SK531 Economics of the rm, Diderik Lund, 1 September 2000 Signaling by means of dividends? Bernheim and Wantz, AER (1995) Background Under classical tax system: Dividends disfavored Better: Retain prots in rm Moreover: Buy back shares (if not taxed) Nevertheless: Do actually observe dividends Page 532, rst column: \rms derive an advantage from the payment of dividends : : : " Interpretation: Tax system implies no dividends. Must exist some other advantage which counteracts tax disadvantage. Contrast to Norway: Tax system after 1992 does not discriminate against dividends. No basis for doing the same kind of research. Another, more specic point: Test in Bernheim and Wantz (1995) is based on variation in personal tax rates for dividends, not present in Norway. 1 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Why dividends in spite of double taxation? Three hypotheses: 1. Signaling high value 2. Limits powers of management 3. (Unexplained) shareholders' preference for dividends (| are they credit rationed?) All hypotheses imply: Higher dividend gives higher value of rm ('s shares). Observing this is not sucient to distinguish between hypotheses. Sequence of events (one rm, one year): Dividends are announced Stock market reacts to announcement, share price usually an increasing function of announced dividends (Some weeks later:) Dividends are paid out (= \share goes ex-dividend") Vt+1 Vt , Dt (time measured in days) t + 1 called \ex-dividend date" 2 SK531 Economics of the rm, Diderik Lund, 1 September 2000 New in Bernheim and Wantz: Test based on variation in tax rates No longer question whether announcement of higher dividend leads to higher share value But: Can we see systematic dierences in the rst derivative of this relationship (\bang-for-the-bucks") when tax rates vary? And: Can this tell which of the hypotheses is right? Presumes variation in tax rates over time (or between rms, or regionally, or : : : ) U.S.A., data 1962{1988: Changes in particular in 1981 and 1986 Signaling: The more dividends are discriminated against by the tax system, the more \bang-for-the bucks" Management preferences: The more dividends are discriminated against by the tax system, the less \bang-for-the-bucks" 3 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Signaling (For simplicity:) Two types of rms Want as high market values as possible Good rms cannot verify that they are good Without information problem: Tax implies no dividend Utility of management, U (v; y), increasing in v (market value of rm), decreasing in y (dividends) Good rm can better aord to pay dividends, can use it as signal that it is good, y = y; v = v Separating equilibrium: Bad rm nds it too expensive to imitate good one, chooses instead y = 0; v = v . 4 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Signaling, contd. Higher tax rate on dividends: Even more expensive for bad rm to imitate the good Indierence curves twisted counter-clockwise, around (0; v) Good rm doesn't need to pay as much dividends, is not imitated anyhow New equilibrium has y = y ; v = v for good rm Increased slope, i.e., greater \bang-for-the-bucks," compared to low-tax situation Summing up: More expensive to pay dividends, good rm does not need to pay so much to avoid imitation Later (p. 540, rst column): Same mechanism if other variables are changed so that dividend payments becomes more expensive 0 0 5 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Alternative hypothesis: Model without signaling (p. 536{537) (\First case" from second column, p. 536 to second column p. 537, somewhat simplied) Market value before dividend distribution is V (Y; y; ) = v(Y , y) + (1 , )y where Y is total resources of rm, y is dividends, is dividend tax rate, and v is an increasing and concave function (slightly convex will also be OK). Utility of management is U (Y; y; w; ) = v(Y , y) + (1 , )y + w(Y , y) where is unobserved by the market, w is increasing and concave. is stochastic (but always positive), with a new outcome each period. Management chooses y to maximize own utility: dU = 0 implies , v + (1 , ) , w = 0 dy Total dierentiation w.r.t. y and : dy = 1 < 0 ,d = (,v , w )dy implies d v + w 0 00 0 00 00 6 00 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Model without signaling, contd. A large implies a choice of low y . Change in the market value (\bang-for-the-buck"): dV = ,v + 1 , = w > 0 dy 0 0 given that management in each period obtains its f.o.c. Exception if = 0. Interpretation: For that y which is chosen by the management, we have dV=dy > 0, i.e., the market value could have been increased by choosing a larger y . The eect of a tax: 0 1 d dVdy d(w ) dy = = w @, A < 0 0 d 00 d opposite the signaling model. 7 d SK531 Economics of the rm, Diderik Lund, 1 September 2000 Empirical: Dependent variable (left hand side side variable) How does a dividend announcement aect the share price change between just before and just after the announcement? Choose to consider change over three-day period Compare with change in market portfolio in same period Left-hand side variable is relative change (= return) in share price over period, minus return for mkt. pf. Could be dierent mechanisms when dividends are increased as compared to decreased, non-symmetric Choose to consider increases only Complete dataset AMEX and NASDAQ 1962{1988 12961 observationes of corporation, traded at one of these, increasing dividend compared with previous 8 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Empirical: Independent variables (explanatory variables, right hand side variables) First suggestion: Share return = 0 + (1 + 2 ) dividend increase + : : : Dividend increase must also be transformed into relative, 0 D it , Dit DIV = p0it it where Dit0 is previous dividend, p0it is share price (average over ten days), i is index for corporation no. i The parenthesis gives \bang-for-the-buck" By letting be included in the parenthesis, we allow for the eect of a dividend change to be dierent when changes Instead of one uses (with opposite0 sign): 1 s 1 , mjt CA X THETAt = wjt B@ 1 , zjt j =1 Previously calculated by Poterba (1987) Weighted average for shareholders with dierent tax rates mjt is personal marginal tax on dividends zjt is personal marginal tax on capital gains, present value wjt is fraction owned by shareholders in tax bracket j In principle weights could be dened for each corporation, but only an average THETA for each year was available 9 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Empirical: Independent variables, contd. Other observables making dividends more expensive? Same eect as tax both in signaling model and alternative model. Two variables are included HIRATEDit shows whether corporation i has high or low \rating" when borrowing in the market Hypothesis: More expensive to pay dividends if expensive to borrow Dummy variable: = 1 if high rating (BB+ or better), = 0 if lower Varies (almost entirely) between corporations, while THETAt only varies between years CAPt measures capacity utilization in industry, seasonally adjusted, for the month of the dividend payment Hypothesis: More expensive to pay dividends when this is high 10 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Empirical: Independent variables, contd. Wish to include other observable, possible explanatory variables \: : : control for other economic conditions : : : " (p. 541) If not: These will end up in error term Problem if they are correlated with those explanatory variables included GROit is relative change in share price since previous dividend payment INFLit is relative change in consumer price index since previous dividend payment Both may explain increased dividends, and thus also eect on share price of a given increase in dividends MONTHSit is number of months since corporation last changed the dividend May explain eect on share price of a given increase in dividends 11 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Summing up, table 1 Coecients of DIV THETA, DIV HIRATED and DIV CAP have the sign predicted by the signaling model. The rst and last of these are signicantly dierent from zero, while the coecient of DIV HIRATED is not, i.e., we cannot reject that it is zero. Since data for HIRATED before 1977 were not so good, there are two new regressions using only after 1977 data, and in these (table 2), HIRATED comes out signicant. 12 SK531 Economics of the rm, Diderik Lund, 1 September 2000 Further modications Doubt whether THETA really measures tax eect accurately Numbers for each corporation would be better than a weigted average Also: Those with very low or high tax rates may not be trading in a share, i.e., the market price may be determined by shareholders with some particular tax rate(s) (\clientele eects") Alternative regressions: Divide 1962{1988 into three periods, three \tax regimes", 1962{1981, 1982{1986, 1987{1988 Two main changes in tax rates, 1982 and 1987 Signaling model still veried 13