Dynamics of Corporate Dividend Policy under Hyperinflation and Dollarization: A Quantile Regression Approach

Purpose: Zimbabwe experienced hyperinflation (2000-2008) followed by dollarization from 2009 onwards which had implications on dividend policy. In this context, this study isolates the main determinants and examines their behaviour across the distribution of dividend policy. Design/methodology/approach: The study employs quantile regression analysis and a sample of 30 firms listed on the Zimbabwe Stock Exchange (ZSE), covering the period 2000 to 2016. The fixed effects (FE) analysis is applied as a base model. Finding(s): The most robust determinants are ownership structure, earnings per share (EPS) and taxation. In our context, results are more informative, than those based on FE analysis by showing the change in the impact of each explanatory variable across the distribution. EPS has a positive and significant impact on dividend policy throughout the distribution in both sample periods. Its effect increases in magnitude as firms move from low to high quantiles. The other variables are useful in explaining dividend policy at selected points of the distribution. Thus, there is clear heterogeneity in the determinants of dividend policy. Research limitations/implications: The study shows the importance of developing dividend policy by focusing on the position of the firm on the distribution. Dividend policy should be developed in view of the earnings potential of the firm, ownership concentration and perceived changes in fiscal policy. A well-designed policy should have a differentiated approach to influencing corporate dividends. Originality/value: This study enhances our understanding of dividend policy in unique markets. It confirms the applicability of dividend relevance theories. Furthermore, It shows that quantile analysis provides more reliable estimates than those obtained using standard panel data models.


Introduction
Zimbabwe's economy experiences structural changes between 1997 and 2019. This is triggered by both political and economic factors. In 1998, the government embarks on a reform to compulsorily acquire land from the white minority and give it to the landless black majority (Mandizha, 2014). Unbudgeted gratuities are paid to war veterans and the government supports the war in the Democratic Republic of Congo. The international community does not support such decisions and multilateral institutions like the International Monetary Fund (IMF), Africa Development Bank and the World Bank (WB) withdraw financial support. In response, the government prints money to finance its activities resulting in hyperinflation from 2000 to 2008. The effects include: the worsening of the exchange rate, loss of import cover, fall in export revenue, negative gross domestic product (GDP) growth, deterioration in the balance of payments position and fall in production levels and a rise in unemployment. By the end of 2008 the official inflation rate reaches 231 million percent (Makochekanwa, 2007, Mandizha, 2014, Kararach Kadenge and Guvheya, 2010. During this period of high inflation, some firms pay dividends to retain investors. More so, the stock market provides a hedge against inflation and trading activities remain high (Jagongo and Mutswenje, 2014). The money market instruments perform badly due to high inflation and investors shift their portfolios to stock market related securities (Njanike, Katsuro, and Mudzura, 2009). Electronic copy available at: https://ssrn.com/abstract=3881056 However, inflation falls to single digits in February 2009 following the adoption of a multicurrency regime (Kanyenze, Chitambara andTyson, 2017, Zhou andZvoushe, 2012). The economy is partially dollarized, and the United States Dollar, South African Rand and Botswana Pula become legal tender. The speculative activities cease, and firms shift their focus to generating production profits (Sikwila, 2013, Njanike et al, 2009). The Reserve Bank of Zimbabwe (RBZ) loses its lender of last resort function which limits the availability of liquidity in the market. The market is still unstable which affects the level of dividends distributed by firms to shareholders as well. Hence formulating corporate dividend policy is still important for firm managers under dollarization period.
Previous studies (Edwards andMagendzo, 2001, Nor, 2012) on hyperinflation and dollarization fail to discuss these structural changes in view of corporate dividend policy. Understanding the determinants and dynamics of dividend policy in these periods adds to the current debate in corporate finance. Literature lacks studies that directly examine dividend dynamics in this context. This is despite the possibility that the predictive power of main dividend theories may be lost under these circumstances. Previous discussions are mainly based on standard panel data models which provide conclusions based on mean values of explanatory variables. They fail to fully exploit the critical information at different points of the distribution of dividend policy. In view of this, this study extends the current literature by employing quantile regression techniques to obtain useful information due to the presence of heterogeneity in the firms' dividend policy. The effect of each explanatory variable is sensitive to the position of the firm on the distribution of dividend policy. This is tested and confirmed using data for Zimbabwe firms.
The study offers some new insights by showing that there is clear heterogeneity in the determinants of dividend policy. The most robust determinants are ownership structure, earnings per share and taxation. In our context, results are more informative, than those based on fixed effects analysis by showing the change in the impact of each explanatory variable across the distribution. Earnings per share has a positive and significant impact on dividend policy throughout the distribution in both sample periods. Its effect increases in magnitude as firms move from low to high quantiles. The other variables are useful in explaining dividend policy at selected points of the distribution. Thus, quantile analysis provides more reliable estimates than those provided by standard panel data models. This study enhances our understanding of dividend policy in unique markets. It confirms the applicability of dividend relevance theories.
The rest of the study is organized as follows: section two summarizes the key theories underpinning this study and the main determinants of dividend policy, section three discusses the methodology applied in this study, section four discusses main results and section five concludes and provides policy implications.

