characteristics are discussed in more detail in the following paragraphs).
In general, the current literature agrees that lenders, which are active in the private credit market, are better suited to provide credit to “information problematic” borrowers compared to public creditors. These so called “private lenders” are also better suited to provide credit to
6 borrowers with a higher risk profile. This is mainly the result of their higher financial compensation and more evolved monitoring skills (Diamond, 1991; Fama, 1985).
2.2 The trade-off between public and private debt in the US market
The following paragraph discusses the current literature regarding the trade-off between public and private debt. The foundation of academic papers that covers the choice between public and private debt is quite large. However, although many authors have extensively explored this topic (Diamond, 1984; Fama, 1985; Rajan, 1992; Lee and Kocher, 2001; Denis and Mihov, 2003), very little research has been conducted with respect to financing choices regarding Real Estate projects. After discussing the most relevant literature regarding the trade-off between public and private debt, this paragraph describes the limited research performed regarding the choice between private and public debt when financing Real Estate projects.
2.2.1 The size hypothesis, public versus private debt
Among others, Diamond (1984), Fama (1985) and Rajan (1992) have discussed the trade-off between public and private debt extensively. Diamond (1984) and Fama (1985) hypothesized that private debt has significant advantages over public debt, respectively monitoring efficiency and access to private information. Rajan (1992) as well as Rauh and Sufi (2010) argued that private debt can also have disadvantages, such as the distortion of management incentives. Banks are able to control the investment decisions of a company through extensive monitoring. By controlling the investment decisions of a firm, the bank distorts a firm’s incentives which can make it a less desirable form of funding (Rajan, 1992)4.
The predominant part of the literature covering the choice between public and private debt does so in the light of information asymmetry, risk aversion, bargaining power of debt holders and the agency costs of debts (Denis and Mihov, 2003). During the following paragraphs these elements are discussed and used to form several hypotheses which are tested during this research.
4
The effect of capital structure on the distortion of management incentives, although being very interesting, is beyond the scope of this research. For deeper insights regarding this topic I refer to the paper of Rauh and Sufi (2010)
7 In their study on behalf of the Board of Governors of the Federal Reserve System Carey, M., Prowse, S., Rea, J., and Udell, G. (1993). come to the following conclusion regarding a firm’s decision of the type of financing it should attract. Borrowers’ access to debt markets is closely related to firm size, with size serving mainly as a proxy for the degree of information-problems that borrowers impose on lenders. In other words, the smaller the size of a company, the more difficult it will be for (public)debt markets to obtain all the desired relevant information in order to decide whether to invest in a company or not.
In their report Carey et al (1993) do not focus on one specific industry but try to form an analysis for the entire US market. Their study’s main findings report that the market differs between firm characteristics. At one end of the scale are small, relatively unknown firms posing significant information problems that require extensive due diligence or loan monitoring by lenders. These firms only tend to have access to relatively short-term loans provided by banks and other bank-like intermediaries, which have the staff and expertise to undertake information-intensive lending and limit borrowers’ risk-taking behaviour through tight covenants or collateral in loan agreements. On the other end of the scale, somewhat less information-problematic, typically larger firms can issue in the private placement market.
According to Krishnaswami, Spindt and Subramaniam (1999), the majority of the lenders in the private placement markets are large institutions, such as commercial banks and (life) insurance companies which, over time, became specialised in executing comprehensive credit evaluations and the monitoring of issued debt. Due to these lender characteristics, the authors hypothesize that smaller firms, which have to cope with severe information problems and will have limited access to public markets due to the relatively high flotation costs, will chose private debt instead of public debt. These findings result in the size hypothesis which predicts that firm size has a positive correlation with public debt and therefore a negative correlation with private debt.
This size-hypothesis is in line with the authors findings regarding flotation costs(which can be seen as a proxy for firm size), which is confirmed in the paper of Blackwell and Kidwell (1988). Both Krishnaswami, Spindt and Subramaniam (1999) and Blackwell and Kidwell (1988) come to the conclusion that public debt issues are in general associated with higher flotation costs compared to private debt issues. These obligatory expenditures form a costly entry barrier for smaller companies. Besides the high flotation costs, Chittenden et al. (1996) conclude that, based on the findings of Backland and Davis (1990), initial public offerings, especially for smaller companies, tend to be subjected to under-pricing. This under-pricing makes public offerings less favourable for these types of companies.
8 Furthermore, after obtaining a flotation on a stock market, small firms often find themselves being subjected to the so called “small firm effect” (Chittenden et al. 1996). The small firm effect refers to the situation when a smaller firm has a higher cost of equity, for a given amount of risk, compared to larger companies (Banz, 1981; Reinganum, 1981).
Chittenden et al. (1996) also provide empirical evidence that owner-managers of smaller firms face a higher takeover threat due to loss of control after an initial public offering. The size hypothesis predicts that firm size has a positive correlation with public debt and therefore a negative correlation with private debt. These findings are confirmed in the more recent study of La Rocca, La Rocca and Cariola (2011).
One of the limitations of the current literature is that it does hardly ever distinguish between bank private debt and non-bank private debt. Fama (1985) is one of the first few authors that specifically acknowledges non-bank private debt.
2.2.1.1 Bank versus non-bank debt
The size hypothesis as described above does not provide any theoretical insight in the influence of size on the choice between bank and non-bank debt. In order to apply the size hypothesis on the choice between bank and non-bank debt, first the specific characteristics of both debt types have to be defined. The main difference between bank debt and non-bank is the difference in concentration and identity of debt holders, regulatory requirements, maturity and placement structure (Denis and Mihov, 2003).
