1
Dividend cuts and omissions in the US banking industry
“How did stock markets react on dividend cuts and omissions in the US banking
industry during the financial crisis of 2007 and 2008?”
Author:
N. van Saase
Student number:
10618511
Thesis supervisor: Dhr. J. Lemmen
Finish date:
31 January 2017
University of Amsterdam
Faculty Economics & business
BSC Economie & Bedrijfskunde
2
Statement of Originality
This document is written by Student Niels van Saase who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for contents.
3
Abstract
Dividend theories suggest that markets react strongly negative after banks cut on their dividends. This research uses event study methodology to measure the market reaction around announcements of dividend cuts and omissions in the US banking industry in the crisis period 2007-2008. This research only finds a significantly negative market reaction on the day after the announcement, while prior research expected a much larger negative reaction around the announcement period. Besides, multiple regression methodology is used to explain the market reaction on the day after the announcement. The most significant and stable result of the regression is the negative relation between the decrease in dividend yields and the market reaction on the day after the event, which is a support of the signalling theory.
4
Table of contents
1. Introduction
5
2. Theory and hypotheses
7
2.1 Research question
7
2.2 Dividend irrelevance
7
2.3 Dividend relevance
7
2.4 Agency-cost theory
8
2.5 Fama & French characteristics of dividend payers
9
2.6 Bank characteristics
9
3. Data and method
12
3.1 Event study method
12
3.2 Event study data
14
3.3 Multiple regression method
16
3.4 Multiple regression data
17
4. Results
20
4.1 Event study
20
4.2 Multiple regression
22
5. Discussion
25
5.1 Discussion of the results
25
5.2 Limitations
28
References
29
5
1. Introduction
Dividend relevance, and the information that dividend payments deliver to investors and financial analysts, has been a topic for several decades. Dividend is the portion of a company’s profit or reserves which is distributed to their shareholders, as a return for the risk that these shareholders carry because of their personal investment in the company. Dividend payments are often seen as an indicator of the performance level of a company over a period of time. At the same time, dividend payments are also seen as future performance indicators.
There have been many studies on the market reaction of dividend payments and the signals that companies send to the market by paying dividends. To create a fair view of these effects and to eliminate biased results, the financial industry and especially banks are mainly excluded in these studies. This industry would distort the results because of the high levels of regulation and leverage which make the dividend levels of banks not comparable to the dividend levels of companies from other industries (Slovin et al., 1992).
In 2007, a worldwide economic crisis started with the burst of a bubble on the American housing market. In the period prior to 2007, homeowners had been granted mortgages that they could not afford. Those mortgages were very risky, but the risk of the mortgages had been passed on to other investors by adding them into so called mortgage-backed securities. After the burst of the bubble, the value of those mortgages and the underlying prices of the houses sharply declined and investors in this industry suffered from large losses.
Banks were highly invested in those mortgages and started to develop liquidity problems. One way to improve liquidity was to cut on dividends. However, many banks kept on paying and even increased dividends at the beginning of the crisis period (Acharaya et al., 2011). An explanation of this bank behavior might be found in the signaling theory, where dividend cuts and omissions are a signal of negative future prospects, a signal that banks not want to send to investors and analysts. A few months later, banks started to cut their dividends because liquidity positions worsened even more and
regulatory pressure increased.
This research will examine the market reactions on those dividend cuts and omissions within the crisis period of 2007 and 2008. The research question will be as follows: “How did stock markets react on dividend cuts and omissions in the US banking industry during the financial crisis of 2007 and 2008?”
Event study methodology of MacKinlay (1997) will be used to capture the stock market reaction around announcements of cuts and omissions. This methodology uses expected returns and realized
6 returns to determine “abnormal returns”. After that, the methodology of Bessler & Nohel (1996) will be used to explain the market reaction. This methodology uses multiple regression to determine whether bank characteristics influence the market reaction.
This research contributes to the prior literature due to the fact that it focuses on the banking industry, which is often excluded in dividend studies. Further, it contributes to existing literature because the market reaction is measured in a very exceptional period, a worldwide economic crisis which led to a very instable and turbulent financial environment. At last, this research not only determines the market reactions but also explains the reactions by using multiple regression.
7
2. Theory and hypotheses
This section will give an overview of several dividend theories and characteristics of dividend payers. Furthermore, this section will state hypothesis based on the discussed literature.
2.1 Research question
The aim of this research is to find out how stock markets reacted on announcements of dividend cuts and omissions in the US banking industry during the crisis period in 2007 and 2008. A second goal is to explain the market reaction by looking at several bank characteristics.
The final research question is the following: “How did stock markets react on dividend cuts and omissions in the US banking industry during the financial crisis of 2007 and 2008?”
Where the market reactions are measured in abnormal returns, a method that compares expected returns with the realized returns around a certain economic event (MacKinlay, 1997). Dividend cuts are events where banks decrease their dividend yields. Dividend omissions are more extreme events where banks completely stop their dividend payments. US stock markets that are taken into account are the New York Stock Exchange (NYSE), American Stock Exchange (AMEX) and National Association of Securities Dealers Automated Quotations (NASDAQ).
2.2 Dividend irrelevance
Why companies pay dividends has been debated for several decades. Researchers have different opinions about the relevance of dividends. Modigliani & Miller (1961) mention that dividend policies are irrelevant in perfect capital markets – when taxes and the effects transactions are not present. However, perfect capital markets are not realistic which makes the theory incomplete. Black & Scholes (1974) researched the relation between dividend yield and stock performance. No significant relation was found which supported the dividend irrelevance theory. They also recommended not to solely invest in high dividend stocks because this would reduce the amount of diversification in a portfolio.
2.3 Dividend relevance
As a counterargument, Gordon (1963) and Lintner (1962) argued the relevance of dividends in the “Bird in hand” theory. Investors are generally risk averse and would rather prefer a certain amount of money now than an uncertain amount in the future. There is always uncertainty about future
performance, which is mainly determined by economic circumstances and managerial capabilities, factors that investors are unable to control and hard to forecast.
8 A second theory that underlines the relevance of dividend is the signalling theory. Bhattacharya (1979) and Miller & Rock (1985) argued that companies use dividend as an instrument to signal future prospects. In contrary to Modigliani & Miller (1961), this theory assumes that markets are not perfect. An important market imperfection is the asymmetric information between inside managers and outside investors about future expectations on company performance. Dividend policies can serve as an instrument to signal those prospects to the market in a credible way. According to the theory, an increase in dividend pay-out sends a positive signal to the market and a decrease in dividend pay-out sends a negative signal.