Literature Review
This study is done under imperfect market conditions as such dividend policy affect firm value (Gordon, 1963, Lintner, 1962. It is underpinned by dividend relevance theories which are summarized as follows: the bird in hand theory, by Lintner and Gordon, shows that investors prefer current dividends which have low risk. They tend to discount future cashflows with a lower rate which increases the value of the firm. A firm that does not pay dividends experiences low firm value (Gordon, 1963, Lintner, 1962; the agency costs theory shows that the payment of dividends reduce the agency problem by removing excess cash which might be misused by managers (Easterbrook, 1984) and the clientele theory (Allen at el, 2000, Seida, 2002 shows that the payment of dividend attracts institutional investors due to low taxation. Investors' preference of dividend payments is determined by the level of taxation. Those in higher tax brackets prefer shares with low or no dividends while those in low tax brackets prefer cash dividends.

Determinants of Dividend Policy Past Dividends
Thus far, literature is clear on the impact of previous dividend payouts on current dividend policy. Studies (Edmund, 2018, Mirbagherijam, 2014, Tran and Nguyen, 2014 have shown that past dividend payments are a good predictor of future dividend policies. Investors can use the payment of dividend as a signal for the future prospects of the firm. Firm managers are reluctant to reduce dividends payments even during inflationary periods. They continue to make disbursements to mimic good prospects for current and potential investors.

Financial Leverage
Highly leveraged firms (LEV) pay less dividends due to high debt service costs (Hosain, 2016, Edmund, 2018, Fliers, 2017. Such firms are exposed to the risk of bankruptcy. More dividends can be paid where a firm relies on other sources of cash flows (Nguyen et al, 2013, Ahmad andJavid, 2009). Payment of dividends may differ according to debt composition. Firms may be willing to acquire more debt to finance dividend payouts which acts as a signaling device to shareholders. However, some studies (Rizqia andSumiati, 2013, Alzomaia andAl-Khadhiri, 2013) argue that financial leverage has no effect on dividend policy.

Investment Outlays
Previous studies (Al-Najjar andBelghitar, 2011, Ahmed andJavid, 2009) show that high investment expenditure (INV) reduces the likelihood of paying dividends. Firms with more investment opportunities may source external funding where access to financial markets is easy and they can still maintain high dividend payouts. Bildik, Fatemi, and Fooladi (2015) argue that large firms can still pay dividends in the absence of credible growth opportunities.
Their study confirms the positive relationship between investment decisions and dividend policy. This is consistent with theoretical propositions (Adediran andAlade, 2013, Lahiri andChakraborty, 2014) that firms can make investment and dividend decisions concurrently.

Earnings per Share
Literature confirms the linkages between earnings per share and dividend policy. For previous studies (Basse and Reddeman, 2011, Adediran and Alade, 2013, Bassey, Asinya, and Elizabeth, 2014, King'wara, 2015 show that high earnings per share (EPS) guarantee the payment of more dividends. Again, firms may not necessarily make huge dividend disbursements as they seek to retain funds for future use. Ahmed and Javid (2009) argue that though dividend policy is dependent on earnings per share (EPS) and past year's dividends, it is more sensitive to the former than the latter.

