The Information Content of Dividend Policy on Future Earnings in ...
International Research Journal of Finance and Economics ISSN 1450-2887 Issue 49 (2010) © EuroJournals Publishing, Inc. 2010 http://www.eurojournals.com/finance.htm
The Information Content of Dividend Policy on Future Earnings in Australia: A VECM Approach King Fuei Lee Senior Fund Manager, Schroder Investment Management 65 Chulia Street #46-00 OCBC Centre Singapore 049513 E-mail: [email protected] Tel: +65-6535 3411; Fax: +65-6535 3486 Abstract The main purpose of this paper is to investigate the existence of the dividend signalling effect in the Australian aggregate market by applying Johansen’s vector errorcorrection model (VECM) to conduct impulse response analysis and Granger-causality test. Our findings show that a unit shock increase in dividend payout leads to a permanent increase in future earnings over time, and that the payout ratio Granger-causes earnings. These results imply that there exists informational/signalling content in dividend payout in the Australian market over the long run. In addition, we also find evidence of Lintner’s (1956) sticky dividends in the impulse response function of payout to a unit shock in earnings, thus supporting the assertion of Kalay (1980) that managerial reluctance to cut dividends is a necessary condition to the existence of information content in dividends policy.
Keywords: Dividend signalling; Future earnings; Dividend payout; Australia; Impulse response function; Sticky dividends JEL Classification Codes: G35
1. Introduction The first prominent suggestion of the existence of information content in dividend policy was mooted in the seminal paper by Lintner (1956) where he examined the dividend behaviour of 28 US companies over the 1918-1941 period and noticed a pattern of sticky dividends and dividend smoothing by managers. In particular, he found that dividends are slow to adjust and tend to lag behind shifts in earnings because managers want to await confirmation of the sustainability of earnings changes before they change dividends. His empirical evidence however ran contrary to the dividend irrelevance theorem by Miller and Modigliani (1961) who hypothesised that in a world of perfect information, full capital mobility, no taxes and no agency costs, the dividend policy adopted by a company should be irrelevant to its market value. This conflict is commonly referred to as the dividend puzzle. Several theories that incorporate considerations for real-world market imperfections of asymmetric information have been advanced to explain Lintner’s observation. One of the key theories is the dividend signalling theory which hypothesises that because of asymmetric information between managers and shareholders, dividends can serve as a signal of the insider’s anticipation of the firm’s future performance. According to the theory, higher payout ratio should signal the manager’s expectations for higher future earnings i.e. there is information content in dividends on future earnings. There are however two pre-conditions for the signalling effect to exist. Firstly, there must be a cost for
International Research Journal of Finance and Economics - Issue 49 (2010)
69
bad companies trying to mimic a good company. Secondly, as shown by Kalay (1980), managerial reluctance to cut dividends is a necessary condition to convey information. This reluctance to cut dividends thus explains the stylised fact of “sticky dividends” in Lintner (1956). Empirical investigations into the relationship between dividends and future earnings have however yielded mixed results, with some studies finding a positive relationship between the variables (Healy and Palepu, 1988; Kao and Wu, 1994; Brooks, Charlton, and Hendershott, 1998), while others finding little or no evidence (DeAngelo, DeAngelo, and Skinner, 1996; Benartzi, Michaely, and Thaler, 1997). Most empirical studies have typically employed simple Ordinary Least Squares (OLS) techniques to perform their analyses. This can be undesirable as the use of OLS regression techniques to test the relationship between current dividend policy and future earnings leads to spurious results if the time series of payout and earnings are non-stationary like many other economic time series. The studies are also primarily focused on the US and European markets, with little work done for the increasingly important Asian-Pacific countries. Lee (2010) attempts to address some of these concerns in his study by investigating the signalling effect of dividend payout on future earnings in Singapore using impulse response functions under a VECM framework. His VECM approach therefore accounts for the non-stationarity of time series, while providing two added advantages. Firstly, the approach is able to include consideration for possible endogeneity between the variables while not making a priori assumptions about the nature of the inter-relationships. For instance, in explaining the trend stationarity of log dividends, Campbell and Shiller (1988) have attributed it to managers’ reluctance to raise dividends in line with earnings increases, while Linter (1956) had explained his observation of the slow move by companies towards a target dividend payout ratio as representing managers’ unwillingness to cut dividends paid to investors. This “stickiness” in dividends implies that current dividends and payout ratio are at least functions of both earnings as well as lagged values of dividends and payout ratios respectively. This means that the relationship between payout and earnings is better investigated within a system of simultaneous equations that are endogenously determined than as OLS regressions. Secondly, the investigation of the information content in dividend policy usually requires the identification of unexpected dividend policy changes which can be problematic. The difficulty in isolating and controlling investors' expectations means that most studies have either assumed a naive dividend expectation model (i.e. any change in dividends is unexpected) or employed some dividend forecasting model to capture investors' expectations. A VECM approach gets around that by applying Johansen’s vector error-correction model (VECM) to derive impulse response functions through which the unanticipated changes in dividend policy can be modelled as unit shocks to the variable in the VECM. Lee (2010) thus found evidence of the existence of information content in the dividend policy of the aggregate Singapore market. Despite his findings, a natural extension of his conclusions to all other Asian-Pacific countries can be difficult due to certain key differences between the tax regime and institutional structure of Singapore versus other countries. Firstly as highlighted by Lin (2002), the legal regime in which companies operate in and the institutional structure through which firms raise capital have significant impact on the extent of agency costs that exist in companies in the Asia Pacific region. These varying degrees of agency costs may affect dividend policy and hence the motivation for dividend signalling. In particular, while both Singapore and Australia operate under very similar common law regimes as a consequence of their Commonwealth backgrounds, the institutional structures between the two countries differ substantially. Lin (2002) found the Singapore institutional structure to be closer to a “Continental-German-Japanese” banking model (Mayer, 1990) while Australia’s structure approaches closer to an “Anglo-Saxon” capital market model. The key difference between a bank-oriented model and a capital market-oriented model lies in the extent of delegation of monitoring activities to banks, with banks in the formal model adopting a more dominant role in the monitoring of companies. As a consequence, the levels of agency costs differ under the two institutional structures, thus leading to different dividend policies existing in the two types of financial systems. Dewenter and Warther (1998) hence attributed the less sticky
70
International Research Journal of Finance and Economics - Issue 49 (2010)
dividends of Japanese companies, compared to their US counterparts, to the lower agency costs of Japanese firms as a result of their closer ties with the banks through cross-holdings. These results also mean that while the conclusion of the existence of informational content in dividends on future earnings in Singapore by Lee (2010) may be reasonably inferred for other common law/bank-oriented countries in the Asia Pacific region like Hong Kong, Malaysia and Thailand, it will not necessarily hold true for countries in the region which are common law/capital market-oriented like Australia and New Zealand. Secondly, the tax regime also differs significantly between the two countries. Australia operates a tax system where dividends are being taxed at the same rate as capital gains at the individual level, while in Singapore personal taxes are only applied to dividends and not to capital gains. As argued by the tax-effect hypothesis of dividends, the difference in tax regimes will have an impact on corporate dividend policy and hence the incentive for dividend signalling in Singapore versus Australia. These key differences make a separate detailed study of dividend signalling in Australia an interesting exercise in its own right. This paper therefore attempts to expand the current literature on dividend signalling by applying cointegration and vector error-correction modelling techniques to investigate the existence of informational content in the dividend payout in the “Anglo-Saxon” capital market model of Australia. In addition, the paper extends the analysis of Lee (2010) to also search for the presence of managerial reluctance to cut dividends which is a necessary pre-condition to the existence of dividend signalling. This pre-condition has typically been implicitly assumed in most dividend-signalling studies without any explicit investigation for it. This paper therefore also expands on current literature by conducting a simultaneous investigation into the existence of both the dividend signalling effect and the prerequisite of managerial reluctance to cut dividends within the same analytical framework. Specifically, this paper will apply Johansen’s cointegration techniques (Johansen, 1988; Johansen and Juselius, 1990) to the Australian market to develop a vector error-correction matrix (VECM) between earnings and dividend payout which is then used to derive the impulse response function for tracing the effect of a unit shock in the payout ratio on earnings over time. We also derive the impulse response function for tracing the effect of a unit shock in earnings on payout ratio, as well as conduct the Granger-causality test to establish the extent to which the lag process in dividend payout explains future earnings. The main findings of this paper can be briefly described as follows: 1. An unanticipated increase in the dividend payout, represented by a one standard deviation increase in dividend payout in the impulse response function, leads to a permanent increase in future earnings of the “Anglo-Saxon” capital-market model of Australia over time. This supports the dividend signalling theory that dividend payout possesses informational content about future earnings, and that the two variables are positively correlated. 2. Lagged values of the dividend payout ratio have statistically significant information about future real earnings, as evidenced by the rejection of the null hypothesis of no granger-causality between payout ratio and real earnings in the Australian market. 3. An unanticipated increase in real earnings is accompanied by an eventual permanent increase in the dividend payout but only after a lag. This supports the observation of sticky dividends by Lintner (1956) where companies only raise their dividend policies when they are sure of the sustainability of earnings increases. It also satisfies the assertion by Kalay (1980) of managerial reluctance to cut dividends as a necessary condition to the existence of information content in dividend policy. The paper is structured as follows: Introduction of this paper is presented in Section 1. Section 2 gives an overview of the econometric procedures employed. In this section, we use the JMulti package by Helmut Lutkepohl and Markus Kratzig, and the EasyReg package by Herman Bierens for the econometric analysis. Section 3 gives a description of the data used. This is followed by the empirical results and their interpretations in Section 4, while Section 5 provides the conclusions of the paper.
