1
Zero interest rates and the impact
on dividend pay-out in the U.S.
Date: 29/01/2018
Supervisor: Robin Doettling Academic year: 2017-2018
Bachelor’s thesis Semester 1 periods 2 & 3
Sven van Kraaij 10996893 Abstract
In this research the main question is whether the zero interest rate policy, which was
implemented by the Federal Reserve Bank in December 2008, has an effect on the dividend pay-out of firms from the United States. In order to come to a conclusion an already existing model is used, while a dummy variable for the zero interest rate policy, the CPI, and quarter and industry fixed effects are added. The tests used in this research are OLS regressions and firm fixed effects regressions, both with robust standard errors. The firm fixed effects regression is the most appropriate regression to use for the panel data regression, so the conclusions result from that regression, while the standard OLS regression might be used as an interpretation. The main result of this research is that a switch to the zero interest rate policy will decrease the dividend pay-out, which is written as the dividend yield, of the U.S. firms. However, the dividend yield itself increases, which is not supported by the theory. Secondly, the zero interest rate policy seems to have a negative effect on the repurchase yield, which is also not in line with theory.
2
Statement of Originality
This document is written by Sven van Kraaij who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
1. Introduction
On December the 17th 2008 the Federal Reserve Bank, abbreviated as the FED,
which is the national bank of the United States, decreased the interest rate to the lowest level ever occurred in the US (Elliott & Seager, 2008). This was the beginning of the nearly zero-interest rate policy the FED implemented. According to Chen (2017), the nearly-ZIRP (nearly-zero interest rate policy) was a response of the FED on the weak recovery of the 2007-2009 economic recession. During this period an interest rate between 0 and ¼ percent was implemented. Because of the fact that the interest rate is nearly zero, it is cheap to borrow money. In this way the FED tries to stimulate
companies to borrow money in order to make an end to the crisis by increasing the economic activity. Chen’s research found that this nearly-ZIRP indeed boosts output and thus stimulates economic activity, but it also increases the inflation.
Finally, on December the 16th 2015, almost 7 years later, the FED increased the
interest rate (Oyedele, 2015). The FED decided to increase the target range of the interest rate from a maximum of ¼ percent to a minimum of ¼ percent and a maximum of ½ percent. The Federal Open Market Committee even expected the interest rate to be 1.375 percent in the end of 2016.
The research question of this paper is: what is the effect of a nearly-ZIRP on the corporate dividend pay-out policy of firms from the U.S.? In order to find an answer to this question, a model will be composed by taking into account some literature. When testing the model, the most important finding would be a significant effect of the nearly-ZIRP on the dividend yield, which is the dividend pay-out as a fraction of the share price.
I add value to this topic because there exists no paper where the zero interest rate policy is mentioned as an effect on corporate dividend pay-out. However, what has been examined in the paper of Khan Khan et al. (2013) is the effect of inflation on the
dividend policy of companies in Pakistan. In this paper Khan Khan et al. (2013) also find that interest rate is positively related with dividend pay-out. So despite shifting the research country from Pakistan to the U.S., I want to find out whether a zero interest rate policy affects the dividend pay-out, which could differ from the findings of Khan Khan et al. (2013).
The first question that has to be answered is: why does the zero-interest rate policy affect the dividend pay-out? This will be answered by making a link between some literature.
4 First of all it must be clear that no direct relation exists between these two.
Therefore, an indirect relation can be used. Copeland et al. (2014, p.615) claims that a higher investment rate lowers the current dividend pay-out. Mundell (1963) and Lutz (1945) make the relation between interest rates and its impact on investments. These articles combined lead to the relation between the interest rate and dividend pay-out. According to Lutz (1945), there is evidence that the interest rate does have an impact on investment. However, this is not always the case. The most important finding of his research, in order to use in this research, is that “the level of the interest rate affects the willingness of financial institutions to grant credit or to float bonds and stocks, so that the interest rate may influence the volume of investment” (Lutz, 1945).
Mundell (1963) claims that the real interest rate decreases as inflation rate rises, and this will lead to investment booms. According to him this is caused by the fact that the inflation will reduce real money balances and the resulting decline in wealth will increase the amount of savings. This is very helpful in this research because now I know that interest rate and inflation do have an impact on each other but also on the
investment rates, and therefore on the dividend pay-out. This is why not only the interest rate but also inflation should also be added to the model.
Lutz (1945) does not mention what the impact of a decrease or increase of the
interest rate on the investment rate is. However, when taking into account the model of Mundell (1963) the relation can be determined. As Mundell (1963) claims: a rise in inflation will cause a decrease in real interest rate, and this causes the investments to rise. Copeland (2014, p.615) claims that a rise in investments will lower the current dividend pay-out. These two claims combined leads to the conclusion that a decrease in interest rate will lower the current dividend pay-out.
This means that a decrease in real interest rates, thus also an interest rate of zero, would increase the total investments, since it is cheaper to borrow money. However, since this money will be used for the investments, less money will be left for the pay-out of dividends. This will lead to a decrease of the dividend pay-out. Since managers are not keen with cutting dividends, they will increase the amount of repurchases in order to give the shareholders at least the same pay-out.
The hypothesis of this research is that I expect that this zero-interest rate policy will have a negative effect on the dividend pay-out of firms, and therefore the dividend pay-out to decrease. This is because of the fact that it is cheaper for companies to borrow money, so instead of paying out dividend it is possible to buy back shares or to invest.