According to Denis and Mihov (2003) the distinction between non-bank private debt and bank private debt is neglected in the current academic literature. Private non-bank debt can be classified as a relatively illiquid form of debt (Kwan and Carelton, 2004). Other characteristic which are often associated with non-bank private debt are relatively low flotation costs and design covenants (Denis and Mihov, 2003). These custom-designed covenants are possible due to the high concentration of ownership5.
The regulatory differences between bank debt and non-bank private debt can be traced back to the large, and continuously increasing, amount of legislation and requirements involved with bank funding due to the potentially substantial macro-economic issues that can occur as
5
Concentration refers to the number of financing parties. Where private bank debt is collected from only several parties. Non-bank private debt can be obtained through multiple parties, mostly institutional investors.
9 a result of falling banks (Allen, Chan, Milne & Thomas, 2012). Due to the increasing compliance obligations of banks, bank debt will become more expensive.
A substantial part of all private placements that occur in the US market is being placed under SEC Rule 144A (Denis and Mihov, 2003). Rule 144A was approved by the Securities and Exchange Commission in April 1990 (Chaplinsky and Ramchand, 2004). The SEC Rule 144A provided companies with the possibility to directly market debt to (private) Qualified Institutional Investors (QIBs). Instead of going through the process of issuing public debt which is much lengthier and time-consuming.
According to Fenn (2000) Rule 144a is broadly used by below-investment grade firms to rapidly raise funds that are subsequently registered. Combining the findings of Diamond (1991), (Denis and Mihov, 2003), Wald (1999) and Fenn (2000) one can conclude that private placements, filled under SEC Rule 144A can serve as a proxy for non-bank private funding.
2.2.2. The credit quality hypothesis
Besides the previously mentioned size hypothesis, the credit quality hypothesis is also a recurring argument in the current literature. This hypothesis states that more riskier firms, which realize the lowest project quality and the highest bankruptcy probability, will be the most suitable candidates for private non-bank financing (Denis and Mihov, 2003). The reasoning behind this statement is that private investors are, in general, more skilled investors compared to the common man buying stock through the public debt market. These private investors are professionals who are more comfortable within certain sectors and with certain levels of risk compared to public investors.
Denis and Mihov (2003) claim that non-bank private debt, issues under SEC Rule 144A, can be associated with several features of both bank debt and low-grade public debt issues. Debt issues under SEC Rule 144A are tightly held and relatively illiquid and have lower flotation costs than publicissues and have custom-designed covenants. According to the authors, this can make non-bank private debt issues very suited for firms with a poor credit quality. Additionally, Carey, Post and Sharpe (1998) present the possibility that bank regulators can potentially discourage the issuing of bank debt to low credit quality firms through the large loan loss reserves requirements which are mandatory for these specific loans.
Diamond (1991) investigates the choice between borrowing directly (issuing publicly traded bonds or commercial paper, without monitoring) and borrowing through a bank that monitors
10 to alleviate moral hazard. The intuition behind this trade-off is that, through monitoring applied by banks, management decisions can be influenced. Firms can experience this as a negative feature of using bank debt. Therefore firms might be tempted to try to finance their self through public funding. One of the author’s findings is that there is a "life cycle" effect in the use of borrowing through intermediaries. New debtors often borrow from banks initially but may later issue debt directly, without using an intermediary. A borrower's credit record acquired when monitored by a bank serves to predict future actions of the borrower when not monitored. Wald (1999) comes to similar conclusions. He finds that especially U.S. capital structures may be more sensitive to default risk compared to other countries. In other words, according to the empirical findings of Wald (1999), US companies are judged more severe on their capital structure compared to other countries. This confirms Diamond’s (1991) findings regarding the credit quality hypothesis that firms with a higher risk profile opt for private funding.
Guedes and Opler (1996) investigate the determinants of corporate debt maturity6, one of their findings is that firms with speculative grade credit ratings, and therefore a higher risk profile, usually attracted debt from the middle of the maturity spectrum. According to the authors, this finding is in line with the theory that companies with poor credit quality are reluctant to issue short-term debt due to the possibility of inefficient liquidations. But the market does not define these companies eligible for long-term debt due to their more risky asset base. The authors claim that the majority of debt from the middle of the maturity spectrum will most likely consist out of private debt. Denis and Mihov (2003) confirm the claim of Guedes and Opler (1996). In their research the they find a private debt maturity median of 8.2 years. This finding supports the claim that, regarding a 20 year horizon used by Guedes and Opler (1996), private debt has a maturity around the middle of the maturity spectrum.
Denis and Mihov (2003) found a public debt maturity median of 10 years. This implies that these type of firms do not issue short-term debt in order to avoid inefficient liquidation, but are screened out of the long-term debt market because of the prospect of risky asset substitution. The findings of Guedes and Opler (1996) provide further empirical foundation for the credit quality hypothesis.
The credit quality hypothesis, hypothesizes that credit quality, or risk profile, is positively correlated with a firms probability to attract private debt. Furthermore, It is hypothesize that companies with a higher risk profile, and therefore a lower credit quality, will seek funding under SEC Rule 144A which can be seen as a proxy for non-bank funding.
11 Although it might seem counter intuitive to claim that private investors prefer riskier debt, this statement is more logical then it appears at first sight. Private investors will be more specialized within certain sectors. Due to this informative advantage they will be more comfortable with certain levels of risk. Public investors, who often lack this informative advantage, will avoid these type of investments (Diamond, 1984 and Denis and Mihov, 2003).