Prior research has been performed on the validity of the signalling theory. Michaely et al. (1996) researched the market reactions after dividend initiations and omissions on the New York Stock Exchange (NYSE) and American Stock Exchange (AMEX). They found a negative cumulative abnormal return (CAR) of -7% on average over a three day (-1, 1) period around omission announcements. This research supports the signalling theory because the market reactions are significantly negative around the announcement. Healy & Palepu (1988) analysed initiations and omissions on the NYSE and AMEX. Over the two day window (-1, 0), they found a positive CAR of 3.9% for the initiation sample and a negative CAR of -9.5% for the omission sample.
Bessler & Nohel (1996) researched US banks that are listed on the NYSE, AMEX and NASDAQ and found a negative CAR of -8.02% over a two-day period (0,1) around the announcement of dividend cuts and omissions. Over the two-week period (-8, 1), a significant CAR of -11.46% was found which could be interpreted as leakages of information prior to the announcement. Poloncheck et al. (1988) also researched the market reaction around dividend omissions in the US banking industry. A significant negative CAR of -7.37% was found over the period (-1, 1) around the announcement.
Above mentioned research shows that the signalling theory holds in the days around the
announcement of dividend cuts and omissions, both for banking and non-banking industries. This leads to the first hypothesis of this research:
Hypothesis 1: abnormal returns are negative around announcements of dividend cuts and omissions.
2.4 Agency-cost theory
The agency-cost theory is about the general problem between a principal and his agents. In the context of this research, the principal refers to shareholders and the agents refer to managers of the company. Shareholders generally want to maximize profits on their investment and it is the task of the managers to fulfil this, within the boundaries of law and regulation (Friedman, 1970). However, managers also
9 have their own interests which can be different. A manager may want to build a large empire and to fulfil this he takes over other companies, even if this is not leading to extra profits for the company.
Jensen (1986) proposes that the occurrence of agency problems is strongly related to the amount of free-cash flows. If the free-cash flows are large, managers have more possibilities to perform activities that are in line with their own interests, and not with the shareholder’s interests. A company can undertake two possible actions to prevent the self-fulfilling behaviour of managers: the first option is to increase monitoring activities. However, this will create costs. The second option is to reduce the amount of free-cash flow, so managers have room to perform self-fulfilling activities (Easterbrook, 1984).
2.5 Fama & French characteristics of dividend payers
Fama & French (2001) argued that dividend policies are mainly determined by several characteristics. The following characteristics were proposed: size, profitability and investment opportunities. They found that large companies with high profitability and low investment opportunities pay higher dividend amounts. Those companies can afford to pay high dividend levels since the cash is not needed for new investments.
2.6 Bank characteristics
Characteristics of banks might influence the market reactions around announcements of dividend cuts and omissions. Bessler & Nohel (1996) used multiple regression methodology to examine how bank characteristics are related to the market reactions. They used the following three characteristics: decrease in dividend yield, capital adequacy and asset size. Those characteristics will be discussed in this section.
The first characteristic is the size of the dividend cut. Bessler & Nohel (1996) expected a negative relation between the size of a dividend cut and the market reaction. A larger dividend cut would send a stronger negative sign to the market. The multiple regression results confirmed the expectation of the size effect. Ghosh & Woolridge (1988) found the same results in their research on the US banking industry between 1962-1984. This leads to the second hypothesis:
Hypothesis 2: the size of a dividend cut is negatively related to the market reaction.
The second characteristic is the capital adequacy of a bank which measures how much capital a bank has in relation to the total assets outstanding. This measurement is used as a predictor for default risk. In which case a higher level of capital adequacy reduces the level of default risk. Bessler & Nohel
10 (1996) expected a positive relation between capital adequacy and the market reaction because
investors would be less worried when a financial stable bank reduces its dividend level. However, the results moved in the opposite direction. So, banks with higher safety levels received a stronger negative market reaction. They argued that this could be the result of the predictability of dividends cuts in financial unstable banks. If the market already predicted a cut in dividends, this information is already priced in the stock and will not result in a large abnormal return on the announcement date.
Different measurements for capital adequacy exists. The Basle agreement divides the total amount of capital in Tier1 and Tier2, where Tier1 only includes common stock and retained earnings and Tier2 includes the more unstable equity parts like undisclosed reserves and revaluation reserves (Basle, 1988). Bessler & Nohel (1996) use a ratio that includes both Tier1 and Tier2 but Buehler et al. (2009) argues that it is better to solely focus on Tier1 because this equity is more stable and is a better predictor of financial distress.
This research therefore uses the Tier1 core capital ratio to the model in order to have the best predictor for financial distress. Despite the obtained results of Bessler & Nohel (1996), the expectation is that less capitalized banks face a larger negative reaction on the stock market. Mainly because this research solely takes place in the crisis period of 2007 and 2008, where the fear for financial distress was substantial in the banking industry. This leads to the third hypothesis:
Hypothesis 3: the Tier1 core capital ratio is positively related to the market reaction.
The third characteristic is the asset size. Asset size can influence the market reaction in different ways. Large banks can be so important for the overall financial stability of the market that they are too big to fail. Bessler & Nohel (1996) argue that “too big to fail” banks therefore send more information to the market when they announce to their dividend levels. It sends a signal to the market that it might be the last possibility to strengthen the financial position. On the other hand, prior research suggests that large companies are strictly followed by the media and financial analysts which results in information leakage prior to the event. If this is the case, the stock market already priced in the information in the stock prices which results in a lower negative reaction on the announcement day. However, the whole financial industry was strictly followed by the media in 2007 and 2008 so it is more likely that the argument about “too big to fail” banks holds. This leads to the fourth hypothesis:
Hypothesis 4: the size of bank is negatively related to the market reaction
Lastly, it is interesting to look at the relative undercapitalization of banks. Investors often determine their investment decisions by comparing different companies in the same industry. This also holds for investments in the banking industry. In this industry, ratios of the financial position give a good overview of the relative performance. In this research, the relative undercapitalization of a bank might
11 have influence on the behaviour of investors around announcements of omissions and cuts in
dividends. It is expected that investors react more negative on banks that are relatively undercapitalized, compared to their peers. This leads to the fifth hypothesis:
12
3. Data and method
Overall, this research uses event study methodology to measure the market reactions around
announcements of dividend cuts and omissions in the US banking industry over the period 2007 and 2008. Event study methodology of MacKinlay (1997) is used to determine the reaction of the market. After that, multiple regression is used to explain the market reactions around the events.