5 The data used in this research is from the CRSP-Compustat merged database of WRDS. It is a longitudinal research, which means that two time-frames will be
compared in this case. The data that is used is computed quarterly from January 2002 until December 2015, consisting of all the companies in this database. In the third paragraph the model will be explained in more detail. The main idea is that one of the added variables is a dummy variable which takes a value of 1 for the period when the nearly-ZIRP is implemented, and 0 otherwise.
This research shows that the nearly-ZIRP dummy variable indeed has a negative effect on the dividend yield, which means that a shift to a policy with nearly zero interest rates will on average decrease the dividend yield. However, what can be seen is that the dividend yield itself increases instead of decreases, which is not in line with what was expected. Secondly, an effect of the industry fixed effects on the coefficient age was detected, which means that the type of industry influences the dividend yield through the age. Finally, a regression on the repurchase yield has taken place. This resulted in a negative effect of the nearly-ZIRP dummy variable on the repurchase yield, which can be confirmed by the decrease of the average repurchase yield. This means that a shift to the nearly-ZIRP would decrease the repurchase yield on average. However, this is not in line with what was expected.
In order to come to this conclusion dividend pay-out and zero-interest rate will be explained in the next paragraph. In the third paragraph the model will be explained in detail. This means that the existing model that is used will be mentioned, but also the additional variables, and why all variables used in the model could have an impact on the pay-out of dividend. Finally, the way of testing and the main results will be
discussed, after which a conclusion can be drawn in the last paragraph.
2. Literature
2.1 Dividend pay-out
According to Copeland et al. (2014) shareholders are indifferent between receiving their money as dividends or as capital gains, as long as no taxes exist in the world. However, in the real world there are corporate and personal taxes, which both have an influence on shareholder’s wealth. This thus influences the decision to pay out dividends instead of capital gains, or the other way round.
The differences between taxes on capital gains and taxes on dividends is explained by Berk & DeMarzo (2014): ‘shareholders must pay taxes on dividends they receive, but
6 they must also pay capital gain taxes when they sell their shares’. This means that dividend pay-out and share repurchases have different tax rates. So when the tax rate on dividends exceeds the tax rate on capital gains, the taxes paid for repurchases is lower than taxes paid for dividends. The firm value of a firm using share repurchases over dividend pay-out, like the case illustrated above, will then increase with the amount of tax savings (Berk & DeMarzo, 2014).
As table 17.2 in the book of Berk & DeMarzo (2014) shows us, tax rates on dividends exceeds tax rates on capital gains from 1971 until 2002. However, as Figure 17.4 shows us, the amount of firms using dividend pay-out, has always been larger than firms repurchasing shares. This will definitely raise some questions, because the tax
advantage should increase firm value if the firm would choose to repurchase its shares. The table below is table 17.2 from Berk & DeMarzo (2014, p.594). This table contains the long-term capital gains tax rates and the dividend tax rates in the US from 1971 until 2012. What can be seen is that the dividend tax rate only equals the capital gains rate since 2003, while it was higher ever before.
The next graph is figure 17.4 from Berk & DeMarzo (2014, p.595). This figure shows the percentage of firms paying dividends and the percentage of firms repurchasing shares. What can be seen is that the percentage of firms paying dividends decreased over time, while the percentage of firms repurchasing shares remained quite stable.
7 DeAngelo et al. (2008) tells us, recent years total pay-outs primarily exists of only two components: dividend pay-out and stock repurchases. They also say that
repurchases will distribute to the earnings of shareholders, without having to higher the dividend pay-out which would ‘infer from an increase in the firm’s regular dividend’ (DeAngelo et al., 2008). This is an advantage of a share repurchase compared to dividend pay-out. In this research the stock repurchases are also expected to increase as the dividend pay-out decreases, vice versa. So, this will eventually be tested.
Managers are most likely to repurchase when favourable transitory earnings, ‘earnings that reflect a material one-time asset write-down or a temporary cash
windfall’, are realized (DeAngelo et al., 2008). Skinner (2008) approves this and indicates that firms that repurchase frequently, primarily buy back shares in years where the firm leads a loss or when there are positive earnings that are smaller than the firm’s total pay-out. These situations are not sustainable in the long run.
Why would firms then choose to pay out dividends instead of repurchasing shares? One of the reasons is that dividends do not transfer cross-stockholder wealth, which is the case with repurchases because it distributes cash to either the selling or the non-selling group of stockholders, depending on whether the repurchase price is below or above intrinsic value (DeAngelo et al., 2008).
Another argument is that the investors who would not sell their stock while the firm is repurchasing stock, will not be harmed by a dividend pay-out when the firm’s shares are overvalued (Chowdhry & Nanda, 1994).
Furthermore, according to Barclay and Smith (1988), ‘repurchases can raise the bid/ask spread’, which rests on the assumption that insiders of the company will
8 sometimes repurchase the firm’s shares when they are undervalued in order to exploit outside investors.
Finally, in the Myers and Majluf’s (1984) asymmetric information model, where managers who are seeking to repurchase shares, investors will be mistrustful because they expect the managers to be opportunistic, and thus only buy back at a price below the firms estimate of the intrinsic value. This causes the investors not voluntarily willing to sell their shares back to the firm, because this might represent an attempt to buy back stock cheaply. The only way to retrieve full value will then be achieved by paying out cash dividends. These arguments might explain why dividends dominate share repurchases.
