1
The impact of the ECB’s Corporate Sector
Purchase Programme on non-financial
companies in the Eurozone
(JEL: E52)
Name: Sander Tiggelaar
1Student number: 2645327
University: Rijksuniversiteit Groningen
Study programme: MSc Finance & MSc Economics
Course: Combined master thesis (EBM000A20)
Supervisor: prof. dr. K.F. Roszbach
Date: 10
thof January 2019
Abstract
This thesis investigates the impact of the Corporate Sector Purchase Programme (CSPP) of the European Central Bank (ECB) on non-financial companies within the Eurozone. It takes a new perspective by looking at the implications of the CSPP on market-valuation of companies using a difference-in-difference estimation methodology. The results show that the programme had a significant effect on market valuation. It is also found that the effect is not smaller for companies in the control group as compared to the treatment group, a finding in line with the theory of cash reinvestment after Quantitative Easing.
Keywords:
• Unconventional monetary policy
• Quantitative Easing
• Corporate Sector Purchase Programme (CSPP)
• Difference-in-difference methodology
2
1 Introduction
In September 2008, the bankruptcy of Lehman brothers marked the start of a downfall of the financial system in the USA, Europe and other regions. Many serious flaws in the financial system that had accumulated in the pre-crisis years caused a crisis of immense proportions. The most important flaws included the low interest rate and therefore cheap mortgages, extra household lending during increasing house prices, unsustainable and non-transparent financial products, weak supervision on financial products, and increased risk-taking of banks. After Lehman brothers, a series of bank bankruptcies and bank bailouts were inevitable. Well-knows cases include Northern Rock and Icesave. Both the Federal Reserve (FED) and the European Central Bank (ECB) broadened their original toolkit, partly in order to overcome the Zero Lower Bound (a phenomenon previously thought to have only theoretical relevance). In the Eurozone another crisis unfolded; several Eurozone members were at risk of bankruptcy, with Greece as the prime example. This so called ‘sovereign debt crisis’ caused the financial crisis in the Eurozone to be considerably longer-lasting than the crisis in the USA and other regions (Stiglitz; 2016). Together with much-needed reforms such as the banking union and better macroprudential banking supervision, the ECB engaged in large-scale unconventional monetary policy (De Grauwe; 2016).
3
Conventional monetary policy SMP/OMT TLTROs/FRFAs Quantitative Easing (QE)
Central Bank Financial Institutions Firms and Households Central Bank Financial Institutions Firms and Households Central Bank Financial Institutions Firms and Households Central Bank Financial Institutions Firms and Households
At a more microeconomic level, the ECB designed so called Fixed Rate Full Allotments, or FRFAs (2008) and Targeted Longer-term Refinancing Operations, or TLTROs (2014). These programmes were designed to help banks with financial difficulties by providing funds at looser than normal conditions. These programmes also served as a replacement for the dried-up interbank market. In order to overcome the Zero Lower Bound, the ECB introduced a package of programmes referred to as ‘Purchases of marketable debt instruments’ or ‘Quantitative Easing/QE’ (ECB; 2018). These programmes are designed to work beyond the Zero Lower Bound by purchasing assets directly in the market. The ECB’s QE strategy contains four pillars: (1) the Corporate Sector Purchase Programme (CSPP), (2) the Public Sector Purchase Programme (PSPP), (3) the Asset-backed Securities Purchase Programme (ABSPP), and (4) the Covered Bond Purchase Programme (CBPP). The distinctive way of how liquidity is provided in the above-mentioned monetary policies is illustrated graphically in Figure 1 below.
FIGURE 1. (Un)conventional monetary policy in the liquidity pyramid
Given that the most recent unconventional policy measures in both the USA and Europe are in place now for several years and are slowly being redeemed, research regarding their effectiveness is becoming of increased relevance. These studies investigate macroeconomic effects, microeconomic effects, cross-country comparisons, and impacts on company behaviour and capital structure. Jäger and Grigoriadis (2017) and Eser and Schwaab (2015) investigate the macroeconomic impact of unconventional monetary policies within the Eurozone by looking at the development of sovereign bond yields. Jäger and Grigoriadis (2017) apply an event study methodology whereas Eser and Schwaab (2015) estimate a Pooled OLS panel model. They both find evidence supporting the Eurozone-wide effectiveness of the SMP (Securities Markets Programme); however, they also highlight that the SMP alone was insufficient to solve the sovereign debt crisis and that programmes such as the OMT where necessary. Jäger and Grigoriadis (2017) also note that where the SMP had a differential effect between crisis and non-crisis countries, the OMT was effective in both non-crisis- and non-non-crisis countries.
4
Other authors investigate the microeconomic effects of (un)conventional monetary policy. Drechsler et al. (2018) develop a dynamic asset pricing model which incorporates monetary policy into the risk premium of companies, and Boermans and Keshkov (2018) investigate the impact of the Eurozone’s Public Sector Purchase Programme (PSPP) on the ownership concentration of ‘PSPP-eligible’ institutions.
Another area of research studies the effect of (un)conventional monetary policy on capital structure decisions and investment behaviour. Foley-Fisher et al. (2016) investigate the effect of the Maturity Extension Program (MEP) in the USA on company-level decisions. This study looks at differential effects of this unconventional monetary policy programme on companies with a different reliance on longer-term debt. They find that stock prices on the announcement date of the MEP for companies with relatively high longer-term debt rose more relative to those companies with lower levels of longer-term debt. Furthermore, the former companies expanded relatively more on employment and investment. Another important phenomenon observed in this study is so called ‘reach for yield’, where both demand and yield for riskier corporate debt increased after the announcement of the MEP. The underlying logic for this effect is that when investors learn that the yield on safer ‘eligible’ bonds has decreased, there might be a larger demand for riskier non-eligible bonds that do offer a higher yield. This demand might result in companies trying to reach a higher yield on their bonds to attract these investors, a tendency confirmed by Chaudron (2016).
5
Primary market
The focus of this thesis is on the effect of the ECB’s Corporate Sector Purchase programme (CSPP) on company value. This programme was announced on March the 10th 2016 and was initiated
starting June the 8th of the same year. According to ECB (2018), ‘‘[the CSPP] helps to further
strengthen the pass-through of the Eurosystem’s asset purchases to financing conditions of the real economy (…)’’. The CSPP was the first and single programme within the ECB toolkit specifically aimed at Quantitative Easing through the corporate sector. It was implemented in combination with three other programmes in the context of ‘Purchases of marketable debt instruments’, as explained above. Gross-Rueschkamp et al. (2017) distinguish six key aspects of the CSPP: (1) the bond issuer has to be incorporated in the Eurozone, (2) the security has to have a minimum maturity of 6 months and a maximum maturity of less than 31 years, (3) an issue has to have a minimum credit rating of BBB from at least four agencies, (4) an eligible security has to be denominated in euros and has to have a YTM larger than the ECB’s deposit facility rate, (5) securities can be purchased both in primary as well as in secondary markets, and (6) the CSPP was set until at least March 2017 (but it was later extended to December 2017). Figure 2 below describes the procedure of the CSPP using stylized balance sheets.