3.1 Event study method
The event study methodology has been developed by Fama et. al (1969) and prior research used this methodology frequently. The methodology is mainly used to measure the stock market reaction in the period around a specific economic event, such as a merger or acquisition or interest announcement, but also for the announcement of dividend payout. This research uses the methodology that is proposed by MacKinlay (1997).
The first step in the event study methodology is the determination of the economic event. In this research the events of interest are announcements of dividend cuts and omissions. Only US commercial banks and savings banks are taken into account over the years 2007 and 2008. Those banks should be listed on one of the three following exchanges: American Stock Exchange (AMEX, New York Stock Exchange (NYSE) or on the National Association of Securities Dealers Automated Quotations (NASDAQ).
The second step is to determine the event window. This window is the period of time over which the market reaction is measured. This research applies several event windows in order to see leakages of information in advance or slower reactions on the market after the event (Haleblian et al., 1999). Bessler & Nohel (1996) used the windows (0, 1) and (-8, 1). The first window consists of the event day and the day after the event and the second window consists of 8 days before and 1 day after the event. Most researchers analyzed the following windows in their research: (-1, 0), (0, 1) and (-1, 1) (Ryan et al., 2000; Poloncheck, 1989). This research analyzes the following windows: (-1, 0), (0, 1), (-1, 1), (-3, 3), (-5, 5). The latter two windows are added to check for leakages before or delays in the market reaction after the event.
Thirdly, the estimation window should be defined. This window is needed to estimate expected returns. This window runs from T1 to T2 (T1, T2), where both days never overlap with the event day to prevent correlation between expected returns and the event. The abnormal returns would be biased if the event window and estimation window overlap with each other.
13 Bessler & Nohel (1996) used the estimation window (-110, -11), which is a period of 100 days that stops 11 days before the event. Ryan et al. (2000) used a wider window that ran from (-261, -82). However, this window is not preferable because the expected returns will be estimated over a time period that end 82 days before the event. This might bias the results of this study because returns change quickly in the crisis period. Therefore, the window of (-110, -11) is used.
The next step is determining the abnormal returns. In formula form:
ARit = Rit − E(Rit|Xt) (1)
Where ARit is abnormal return for stock i on day t, Rit is the return for stock i on day t and E(Rit|Xt) is the return that is expected when there had been no event.
Expected returns are determined by the market model. This model assumes a linear relation between stocks and a certain benchmark, in this case the market index. The CRSP equal weighted market index will be used as a benchmark. The market model uses Ordinary Least Squares (OLS) over the
estimation period to determine the linear relation between stock prices and the CRSP index. A second possibility is the use of a constant mean return, which assumes the return to be constant over time. However, this estimation method is less precise and results in higher variance levels which is less desirable (Brown & Warner, 1985).
The market model has the following formula form:
Rit = ai + BiRmt + εi (2)
(εit) = 0 and Var(εit) = σ2ε (3)
Where ai and Bi are OLS estimates, Rit is return of stock i on day t and Rmt is return on the market and εit is the error term with a mean value of zero. This research uses the Equally Weighted market index as a proxy for the market.
The return information of this index is extracted from CRSP. The information of the abnormal returns can be used to determine Cumulative Abnormal Returns (CAR). This is a measurement of abnormal returns over a certain period. CAR in formula form:
14 Before analyzing the CARs and ARs, it is important to test for overall significance. A cross-sectional t-test on the average CAR (CAAR) and average AR (AAR) must be conducted to check for
significance over the event periods. Cross sectional t-test for the CAR:
𝑡 = CAAR(t1,t2)sCAAR(t1,t2) √n
(5)
Where CAAR(t1,t2) is the average of the CARs over the event window, sCAAR is the standard deviation of the average CARs over the event window and √n is the square root of the number of observations. Cross sectional t-test for the AR:
𝑡 = sAARtAARt √n
(6)
Where AARt is the average abnormal return on event day t, sAARt is the standard deviation of the average ARs and √n is the square root of the number of observations.
3.2 Event study data
In order to conduct an event study, several data items are needed. This section describes what data is required and how the data is treated.
This research only considers commercial and saving banks that are listed on the NYSE, AMEX and NASDAQ. Security Industrial Classification codes (SIC) are used to filter the banks from the rest of the dataset. SIC codes that start with 602 and 603 are respectively commercial and saving banks. Important data for the event study are daily stock prices and announcements of dividends. This data is extracted from CRSP, a database that is included in the Wharton Research Data Center (WRDS). This database has also been used by other researchers (Bessler & Nohel, 1996; Healy & Palepu, 1988; Poloncheck et al., 1989). This research only considers announcements of quarterly dividends.
The procedure for dividend cuts and dividend omissions is different. First, the criteria of dividend cuts will be discussed. This research uses the following definition for a dividend cut: A bank decreases the amount of dividends after paying a stable stream of dividends for at least 4 years in a row. Michaely et al. (1995) restricted to only 3 years but Healy & Palepu (1988) restricted to even 10 years, which would reduce the sample of our research too much.
15 A dividend cut is added to the dataset only if:
- The cut is not a result of stock splits - Stock price information is available
- Accounting information is available about the bank characteristics which are needed for the multiple regression model
- A bank paid quarterly dividend
Before the mentioned restrictions, the sample consisted of 96 announcements of dividend cuts but afterwards only 71 announcements are left.
This research uses the following definition for a dividend omission: A banks stops the payment of dividends after paying a stable stream of dividends for at least 4 years in a row.
However, omissions are harder to define because no announcement can be found in the CRSP database. The database only includes information about payments and not about omissions. So the following steps are undertaken to determine an omission:
1. Determine banks that stopped paying dividends
2. Count further to the date that the bank normally announces its dividends and take this date as the event date
3. Search online newspapers and annual reports for announcements of the omission
A dividend omission is only added to the dataset if:
- The omission is announced in an article or annual report - Stock price information is available
- Accounting information is available about the bank characteristics which are needed for the multiple regression model
- A bank paid quarterly dividends
Before the above mentioned restrictions, the sample consisted of 28 announcements of omissions but afterwards only 20 announcements are left. So, the final sample consists of 20 dividend omission announcements and 71 dividend cut announcements.
16
Table 1: Distribution of dividends cuts and omissions over the years 2007 and 2008.
Year Number of dividend omissions Number of dividend cuts Percentage of sample 2007 0 8 8.6% 2008 20 63 91.4%
This table shows when the cuts and omissions took place. The most cuts and omission took place in 2008. In 2007, no bank omitted dividends. These findings are in line with the paper of Acharya et al., (2011). This research mentioned that banks started to cut their dividends later in the crisis period.