The argument given above has to do with dividend signalling. For example, in
rational markets, there is a chance that the share price might increase at announcement of a dividend increase. If the dividend change reveals new earnings information, this might be a side effect of the pay-out decision of the manager. However, the dividend may also reflect to a managers choice to communicate his/her prospect about future earnings to investors outside the company (DeAngelo et al., 2008).
While there might be evidence that a cut in dividends or a change from dividend pay-out to the repurchase of shares would sometimes be better for the firm, the managers often decide not to change their pay-out policy. As the survey of Brav et al. (2005)
indicates, managers of dividend paying firms place priority on maintaining the dividends at this rate. These managers would often, in order to maintain the dividend, even cede some positive NPV investment projects. However, these same managers would decide how much shares to buy back only after the investment decisions were made, while the dividend pay-out policy will be determined before the investment decisions are made. This confirms the fact that managers view dividend pay-out as a permanent cash
distribution, whereas share repurchases are temporary, mostly single-period, transitory pay-outs (DeAngelo et al., 2008).
Another reason why dividend cuts are not ideal, is that the stock market responses negatively to these cuts (Ghosh & Woolridge, 1989), even if the managers could explain that these cuts are in the favour of the outside stockholder, e.g., to keep cash which can be used to purchase a machine that increases net profitability. The stockholders are reluctant to a cut in dividends because this often refers to a firm being in financial trouble.
These arguments could be the reason to deny the hypothesis. If the managers are not willing to buy back shares, they are most likely to at least keep the dividends at the
9 same level. In that case, the nearly-ZIRP thus does not have an influence on the
dividend pay-out.
2.2 Zero interest rate policy
In this paragraph the zero interest rate policy and its effects will be explained. A Nearly Zero Interest Rate Policy means that the interest rate, which is decided by the Federal Reserve Bank, will be very close to zero. The maximum interest rate during this nearly-ZIRP period equalled 0.25%. This means that companies and consumers could borrow money against this rate, which is a cheap deal compared to previous years and even decades. This policy should boost economic activity (Chen, 2017), which is not a surprising thing to mention, since the financial crisis decreased the economic activity due to a lack of confidence.
According to Bauer and Rudebusch (2015) it is important to divine the path of future (expected) monetary policy, which is primarily obtained from the ‘term structure of interest rates’, foremost during the financial crisis and its aftermath. This term structure of interest rates takes into account the expected path of short-term interest rates, according to financial market participants. And as mentioned before, the short-term interest rate can be decided by the FED. There already existed dynamic short-term structure models (DTSMs) which were used to ‘extract such short-rate expectations’ (Bauer & Rudebusch, 2015), however, these models might not be representative because of the near-zero interest rate policies. Standard DTSMs do not take into account that ‘interest rates are bounded below by zero because negative nominal interest rates would lead to riskless arbitrage opportunities’ (Bauer & Rudebusch, 2015). Because the
nominal interest rates are now nearly zero, these models cannot be used because they lack in having a nonnegativity restriction. Bauer and Rudebusch (2015) therefore use a shadow-rate, which respects the Zero Lower Bound (ZLB). If the shadow rate obtains a positive value, it replaces the observed short rate; otherwise, the short rate is set to a near-zero value. This can be seen in the graph below, as you can see the fitted yield curve does not drop below zero, even if the shadow yield curve does. What can be concluded from this paper is that the information content of the yield curve will be limited by the ZLB because of the minimum value of zero. If this is the case, predicting how long the interest rate will remain near zero will be more importantly influenced by macro variables, because these give you important additional information for predicting future yields (Bauer & Rudebusch, 2015). This might play a role in the decision to
10 borrow money. When the interest rates are expected to rise, the chance of repurchasing shares will increase until the interest rates actually increase.
The graph below is from the article of Bauer and Rudebusch (2015), this graph shows the shadow-rate and the fitted yield curve belonging to this rate.
Despite the near-zero interest rate, there are also some disadvantages of a nearly-ZIRP (Bernanke & Reinhart, 2004). The first disadvantage is that rates on financial instruments which are typically priced below the overnight rate, would decrease to a value of zero, and therefore investors should look for other alternatives. Examples are liquid deposits and shares in money-market mutual funds (Bernanke & Reinhart, 2004). These changes do not only cause the investors to seek for an alternative, but also banks might have to change their capital requirements because liquidity in certain markets might be affected.
The second argument is that consumers and companies might see this nearly-ZIRP as an ineffective way of conducting the monetary policy. This would be a misimpression, as the FED only tries to stimulate the economic activity. Moreover, because the
expectations play an important role in order to make this policy work, communication plays an important role (Bernanke & Reinhart, 2004).
3. Data and Methodology 3.1 The model
In this paragraph the model, data and methodology will be explained. The basis of the model used in this research is the already existing model of Gupta and Singhania (2012). However, this research is aiming at the effect of the nearly-ZIRP, so that should be included in the model as well. As shown by Mundell (1963), the interest rate indeed has got an influence on the dividend pay-out, however he also examines that this is
11 related to the inflation, which therefore should be included in the model as well. This is agreed upon by Khan Khan et al. (2013) who find a negative impact of inflation on the dividend yield, and a positive relation of the interest rate on the dividend yield, which is in line with this thesis’ hypothesis. This means that they expect the dividend to decrease as inflation rate rises, and the dividend to increase as the interest rate rises. Therefore, the dividend will decrease when the interest rate decreases.