International Research Journal of Finance and Economics - Issue 49 (2010)
71
2. Econometric Methodology We define the two variables of real earnings as EPSt, and dividend payout ratio as, POUTt at time t. We then pursue the following econometric procedure. First, we test for non-stationarity and the order of integration for both POUTt and EPSt by employing the Augmented Dickey-Fuller (ADF) procedure (Dickey and Fuller, 1979, 1981), the Phillips-Perron (PP) unit root test (Phillips and Perron, 1988) and the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (Kwiatkowski, Phillips, Schmidt and Shin, 1992) which uses a null hypothesis of stationarity. The tests are applied to both the model with constant and the model with constant and time trend, and performed for 0 to 60 lags i.e. five years. For the ADF test, the choice of the number of lagged differenced terms is made based on the four information criterion: Akaike Information Criterion (Akaike, 1971), Hannan-Quinn Criterion (Hannan and Quinn, 1979), Schwarz Criterion (Schwarz, 1978) and Final Prediction Error (Akaike, 1969). For the PP test and KPSS test, the number of truncation lags (l) to be used to evaluate the serial correlation for the NeweyWest correction (Newey and West, 1987) is computed by l = q(T/100)^0.25 where q = 4 or 12. For all the unit root tests, if non-stationarity is not rejected, the variable is differenced once and the unit root tests are performed again. This is repeated until stationarity is achieved. The number of differences taken before the series becomes stationary is then the order of integration ie I(d). If the two time series are found to be integrated of the same order, we then proceed to test for the existence of cointegration vectors among them by performing the Johansen maximum eigenvalue test, the Johansen trace test and the Saikkonen & Lütkepohl cointegration test (Saikkonen & Lütkepohl, 2000a; 2000b; 2000c; 2002). The cointegration tests involve the determination of the rank of the matrix Π which reflects the number of cointegrating vectors in the process governing movements of POUTt and EPSt. In this application, there are three possible ranks of Π. If r = 2 (i.e. Π is full rank), that means that both POUTt and EPSt are stationary processes. This would however contradict the earlier unit root tests findings that they are both integrated of order one or higher. If r = 0, then there is no cointegration between the two variables, and no stationary long run relationship exists. When r = 1, there is a single cointegrating vector between POUTt and EPSt such that Π = αβ’, where α contains the cointegrating vector and β is the corresponding error-correction coefficients. This supports the presence of a cointegrating relation between payout ratio and real earnings. If we find the existence of one cointegrating relation between the two variables of POUTt and EPSt, we can then proceed to derive the vector error-correction matrix (VECM) of forms: ∆EPSt = µ2 + Σk-1j=1 Г21(j) ∆POUTt-j + Σ k-1j=1 Г22(j) ∆EPSt-j + Π 21 POUTt-k + Π 22 EPSt-k (1) ∆POUTt = µ1 + Σk-1j=1 Г11(j) ∆POUTt-j + Σ k-1j=1 Г12(j)∆EPSt-j + Π11 POUTt-k + Π12 EPSt-k (2) where the matrix Г represents the short run dynamics of the relationship between POUTt and EPSt, and the matrix Π captures the long run information in the data. The order (lag length) of the Vector Autoregressive (VAR) form is chosen by selecting the VAR(p) model based on the four information criterion: Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. Lags of 0 to 60 months are tested. We perform two particular analyses to investigate the existence of informational content in dividend policy about future earnings. First, we test for causality in the VECM by applying the Granger-causality test. Granger (1969) defined a variable as being causal for another variable if inclusion of the lagged values of the former helps to improve the forecasts of the latter. If the dividend payout ratio contains information content about future earnings as hypothesised by the dividend signalling theory, then payout should granger-cause earnings. We next proceed to employ impulse response analysis which provides a fuller illustration of the dynamic interactions between the variables in the system by tracing the effect of innovations in one variable on the other variable over time. We are particularly interested in analysing the impulse response function of earnings to a unit shock in payout. We also examine the impulse response function of payout to a unit shock in earnings for signs of sticky dividends.
72
International Research Journal of Finance and Economics - Issue 49 (2010)
3. Data Sample This paper focuses on the Australian market. We have chosen to use the MSCI Australia index as it has the longest history of available data while providing a good representation of the country’s aggregate equity market given the index’s minimum 85% target representation guideline. The time period employed is therefore from December 1969 to March 2009. The data required are the monthly values of price-earnings ratio, dividend yield, the stock market index level and the consumer price index. We derive the 12-month trailing earnings in index points for the market by dividing the price series through by the price-earnings ratio. The real earnings series is subsequently obtained by dividing the earnings series through by the consumer price index. The payout ratio is derived by multiplying the dividend yield with the price-earnings ratio. The methodology for the derivation of the real earnings and payout series are consistent with the procedure of Arnott and Asness (2003) and Gwilym et al (2006). The logarithms of real earnings and payout ratio are used for this analysis. All data is downloaded from Datastream. The issue to note with these indices is that the survivorship bias in any index construction process affects the variation in index compositions over time. This means that the EPS of the index is likely to lag the GDP growth of the economy.
4. Empirical Findings 4.1. Results Figures 1 and 2 show the time series plots of the payout ratio (POUT) and real earnings (EPS) of the MSCI Australia index respectively versus the 1-year moving average over time. A visual inspection of the graphs suggests a lack of existence of time trending properties in both time series. Figure 1: Time Series Plot of Payout Ratio of MSCI Australia Index 4.70
4.60
4.50
4.40
4.30
4.20
4.10
4.00
3.90
3.80
Logarithm of Dividend Payout of MSCI Australia Index
c05
ec -0 7 D
De
ec -0 3
c01
D
De
ec -9 9 D
c95
ec -9 7 D
De
ec -9 3 D
c89
ec -9 1 D
De
c87 De
ec -8 5 D
ec -8 3 D
c81 De
ec -7 9 D
ec -7 7 D
c75 De
c73 De
ec -7 1 D
D
ec -6 9
3.70
Moving Average (12months)
Table 1 shows the results of the various unit root tests. We have tested all the variables in both levels and first-difference terms for non-stationarity using unit root tests with (1) a constant term, and (2) a constant term with time trend.
73
International Research Journal of Finance and Economics - Issue 49 (2010) Figure 2: Time Series Plot of Real Earnings of MSCI Australia Index -0.50
-0.70
-0.90
-1.10
-1.30
-1.50
-1.70
-1.90
-2.10
D ec D 69 ec D 70 ec D 71 ec D 72 ec D 73 ec D 74 ec D 75 ec D 76 ec D 77 ec D 78 ec D 79 ec -8 D 0 ec D 81 ec D 82 ec D 83 ec D 84 ec D 85 ec D 86 ec -8 D 7 ec D 88 ec D 89 ec D 90 ec D 91 ec D 92 ec D 93 ec D 94 ec D 95 ec D 96 ec D 97 ec D 98 ec D 99 ec D 00 ec -0 D 1 ec D 02 ec D 03 ec D 04 ec D 05 ec D 06 ec D 07 ec -0 8
-2.30
Logarithm of Real Earnings of MSCI Australia Index (rhs)
Table 1:
Moving Average (12months)
Unit Root Test Variables in Levels POUT EPS
Augmented Dickey-Fuller H0: Unit Root H1: Level Stationary H0: Unit Root H1: Trend Stationary with Drift Phillips-Perron H0: Unit Root H0: Unit Root with Drift
ADF(c) ADF(c+t)
-2.65 (7) -3.64 (7)*
-2.54 (6) -3.65 (6)*
-6.38 (6)** -6.40 (6)**
-5.78 (5)** -5.76 (5)**
H1: Level Stationary
PP(c)
H1: Trend Stationary
PP(c+t)
-12.46 (6) -13.95 (18) -21.35 (6)
-5.93 (6) -8.41 (18) -12.67 (6)
-543.33 (6)** -530.01 (18)** -542.45 (6)**
-510.02 (6)** -597.10 (18)** -509.49 (6)**
-24.26 (18)*
-17.14 (18)
-527.78 (18)**
-593.38 (18)**
2.63 (6)**
3.19 (6)**
0.08 (6)
0.06 (6)
1.10 (18)** 0.36 (6)**
1.30 (18)** 0.39 (6)**
0.09 (18) 0.03 (6)
0.04 (18) 0.03 (6)
0.16 (18)**
0.17 (18)**
0.04 (18)
0.03 (18)
Kwaitowski, Phillips, Schmidt and Shin H0: Level H1: Unit Root KPSS(c) Stationary H0: Trend Stationary
Variables in First Difference ∆POUT ∆EPS
H1: Unit Root with Drift
KPSS(c+t)
Notes: All variables in logarithms. Period: Jan 1990 – Jun 2007. Significance levels: ** = 1%, * = 5%. ADF-Tests: Critical values provided by JMulti and EasyReg. Lag lengths for ADF test according to the Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. Numbers in parentheses are the truncation lags (l) use to evaluate the serial correlation for the Newey-West correction for PP and KPSS tests, and are computed by l = q(T/100)^0.25 where q = 4 or 12.
In general, the unit root tests for non-stationarity (i.e. ADF and PP) fail to reject the null hypothesis of non-stationarity at 5% significance levels for both POUT and EPS in level terms, while
74
International Research Journal of Finance and Economics - Issue 49 (2010)
rejecting the null hypothesis at 1% significance levels for both variables in first-differenced terms. For the unit root test for stationarity i.e. KPSS, the null hypothesis is rejected at 1% significance levels for both POUT and EPS in level terms while the null hypothesis is not rejected at 5% significance levels for both variables in first-differenced terms. The unit root tests results therefore show strong evidence that the payout ratio (POUT) and real earnings (EPS) of the MSCI Australia index are non-stationary and are integrated of order one i.e. I(1). It is worth noting that when estimating the models with constant and time trend for use in the ADF tests, the coefficients of the time trend for both POUT and EPS are not significant. In addition, the inclusion of a time trend in the unit root tests yielded very similar results as the tests without time trends. This points at the lack of time-trending properties in both time series, and corroborates the earlier conclusions from our visual inspection of the graphs. We have however chosen to continue conducting our analyses for both the inclusion and exclusion of time trends for completeness. Having confirmed that both processes are I(1), we then proceed to determine the lag lengths to be used in the cointegration tests. This is done based on the four information criteria: Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. We tested a maximum lag length of 50 lags and an optimal lag length of 7 lags is recommended. Table 2 shows the results of the Johansen maximum eigenvalue test, the Johansen trace test and the Saikkonen-Lütkepohl cointegration tests for both a model with intercept, and a model with intercept and time trend. All the cointegration tests reject the null hypothesis of no cointegrating relation at 5% significance level while accepting the null hypothesis of one cointegrating relation at 10% significance level. Table 2:
Cointegration Analysis of Dividend Payout and Earnings for MSCI Australia Index With Intercept
With Intercept and Time Trend
Test for the cointegration rank H0
H1
Test statistic
Critical Values 5%
10%
Test statistic
Critical Values 5%
10%
Johansen's Cointegration Maximum Eigenvalue Test r=0
r≥1
27.90
15.80
13.80
28.00
19.20
16.90
r≤1
r≥2
1.00
9.10
7.60
8.40
12.50
10.60
Johansen's Cointegration Trace Test r=0
r≥1
28.91
20.16
17.98
36.45
25.73
23.32
r≤1
r≥2
1.02
9.14
7.60
8.42
12.45
10.68
Saikkonen & Lütkepohl Cointegration Test r=0
r≥1
24.80
12.26
10.47
26.06
15.76
13.88
r≤1
r≥2
0.03
4.13
2.98
1.10
6.79
5.47
Number of cointegrating relations:
1
Choice of rank of Var(p): 7 Notes: r is the number of cointegrating relations. The rank of Var(p) is chosen based on Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error.