FIGURE 2. The procedure of the CSPP
Corporate Balance Sheet ECB Balance Sheet
v
The actual purchases as described in Figure 2 are predominantly executed by the National Central Banks (NCB) following instructions of the ECB, according to De Haan et al. (2015). In terms of isolating a unique causal effect, the CSPP has two favourable characterisics. The first is that it is the only programme within both QE and within the broader array of unconventional policies that is specifically aimed at alleviating financial constraints through buying corporate bonds. The second is that the start date of the CSPP (8 June 2016) is considerably isolated. The closest starting dates are 21 November 2014 (ABSPP) and 9 March 2015 (PSPP).
Previous research regarding the CSPP predominantly looks at the effects on capital structure. De Santes et al. (2018), Gross-Rueschkamp et al. (2017), and Arce et al. (2017) conclude that since the announcement of the CSPP, non-financial corporations in the Eurozone have started issuing more ‘CSPP-eligible’ bonds at the expense of bank loans.
6
De Santes et al. (2018) also find that eligible companies issue bonds at larger maturities, which is in line with the results found by Foley-Fisher et al. (2016) discussed above. Furthermore, the authors present some evidence that the reduction in bank loans has freed up bank funds for smaller SMEs not issuing bonds. Supplementing previous research investigating the effects on capital structure, this thesis looks at the effects on company value using a difference-in-difference methodology.
This thesis uses data from the ECB to split the sample into companies which actually sold corporate bonds during the CSPP and those which have not. It thereby creates a true treatment and control group rather than groups only based on programme-eligibility. The key alternative hypothesis in this thesis is that companies which have been part of the CSPP have experienced lower bond yields and therefore lower financing costs compared to companies which have not been part of the programme. If this is the case, spillovers and cash reinvestments are insufficient to entirely eliminate a diverging effect between companies selling corporate bonds to the central bank and those which have not done so. This result would imply that central bank purchases of corporate bonds might create market distortions and potential inefficiencies.
The idea is motivated by the fact that existing research has mainly focused on the effects of the CSPP on capital structure rather than company value. Furthermore, previous research applying a difference-in-difference methodology to study the effects of the ECB’s QE programmes have looked at differences between programme-eligible and programme-ineligible companies - see for example De Santes et al. (2018) and Boermans and Keshkov (2018) - rather than at a potential divergence or spillovers between companies which actually sold corporate bonds to the central bank compared to companies which have not. By looking at actual corporate bond sales, a more accurate treatment group can be created. A potential differential effect between treated and untreated companies raises important concerns, since the choice of which instruments are purchased has an impact on the outcome and spillovers are only limited. Grounded in these research objectives, the following research question is formulated:
• Has the ECB’s Corporate Sector Purchase Programme impacted the market valuation of non-financial companies in The Eurozone?
7
(1)
2 Methodology
2.1 Theoretical framework
In order to answer the research question presented in the previous section, it is important to choose a methodology that is able to reveal the specific effect of the CSPP. At the most basic level, a choice needs to be made between one of three basic methodological options: (1) a time-series estimation, (2) a cross-sectional model, or (3) a panel methodology. The first one looks at the development of one unit (i.e. a company) over multiple periods, whereas the second one looks at multiple units (i.e. multiple companies) over one single time period. The advantage of using a panel methodology is that the analysis is neither restricted to one time period nor to one unit. A panel incorporates both a temporal dimension as well as a cross-sectional dimension. Therefore, in a panel, an extensive list of Eurozone companies over a period starting well before the start of the CSPP until the end of implementation can be used as the sample. Differences between companies that are already present before the start of the treatment can be controlled for. This thesis uses a difference-in-difference application of the Fixed Effects panel methodology to estimate the panel. Regarding the difference-in-difference estimation method; this methodology derives from the standard Fixed Effects panel. This standard Fixed Effects panel can be described as follows:
∆𝑦𝑖𝑡 = 𝛿∆𝑟𝑖𝑡+ ∆𝜇𝑡+ 𝛼𝑖 + ∆𝜇𝑖𝑡,
where i refers to the unit and t to the period. 𝑦𝑖𝑡 is the dependent variable of interest. The term 𝜇𝑡
refers to time fixed effects and the term 𝜇𝑖𝑡 refers to individual fixed effects. The term 𝛿 refers to the effect of the treatment variable. 𝑟𝑖𝑡 is the treatment variable. This treatment is the intervention that has directly impacted only units in the so called treatment group, but not those units in the control group. Using the formulation of Verbeek (2017), the treatment variable (𝑟𝑖𝑡) can be defined
as follows:
𝑟𝑖𝑡 = 1 if unit i has been part of the treatment during period t
= 0 otherwise, in case of a binary (dummy) variable, and:
𝑟𝑖𝑡 = magnitude of the treatment if unit i has been part of the treatment during period t
8 (1A) (1B) (1C) (2) 𝑟𝑖𝑡
A key intermediate step in estimating the difference-in-difference estimator is to estimate Equation 1 for the treatment group and control group separately as follows:
∆𝑦̅𝑖𝑡𝑡𝑟𝑒𝑎𝑡𝑒𝑑 = ∆𝑦𝑡,𝑇= 𝛿∆𝑟𝑡,𝑇+ ∆𝜇𝑡+ 𝛼𝑇+ ∆𝜇𝑡,𝑇
𝑦̅𝑖𝑡𝑛𝑜𝑛𝑡𝑟𝑒𝑎𝑡𝑒𝑑 = ∆𝑦𝑡,𝐶= ∆𝜇𝑡+ 𝛼𝐶+ ∆𝜇𝑡,𝐶,
which can then be rewritten by deducting Equation 1A from Equation 1B.