This results in the following model:
DividendYieldit= β0 + β1MPBit + β2AGEit + β3ln_MCAPit + β4D/Eit + β5EPSit +
β6ZIRP_Dummyit + β7CPIt + ηIndustryj + θquarter + ε
The dividend yield is the dividend paid out as a percentage of the share price. The dividend yield has to be determined per year-quarter (t), and per company (i), since it differs per company. Year-quarter is the year and the quarter of the year it is in.
The variables EPS and age are both following from Lintner’s (1956) model. Lintner (1956) claims that dividend depends on two factors: ‘current earnings’ and dividends paid out in the previous year, which on its term depends on the earnings of that year and the dividends in the year before. This of course continues until the date of the first
dividend pay-out, which is almost similar to the Initial Public Offering (IPO) date, because since that date the shares of the company are publicly offered. This is the first reason for taking into account the age of the firm, which is in this research calculated by counting the years from the IPO date. The second reason to include the age is that the earnings of mature companies are much more stable than those of growing companies, which might lead to a higher pay-out (Gupta & Singhania, 2012).
The EPS, earnings per share, are the total earnings of a company divided by the total shares outstanding. If this amount is higher, more dividend could be paid out. It is included because it includes the earnings, which is an important measure for the dividend pay-out according to Lintner (1956).
The MPB, Market Price to Book value, is the market value of equity divided by the book value of equity of a firm. This should be included following the pecking order theory of Myers and Majluf (1984). This theory states that companies with high growth
expectations prefer investing with internal funds, however, these companies rather finance with debt than with issuing equity. This results in a lower dividend pay-out. The relation between MPB and high growth expectations is in the fact that companies with high growth expectation will have a higher MPB value, which is caused by this growth.
12 The MCAP, market capitalization, is calculated by multiplying the share price with the total shares outstanding. The data that is used already divided the total shares outstanding by one million, but still the market capitalization is too high to normally adjust in the model. Therefore the natural logarithm (ln) of this market capitalization will be taken. The market capitalization is included because the larger the company, the better the access to capital markets, which makes it less dependent on internal financing (Deshmukh, 2013). For this reason, large companies, which means companies with a large market capitalization, are more likely to pay a high dividend.
The D/E is the Debt to Equity ratio. This is calculated by dividing the total debt by the total equity of the firm, where equity is calculated by subtracting total liabilities from the total assets. Taking into account this variable is following from the leverage argument. This argument claims that firms with a high debt may not be able to pay out dividend, or at least pay out a high dividend, because it has to repay its creditors first.
As told in the beginning of the paragraph, the variables above are from the already existing model of Gupta and Singhania (2012). The following variables are added in this particular research.
The ZIRP_Dummy is a dummy variable used for the zero interest rate policy. This dummy takes a value of 1 if the policy is implemented, and a value of 0 otherwise.
Because this research uses the first quarter of 2002 as the starting date, a value of 0 will be given to this dummy from the first quarter of 2002 until the last quarter of 2008, and a value of 1 for the first quarter of 2008 until the last quarter of 2015.
As Mundell (1963) and Khan Khan et al. (2013) mention, the dividend yield does not only depend on the interest rate, but also on the inflation rate. That is why the CPI, consumer price index, is included, because the CPI is a measure of inflation. The study of Pilotte (2003) mentions that a relation between stock prices and inflation exists, because stock prices and the price level react differently to changes in expected future output. However, Pilotte (2003) also mentions that the effects aren’t either positive or negative, but they might differ depending on the year. So no clear-cut direction can be examined on forehand, but it has an influence and therefore should be included in the model.
The industry variable, industryj, is an industry fixed effect. This means that this
value adopts a value of 1 for the industry the company is in. What is used in this case are the first three digits of the NAICS, North American Industry Classification System. If the company has a NAICS which starts with 311 for example, this means that the company is part of the food manufacturing industry, while 334 stands for the computer and electronic product manufacturing industry. By accounting for the industries, a
13 distinction between the pay-out in certain industries can be made, which might change the results.
Finally a variable called quarter, is included. This variable is also a fixed effect, however, this one controls for the quarter of the year. It has to be included because the interest rate dummy variable otherwise correlates with the variables year or year-quarter, and will therefore be omitted in the regression that is needed for testing panel data. It is also valuable in the way that it can change the results, which would mean that the dividend yield also depends on the time of the year.
3.2 Data
Next, the way of gathering data and the way of testing will be described.
All the data is coming from WRDS. All the variables which were needed are from the CRSP-Compustat merged database, except for the CPI which is from the CRSP US treasury and inflation Indexes. The CRSP-Compustat data is all quarterly compounded, while the CPI is annually compounded. Not all the ratios and variables were directly on the CRSP-Compustat merged database, but had to be calculated by using other variables which were available. The way of calculating the variables used in the model, are
explained in the first part of this paragraph. 3.3 Methodology
Before starting to test, the outliers has to be found and get rid of. The only variables that has to be considered are the ones written as a fraction or as a percentage. This means that the Dividend Yield, MPB, D/E and the EPS has to be considered for the removal of outliers. When the highest and/or lowest values of a specific variable differ a lot from the lowest or highest percentiles, these values has to be excluded. This is done by the winsor2 tool. The outliers will then be replaced by the lower and upper bound, which are the values of the percentiles you want the outliers to be.
