1 | P a g e
University of Amsterdam MSc Business Economics
Dual track: Finance & Real Estate Finance
Can a disconnect be observed
between investment decisions and
financing decisions during
Quantitative Easing?
Author: B. Vujanovic
Student number: 10189815
Thesis supervisor: dr. Erasmo Giambona
2 | P a g e
Abstract
Since 2008, the Federal Funds rate has been at its lower bound, effectively leaving the Federal Reserve without a policy instrument to stimulate the economy. In order to stimulate the economy, as the economic outlook deteriorated, the Federal Reserve purchased
substantial quantities of treasuries and mortgage-backed securities with medium- and long-term maturities. From a corporate finance point of view, the benefits and costs of
Quantitative Easing for US non-financial firms are still unclear. To fill the gap in the literature this paper does an analysis on leverage ratios, cash holdings and capital expenditures to test the hypothesis of a disconnect between investments decisions and financing decisions, for non-financial firms. By using balance sheet data this paper finds that the leverage ratios of non-financial firms in the US were higher during Quantitative Easing. Further analysis shows that during the same period cash holdings increased and capital expenditures where not influenced through Quantitative Easing. After several robustness checks on alternative control variables, outliers and Driscoll-Kraay standard errors the results still hold. Therefore, a disconnect between firms’ investments decisions and financing decisions is shown. This hence confirms the concerns about the effectiveness of the Quantitative Easing program for non-financial firms in the US.
NON-PLAGIARISM STATEMENT
By submitting this thesis, the author declares to have written this thesis completely by himself/herself, and not to have used sources or resources other than the ones mentioned. All sources used, quotes and citations that were literally taken from publications, or that were in close accordance with the meaning of those publications, are indicated as such.
COPYRIGHT STATEMENT
The author has copyright of this thesis, but also acknowledges the intellectual copyright of contributions made by the thesis supervisor, which may include important research ideas and data. Author and thesis supervisor will have made clear agreements about issues such as confidentiality.
Electronic versions of the thesis are in principle available for inclusion in any EUR thesis database and repository, such as the Master Thesis Repository of the Erasmus University Rotterdam
3 | P a g e
Content
1.0 Introduction ... 5 2.0 Literature review ... 8 2.1 Why QE? ... 8 2.2.0 QE transmission channels ... 8 2.2.1 Signalling/Confidence channels ... 8 2.2.2 Money channel ... 9 2.2.3 Liquidity channel ... 9 2.2.4 Rebalancing channel ... 102.2.5 Two phases of transmission ... 11
2.3 Risks of QE – The disconnect ... 12
3.0 Methodology ... 14
3.1 Hypothesis 1: Cross-sectional variation between leverage ratio and QE ... 16
3.2 Hypothesis 2: Cross-sectional variation between QE, investment and cash holding .... 17
4.0 Data and descriptive statistics... 18
4.1 Frequency and time period ... 18
4.2 Summary statistics ... 18
4.3 Cross-correlation ... 20
4.4 Summary Statistics Robustness Check ... 21
5.0 Results ... 23
5.1 Results Hypothesis 1 ... 23
5.2 Results Hypothesis 2 ... 26
5.3 Reasoning behind the disconnection ... 29
6.0 Robustness checks ... 31
6.1 Tobin’s Q and US 10-year yield ... 31
6.2Winsorized data ... 32
6.3 Driscoll-Kraay regressions ... 33
6.4 Overview of the Robustness Checks ... 33
8.1 Appendix: Yield curve, Fed balance sheet and term premiums ... 37
8.2 Appendix: QE transmission channel ... 38
4 | P a g e
8.4 Appendix: LnMB and lnA ... 40
8.5 Appendix: Cross correlation matrix ... 41
8.6 Appendix VIF test for collinearity with Purchase variable ... 42
8.7 Appendix VIF test for collinearity without Purchase variable ... 43
8.8.0 Robustness check- Tobin’s Q ... 44
8.8.1 Robustness check-US10Y ... 45
8.8.2 Robustness check-Winsorized ... 46
8.8.3 Robustness Check-Driscroll-Kraay ... 47
5 | P a g e
1.0 Introduction
In response to the financial crisis in 2007, the Federal Reserve (Fed) implemented the Quantitative Easing program (QE). According to the Fed, this is the only way, in a zero interest rate environment, to stimulate the economy in order to reach the dual mandate of the Fed, i.e. a low unemployment rate and a stable inflation target. The policy implications of the QE program on the economy are still unclear.
Lo Duca et al. (2015) show that there was a portfolio-rebalancing effect into corporate bonds during QE, where non-financial corporations in the US issued corporate bonds to provide liquidity, which would decrease the cost of borrowing for long-term debt and hence should increase investments (Joyce et al., 2011). However, Lo Duca et al. (2015) argue, that
investment sentiment was not the main reason for the bond issuance. This has led Stein (2012) and Lo Duca et al. (2015) to suggest that there might be a disconnect between firms’ investment decisions and firms’ financing decisions during QE. Stein (2012) argues that large leverage ratios and cash holdings during QE suggest a disconnect between financing and investment decisions. This paper measures the cross-sectional variance of firms’ leverage ratios, investment ratio and cash holdings ratio during QE to show the disconnect revealed by Stein (2012) and Lo Duca et al. (2015). In this paper the disconnect is defined as follows: the QE-related factor translates into a positive relation with debt-to-asset ratios, a positive relation with the cash holdings to assets ratio and an insignificant relation with the capital expenditures to assets ratio.
The cross-sectional variance analysis of non-financial firms in the US will show how the non-financial firms, in general, reacted to the QE program. This is relevant, since the true effects of the QE policy on the non-financial part of the economy are still unclear, from a corporate finance perspective. Furthermore, the non-financial part of the economy is
relevant for the Fed in achieving one part of the dual mandate, i.e. low unemployment, since the non-financial part counts for a significant part of the job-market. Finally, this paper might also have implications for the models that the Fed used to examine the policy effects of QE on the economy.
6 | P a g e
advance of applying the QE program, the Fed analysed the implications of the QE program on the economy, using the macro FRB/US model. This model does not differentiate between whether the long-term borrowing costs decrease due to changes in the expected future path of short-term rates, or due to a change in the term premium, as these would have similar effects on investments in this model. However, according to Stein (2012), this is a false assumption during QE. A corporation will only engage in new investment, when the internal rate of return (IRR) exceeds the hurdle rate. Stein (2012) argues that the hurdle rate is completely dependent on the expected path of short-term rates, and won’t be influenced by changes in the term premium. He explains:
“A risk-neutral firm faces a rate on its 10-year bonds of 2 percent. At the same time, it expects that the sequence of rolled-over
short-term rates over the next 10 years will average 3 percent. Hence, there is a term premium of minus 1 percent. What should the firm do? Clearly, it should take advantage of the cheap long-term debt by issuing bonds. But it is less obvious that the bargain 2 percent rate on these bonds should exert any influence on its capital spending plans. After all, it can take the proceeds of the bond issue and use these to pay down short-term debt, repurchase stock, or buy short-term securities. These capital-structure adjustments all yield an effective return of 3 percent. As a result, the hurdle rate for new investment should remain pinned at 3 percent. In other words, the negative term premium matters a lot for financing behaviour, but in this stylized world, investment spending is decoupled from the term premium and is determined instead by the expected future path of short rates.”