After establishing the existence of a cointegrating relation between the variables of POUT and EPS, we proceed to estimate a vector error correction model for analysis. The derived VECM is shown in Table 7 and Table 8 in the appendices. While an exhaustive interpretation of the individual coefficients is not within the scope of this paper, we note that the signs of the coefficients in the estimated cointegrating vector (with the coefficient of EPS normalised) are exactly as hypothesised by
75
International Research Journal of Finance and Economics - Issue 49 (2010)
the dividend signalling theory. In the long-run, real earnings is positively related to the payout ratio, hence suggesting the existence of informational content in dividend payout. Before the VECM is being applied for further analysis, we check whether the model is an adequate representation by conducting the Portmanteau test for autocorrelation and the LM test. The Portmanteau test for autocorrelation tends to be more useful for longer lag lengths, while the LM test is usually more applicable to the testing of low order residual autocorrelation. We therefore apply a lag order of 16 for our Portmanteau test, and a lag length of 5 for our LM test. The results are presented in Table 3 where it shows that the null hypotheses of no residual autocorrelation are not rejected at 10% significance level for both tests under the model with intercept and the model with intercept and time trend. Table 3:
Tests for Autocorrelation With Intercept
With Intercept and Time Trend
Test statistic
43.22
42.24
p-value
0.26
0.29
Adjusted test statistic
44.24
43.23
p-value
0.25
0.26
Lag order
16
16
Degrees of freedom
38
38
LM statistic
25.93
25.17
p-value
0.17
0.20
Lag order
5
5
Degrees of freedom
20
20
Portmanteau Test H0: No residual autocorrelation up to lag order H1: At least one autocovariance is nonzero
LM Test H0: No residual autocorrelation at lag order H1: At least one autocovariance is nonzero
We now proceed to investigate the existence of informational content in dividend payout on future earnings. We first perform the Granger-causality test to check for the existence of granger-causal relations between the payout ratio and real earnings of the Australian market, with particular focus on the test of the null hypothesis that POUT does not granger-cause EPS. The results of the test are shown in Table 4. The null hypothesis of no granger-causality between POUT and EPS is rejected at 5% significance level. This means that POUT granger-causes EPS i.e. lagged values of POUT provide statically significant information about the future values of EPS, and indicates the existence of an informational/signalling effect in current dividend policy on future earnings. Table 4:
Granger-Causality Analysis of Dividend Payout and Earnings for MSCI Australia Index
With Intercept H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS With Intercept and Time Trend
Test statistic
pval-F(1,7,894)
2.13
0.04
76
International Research Journal of Finance and Economics - Issue 49 (2010)
H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS
2.13
0.04
Figure 3: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Earnings and Dividend Payout 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
We next examine the effect of an unanticipated increase in dividend payout on future earnings. This is done by first deriving an impulse response function (IRF) tracing the response of EPS to a one standard deviation innovation in POUT. Figure 3 shows the impulse response function with the upper and lower limits of the 95% confidence interval. It can be seen that a shock increase in dividend payout leads to a positive and permanent increase in future real earnings, with the strongest positive impact experienced in real earnings that is 18 months forward. This confirms our earlier finding that changes in dividend policy provide information/signals about future real earnings. The impulse response function also shows that increases in the dividend payout leads to increases in future earnings. These findings are consistent with the dividend signalling theory. Figure 4 shows the impulse response function tracing the response of POUT to a one standard deviation innovation in EPS. It can be seen that a shock increase in real earnings leads to an eventual permanent positive increase in the payout ratio. However, it is also observed that the payout ratio initially drops in the first 7 months after the innovation before staging a recovery to an eventual permanent increase of approximately 0.65 standard deviation 43 months later. This suggests that even after an unanticipated increase in real earnings, management have typically allowed the payout ratio to drop initially, as dividends are kept unchanged while earnings rise, before they incorporate the permanent increase in real earnings into a permanent increase in dividend payout. This is consistent with the observation of sticky dividends by Lintner (1956) where he noticed that dividend policy tracks earnings with a lag as management awaits confirmation of the sustainability of earnings increases before they increase dividends. It also supports the assertion of Kalay (1980) that managerial reluctance to cut dividends is a necessary condition for the existence of informational content in dividend policy.
77
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 4: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Earnings and Dividend Payout 1.00
0.80
0.60
Impulse Response Function
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
Figure 5 and 6 show the impulse response function tracing the response of EPS to an innovation in POUT, and the response of POUT to an innovation in EPS respectively for the VECM with intercept and time trend. The findings are almost identical to the VECM with intercept only, and reinforce the dividend signalling theory as well as the observation of sticky dividends. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The stylised fact of sticky dividends observed by Lintner (1956) is also found in our analysis. These results support the dividend signalling theory or the information content of dividends hypothesis. Figure 5: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Earnings and Dividend Payout 1.40
1.20
Impulse Response Function
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
Forecast Horizon
36
38
40
42
44
46
48
50
52
54
56
58
60
78
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 6: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Earnings and Dividend Payout 1.00
0.80
0.60
Impulse Response Function
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
4.2. Robustness Test As with most VECM analysis, because the model is under-identified, if there exists a contemporaneous relation between the shocks of real earnings and dividend payout ratio, the results of the impulse response analysis will be very sensitive to the ordering of the variables in the VECM model. One of the ways to resolve this issue is to try a different ordering of the variables and see whether the results are significantly affected. We therefore repeat the construction of the VECM for both the models including and excluding time trends for a different ordering of the variables of EPS and POUT. Table 5:
Tests for Autocorrelation
Portmanteau Test H0: No residual autocorrelation up to lag order H1: At least one autocovariance is nonzero Test statistic p-value Adjusted test statistic p-value Lag order Degrees of freedom LM Test H0: No residual autocorrelation at lag order H1: At least one autocovariance is nonzero LM statistic p-value Lag order Degrees of freedom
With Intercept
With Intercept and Time Trend
43.03 0.26 44.04 0.23 16 38
42.24 0.29 43.23 0.26 16 38
28.53 0.10 5 20
25.17 0.20 5 20
79
International Research Journal of Finance and Economics - Issue 49 (2010) Table 6:
Granger-Causality Analysis of Dividend Payout and Earnings for MSCI Australia Index Test statistic
pval-F(1,7,894)
2.13
0.04
1.94
0.06
1.76
0.09
1.97
0.06
With Intercept H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS H0: EPS Does Not Granger-cause POUT H1: EPS Granger-causes POUT With Intercept and Time Trend H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS H0: EPS Does Not Granger-cause POUT H1: EPS Granger-causes POUT
Figure 7: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Dividend Payout and Earnings 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
Figure 8: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Dividend Payout and Earnings 1.20
1.00
0.80
Impulse Response Function
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
36
38
40
42
44
46
48
50
52
54
56
58
60
80
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 9: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Dividend Payout and Earnings 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
Figure 10: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Dividend Payout and Earnings 1.20
1.00
0.80
Impulse Response Function
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
Tables 9 and 10 in the appendices show the derived VECM models for the exclusion and inclusion of time trends. It can be seen that the signs of the coefficients in the estimated cointegrating vector is as hypothesised by the dividend signalling model. Table 5 shows the tests for autocorrelation, and the results demonstrate a lack of residual autocorrelation and the suitability of the models. The granger-causality tests results are shown in Table 6 where we see strong evidence that payout ratio granger-causes real earnings in the Australian market. The impulse responses functions of real earnings and payout to unit shocks in payout and real earnings respectively for the VECM models with intercepts, and with intercepts and time trend, are shown in Graphs 7-10. The results are the same as the models before, showing both informational content in dividend policy as well as signs of sticky dividends. When we compare the similar findings
International Research Journal of Finance and Economics - Issue 49 (2010)
81
between this paper and Lee (2010), it is interesting to note that despite the differences in tax regime and institutional structure between Singapore and Australia, the accompanying varying effects of agency costs and tax effect has not prevented the assertion of the dividend signalling effect. Our results are similar to that of Lin (2002) who found no statistically significant differences in the dividend policies of companies in common-law/bank-oriented countries versus firms in common law/capital market-oriented countries. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings in the “Anglo-Saxon” capital market model of Australia, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The stylised fact of sticky dividends observed by Lintner (1956) is also found in our analysis. These results support the dividend signalling theory or the information content of dividends hypothesis. Our results are robust to the under-identification of the VECM models.
5. Summary and Concluding Remarks This paper applies Johansen’s vector error-correction model (VECM) to investigate the existence of the dividend signalling effect in the Australian aggregate market using impulse response analysis and Granger-causality test. The main results for this paper can be summarized as follows: 1. An unanticipated increase in the dividend payout, represented by a one standard deviation increase in dividend payout in the impulse response function, leads to a permanent increase in future earnings of the Australian market over time. This supports the dividend signalling theory that dividend payout possesses informational content about future earnings, and that the two variables are positively correlated. 2. Lagged values of the dividend payout ratio have statistically significant information about future real earnings, as evidenced by the rejection of the null hypothesis of no granger-causality between payout ratio and real earnings in the Australian market. 3. An unanticipated increase in real earnings is accompanied by an eventual permanent increase in the dividend payout after a lag. This supports the observation of sticky dividends by Lintner (1956), and satisfies the assertion by Kalay (1980) of managerial reluctance to cut dividends as a necessary condition to the existence of information content in dividend policy. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The prerequisite of managerial reluctance to cut dividends to the dividend signalling theory is also satisfied by our empirical finding. These results support the dividend signalling theory or the informational content of dividends hypothesis, and are robust to the under-identification of the VECM models.
References [1] [2] [3] [4] [5]
Akaike, H., 1969. “Fitting Autoregressive Models for Prediction”, Annals of the Institute of Statistical Mathematics 21, pp. 243-247. Akaike, H., 1971. “Autoregressive Model Fitting for Control”, Annals of the Institute of Statistical Mathematics 23, pp. 163-180. Arnott, R. and C. Asness, 2003. “Surprise! Higher Dividends = Higher Earnings Growth”, Financial Analysts Journal 59, pp. 70-87. Ball, R. and R. Watts, 1972. “Some Time-Series Properties of Accounting Income”, Journal of Finance 27, pp. 663-81. Bernartzi, S., R. Michaely and R. Thaler, 1997. “Do Changes in Dividends Signal the Future or the Past?”, Journal of Finance 52, pp. 1007-1034.