Thereafter, the correlation between the variables has to be examined. When the correlation is larger than 0.50, these variables has to be looked into, because it is
possible that these variables are more or less the same, or measure something the same way. It does not mean that highly correlated variables are therefore wrong and should be omitted, because it could be caused by luck, and sometimes it even should be correlated to measure the dependent variable.
What has to be reminded, is that panel data has to be used. With panel data, the data is sorted per company and per date. If it is not sorted in this way, the variables do
14 not specifically match to this company and this date, and this will cause the results to be less specific and sometimes even incorrect.
As told in the introduction, the hypothesis of this research is that I expect the nearly-ZIRP to decrease the dividend pay-out. However, what should be kept in mind, is that it is not easy for companies to cut its dividends, since it gives a bad signal to shareholders (Myers and Majluf, 1984). Therefore, I expect the companies to either invest and pay out the same amount of cash dividend or just buy back its shares. Floyd et al. (2015)
confirms this. They say that pay-out ratios were stable over the past 30 years, however, they say that industrial companies use share repurchases more often to supplement dividend pay-outs. This causes the pay-out ratios to remain stable, because a repurchase can be seen as a pay-out.
Another reason why dividends would decrease, and therefore the repurchases to increase, has to deal with the financial situation. The nearly-ZIRP was implemented after the beginning of the financial crisis in 2008, which caused the firms to have financial difficulties. Since it is not ideal to cut dividends, it might be better to
repurchase shares, which is agreed upon by DeAngelo et al. (2008), who say that firms mostly repurchase stock when firms are likely to have risky earnings streams.
3.4 Testing
To start with, some standard OLS regressions with robust standard errors take place. An OLS regression is the most simple form of running a regression, and it estimates the dependent variable in a linear regression model, using a normal
distribution. Using the robust standard errors will take into account some outliers which cannot be removed, while an OLS regression with homoscedasticity standard errors will remove these. For example the market capitalization of Apple is much higher than those of most of the other companies, and therefore it could affect the OLS regression results, since that takes into account the standard deviation. Because the robustness does not use this, the estimates will be more precise.
The second kind of regressions are Firm fixed effects regressions. These regressions do take into account all the panels, with the xtset command. This means that the data is now sorted per company, and per date. Because it takes into account the company code, which is correlated with the industry, the industry fixed effects cannot be used due to collinearity.
15 Thereafter, an interaction variable will be included: Dum*CPI, which is the nearly-ZIRP dummy multiplied with the CPI. This will be included because of the high correlation between the two variables, and the effect they have on each other. This interaction variable tests whether the inflation has an influence during the nearly-ZIRP. The test used for this will be the firm fixed effects regression, because this is the most exact way of testing panel data.
At last, the rest of the hypothesis will be tested. This means that the average dividend yield and the average repurchase yield, which is calculated by the amount of repurchases times the share price divided by total shares outstanding times 100%, to get it in the same form as the dividend yield, will be graphed and examined. Subsequently, the repurchase yield will also be tested with the firm fixed effects regressions, to see whether the findings are indeed true.
4. Results 4.1 Correlation
In the following table, the correlation between all of the variables can be seen. The higher the correlation, the more the variables depend on each other. The correlation between the Dividend Yield and age is for example -0.0191, which means that an
increase in the age will decrease the dividend yield. However, it does not say anything about the increase or decrease as a number.
The correlation between CPI and the zero interest rate dummy variable is high, however, this is not a reason to omit the CPI variable. This can be concluded from the Mundell-Fleming model, also called the IS-LM-BP model, which is explained by Pilbeam (2013, pp.71-77). Here, both variables are in the LM-curve, which is the money market. Pilbeam (2013, pp.71-77) claims that the interest rate influences the money demand in a way that excess cash will decrease as the interest rates rises, and this will decrease the money demand. Secondly, the inflation influences the money demand as well, because the demand of money will increase as inflation rises, since products are now more
logMark_cap 0.1067 0.0974 0.1899 0.3766 0.1272 0.1418 0.1562 1.0000 MPB -0.1090 -0.0173 -0.0602 0.0059 0.1332 0.0126 1.0000 CPI 0.0800 0.8398 0.2069 0.0237 0.0455 1.0000 DE 0.1976 0.0254 0.0041 0.0012 1.0000 EPS 0.0483 0.0175 0.1415 1.0000 Age -0.0191 0.1931 1.0000 ZIRP_dummy 0.0528 1.0000 DY 1.0000 DY ZIRP_d~y Age EPS DE CPI MPB logMar~p
16 expensive. This means that both variables have an opposite effect on the money demand, and therefore should be included.
4.2 Standard OLS regression
The first regression is a standard OLS regression, with robust standard errors. This regression does not take into account the quarter and industry fixed effects, which can be seen in the last two rows of the variables (both fixed effects are denoted as ‘No’). The second regression does take into account the industry fixed effects, but it does not take into account the quarter fixed effects, while the third regression switches that around. Finally, in the fourth regression, both the quarter and the industry fixed effects are included.