Speech at the Brookings Institution, Washington, D.C. (Stein, 2012)
As follows from Stein (2012), the QE policy effectively decreased the term premium, while leaving short-term rates unaffected and hence left the hurdle rate, and the investment rate, unaffected. This implies that conditions for issuing long-term debt improved, as showed by Lo Duca et al. (2015), who found an increase in corporate bond issuance during QE. The finding of Lo Duca et al. (2015), and Stein’s theory (2012), form the foundation of the theory of a disconnect between investment and financing decisions.
The empirical analysis of this paper is divided into two, which both contribute to answering the research question: can a disconnect be observed between investment decisions and financing decisions during QE?
Using balance sheet data from Compustat (Capital IQ, North America) database from 1975 - 2015, the first part of the empirical analysis of this paper attempts to extend the finding of Lo Duca et al. (2015) of increased debt issuance during QE. The key difference
7 | P a g e
between this paper, and the paper of Lo Duca et al. (2015), is that this paper tries to find whether corporations increased their leverage ratio during QE. By looking at the cross-sectional variation in the leverage ratio of firms, and not the total bonds to GDP issuance, as Lo Duca et al. (2015) did, this paper is able to incorporate specific firm characteristics of non-financial firms into the analysis. Since this analysis looks at the total effect, it will also
address whether there is any substitution effect from bank lending to corporate bond lending, this might be the case after the financial crisis as banks were reluctant to lend money due to uncertainty.
The second part of the empirical analysis of this paper attempts to show the
disconnect between firms’ investment decisions and firms’ financing decisions, by looking at the cross-sectional variation of firms in the panel data during QE. Only when, from the empirical analyses, it follows that QE-related factors translated into a positive relation with debt to asset ratios, a positive relation with cash holdings to assets ratio and an insignificant relation with the investments, will the disconnect will be shown.
This paper is organised by first reviewing what the current literature has to say about the benefits and risk of QE. Then it discusses the methodology of its empirical analysis. Next, the data that is used in the empirical analysis will be discussed. Finally, showing the results of the empirical analysis with several robustness checks, and reviewing if there is a disconnect between investment and financing decisions.
8 | P a g e
2.0 Literature review
First, this literature review addresses why the Fed was forced to use QE as a policy measure to stimulate the economy. Then, it discusses the theoretical framework of the QE
transmission channels, as addressed by Joyce (2011), through which the economy and the financial markets might be affected. This is followed by addressing the implications of the QE program, and how it has segmented the bond market creating a situation where Stein (2012) and Lo Duca (2015) find signs of a potential disconnect between financing and investment decisions.
2.1 Why QE?
Following the financial crisis, in 2007, the Fed responded by decreasing the effective Federal Funds Rate (FFR), to stimulate the economy. The FFR, which determines short-term interest rates, was almost zero at the end of 2008, which is the lower bound. At this point, the Fed had no policy mechanisms left to effectively decrease the expected path of short-term rates, in an attempt to stimulate the economy. Therefore, Ben Bernanke, the Fed chairman at the time, was forced to use an unconventional monetary policy, i.e. “QE”. This policy involved the Fed buying treasuries and MBS with medium- to long-term maturity, in order to effectively lower the long-term yields to increase investments and hence to stimulate the economy.
2.2.0 QE transmission channels
This section looks at the explanation of Joyce et al. (2011), to get a better understanding of the theoretical framework of how QE affects the broader economy, and financial conditions. They elaborate five means by which QE affects the economy. These five channels are: policy signalling, confidence, money, market liquidity and portfolio rebalancing1. They also
elaborate the two phases in which these channels affect the economy. 2.2.1 Signalling/Confidence channels
The signalling channel addresses how QE influences the economy, by information provided to market participants about the likely path of future QE monetary policy, which is also referred to as “forward guidance” by the Fed. Depending on the maturities of the assets that the Fed buys, once an announcement is made, the effects will immediately be seen in the
9 | P a g e
interest rates across the yield curve (Krishnamurthy and Vissing-Jorgensen 2011). The signalling channel has a direct effect on asset prices, due to the fact that speculators will try to make a profit after announcements by the Fed. The effectiveness of this channel is dependent on the confidence that market participants have in the central bank. If confidence is high enough an announcement might not only result in a change of asset prices, but might also directly influence consumer spending or the willingness to take on debt. This, again, would result in even larger changes in asset prices and risk premiums (Joyce, 2011).
2.2.2 Money channel
The money channel addresses how QE influences the banking sector through newly-gained reserves at the central bank and increased customer deposits. These new reserves are the result of a direct purchase of assets from the banking sector. The increased customer
deposits are due to indirect asset purchases from non-banks, that deposit their newly-gained funding at one of the banks. Joyce et al. (2011) argue that this newly-gained funding might increase bank lending. However, both Joyce (2011) and Lo Duca (2015) argue that, due to the financial crisis of 2008 banks are reluctant to grant loans. Due to new regulations for banks, it can be difficult to capture the effects QE had through the money channel.
2.2.3 Liquidity channel
The liquidity channel addresses how QE influences the economy through increases in the liquidity of the assets bought by the Fed. In a way, QE creates a scarcity effect, where the supply of a particular asset in the private sector is diminished. Whilst creating enhanced liquidity, the Fed also withdraws bad debt from the private market, which improves market functioning, e.g. MBS. Joyce (2011) asserts that the increased market functioning will lower the premium for illiquidity, and hence increase asset prices. Joyce (2011) further argues that the effects of QE through the liquidity channel may only persist while central banks are conducting asset purchases. The effectiveness of this channel is hard to capture, in particular for the MBS market. The reason for this is that simultaneous policies affected the market during QE. For example, the guarantees given on Agency MBS2 by government-sponsored enterprises (GSE), have caused Agency MBS to have no credit risk. These simultaneous
10 | P a g e
effects can make it difficult to capture the effects of QE, through the liquidity channel (Hancock and Passmore, 2015).
2.2.4 Rebalancing channel
The rebalancing channel addresses how QE influences the economy, by investors searching for assets with similar risk exposure to those bought by the fed, which induces a downward pressure on their yield. Whenever the central bank purchases large quantities of particular assets from the market, they effectively decrease the relative supply of these assets. The base money issued, and the financial assets purchased are not perfectly substitutable. In order to regain the same risk exposure, the firms who sold those particular assets may attempt to rebalance their portfolios by purchasing assets that have a similar risk exposure. Through this rebalancing channel QE pushes up the prices of these similar assets and hence induces a downward pressure on their yields.
Lo Duca et al. (2015) prove that during QE, corporations in the advanced economies, and emerging markets, issued bonds to investors as substitutes for particular assets bought by the Fed. This finding is proof of the rebalancing phenomenon, which is similar to the “Gap-filling” theory provided by Greenwood et al. (2010). This theory explains how corporate bonds could replace the assets removed from the market, e.g. by QE. When assets with a long-term maturity are bought, as under the QE program, the corporate bond sector would act as a liquidity provider enhancing the long-term bond supply3, which is then absorbed by the market. This is the foundation of the “Gap-Filling theory” of corporate debt maturities.
Krishnamurthy and Vissing-Jorgensen (2011) find that when QE was announced, triple-A rated bonds fell 77 basis points in a two-day period. They explain this drop by a possible reduction in default risk, due to an increase in bond liquidity during the peak of the financial crisis. The purchasing of treasuries and MBS significantly increased the liquidity of corporate bonds through the rebalancing channel and this in turn significantly decreased the default risk of these bonds. This is supported by Joyce et al. (2011) who found that, after the large asset-purchasing program of the Bank of England, institutional investors increased their holdings of corporate bonds, and this lowered the yields on corporate bonds.