82
International Research Journal of Finance and Economics - Issue 49 (2010)
[6]
Brooks, Y., R. Charlton and W., Hendershott, 1998. “Do Firms Use Dividends to Signal Large Future Cash Flows?”, Financial Management 27, pp. 46-57. Campbell, J. and R. Shiller, 1988. “Stock Prices, Earnings, and Expected Dividends”, Journal of Finance 43, pp.661-76. DeAngelo, H., L. DeAngelo and D. Skinner, 1996. “Reversal of Fortune, Dividend Signaling, and the Disappearance of Sustained Earnings Growth”, Journal of Financial Economics 40, pp. 340-371. Dewenter K. and V. Warther, 1998. “Dividends, Asymmetric Information and Agency Conflicts: Evidence from A Comparison of Dividend Policies of Japanese and US firms”, Journal of Finance 53, pp. 879-904. Dickey, D. A. and W.A. Fuller, 1979. “Estimators for Autoregressive Time Series With a Unit Root”, Journal of the American Statistical Association 74, pp. 427–431. Dickey, D.A. and W.A. Fuller, 1981. “Likelihood Ratio Statistics for Autoregressive Time Series With a Unit Root”, Econometrica 49, pp. 1057-72. Granger, C.W.J., 1969. “Investigating Causal Relation By Econometric and Cross-sectional Method”, Econometrica 37, pp. 424–438. Gwilym, O., J. Seaton, K. Suddason and S. Thomas, 2006. “International Evidence on the Payout Ratio, Earnings, Dividends, and Returns”, Financial Analysts Journal 62, pp. 36-53. Hannan, E. J. and B.G. Quinn, 1979. “The Determination of the Order of an Autoregression”, Journal of the Royal Statistical Society B, pp. 190-195. Healy, P. and K. Palepu, 1988. “Earnings Information Conveyed by Dividend Initiations and Omissions”, Journal of Financial Economics 21, pp. 149-176. Johansen, S., 1988. “Statistical Analysis of Cointegration Vectors”, Journal of Economic Dynamics and Control 62, pp. 36-53. Johansen, S. and K. Juselius, 1990. “Maximum Likelihood Estimation and Inference on Cointegration - With Applications to the Demand for Money”, Oxford Bulletin of Economics and Statistics 52, pp. 169-210. Kalay, A., 1980. “Signaling, Information Content, and the Reluctance to Cut Dividends”, Journal of Financial and Quantitative Analysis 15, pp. 855-869. Kao, C. and C. Wu, 1994. “Rational Expectations, Information Signalling and Dividend Adjustment to Permanent Earnings”, Review of Economics and Statistics 76, pp. 490-502. Kwiatkowski, D., P.C.B. Phillips, P. Schmidt and Y. Shin, 1992. “Testing the Null of Stationarity Against the Alternative of a Unit Root”, Journal of Econometrics 54, pp. 159-178. Lee, K.F., 2010. “An Empirical Study of Dividend Payout and Future Earnings in Singapore”, Review of Pacific Basin Financial Markets and Policies, doi:10.1142/S0219091510001949 Lin, C.T., 2002. “Dividend Policies, Legal Regimes and Institutional Structures in the Asia Pacific Region”, School of Finance and Business Economics Working Paper Series 02.04 Lintner, J., 1956. “Distribution of Incomes of Corporations Among Dividends, Retained Earnings, and Taxes”, American Economic Review 46, pp. 97-113. Mayer, C., 1990. Financial Systems, Corporate Finance and Economic Development, in: Hubbard, R.G. (Ed.), Asymmetric Information, Corporate Finance and Investment, University of Chicago Press, Chicago, IL. pp. 307.332. Miller, M. and F. Modigliani, 1961. “Dividend Policy, Growth, and the Valuation of Shares”, Journal of Business XXXIV, pp. 235-64. Newey, W.K. and K.D., West, 1987. “A Simple Positive Definite Heteroskedasticity and Autocorrelation Consistent Covariance Matrix”, Econometrica 55, pp. 703–708. Penman, S., 1983. “The Predictive Content of Earnings Forecasts and Dividends”, Journal of Finance 38, pp. 1181-1199. Phillips, P.C.B. and P. Perron, 1988. “Testing For a Unit Root in Time Series Regression”,.Biometrika 75, pp. 335-346.
[7] [8]
[9]
[10] [11] [12] [13] [14] [15] [16] [17]
[18] [19] [20] [21] [22] [23] [24]
[25] [26] [27] [28]
International Research Journal of Finance and Economics - Issue 49 (2010) [29] [30] [31]
[32] [33]
83
Saikkonen, P. and H. Lütkepohl, 2000a. „Testing for the Cointegrating Rank of a VAR Process With an Intercept”, Econometric Theory 54, pp. 159-178. Saikkonen, P. and H. Lütkepohl, 2000b. „Testing for the Cointegrating Rank of a VAR Process With Structural Shifts”, Journal of Business & Economic Statistics 18, pp.451-464. Saikkonen, P. and H. Lütkepohl, 2000c. „Trend Adjustment Prior to Testing for the Cointegrating Rank of a Vector Autoregressive Process”, Journal of Time Series Analysis 21, pp. 435-456. Saikkonen, P. and H. Lütkepohl, 2002. „Testing For a Unit Root in a Time Series With a Level Shift at Unknown Time.” Econometric Theory 18, pp. 313-348. Schwarz, G., 1978. “Estimating the Dimension of a Model”, Annals of Statistics 6, pp. 461-464.
84
International Research Journal of Finance and Economics - Issue 49 (2010)
Appendices Table 7:
Vector Error Correction Matrix (VECM) With Intercept for Earnings and Dividend Payout
∆EPSt
[
]
=
-0.03
[
∆Payoutt
]
[
[
+
[
[
-1.36
0.16
0.22
-0.22
-0.29
0.13
0.09
]
[
]
[
0.03
+
Table 8:
1.00
[
+
[
+
[
-0.02
-0.01
0.10 -0.05
0.04 -0.02
u1t u2t
]
[
]
[
EPSt-1 Payoutt-1
∆EPSt-1 ∆Payoutt-1 ∆EPSt-3 ∆Payoutt-3 ∆EPSt-5 ∆Payoutt-5
]
+
]
+
]
[
+
]
[
+
[
[
7.07
0.03
0.06
-0.08
-0.11
0.07
0.02
-0.10
-0.07
0.40 -0.20
0.11 0.02
]
[
]
[
]
[
]
[
CONST ]
∆EPSt-2 ∆Payoutt-2 ∆EPSt-4 ∆Payoutt-4 ∆EPSt-6 ∆Payoutt-6
]
]
]
]
]
Vector Error Correction Matrix (VECM) With Intercept and Time Trend for Earnings and Dividend Payout
∆EPSt ∆Payoutt
]
=
[
-0.02 0.03
]
[
[
+
[
+
1.00
-2.12
0.16
0.21
-0.22
-0.29
0.13
0.08
[
]
[
]
[
-0.01
]
+
]
+
∆EPSt-1
∆EPSt-3 [
]
+
[
0.10 -0.05
0.04 -0.02
]
+
[
-0.20 0.26
]
[
[
0.00
0.03
0.05
-0.08
-0.11
0.07
0.01
[
∆Payoutt-3
+
[
[
∆Payoutt-1
] -0.03
EPSt-1 Payoutt-1
∆EPSt-5 ∆Payoutt-5
]
+
[
CONST
]
+
[
]
[
]
[
-0.07
0.39 -0.20 u1t
0.10 0.02
u2t
]
]
∆EPSt-2 ] ∆Payoutt-2 ∆EPSt-4 ]
-0.10
TRENDt-1]
[
] ∆Payoutt-4
]
[
∆EPSt-6 ∆Payoutt-6
]
85
International Research Journal of Finance and Economics - Issue 49 (2010)
Table 9:
Vector Error Correction Matrix (VECM) With Intercept for Dividend Payout and Earnings -0.05
∆POUTt [
]=[
(-5.36)
POUTt-1 ][[
0.04
∆EPSt
1.00
-0.74
( --- )
('4.64)
][
]+
-5.22
[
EPSt-1
][
] CONST ]
(-21.24)
(4.09)
+[
+[
+[
-0.29
-0.22
(-4.11)
(-3.27)
0.22
0.16
(2.86)
(2.26)
-0.01 (-0.18) 0.09
-0.02 (-0.35) 0.13
(1.10)
(1.81)
-0.02 (-0.28) 0.04 (0.55)
-0.05 (-0.72) 0.10 (1.44)
∆POUTt-1 ][
]+[ ∆EPSt-1
∆POUTt-3 ][
]+[ ∆EPSt-3
∆POUTt-5 ][
]+[ ∆EPSgt-5
-0.11
-0.08
(-1.54)
(-1.17)
0.06
0.03
(0.74)
(0.42)
-0.07 (-0.93) 0.02
-0.10 (-1.44) 0.07
(0.25)
(1.00)
0.02 (0.23) 0.11 (1.48)
-0.20 (-2.96) 0.40 (5.57)
∆POUTt-2 ][
] ∆EPSt-2
∆POUTt-4 ][
] ∆EPSt-4
∆POUTt-6 ][
] ∆EPSt-6
u1t +[
] u2t
Table 10: Vector Error Correction Matrix (VECM) With Intercept and Time Trend for Dividend Payout and Earnings ∆POUTt [
]=[ ∆EPSt
-0.06 (-5.36) 0.04 (3.83)
][[
+[
1.00
-0.47
( --- )
(-2.81)
-0.29 (-4.07)
-0.22 (-3.29)
POUTt-1 ][
]+
[
EPSt-1
][
∆POUTt-1
0.00
][
(-1.27)
]+[
-0.11 (-1.50)
-0.08 (-1.19)
] TRENDt-1]
][
∆POUTt-2
]
86
International Research Journal of Finance and Economics - Issue 49 (2010)
+[
+[
+[
0.21 (2.77)
0.16 (2.19)
-0.01 (-0.16) 0.08 (1.00)
-0.03 (-0.37) 0.13 (1.74)
-0.02 0.27 0.04 (0.47)
-0.05 (-0.75) 0.10 (1.36)
0.26 (5.38) -0.20 (-3.82)
∆EPSt-1
∆POUTt-3 ][
]+[ ∆EPSt-3
∆POUTt-5 ][
]+[ ∆EPSgt-5 +[
]
[
0.05 (0.65)
0.03 (0.35)
-0.07 (-0.89) 0.01 (0.14)
-0.10 (-1.46) 0.07 (0.91)
][
0.02 (0.26) 0.10 (1.39)
-0.20 (-2.97) 0.39 (5.48)
][
u1t
CONST ]
] u2t
∆EPSt-2
∆POUTt-4 ] ∆EPSt-4
∆POUTt-6 ] ∆EPSt-6
The Information Content of Dividend Policy on Future Earnings in Australia: A VECM Approach King Fuei Lee Senior Fund Manager, Schroder Investment Management 65 Chulia Street #46-00 OCBC Centre Singapore 049513 E-mail: [email protected] Tel: +65-6535 3411; Fax: +65-6535 3486 Abstract The main purpose of this paper is to investigate the existence of the dividend signalling effect in the Australian aggregate market by applying Johansen’s vector errorcorrection model (VECM) to conduct impulse response analysis and Granger-causality test. Our findings show that a unit shock increase in dividend payout leads to a permanent increase in future earnings over time, and that the payout ratio Granger-causes earnings. These results imply that there exists informational/signalling content in dividend payout in the Australian market over the long run. In addition, we also find evidence of Lintner’s (1956) sticky dividends in the impulse response function of payout to a unit shock in earnings, thus supporting the assertion of Kalay (1980) that managerial reluctance to cut dividends is a necessary condition to the existence of information content in dividends policy.