It is important to mention that the data of the abnormal returns has been winsorized. Some data points showed very large positive and negative abnormal returns that would definitely bias the results of the research. By observing the dataset, it became clear that the largest part of the return was between -15% and +15%. It is of course possible to observe larger returns. Therefore, the choice has been made to replace returns that exceeded the level of 25%, by 25%. For example, on event day -5, a bank showed a positive abnormal return of 69.5%. This returns have been replaced by a positive return of 25%. Over the whole sample, 10 returns have been winsorized. However, winsorizing has also its
drawbacks. According to Sorokina et al. (2013), winsorizing can remove important information out of the dataset. However, this dataset only contains 91 observations. So, large outliers would bias the results too much.
3.3 Multiple regression method
In the prior section, it is explained how the market reaction around announcements dividend cuts and omissions is measured. Besides, this research investigates whether bank characteristics influence the market reaction around the event. Prior research used multiple regression models to determine which characteristics influence the market reaction (Ghosh & Woolridge, 1988; Bessler & Nohel, 1996; Bashir et al., 2013).
This research uses the same methodology to explain the abnormal returns around announcements of dividend cuts and omissions. The multiple regression model has the following form:
𝑌𝑖 = 𝛼𝑖 + 𝛽1 ∗ 𝑋1 + 𝛽2 ∗ 𝑋2 … 𝛽𝑖 ∗ 𝑋𝑖 + 𝜀𝑖 (7)
Where 𝑌𝑖 is the dependent variable, 𝛼𝑖 is the constant term, 𝛽1 to 𝛽𝑖 are the model regression coefficients, 𝑋1 to 𝑋𝑖 are the explanatory variables and 𝜀𝑖 is the error term of the model.
17 Robust standard errors are used in Stata because it is assumed that the standard errors are not
homoscedastic. Before regressing the models, the variables are checked on multicollinearity with the option VIF in Stata (See Appendix table 4-8)
In this research, the dependent variable will be the market reaction measured in abnormal returns (AR) or cumulative abnormal returns (CAR). Which event window of the CAR is used in the regression depends on the significance of the windows.
3.4 Multiple regression data
Several data is needed for the regression model. As mentioned before, the multiple regression model uses the event study results as input for the dependent variable. So, the cumulative abnormal returns (AR) and cumulative abnormal returns (CAR) serve as input for the dependent variable.
The following bank characteristics are used as independent variables: decrease in dividend yield, capital adequacy and asset size. These data is extracted from Compustat, a database that is available in the Wharton Research Data Center (WRDS).
Descriptions of the independent variables and the expectation of the relation with the dependent variable will follow now. The first variable is the decrease in dividend yield (DY). This variable is calculated by taking the difference between dividends in the current and prior period and divide this difference by the stock price at the announcement date. If the signalling theory holds, decreases in dividend yields signal a negative message to the market. It is expected that a larger cut in dividends sends a stronger negative message.
The second variable is a measure of capital adequacy. Input for the level of capital adequacy comes from the Tier 1 core capital ratio (Tier1). This ratio calculates the amount of core capital (retained earnings + common stock) and divides this by the amount of risk-weighted assets outstanding. It is assumed that a lower Tier 1 ratio increases the level of default risk. Investors dislike this risk and the expectation is that the market reaction will be stronger negative after a bank with a low Tier1 ratio announces to cut on its dividends.
The third variable is a dummy that determines whether a bank is undercapitalized (Undercap). In this research, a bank is undercapitalized when it has a lower capital ratio than the median Tier1 of the sample. The median level of Tier1 in the sample is 9.38. The expectation is that undercapitalized banks face a stronger negative market reaction.
18 The fourth measures the size of a bank. This variable is calculated by taking the logarithm of assets (Logassets). It is expected that the market reaction is more negative after large banks announce to cut their dividends. The expectation follows from the thought that large banks can be too big to fail and reveal more information when they announce a cut in dividends.
The table below gives an overview of the independent variables that are used in the regression model. It gives a description of the variables and displays the expected relation with the dependent variable.
Table 2: description of independent variables
This table shows the independent variables that are used in the multiple regression models and describes those variables in the second column . In the third column, the expected relation with the market reaction (measured in abnormal returns) is provided. Dividend yield decrease (DY) reflects the size of the dividend cut. The signalling theory predicts a negative relation between DY and the market reaction. Tier 1 (Tier1) is a ratio that tells how much core capital a banks has in relation to the total amount of risk weighted assets outstanding. Banks with a lower Tier 1 ratio are more vulnerable for shocks in the economy because they have little capital to absorb losses. The third variable measures whether a bank is undercapitalized (Undercap). The last variable measures the bank size by taking the logarithm of the total assets (Logassets).The expectation is that larger banks provide more information to the market when they cut their dividend levels.
Variable & name Description Expected sign of the variable
Dividend yield decrease (DY)
Decrease in dividend/stock price at the moment of the announcement
-, Signalling theory
Tier 1 (Tier1)
Core Tier 1 ratio in the quarter before the announcement
+, Default risk
Undercapitalized (Undercap)
Dummy variable that is 1 if a bank has a lower Tier 1 ratio than the median level of the sample. Measured over the quarter before the
announcement
-, Default risk
-, Too big to fail -, signalling theory Size (Logassets) Natural logarithm of total
19 A second relevant table gives an overview of the descriptive statistics of the independent variables. This table follows below.
Table 3: Descriptive statistics
Variable N Min. Max. Mean Std. Deviation median
Arday1 91 -0.278 0.152 -0.0215 0.074 -0.01 Tier1 91 6.530 17.02 9.4554 1.907 9.38 Undercap 91 0 1 0.4945 0.503 0 Assets 91 259.920 812433 29215.9 96288.76 2193 Logassets 91 5.560 13.607 8.2723 1.809 7.69 DY 91 0.0002 0.044 0.0107 0.008 0.008
This table gives an overview of the descriptive statistics of the dependent and independent variables. Where N is the number of observations, which is equal for all variables. Secondly, statistics of the minimum and maximum levels are provided. For banks, the asset size is presented in millions dollars. The fourth column shows the spread of the data, measured in the standard deviation. In the last column, the median levels of the sample are stated.
The table above shows that the total number of descriptive statistics of every characteristic equals the total number of 91 dividend announcements. The safest bank in the sample has almost the triple amount of Tier 1 core capital in relation to the risk-weighted assets outstanding. Further, it becomes obvious that the spread between the largest and smallest bank is very big. Lastly, the average decrease in dividend yield is a little more than 1%. However, the largest cut in dividend yield is even 4%.