11 | P a g e
2.2.5 Two phases of transmission
Joyce (2011) et al. show that this QE transmission channel can be divided into two stages, the impact phase and the adjustment phase. Similar stages can be found in the papers of D’Amico et al. (2013) and Lo Duca et al. (2015), where they differentiate between these phases by distinguishing between an asset-purchasing period, i.e. the impact phase, and an asset-holding period, i.e. the adjustment phase.
Joyce (2011) argues that the impact phase creates an imbalance in the portfolios held by the private sector, increased the broad money in the economy and decreases the supply of those particular assets bought by the central bank. The broad money and the assets bought by the central bank are imperfect substitutes, which forces investors to rebalance their portfolio’s to regain the same risk exposure. As they try to rebalance their portfolios, these firms effectively bid up the prices of similar assets, and increase the liquidity of those assets. In this phase, the QE program affects the economy the most through the
Signalling/confidence channel, liquidity channel and rebalancing channel.
Further, Joyce (2011) argues that, at the start of the adjustment phase, firms are still rebalancing their portfolios and hence there is no equilibrium on the money and asset market. As firms are searching for long-term assets, similar to those bought by the central bank, in order to rebalance their portfolios, they effectively increase the demand for these assets and as a result the economy starts supplying. The increased demand pushes up the prices of those similar assets until the equilibrium on the money and asset markets are restored again. In theory this should decrease the borrowing costs and should increase the total wealth due to increased asset prices. In this phase the QE program affects the economy mainly through the rebalancing channel.
Francisco Ruano (2014) provides evidence for the increase in total wealth, due to increased asset prices, by examining the effects QE had on the North American stock market, using a Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model.
Joyce et al. (2011) find that long-term borrowing costs decreased in the UK, as a consequence of the QE program. Gagnon et al. (2011) found that QE has decreased long-term interest rates on a range of securities in the US, including treasuries and short-long-term repurchase agreements. Gagnon et al. (2011) also argue, just like Stein (2012), that QE
12 | P a g e
caused lower risk premiums and term premiums, but did not affect the expected path of future short-term interest rates.
2.3 Risks of QE – The disconnect
A yield curve is a line that shows the yield, at a point in time, of the same particular bond, but with differing maturity dates. A yield curve4 in a stable market tends to be upward sloping, since investors demand a term premium for potential future changes in the interest rate. Gagnon et al. (2011) and Stein (2012) argued that QE lowered the risk premiums and the term premiums of long-term assets. This, in turn, effectively decreased the long-term yield, as Gagnon et al. (2011) shows.
In theory, when rolling over short-term maturity treasuries, they should be a perfect substitute for long-term maturity treasuries. When there is a mismatch between short- and long-term maturity treasury yields, arbitrageurs attempt to make a profit since they should be perfect substitutes. The QE program lowered the risk premiums, and the term premiums, of only the term assets, which made short-term assets an imperfect substitute for long-term assets.
Through an event study, Michael Cahill et al. (2013), Gagnon et al. (2011), Neely (2015) and Krishnamurthy and Vissing-Jorgenson (2012) find evidence that these have become imperfect substitutes, by showing that yields on long-term treasuries fell further than short-term treasuries. This suggests that due to the QE program there is a segmenting effect between short- and long-term maturity assets5. In the case of the US, this
segmentation is a result of the Fed buying and eventually holding over 4.5 trillion dollars6 of MBS and treasuries, which is roughly 20 percent of the total MBS and treasury market in 2015. Joyce and Tong (2012), McLaren and Daros (2012), find similar results for Gilts and hence confirm the same segmentation due to the asset purchasing program conducted by the Bank of England. This segmentation effect can be explained by an increase of the total risk tolerance of investors towards long-term maturity assets. Adrian et al. (2013) show that the term premium for 1 year and 10 year treasuries is decreasing, and is even negative, once
4 Appendix 8.0 – Figure 1 – Term structure graphical representation of yields for a particular asset with
different maturities.
5 Appendix 8.0 - Figure 1 Yield curve 11/03/2008 shows a nod between 20 and 30-year maturity assets i.e.
negative term premium.
13 | P a g e
the Fed bought enough assets, as can be seen in figure two7. This finding supports Stein’s (2012) view that a decrease in long-term yields would not necessarily decrease the expected path of short-term yields. The decreasing, and even negative, term premium has caused a sudden increase in bond issuance, as this created favourable condition to issue long-term corporate bonds (Lo Duca et al., 2015). Furthermore, Lo Duca et al. (2015) state that, under normal market conditions, the financing decisions of corporations should be based on their investment opportunities. However, Lo Duca et al. (2015) also mention that the investment opportunities are not necessarily the main driver of borrowing decisions during QE, as normal market conditions did not pertain. This could indicate a disconnect between firms’ investment decisions and firms’ financing decisions, in the case that investments do not increase.
The next section explains the methodology of the empirical analysis that looks at whether a QE-related factor translates into a positive relation with debt to asset ratios, a positive relation with cash holdings and an insignificant relation with investments, as this would implicate that there is a disconnect between firms’ investment decisions and firms’ financing decisions. If this is shown, this would indicate that the increasing leverage ratios are not due to the investment sentiment. This would further imply that QE is not an effective measure to support the dual mandate, and the assumptions of the macro FRB/US model are not correct for non-financial firms.
14 | P a g e
3.0 Methodology
This paper evaluates the impact of QE on non-financial firms, by analysing the cross-sectional variation, using the following panel data regression setting:
𝒚𝒊,𝒕= 𝛃 ∗ 𝐐𝐄 + 𝛄 ∗ 𝐅 + 𝛄 ∗ 𝐌 + 𝛆𝒊,𝒕 (𝟑. 𝟏)
There are five dependent variables in this benchmark specification; total debt to assets, long-term debt to assets, short-term debt to assets, capital expenditures to assets and cash to asset ratio, of company i. This paper will analyse the cross-sectional variation in the three leverage ratios of non-financial firms during QE, which would indicate how non-financial firms adjusted their leverage during QE. Further, this paper will analyse the cross-sectional variation in the capital expenditures to assets ratio, as a measure of investment, of non-financial firms during QE, which would indicate how non-non-financial firms adjusted their capital spending plans, in response to the QE program. Finally, this paper will analyse the cross-sectional variation in the cash to assets ratio of non-financial firms during QE, which would indicate how non-financial firms adjusted their cash holdings in response to the QE program. It is possible to extend the findings of Lo Duca et al. (2015), by measuring the cross-sectional variation of these five dependent variables. In this way, it is possible to state how these non-financial firms reacted during QE. If it follows from the results, that the cross-sectional variation in the leverage ratios and the cash holding ratios increased, while at the same time the cross-sectional variation in the investment ratio did not change, then there is the
disconnect as suspected. Although there may be a disconnect, this does not imply that there is a causal relation between the dependent variables and the QE program.