When ∆𝑦̅𝑖𝑡𝑡𝑟𝑒𝑎𝑡𝑒𝑑− ∆𝑦̅𝑖𝑡𝑛𝑜𝑛𝑡𝑟𝑒𝑎𝑡𝑒𝑑 is defined as 𝛿𝑖𝑡, the following difference-in-difference equation is obtained:
𝛿𝑖𝑡 =𝛿∆𝑟𝑡,𝑇+[∆𝜇𝑡,𝑇− ∆𝜇𝑡,𝐶]+[𝛼𝑇− 𝛼𝐶]
Note that Equation 1C does not contain any time fixed effects since these are independent of the unit i and thus cancel out when deducting 1A from 1B (∆𝜇 − ∆𝜇 = 0). This is the key advantage of using a difference-in-difference model rather than a standard Fixed Effects panel, according to Woolridge (2012). Equation 1C has an important implicit assumption; any difference in difference of the dependent variable is only driven by a constant, the treatment, and by individual-level controls. This assumption only holds if indeed the treatment group and control group only differ in terms of the treatment, so that ∆𝜇 − ∆𝜇 = 0. It is however impossible to know how the dependent variable of a treated unit would have behaved if there had been no treatment and whether it behaved sufficiently in tandem with a similar nontreated unit. Fortunately, it is possible to compare the dependent variable of different treated and nontreated units before the treatment begins (and when both units are thus nontreated). This is a common way to estimate the validity of the parallel trend assumption. Hill et al. (2012) present a useful method to estimate a difference-in-difference model in a regression format, which is summarized in Equation 2 below:
∆𝑦𝑖𝑡 = 𝛽1+𝛽2(𝑟𝑡) + 𝛽3(𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡) + 𝑒𝑖𝑡,
where the term 𝑇𝑅𝐸𝐴𝑇𝑖 is equal to 1 for units in the treatment group and the term 𝑟𝑡 contains the treatment. The coefficient for the third term (𝛽3) is essentially the estimated value for 𝛿𝑖𝑡 of
equation 1C. The term 𝑒𝑖𝑡 represents the error term.
9
(3) Note that the interaction between the dummy indicating whether a unit is in the treatment group and the treatment (𝑇𝑅𝐸𝐴𝑇𝑖 × 𝑟𝑡) is equal to the treatment variable 𝑟𝑖𝑡 of Verbeek (2017) mentioned above. Intuitively, Verbeek (2017) incorporates the variable 𝑇𝑅𝐸𝐴𝑇𝑖 by adding a
unit-specific indicator i to the treatment. Equation 2 can be estimated using Fixed Effects. Hill et al. (2012) also present an alternative equation which reduces the temporal dimension to two periods: (1) before treatment, and (2) after treatment. The period (before = 0/after = 1) can be added as an explanatory variable, and the regression can be estimated using OLS. This two-period difference-in-difference model is summarized in Equation 3 below:
𝑦𝑖𝑡
̅̅̅̅ = 𝛽1+ 𝛽2𝑇𝑅𝐸𝐴𝑇𝑖+ 𝛽3𝐴𝐹𝑇𝐸𝑅𝑡+ 𝛽4(𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡) + 𝑒𝑖𝑡,
where 𝑦̅̅̅̅ is the average value of the dependent variable for the relevant period (‘before’ or 𝑖𝑡
‘after’). 𝑇𝑅𝐸𝐴𝑇𝑖 represents a dummy for whether unit i was part of the treatment and 𝐴𝐹𝑇𝐸𝑅𝑡 is
a dummy variable equal to 1 when t is during the treatment period. 𝑒𝑖𝑡 represents the error term. Similarly to Equation 2, the coefficient for the fourth term (𝛽4) is essentially the estimated value for 𝛿𝑖𝑡 of equation 1C. Both equations can also be specified in terms of a difference-in-difference-in-difference model. In this model, some individual-specific effects are added to equations 2 and 3. Adding these individual-specific effects reveals the unique effect of the treatment on units of the treatment group with particular characteristics. These extra variables are constructed as an interaction term of a certain characteristic with the indication of the treatment group and the treatment itself.
2.2 Empirical strategy
10
This frequency of reporting is an important consideration. Although share prices are not a perfect substitute for ‘real’ financial indicators they are able to show developments at a very high frequency. Especially for more recent developments (such as the CSPP) it would not be informative to look at the development of variables such as profit margins which are usually reported at semi-annual frequencies. Such an indicator would only yield 4 values for the period after the start of the treatment for each company in the sample (and even less when accounting for the lag in data availability). In order to properly account for ‘noise’ due to dividends, share prices are converted to stock return indices; furthermore, they are normalized to 100 at the start of the treatment to allow for comparison; that is, all stock returns intersect at T = 0 (start of the treatment). The treatment subgroup and control subgroup are based on an equal-size sample of companies from the Eurostoxx-index. All companies from this index which have been part of the CSPP are used as treatment group. The control group is created by randomly choosing companies from the remainder of the index provided that the proportion of each industry classification is exactly the same between treatment group and control group. Section 3 discusses the procedure and the underlying data in more detail.
Another important choice relates to the way in which a difference in outcome between the treatment subgroup and the control subgroup can be linked to the CSPP. For this, an explanatory variable containing the CSPP intervention needs to be included. There are two ways of doing this: (1) adding an impulse dummy equal to 1 for the period after the start of the CSPP and zero otherwise into the regression (binary), or (2) adding a variable containing the exact amount of corporate bond purchases corresponding to period ‘t’ (continuous). Both measures are often used together in studies of monetary policy, see for example Jäger and Grigoriadis (2017) and Eser and Schwaab (2015). The model in Equation 4 below is estimated for both of these methods separately. Using again the terminology of Verbeek (2017); the treatment variable (𝑟𝑖𝑡) can be defined as follows:
𝑟𝑖𝑡 = 1 if company i has sold bonds to the ECB as part of the CSPP during period t
= 0 otherwise, in case of a binary (dummy) variable, and:
𝑟𝑖𝑡 = Eurozone-wide corporate bond purchases if company i has sold bonds to the ECB as part of the CSPP during period t
11
(4)
(5) These Eurozone-wide corporate bond purchases are calculated by dividing the magnitude of the purchases during the relevant period by the average value over the period June 2016 until December 2017. Since the ECB only provides a list of companies of which corporate bonds have been purchased during the treatment period (not an indication of the amount of purchases of each single company’s bonds), the assumption is made that purchases for each company were changing from month to month proportional to total Eurozone-wide purchases. The magnitude of these aggregate purchases is discussed and summarized in section 3. The analysis above results in the following equation, which is an application of the more generic Equation 2 of section 2.1:
∆𝑅𝑖𝑡 = 𝛽1+ 𝛽2(𝑟𝑡) + 𝛽3(𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡) + 𝑒𝑖𝑡
In this equation, i refers to a company and t refers to the period. The dependent variable R is the stock return indexed to 100 (100 = start treatment/June 2016). The period refers to a month within the time span 2014-2017 (48 months). This time frame includes the introduction and implementation of the CSPP (June 2016 - December 2017). The term 𝑇𝑅𝐸𝐴𝑇𝑖 is equal to 1 for companies in the treatment group, and the term 𝑟𝑡 represents the treatment (either a dummy or a continuous variable). The term 𝑒𝑖𝑡 represents the error term. Equation 5 uses the OLS method of
Equation 3 of section 2.1:
∆𝑅
̅̅̅̅𝑖𝑡 = 𝛽1+ 𝛽2𝑇𝑅𝐸𝐴𝑇𝑖+ 𝛽3𝐴𝐹𝑇𝐸𝑅𝑡+ 𝛽4(𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡) + 𝑒𝑖𝑡
In Equation 5, i refers to the company and t refers to the period. 𝑇𝑅𝐸𝐴𝑇𝑖represents a dummy for