Managerial Ownership
The agency theory shows that managerial ownership is related to dividend policy. Studies (Björn andLantz, 2016, Ahmed andJavid, 2009) show that more dividends are paid where managers seek to reward themselves using free cash flows. On the other hand, high managerial ownership (OWN1) may cause managers to postpone the payment of dividends and invest so as to increase the firm's future income generating capacity (Kania and Bacon, 2005, Mirza and Azfa, 2010, Bushra and Mirza, 2015. Low dividends in firms with high inside ownership are explained by the desire by management to increase the expected value of their stock options which they receive as compensation. However, studies like Arshad et al (2013) and Hosain (2016) show that in the case of Pakistan firms inside ownership and dividend policy have no relationship

Institutional Ownership
The presence of institutional shareholders brings discipline among managers who are deterred from overinvesting a firm's excess funds. Past studies (Bozec et al, 2010, Björn andLantz, 2016) show that institutional ownership (OWN5) promotes the payment of dividends where a firm is at the high quantile of its growth opportunities. In this case the firm may be having more excess cashflows which are useful for rewarding owners. On the other hand, institutional shareholders may restrict payment of dividends and advocate for more money to be spent on growth opportunities (Kania andBacon, 2005, Bushra andMirza, 2015). However, Mossadak, Fontaine, and Khemakhem (2016) argue that institutional ownership has no effect on dividend policy.

Taxation
The taxation policy of the firm affects investor choices. Institutional investors and the elderly prefer dividend paying stocks since tax on dividends is low. Taxation (TP) reduces funds available for payment of dividends (Berzins, Bohren and Stacescu, 2017, Morck and Yeung, 2005). On the other hand, previous studies (Atia, 2017, Amidu andAbor, 2010) show that taxation has a positive relationship with dividend payout where firm managers have chosen a certain dividend policy, desire to use dividends as an investor retention strategy or have access to other financing alternatives. Chetty and Saez (2010) argue that corporate taxation does not distort the ability of a firm to pay more dividends in contrast to the agency cost theory. It has an insignificant effect on dividends (Gul et al, 2012, Khan, Jehan andShah, 2017).

Firm Size
Large sized firms (SIZE2) pay more dividends as they are likely to be financially stable (Arshad et al, 2013, Michaely andRoberts, 2012). Similarly, Bildik et al (2015) show that large firms can still pay dividends in the absence of credible growth opportunities, but they have to be profitable (Kowalewski, Stetsyuk and Talavera, 2007). On the other hand, King'wara (2015) argues that large firms could have taken more debt to finance their current levels of growth. This reduces the ability to make dividend payments in the short term.

Inflation and Money Supply
Inflation (INFN) and money supply (MSP) are useful in controlling for hyperinflation and dollarization respectively as firms design their dividend policy. Basse and Reddeman (2011) argue that firms pay more dividends even when they are faced with high inflation. They can still rely on high nominal earnings. Firms are expected to have reduced dividends payout under hyperinflation (Elly andHellen, 2013, Edmund, 2018) and more payouts during dollarization period. On the contrary, Mirbagherijam (2014) argues that dividends signal bad future prospects. Chronic high inflation results in a fall in the earnings as well as dividends. Pandey and Bhat (2004) show that money supply has a positive effect on dividend policy. On the contrary, Akyildirim et al (2013) supports the proposition that high money supply reduces the payment of dividends. Furthermore, Mambo (2012) argues that monetary policy activities have no effect on dividend policy.

Model Specification
Dividend policy (PR) is measured using dividend per share. The choice is determined by its usage in literature, diagnostic tests and giving of better results consistent with Zimbabwean context. The dependent variable, PR, is specified as function of the firm and macroeconomic variables and their expected signs are guided by literature. The generalized model is stated as: x as represented by the firm and macro variables.
Quantile regression model (Koenker and Basset, 1978) is used to analyse the effects of each explanatory variable on corporate dividend policy in different quantiles. It helps in exploring, accurately, the determinants of dividend policy. This approach helps in understanding the effects of each variable by looking at the sign of the coefficient, size and level of significance across the distribution. It gives better results than those given by OLS models (Fattouh, Harris and Scaramozzino, 2008). The design matrix bootstrap method is used to estimate standard errors for coefficients (Buchinsky, 1998). Confidence intervals are constructed using the percentile method (Koenker and Hallock, 2001). Estimations are done using nine quantiles: 0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70, 0.80 and 0.90. The model estimated is specified as follows: The study also employs the fixed effects model to provide base results for comparison with quantile regression. Robust standard errors are employed. Potential endogeneity is tested by checking the robustness of estimates. This is done by removing or adding explanatory variables to see if results remain unchanged (See results in Tables 5 and 6 in the appendix). A sample of 30 non-financial firms, is used, that are listed on the ZSE, giving a total of 510 firm years. Annual data on firm characteristics and macroeconomic variables is extracted from financial statements of firms and WB (2017)  Electronic copy available at: https://ssrn.com/abstract=3881056 *The Table contains definitions of main variables as they are applied in this study.