In order to fully capture the cross-sectional variation of the five dependent variables, during QE, this paper divides the independent variables in three, following the example of Lo Duca et al. (2015). The “QE” variables are related to the Feds’ balance sheet and capture the cross-sectional variation during QE. Among the QE-related variables are the Feds’ balance sheet to total outstanding treasuries and MBS, and the QE purchases of the Fed, which is the change in the Feds’ balance sheet to total outstanding treasuries and MBS. In order to fully capture the cross-sectional variation of the five dependent variables during QE, this paper controls for firm specific characteristics in the “F” variable. Among the F variables there are control variables such as: property plant and equipment to total assets (as a measure of
15 | P a g e
collateral), dividends to total assets (as a measure of financial constraints), operating income before depreciation to total assets (as a measure of profitability), the logarithm of total assets (as a measure of firms’ size), the logarithm of the market to book ratio (as a measure of long-term performance and growth opportunities), and the cash flow to assets (a measure of self-financing)8. In order to fully capture the cross-sectional variation of the five
dependent variables during QE, this paper also controls for the different components of the yield curve in the “Y” variable. Among the Y variable are control variables such as: 10-year risk neutral yield which is an estimation of the expected path of short-term rates and 10-year term premium9.
This paper uses a dataset from 1975 - 2015. The start of the period that is analysed is before the implementation of the QE program, in order to have a reference model. The analysis is divided into two parts to answer the research question of this paper: can a disconnect be observed between investment decisions and financing decisions during QE?
The first part attempts to extend the finding of Lo Duca et al. (2015), by identifying a link between all three leverage ratios and the QE-related factors, using balance sheet data of non-financial corporations in the US. The second part of this paper attempts to show that QE-related factors had no effect on the investment ratio, while (at the same time) the cash ratio and the leverage ratios increased, as Stein (2012) suggested. This would illustrate a disconnect between firms’ financing decisions and firms’ investment decisions during QE. The next section will explain the two hypotheses of this paper. This is followed up by the empirical results, from which it is possible to define what is observed during QE and the effect QE had on the non-financial economy, and hence the effectiveness of the QE program from a corporate finance perspective.
8 These variables have been selected on basis of the paper of Korajczyk & Levy (2003).
9 These control variables have been selected on basis of the paper of Lo Duca et al. (2015), calculated by Adrian
16 | P a g e
3.1 Hypothesis 1: Cross-sectional variation between leverage ratio and QE
“QE-related factors can explain the positive cross-sectional variation in the leverage ratio of all non-financial firms in the US”
Lo Duca et al. (2015) identified a link between bond issuance to GDP ratio and the QE
program. This paper will first analyse empirically whether this finding of Lo Duca et al. (2015) is consistent on a firm level, i.e. leverage ratio of firm i and the QE program. This is an
extension to the finding of Lo Duca et al. (2015) as, in this paper, the effects through the money channel, as described by Joyce (2011), are taken into account. In this way, this paper effectively controls for the substitution effect from bank financing to bond financing. In this analysis this paper will use the three leverage ratios as the dependent variable 𝒚𝒊,𝒕.
When the regression results show a significant positive effect of QE factors on the total- and long-term leverage ratios, and an insignificant effect on the short-term leverage ratio, the link between the leverage ratio and QE is identified. This would imply that this hypothesis is true and would show how non-financial corporations adjusted their leverage ratio in response to QE. When the regression results show an insignificant effect of the QE factors on the total- and long-term leverage ratios, the link between the leverage ratio and the QE program is not identified and hence this hypothesis is false. This would imply that non-financial firms did not take on more risk in response to QE i.e. have a higher leverage ratio. However, it shows that the increase in corporate bonds issuance, which was shown by Lo Duca et al. (2015), is explained by the substitution effect from bank financing to corporate bond financing.
17 | P a g e
3.2 Hypothesis 2: Cross-sectional variation between QE, investment and cash holding
“QE-related factors can explain a positive cross-sectional variation in cash holdings but cannot explain the cross-sectional variation in investment for all non-financial firms in the US”
The second part of the empirical analysis attempts to identify the disconnect between firms’ investment decisions and firms’ financing decisions, as suggested by Lo Duca et al. (2015) and Stein (2012). In the analysis, the investment ratio and the cash ratio are used as the two dependent variables. In this part of the empirical analysis, this paper assumes that
hypothesis 1 is true. If so, the combination of these hypotheses confirm the disconnect, as suggested by Stein (2012) and Lo Duca et al. (2015).
When the regression results show that the QE-related factors have a significant positive effect on the cash ratio, and an insignificant effect on the investment ratio, then Hypothesis 2 is correct. As long as hypothesis 1 is true, the combination of the hypotheses implies a disconnect between firms’ investment decisions and firms’ financing decisions. This would implicitly confirm the suggestion of Stein (2012) and Lo Duca (2015). When the
regression results show that the QE-related variables have a significant positive effect on the investment ratio, and a significant negative effect on the cash ratio, then hypothesis 2 is false. This implies that the QE program of the Fed brought confidence to US corporations, which made them adjust their capital spending plans. This would implicitly contradict the suggestion of Stein (2012) and Lo Duca (2015).
Furthermore, in an attempt to strengthen the results, this paper also tests the hypothesis of Stein (2012). Where the term premium is expected to have a significant negative effect on the investment ratio during “normal market” conditions (1975 – 2009), and no significant effect on the investment ratio during QE (2009 – 2015). Furthermore, this paper expects the risk-neutral yield to have a significant positive effect on the investment ratio since this influences the hurdle rate.
18 | P a g e
4.0 Data and descriptive statistics
This section discusses the data used in this paper by looking at the frequency and time period, the summary statistics of the variables, the cross-correlation table and the summary statistics of the variables used in the robustness checks.
4.1 Frequency and time period
The balance sheet data used in this paper is acquired from Wharton Research Data Services capital IQ North America. The other data from St. Louis Fed, SIFMA and DataStream is merged into the data set. This collection of data has 17,779 different companies excluding the companies with an asset value below one. This paper takes the annual balance sheet data, because some variables are not reported quarterly, especially at smaller firms. In other words, quarterly balance sheet data would create a bias towards larger firms. This paper uses the annualized average of monthly data on the 10-year risk-neutral yield, the 10-year term premium and the 10-year US treasury yield. The time period of the data is from 1975 to 2015. Compared with the paper of Lo Duca et al. (2015), who use quarterly data from 2000 to 2013, this paper has a larger reference model.
4.2 Summary statistics
This section will elaborate the summary statistics of the variables used in the empirical analyses, which are shown in table 4.1 below.