whether company i was part of the treatment and 𝐴𝐹𝑇𝐸𝑅𝑡 is a dummy variable equal to 1 when t
is during the treatment period. The bar above the dependent variable indicates that the average is taken over either the period before June 2016 or over the period after June 2016. The dependent variable (R) in both equations is transformed into a change in percentage points between the current period and the preceding period. Boermans and Keshkov (2018) use a similar Two-period OLS regression to study the impact of the PSPP-programme on bond ownership concentration; however, they base the variable 𝑇𝑅𝐸𝐴𝑇𝑖 on programme eligibility rather than on actual bond sales
12
As an extension of the difference-in-difference model, both equations 4 and 5 are also transformed into a difference-in-difference-in-difference model by adding some individual effects to the existing regressions. The following company characteristics are used in constructing the difference-in-difference-in-difference model: (1) leverage (as estimated by the average debt to common equity ratio over the period 2010-2015), (2) size (as estimated by the average market capitalization in euros over the period 2010-2015), and (3) the average market-to-book ratio over the same period. The first one is motivated by the fact that a change in debt yield has a different effect on companies depending on its reliance on debt relative to equity. The other two are based on Fama and French (1993), who stress the importance of size and market-to-book ratio on stock returns. The values of all companies (the full sample of N = 200, see section 3 for details) are assembled and divided into one of the four relevant quartiles of 50 companies, each with its own dummy variable. In this way, each extra control requires the estimation of three extra explanatory dummy variables compared to the baseline model (the first quartile dummy is the omitted group). It is important not to use data during the period after March 2016 (when the CSPP was announced) for estimating the appropriate quartile in order to avoid endogeneity issues, as is stressed by for example Foley-Fisher et al. (2016). Companies respond to announcements of central bank programmes by increasing for example the leverage ratio, which then becomes an endogenous variable.
The variables of interest in the formulation of the hypotheses derive from two key objectives: (1) measuring the effect of the CSPP on the entire sample, and (2) determining the differential effect of the CSPP on the treatment group compared to the control group. Where the first objective looks at the total effect, the second objective looks at whether the group of companies selling bonds to the ECB benefitted more from the programme as compared to the group of companies which have not sold corporate bonds during the implementation of the CSPP. In equation 4, the variable of interest for the first objective is the term 𝑟𝑡, and the term relevant for the second objective is the
interaction term 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝑟𝑡 (with 𝑟𝑡 either in a binary or in a continuous form). In equation 5, the first objective is contained into the term 𝐴𝐹𝑇𝐸𝑅𝑡, and the second objective is represented into the interaction term between belonging to the treatment group (𝑇𝑅𝐸𝐴𝑇𝑖) and the term 𝐴𝐹𝑇𝐸𝑅𝑡.
13
In order to investigate the general effect of the CSPP-programme on the market valuation (the aforementioned first objective), the relevant alternative hypothesis is that 𝛽2 in Equation 4 and 𝛽3 in Equation 5 are significantly positive (again at the usual levels of significance). The Fixed Effects panel model with the continuous treatment variable also allows for a more precise estimation of the actual magnitude of the effect. The next section elaborates on the exact data requirements and data sources.
3 Data
The basis for the sample of companies used in the empirical analysis is the Eurostoxx-index. This index contains the largest listed companies for 35 different industry classifications within a selection of 11 member countries of the Eurozone (Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, The Netherlands, Portugal, and Spain). After eliminating 61 companies for which at least one of the required data types is unavailable and another 45 companies which are in the ‘financial sector’, a sample of 200 companies remains as the basis for the analysis. Companies from the financial sector are omitted because the focus of the CSPP is on non-financial corporations. Furthermore, financial corporations often concern banks or investment funds linked to banks. These companies are targeted through other programmes of the ECB (especially in the case of systematic banks). The following four industry classifications are treated as part of the financial sector: (1) banks, (2) financial services, (3) insurance, and (4) investment trusts. Of the 200 companies, 77 have sold bonds to the ECB during the implementation of the CSPP. From the remaining 123 companies, 77 companies were randomly chosen conditional to achieving a perfectly similar division of industry classification between treatment and control. More precisely, when the number of control companies exceeds the number of treatment companies within a particular industry classification, the number necessary to equalize both groups is deducted from the pool of control companies (in alphabetical order). In a few cases when the number of treatment companies exceeded the number of control companies, a similar procedure is applied on treated companies which resulted in 12 treated companies to be omitted from the sample (so that 65 companies remain in each group).
14
Figure 3 shows the magnitude of corporate bond purchases within the Eurozone as part of the CSPP. This figure shows that the purchases during the CSPP were ranging between roughly 4 billion euros and 10 billion euros until termination in December 2017. After that, purchases have been reduced at a steady pace (see also Fletcher and Smith; 2018).
FIGURE 3. Magnitude of Eurozone-wide corporate bond purchases during CSPP
Panel A. Monthly corporate bond purchases
Panel B. Cumulative corporate bond purchases
Return indices are used to approximate share prices. Return indices differ from ordinary share prices in that Return indices assume that dividends are reinvested at the then current stock price in order to avoid fluctuations in the stock price (and in unadjusted stock returns) exclusively due to dividends2. These return indices are made comparable by normalizing the value of June 2016 for each company (the start of the treatment) to 100 and indexing all other values to this base value.
2 Datastream calculates Return Indices as follows: 𝑅𝐼
𝑡= 𝑅𝐼𝑡−1∗ 𝑃𝑡
𝑃𝑡−1 where 𝑅𝐼𝑡 is the return index of the current period and 𝑅𝐼𝑡−1
is the return index of the previous period. 𝑃𝑡 is the price of the current period and 𝑃𝑡−1 is the price of the previous period. When t corresponds
to an ex-dividend date, the formula changes to 𝑅𝐼𝑡= 𝑅𝐼𝑡−1∗ 𝑃𝑡+ 𝐷𝑡
𝑃𝑡−1, where 𝐷𝑡 is the dividend payment associated with ex-date t. Note Fig. 3
15
Return indices are used at a monthly frequency and the time span in both the panel and the Two-period OLS is January 2014 until December 2017; that is 29 months before the start of the treatment and 19 months after the start of the treatment (T = 48 in total). Since Equation 5 only distinguishes two time periods, the changes in return index for each company are averages of either the period before June 2016 (before treatment) or of the period after June 2016 (during treatment).