2.3 The trade-off between public and private debt in the Real Estate sector
Where the previous paragraph covered the trade-off between public and private debt in the US market in general, the following paragraph covers the same trade-off in the Real Estate sector. By doing so it will be investigate whether Real Estate can be subjected to the same hypotheses discussed in the previous paragraph. There is little research on the trade-off between public and private debt in the Real Estate sector.One of the difficulties of Real Estate research is that most Real Estate markets are characterised by incomplete information, costly search, and varying expectations as a result of the appraisal based nature of the value of underlying assets (Quan and Quigley, 1991). In order to mitigate these specific Real Estate market characteristics, Real Estate Investment Trusts (REITs) will be used as a proxy for the Real Estate sector. Using REITs as a proxy for the Real Estate sector is in line with the research and findings of Giliberto (1993) who measured Real Estate returns with a REIT index, Liang and Webb (1996) who analysed the role of Real Estate in a mix portfolio and used REITs as Real Estate proxy and Benefield, Anderson and Zumpano (2007) who investigated the position of the market portfolio with an implementation of REITs. Due to their transparent nature, a typical REIT discloses an abundance of standardized financial data, REITs provide ideal data for cross-section analyses.
REITs entered the investment scene in the late 1960s after the Congress of the United states passed the Real Estate Investment Trust act. Their main goal was to broaden the investment spectrum beyond securities such as stocks or bonds. Currently, the REIT act has been acknowledged in over 30 counties (Brounen and de Koning, 2012). When used as a proxy for the Real Estate sector one should take some REIT specific characteristics into account.
Geltner, Miller, Clayton and Eichholtz (2007) provide a description of characteristics which are REIT specific. The authors state that the most important feature of a REIT, when compared with different types of stock, is the exemption from corporate income tax. Although
12
this statement is accepted throughout the academic literature, it is technically not correct. The dividends which are being paid by REITs to their investors are tax deductible. However, since REITs have the obligation to pay out the vast majority of their earnings in the form of dividends, this results in a corporate tax bill of approximately zero.
The reasoning behind this “tax exemption” is that REITs are classified as investment vehicles, comparable to mutual funds. The deductibility for tax purposes of the dividends paid by REITs enables investors to avoid double taxation of corporate income that characterizes the majority of stocks (Geltner et al, 2007). However, this somewhat encouraging tax treatment comes with a price in the form of multiple regulatory constraints. The rationale behind these constraints is to keep REITs as a more passive investment vehicle and to make sure it remains and accessible to small individual investors. If a company want to qualify as a REIT, it must pass 4 tests on an ongoing basis.
To be qualified as a REIT a company has to pass the so called “Ownership Test”. A REIT cannot be a corporation with concentrated ownership. This is defined by Geltner et al (2007) as follows: “no five or fewer individuals (and certain trusts) may own more than 50% of the REIT’s stock, and there must be at least 100 different shareholders”. US legislation contains an exemption on this rule regarding pension funds. Pension funds are considered to represent all the participants in the pension plan. This exemption is known as the “look-through” provision.
Besides the previous described Ownership Test, a company also has to pass the so called “Asset Test”. To pass this test, at least 75% of the total assets must be real estate, mortgages, cash or federal government securities. Besides this asset constriction, at least 75% of the annually gross income must, directly or indirectly, be the result of real property. A relatively new development in the “asset test” is that as of 2001 REITs are allowed to form a so called “taxable REIT subsidiary”. Through this subsidiary REITs can provide services to
tenants which were previously not allowed7. A maximum of 20% of the assets can be made
up of stock of a taxable REIT subsidiary.
Next to the Ownership Test and the Asset Test a company also has to pass the “Income Test”. In order to pass this test, a company must derive the majority of their income, at least 75%, from passive sources. According to Geltner et al (2007), the term “passive sources” in this definition relates to: “rents and mortgage interests, as distinct from short-term trading or sale of property assets”. In other words, companies are not allowed to use the favourable tax legislation, which are a characteristic of REITs, to shield income from a non-real-estate
13
nature. In order to force the Ownership Test, a REIT will face a tax level of 100% on all net income which is the result of prohibited transactions like the disposition or sale of certain properties which are solely held for disposal in the ordinary course of its trade of business. The investment horizon of a no prohibited transaction is at least four years. This forces REITs to have a more long-term investment horizon.
The fourth test a company has to pass to gain REIT status is the “Distribution Test”. In order to pass this test at least 90% of the yearly taxable income must be paid out as dividends. Because of the Distribution Test, REITs are very, if not completely, dependent on external funding since they cannot reinvest their earnings.
Only when a company passes the four tests described above, on a continuous basis, it can gain the status of REIT.
2.3.1 The size hypothesis
Although REITs exist over 50 years, the current literature only briefly addresses capital structure and the trade-off between private and public debt (especially with the distinction being made between private bank debt and private non-bank debt). One of the few papers covering REIT capital structure, with a focus on private bank debt is the paper of Hardin and Wu (2010).
The authors focus their research on the evolution of capital structures applied by REITs. They specifically target the effects of banking relationships and the use of bank debt on capital structures applied by REITs. In order to analyse the evolution of REIT capital structure Hardin and Wu (2010) pose the following research questions. Are REITs with banking relationships more likely to have access to public debt markets? Does the development of banking relationships help reduce the use of secured debt in a REITs liability structure? Do REITs with banking relationships have bigger or lower leverage? To answer these question the authors use empirical analyses on a REIT dataset which is constructed with the use of three different data sources.
To find an answer on the first question, the relationship between banking relationships and access to public debt markets, the authors use a probit model8. Hardin and Wu (2010) find a positive relationship which is significant at the 5% level. To find an answer on the second question, the relationship between bank relationships and the use of secured debt, the
8
A probit model is similar to the logit models which I use in my thesis. These models will be discussed in more detail during this thesis.