As follows from table 4.1, the total leverage ratio has a mean of 33%, the long-term leverage ratio has a mean of 24%, the short-term leverage ratio has a mean of 9%, the investment ratio has a mean of 7%, and the cash ratio has a mean of 9%. The standard deviation of these five dependent variables is high. This is an indication of outliers, which can be observed in the MAX column. Furthermore, note that the minimum of the investment ratio is negative, which would imply a net gain from property, plant and/or equipment. Also note that the minimums of the short-term leverage ratio and the cash ratio are negative, which could be due to measurement errors. Figures 7 to 9 in appendix 8.3 illustrate the graphical
visualisation of the dependent variables in relation to the Feds’ balance sheet over time. As follows from table 4.1 the cash flow variable shows the most obvious skewness, which is to the left, and can be seen in the mean and the median. Furthermore, this paper uses the logarithms of asset value and market to book value. The reason being is that these
19 | P a g e
Table 4.1 Summary statistics
Variable Description Source Mean P50 SD P25 P75 Min Max N
TBL Total leverage to asset value WRDS 0.33 0.28 0.57 0.13 0.43 0.00 119.42 186147
LBL Total long-term leverage to asset value WRDS 0.24 0.19 0.48 0.05 0.34 0.00 118.56 186404 SBL Total short-term leverage to asset value WRDS 0.09 0.03 0.29 0.01 0.09 -0.04 29.30 186540 INV Capital expenditures to total asset value WRDS 0.07 0.05 0.09 0.02 0.09 -0.44 2.35 184357 Cash Cash holdings to total asset value WRDS 0.09 0.03 0.14 0.01 0.10 -0.08 1.00 172477 Fed Pur Amount of asset purchased by the Fed to total
treasuries and MBS outstanding
St. Louis Fed / SIFMA
0.02 0.02 0.02 0.00 0.02 0.00 0.05 17671 FED BS Total asset value held by the Fed to total treasuries
and MBS outstanding
St. Louis Fed / SIFMA
0.11 0.12 0.05 0.06 0.14 0.05 0.19 45146 PPE Total property plant and equipment to total asset
value
WRDS 0.14 0.08 0.19 0.00 0.19 -0.11 3.75 93458 Div Common/Ordinary dividends to asset value WRDS 0.01 0.00 0.05 0.00 0.01 -0.01 4.85 185178 Prof Operating income before depreciation to total asset
value
WRDS 0.05 0.11 0.37 0.04 0.17 -26.86 3.63 185854
LnA Logarithm of total asset value WRDS 4.86 4.67 2.28 3.08 6.49 0.69 13.59 186848
lnMB Logarithm of market value to total asset value WRDS 0.49 0.32 0.77 -0.01 0.83 -5.28 8.14 143353 CF Income before extraordinary items plus depreciation
and amortisation to total asset value
WRDS -0.02 0.07 1.07 0.01 0.11 -55.81 400.70 185818 RN10 Monthly average 10-year risk neutral yield
(annualised)
Adrian et al. (2013) 0.05 0.05 0.02 0.04 0.06 0.01 0.10 186848 TP10 Monthly average 10-year term premium (annualised) Adrian et al. (2013) 0.02 0.02 0.01 0.01 0.03 0.00 0.04 186848 Sample period 1975-2015, annual data.
20 | P a g e
variables don’t have a linear relation with the dependent variables, i.e. have diminishing effects for larger values10. From table 4.1 it follows that some independent variables have a high standard deviation. This is an indication of outliers, which can be observed in the columns Min and Max.
4.3 Cross-correlation
This section will elaborate the cross-correlation in appendix 8.5. It is necessary to address a couple of correlations between the independent variables, because of collinearity concerns. First of all, from the correlation table it follows that the Feds’ purchasing variable and the Feds’ balance sheet variable have a high correlation of 76%. This is as expected, since the purchases of the Fed directly increases its balance sheet. However, Lo Duca et al. (2015) applied these two variables in the same regression analysis, which questions the validity of their analysis and results. To check whether one of these variables might have to be omitted from the regression, due to collinearity, this paper uses the Variance Inflation Factor (VIF). This measure tests for collinearity between the independent variables in a regression. From the results11 of the VIF test it follows that all the regressions contain collinearity, once the yield variables are added. However, when the Feds’ purchasing variable is omitted from the regressions, the results12 of the VIF test show no more indication of collinearity. This implies that no more variables have to be omitted from the regression whatever their correlation. Therefore, to protect the validity of the analysis, and the results of this paper, the Feds’ purchasing variable will be omitted from the regressions.
Furthermore, appendix 8.5 shows a high correlation between the Feds’ balance sheet variable and the risk-neutral yield, the term premium and the US treasury yield. This is as expected since the aim of the QE program was to decrease the yields, which apparently affected the 10-year risk-neutral yield, 10-year term premium and 10-year US treasury yield. From appendix 8.5, the correlation between the profitability variable and the cash flow variable is 82%, which is as expected since cash-flow indirectly determines the profitability. From appendix 8.5 it also follows that the correlation between the logarithm of the market to book ratio and Tobin’s Q is 44%. This is as expected, due to the fact that these two
10 Appendix 8.4 shows the improvement of the normal distribution once the log is taken 11 Appendix 8.6
21 | P a g e
variables are similar. In this paper, Tobin’s Q is the ratio of the market value of equity and the book value of liabilities to total asset value, while the market to book ratio is the market value of the firm to the book value of the firm. The only difference is that Tobin’s Q adds the book value of liabilities in the numerator. The correlation might look low for variables that are so similar, which could be due to the use of a logarithmic function on the market-to-book ratio. Since these variables are almost similar and have a substantial amount of
correlation, these variables will not be used in the same regression. Tobin’s Q will be used in the regressions of the robustness checks, while the logarithm of the market to book ratio will be used in the regressions of the empirical research. Furthermore, appendix 8.5 shows a high correlation between the 10-year US treasury yield and both the 10-year risk-neutral yield and the 10-year term premium, of respectively 96% and 80%. This is as expected as the 10-year risk-neutral yield and the 10-year term premium are components of the 10-year US treasury yield. Therefore, the 10-year risk-neutral yield and the 10-year term premium will not be used in the same regression as the 10-year US treasury yield, which will be used in the regressions of the robustness checks. It also follows from appendix 8.5 that the 10-year risk-neutral yield and the 10-year term premium have a high correlation of 61%, which is as expected as the 10-year risk-neutral yield directly influences the 10-year term premium. Since the VIF gives results under the ten, collinearity is of no more concern. These variables tested by the VIF can be used in the same regressions in the empirical research. Based on the same reasoning, the probability variable and the cash flow variable can be used in the same regressions in the empirical research.
4.4 Summary Statistics Robustness Check
The summary statistics of the variables used in the regressions of the robustness checks are shown in table 4.2 below. These variables have been winsorized by 1%, which effectively reduces the effect of the 1% spurious outliers on both sides of the normal distribution. As follows, from table 4.2, the mean of the variables did not change. However, the maximums did change, they became more realistic after correcting for spurious outliers. Furthermore, as follows from the table, as well as section in 4.2, Tobins’ Q and the 10-year US treasury yield are used instead of, respectively, the logarithm of the market to book ratio, and the 10-year risk-neutral yield and the term premium.
22 | P a g e
Table 4.2 Summary statistics – robustness check variables
Variable Description Source Mean P50 SD P25 P75 Min Max N
TBL_win Total leverage to total asset value WRDS 0.32 0.28 0.26 0.13 0.43 0.00 2.26 186147 LBL_win Total long-term leverage to total asset value WRDS 0.23 0.19 0.22 0.05 0.34 0.00 1.46 186404 SBL_win Total short-term leverage to total asset value WRDS 0.08 0.03 0.12 0.01 0.09 0.00 0.70 186540 INV_win Capital expenditures to total asset value WRDS 0.07 0.05 0.08 0.02 0.09 0.00 0.55 184357 Cash_win Cash holdings to total asset value WRDS 0.09 0.03 0.14 0.01 0.10 0.00 0.97 172477 PPE_win Total property plant and equipment to total asset value WRDS 0.13 0.08 0.18 0.00 0.19 0.00 1.08 93458 Div_win Common/Ordinary dividends to asset value WRDS 0.01 0.00 0.02 0.00 0.01 0.00 0.35 185178 Prof_win Operating income before depreciation to total asset value WRDS 0.05 0.11 0.27 0.04 0.17 -2.98 0.48 185854 LnA_win Logarithm of total asset value WRDS 4.85 4.67 2.26 3.08 6.49 0.79 11.34 186848 lnMB_win Logarithm of market value to total asset value WRDS 0.49 0.32 0.76 -0.01 0.83 -1.34 6.06 143353 CF_win Income before extraordinary items plus depreciation and
amortisation to total asset value
WRDS -0.01 0.07 0.31 0.01 0.11 -3.61 0.45 185818 Q Tobins Q; total market value of equity and debt to total asset
value
WRDS 2.17 1.60 2.19 1.27 2.23 0.46 44.02 148982 US10Y Monthly average 10 year US treasury yield (annualised) Datastream 0.07 0.07 0.03 0.05 0.09 0.02 0.14 186848 Sample period 1975-2015, annual data.