4 Empirical application and results
4.1 Creating the treatment group and control group
Applying the procedure outlined in sections 2 and 3, a group of 130 companies is created, with 65 belonging to the treatment group and 65 belonging to the control group. As explained in the previous section, 61 companies were omitted from the initial sample due to missing data, and another 45 companies because they belong to the financial sector. The remainder of omitted companies is due to the fact that the fraction of each industry classification in each group is required to be equal; this also causes some attrition in the sample, as explained in section 33. Table 1 below shows the division over the various industry classifications of the constructed treatment group and control group.
Table 1. Industry classification division of treatment group and control group
This table shows the breakdown of the sample into 26 different industry classifications. It shows the division of the treatment group and of the control group separately. The number of companies within each industry classification is equal for the treatment group and the control group. All underlying data is downloaded from Datastream and processed using own calculations.
TREATMENT CONTROL
Industry classification No. Obs. No. obs.
Aerospace and defence Automobiles and parts Beverages
Chemicals
Construction and materials Electricity 3 6 2 6 4 4 3 6 2 6 4 4
3 The Eurostoxx index distinguishes 31 different industry classifications (when omitting the financial sector); however, 5 industry
16
Table 1 Cont’d
Electronic and electrical equipment Fixed line telecommunications Food and drug retailers Food producers
Gas, water and multiutilities General industrials
General retailers
Health care equipment and services Household goods and home construction Industrial engineering
Industrial transportation Media
Mobile telecommunications Oil and gas producers Oil equipment and services Personal goods
Pharmaceutical and biotechnology Software and computer services Technology hardware and equipment Travel and leisure
Total 2 1 2 2 3 1 1 3 1 2 2 4 1 2 1 3 3 2 2 2 65 2 1 2 2 3 1 1 3 1 2 2 4 1 2 1 3 3 2 2 2 65
4.2 The validity of the parallel trend assumption
17
The period investigated in this section is January 2014 until and including May 2016 plus 3 months after the start of the treatment (until 1-9-2016). In Figure 4, the average normalized return index of the treatment group and control group is depicted, where the value at the start of the treatment is normalized to 100 for all companies before calculating the average values. Naturally, this implies that both groups intersect at the vertical line ‘T = 0’ in the figure (where the return index equals 100 for both groups). It can be seen from Figure 4 that the parallel trend assumption holds rather strongly for the period before T = 0 (start of treatment), implying that the constructed groups as summarized in the previous section are appropriate to use for difference-in-difference estimation.
FIGURE 4. Monthly normalized return index for treatment- and control group
4.3 The effect of the CSPP on stock market returns
This section presents the results from the empirical analysis. Table 2 contains the estimated parameters from Equation 4. The table presents separately the results from the two aforementioned methods of incorporating the treatment variable; panel A presents the results using an impulse dummy equal to 1 for periods during the CSPP, and panel B presents the results when a continuous variable is used instead.
Note Fig. 4
18
TABLE 2. Panel with fixed effects
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the change in stock return indexed to 100 (100 = start treatment/June 2016) where stock returns are based on Return Indices. The model is estimated using Fixed Effects. Row (1) reports the results without controls, row (2) reports the results with an interaction between treatment and size, row (3) reports the results with an interaction term between treatment and leverage, and row (4) reports the results with an interaction between treatment and market-to-book ratio. Row (5) reports the results with all aforementioned controls included. The treatment variable (𝑟𝑡) refers to
an impulse dummy in Panel A and to a continuous variable representing the monetary amount of purchases (calculated as the value divided by the average over the period June 2016 – December 2017) in Panel B. Standard errors are depicted in parentheses, and statistical significance is indicated with *, **, and *** for the 10 percent, 5 percent, and 1 percent significance level respectively.
Panel A. Impulse dummy
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
19
Panel B. Continuous variable
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) 0.430*** (0.115) 1.644*** (0.249) -0.437 (0.355) 0.430*** (0.115) 1.645*** (0.249) -0.076 (0.431) -0.283 (0.732) -0.966 (0.566) 0.000 (0.000) 0.430*** (0.115) 1.645*** (0.249) -0.461 (0.720) 0.417 (0.835) 0.731 (0.813) -1.120 (0.827) 0.430*** (0.115) 1.644*** (0.249) -0.820 (0.565) 0.540 (0.706) 0.135 (0.688) 1.070 (0.774) 0.430*** (0.115) 1.645*** (0.249) -0.521 (0.940) -0.118 (0.745) -0.836 (0.577) 0.000 (0.000) 0.122 (0.856) 0.648 (0.845) -1.201 (0.864) 0.824 (0.714) 0.217 (0.714) 1.046 (0.798) No. obs. 𝑅2 6,240 0.0107 6,240 0.0112 6,240 0.0121 6,240 0.0111 6,240 0.0129
20
Note that belonging to the treatment group does not matter significantly according to the results in Panel B (the interaction between belonging to the treatment group and the treatment is insignificant in all columns). In Panel A, however, the general effect of the treatment is negative, and belonging to the treatment group compensates for this to a small extent.
Table 3 contains the estimated parameters for Equation 5. This equation essentially has two dependent variables per company; one containing the average change in normalized stock returns for the period before June 2016, and one representing the average change in normalized stock returns for the period after June 2016. Each value for the period after June 2016 corresponds to a 1 under the variable 𝐴𝐹𝑇𝐸𝑅𝑡 , and both values of a treatment company are accompanied by a 1 in
both periods for the variable 𝑇𝑅𝐸𝐴𝑇𝑖.
TABLE 3. Two-period OLS
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the average change in stock return indexed to 100 (100 = start treatment/June 2016) for the respective period (before or after June 2016). Stock returns are based on Return Indices. 𝑇𝑅𝐸𝐴𝑇𝑖 represents a dummy for whether
company i was part of the treatment and 𝐴𝐹𝑇𝐸𝑅𝑡 is a dummy variable equal to 1 when t is during the treatment period
(period after June 2016). The model is estimated using OLS. Row (1) reports the results without controls, row (2) reports the results with an interaction between treatment and size, row (3) reports the results with an interaction term between treatment and leverage, and row (4) reports the results with an interaction between treatment and market-to-book ratio. Row (5) reports the results with all aforementioned controls included. Standard errors are depicted in parentheses and refer to Heteroskedasticity and Autocorrelation Consistent (HAC) standard errors. Statistical significance is indicated with *, **, and *** for the 10 percent, 5 percent, and 1 percent significance level respectively.