14 authors use a standard OLS model with a secured debt ratio as dependent variable. In order to correct for correlations across their observations the authors also construct an instrumental variable model. The authors find a strong negative relationship which is significant at the 1% level between bank relationships and the use of secured debt. To find an answer on the final question, the authors use the same model as was used for the second question only with a different dependent variable, leverage. They find a negative relationship which is significant at the 5% level.
Furthermore the authors find that from 1992 until 2003, REITs drastically increased their use of bank debt from $329 million to $13,405 million. According to the authors, this increase is mainly the result of the rapid growth in property acquisitions, development and mergers representing the core growth strategies employed during the horizon used for the investigation. Due to the previously discussed distribution test, REITs have relied enormously on funding provided by banks.
Additionally, the authors found that size (measured as the natural logarithm of total revenue) significantly positively influences if a REIT has more public debt. The authors use long-term S&P issues as a proxy for public debt (Hardin and Wu, 2010). The significance of size found by Hardin and Wu (2010) is in line with the size hypotheses described in the previous paragraphs. Unfortunately however, the authors limited their research to public and private bank debt and do not provide an insight in different types of private debt.
15 Next to this rapid growth, the authors conclude that REITs significantly reduced their use of mortgages as means of funding. Instead they shifted to public and bank debt.
Figure 1: Total amount of mortgage debt in $m
Source: Hardin and Wu (2010)
Figure 2: Total amount of public and bank debt in $m
Source: Hardin and Wu (2010)
The conclusion of Hardin and Wu (2010) that REITs attract increasingly more bank debt acknowledges the need for a more, in-depth, knowledge of this topic which is another justification of this thesis.
0 500 1000 1500 2000 2500 3000 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
16 The current academic literature, covering REITs, only provides limited evidence that REITs are subjected to the above described size hypotheses. Hardin and Wu (2010) limit their research to the trade-off between public and private debt. By doing so, the authors neglect the relationship between different types of private debt and the size hypothesis This lack of coverage in the current academic papers, again, indicates the relevance of this thesis. By investigating the potential impact of REIT size on its capital structure with a focus on non-bank private debt, this research provides an insight in whether non-non-bank private debt REIT funding is related to firm size. It therefore contributes to the continuous growth of capital structure literature.
2.3.2 The credit quality hypothesis and REITs
Devos, Spieler and Tsang. (2011) investigate the REITs institutional ownership dynamics influenced by the crisis. The authors provide evidence that institutional investors seek to limit their exposure to firm specific risk and therefore shift to REITs with a higher credit quality (Devos et al, 2011). According to the findings of the authors, institutional investors ownership of REITs declines significantly during times of market stress. Furthermore, Devos et al (2011) conclude that REITs actually actively manage their portfolio’s and display a so called “flight to quality” and are avoiding REITs with lower credit quality.
These results imply that riskier Real Estate firms, having the lowest project quality and highest probability of default, are not the best candidates for private, non-bank debt financing. These findings are not in line with the credit risk hypotheses that firms with the lowest quality projects, and therefore a higher risk profile, will attract private debt. Chikolwa (2011) investigated the capital structure of Australian REITs. One of his key findings was the significant negative correlation between risk, measured as the variability of expected earnings, and leverage. This finding can be seen as counter intuitive however Chikolwa (2011) claims that, when it comes to REITs, firms with low growth potential will be more vulnerable to business fluctuations in business outlook and therefore will be sensitive to financial distress. These companies will chose a lower leverage ratio. In other words, the lower the risk profile of a company the more debt it will attract. Unfortunately Chikolwa (2011) does not conduct any research regarding the type of debt. According to the author, information covering the specific private debt types is very difficult to come by in Australia. The author acknowledges the future research potential regarding the distinction between different debt sources. This finding provides more empirical evidence for the argument that credit quality has an influence on capital structure of REITs.
17 Although the current literature covering the credit quality hypothesis regarding REITs is very limited, there seems to be evidence which supports the hypothesis. Based on the findings of Chikolwa (2011) one can argue that the credit quality hypothesis also applies to the Real Estate sector. However, the findings of Devos et all (2011) contradict these findings.
2.4 Causality
Unfortunately none of the papers discussed above mention the possibility of endogeneity. This makes it very difficult to assume causality regarding their findings. It also poses several difficulties for this specific research since the estimation techniques, sample selection and models used in this research are based on the theoretical framework constructed through these papers.
It is therefore very important to distinguish between correlation and causality during this research. Although the papers used for this research take causality for granted I will be limited to more general conclusions regarding my empirical findings.
2.5 Hypotheses
Numerous studies have been conducted regarding the trade-off between public and private debt. However, very few took into account the Real Estate sector on its own nor did they specify the specific type of private debt. This research will therefore significantly contribute to the current literature regarding the trade-off between public and private debt. Based on the current literature and prevailing theories previously described the two following hypotheses have been formed:
1. Smaller firms which have to cope with information problems will chose private debt instead of public debt.
2. The riskiest firms, having lowest project quality and highest probability of bankruptcy will be the most likely candidates for private, non-bank debt financing.
The third paragraph will describe what type of sample will be used to test these two hypotheses. The fourth paragraph will describe what methodology was used to test the two hypotheses. The fifth paragraph will provide an insight in the results and the sixth paragraph will form a conclusion and discussion for further research.
18
3 Data Collection and Sample Description
The following chapter describes the construction of the sample used to test the size and credit quality hypotheses. Since the hypotheses were tested for both the US corporate market as well as the Real Estate sector I needed to collect data of both a cross section of the US market as well as REITs. Since it was a rather difficult technical process to merge all data source in to one sample, I chose to construct two data samples9.
3.1 US market sample.
The US market sample is constructed out of finance applications from US listed companies between 1990 and 2005. This timespan was specifically chosen to generate sufficient data points to perform a regression, the regression approach will be discussed in more detail in the following paragraphs, without any interference caused by the current financial crises.