In all the regressions, the dummy variable is negative and significantly different from zero, with an alpha of 1 percent, which is what was predicted. This means that the dividend yield will on average decrease when the zero interest rate policy is
implemented, for example in the first column, the dividend yield will decrease with 0.183% because of the implementation of the nearly-ZIRP. As you can see in the second and fourth column, including the industry fixed effect changes the effect of age and EPS. The variables change from significant at a 1% level, to non-significant at all, except for the variable age in the last model, which is now significant at a 5% level. This means
(1) (2) (3) (4) Dividend Yield Dividend Yield Dividend Yield Dividend Yield ZIRP_dummy -0.183*** -0.207*** -0.148*** -0.170*** (-6.98) (-8.50) (-5.70) (-7.07) Age -0.0176*** 0.000847 -0.0158*** 0.00266** (-18.20) (0.96) (-16.68) (3.08) EPS 0.0602*** -0.00756 0.0631*** -0.00607 (4.24) (-0.54) (4.52) (-0.44) D/E 0.542*** 0.248*** 0.542*** 0.246*** (49.17) (20.89) (50.02) (21.13) CPI 0.00526*** 0.00394*** 0.00474*** 0.00340*** (18.08) (14.71) (16.45) (12.8) MPB -0.0846*** -0.0387*** -0.0832*** -0.0370*** (-48.62) (-23.44) (-48.48) (-22.65) logMark_cap 0.102*** 0.0472*** 0.100*** 0.0455*** (27.86) (12.91) (27.7) (12.6) Constant -2.098*** -1.346*** -2.318*** -1.495*** (-15.92) (-5.03) (-17.71) (-5.94) Firm FE No No No No
Quarter FE No No Yes Yes
Industry FE No Yes No Yes
N 93409 93409 93409 93409
t statistics in parentheses * p<0.05,** p<0.01, *** p<0.001
17 that different types of industries all have a different impact on the dividend yield,
namely through the age of the companies in these industries and through the earnings per share of these companies.
4.2 Firm fixed effects regression
The fixed effects used in these regressions will be the company code and the date. This orders the panels in the right way. In this case, only two types of tests can be used. The first one will be a regression with robust standard errors, and without the inclusion of the quarter fixed effects, while the second test will include these quarter fixed effects. The industry fixed effects cannot be used, since the company code is used as a firm fixed effect, so that will not be considered at all because of the collinearity.
The results can be seen in the table below:
As you can see in the table, the change to the nearly-ZIRP will decrease the dividend yield significantly again, at a 1% level, and therefore the hypothesis is agreed upon. The next thing that can be concluded is that the quarter fixed effects have an influence on the significance of the variable age, however, as you can see from the standard OLS regression, this is caused by the industry fixed effects. Finally, the negative and significant value of the natural logarithm of the market capitalization is actually surprising, because the firms with a large market cap were expected to pay out
(1) (2) Dividend Yield Dividend Yield ZIRP_dummy -0.165*** -0.209*** (-4.40) (-5.55) Age -0.0792*** 0.00193 (-5.06) (0.12) EPS -0.109*** -0.103*** (-6.11) (-5.95) D/E 0.100** 0.0866** (3.26) (2.84) CPI 0.0115*** 0.00457*** (9.02) (3.38) MPB -0.00676 -0.00475 (-1.66) (-1.18) logMark_cap -0.103*** -0.107*** (-4.89) (-5.10) Constant -3.257*** -1.055* (-6.50) (-1.99)
Firm FE Yes Yes
Industry FE No No Quarter FE No Yes N 93409 93409 t statistics in parentheses * p<0.05, **p<0.01, *** p<0.001
18 more dividend, while this negative value means that the higher the market
capitalization is, the lower the dividend yield will be. 4.3 Interaction variable
Next, the firm fixed effects model will be examined again, however now an interaction variable Dum*CPI, the nearly-ZIRP dummy times the CPI, will be included to test whether the inflation has an impact on the dividend yield during the nearly-ZIRP. The model will now look as follows:
DividendYieldit= β0 + β1MPBit + β2AGEit + β3ln_MCAPit + β4D/Eit + β5EPSit +
β6ZIRP_Dummyit + β7CPIt + β8Dum*CPI + ηIndustryj + θquarter + ε
The first regression will be a firm fixed effects regression with robust standard errors. The second regression is the same, however it includes the interaction variable. The third regression will be the same as the first, however the quarter fixed effects will be included. And the last regression also adds the interaction variable to this third regression.
The results can be seen below:
(1) (2) (3) (4) DY DY DY DY ZIRP_dummy -0.165*** 0.970* -0.209*** 3.230*** (-4.40) (1.97) (-5.55) (6.33) Age -0.0792*** -0.0518*** 0.00193 0.0952*** (-5.06) (-3.36) (0.12) (5.7) EPS -0.109*** -0.108*** -0.103*** -0.101*** (-6.11) (-6.06) (-5.95) (-5.82) D/E 0.100** 0.0994** 0.0866** 0.0829** (3.26) (3.22) (2.84) (2.73) CPI 0.0115*** 0.0101*** 0.00457*** -0.0005 (9.02) (8.85) (3.38) (-0.41) MPB -0.00676 -0.00667 -0.00475 -0.0044 (-1.66) (-1.64) (-1.18) (-1.09) logMark_cap -0.103*** -0.102*** -0.107*** -0.105*** (-4.89) (-4.85) (-5.10) (-5.00) CPIxDummy -0.00228* -0.00691*** (-2.31) (-6.75) Constant -3.257*** -2.816*** -1.055* 0.593 (-6.50) (-6.23) (-1.99) (1.24)
Firm FE Yes Yes Yes Yes
Industry FE No No No No
Quarter FE No No Yes Yes
N 93409 93409 93409 93409
adj. R-sq 0.012 0.012 0.065 0.066
t statistics in parentheses * p<0.05, **p<0.01, *** p<0.001
19 The addition of the interaction variable has changed the results, especially when taking into account the quarter fixed effects. Instead of being negative and significant, the change to the zero interest rate policy now increases the dividend yield significantly, at a 1% significance level. Furthermore, it causes the CPI to not being significant
anymore, which means that an increase or decrease in inflation does not change the dividend yield. The age variable changes from negative to positive, while in both
regressions this variable is significant. This means that the quarter fixed effects indeed influence the age to be positive, or the other way round, the industry fixed effects causes the age to have a negative effect on the dividend yield, which is in line with the OLS regression. Finally, the interaction variable turns out to be negative and significant. This means that the interaction between these two variables have a reverse effect on the dividend yield. So if the CPI rises, the change to the nearly-ZIRP reduces the dividend yield, vice versa. However, the adjusted R-squared, the amount of variation of the dependent variable which is estimated by the model, does not change or changes just by 0.001 in the second case. This change is so small, that it is questionable whether the interaction variable adds value, and therefore these regressions will not be used to come to a conclusion.