Variable (1) (2) (3) (4) (5) C 0.593*** (0.127) 0.593*** (0.128) 0.593*** (0.128) 0.593*** (0.128) 0.593*** (0.129) 𝑇𝑅𝐸𝐴𝑇𝑖 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡×
Market cap. (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 ×𝐴𝐹𝑇𝐸𝑅𝑡×
Market cap. (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡×
Market cap. (Quartile 4)
21 Table 3 Cont’d 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Leverage (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Leverage (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Leverage (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖 ×𝐴𝐹𝑇𝐸𝑅𝑡× Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Market-to-book (Quartile 4) 0.188 (0.734) 0.325 (0.644) -1.044* (0.627) 0.533 (0.488) 0.635* (0.362) 1.703** (0.713) 0.129 (0.607) 0.582 (0.588) -0.815 (0.522) 0.652 (0.408) 0.605* (0.311) 1.667** (0.694) No. obs. 260 260 260 260 260 𝑅2 0.1225 0.1239 0.1547 0.1517 0.1806
Table 3 confirms the findings of the second panel of Table 2 that companies in the sample experienced an increase in average stock returns when comparing the ‘before’-period to the ‘after’-period (independent of actually being treated). The fact that Panel A of Table 2 gives different results might be due to the fact that the impulse dummy does not contain enough information to give an adequate interpretation of the effect of the CSPP. Section 5 will perform some robustness checks in order to see whether the results are different with different time frames.
22
Portfolio rebalancing is likely to have increased demand for corporate bonds of companies not selling bonds to the ECB as well. In this way, bond yields of both treated and untreated companies will be adjusted downwards, which translates into similar movements in stock returns. Both groups experienced a similar increase in stock returns, independent of which sample of companies actually sold corporate bonds to the central bank. The theory of cash reinvestment only holds for secondary purchases, not for primary purchases since only secondary purchases results in investor’s portfolio rebalancing. This provides an opportunity to further test the validity of this theory. A way to do so is to estimate the Fixed Effects panel model of Equation 4 with the continuous treatment variable again, but now the continuous treatment variable is based on primary purchases only. In this case, the significance of the coefficient related to the treatment should be smaller compared to the baseline model presented in Table 2. Table 4 shows the results for this alternative model. Comparing the coefficient for the term 𝑟𝑡 of Table 2 and Table 4 reveals that omitting the secondary
purchases considerably reduced the significance of the general effect of the CSPP. This finding provides evidence in favour of the cash reinvestment theory.
TABLE 4. Panel with fixed effects (primary purchases only)
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the change in stock return indexed to 100 (100 = start treatment/June 2016) where stock returns are based on Return Indices. The model is estimated using Fixed Effects. Row (1) reports the results without controls, row (2) reports the results with an interaction between treatment and size, row (3) reports the results with an interaction term between treatment and leverage, and row (4) reports the results with an interaction between treatment and market-to-book ratio. Row (5) reports the results with all aforementioned controls included. The treatment variable (𝑟𝑡) refers to
a continuous variable representing the monetary amount of primary purchases (calculated as the value divided by the average over the period Jun 2016 – December 2017). Standard errors are depicted in parentheses, and statistical significance is indicated with *, **, and *** for the 10 percent, 5 percent, and 1 percent significance level respectively.
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
23 Table 4 Cont’d 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) 0.454 (0.579) -0.657 (0.589) 0.160 (0.503) 0.321 (0.490) 0.702 (0.551) 0.492 (0.602) -0.566 (0.615) 0.332 (0.508) 0.418 (0.508) 0.663 (0.568) No. obs. 𝑅2 6,240 0.0011 6,240 0.0018 6,240 0.0023 6,240 0.0014 6,240 0.0031
Some potential alternative explanations for the results found in this section could be that investors anticipated correctly which companies would become part of the programme after the announcement of the programme in March 2016. This would imply that if the treatment period is set to begin in March rather than June, a differential effect would be observable. Another potential explanation (related to these anticipation effects) is that the time frame after the start of the treatment is too long and reducing the period after the treatment would reveal an effect greater for treated companies as compared to control companies. Furthermore, altering the time frame also helps to distinguish the more dynamic effects of the CSPP by looking at how the effect changes over time. These issued are adressed in the next section.
5 Robustness checks
24
TABLE 5. Two-period OLS with anticipation effect
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the average change in stock return indexed to 100 (100 = announcement treatment/March 2016) for the respective period (before or after March 2016). Stock returns are based on Return Indices. 𝑇𝑅𝐸𝐴𝑇𝑖 represents a
dummy for whether company i was part of the treatment and 𝐴𝐹𝑇𝐸𝑅𝑡 is a dummy variable equal to 1 when t is during
the treatment period (period after June 2016). The model is estimated using OLS. Row (1) reports the results without controls, row (2) reports the results with an interaction between treatment and size, row (3) reports the results with an interaction term between treatment and leverage, and row (4) reports the results with an interaction between treatment and market-to-book ratio. Row (5) reports the results with all aforementioned controls included. Standard errors are depicted in parentheses and refer to Heteroskedasticity and Autocorrelation Consistent (HAC) standard errors. Statistical significance is indicated with *, **, and *** for the 10 percent, 5 percent, and 1 percent significance level respectively. Variable (1) (2) (3) (4) (5) C 0.684*** (0.153) 0.684*** (0.154) 0.684*** (0.154) 0.684*** (0.154) 0.684*** (0.156) 𝑇𝑅𝐸𝐴𝑇𝑖 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
25
Table 5 Cont’d
No. obs. 260 260 260 260 260
𝑅2 0.1187 0.1324 0.1305 0.1340 0.1585
Comparing Table 5 to Table 3, changing the start of the treatment period to March 2016 rather than June 2016 does not change the results considerably. Still, the model finds significant evidence that stock returns were higher after the treatment than before the treatment; furthermore, the interaction between 𝐴𝐹𝑇𝐸𝑅𝑡 and 𝑇𝑅𝐸𝐴𝑇𝑖 remains insignificant. The magnitude of the coefficient
for the 𝐴𝐹𝑇𝐸𝑅𝑡-term has decreased slightly in Table 5. This means that the average effect on stock returns was greater during the period after June 2016 compared to the period after March 2016, implying that anticipation effects were small. Another interesting robustness check is to investigate whether changing the time frame after the start of the treatment alters the results significantly. Such a test could yield different results for different time frames if for example the impact of purchases on investor sentiments decreases over time. Table 6 again applies Equation 4; however, now the time period after the start of the treatment is gradually extended. The same is done for Equation 5 in Table 7. In both tables, three different time frames are used: 3 months (T = 3), 6 months (T = 6), and 9 months (T = 9) after the start of the treatment. The period before the treatment is still starting at January 2014.
