Does a change in dividend tax rates in the U.S. affect equity prices of non-U.S. stocks?

Research in this area primarily uses announcements by the Federal Open Market Committee (FOMC) to capture changes in U.S. monetary policy. The two FOMC announcements closest to the event window used in this paper are May 6, 2003, and June 25, 2003, making it unlikely I am picking up any effects from announcements about U.S. monetary policy (Gurkaynak et al. 2005).

Desai and Dharmapala (2011) assume the tax rate on capital gains is zero, which is why the expected equity price, E[P_T], is not reduced by taxes. This is not an unreasonable assumption, as Constantinides (1984) and Dammon et al. (2001) show that investors can reduce or eliminate capital gains taxes by accelerating the realization of capital losses and deferring the realization of capital gains. Empirically, Sialm (2009) shows that capital gains yields are relatively small, and Sialm and Starks (2012) document that the highest capital gains yields are realized by investors who will not pay current tax on the realizations (because the assets are held in tax-qualified retirement accounts).

The tax rate is the same for U.S. investors on U.S. and treaty country equities. Therefore the implications of the dividend tax rate cut for treaty country stocks will also apply to U.S. equities.

Until fairly recently, predictions about tax capitalization in the accounting literature were based on the logic of the marginal investor approach. Under the marginal investor approach, it is the marginal investor’s tax rate that determines the dividend tax premium (Ayers et al. 2002; Dhaliwal et al. 2003; Dhaliwal et al. 2007), rather than the weighted average tax rate. Guenther and Sansing (2010) correctly assert that the marginal investor approach is inconsistent with equilibrium pricing for a market with risky assets, which is why I motivate the hypotheses in this paper using the open-economy after-tax CAPM from Desai and Dharmapala (2011).

The tax rate change is assumed to apply only to the near term dividend, D, which implies the expected equity price, E[P], is unaffected.

Equation (2) does not change when allowing the stock prices for treaty and nontreaty country equities to covary (see equation (22) in Appendix 1).

For research about the relation between investor-level taxes and U.S. equities see Litzenberger and Ramaswamy (1979), (1980), Gordon and Bradford (1980), Miller and Scholes (1982), Naranjo et al. (1998), Dhaliwal et al. (2003), Dhaliwal et al. (2005), Sialm (2009), Auerbach and Hassett (2007) and Amromin et al. (2008).

Specifically, Wolff (2010) shows that, as a percentage of total wealth owned by U.S. households in 2001, the top 10% owned 85% of outstanding stocks. Given this finding, it seems likely that the wealthiest 10% of U.S. households, who owned a large portion of U.S. equities and by extension global wealth, were taxable U.S. investors. Importantly, the U.S. data in the Credit Swiss Global Wealth Databook (2010) and the data in Wolff (2010) both come from the U.S. Survey of Consumer Finance. In addition, individual net worth is defined in a similar way in both the Credit Swiss Global Wealth Databook (2010) and Wolff (2010), making comparisons between the studies appropriate.

There is variation in how much equity is held by U.S. investors in treaty countries. For example, based on 2003 holdings, U.S. investors owned 17% of equities in the United Kingdom but only 6% of the equities in Italy.

Another important assumption is that the dividend flow is certain. If the dividend flow is not certain, then the dividend tax rate change would affect the variances and covariances of the cash flows, and H3 may not hold. Nonetheless, prior literature suggests the assumption that dividend flow is certain is reasonable. For example, Linter (1956), DeAngelo et al. (2004), and Denis and Osobov (2008) show that dividends paid by firms in the United States and other large economies are sticky, so any variance effects are likely to be second order. Also, based on the observation that dividends tend to be sticky, researchers typically measure future dividends based on past dividends (Dhaliwal et al. 2007; Sialm 2009; Sialm and Starks 2012).

Desai and Dharmapala (2011) model the equilibrium price as a function of dividends paid (D_T). In my empirical specifications, I essentially transform equation (1) into a model of expected returns, similar to the original model developed by Brennan (1970), and examine the relation between stock returns and dividend yields.

Ideally I would include SMB and HML adjustments in my calculation of CAR, but unfortunately I cannot obtain daily SMB and HML factors for all of the countries in my sample. However, Ken French does provide regional monthly factors on his website. (The construction of these factors is detailed in Fama and French 2012.) I use these monthly factors to calculate each firm’s factor risk loadings for the regional SMB and HML factors and include these loadings as controls (β_SMB and β_HML, which are defined in Appendix 2). Other papers that have included a firm’s factor risk loading/historical sensitivity to the SMB and HML factors as control variables include Dhaliwal et al. (2007) and Konchitchki et al. (2016).