An overview of the correlations between dependent and independent variables is present in the Appendix (See Appendix table 2).
20
4. Results
This section discusses the results. The first part is about the event study analysis and the second discusses the multiple regression.
4.1 Event study
Event study methodology is used to determine the abnormal market reaction around announcements of dividend cuts and omissions in the US banking industry. The abnormal market reactions are
determined by the market model, which calculates expected returns by looking at the relation between stock prices and market indexes in the near past. This research used an estimation period of 100 days which runs from 110 days to 10 days before the actual event (-110, -11). The event period runs from 5 days before to 5 days after the actual event (-5, 5). Stock prices and market indexes are subtracted from CRSP. The final sample of the research consists of 91 announcements of which 71 are cuts and 20 are omissions in dividends
After testing the cumulative average abnormal returns (CAAR) over the event windows, no significant negative were found for any of the event windows. First, the widest 11 day event window (-5, 5) has been tested. Over this window, the CAAR is -0.68% which is insignificantly different from zero (p>0.10). The second event window (-3, 3) has CAAR of -2.59% which is insignificantly different from zero (p>0.10). For the third window (-1, 1), a CAAR of -1.30% has been found which is also not significantly different from zero (p>0.10). The event window (-1, 0) even shows a positive CAAR of 0.85%, however not significant (p>0.10). The event window (0, 1) shows a CAAR of 1.74% which is significant at the 10% level (p<0.10). However, this period is only significant at the 10% level because of the significant negative abnormal returns at event day 1. The average abnormal returns (AAR) on day 0 are even positive. The average negative return on event day 1 is -2.15%, which is significant at the 1% level (p<0.01). The delay of the market reaction could be related with the moment of the day that banks announce the cuts and omissions. It could even be the case that the announcements occur after the stock markets closed. The table below shows the CAARs and the levels of significance for the above mentioned results.
21
Table 4: CAAR and significance levels
(t1,t2) CAAR Std. dev. t-statistic p-value N (-5,5) -0.68 .2310 -0.2827 0.7781 91 (-3,3) -2.60 .1667 -1.4829 0.1416 91 (-1,1) -1.30 .1125 -1.0996 0.2744 91 (-1,0) 0.85 .0855 0.9505 0.3444 91 (-0,1) -1.73 .0940 -1.7635* 0.0812 91
This table shows the Cumulative Average Abnormal returns (CAAR) and its level of significance over several event windows that are tested in this research. The interpretation of this table is as following. The first column (t1,t2) stands for the length of the event period. CAAR stands for the Cumulative Average Abnormal return. The Std. dev. is a measurement of the spread of the CAAR around the average.. The t-statistic and p-value show the level of significance. Where *, **, *** mean respectively 10%, 5% and 1% significance. N displays the number of observations. This table shows that the CAAR is only significant at the event period (0, 1). However, this window is only significant because the Average Abnormal Return (AAR) is significant at event day 1. The AAR on day 0 even shows a positive AAR.
The figure below shows the average abnormal market reaction over the event period. The returns are positive until event day -2. Then the return becomes negative on day -2. However, it increases after that day until event day 1. On that day, the returns are significantly negative. But after that day, the returns do not show any significant results.
Figure 1: Average abnormal returns: omissions and decreases
This figure shows the average abnormal returns (AAR) over the largest event window that runs from event day -5 to event day +5 (-5, 5).On the vertical axis, the level of AAR is displayed. The horizontal axis displays the event date. The AAR is only negative on event days -2, +1 and 3. The returns are only significant on event day +1 (p<0.01).
22
4.2 Multiple regression
This research used OLS multiple regression methodology to express the relation between the abnormal returns around the events (measured in AR and CAR) and the explanatory variables Dividend yield decrease, Tier 1, Undercapitalized and asset size.
ARday 1
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Tier1 - 0.0063* - 0.0086*** -0.0058 (0.0034) (0.0032) (0.0035) Undercap 0.0152 (0.0148) Independent variables Logassets 0.0064 0.0087 0.0099** (0.0041) (0.0057) (0.0047) DY -1.7175 -2.1685 - 2.7029** -2.7725** (1.3609) (1.3243) (1.2862) (1.318) Constant 0.0377 -0.0743** -0.0031 0.0833** -0.0104 -0.0815** (0.0346) (0.0347) (0.0135) (0.0336) (0.0616) (0.0367) N 91 91 91 91 91 91 F-Statistic 3.31 2.4 1.59 4.97 3.97 2.6 Sig. 0.0722 0.1251 0.2102 0.009 0.0105 0.0571 R-Square 0.0263 0.0245 0.0414 0.0884 0.1231 0.1136 Adj. R-Square 0.0153 0.0136 0.0306 0.0677 0.0929 0.083
This table shows 6 regressions with the abnormal return on day 1 (ARday1) as the dependent variable and independent variables about capital adequacy (Tier1), undercapitalization (Undercap), size (Logassets) and decrease in dividend yield (DY). N displays the number of observations, F-statistic displays the outcome of the model F-test, Sig. is the level of significance according to the model F-test and R-square tells how much the independent variables explain the dependent variable. All models use robust standard errors to correct for heteroskedasticity in the standard errors. Lastly, *, **,*** are significance levels 10%, 5% and 1%.
Model 1 shows the relation between the abnormal return on event day 1 (ARday1) and the independent variable Tier 1. This day is relevant because it shows a significant negative market reaction after the announcement day. Tier 1 core capital is an important bank characteristic because
23 this capital is needed to absorb losses in crisis periods. Other capital ratios perform worse in absorbing losses and are less precise in determining the risk of default. Therefore, a positive relation was
expected between Tier 1 and Arday 1. However, the regression shows a negative significant relation between the variables (P<0.10). So, the market reacts less negative on riskier banks that cut their dividend levels. The R2 of the model is 0.0263, which is not very high but also not rare for a model with only one explanatory variable.
Model 2 shows the relation between Arday 1 and the logarithm of assets (Logassets). The expectation was to find a negative relation between large banks Arday 1 because more information is revealed when a large bank announces to cut its dividends. The regression results show a contrary relation between Logassets and Arday 1, which means that market react less negative on large banks that cut their dividend levels. However, the relation is not significant at the 10 % level so no strong
conclusions can be drawn on this outcome. The only significant coefficient is the constant term (P<0.05) The R2 is 0.0245, which means that the variable Logassets explains 2.45% of the variance in the abnormal returns on day 1.