23 | P a g e
5.0 Results
This section discusses the empirical results of the regressions, for the first and second hypothesis respectively. This will be followed by a general overview of the findings, in order to answer the research question whether a disconnect between firms’ financing decisions and firms’ investment decisions is observed during QE in the US.
5.1 Results Hypothesis 1
Table 5.1 shows the results of the regressions with as the dependent variables TBL, LBL and SBL. Columns 1, 3 and 5 only include the QE-related variable and the firm-specific
characteristics variables, while columns 2, 4 and 6 also include the yield curve-related variables.
From table 5.1, it follows that the FED BS variable has a positive significant relation with all three leverage ratios, in every column, which implies that firms took on a higher leverage ratio during QE. From column 1 it follows that when the FED BS increases by one percentage point, the TBL will increase by 0.00379. From column 2 it follows that, once controlled for the yield curve-related variables, the FED BS coefficient slightly decreases to 0.335, although the relation is still significant at the 1% level. Furthermore, it follows from column 2 that the TP10 has a negative significant relation with the TBL, which implies that the TP10 explains a significant part of the TBL, while the RN10 has an insignificant relation with the TBL.
From columns 3 and 4 it follows that the relation of the FED BS with the LBL is positive and significant at the 10% level, but when controlled for yield curve-related variables this relation is significant at the 1% level. This shows the importance of the yield curve-related control variables. Furthermore, when controlled for yield curve-related variables the table shows an increase in the FED BS coefficient from 0.113 to 0.146. From column 4 it follows that the TP10 has a negative significant relation with the LBL at the 1% level, while the RN10 has a positive significant relation with the LBL at the 10% level. The latter is unexpected, as it becomes less appealing for firms to issue long-term debt when the RN10 increases, since this decreases the price of corporate bonds and hence the amount of funds raised.
24 | P a g e
Table 5.1 Determinants of Total (TBL), long-term(LBL) and short-term(SBL) leverage.
This table looks at the determinants of the leverage ratios of non-financial firms, in the US from 1975 - 2015 (annual data). The variables used are based on the papers of Korajczyka and LevBy (2003) and Lo Duca et al. (2015). By doing this analysis this paper can argue how corporations adjusted their leverage ratios during QE by looking at the cross section of the variation caused by the QE-related factor FED BS. The regressions with even numbers 2,4 and 6 include the 10-year risk neutral yield and the 10-year term premium that represent the term structure of the yield curve at that moment. The regressions are absorbed by Gvkey and standard errors are group clustered at state and year level. The stars indicate
significance at: *** p<0.01, ** p<0.05 and * p<0.1 respectively.
Leverage ratios (1) (2) (3) (4) (5) (6) VARIABLES TBL TBL LBL LBL SBL SBL FED BS 0.379*** 0.335*** 0.113* 0.146*** 0.263*** 0.187*** (0.075) (0.060) (0.061) (0.036) (0.044) (0.045) PPE 0.109*** 0.105*** 0.070*** 0.074*** 0.039*** 0.031*** (0.014) (0.013) (0.010) (0.008) (0.011) (0.011) Div 0.101* 0.101* 0.115** 0.113** -0.014 -0.013 (0.059) (0.059) (0.054) (0.054) (0.011) (0.011) Prof -0.412*** -0.409*** -0.199*** -0.199*** -0.212*** -0.210*** (0.076) (0.076) (0.067) (0.067) (0.031) (0.031) LnA -0.015*** -0.019*** 0.006 0.004 -0.021*** -0.023*** (0.006) (0.007) (0.005) (0.006) (0.002) (0.002) lnMB 0.020** 0.020** 0.013 0.013 0.007** 0.007** (0.010) (0.010) (0.009) (0.009) (0.003) (0.003) CF -0.026 -0.026 -0.020 -0.020 -0.007 -0.007 (0.019) (0.019) (0.015) (0.015) (0.006) (0.006) RN10 -0.013 0.289* -0.299*** (0.160) (0.151) (0.058) TP10 -1.453*** -0.916*** -0.532*** (0.266) (0.248) (0.090) Constant 0.367*** 0.421*** 0.187*** 0.199*** 0.181*** 0.222*** (0.023) (0.040) (0.021) (0.038) (0.009) (0.012) Observations 137,682 137,682 137,824 137,824 137,988 137,988 Adjusted R-squared 0.287 0.288 0.220 0.221 0.281 0.281
25 | P a g e
positive and significant at the 1% level in both columns 5 and 6, although the coefficient of the FED BS decreases from 0.267 to 0.187 when controlled for the yield curve-related variables. Furthermore, it follows from column 6 that the TP10 and the RN10 have a significant negative relation with the SBL, which is in line with the theory.
From table 5.1 it follows that the PPE has a significant positive relation with all three leverage ratios. This can be explained by the fact that PPE is an indication of the amount of collateral a firm has and collateral creates favourable conditions for taking on debt.
Furthermore, it also follows that the Div is an important variable in explaining the TBL and the LBL, as these relations are significant. The positive relation is as expected since firms that issue large amounts of dividends signal confidence to the market that they will have stable cash flows, which creates favourable conditions for taking on debt. The Prof has a negative significant relation with all three leverage ratios. This relation can be explained by the fact that firms with high profitability can finance themselves and do not need external funding to take on investments. The LnA has a negative significant relation with both the TBL and the SBL, as firms with higher asset values tend to be more mature and hence do not need a high levels of capital expenditures and hence funding. The lnMB has a significant positive relation with both the TBL and the SBL, as firms with a high lnMB signals to investors that they have high potential growth and hence creates favourable conditions for debt financing. Note that the CF does not have a significant relation with any of the three leverage ratios.
As follows from this analysis the FED BS variable shows that during QE the observed leverage ratios were higher. This significant relation supports the findings of Lo Duca et al. (2015) concerning the link between bond issuance and the QE program. It also implies that the QE-related factors can explain the positive cross-sectional variation in the leverage ratios of all non-financial firms in the US and hence Hypothesis 1 is true. Furthermore, if there would have been a substitution effect the leverage ratio would be constant over time, ceteris paribus. However, the significant relation implies that there was no substitution effect from bank financing to corporate bond financing. Surprisingly the observed short-term leverage ratio also has a significant positive relation with the FED BS. This is surprising since Stein (2012) argued that firms might use the gained funding from long-term debt to buy back short-term debt, which would imply a negative relation. The strong negative relation of the
26 | P a g e
term premium with the leverage ratios gives an indication that QE might also increase the leverage ratio by effectively decreasing the term premium as Stein (2012) suggested. This relation is as expected, a decrease in the term premium would make it attractive for firms to issue long-term debt since investors demand a lower risk premium for longer maturity assets. The next section will discuss the second hypothesis.