Through the Thompson One data base 26,304 finance applications where selected in the time span of 1990 until 2005. From these 26,304 selected finance applications, 19,009 disclosed if they concerned a private or public placement. All companies which did not disclose if their finance applications where either private or public were discarded since private placement dummies were essential, dependent variables, for this research.
After controlling for SEC Rule 144A disclosure, 7,387 debt applications remained useful, i.e. disclosed if SEC Rule 144A was applicable or not, all others were discarded. In order to match financial and firm characteristics with each individual finance application, all 7,387 debt applications were linked with an unique ticker symbol. All debt applications that did not disclose a ticker symbol were discarded. From the 7,387 applications, 1,604 did not provide a ticker symbol. This resulted in 5,783 remaining debt applications. These remaining 5,783 applications were sorted per year and, through ticker symbols and matched with relevant financial data.
In line with the research of Denis and Mihov (2003) the COMPUSTAT database was used to provide the relevant accounting data needed for the regression analysis. After discarding all debt applications that did not provide all accounting data needed, 2,709 debt applications remained forming the US-Market sample. During the composition of this sample no companies active in either the Financial or Energy & Power sector were taken into account to avoid large outliers as a result of the capital nature of these sectors. Besides the exclusion of
19 the Financial and Energy & Power sectors, companies active in the Real Estate sector were also not taken in to account to avoid dilution of the cross-section analyses10.
As is described in the previous paragraph, the main purpose of this research is to test the size and credit quality hypotheses on the Real Estate sector of the US compared to a cross-section of the US market, through this comparison insight will be provided in the role of private and non-bank private debt in REIT capital structure. To test the size hypothesis, the variable “SIZE” where size is measured as the natural logarithm of sales is used. In order to test the credit quality, in line with the research of Denis and Mihov (2003), the Altman score for each individual company as a proxy for credit quality was computed. The Altman Z-score will be discussed in more detail in appendix 1.
Besides the “SIZE” variable and the “Altman Z-score” variable, multiple control variables have been constructed which are the following: the natural logarithm of total assets and liabilities ratio ln(Asset Liability ratio), the level of research and development cost (R&D) and the number of employees (Employees).The table displayed below provides brief descriptive statistic of the sample variables. These control variables have been selected based on the research of Denis and Mihov (2003), Walt (1999) and Diamond (1991). These control variables will be discussed in more detail later this research.
Table 1 provides the number of observations Obs, the Mean, the standard deviation Std. Dev. As well as the minimum and maximum values of the variables incorporated in the model. The computation of the Altman Z-score is addressed in more detail in the appendix. Size is the natural logarithm of Sales in 1,000$. R&D is measured in 1,000,000$. Employees is the number of employees measured in thousands.
Based on the current academic described in the previous section, it is expected that the Altman Z-score, being a proxy for credit quality, will have a negative relationship with the non-bank private debt dummy in the regressions using the data based on the cross section of the US market. Furthermore, it is anticipated that Size will have a negative relationship with the private placement dummy when analysing the cross section of the US market.
10 I want to investigate the difference between real estate and non-real-estate companies.
Obs Mean Std. Dev. Min Max
Altman Z-score 2,779 5.6925 7.9725 0.2006 50.4006
Size 2,717 4.7246 2.5609 -2.2600 9.9700
R&D 2,053 47.6479 123.5083 0 956
Employees 2,707 8 27 0 480
Ln( Asset Liability ratio) 2,779 -0.9337 0.7966 -4.6635 2.0839
Table 1
20 Based on the empirical findings of Walt (1999) it is expected that R&D expenditures will have a positive relationship with private funding. R&D can be seen as a proxy for firm specificity and, according to the findings of Titman (1984), firm specificity increases a firms debt/asset ratio. In other words, increasing firm specificity increases a firm’s debt level. Titman (1984) and Walt (1999) do not make a distinction between private and public debt. Since this specific distinction will be made in this thesis, additional insight can be obtained. As described in the previous sections covering the credit quality hypothesis, it is to be expected that private debt is positively correlated which higher, more specific, risk profiles. Extrapolating this statement results in the anticipation that firms with higher specificity or, using the definition of Titman (1984), higher R&D expenses will have higher (non-bank) private debt levels.
According to the findings of Hanka (1998), firms with higher debt levels have a history of reducing their employment levels more often. Therefore it is expected that number of employees will have a negative relationship with debt levels.
The variable Ln(Asset Liability ratio) is the natural logarithm of the asset-liability ratio. This ratio is used as a proxy for debt history of a company, the rationale behind this is as follows. The higher the asset-liability ratio is the less debt a company has. Both Denis and Mihov (2003) as well as Walt (1999) provided evidence that the current debt mix of a company has an impact on new debt applications11. Therefore it is expected that the natural logarithm of the asset-liability ratio will have a positive relationship with private debt applications.
All relevant variables have been corrected for outliers through winsorization at the 1% level. This correction technique replaces the outliers which are located in the 1% level in order to avoid the loss of data points. This technique is available in most econometric software packages To provide extra insight in the data and to be able to correct for sector fixed effects in the regression, sector dummies were addressed to the finance applications.
11
Denis and Mihov (2003) found that not having issued (rated) debt previously, ceteris paribus, is
indeed an indicator of lower credit quality, because reputation plays a very important role in establishing the firm’s creditworthiness.
21 A distinction was made between: Healthcare, Industrials, Materials, High Technology, Consumer Products and Services, Telecommunication, Media and Entertainment, Retail and Consumer Staples. Graph 1, displayed below, provides a summary of finance application per sector.
The displayed graph reveals that the sectors Healthcare and Information technology generate the majority of finance applications in the US market followed by consumer discretionary.