The third model explains the relation between ARday1 and the decrease in dividend yield (DY). A negative relation has been found between DY and ARday1, which was also expected based on the signalling theory. So, when the decrease in yield is stronger, ARday1 is more negative on average. However, the coefficient of DY is insignificant (p>0.10). The R2 in model 3 is 0.0414 which means that the variable DY is the best variable in explaining the variance in ARday1.
In the fourth model, the variables Tier1 and DY are added to explain the dependent variable ARday1. The signs of the coefficients remain the same as in model 1 and 3. Tier 1 shows even larger
significance in the unexpected way (p<0.01) and DY remains insignificantly different from zero. Besides, the constant term is significantly different from zero (p<0.05). Based on the model F-test, this model has the highest level of significance. (p<0.01)
In the fifth model, the variables from the models 1,2 and 3 are added to the model. Tier1 is no longer significant in this model (p>0.10). The variable DY becomes significant at the 5% level and the variable Logassets remains insignificant. This model has the highest predictive power with an R2 and adjusted R2 of respectively 0.1231 and 0.0929.
In the last model, the variable Tier 1 is replaced by a dummy which measures whether a bank is undercapitalized (Undercap) or not. The variables Tier 1 and Undercap are not included in the same regression because they have a correlation of -0.709, which is unacceptable. A bank is
24 variable Undercap shows an insignificant positive relation with ARday 1 (p>0.10), which means that banks with higher than median default risk receive a larger negative market reaction on the
announcement day but it is insignificant so it does not show anything. This is not what was expected according to the default risk theory. Both Logassets and the constant term become significant in this model (p<0.05) but the level of significance for the full model lowers in comparison to model 5.
25
Discussion
This section discusses the results an provides a conclusion. Further, the limitations of this research will be discussed.
5.1 Discussion of the results
Research on dividends and its impact on shareholder value exists for several decades now. However, most of this research focused on all industries except the financial industry because this industry would bias the results (Slovin et al., 1992). During and after the financial crisis in 2007, research on the dividend policies in the financial industry became more popular.
This research determines the abnormal market reaction around announcements of dividend cuts and omissions and attempts to explain this reaction by using multiple regression methodology.
First , throughout this research the market reaction towards dividend cuts and omissions has been measured by testing the CAAR on significance around the announcement of these cuts and omissions in several event windows. The only significant returns are observed over the two-day period (0, 1) and event day 1. The two-day event window (0, 1) is significant (p<0.10), because of the negative
abnormal returns on event day 1 which are -2.15% on average (P<0.01). The returns on event day 0 within the two-day event window (0, 1) show actually positive returns instead of negative returns, but these positive returns are compensated by the stronger negative returns on day 1.
The difference in market reaction between day 0 and 1 could be due to a slow reaction of investors or due to banks that announce dividend cuts and omissions late on a trading day. If banks do this, the reaction on the stock market will simply be delayed to the next trading day.
Overall, the market reaction is less intense than prior research expected. This could have several causes. The first one is that investors in banks prefer a stable capital position over the payment of dividends in a period of financial crisis. A direct positive effect of a cut in dividends is namely a strengthening of the capital position, because money is kept within the firm instead of distributed to shareholders .
Other researchers did not solely focus on crisis periods which might explain part of the weaker market reaction. A second explanation could be that investors already expected that banks stopped the
payment of dividends. If this is the case, the market reaction will not be that strong on the day of the actual announcement and the research method will not detect an abnormal reaction.
26 A third explanation could be found in the determination of abnormal returns, where expected returns are estimated over a 100-day period prior to the event. In this period, the returns on stock markets were already declining because of bad economic and financial prospects . Because of these declines in share prices, the share prices of banks had to decline even stronger in order to become “abnormal” compared to the expected return. In the end, this declining trend made it ‘tougher’ for bank shares to decline in an abnormal way.
A last argument for the weaker market reaction is that many banks cut their dividends in the same time window, which makes a dividend cut or omission a less special event. Besides, it is plausible that investors and analysts expect dividend cuts and omissions at a bank when comparable banks also undertake these actions.
In the second part of the research, it has been attempted to find a relation between several bank characteristics and the stock market reaction around the announcement of cuts and omissions. The only significant reaction of abnormal returns was found at event day 1 so the regression takes the abnormal return (ARday1) at this day as the dependent variable.
The multiple regression results do not show the results that were expected. The expectation was to find a positive relation between the safeness of a bank – measured in Tier 1 core capital ratio – and the market reaction. However, a negative relation was found (P<0.10) between the safety ratio and the market reaction on day 1. Bessler & Nohel (1996) obtained the same results and concluded that the surprise is larger when a safe bank announces a cut in dividends. So, the larger surprise may result in larger negative abnormal returns around the event.
Further, it was expected to find a negative relation between asset size – measured in log(Assets) – and the market reaction. Only in model 6, the variable shows a significant coefficient (p<0.05). However, the sign is positive and the expectation was to find a negative relation. The argument for this was mainly based on the assumption that banks that are “too big to fail” provide a stronger negative signal to the market when it announces a cut in the level of dividends. Apparently, investors are less worried when a large bank announces to cut dividends. A possible explanation could be the safety net that a government provides when a bank is close to bankruptcy. Despite the negative signal of a cut, investors and analysts trust that the large banks will survive the tough financial environment, and otherwise a government will not allow a big bank to fail, which is why they react less negative around announcements of those banks
The third expectation was to find a negative relation between the decrease in dividend yield and the abnormal market reaction. According to the signalling theory, larger decreases send stronger signals to
27 the market. This hypothesis is supported by the regression models 5 and 6. The variable decrease in yield (DY) is negative and significant at the 5% level (p<0.05) in this models. However, DY becomes insignificant different from zero when it is added to model 3 where it is the only explanatory factor in the determination for the abnormal market reaction. So, the signalling theory partly holds in the regressions.
Lastly, it was hypothesized that relatively undercapitalized banks would show larger negative returns. The regression results shows a contrary relation between undercapitalized banks and the market reaction. However, model 6 shows that this relation is not significant (p<0.10).
This leads to the final conclusion of this research. First of all, the abnormal market reaction around a dividend cut or omission is on average less negative than expected. No significant event windows are found with the event study methodology, and the only negative significant abnormal returns are found on event day 1. On the other hand, this research confirms, just like existing literature, that there is a negative relation between dividend cuts and omissions and the market reaction towards these events. An important factor to take into account in this research is the time period, the global financial crisis of 2008, in which the announcements took place and the market reactions have been measured.
