Stock prices and dividends: A South African perspective

ABSTRACT

The present value model (PVM) is considered a cornerstone of financial theory and practice, yet the limited empirical tests thereof have produced varying results depending on the method used. In addition, these tests have focused principally on developed countries, with little knowledge of the validity of the PVM for developing markets. This study employs firm-level data and advanced panel unit root and cointegration tests to assess the PVM for South Africa using a sample of firms that have consistently paid dividends over the period 1999 to 2018. The results largely support the assertion of the model that the stock price is a function of future dividends but the relationship does not move in perfect harmony.

KEYWORDS:

• panel cointegration techniques
• present value model
• stock prices
• dividends
• firm-level data

Acknowledgements

The authors would like to thank the anonymous reviewers for helpful comments on the previous drafts of this paper.

Notes

1 Gross rather than net dividends were used as they fully reflect the value of the firm’s distribution of profits to shareholders. In addition, they are also unaffected by changes in the tax laws in South Africa such as the replacement of the Secondary Tax on Companies in 2012 with the Dividend Tax and the subsequent increase in the Dividend Tax rate from 15% to 20% in 2017.

2 To a maximum of four lags due to the small number of time series observations.

3 A Brownian bridge is a stochastic process that has a conditional distribution of a Weiner process (W(t)), also referred to as a Brownian motion (Chow, 2009). While a Brownian motion is pinned at the origin at t=0 (there is a known starting point), a Brownian bridge is tied down at the origin at t=0 and at t=T (at both the start and end points) (Chow, 2009). The process is thus characterised by uncertainty in the middle of the bridge but no uncertainty at either end (Chow, 2009).

4 While Kao (1999) also proposed two Dickey-Fuller based t-statistics, only the ADF test is used as it accounts for autocorrelation in the residuals.

5 The optimal number of leads and lags were determined using the AIC but only to a maximum of two as there were insufficient observations to allow for a maximum of four leads and lags.

6 A Bartlett kernel window of three was computed for the semi-parametric estimation of the long run variances based on the number of time series observations (Persyn & Westerlund, 2007).

7 For the purposes of brevity, the details of FMOLS are not discussed but the reader is directed to Pedroni’s (2001) paper for more information.

8 To calculate the average aggregated stock price and dividend series, the annual dividend-price ratio and compound growth rate in the share price were calculated for every year. Each firm’s stock price was then equated to 100 in 1999, with the price in each year thereafter computed by multiplying the new value by one plus the growth in the original stock price series over the year. Thereafter a new dividend value was obtained by multiplying the new stock price series by the dividend-price ratio. These re-indexed dividend and price series were then averaged across the 22 firms and the natural log of each series obtained, as shown in Figure 1.

9 These results are not presented for the purposes of brevity but are available from the authors upon request.

10 The FMOLS and DOLS regressions were also estimated for two sub-periods: 1999-2008 and 2009-2018 based on the visual evidence presented in Figure 1, which suggested there may be a breakpoint in the stock price-dividend relationship after the financial crisis. However, the long run coefficient did not differ materially across the two sub-periods for any of the specifications. These results are available from the authors upon request.