Besides the sector dummies, each application was linked with a “year dummy” to correct for any possible time fixed effects during the regression process.
3.2 Real Estate sample
The Real Estate sample is constructed out of finance applications from US REITs between 1990 and 2005. This timespan was specifically chosen to generate sufficient data points to perform a regression, the regression approaches will be discussed in more detail in the following paragraphs, without any interference caused by the current financial crises.
The Real Estate sample has been constructed in a similar way as the US market sample has been composed. For the same time span, 1990 – 2005, finance applications were selected from US based companies through the Thompson One database. Only this time, a Real Estate sector filter was applied to obtain only the applications that were related to the US Real Estate Sector. This initial filtered query resulted in 1.959 Real Estate sector related debt applications. After discarding all applications that did not disclose if they related to public or
0 100 200 300 400 500 600 700 800 900 1000
Graph 1 - US sample
Finance applications per sector
22 private placements, 1,580 applications remained. From those 1,580 applications, 374 disclosed if SEC Rule 144a was applicable or not all others were discarded. From these remaining applications, 236 disclosed ticker symbols. After sorting the remaining debt applications per year, the COMPUSTAT data base was used to provide the relevant financial company data. Besides the COMPUSTAT data base, the Ziman REIT database was used to define REIT property types as well as REIT types, Equity, Mortgage of Hybrid. The difference between Equity, Mortgage and Hybrid REITs can be stated as follows.
When a REIT is labelled as Equity REIT, it will invest in its properties and therefore be the legal owner of them. In other words they are responsible for the equity value of their properties.
When a REIT is labelled as Mortgage REIT, it will invest in property mortgages. In other words, they will provide funding for owners of Real Estate, or acquire existing mortgages as well as mortgages-backed securities.
When a REIT is labelled as Hybrid REIT, it will simply be active in both the Equity as well as the mortgage market.12
Graph 2 provides an overview of the REIT types present in the sample. As can be seen in graph 2, the majority of REITs in the Real Estate sample were equity. The percentages of mortgage and Hybrid REITs are similar. Since the Equity REITs form the vast majority of the US REIT industry, this sample can be seen as good comparison with the US REIT population.13
12
All REIT type information has been acquired through the sec website, www.sec.gov.
13 US REIT industry compilation information was obtained through www.reit.com.
89%
5% 6%
Graph 2 - REIT types in sample
Equity Mortgage Hybrid
23 After discarding any application which did not provide sufficient financial data, 102 Real Estate orientated debt applications remained.
In order to test the size hypothesis, a “SIZE” variable was constructed which was measured by natural logarithm of revenue. This proxy for REIT size is in line with the research of Hardin and Wu (2010) and provides a good comparable for the size proxy used in the US Market sample, the natural logarithm of sales.
Although the size hypothesis can be tested with similar variables for both samples, the credit quality hypothesis requires a slightly different approach. As described in the previous sub paragraph, the credit quality of the companies based in the US Market sample was measured with the Altman Z-score as credit quality proxy. However, due to its working capital nature (see appendix 1) the Altman Z-score cannot be applied to REITs due to the absence of required balance sheet elements like, among others, current assets and current liabilities. As a substitute for the Altman Z-score, REIT stock price variance was used to determine credit quality. Price variance as proxy for REIT total risk is in line with the research of Gyourko and Nelling (1996) Besides the stock price variance, an “interaction variable” between the stock price variance and REIT beta was constructed to incorporate systematic risk in the model as well. Through this “interaction variable” additional insight might be obtained regarding the credit quality of REITs and its effect on the capital structure. REIT beta levels and stock variances were obtained through ticker symbols from the CRSP database.
Besides the “SIZE” variable and the “Variance” variable, multiple control variables have been constructed which are the following: the number of employees (Employees), The natural logarithm of the sale of stock in the year of the application ( ln(SoS) ), the level of retained earnings (RE), and total liabilities (TL) as well as an interaction term between a firm’s variance and a firm’s beta.
Similar to the regression model used for the cross-section of the US market, the number of employees is used as a control variable. This variable was chosen based on the findings of Hanka (1998). The author provided evidence that firms with higher debt levels have a history of reducing their employment levels more often14. Therefore, as previously mentioned, it is expected that number of employees will have a negative relationship with debt levels.
The natural logarithm of sale of stocks is used as a proxy for funding history. When a company increases its sale of stock, less funding from other sources will be required.
14
Hanka (1998) found that a 10th to 90th percentile difference in debt is associated with 35% greater expected workforce reductions.
24 Therefore, it is expected that the sale of stocks has a negative relationship with debt applications.
Retained Earnings is expected to have a negative relationship with funding applications due to the reduced need for external funding. Although the previous discussed distribution test, REITs have to pay-out at least 90% of their earnings as dividends, Ott, Riddiough and Yi (2005) argue that REITs still rely, be it minimal15, on their retained earnings.
Besides the control variables described above an additional interaction variable between a firm’s variance and a firm’s beta was incorporated in the regression model. Through this interaction variable, the credit quality hypothesis will be tested on idiosyncratic risk as well as non-idiosyncratic risk.
The table displayed below provides brief descriptive statistic of the sample variables. All relevant variables have been corrected for outliers through winsorization at the 1% level.
Table 2 provides the number of observations Obs, the Mean, the standard deviation Std. Dev. As well as the minimum and maximum values of the variables incorporated in the model.
To provide extra insight, property sector dummies were address to the finance applications. A distinction was made between: Unclassified, Diversified, Health Care, Industrial & Office, Lodging & Resorts, Mortgage, Residential, Retail and Self-Storage.