Diagnostic Tests and Descriptive Statistics
The study employs Levin, Lin and Chu (LLC) and Im, Pesaran and Shin (IPS) to test for unit root at 5% level of significance and results show that all variables are stationary at levels. The study considers multicollinearity using Pearson correlation matrix. Correlation coefficients are mostly less than 0.5 which implies that there is limited multicollinearity between any pair of variables. Thus, all the variables could be used in the same model without giving spurious results. All the variables (Table 2) are positively and highly skewed except for the measure of investment decisions. Variables in the analysis are leptokurtic with a measure of kurtosis higher than 3. The study shows that the average dividend per share is US$0.025. The average firm growth rate is 11.9% and firms are not highly geared since the average level is slightly above 50%. The average inflation rate is 136190.1% during the period of analysis. Management hold about 8.6% of shares while institutional shareholders hold 74.4% of shares on average. The average of money supply is 55% of GDP for Zimbabwe. Earnings per share are US$0.041 while the tax paid is around 12% on average. The level of investment fell, on average, by about 16%, during the period of analysis.  Table 3 and Figure 1 show results during hyperinflation period while Table 4 and Figure 2 show results under dollarization. The most robust determinants of corporate dividend policy, in both periods, are ownership structure variables, earnings per share and taxation. Findings show that investment decisions, firm growth (Edmund, 2018), leverage (Rizqia and Sumiati, 2013  -5.03E-10 -9.97e-11 -6.25e-10 -4.09e-10 -7.37e-10 -6.69e-10 -7.73e-10 -1.94e-09 -2.05e-09 -1.73e-09 OWN1 0.0505*** 0.0248*** 0.0249*** 0.0216*** 0.0219*** 0.0223*** 0.0152*** 0.0252*** 0.0393*** 0.0491*** SIZE2 -0.0012 -3.71e-05 -8.32e-05 -6.91e-05 -0.  Electronic copy available at: https://ssrn.com/abstract=3881056  The study shows that insider ownership has a positive effect on dividend policy which is consistent with the results based on the FE model. However, quantile regression shows clear variations in the magnitude of coefficients as firms move from low to high quantiles. Under hyperinflation, there is evidence of managerial entrenchment throughout the distribution. Under dollarization, insider ownership is important from 30 th quantile onwards. In both periods, the effect of inside ownership increases as firms move to higher levels of the distribution of dividend policy. Findings are consistent with past studies (Mossadak, Fontaine andKhemakhem, 2016, Gowri andSaravanan, 2016) which support strong managerial entrenchment as firms move towards higher levels of dividend policy.