Keywords: Dividend signalling; Future earnings; Dividend payout; Australia; Impulse response function; Sticky dividends JEL Classification Codes: G35
1. Introduction The first prominent suggestion of the existence of information content in dividend policy was mooted in the seminal paper by Lintner (1956) where he examined the dividend behaviour of 28 US companies over the 1918-1941 period and noticed a pattern of sticky dividends and dividend smoothing by managers. In particular, he found that dividends are slow to adjust and tend to lag behind shifts in earnings because managers want to await confirmation of the sustainability of earnings changes before they change dividends. His empirical evidence however ran contrary to the dividend irrelevance theorem by Miller and Modigliani (1961) who hypothesised that in a world of perfect information, full capital mobility, no taxes and no agency costs, the dividend policy adopted by a company should be irrelevant to its market value. This conflict is commonly referred to as the dividend puzzle. Several theories that incorporate considerations for real-world market imperfections of asymmetric information have been advanced to explain Lintner’s observation. One of the key theories is the dividend signalling theory which hypothesises that because of asymmetric information between managers and shareholders, dividends can serve as a signal of the insider’s anticipation of the firm’s future performance. According to the theory, higher payout ratio should signal the manager’s expectations for higher future earnings i.e. there is information content in dividends on future earnings. There are however two pre-conditions for the signalling effect to exist. Firstly, there must be a cost for
International Research Journal of Finance and Economics - Issue 49 (2010)
69
bad companies trying to mimic a good company. Secondly, as shown by Kalay (1980), managerial reluctance to cut dividends is a necessary condition to convey information. This reluctance to cut dividends thus explains the stylised fact of “sticky dividends” in Lintner (1956). Empirical investigations into the relationship between dividends and future earnings have however yielded mixed results, with some studies finding a positive relationship between the variables (Healy and Palepu, 1988; Kao and Wu, 1994; Brooks, Charlton, and Hendershott, 1998), while others finding little or no evidence (DeAngelo, DeAngelo, and Skinner, 1996; Benartzi, Michaely, and Thaler, 1997). Most empirical studies have typically employed simple Ordinary Least Squares (OLS) techniques to perform their analyses. This can be undesirable as the use of OLS regression techniques to test the relationship between current dividend policy and future earnings leads to spurious results if the time series of payout and earnings are non-stationary like many other economic time series. The studies are also primarily focused on the US and European markets, with little work done for the increasingly important Asian-Pacific countries. Lee (2010) attempts to address some of these concerns in his study by investigating the signalling effect of dividend payout on future earnings in Singapore using impulse response functions under a VECM framework. His VECM approach therefore accounts for the non-stationarity of time series, while providing two added advantages. Firstly, the approach is able to include consideration for possible endogeneity between the variables while not making a priori assumptions about the nature of the inter-relationships. For instance, in explaining the trend stationarity of log dividends, Campbell and Shiller (1988) have attributed it to managers’ reluctance to raise dividends in line with earnings increases, while Linter (1956) had explained his observation of the slow move by companies towards a target dividend payout ratio as representing managers’ unwillingness to cut dividends paid to investors. This “stickiness” in dividends implies that current dividends and payout ratio are at least functions of both earnings as well as lagged values of dividends and payout ratios respectively. This means that the relationship between payout and earnings is better investigated within a system of simultaneous equations that are endogenously determined than as OLS regressions. Secondly, the investigation of the information content in dividend policy usually requires the identification of unexpected dividend policy changes which can be problematic. The difficulty in isolating and controlling investors' expectations means that most studies have either assumed a naive dividend expectation model (i.e. any change in dividends is unexpected) or employed some dividend forecasting model to capture investors' expectations. A VECM approach gets around that by applying Johansen’s vector error-correction model (VECM) to derive impulse response functions through which the unanticipated changes in dividend policy can be modelled as unit shocks to the variable in the VECM. Lee (2010) thus found evidence of the existence of information content in the dividend policy of the aggregate Singapore market. Despite his findings, a natural extension of his conclusions to all other Asian-Pacific countries can be difficult due to certain key differences between the tax regime and institutional structure of Singapore versus other countries. Firstly as highlighted by Lin (2002), the legal regime in which companies operate in and the institutional structure through which firms raise capital have significant impact on the extent of agency costs that exist in companies in the Asia Pacific region. These varying degrees of agency costs may affect dividend policy and hence the motivation for dividend signalling. In particular, while both Singapore and Australia operate under very similar common law regimes as a consequence of their Commonwealth backgrounds, the institutional structures between the two countries differ substantially. Lin (2002) found the Singapore institutional structure to be closer to a “Continental-German-Japanese” banking model (Mayer, 1990) while Australia’s structure approaches closer to an “Anglo-Saxon” capital market model. The key difference between a bank-oriented model and a capital market-oriented model lies in the extent of delegation of monitoring activities to banks, with banks in the formal model adopting a more dominant role in the monitoring of companies. As a consequence, the levels of agency costs differ under the two institutional structures, thus leading to different dividend policies existing in the two types of financial systems. Dewenter and Warther (1998) hence attributed the less sticky
70
International Research Journal of Finance and Economics - Issue 49 (2010)
dividends of Japanese companies, compared to their US counterparts, to the lower agency costs of Japanese firms as a result of their closer ties with the banks through cross-holdings. These results also mean that while the conclusion of the existence of informational content in dividends on future earnings in Singapore by Lee (2010) may be reasonably inferred for other common law/bank-oriented countries in the Asia Pacific region like Hong Kong, Malaysia and Thailand, it will not necessarily hold true for countries in the region which are common law/capital market-oriented like Australia and New Zealand. Secondly, the tax regime also differs significantly between the two countries. Australia operates a tax system where dividends are being taxed at the same rate as capital gains at the individual level, while in Singapore personal taxes are only applied to dividends and not to capital gains. As argued by the tax-effect hypothesis of dividends, the difference in tax regimes will have an impact on corporate dividend policy and hence the incentive for dividend signalling in Singapore versus Australia. These key differences make a separate detailed study of dividend signalling in Australia an interesting exercise in its own right. This paper therefore attempts to expand the current literature on dividend signalling by applying cointegration and vector error-correction modelling techniques to investigate the existence of informational content in the dividend payout in the “Anglo-Saxon” capital market model of Australia. In addition, the paper extends the analysis of Lee (2010) to also search for the presence of managerial reluctance to cut dividends which is a necessary pre-condition to the existence of dividend signalling. This pre-condition has typically been implicitly assumed in most dividend-signalling studies without any explicit investigation for it. This paper therefore also expands on current literature by conducting a simultaneous investigation into the existence of both the dividend signalling effect and the prerequisite of managerial reluctance to cut dividends within the same analytical framework. Specifically, this paper will apply Johansen’s cointegration techniques (Johansen, 1988; Johansen and Juselius, 1990) to the Australian market to develop a vector error-correction matrix (VECM) between earnings and dividend payout which is then used to derive the impulse response function for tracing the effect of a unit shock in the payout ratio on earnings over time. We also derive the impulse response function for tracing the effect of a unit shock in earnings on payout ratio, as well as conduct the Granger-causality test to establish the extent to which the lag process in dividend payout explains future earnings. The main findings of this paper can be briefly described as follows: 1. An unanticipated increase in the dividend payout, represented by a one standard deviation increase in dividend payout in the impulse response function, leads to a permanent increase in future earnings of the “Anglo-Saxon” capital-market model of Australia over time. This supports the dividend signalling theory that dividend payout possesses informational content about future earnings, and that the two variables are positively correlated. 2. Lagged values of the dividend payout ratio have statistically significant information about future real earnings, as evidenced by the rejection of the null hypothesis of no granger-causality between payout ratio and real earnings in the Australian market. 3. An unanticipated increase in real earnings is accompanied by an eventual permanent increase in the dividend payout but only after a lag. This supports the observation of sticky dividends by Lintner (1956) where companies only raise their dividend policies when they are sure of the sustainability of earnings increases. It also satisfies the assertion by Kalay (1980) of managerial reluctance to cut dividends as a necessary condition to the existence of information content in dividend policy. The paper is structured as follows: Introduction of this paper is presented in Section 1. Section 2 gives an overview of the econometric procedures employed. In this section, we use the JMulti package by Helmut Lutkepohl and Markus Kratzig, and the EasyReg package by Herman Bierens for the econometric analysis. Section 3 gives a description of the data used. This is followed by the empirical results and their interpretations in Section 4, while Section 5 provides the conclusions of the paper.