4.4 Repurchases
In the repurchases part, the final part of the hypothesis will be tested, namely whether the nearly-ZIRP increases the share repurchases.
First of all, the Dividend Yield and the ‘Repurchase Yield’, amount of shares repurchased multiplied with the share price divided by the total amount of shares times 100%, will be graphed and looked upon.
20 This graph shows us the trend of the average dividend yield and that of the average repurchase yield. As you can see, the average dividend yield rises, which means that the dividend yield did not decrease during the zero interest rate period, which is not in line with the hypothesis. Secondly, you can see that the repurchase yield does decrease instead of increase, and again, this is not in line with what was expected.
The second question that arises is if this trend can be supported by the regression model? To find an answer to this, the firm fixed model with and without the interaction variable will be used, however, now the Repurchase yield will be the dependent variable.
21 As you can see from this table, the effect of the interaction variable is not
significant, and it changes the dummy variable from significant to not significant (columns 2 and 4). This means that the interaction between the zero interest rate policy and the inflation does not change the results significantly, and it therefore also causes the zero interest rate policy to have no impact on share repurchases anymore.
When the interaction variable is not taken into account, the change to a nearly-ZIRP reduces the repurchase yield significantly, which means that the repurchase yield decreases when the nearly-ZIRP is implemented. This is in line with the graph showed above, however, it is not in line with the hypothesis, where the repurchases were
expected to increase. In this case less variables are significant compared to the tests with the dividend yield, which means that they might not be the right estimators for the repurchase yield.
The effect of the interaction variable on the adjusted R-squared is again destroyable. This means that it will be better not to include the interaction variable, since it does not add value to the test.
(1) (2) (3) (4) Repurchase Yield Repurchase Yield Repurchase Yield Repurchase Yield ZIRP_dummy -0.439*** -0.220 -0.445*** 0.446 (-12.28) (-0.37) (-12.38) (0.74) Age 0.00227 0.00638 0.0217 0.0404* (0.16) (0.42) (1.41) (2.33) EPS 0.119*** 0.119*** 0.117*** 0.118*** (7.46) (7.43) (7.37) (7.42) D/E 0.0284 0.0281 0.0268 0.0255 (1.09) (1.08) (1.03) (0.98) CPI 0.00317* 0.00309* 0.00132 0.000774 (2.55) (2.57) (0.94) (0.57) MPB -0.00777 -0.00774 -0.00748 -0.00735 (-1.57) (-1.57) (-1.51) (-1.49) logMark_cap -0.0583** -0.0572** -0.0595** -0.0554** (-2.84) (-2.77) (-2.90) (-2.69) CPIxDummy -0.000443 -0.0018 (-0.37) (-1.49) Constant -0.377 -0.379 0.315 0.378 (-0.75) (-0.75) (0.57) (0.69)
Firm FE Yes Yes Yes Yes
Industry FE No No No No
Quarter FE No No Yes Yes
N 40073 40073 40073 40073
adj. R-sq 0.014 0.014 0.015 0.015
t statistics in parentheses * p<0.05, **p<0.01, ***p<0.001
22 5. Conclusion
The purpose of this research is to find an answer to the question whether the dividend pay-out policy of firms from the United States has changed because of the implementation of the zero interest rate policy. The expectation was that the dividend pay-out would decrease, while the repurchases were expected to increase, because it is cheap to borrow money. This also has to deal with the fact that a cut in dividend pay-out gives a bad signal to shareholders, so the dividend will remain the same or the amount of repurchases increases in order to at least give the shareholders the same pay-out.
To come to a conclusion regressions with robust standard errors are used. The first kind of regressions, the pooled OLS regressions with and without time or industry fixed effects are not the ideal test for panel data regressions. However, they still give insight in the effect of certain variables, which would otherwise not be obvious with the final regression. This final regression is the firm fixed effects regression, which takes the company code, and the date as fixed effects.
When testing the firm fixed effects regressions, the same results come up after all sort of tests. You can indeed see that the zero interest rate policy dummy has a negative effect on the dividend yield, which means that the dividend pay-out should decrease due to the switch to the zero interest rate policy. This is in line with the hypothesis of this research.
Secondly, the inclusion of the quarter fixed effects changes the age from having a negative and significant impact on the dividend yield, to a positive and non-significant effect. One can see this as a significant effect of the quarter fixed effects on the age, that on its turn affects the dividend yield. However, what can be seen from the first two kinds of regressions is that it is not due to the quarter fixed effects that the coefficient of age changes from negative to positive, but that the effect of the industry fixed effects changes the coefficient of the age from positive to negative. This thus means that there is a
difference in the age of companies in each industry, and this is related to the dividend pay-out.