TABLE 6. Panel with fixed effects and different time frames
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the change in stock return indexed to 100 (100 = start treatment/June 2016) where stock returns are based on Return Indices. The model is estimated using Fixed effects. Row (1) reports the results without controls, row (2) reports the results with an interaction between treatment and size, row (3) reports the results with an interaction term between treatment and leverage, and row (4) reports the results with an interaction between treatment and market-to-book ratio. Row (5) reports the results with all aforementioned controls included. The treatment variable (𝑟𝑡) refers to
an impulse dummy in Panel A and to a continuous variable representing the monetary amount of purchases (calculated as the value divided by the average over the period Jun 2016 – December 2017) in panel B. T = 3 refers to three months after the start of the treatment. Similarly, T = 6 refers to six months, and T = 9 refers to 9 months. Standard errors are depicted in parentheses, and statistical significance is indicated with *, **, and *** for the 10 percent, 5 percent, and 1 percent significance level respectively.
T = 3 Panel A. Impulse dummy
26
Table 6 Cont’d
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) -0.393 (1.341) -1.310 (1.037) 0.000 (0.000) 1.695 (1.534) 1.096 (1.491) -1.338 (1.517) 1.670 (1.293) 1.791 (1.264) 1.961 (1.417) -0.225 (1.364) -1.132 (1.058) 0.000 (0.000) 1.610 (1.574) 1.406 (1.551) -1.090 (1.586) 2.005 (1.307) 1.961 (1.314) 1.767 (1.460) No. obs. 𝑅2 4,290 0.0001 4,290 0.0005 4,290 0.0017 4,290 0.0008 4,290 0.0027
T = 3 Panel B. Continuous variable
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
27 Table 6 Cont’d 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) 1.443 (1.357) -0.960 (1.380) 1.574 (1.178) 1.350 (1.154) 1.374 (1.292) 1.637 (1.411) -0.874 (1.443) 1.892 (1.191) 1.538 (1.199) 1.242 (1.331) No. obs. 𝑅2 4,290 0.0005 4,290 0.0008 4,290 0.0022 4,290 0.0011 4,290 0.0030
T = 6 Panel A. Impulse dummy
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
28 Table 6 Cont’d 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) 1.187 (1.113) 0.947 (1.146) No. obs. 𝑅2 4,680 0.0003 4,680 0.0008 4,680 0.0025 4,680 0.0009 4,680 0.0034
T = 6 Panel B. Continuous variable
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
29
T = 9 Panel A. Impulse dummy
Variable (1) (2) (3) (4) (5)
C
𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) 0.951*** (0.144) -0.408** (0.197) 0.795 (0.320) 0.951*** (0.144) -0.407** (0.197) 1.281*** (0.445) -0.150 (0.921) -1.407** (0.712) 0.000 (0.000) 0.949*** (0.134) -0.403** (0.197) 0.526 (0.850) 0.987 (1.053) 1.010 (1.025) -1.097 (1.043) 0.949*** (0.144) -0.403** (0.197) 0.219 (0.638) 1.122 (0.889) 0.183 (0.867) 1.190 (0.974) 0.947*** (0.144) -0.400** (0.198) 0.481 (1.142) 0.112 (0.937) -1.213* (0.726) 0.000 (0.000) 0.541 (1.079) 0.787 (1.066) -1.362 (1.090) 1.476 (0.898) 0.284 (0.902) 1.128 (1.003) No. obs. 𝑅2 5,070 0.0020 5,070 0.0028 5,070 0.0036 5,070 0.0025 5,070 0.0049
T = 9 Panel B. Continuous variable
30
Table 6 Cont’d
𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡×
Market capitalization (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 ×
Market capitalization (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡× Debt-to-equity (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Debt-to-equity (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖× 𝑟𝑡 × Market-to-book (Quartile 4) -0.063 (0.820) -1.178* (0.634) 0.000 (0.000) 0.677 (0.936) 0.968 (0.911) -0.829 (0.926) 0.792 (0.791) -0.179 (0.771) 0.623 (0.867) 0.135 (0.834) -1.017 (0.646) 0.000 (0.000) 0.260 (0.960) 0.676 (0.947) -1.157 (0.968) 1.106 (0.799) -0.079 (0.801) 0.620 (0.893) No. obs. 𝑅2 5,070 0.0049 5,070 0.0057 5,070 0.0063 5,070 0.0053 5,070 0.0074
TABLE 7. Two-period OLS with different time frames
This table studies the impact of the CSPP on stock market returns of companies in the Eurozone. The dependent variable is the average change in stock return indexed to 100 (100 = start treatment/June 2016) for the respective period (before or after June 2016). Stock returns are based on Return Indices. 𝑇𝑅𝐸𝐴𝑇𝑖 represents a dummy for whether
company i was part of the treatment and 𝐴𝐹𝑇𝐸𝑅𝑡 is a dummy variable equal to 1 when t is during the treatment period
31
Table 7 Cont’d
𝐴𝐹𝑇𝐸𝑅𝑡
𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡
𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market cap. (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
32
Table 7 Cont’d
𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market cap. (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡× Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market-to-book (Quartile 4) -0.740 (0.681) 0.000 (0.000) 0.110 (0.651) 1.584** (0.760) 2.140** (0.922) 1.532** (0.693) 0.401 (0.835) -1.029 (0.908) -0.748 (0.864) 0.000 (0.000) 0.931 (0.689) 0.851 (0.700) 1.416* (0.776) 1.897** (0.824) 1.205* (0.719) 0.361 (0.828) -0.729 (0.888) -0.626 (0.909) No. obs. 260 260 260 260 260 𝑅2 0.0103 0.0177 0.0446 0.0373 0.0692 T = 9 (1) (2) (3) (4) (5) C 0.593*** (0.127) 0.593*** (0.128) 0.593*** (0.128) 0.593*** (0.128) 0.593*** (0.129) 𝑇𝑅𝐸𝐴𝑇𝑖 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 x 𝐴𝐹𝑇𝐸𝑅𝑡 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 2)
33
Table 7 Cont’d
𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 ×
Market cap. (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Leverage (Quartile 4) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market-to-book (Quartile 2) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market-to-book (Quartile 3) 𝑇𝑅𝐸𝐴𝑇𝑖 × 𝐴𝐹𝑇𝐸𝑅𝑡 × Market-to-book (Quartile 4) 0.000 (0.000) 0.312 (0.652) 0.954 (0.683) 1.853* (1.113) 1.639** (0.706) -0.242 (1.055) -1.659 (1.094) -1.529 (1.062) 1.070 (0.677) 1.274 (0.861) 0.707 (0.635) 1.439 (0.895) 1.209* (0.658) -0.311 (1.006) -1.339 (0.985) -1.364 (1.004) No. obs. 260 260 260 260 260 𝑅2 0.1123 0.1246 0.1427 0.1557 0.1845
When comparing the three time frames (T = 3, T = 6, and T = 9) in tables 6 and 7, both the panel with the continuous treatment variable and the Two-period OLS model reveal that the effect of the CSPP on stock returns of companies is most significant when using the time frame of 9 months after the start of the treatment. The results of tables 6 and 7 also suggest that a shorter time frame does not reveal a differential effect between the treatment group and the control group. Combining the lower magnitude of the effect of the CSPP when using the announcement of the CSPP in Table 5 and the fact that the effect only becomes significant after 9 months in tables 6 and 7 highlights the fact that anticipation effects were not very strong. Instead, the effect on market valuation closely followed the actual purchases over time (as indicated by the panel with the continuous treatment variable).