International Research Journal of Finance and Economics - Issue 49 (2010)
71
2. Econometric Methodology We define the two variables of real earnings as EPSt, and dividend payout ratio as, POUTt at time t. We then pursue the following econometric procedure. First, we test for non-stationarity and the order of integration for both POUTt and EPSt by employing the Augmented Dickey-Fuller (ADF) procedure (Dickey and Fuller, 1979, 1981), the Phillips-Perron (PP) unit root test (Phillips and Perron, 1988) and the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test (Kwiatkowski, Phillips, Schmidt and Shin, 1992) which uses a null hypothesis of stationarity. The tests are applied to both the model with constant and the model with constant and time trend, and performed for 0 to 60 lags i.e. five years. For the ADF test, the choice of the number of lagged differenced terms is made based on the four information criterion: Akaike Information Criterion (Akaike, 1971), Hannan-Quinn Criterion (Hannan and Quinn, 1979), Schwarz Criterion (Schwarz, 1978) and Final Prediction Error (Akaike, 1969). For the PP test and KPSS test, the number of truncation lags (l) to be used to evaluate the serial correlation for the NeweyWest correction (Newey and West, 1987) is computed by l = q(T/100)^0.25 where q = 4 or 12. For all the unit root tests, if non-stationarity is not rejected, the variable is differenced once and the unit root tests are performed again. This is repeated until stationarity is achieved. The number of differences taken before the series becomes stationary is then the order of integration ie I(d). If the two time series are found to be integrated of the same order, we then proceed to test for the existence of cointegration vectors among them by performing the Johansen maximum eigenvalue test, the Johansen trace test and the Saikkonen & Lütkepohl cointegration test (Saikkonen & Lütkepohl, 2000a; 2000b; 2000c; 2002). The cointegration tests involve the determination of the rank of the matrix Π which reflects the number of cointegrating vectors in the process governing movements of POUTt and EPSt. In this application, there are three possible ranks of Π. If r = 2 (i.e. Π is full rank), that means that both POUTt and EPSt are stationary processes. This would however contradict the earlier unit root tests findings that they are both integrated of order one or higher. If r = 0, then there is no cointegration between the two variables, and no stationary long run relationship exists. When r = 1, there is a single cointegrating vector between POUTt and EPSt such that Π = αβ’, where α contains the cointegrating vector and β is the corresponding error-correction coefficients. This supports the presence of a cointegrating relation between payout ratio and real earnings. If we find the existence of one cointegrating relation between the two variables of POUTt and EPSt, we can then proceed to derive the vector error-correction matrix (VECM) of forms: ∆EPSt = µ2 + Σk-1j=1 Г21(j) ∆POUTt-j + Σ k-1j=1 Г22(j) ∆EPSt-j + Π 21 POUTt-k + Π 22 EPSt-k (1) ∆POUTt = µ1 + Σk-1j=1 Г11(j) ∆POUTt-j + Σ k-1j=1 Г12(j)∆EPSt-j + Π11 POUTt-k + Π12 EPSt-k (2) where the matrix Г represents the short run dynamics of the relationship between POUTt and EPSt, and the matrix Π captures the long run information in the data. The order (lag length) of the Vector Autoregressive (VAR) form is chosen by selecting the VAR(p) model based on the four information criterion: Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. Lags of 0 to 60 months are tested. We perform two particular analyses to investigate the existence of informational content in dividend policy about future earnings. First, we test for causality in the VECM by applying the Granger-causality test. Granger (1969) defined a variable as being causal for another variable if inclusion of the lagged values of the former helps to improve the forecasts of the latter. If the dividend payout ratio contains information content about future earnings as hypothesised by the dividend signalling theory, then payout should granger-cause earnings. We next proceed to employ impulse response analysis which provides a fuller illustration of the dynamic interactions between the variables in the system by tracing the effect of innovations in one variable on the other variable over time. We are particularly interested in analysing the impulse response function of earnings to a unit shock in payout. We also examine the impulse response function of payout to a unit shock in earnings for signs of sticky dividends.
72
International Research Journal of Finance and Economics - Issue 49 (2010)
3. Data Sample This paper focuses on the Australian market. We have chosen to use the MSCI Australia index as it has the longest history of available data while providing a good representation of the country’s aggregate equity market given the index’s minimum 85% target representation guideline. The time period employed is therefore from December 1969 to March 2009. The data required are the monthly values of price-earnings ratio, dividend yield, the stock market index level and the consumer price index. We derive the 12-month trailing earnings in index points for the market by dividing the price series through by the price-earnings ratio. The real earnings series is subsequently obtained by dividing the earnings series through by the consumer price index. The payout ratio is derived by multiplying the dividend yield with the price-earnings ratio. The methodology for the derivation of the real earnings and payout series are consistent with the procedure of Arnott and Asness (2003) and Gwilym et al (2006). The logarithms of real earnings and payout ratio are used for this analysis. All data is downloaded from Datastream. The issue to note with these indices is that the survivorship bias in any index construction process affects the variation in index compositions over time. This means that the EPS of the index is likely to lag the GDP growth of the economy.
4. Empirical Findings 4.1. Results Figures 1 and 2 show the time series plots of the payout ratio (POUT) and real earnings (EPS) of the MSCI Australia index respectively versus the 1-year moving average over time. A visual inspection of the graphs suggests a lack of existence of time trending properties in both time series. Figure 1: Time Series Plot of Payout Ratio of MSCI Australia Index 4.70
4.60
4.50
4.40
4.30
4.20
4.10
4.00
3.90
3.80
Logarithm of Dividend Payout of MSCI Australia Index
c05
ec -0 7 D
De
ec -0 3
c01
D
De
ec -9 9 D
c95
ec -9 7 D
De
ec -9 3 D
c89
ec -9 1 D
De
c87 De
ec -8 5 D
ec -8 3 D
c81 De
ec -7 9 D
ec -7 7 D
c75 De
c73 De
ec -7 1 D
D
ec -6 9
3.70
Moving Average (12months)
Table 1 shows the results of the various unit root tests. We have tested all the variables in both levels and first-difference terms for non-stationarity using unit root tests with (1) a constant term, and (2) a constant term with time trend.
73
International Research Journal of Finance and Economics - Issue 49 (2010) Figure 2: Time Series Plot of Real Earnings of MSCI Australia Index -0.50
-0.70
-0.90
-1.10
-1.30
-1.50
-1.70
-1.90
-2.10
D ec D 69 ec D 70 ec D 71 ec D 72 ec D 73 ec D 74 ec D 75 ec D 76 ec D 77 ec D 78 ec D 79 ec -8 D 0 ec D 81 ec D 82 ec D 83 ec D 84 ec D 85 ec D 86 ec -8 D 7 ec D 88 ec D 89 ec D 90 ec D 91 ec D 92 ec D 93 ec D 94 ec D 95 ec D 96 ec D 97 ec D 98 ec D 99 ec D 00 ec -0 D 1 ec D 02 ec D 03 ec D 04 ec D 05 ec D 06 ec D 07 ec -0 8
-2.30
Logarithm of Real Earnings of MSCI Australia Index (rhs)
Table 1:
Moving Average (12months)
Unit Root Test Variables in Levels POUT EPS
Augmented Dickey-Fuller H0: Unit Root H1: Level Stationary H0: Unit Root H1: Trend Stationary with Drift Phillips-Perron H0: Unit Root H0: Unit Root with Drift
ADF(c) ADF(c+t)
-2.65 (7) -3.64 (7)*
-2.54 (6) -3.65 (6)*
-6.38 (6)** -6.40 (6)**
-5.78 (5)** -5.76 (5)**
H1: Level Stationary
PP(c)
H1: Trend Stationary
PP(c+t)
-12.46 (6) -13.95 (18) -21.35 (6)
-5.93 (6) -8.41 (18) -12.67 (6)
-543.33 (6)** -530.01 (18)** -542.45 (6)**
-510.02 (6)** -597.10 (18)** -509.49 (6)**
-24.26 (18)*
-17.14 (18)
-527.78 (18)**
-593.38 (18)**
2.63 (6)**
3.19 (6)**
0.08 (6)
0.06 (6)
1.10 (18)** 0.36 (6)**
1.30 (18)** 0.39 (6)**
0.09 (18) 0.03 (6)
0.04 (18) 0.03 (6)
0.16 (18)**
0.17 (18)**
0.04 (18)
0.03 (18)
Kwaitowski, Phillips, Schmidt and Shin H0: Level H1: Unit Root KPSS(c) Stationary H0: Trend Stationary
Variables in First Difference ∆POUT ∆EPS
H1: Unit Root with Drift
KPSS(c+t)
Notes: All variables in logarithms. Period: Jan 1990 – Jun 2007. Significance levels: ** = 1%, * = 5%. ADF-Tests: Critical values provided by JMulti and EasyReg. Lag lengths for ADF test according to the Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. Numbers in parentheses are the truncation lags (l) use to evaluate the serial correlation for the Newey-West correction for PP and KPSS tests, and are computed by l = q(T/100)^0.25 where q = 4 or 12.
In general, the unit root tests for non-stationarity (i.e. ADF and PP) fail to reject the null hypothesis of non-stationarity at 5% significance levels for both POUT and EPS in level terms, while
74
International Research Journal of Finance and Economics - Issue 49 (2010)
rejecting the null hypothesis at 1% significance levels for both variables in first-differenced terms. For the unit root test for stationarity i.e. KPSS, the null hypothesis is rejected at 1% significance levels for both POUT and EPS in level terms while the null hypothesis is not rejected at 5% significance levels for both variables in first-differenced terms. The unit root tests results therefore show strong evidence that the payout ratio (POUT) and real earnings (EPS) of the MSCI Australia index are non-stationary and are integrated of order one i.e. I(1). It is worth noting that when estimating the models with constant and time trend for use in the ADF tests, the coefficients of the time trend for both POUT and EPS are not significant. In addition, the inclusion of a time trend in the unit root tests yielded very similar results as the tests without time trends. This points at the lack of time-trending properties in both time series, and corroborates the earlier conclusions from our visual inspection of the graphs. We have however chosen to continue conducting our analyses for both the inclusion and exclusion of time trends for completeness. Having confirmed that both processes are I(1), we then proceed to determine the lag lengths to be used in the cointegration tests. This is done based on the four information criteria: Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error. We tested a maximum lag length of 50 lags and an optimal lag length of 7 lags is recommended. Table 2 shows the results of the Johansen maximum eigenvalue test, the Johansen trace test and the Saikkonen-Lütkepohl cointegration tests for both a model with intercept, and a model with intercept and time trend. All the cointegration tests reject the null hypothesis of no cointegrating relation at 5% significance level while accepting the null hypothesis of one cointegrating relation at 10% significance level. Table 2:
Cointegration Analysis of Dividend Payout and Earnings for MSCI Australia Index With Intercept
With Intercept and Time Trend
Test for the cointegration rank H0
H1
Test statistic
Critical Values 5%
10%
Test statistic
Critical Values 5%
10%
Johansen's Cointegration Maximum Eigenvalue Test r=0
r≥1
27.90
15.80
13.80
28.00
19.20
16.90
r≤1
r≥2
1.00
9.10
7.60
8.40
12.50
10.60
Johansen's Cointegration Trace Test r=0
r≥1
28.91
20.16
17.98
36.45
25.73
23.32
r≤1
r≥2
1.02
9.14
7.60
8.42
12.45
10.68
Saikkonen & Lütkepohl Cointegration Test r=0
r≥1
24.80
12.26
10.47
26.06
15.76
13.88
r≤1
r≥2
0.03
4.13
2.98
1.10
6.79
5.47
Number of cointegrating relations:
1
Choice of rank of Var(p): 7 Notes: r is the number of cointegrating relations. The rank of Var(p) is chosen based on Akaike Information Criterion, Hannan-Quinn Criterion, Schwarz Criterion and Final Prediction Error.