15 Ott et all (2005) found that REIT’s retained earnings accounted for little over 7% of their new investments.
Obs Mean Std. Dev. Min Max
Variance 102 11.2246 24.6934 0.1100 142.7400 Size 101 4.7771 1.1045 3.0246 6.5916 Employees 87 3.1805 17.2988 0 115 Ln (SoS) 94 4.4373 1.4851 1.4100 6.1469 RE 102 -48.9601 64.0638 -185.9660 2.3140 Beta * Variance 101 1.5226 2.0551 0 11.0224 Table 2
25 The stack chart displayed below provide a summary of finance application per property sector.
Graph 3 provides an insight in the finance applications from the Real Estate sample. Healthcare, Industrial/office and retail generate the majority of the applications. This appears to be in line with the findings from graph 1.
Similar to the US sample, each application was matched with a “year dummy”. The sector dummies and year dummies will be used in the regression process to control for possible sector as well as time fixed effects.
0 5 10 15 20 25
Graph 3 - Real Estate sample
Finance applications per sector
26
4 Methodology
As is described in the previous sections of this paper, the goal of this research is to compare finance applications done by REITs (as a proxy for companies active in the Real Estate sector) with finance applications done by companies from a cross-section of the total US market (excluding any companies active within the finance of energy sector). This comparison will form the foundation of the research whether firms active in the Real Estate sector follow the, previously mentioned, size and credit hypotheses in a similar way as non-real-estate companies. By examining this, a better insight will be gained regarding the capital structure of real estate companies.
The following section describes the methodology used in order to make this comparison. The first subparagraph describes the regression approach, the second subparagraph describes he US market model and the Real Estate model. The third subparagraph concludes.
4.1 regression approach
Due to the binary nature of the dependent variable (private placement dummy and SEC Rule 144a dummy), the regression should be interpreted as modelling the probability that the dependent variable equals 1. In order to solve this OLS shortcoming, the PROBIT and LOGIT model have been designed. PROBIT and LOGIT regression are nonlinear regression models. Since regression with a binary dependent variable Y models the probability that Y=1, these regression models use cumulative probability distributions. During this research, the LOGIT regression is used.
One thing that has be taken into consideration is the “measures of fit” of the PROBIT and LOGIT models. Due to their binary nature the becomes as poor measure of fit. This can be solved through the use of the so called “pseudo- ”. This measurement of fit uses the likelihood function to determine the fit of the model. The pseudo- can be interpreted in similar ways as the or adjusted- . Appendix 2 provides further elaboration regarding the interpretation of the LOGIT model.
27
4.2 Models used
During this thesis the robust version of the standard LOGIT regression will be used. In order to compare private debt with private non-bank debt both the Private Placement dummy as well as the R144a dummy will be regressed as independent variables. In order correct for possible sector fixed effects or/and time fixed effect, sector and time dummies will be incorporated in the regression.
4.2.1 The US market model
The US market model with the Private Placement/R144A dummy as independent variable including sector and time fixed effects is as follows:
Where is the dummy variable for private placement/R144A, is the intercept, reflects
the variable “Altman Z-score”, reflects the variable “Size” , reflects the variable “R&D”, reflects the variable “Employees”, reflects the variable “Ln( Assets Liability ratio). For simplicity, the seven sector dummies have been summarized by . The sixteen year dummies have been summarized by . The error term of the model is represented by .
Each X variable is linked with a which is the LOGIT coefficient of those specific variables.
Although the academic research discussed in the previous section, which was used as a foundation for this model, did not specifically address the possibility nor occurrence of endogeneity, the possible presence of endogeneity should nevertheless still be discussed.
Endogeneity occurs when a variable is correlated with the error term. This can be the result of, among others, a causality loop (Stock and Watson, 2012). The question whether the independent variables are causing the shifts in the dependent variable or rather vice a versa has been, similar to the studied literature, put aside during this research and might be an interesting topic for further research.
28
4.2.2 The Real Estate model
The Real Estate model with the Private Placement/R144A dummy as independent variable including sector and time fixed effects is as follows:
Where is the dummy variable for private placement/R144A, is the intercept, reflects
the variable “Variance”, reflects the variable “Size” , reflects the variable “Employees”, reflects the variable “Ln (Sale of Stock)”, reflects the variable “Retained Earnings” and reflects the interaction variable “Beta * Variance. For simplicity, the nine sector dummies have been summarized by . The sixteen year dummies have been summarized by . The error term of the model is represented by . Each X variable is linked with a which is the
LOGIT coefficient of those specific variables.
4.3 Some concluding remarks
The previous subparagraphs described the specific models used. During the entire regression process, robust estimators have been used through pre-determined functions in STATA (the software which was used for this thesis).
In order to correct for any sector or time fixed effects, multiple sector and time dummies have been incorporated in the models. To check for relevance of these dummies, regressions have been performed with and without these dummies. The results will be discussed in the following paragraph.
As described above, an interaction variable between REIT beta and REIT stock price variance was included in the Real Estate model. This interaction variable will, by definition, be highly correlated with the variable “variance”. This high correlation, or “multicollinearity”, will result in higher standard errors of both variables which can affect the significance levels of these variables. However, as will be seen in the following paragraph both variables are still significant despite the high correlation. Therefore the incorporation of the interaction variable does not pose any problems to the model.
29 Furthermore, each model contains three regressions with different dummy variables included. Regression one until three uses the Private Placement dummy (PP) as dependent variable. Regression four until six uses the SEC Rule 144A dummy (R144A), which is used as a proxy for non-bank private debt, as dependent variable.
5 Results
The following section will describe and interpret the results of the individual regressions and compare the Real Estate regression output with the regression output of the cross-section of the US market. Besides the regression output analysis, this section will also cover the power and predictability of the models used. During the interpretation of the results lack of proven causality has to be taken in to account.