5.2 Results Hypothesis 2
Table 5.2 shows the results of the regressions with as the dependent variables INV and Cash. Columns 1 and 3 only include the QE-related variable and the firm specific characteristics variables, while columns 2 and 4 and 6 also include the yield curve-related variables. Columns 5 and 6 contain two different sub-periods for the term premium. In other words, the TP10 has been replaced by respectively the TP7509 and the TP0915. In this way, column 5 provides the relation of INV and the term premium under “normal market” conditions, while column 6 provides the relation of INV and the term premium during QE.
From columns 1 and 2 it follows that the FED BS has a significant positive relation with the cash ratio. From column 1 it follows that when the FED BS would increase by one percentage point, the observed Cash would increase by 0.00248. From column 2 it follows that once controlled for the yield curve-related variables, the FED BS coefficient decreases to 0.099, although the relation is still significant at the 1% level. Furthermore, it follows from column 2 that the TP10 and the RN10 have a significant negative relation with the Cash. From column 3 it follows that the FED BS has a significant negative relation with the INV, which cannot be explained with economic reasoning. However, once controlled for the yield curve-related variables in column 4, this relation becomes insignificant. Again, this shows the importance of the yield curve-related control variables. From column 4 it also follows that the RN10 has a positive significant relation with the INV, while the TP10 has an insignificant relation with the INV. Furthermore, after dividing the data on the TP10 into two sub-periods in columns 5 and 6, the results still show an insignificant relation between the FED BS and the INV. From columns 5 and 6 it also follows that the RN10 still has a positive significant relation with the INV. However, the TP7509 in column 5 has a
27 | P a g e
Table 5.2 Determinants of investment (INV) and cash holdings (Cash).
This table looks at the determinants of investment and cash holdings of the non-financial firms in the US from 1975 - 2015 (annual data). The variables used are based on the papers of Korajczyka and LevBy (2003) and Lo Duca et al. (2015). By doing this analysis this paper can argue how corporations adjusted their investment ratio and cash holdings during QE by looking at the cross section of the variation caused by the QE-related factor FED BS. The regressions with even numbers 2,4 and 6 include the year risk neutral yield and the 10-year term premium that represent the term structure of the yield curve at that moment. The regressions are absorbed by Gvkey and standard errors are group clustered at state and year level. The stars indicate significance at: *** p<0.01, ** p<0.05 and * p<0.1 respectively.
(1) (2) (3) (4) (5) (6)
VARIABLES Cash Cash INV INV INV INV
FED BS 0.248*** 0.099*** -0.125*** 0.013 0.008 -0.003 (0.012) (0.013) (0.007) (0.009) (0.010) (0.010) PPE -0.019*** -0.034*** 0.008*** 0.023*** 0.023*** 0.023*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Div 0.050*** 0.053*** 0.008 0.005 0.005 0.006 (0.013) (0.013) (0.008) (0.008) (0.008) (0.008) Prof -0.002 0.000 -0.006** -0.008*** -0.008*** -0.008*** (0.002) (0.002) (0.003) (0.003) (0.003) (0.003) LnA -0.006*** -0.009*** -0.006*** -0.004*** -0.004*** -0.003*** (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) lnMB 0.027*** 0.027*** 0.017*** 0.017*** 0.017*** 0.017*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) CF 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) RN10 -0.699*** 0.671*** 0.685*** 0.682*** (0.035) (0.033) (0.034) (0.034) TP10 -0.249*** -0.022 (0.051) (0.044) TP7509 -0.001* (0.000) TP0915 0.003*** (0.001) Constant 0.100*** 0.157*** 0.096*** 0.048*** 0.049*** 0.046*** (0.003) (0.004) (0.002) (0.004) (0.004) (0.004) Observations 127,155 127,155 136,728 136,728 136,728 136,728 Adjusted R-squared 0.590 0.594 0.497 0.505 0.505 0.505
28 | P a g e
significant negative relation with the INV at the 10% level, while the TP0915 in column 6 has a significant positive relation with the INV at the 1% level.
From table 5.2 it follows that the PPE has a significant positive relation with the INV and a significant negative relation with the Cash. The positive relation between the PPE and the INV can be explained by the fact firms with a relative high PPE tend to have relative high maintenance and replacement costs, which forces them to reinvest. The negative relation between PPE and the Cash can be explained by the fact that a high PPE creates favourable conditions for debt financing and hence holding cash is not optimal. Furthermore, it also follows from table 5.2 that the Div has a significant positive relation with the Cash. Firms with a relative high dividends ratio tend to have stable and excessive profits, which they can disburse to shareholders. Therefore, the firms that do disburse dividends tend to have a higher cash ratio due to their stable and excessive profits, which explains the positive relation between the Div and the Cash. The Prof has a significant negative relation with the INV, which could be explained by the fact that high profitable mature firms do not need to expand their investments. Also note that Div has an insignificant relation with INV, while Prof has an insignificant relation with the Cash. Furthermore, the relation between LnA and Cash as well as INV is significant negative. The negative relation with INV can be explained as firms with higher asset values tend to be more mature and hence do not need a high level of capital expenditures. The negative relation with Cash can be explained by the fact that there is less variance in their cash flows, which enables them to lower their cash holdings. The lnMB has a significant positive relation with Cash and INV. The positive relation with INV can be explained by the fact that high lnMB firms have growth opportunities, which can be achieved through investing. The CF has a significant positive relation with Cash and INV.
As follows from this analysis the FED BS shows that, during QE, the observed cash ratios were higher, while the observed investment ratio during QE did not significantly change. This finding in combination with the observed higher leverage ratios implies a disconnect
between firms’ financing decisions and firms’ investment decisions and hence that hypothesis 2 is true. It is worth noticing that this is not a causal relationship, but a finding concerning this specific QE period and this specific dataset. In other words, our finding cannot be generalized to QE programs in different periods or countries. Furthermore, it also
29 | P a g e
follows from the analysis that there is a disconnect between the term premium and investments. Over the whole time period the term premium has an insignificant relation with INV. While, the RN10 has a significant positive relation with INV over the whole time period. The RN10 is a proxy of the expected path of short-term rates. Therefore, this finding is in line with the hypothesis of Stein (2012) as the investment spending is mostly influenced by the expected path of short-term rates. These findings are in line with the speech of Stein (2012) that investment spending is decoupled from the term premium and is determined instead by the expected path of short-term rates.
When the term premium is divided into two sub-periods the relation with INV changes. In the first sub-period, 1975 - 2009, the relation between the term premium (TP7509) and INV is significantly negative, which is in line with the theory under “normal” market conditions. In the second sub-period, 2009 - 2015, the relation between the term premium (TP0915) and INV is significantly positive, which cannot be explained by economic theory. The change of the relation between the term premium and INV over time can be the reason for the insignificant relation between TP10 and INV over the whole time period. These findings amplify the disconnect between firms’ financing decisions and firms’ investment decisions.
5.3 Reasoning behind the disconnection
From the previous sections, it follows that the observed leverage ratios and cash ratios were higher during QE, while FED BS had an insignificant relation with the investment ratio. This supports the disconnect as defined in the introduction. The change in the relation between the term premium and investments supports the disconnect and shows that Steins (2012) theory holds, i.e. the term premium is decoupled from investments.