34
Combined with the fact that the Two-period OLS and the Panel with the continuous variable give similar results (also in a more dynamic approach with changing time frames), it is assumed that the impulse dummy contains too little information to really isolate the effect. This problem is also magnified by the fact that the panel uses relatively high-frequency data in which periods are short and therefore the effect per period is small. Since the impulse dummy only distinguishes between during treatment (1) and not during treatment (0), it is not able to detect more subtle changes.
Comparing the results of Table 2 until Table 7, the columns containing a difference-in-difference-in-difference analysis do not point to a specific financial fundamental (size, leverage, or market-to-book ratio) that has consistently an individual-specific effect. It can be concluded that not only there is no differential effect between belonging to the treatment group or control group, but the effect is also not related to a company’s size, leverage ratio, and market-to-book ratio. Note however that these individual effects are only estimated for companies in the treatment group. When estimating individual effects for the control group as well, the results could be different.
6 Conclusion and discussion
35
These results are strengthened by the fact that the ‘treatment’ is based on actual corporate bond sales rather than on programme-eligibility. The results are also relevant from the perspective of economic policy because it provides evidence that Quantitative Easing does not have a distortive effect in terms of market valuation. This implies that the choice of which particular financial instruments are purchased by the central bank should only be driven by criteria such as risk and maturity, not by the specific companies they are purchased from (either directly or indirectly).
A limitation of this thesis lies in the fact that this thesis relies on data from larger and listed companies. Further research should therefore also focus on the (spillover) effects of unconventional monetary policies on smaller (unlisted) companies for they are in the end most strongly affected by a financial crisis (and by a downturn in general). This would however require the creation and integration of more regional-level datasets. As opposed to a difference-in-difference model – which is designed to compare similar companies – further research could focus on potential vertical spillovers rather than horizontal spillovers by comparing dissimilar companies in terms of for example size or risk-profile. Another limitation is related to the fact that the programme under investigation in this thesis is very recent. It would be interesting to see at a later stage what the long-term effects of this programme and similar programmes are; for example, have companies become persistently more reliant on debt rather than equity, or have real interest rates decrease so far that they facilitate inefficient economic activity? These are relevant questions to be investigated in the future. Further research is essential to thoroughly evaluate the monetary strategies applied during the financial crisis.
REFERENCES
Arce, O, Gimeno, R. & Mayordomo, S. (2017). Making room for the needy: the credit-reallocation effects of the ECB’s corporate QE. Banco de Espana Documentos de Trabajo Nr. 1743(2017) Boermans, M. & Keshkov, V. (2018). The impact of the ECB asset purchases on the European bond market structure: granular evidence of ownership concentration. Publication of De Nederlandsche Bank (DNB)
Brooks, C. (2014). Econometrics for finance. Cambridge University Press, Cambridge (United Kingdom)
36
Drechsler, I., Savov, A. & Schnabl, P. (2018). A model of monetary policy and risk premia. The journal of finance, 73(1); pp. 317-373
ECB (2018). Unconventional monetary policy in the Eurozone. Retrieved from: https://www.ecb .europa.eu/mopo/implement/omt/html/index.en.html
Eser, F. & Schwaab, B. (2015). Evaluating the impact of unconventional monetary policy measures: empirical evidence from the ECB’s Securities Markets Programme. The journal of financial economics, 119(1); pp. 147-167
Fama, E.F. & French, K.R. (1993). Common risk factors in the returns on stocks and bonds. The journal of Financial Economics, 33(1), pp. 3-56
Fletcher, L. & Smith, R. (2018). ECB’s corporate exit leaves bond investors on edge. Publication of The Financial Times (November 27, 2018). Retrieved from: https://www.ft.com/content/ 3c591338-f193-11e8-9623-d7f988 1e729f
Foley-Fisher, N., Ramcharan, R. & Yu, E. (2016). The impact of unconventional monetary policy on firm financing constraints: evidence form the maturity extension program. The journal of financial economics, 122(2); pp. 409-429
Grauwe, P. & Ji, Y. (2015). Correcting for the Eurozone Design Failures: the role of the ECB. Journal of European Integration, 37(7), 739-754
Grauwe, P. de (2016). Economics of monetary union. Oxford University Press, Oxford (United Kingdom)
Gross-Rueschkamp, B., Steffen, S. & Streitz, D. (2017). Cutting out the middleman – the ECB as corporate bond investor. SSRN, October 2017.
Haan, J. de, Oosterloo, S., and Schoenmaker, D. (2015). Financial markets and institutions. Cambridge University Press, Cambridge (United Kingdom)
Hill, R.C., Griffiths, W.E., & Lim, G.C. (2012). Principles of econometrics. John Wiley & Sons Inc., Hoboken (United States)
Jäger, J. & Grigoriadis, T. (2017). The effectiveness of the ECB’s unconventional monetary policy: comparative evidence from crisis and non-crisis Euro-area countries, The journal of international money and finance, 78; pp. 21-43
37
Santes, R.A. de, Geis, A., Juskaite, A. & Cruz, L.V. (2018). The impact of the corporate sector purchase programme on corporate bond markets and the financing of euro area non-financial corporations. ECB Economic Bulletin, Issue 3(2018)
Stiglitz, J.E. (2016). The Euro. Pinguin Books, London (United Kingdom)
Verbeek, M. (2017). A guide to modern econometrics. John Wiley & Sons Inc., Hoboken (United States)