5.1 Cross-section of the US market.
Both size and credit quality hypotheses have been tested with a LOGIT model. Before the regression outputs can be analyzed the power of the model has to be addressed. Table 3 contains the correlation matrix of the US market model.
Table 3 provides the correlation coefficients of the variables incorporated in the US market model. Where the Altman Z-score is a ratio analyses based on a company’s financial statements (see appendix 1). Size is the natural logarithm of sales. R&D are the research and development expenditures of a company. Employees is the number full time employees active in the relevant financial year. Ln(Asset Liability ratio) is the natural logarithm of asset-liability ratio.
Table 3 shows that the model does not suffer from perfect multicollinearity (a perfect linear combination of one regressor of the other regressors). Special care has been taken with the sector dummy variables and time dummy variables (which are incorporate further in this
Altman
Z-score Size R&D Emp Ln( TL)
Altman Z-score 1.0000 - - -
-Size -0.5630 1.0000 - -
-R&D -0.1613 0.3695 1.0000 -
-Employees -0.1588 0.4291 0.3637 1.0000
-Ln( Asset Liability ratio) -0.1450 0.3336 0.0808 0.1597 1.0000
Table 3
30 research) to not become a victim of the so called “dummy variable trap” which is one of the main causes of perfect multicollinearity.
The Altman Z-score is negatively correlated with all the other variables. From these outputs it can be argued that larger firms with higher R&D expenses and a larger employee base are associated with a higher risk profile. However to prove this argument, causality needs to be proven. Since this is unfortunately not the case, the correlation can only be proven and not a causal relationship.
Size is positively related to R&D, Employees and Ln(Asset Liability ratio) which is also in line with expectations that larger firms will have more expenses and a larger employee base. R&D and Employees follow the same pattern as size. Similar to the Altman Z-score, this finding also regrettably lacks a causal foundation.
5.1.1 US market regression output
The following section describes the regression output from the US market model. Due to the use of LOGIT regression models, the interpretation of these results is a little bit tricky Appendix 2 provides a more detailed explanation of LOGIT regression interpretation. The final subparagraph of this paragraph discusses the interpretation of LOGIT coefficients obtained during the regression process.
Table 4 provides the regression output of the US market sample. As is explained above, LOGIT models have been used. The Altman Z-score shows a positive coefficient in all the regressions with the Private Placement dummy as dependent variable (i.e. regression one until three). When the models does not correct for neither sector fixed effects nor time fixed effect, the Altman Z-score has a LOGIT coefficient of 0.0914 which is significant at the 1% level. When the model corrects for sector fixed effects, the LOGIT coefficient increases slightly to 0.0918 and remains significant at the 1% level. However, when the model correct for both sector fixed effects as well as time fixed effects, the LOGIT coefficient of the Altman Z-score increases significantly to 0.1621 at a significance level of 1%. This large shift after the correction for fixed effects again indicates the presence of these effects.
The Altman Z-score shows a negative LOGIT coefficient for all the regressions with the R144A dummy as dependent variable (i.e. regression four until six). Similar to the regressions with the Private Placement variable, the LOGIT coefficient increase (negatively) in magnitude after correcting for sector and time fixed effects. When the model does not correct for any fixed effects the LOGIT coefficient of the Altman Z-score amounts to -0.2086,
31 significant at the 1% level. When the model corrects for sector fixed effects, the LOGIT coefficient negatively increases to -0.2156 and remains significant at the 1% level. After correcting for both sector and time fixed effects the LOGIT coefficient increases (negatively) to -0.2875 and remains significant at the 1% level.
The finding regarding the Altman Z-score support the credit quality hypotheses that more riskier firms, which realize the lowest project quality and highest bankruptcy probability, will be the most suitable candidates for private non-bank financing. The interpretation of the Altman Z-score LOGIT coefficients will be discussed in more detail further on in this section.
Table 4 provides the regression output of the US market sample LOGIT regressions are being displayed. The table contains six regressions: three LOGIT regressions with the Private Placement dummy (PP) as dependent and three LOGIT regressions with the non-bank finance dummy (R144A) as dependent. P-values are in parentheses and significance is stated with stars: *** significant at the 1% level, ** significant at the 5% level and * significant at the 10% level. All regression are robustly estimated with the use of STATA.
1 2 3 4 5 6 Independent Intercept 3.2707 2.9104 2.9719 -0.7109 -2.0508 -1.6666 (0.000***) (0.000***) (0.000***) (0.148) (0.016**) (0.094*) Altman Z-score 0.0914 0.0918 0.1621 -0.2086 -0.2156 -0.2875 (0.001***) (0.001***) (0.000***) (0.002***) (0.002***) (0.000***) Size -0.4011 -0.3244 -0.2992 0.1426 0.3248 0.2882 (0.000***) (0.000***) (0.000***) (0.021**) (0.000***) (0.000***) R&D 0.0026 0.0019 0.0020 0.0028 0.0011 0.0011 (0.000**) (0.001***) (0.000***) (0.000***) (0.066*) (0.059*) Employees 0.0055 0.0067 0.0055 -0.0112 -0.0061 -0.0057 (0.175) (0.159) (0.181) (0.015**) (0.115) (0.163) Ln( Asset Liability ratio) 1.3067 1.3715 1.6259 0.6655 0.8195 0.8227 (0.000***) (0.000***) (0.000***) (0.000***) (0.000***) (0.000***)
Observations 1947 1947 1896 1947 1947 1896
Pseudo R-square 0.1631 0.1793 0.1966 0.2033 0.2519 0.261
Sector fixed effects √ √ √ √
Time fixed effects √ √
Table 4
Regression output - US market sample
LOGIT LOGIT