Since 2009, Caballero et al. (2014) and Shin (2013) argue that emerging market borrowers have issued an increasingly large amount of US dollar-denominated bonds and deposited the proceeds in domestic banks, effectively earning an interest rate margin. Lo Duca et al. (2015) suggest that this might be due to the fact that non-financial corporations in the US decided to engage in carry trades with the proceeds from their bond issuance. This would imply that non-financial firms in the US substituted their standard business model for carry trades due to the interest rate margin earned on these carry trades. Shin (2013) supports this view by showing that the Chinese and Brazilian outstanding US denominated
30 | P a g e
international debt securities have significantly increased since the financial crisis. This is also supported by one of Lo Duca et al. (2015) analyses, which shows, with a counterfactual analysis approach that without the QE program of the Fed, the corporate bond issuance in the emerging corporate bond market would have been half of the actual total. The increase in the cash ratio and the leverage ratio might be explained by these findings. Therefore, Shin (2013) and Lo Duca et al. (2015) implicitly support the theory about the disconnect between firms’ investment decisions and firms’ financing decisions.
Another possibility is that corporations issued debt in times of favourable financing conditions and are holding funds as a buffer in response to the financial crisis. This view is supported by the “market-timing hypothesis” in Bagker and Qurgler (2002) paper, who show that the capital structure is related to past market values. This hypothesis in combination with the findings of Francisco Ruano (2014) and Joyce (2011), who showed that QE
influenced the equity market, might explain that non-financial firms simply issued debt and increased cash holdings due to favourable financing conditions. This supports the “market timing hypothesis” as a possible explanation of the increase in the leverage ratio. However, this is beyond the scope of this thesis.
In order to improve the validity of the empirical results, the next section contains some robustness checks to correct for measurement errors.
31 | P a g e
6.0 Robustness checks
In this section, there will be four alternative regressions to correct for possible measurement errors. The first two robustness checks use different comparable variables to those used in the empirical analysis. The risk neutral yield and the term premium are replaced by the US 10-year yield and the market to book ratio is replaced by Tobin’s Q. The third robustness check involves winsorized data to check whether spurious outliers affect the results from the empirical analysis. The last robustness check repeats the empirical analysis using Driscoll-Kraay standard errors.
6.1 Tobin’s Q and US 10-year yield
The first robustness check replaces the market to book ratio by Tobin’s Q. As follows from the regression results in appendix 8.8.0, Tobin’s Q has a similar relation with all three leverage ratios as the market to book ratio. However, it did improve the adjusted R-squared measure. Furthermore, the results show that Tobin’s Q has a similar relation with the cash and investment ratio as the market to book ratio. The coefficients decreased in value, while the adjusted R-squared measure slightly deteriorated. The FED BS does not show noticeable differences in the significance nor in the relation to all five dependent variables. However, the relation between the term premium and the investment ratio does show a noticeable difference in the sub-periods. The 1975 – 2009 term premium still has a negative relation with the investment ratio and the 2009 – 2015 term premium still shows a positive relation with investment ratio. However, they are no longer significantly related. This shows that the proposition of Stein (2012) does not hold when controlling for Tobin’s Q and hence the term premium is not decoupled from investment.
The second robustness check replaces the 10-year term premium and the 10-year risk neutral yield with the 10-year US treasury yield. As follows from the regression results in appendix 8.8.1, the 10-year US treasury yield has a similar relation with the total and short-term leverage ratios and have similar adjusted R-squared values. However, the relation with the long-term leverage ratio in column 4 became insignificant. The coefficients in column 1, 3 and 5 are the same as could be expected, while the coefficients in columns 2, 4 and 6 decreased. Furthermore, as can be seen in columns 8 and 10, the relation with the cash ratio and the investment ratio over the whole time period are similar, although the coefficients
32 | P a g e
decreased. In the first sub-period regression in column 11, the 10-year US treasury yield now has a significant positive relation with the investment ratio, while the second sub-period regression still shows a significant positive relation. Since the 10-year US treasury yield is composed of the 10-year term premium and the 10-year risk neutral yield, it is impossible to distinguish between those and to form a conclusion. However, the FED BS still has a similar relation with all three leverage ratios and the cash ratio. From columns 9 to 12 it follows that the Fed BS has a negative relation with the investment ratio during QE. This shows that when not controlled for the term premium and the risk neutral yield, the cross-sectional variance of the FED BS variable and the investment ratio is negative during QE. A decrease in the investment ratio during the QE might be an indication of carry trades. However, due to the scope of this thesis it is not reasonable to make this proposition. It is clear that a higher investment ratio during QE could not have been observed.
6.2 Winsorized data
The third robustness check replaces the data by winsorized data. This will control for the 1% spurious outliers on both sides of the normal distribution. As follows from the regression results in appendix 8.8.2 the regression setting from equation 3.1 has a better fit using winsorized data, as columns 1 to 6 show higher adjusted R-squared values when compared to the empirical analysis. Although, the adjust R-squared values in columns 7 to 12 did not significantly change. The relation of the 10-year risk neutral yield with all three leverage ratios is now significant at the 1% level, even though the coefficient changed. The relation of the 10-year risk neutral yield with the other two dependent variables did not change, nor did the significance level or the coefficients. Furthermore, the relation of the 10-year term premium with the dependent variables did not significantly change, nor did the significance level. The coefficients of the TP10 in all three leverage ratio settings decreased compared to the empirical analysis. As follows from columns 1 to 5 the relation of the FED BS with both the total and long-term leverage ratios did not change. The relation of the FED BS with the short-term leverage ratio is significantly negative now, as Stein (2012) suggested. He suggested that firms might find it interesting to issue long-term debt and repurchase short-term debt, which implies a negative relation between the FED BS and the short-short-term leverage ratio. The coefficients of the FED BS decreased compared with the empirical
33 | P a g e
analysis, while the FED BS also lost some significance in columns 1, 3 and 5. The relation of the FED BS with both the cash ratio and the investment ratio did not change, only column 10 now shows a significant result at the 10% level. The coefficients of the FED BS did not
significantly change compared with the empirical analysis.
6.3 Driscoll-Kraay regressions
In the analysis of Lo Duca et al. (2015) they apply Driscoll-Kraay standard errors, which implies that the error structure is assumed to be heteroskedastic. It also corrects for
autocorrelation and possible correlation between groups. However, because the estimator is based on an asymptotic model, one should be somewhat careful with applying this estimator to panel data that contains a short time period (Hoechle, 2007). Since our reference period is larger than that of Lo Duca et al. (2015), it seems appropriate to use this method for the robustness check.
The results in appendix 8.8.3 show that the total book leverage and the FED BS variable are still significant at 1% and that most of the firm characteristics variables are still significant. Further, it can be seen that the term premium is still significant at 1%. For long-term leverage ratio, it can be seen that when not controlling for the long-term premium in column 3, the FED BS variable is not significant anymore. However, when the analysis is continued in column 4, the variables are significant again just like in table 5.1 and 5.2. The analysis does not show notable differences in the short-term leverage ratios. When column 7 - 12 are analysed for the dependent variables Cash and INV, it follows that there are no real changes in the significance of the important variables can be noticed. However, it can be seen that the term premium subsample variables are not significant any more on
investments. Indicating that when using the Driscoll-Kraay standard errors the change in the relation with the term premium is not significant anymore.
6.4 Overview of the Robustness Checks
When the market to book ratio was replaced by Tobin’s Q, the relation of the FED BS with all dependent variables did not change. The relation of Tobin’s Q is similar to that of the market to book ratio. Furthermore, the relation of the 10-year risk neutral yield and the 10-year term premium with the leverage ratios and the cash ratio did not change. However, the relation between the 10-year term premium and the investment ratio in both sub-periods