Quantile Regression Analysis
Institutional ownership has a negative effect in both structural periods. Firms with high institutional ownership can successfully reduce the payment of dividends. This is consistent with previous studies (Reyna, 2017, Yusof andIsmail, 2016). This study shows that their monitoring role is effective for firms with high dividend policy or payout ratios. The significance of institutional ownership is in the 90 th quantile and starts from 60 th quantile under hyperinflation and dollarization respectively.
Under hyperinflation, expansionary monetary policy reduces the payment of dividend at the higher level of the distribution of dividend policy. This is consistent with Akyildirim et al, (2013) who suggest that high money supply is inflationary which further erodes cashflows meant for dividend payouts. Under dollarization, the effect of money supply is insignificant. Consistent with Mohsin and Ashraf (2011), the results show that expansionary monetary policy has no effect on dividend policy. Firms have alternative sources of finance to improve dividend payouts.
The positive impact of earnings per share differs across the conditional distribution of firms' dividend policy in both periods. The magnitude of the coefficient increases as firms move from lower to higher quantiles. This shows the persistent impact of earnings per share on dividend payout as predicted by theory. Thus, at higher levels of dividend policy, firms with higher earnings pay more dividends as supported by previous studies (King'wara, 2015, Bassey et al, 2014. Taxation has a positive impact on dividend payout at lower quantiles, under hyperinflation upto the 60 th quantile. The positive effect is consistent with past studies (Atia, 2017, Amidu andAbor, 2010). This may indicate their desire to maintain a certain level of dividend policy and more so their ability to source funding elsewhere. At high levels of dividend policy, profitable firms are not considering the effects of taxation when deciding to pay or not to pay dividends. However, under dollarization, taxation has a negative effect within the 90 th quantile. In our context, this shows that firms face the real burden of taxation at higher quantiles considering that the economy is using a stable currency. Previous studies (Arif andAkbarshah, 2013, Chuang et al, 2018) show that higher taxation adversely affects remuneration for shareholders.
The study shows that firm size explains dividend policy under hyperinflation and it is significant in the 90 th quantile. Thus, at higher levels of dividend policy, larger firms are likely to increase their dividend payout ratio and move towards the upper levels. This confirms the argument by Arif and Akbarshah (2013) that large firms have more access to debts market and hence experience fewer external constraints. Dividend policy is not sensitive to firm size under dollarization.
Investment decisions are important under dollarization and have a positive effect between the 40 th and 70 th quantiles only. The positive effect is consistent with Bildik et al (2015). Thus, policy makers' focus should be on firms that are within this range since they are likely to withstand any financial constraints within the economy and continue to pay dividends. Such firms can access alternative sources of investment funds like debt and retained profits. Our results suggest that a study like Lestari (2018), showing that investment has no effect on dividend policy may have analysed behavior of firms at the lowest parts of the distribution.
Consistent with past studies (Bushra and Mirza, 2015, Cristea andCristea, 2017), this study shows that firm growth has a negative effect on dividend policy under dollarization. Thus, firms at the high level of the distribution reduce payment of dividend to free up funds for taking up new opportunities. More so, the study shows that leverage has a positive effect on dividend policy from the 70 th quantile. This is consistent with past studies (Thirumagal andVasantha, 2017, Gowri andSaravanan, 2016) which show that firms are willing to acquire more debt to finance dividend payouts which acts as a signaling device to shareholders. This behavior is expected where firms are not afraid of the risk of bankruptcy as they are exposed to more debt.

Conclusion
This study contributes to corporate finance literature by examining dynamics of dividend policy under unique market conditions. It identifies the major determinants and examines their contribution at different positions of the firm's dividend distribution. By using quantile regression analysis, the study brings useful information which is critical for policy making. The study confirms the importance of dividend relevance theories by showing role played by institutional shareholders, managerial share ownership and taxation. The results are more informative than those from previous studies which are based on OLS methodology. They indicate the points, on the distribution, at which key variables affect dividend policy. For example, under hyperinflation firm size, money supply and institutional ownership have a positive, negative and negative effect, respectively, on dividend policy at 90 th quantile. Earnings per share and inside ownership are positive and significant throughout the distribution while taxation has a positive effect at lower quantiles. Under dollarization, the behavior of institutional ownership and firm growth suggests a non-linear relationship with dividend policy since the sign of the parameter changes from positive to negative. Insider ownership is significant from the 30 th quantile onwards. Other variables like leverage, taxation and investment decisions are important at specific points on the distribution. These results provide a firm foundation for understanding dividend policy in markets under unique conditions. They show the importance of developing policies by focusing on the position of the firm on the distribution of dividend policy. A dividend policy that focus on reducing informational inefficiencies would be desirable for the Zimbabwean market.  239 *** significant at 1%; ** significant at 5%; *significant at 10% The models are estimated using FE and with robust standard errors. The first model contains all explanatory variables which are removed one at a time to check for robustness. The results remain fairly stable for all models. 0.0425** 0.0367** 0.0413** 0.0458*** 0.0352** 0.0217*** 0.0351** 0.0453** FG -0.0021*** -0.0023*** -0.0019*** -0.0022*** -0.0021*** -0.0024*** -0.0008** FLEV6 -0.0013** -0.0016** -0.0014** -0.0014* -0.0010** -0.0013* -0.0017** INFLN -5.03E-10 -5.18E-10 -6.59E-10 -5.24E-10 -5.02E-10 -4.33E-10 -3.65E-10 OWN1 0.0505*** 0.0526*** 0.0504*** 0.0506*** 0.0487*** 0.0509*** 0.0529*** SIZE2 -0.