Finally, the effect of the market capitalization turns out to be negative instead of positive, this might mean that the firms with large market capitalizations will fund less with equity and therefore pay-out less.
When taking into account the interaction variable Dum*CPI, dummy times CPI, you can see a total different effect of the dummy variable on the dividend yield. However, this interaction variable will eventually not be included because the adjusted R squared
23 changes with 0.001 at max. This is not high enough to change the results this
drastically.
The second problem has been found after graphing the dividend yield and the repurchase yield. This graph shows us that the dividend yield rises, and the repurchase yield decreases, while the opposite was expected.
Finally, the firm fixed effects test was used, but now while using the Repurchase yield instead of the dividend yield. Here, the dummy variable has a negative and
significant effect, which is not in line with the hypothesis, but it is in line with the graph described above. Compared to the regression of the dividend yield, more variables are insignificant. This could mean that other variables has to be examined to estimate the repurchase yield.
For next research, it would be a good thing to find out the variables that influence the repurchase yield, so that it can be better compared to the dividend yield. Further research might also try to examine this model for other countries, because there are several countries which implemented the zero interest rate policy, while this research is restricted to the US. And finally, the reason why the dividend yield does not decrease has to be found and can then be used in an addition to this model.
Reference list:
Barclay, M.J., & Smith, C.W. (1988). Corporate payout policy: cash dividend versus open market repurchases. Journal of financial economics, 22, 61-82.
Bauer, M.D., & Rudebusch, G.D. (2015). Monetary Policy Expectations at the Zero Lower Bound. Federal reserve bank of San Francisco working paper series, 2013(18). Berk, J., & DeMarzo, P. (2014). Corporate Finance (3rd edition). Harlow: Pearson
Education.
Bernanke, B.S., & Reinhart, V.R. (2004). Conducting Monetary Policy at Very Low Short-Term Interest Rates. American Economic Review, 94(2), 85-90.
Brav, A., Graham, J.R., Harvey, C.R., & Michaely, R. (2005). Payout policy in the 21st century. Journal of Financial Economics, 77, 483–528.
24 Chen, H. (2017). The effects of the near-zero interest rate policy in a regime-switching
dynamic stochastic general equilibrium model. Journal of Monetary economics, 90, 176-192.
Chowdhry, B., & Nanda, V. (1994). Repurchase premia as a reason for dividends: A dynamic model of corporate payout policies. Review of Financial Studies, 7, 321– 350.
Copeland, T.E, Weston, J.F., & Shastri, K. (2014). Financial theory and corporate policy (4th edition). Harlow: Pearson Education Limited
DeAngelo, H., DeAngelo, L., & Skinner, D.J. (2008). Corporate payout policy. Foundations and trends in finance, 3(2-3), 95-287.
Deshmukh, S. (2003). Dividend initiations and asymmetric information: A hazard model. Financial Review, 38(3), 351–368.
Elliott, L., & Seager, A. (2008, December 17th). Federal Reserve slashes interest rates to
nearly zero. The Guardian. Retrieved from:
https://www.theguardian.com/business/2008/dec/16/federal-reserve-interest-rates-cut
Floyd, E., Li, N., & Skinner, D.J. (2015). Payout policy through the financial crisis: The growth of repurchases and the resilience of dividends. Journal of Financial Economics, 118(2), 299-316.
Ghosh, C. and J. R. Woolridge (1989). Stock-market reaction to growth-induced dividend cuts: Are investors myopic? Managerial and Decision Economics, 10, 25–35. Gupta, A., & Singhania, M. (2012). Determinants of Corporate Dividend Policy: A Tobit
Model Approach. SAGE publications, 16(3), 153-162.
Khan Khan, M.I., Kashif, S.M., & Mehar Meher, M.A.K. (2013). Impact of Inflation on Dividend Policy: Synchronization of Capital Gain and Interest Rate. MPRA paper, 51593.
25 Lintner, J. (1956). Distribution of income of corporation among dividends, retained
earnings and taxes. American Economic Review, 46(2), 97–113.
Lutz, F.A. (1945). The Interest Rate and Investment in a Dynamic Economy. The American Economic Review, 35(5), 811-830.
Mundell, R. (1963). Inflation and Real Interest. Journal of Political Economy, 71(3), 280-283.
Myers, S. and N. Majluf (1984). Corporate financing and investment decisions when firms have information that investors do not have. Journal of Financial Economics, 12, 187–221.
Oyedele, A. (2015, December 16th). The Fed just raised rates, ending 7 years of crisis-era
monetary policy. Business insider. Retrieved from:
http://www.businessinsider.com/federal-reserve-announcement-december-16-2015-12?international=true&r=US&IR=T
Pilbeam, K. (2013). International Finance (4th edition). Hampshire: Palgrave Macmillan
Pilotte, E.A. (2003). Capital Gains, Dividend Yields, and Expected Inflation. The Journal of Finance, 58, 447-466.
Skinner, D.J. (2008). The evolving relation between earnings, dividends, and stock repurchases. Journal of financial economics, 87, 582-609
Wu, J.C., & Xia, F.D. (2015). Measuring the Macroeconomic Impact of Monetary Policy at the Zero Lower Bound. Chicago Booth Working Paper, 13(77).