After establishing the existence of a cointegrating relation between the variables of POUT and EPS, we proceed to estimate a vector error correction model for analysis. The derived VECM is shown in Table 7 and Table 8 in the appendices. While an exhaustive interpretation of the individual coefficients is not within the scope of this paper, we note that the signs of the coefficients in the estimated cointegrating vector (with the coefficient of EPS normalised) are exactly as hypothesised by
75
International Research Journal of Finance and Economics - Issue 49 (2010)
the dividend signalling theory. In the long-run, real earnings is positively related to the payout ratio, hence suggesting the existence of informational content in dividend payout. Before the VECM is being applied for further analysis, we check whether the model is an adequate representation by conducting the Portmanteau test for autocorrelation and the LM test. The Portmanteau test for autocorrelation tends to be more useful for longer lag lengths, while the LM test is usually more applicable to the testing of low order residual autocorrelation. We therefore apply a lag order of 16 for our Portmanteau test, and a lag length of 5 for our LM test. The results are presented in Table 3 where it shows that the null hypotheses of no residual autocorrelation are not rejected at 10% significance level for both tests under the model with intercept and the model with intercept and time trend. Table 3:
Tests for Autocorrelation With Intercept
With Intercept and Time Trend
Test statistic
43.22
42.24
p-value
0.26
0.29
Adjusted test statistic
44.24
43.23
p-value
0.25
0.26
Lag order
16
16
Degrees of freedom
38
38
LM statistic
25.93
25.17
p-value
0.17
0.20
Lag order
5
5
Degrees of freedom
20
20
Portmanteau Test H0: No residual autocorrelation up to lag order H1: At least one autocovariance is nonzero
LM Test H0: No residual autocorrelation at lag order H1: At least one autocovariance is nonzero
We now proceed to investigate the existence of informational content in dividend payout on future earnings. We first perform the Granger-causality test to check for the existence of granger-causal relations between the payout ratio and real earnings of the Australian market, with particular focus on the test of the null hypothesis that POUT does not granger-cause EPS. The results of the test are shown in Table 4. The null hypothesis of no granger-causality between POUT and EPS is rejected at 5% significance level. This means that POUT granger-causes EPS i.e. lagged values of POUT provide statically significant information about the future values of EPS, and indicates the existence of an informational/signalling effect in current dividend policy on future earnings. Table 4:
Granger-Causality Analysis of Dividend Payout and Earnings for MSCI Australia Index
With Intercept H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS With Intercept and Time Trend
Test statistic
pval-F(1,7,894)
2.13
0.04
76
International Research Journal of Finance and Economics - Issue 49 (2010)
H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS
2.13
0.04
Figure 3: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Earnings and Dividend Payout 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
We next examine the effect of an unanticipated increase in dividend payout on future earnings. This is done by first deriving an impulse response function (IRF) tracing the response of EPS to a one standard deviation innovation in POUT. Figure 3 shows the impulse response function with the upper and lower limits of the 95% confidence interval. It can be seen that a shock increase in dividend payout leads to a positive and permanent increase in future real earnings, with the strongest positive impact experienced in real earnings that is 18 months forward. This confirms our earlier finding that changes in dividend policy provide information/signals about future real earnings. The impulse response function also shows that increases in the dividend payout leads to increases in future earnings. These findings are consistent with the dividend signalling theory. Figure 4 shows the impulse response function tracing the response of POUT to a one standard deviation innovation in EPS. It can be seen that a shock increase in real earnings leads to an eventual permanent positive increase in the payout ratio. However, it is also observed that the payout ratio initially drops in the first 7 months after the innovation before staging a recovery to an eventual permanent increase of approximately 0.65 standard deviation 43 months later. This suggests that even after an unanticipated increase in real earnings, management have typically allowed the payout ratio to drop initially, as dividends are kept unchanged while earnings rise, before they incorporate the permanent increase in real earnings into a permanent increase in dividend payout. This is consistent with the observation of sticky dividends by Lintner (1956) where he noticed that dividend policy tracks earnings with a lag as management awaits confirmation of the sustainability of earnings increases before they increase dividends. It also supports the assertion of Kalay (1980) that managerial reluctance to cut dividends is a necessary condition for the existence of informational content in dividend policy.
77
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 4: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Earnings and Dividend Payout 1.00
0.80
0.60
Impulse Response Function
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
Figure 5 and 6 show the impulse response function tracing the response of EPS to an innovation in POUT, and the response of POUT to an innovation in EPS respectively for the VECM with intercept and time trend. The findings are almost identical to the VECM with intercept only, and reinforce the dividend signalling theory as well as the observation of sticky dividends. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The stylised fact of sticky dividends observed by Lintner (1956) is also found in our analysis. These results support the dividend signalling theory or the information content of dividends hypothesis. Figure 5: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Earnings and Dividend Payout 1.40
1.20
Impulse Response Function
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
Forecast Horizon
36
38
40
42
44
46
48
50
52
54
56
58
60
78
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 6: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Earnings and Dividend Payout 1.00
0.80
0.60
Impulse Response Function
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
4.2. Robustness Test As with most VECM analysis, because the model is under-identified, if there exists a contemporaneous relation between the shocks of real earnings and dividend payout ratio, the results of the impulse response analysis will be very sensitive to the ordering of the variables in the VECM model. One of the ways to resolve this issue is to try a different ordering of the variables and see whether the results are significantly affected. We therefore repeat the construction of the VECM for both the models including and excluding time trends for a different ordering of the variables of EPS and POUT. Table 5:
Tests for Autocorrelation
Portmanteau Test H0: No residual autocorrelation up to lag order H1: At least one autocovariance is nonzero Test statistic p-value Adjusted test statistic p-value Lag order Degrees of freedom LM Test H0: No residual autocorrelation at lag order H1: At least one autocovariance is nonzero LM statistic p-value Lag order Degrees of freedom
With Intercept
With Intercept and Time Trend
43.03 0.26 44.04 0.23 16 38
42.24 0.29 43.23 0.26 16 38
28.53 0.10 5 20
25.17 0.20 5 20
79
International Research Journal of Finance and Economics - Issue 49 (2010) Table 6:
Granger-Causality Analysis of Dividend Payout and Earnings for MSCI Australia Index Test statistic
pval-F(1,7,894)
2.13
0.04
1.94
0.06
1.76
0.09
1.97
0.06
With Intercept H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS H0: EPS Does Not Granger-cause POUT H1: EPS Granger-causes POUT With Intercept and Time Trend H0: POUT Does Not Granger-cause EPS H1: POUT Granger-causes EPS H0: EPS Does Not Granger-cause POUT H1: EPS Granger-causes POUT
Figure 7: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Dividend Payout and Earnings 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
Figure 8: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept for Dividend Payout and Earnings 1.20
1.00
0.80
Impulse Response Function
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
36
38
40
42
44
46
48
50
52
54
56
58
60
80
International Research Journal of Finance and Economics - Issue 49 (2010)
Figure 9: Impulse Response Function of Real Earnings to a Unit Shock in Payout for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Dividend Payout and Earnings 1.60
1.40
Impulse Response Function
1.20
1.00
0.80
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
Forecast Horizon
Figure 10: Impulse Response Function of Payout to a Unit Shock in Real Earnings for MSCI Australia Index (with 95% Confidence Interval) for VECM with Intercept and Time Trend for Dividend Payout and Earnings 1.20
1.00
0.80
Impulse Response Function
0.60
0.40
0.20
0.00 0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
-0.20
-0.40
-0.60
-0.80 Forecast Horizon
Tables 9 and 10 in the appendices show the derived VECM models for the exclusion and inclusion of time trends. It can be seen that the signs of the coefficients in the estimated cointegrating vector is as hypothesised by the dividend signalling model. Table 5 shows the tests for autocorrelation, and the results demonstrate a lack of residual autocorrelation and the suitability of the models. The granger-causality tests results are shown in Table 6 where we see strong evidence that payout ratio granger-causes real earnings in the Australian market. The impulse responses functions of real earnings and payout to unit shocks in payout and real earnings respectively for the VECM models with intercepts, and with intercepts and time trend, are shown in Graphs 7-10. The results are the same as the models before, showing both informational content in dividend policy as well as signs of sticky dividends. When we compare the similar findings
International Research Journal of Finance and Economics - Issue 49 (2010)
81
between this paper and Lee (2010), it is interesting to note that despite the differences in tax regime and institutional structure between Singapore and Australia, the accompanying varying effects of agency costs and tax effect has not prevented the assertion of the dividend signalling effect. Our results are similar to that of Lin (2002) who found no statistically significant differences in the dividend policies of companies in common-law/bank-oriented countries versus firms in common law/capital market-oriented countries. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings in the “Anglo-Saxon” capital market model of Australia, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The stylised fact of sticky dividends observed by Lintner (1956) is also found in our analysis. These results support the dividend signalling theory or the information content of dividends hypothesis. Our results are robust to the under-identification of the VECM models.
5. Summary and Concluding Remarks This paper applies Johansen’s vector error-correction model (VECM) to investigate the existence of the dividend signalling effect in the Australian aggregate market using impulse response analysis and Granger-causality test. The main results for this paper can be summarized as follows: 1. An unanticipated increase in the dividend payout, represented by a one standard deviation increase in dividend payout in the impulse response function, leads to a permanent increase in future earnings of the Australian market over time. This supports the dividend signalling theory that dividend payout possesses informational content about future earnings, and that the two variables are positively correlated. 2. Lagged values of the dividend payout ratio have statistically significant information about future real earnings, as evidenced by the rejection of the null hypothesis of no granger-causality between payout ratio and real earnings in the Australian market. 3. An unanticipated increase in real earnings is accompanied by an eventual permanent increase in the dividend payout after a lag. This supports the observation of sticky dividends by Lintner (1956), and satisfies the assertion by Kalay (1980) of managerial reluctance to cut dividends as a necessary condition to the existence of information content in dividend policy. Our analyses have therefore established that changes in dividend policy provide statistically significant information about future earnings, with unanticipated increases in dividend payout leading to positive and permanent increases in future real earnings. The prerequisite of managerial reluctance to cut dividends to the dividend signalling theory is also satisfied by our empirical finding. These results support the dividend signalling theory or the informational content of dividends hypothesis, and are robust to the under-identification of the VECM models.
References [1] [2] [3] [4] [5]
Akaike, H., 1969. “Fitting Autoregressive Models for Prediction”, Annals of the Institute of Statistical Mathematics 21, pp. 243-247. Akaike, H., 1971. “Autoregressive Model Fitting for Control”, Annals of the Institute of Statistical Mathematics 23, pp. 163-180. Arnott, R. and C. Asness, 2003. “Surprise! Higher Dividends = Higher Earnings Growth”, Financial Analysts Journal 59, pp. 70-87. Ball, R. and R. Watts, 1972. “Some Time-Series Properties of Accounting Income”, Journal of Finance 27, pp. 663-81. Bernartzi, S., R. Michaely and R. Thaler, 1997. “Do Changes in Dividends Signal the Future or the Past?”, Journal of Finance 52, pp. 1007-1034.