First of all, the market returns were relatively low at that time (2007-2008), because of bad financial and economic prospects. So, reactions on dividend announcements should be even more negative than the expected returns already were in order to become abnormal.
Furthermore, standard errors were higher in this period due to the large volatility of share prices in the research period. This results in abnormal returns that are less significant than they would be in
“normal periods”.
Besides, it can be concluded that only one hypothesis about the bank characteristics holds. Namely, the negative relation between the magnitude of the dividend yield cut (DY) and the abnormal market reaction. However, the regression model only takes into account the market reaction on event day 1. This makes it hard to draw strong and robust conclusions about the relation between bank
characteristics and the market reaction around the announcement.
In short, this research does not contribute to existing research in a way it indicates new and significant relations. On the other hand, the research is carried out in a very extraordinary period, one of the biggest global financial crises with a massive financial impact, and shows us what the impact of a certain crises is on the market reaction towards dividend cuts and omissions – which provides us with new insights.
28
5.2 Limitations
First of all, the number of observations are limited. The total sample consists of 91 which are extracted from the NYSE, AMEX, and NASDAQ. The number of observations is low for several reasons: the relatively short time period (2007-2008), the selection of only the banking industry, and thirdly because the research only takes into account American banks. The focus on only the US banking industry also makes that this research cannot be generalized to other banks over the world. Of course, banks over the world have large similarities, but the business climate differs over the continents which might have an impact on the way investors react to these kind of events.
Secondly, the amount of variables that are taken into account is limited. This research used the variables that were proposed by Bessler & Nohel (1996). However, more variables could be added the model such as: more capital ratios, future growth factors etc. When more variables are added to the model, the model is able to explain more variance in the market reaction around the events.
The third limitation is the time period of this research. The crisis period 2007-2008 is chosen because this is the period where many banks had to cut their dividend payments. The period would also provide new insights about the market reaction around dividend announcements. However, the choice of the original time period is also a limitation. Financial markets react differently on dividend
announcements in crisis periods, so the outcomes are only generalizable to other comparable crisis periods, and not to ‘normal’ economic and financial environments.
The fourth limitation is the use of an equally-weighted market index. It would be interesting to see whether the regression results change when a value-weighted index is used. This index takes into account the difference between small and big banks, which is very large in this research.
The fifth limitation lies in the nature of event studies. An event study measures the difference between the expected market reaction and the realized market reaction around a certain event. However, it is very hard to determine the exact event date in a study that measures the market reaction on
announcements of dividend cuts and omissions. This study used the announcement days in CRSP for the determination of the event but it is possible that the information was already leaked in front of this event. Even more tricky is the determination of omission events. Those announcements are not present in the CRSP database, so they are determined by analysing the announcement pattern in prior periods and use this information for the expectation of future announcements.
29
References
Acharya, V., Gujral, I., Kulkarni, N., Shin, S.H., 2011, Dividends and Bank Capital in the Financial Crisis of 2007-2009, NBER Working Paper, No. 16896
Basle, 1988, International Convergence of Capital Measurement and Capital Standards, pp 1-30 Bessler, W., Nohel, T., 1996, The Stock-Market Reaction to Dividend Cuts and Omissions by Commercial Banks, Journal of Banking & Finance, Vol. 20, No. 9, pp 1485-1508
Black, F., Scholes, M., 1974, The Effects of Dividend Yield and Dividend Policy on Common Stock Prices and Returns, Journal of Financial Economics, Vol. 1, No. 1, pp 1-22
Bhattacharya, S., 1979, Imperfect Information, Dividend Policy, and The Bird in the Hand Fallacy,
Bell Journal of Economics, Vol. 10, No. 1, pp 259-270
Brown, S.J., Warner, J.B., 1985, Using Daily Stock Returns: The Case of Event Studies, Journal of
Financial Economics, Vol. 14, No. 1, pp 3-31
Buehler, K., Samandari, H., Mazingo, C., 2009, Capital Ratios and Financial Distress: Lessons from the Crisis, McKinsey working papers on risk, Vol. 15, pp 1-15
Easterbrook, F.H., 1984, Two Agency-Cost Explanations of Dividends, The American Economic
Review, Vol. 74, No. 4, pp 650-659
Fama, E.H, Fisher, L., Jensen, M., Roll, R., 1969, The Adjustment of Stock Rrices to New Information, International Economic Review, Vol. 10, No. 1, pp 1-21
Fama, E.H., French, K.R., 2001, Disappearing Dividends: Changing Firm Characteristics or lower Propensity to Pay? Journal of Financial Economics, Vol. 60, No. 1, pp 3-43
Friedman, M., 1970, The Social Responsibility of Business is to Increase its Profits, New York Times Magazine, September 13, pp 32-33
Ghosh, C., Woolridge, R., 1988, An Analysis of Shareholder Reaction to Dividend Cuts and Omissions, Journal of Financial Research, Vol. 11, No. 4, pp 281-294
Gordon, M.J., 1963, Optimal Investment and Financing Policy, Journal of Finance, Vol. 18, pp 264-272
Haleblian, J., Finkelstein, S., 1999, The Influence of Organizational Acquisition Experience on Acquisition Performance: A Behavior Learning Perspective, Administrative Science Quarterly, Vol. 44, No. 1, pp 29-56