I thank Ken French for providing this information on his website http://​mba.​tuck.​dartmouth.​edu/​pages/​faculty/​ ken.​french/​data_​library.​html (accessed March 22, 2018).

Returns for Canadian listed firms are not available in Compustat Global or CRSP, so I obtain returns for these firms from DataStream.

More detail about the STOXX 1800 index can be obtained at https://​www.​stoxx.​com/​index-details?​symbol=​SXW1GR (accessed March 22, 2018).

Because dividend yield is, on average, positive for treaty country firms, Equation (2) suggests the mean CAR for the event window will be positive. One reason this may not be the case is transaction costs. Specifically, if transaction costs in treaty countries are large enough, the benefit of investing in the firm with the average dividend yield may not outweigh the cost, leading to a small amount of investment. However, as a stock’s dividend yield increases, the benefit of investing in that stock outweighs the cost of making the investment, which could lead the stock to experience abnormal returns. If this is a plausible explanation, I would expect the unconditional mean abnormal return for treaty country stocks to become positive for stocks above a certain level of dividend yield (which is when the benefits of investing outweigh the transaction costs). This is different than the main prediction in my paper, which predicts a positive association between CAR and dividend yield, because this association could exist, even if all returns for treaty country stocks are negative. I find that for firms with a dividend yield above the 75th percentile the mean CAR is positive (untabulated). Also, using data from Domowitz et al. (2001), Table 1, which contains the one-way cost of an equity transaction for most of the treaty countries, I estimate the weighted transaction cost for firms in my sample is 48.27 basis points, where the weight is determined using the number of observations from a country. Considering the size of the returns I document in Tables 3 and 4, the weighted transaction cost of 48.27 basis points is fairly large, suggesting transactions costs are likely to be a salient component of investment decisions, especially for lower dividend yield stocks. Thus transaction costs could partially explain why the mean CAR for treaty country stocks is not positive.

Examining raw returns for the portfolio tests is important, because firms included in the STOXX 1800 Indexes may be effected by the dividend tax rate cut. If this is the case, then removing the market effect from the raw portfolio return to calculate the abnormal return could be discarding some of the effect I am testing for; hence the importance of examining both the raw and abnormal return.

I find a similar result if I calculate cumulative abnormal returns by adjusting the firm return by the local market return, rather than using the market model. This result is also unchanged if I use local currency, rather than U.S. dollar-denominated returns, cluster the standard errors by country or industry, calculate firm specific betas using global rather than local market returns, and if I do not truncate the dependent variable (untabulated).

Important good news occurred on Friday May 23, 2003, when the World Health Organization lifted its travel advisory for Hong Kong (Dean and Richardson 2003). This travel advisory was lifted after the markets closed on Friday (Sanchanta 2003). Further, on Monday May 26, 2003, a Wall Street Journal article stated that, due to only a few new cases of SARS being reported, “Taiwan’s SARS outbreak showed further signs of stabilizing” (Dean and Richardson 2003). I examine mean daily abnormal returns for my sample of nonmicrocap, nontreaty stocks and find that the majority (0.0056) of the mean CAR reported in Table 6 (0.007) occurred on May 26, 2003, consistent with the large CAR being related to good news about the abatement of SARS.

Assuming dividends are deterministic is important for the following reason. As shown in Sikes and Verrecchia (2012), an investor-level tax cut will have two effects on equity prices. First, the tax cut will increase expected after-tax cash available for investors (the payoff effect). An increase in expected after-tax cash drives up share prices and decreases expected pre-tax rates of return. “However, in a diversified market setting along the lines of the Capital Asset Pricing Model (CAPM), [a decrease in investor-level tax rates] also [increases] risk because the tax authority absorbs [less] of the risk associated with firms’ residual cash flows (variance of the payoff effect). When risk [increases], share prices [decrease] and, thus, expected pre-tax rates of return [increase]” (Sikes and Verrecchia 2012, p. 1068). Thus the payoff effect and the variance of the payoff effect could offset each other. However, assuming dividends are deterministic ensures the expected payoff effect of a dividend tax rate change will dominate any effects to the variance of the payout.

An important assumption of the model is that the dividend tax rate cut is not permanent and only applies to the next dividend. If this assumption was not made, I shouldn’t see different stock price reactions for firms that are currently paying high dividends verses firms that are currently paying low dividends, because the effect of the dividend tax rate cut on share price should be independent of when the dividends are paid. Importantly, this assumption fits the expectations investors had about the permanence of JGTTRA. Specifically, the tax rate decrease for dividends was originally set to expire on December 31, 2008. However, it has been extended several times and finally rose to 20% for individuals making more than $400,000 in 2013. The expectation that the dividend tax rate would rise in the future is important, as it means the timing of dividend payments matters and suggests firms with high dividend payouts would have greater increases in stock price.