82
International Research Journal of Finance and Economics - Issue 49 (2010)
[6]
Brooks, Y., R. Charlton and W., Hendershott, 1998. “Do Firms Use Dividends to Signal Large Future Cash Flows?”, Financial Management 27, pp. 46-57. Campbell, J. and R. Shiller, 1988. “Stock Prices, Earnings, and Expected Dividends”, Journal of Finance 43, pp.661-76. DeAngelo, H., L. DeAngelo and D. Skinner, 1996. “Reversal of Fortune, Dividend Signaling, and the Disappearance of Sustained Earnings Growth”, Journal of Financial Economics 40, pp. 340-371. Dewenter K. and V. Warther, 1998. “Dividends, Asymmetric Information and Agency Conflicts: Evidence from A Comparison of Dividend Policies of Japanese and US firms”, Journal of Finance 53, pp. 879-904. Dickey, D. A. and W.A. Fuller, 1979. “Estimators for Autoregressive Time Series With a Unit Root”, Journal of the American Statistical Association 74, pp. 427–431. Dickey, D.A. and W.A. Fuller, 1981. “Likelihood Ratio Statistics for Autoregressive Time Series With a Unit Root”, Econometrica 49, pp. 1057-72. Granger, C.W.J., 1969. “Investigating Causal Relation By Econometric and Cross-sectional Method”, Econometrica 37, pp. 424–438. Gwilym, O., J. Seaton, K. Suddason and S. Thomas, 2006. “International Evidence on the Payout Ratio, Earnings, Dividends, and Returns”, Financial Analysts Journal 62, pp. 36-53. Hannan, E. J. and B.G. Quinn, 1979. “The Determination of the Order of an Autoregression”, Journal of the Royal Statistical Society B, pp. 190-195. Healy, P. and K. Palepu, 1988. “Earnings Information Conveyed by Dividend Initiations and Omissions”, Journal of Financial Economics 21, pp. 149-176. Johansen, S., 1988. “Statistical Analysis of Cointegration Vectors”, Journal of Economic Dynamics and Control 62, pp. 36-53. Johansen, S. and K. Juselius, 1990. “Maximum Likelihood Estimation and Inference on Cointegration - With Applications to the Demand for Money”, Oxford Bulletin of Economics and Statistics 52, pp. 169-210. Kalay, A., 1980. “Signaling, Information Content, and the Reluctance to Cut Dividends”, Journal of Financial and Quantitative Analysis 15, pp. 855-869. Kao, C. and C. Wu, 1994. “Rational Expectations, Information Signalling and Dividend Adjustment to Permanent Earnings”, Review of Economics and Statistics 76, pp. 490-502. Kwiatkowski, D., P.C.B. Phillips, P. Schmidt and Y. Shin, 1992. “Testing the Null of Stationarity Against the Alternative of a Unit Root”, Journal of Econometrics 54, pp. 159-178. Lee, K.F., 2010. “An Empirical Study of Dividend Payout and Future Earnings in Singapore”, Review of Pacific Basin Financial Markets and Policies, doi:10.1142/S0219091510001949 Lin, C.T., 2002. “Dividend Policies, Legal Regimes and Institutional Structures in the Asia Pacific Region”, School of Finance and Business Economics Working Paper Series 02.04 Lintner, J., 1956. “Distribution of Incomes of Corporations Among Dividends, Retained Earnings, and Taxes”, American Economic Review 46, pp. 97-113. Mayer, C., 1990. Financial Systems, Corporate Finance and Economic Development, in: Hubbard, R.G. (Ed.), Asymmetric Information, Corporate Finance and Investment, University of Chicago Press, Chicago, IL. pp. 307.332. Miller, M. and F. Modigliani, 1961. “Dividend Policy, Growth, and the Valuation of Shares”, Journal of Business XXXIV, pp. 235-64. Newey, W.K. and K.D., West, 1987. “A Simple Positive Definite Heteroskedasticity and Autocorrelation Consistent Covariance Matrix”, Econometrica 55, pp. 703–708. Penman, S., 1983. “The Predictive Content of Earnings Forecasts and Dividends”, Journal of Finance 38, pp. 1181-1199. Phillips, P.C.B. and P. Perron, 1988. “Testing For a Unit Root in Time Series Regression”,.Biometrika 75, pp. 335-346.
[7] [8]
[9]
[10] [11] [12] [13] [14] [15] [16] [17]
[18] [19] [20] [21] [22] [23] [24]
[25] [26] [27] [28]
International Research Journal of Finance and Economics - Issue 49 (2010) [29] [30] [31]
[32] [33]
83
Saikkonen, P. and H. Lütkepohl, 2000a. „Testing for the Cointegrating Rank of a VAR Process With an Intercept”, Econometric Theory 54, pp. 159-178. Saikkonen, P. and H. Lütkepohl, 2000b. „Testing for the Cointegrating Rank of a VAR Process With Structural Shifts”, Journal of Business & Economic Statistics 18, pp.451-464. Saikkonen, P. and H. Lütkepohl, 2000c. „Trend Adjustment Prior to Testing for the Cointegrating Rank of a Vector Autoregressive Process”, Journal of Time Series Analysis 21, pp. 435-456. Saikkonen, P. and H. Lütkepohl, 2002. „Testing For a Unit Root in a Time Series With a Level Shift at Unknown Time.” Econometric Theory 18, pp. 313-348. Schwarz, G., 1978. “Estimating the Dimension of a Model”, Annals of Statistics 6, pp. 461-464.
84
International Research Journal of Finance and Economics - Issue 49 (2010)
Appendices Table 7:
Vector Error Correction Matrix (VECM) With Intercept for Earnings and Dividend Payout
∆EPSt
[
]
=
-0.03
[
∆Payoutt
]
[
[
+
[
[
-1.36
0.16
0.22
-0.22
-0.29
0.13
0.09
]
[
]
[
0.03
+
Table 8:
1.00
[
+
[
+
[
-0.02
-0.01
0.10 -0.05
0.04 -0.02
u1t u2t
]
[
]
[
EPSt-1 Payoutt-1
∆EPSt-1 ∆Payoutt-1 ∆EPSt-3 ∆Payoutt-3 ∆EPSt-5 ∆Payoutt-5
]
+
]
+
]
[
+
]
[
+
[
[
7.07
0.03
0.06
-0.08
-0.11
0.07
0.02
-0.10
-0.07
0.40 -0.20
0.11 0.02
]
[
]
[
]
[
]
[
CONST ]
∆EPSt-2 ∆Payoutt-2 ∆EPSt-4 ∆Payoutt-4 ∆EPSt-6 ∆Payoutt-6
]
]
]
]
]
Vector Error Correction Matrix (VECM) With Intercept and Time Trend for Earnings and Dividend Payout
∆EPSt ∆Payoutt
]
=
[
-0.02 0.03
]
[
[
+
[
+
1.00
-2.12
0.16
0.21
-0.22
-0.29
0.13
0.08
[
]
[
]
[
-0.01
]
+
]
+
∆EPSt-1
∆EPSt-3 [
]
+
[
0.10 -0.05
0.04 -0.02
]
+
[
-0.20 0.26
]
[
[
0.00
0.03
0.05
-0.08
-0.11
0.07
0.01
[
∆Payoutt-3
+
[
[
∆Payoutt-1
] -0.03
EPSt-1 Payoutt-1
∆EPSt-5 ∆Payoutt-5
]
+
[
CONST
]
+
[
]
[
]
[
-0.07
0.39 -0.20 u1t
0.10 0.02
u2t
]
]
∆EPSt-2 ] ∆Payoutt-2 ∆EPSt-4 ]
-0.10
TRENDt-1]
[
] ∆Payoutt-4
]
[
∆EPSt-6 ∆Payoutt-6
]
85
International Research Journal of Finance and Economics - Issue 49 (2010)
Table 9:
Vector Error Correction Matrix (VECM) With Intercept for Dividend Payout and Earnings -0.05
∆POUTt [
]=[
(-5.36)
POUTt-1 ][[
0.04
∆EPSt
1.00
-0.74
( --- )
('4.64)
][
]+
-5.22
[
EPSt-1
][
] CONST ]
(-21.24)
(4.09)
+[
+[
+[
-0.29
-0.22
(-4.11)
(-3.27)
0.22
0.16
(2.86)
(2.26)
-0.01 (-0.18) 0.09
-0.02 (-0.35) 0.13
(1.10)
(1.81)
-0.02 (-0.28) 0.04 (0.55)
-0.05 (-0.72) 0.10 (1.44)
∆POUTt-1 ][
]+[ ∆EPSt-1
∆POUTt-3 ][
]+[ ∆EPSt-3
∆POUTt-5 ][
]+[ ∆EPSgt-5
-0.11
-0.08
(-1.54)
(-1.17)
0.06
0.03
(0.74)
(0.42)
-0.07 (-0.93) 0.02
-0.10 (-1.44) 0.07
(0.25)
(1.00)
0.02 (0.23) 0.11 (1.48)
-0.20 (-2.96) 0.40 (5.57)
∆POUTt-2 ][
] ∆EPSt-2
∆POUTt-4 ][
] ∆EPSt-4
∆POUTt-6 ][
] ∆EPSt-6
u1t +[
] u2t
Table 10: Vector Error Correction Matrix (VECM) With Intercept and Time Trend for Dividend Payout and Earnings ∆POUTt [
]=[ ∆EPSt
-0.06 (-5.36) 0.04 (3.83)
][[
+[
1.00
-0.47
( --- )
(-2.81)
-0.29 (-4.07)
-0.22 (-3.29)
POUTt-1 ][
]+
[
EPSt-1
][
∆POUTt-1
0.00
][
(-1.27)
]+[
-0.11 (-1.50)
-0.08 (-1.19)
] TRENDt-1]
][
∆POUTt-2
]
86
International Research Journal of Finance and Economics - Issue 49 (2010)
+[
+[
+[
0.21 (2.77)
0.16 (2.19)
-0.01 (-0.16) 0.08 (1.00)
-0.03 (-0.37) 0.13 (1.74)
-0.02 0.27 0.04 (0.47)
-0.05 (-0.75) 0.10 (1.36)
0.26 (5.38) -0.20 (-3.82)
∆EPSt-1
∆POUTt-3 ][
]+[ ∆EPSt-3
∆POUTt-5 ][
]+[ ∆EPSgt-5 +[
]
[
0.05 (0.65)
0.03 (0.35)
-0.07 (-0.89) 0.01 (0.14)
-0.10 (-1.46) 0.07 (0.91)
][
0.02 (0.26) 0.10 (1.39)
-0.20 (-2.97) 0.39 (5.48)
][
u1t
CONST ]
] u2t
∆EPSt-2
∆POUTt-4 ] ∆EPSt-4
∆POUTt-6 ] ∆EPSt-6