Healy, P.M., Palepu, K.G., 1988, Earnings Information Conveyed by Dividend Initiations and Omissions, Journal of Financial Economics, Vol. 21, No. 2, pp 149-175
Jensen, M., 1986, Agency Costs of Free Cash Flow, Corporate Finance, and Takeover, American
Economic Review, Vol. 76, No. 2, pp 323-329
Lintner, J., 1962, Dividends, Earnings, Leverage, Stock Prices and Supply of Capital to Corporations,
The Review of Economics and Statistics, Vol. 64, pp 243-269
MacKinlay, A.C., 1997, Event Studies in Economics and Finance, Journal of Economic Literature, Vol. 35, No. 1, pp 13-39
30 Michaely, R., Thaler, R.H., Womack, K.L., 1995, Price Reactions to Dividend Initiations and
Omissions: Overreaction or Drift? The Journal of Finance, Vol. 50, No. 2, pp 573-608
Miller, M.H., Modigliani, F., 1961, Dividend Policy, Growth, and the Valuation of Shares. The
Journal of Business, Vol. 34, No. 4, pp 411-433
Miller, M.H. , & Rock, K., 1985, Dividend Policy Under Asymmetric Information. The Journal of
Finance, Vol. 40, No. 4, pp 1031-1051
Polonchek, J., Slovin, M., Sushka, M., 1989, Valuation Effects of Commercial Bank Securities Offerings: A Test of the Information Hypothesis, Journal of Banking and Finance, Vol. 13, No. 3, pp 443-461
Ryan, P.A., Besley, S., Lee, H.W., 2000, An Empirical Analysis of Reaction to Dividend Policy Changes for Nasdaq Firms, Journal of Financial and Strategic Decisions, Vol. 13, No. 1, pp 35-44 Slovin, M., Sushka, M., Poloncheck, J., 1992, Information Externalities of Seasoned Equity Issues: Differences Between Banks and Industrial Firms, Journal of Financial Economics, Vol. 32, No. 1, pp 87-101
Sorokina, N., Boot, D.E., Thornton, J.H., 2013, Robust Methods in Event Studies: Empirical Evidence and Theoretical Implications, Journal of Data Science, Vol. 11, pp 575-606
31
Appendix
Appendix table 1: Average abnormal returns (AAR)
t AAR Std. dev. Min. Max. t-stat p-value N -5 0.60% 0.0546 -0.119 0.25 1.0542 0.2946 91 -4 1.09% 0.0777 -0.163 0.25 1.3412 0.1832 91 -3 0.02% 0.0659 -0.226 0.245 0.0350 0.9722 91 -2 -0.84% 0.0618 -0.183 0.192 -1.2976 0.1977 91 -1 0.44% 0.0548 -0.111 0.244 0.7692 0.4438 91 0 .0.41% 0.0654 -0.25 0.25 0.5981 0.5513 91 1 -2.15% 0.0737 -0.25 0.152 -2.7808*** 0.0066 91 2 0.32% 0.0623 -0.158 0.193 0.5015 0.6172 91 3 -0.81% 0.0583 -0.242 0.14 -13.179 0.1909 91 4 0.32% 0.0560 -0.123 0.25 0.5462 0.5863 91 5 0.21% 0.0749 -0.25 0.25 0.2688 0.7887 91
This table shows the average abnormal return (AAR) over each day of the largest event window (-5,5). Where *,**,*** are the significance levels 10%, 5% and 1%. The AARs are in percentages. Min. and Max. represent the largest observations of the abnormal returns (AR), over which the AARs are calculated. For example, the AAR on event day -5 is +0.6%. The only significant AAR has been found on event day 1. On this day, the AAR is -2.15% on average (p<0.01).
32 Variable 1 2 3 4 5 6 7 8 9 10 11 12 1. Arday 1 1 2. Arday 0 0.0432 1 3. CAR (-5,5) 0.484 0.2915 1 4. CAR (-3,3) 0.5311 0.2466 0.7979 1 5. CAR (-1,1) 0.6396 0.5246 0.6139 0.7146 1 6. CAR (-1,0) 0.0848 0.7676 0.3758 0.4596 0.7555 1 7. CAR (0,1) 0.718 0.6251 0.5954 0.6085 0.8742 0.5285 1 8. Tier 1 -0.1621 -0.0597 -0.1555 -0.0912 -0.1077 -0.0301 -0.143 1 9. Undercap 0.0833 0.032 0.1438 0.1142 0.0803 0.0548 0.0685 -0.719 1 10. Assets 0.185 0.2866 0.3045 0.321 0.3008 0.237 0.3436 -0.3164 0.2528 1 11. Logassets 0.1566 0.0499 0.2869 0.2982 0.2318 0.1724 0.1553 -0.4125 0.3514 0.6371 1 12. DY -0.2034 0.0879 0.0496 -0.0628 -0.0667 0.1334 -0.1399 -0.239 0.3222 0.2815 0.3759 1
Appendix table 2:Correlation matrix for all variables
This table shows the correlation matrix for all dependent and independent variables. The table also contains the correlations between the event windows and independent variables that have not been added to the model. As positive correlation means that an increase in X causes an increase in Y. A negative correlation means that an increase in X causes a decrease in Y. Two variables are have a weak correlation when the correlation is between -0.30 and +0.30. If the correlation is below -0.70 or above +0.70, the correlation is high. The correlation between Tier1 and Undercap is -0.719, which is lower than -0.70 so those variables should not be added in the same model.
Appendix table 3: Stata output – ARday1, Tier1, Logassets, DY, robust
_cons -.0103827 .0615462 -0.17 0.866 -.1327125 .1119471 DY -2.702852 1.286231 -2.10 0.039 -5.259376 -.146328 Logassets .0087628 .0050487 1.74 0.086 -.0012721 .0187977 Tier1 -.005788 .0035485 -1.63 0.106 -.0128409 .001265 ARday1 Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .07019 R-squared = 0.1231 Prob > F = 0.0105 F(3, 87) = 3.97 Linear regression Number of obs = 91 . regress ARday1 Tier1 Logassets DY, robust
33
Appendix table 4: Test on multicollinearity – Logassets, Tier1, DY
Appendix table 5:Stata output – ARday1, Undercap, Logassets, DY, robust
Appendix table 6: Test on multicollinearity – Undercap, Logassets, DY
Mean VIF 1.24 DY 1.18 0.850187 Tier1 1.22 0.821609 Logassets 1.34 0.748247 Variable VIF 1/VIF . vif _cons -.0814849 .0367595 -2.22 0.029 -.1545484 -.0084215 DY -2.772475 1.317993 -2.10 0.038 -5.39213 -.1528209 Logassets .0099257 .004671 2.12 0.036 .0006415 .0192099 Undercap .0151614 .0148178 1.02 0.309 -.0142905 .0446134 ARday1 Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .07057 R-squared = 0.1136 Prob > F = 0.0571 F(3, 87) = 2.60 Linear regression Number of obs = 91 . regress ARday1 Undercap Logassets DY, robust
Mean VIF 1.22 Undercap 1.20 0.834432 DY 1.22 0.817471 Logassets 1.25 0.799485 Variable VIF 1/VIF . vif
34
Appendix table 7: Stata output – ARday1, Tier1 DY, robust
Appendix table 8: Test on multicollinearity – DY, Tier1
_cons .0833041 .0336553 2.48 0.015 .0164214 .1501869 DY -2.168489 1.324301 -1.64 0.105 -4.80026 .4632811 Tier1 -.0086329 .0031821 -2.71 0.008 -.0149567 -.002309 ARday1 Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .07116 R-squared = 0.0884 Prob > F = 0.0090 F(2, 88) = 4.97 Linear regression Number of obs = 91 . regress ARday1 Tier1 DY, robust
Mean VIF 1.06
Tier1 1.06 0.942867 DY 1.06 0.942867 Variable VIF 1/VIF . vif
