Solving the Dividend Puzzle

For most individuals, dividend taxation is higher than capital gains taxation.  In theory, one would believe that corporations would thus elect to hold all their profits as retained earnings (or repurchase outstanding shares) in order to increase the price of the stock as opposed to paying out the profits as dividends, since the tax treatment of capital gains is more favorable for investors.  If there were no problems of asymmetric information, the stock price would rise more than the the value of the dividend because of the tax rules.

In the data, however, most firms do pay dividends, despite the tax disadvantage.  This is the ‘Dividend Puzzle.’

In their 2006 NBER paper (“Dividends and Taxes“), Roger Gordon and Martin Dietz offer three traditional theories to explain this phenomenon.  I will briefly review each model.

  • “New View”: The new view assumes that share repurchases are not allowed.  This is not a very reasonable assumption, but share repurchases are outlawed in the UK, and in the US periodic share repurchases are treated as dividends for tax purposes.  The theory says that if an individual has potential projects whose rate of return is high, then Tobin’s q will be greater than 1.  The firm should then sell new shares to finance these high-return projects until q=1. If future projects have a lower return, then the firm should pay out profits as dividends.
  • Agency Model: Shareholders face a tradeoff.  If they choose to keep profits within the firm in the form of retained earnings, managers may be tempted to use the money on projects with a sub-optimal rate of return.  It is assumed that managers (like bureaucrats) are empire builders and wish to increase the portion of the company over which they have control.  Investors are assumed to have less than full knowledge of the profitability of a project and thus must to some degree rely on the manager’s judgement.  In order to limit the capital available to the managers so they will only be able to finance high return projects, investors elect to have a constant stream of dividend payouts.  The drawback to this is that managers who find high-return projects are capital constrained and must go to the capital markets for more funds, which are generally more costly than procuring funds from within.  The tradeoff where dividends limit the capital available to empire-building managers but also increase the cost of financing future investment is the essence of this model.
  • Signaling: In the signaling model, the manager cares about the share price of the firm as well as its true value.  We are no longer in an empire-building set-up.  Investors are not certain whether or not a firm is a high or low value firm, but managers do know the true value of the company.  Dividends are costly, so highly profitable firms will be able to finance paying a constant stream of dividends.  The dividend stream is a signal to the market that this firm is profitable one.

This theories aim to explain some stylized facts that we see in the data.  Let’s see how they do:

  • Dividends are often positive, and stable over time.  The new view would not predict this.  Dividends should vary according to changes in profitability.  The Agency and Signalling models would both display this phenomenon as described above.
  • New Share offerings are seen concurrently with dividend payouts: The New View and the Signaling models make it seem non-sensical to raise capital by increasing the number of shares, but decrease capital through dividend payouts.  The agency theory would support this since a steady stream of dividends over time would limit capital, but occasional needs for the financing of lucrative projects would lead occasional share offerings.
  • Dividend Tax rate changes: The new view would hold that there is no change in efficiency.  The firm pays out its profits through dividends if possible, so an increase in the dividend tax lowers firm value, but does not change efficiency.  In the agency cost model, an increase in the dividend tax lowers efficiency.  Lowering the dividend tax will lead to decreased dividend payments and more retained earnings; but this leaves more cash in the hands of empire-building managers.  The signaling model would predict an efficiency improvement.  Since dividends payouts are more expensive, a lower level signal would be needed to communicate to investors that the firm is high quality.

Gordon and Dietz conclude that the Agency model best fits the data, but more investigation is required.  The do not support this conclusion with econometric proof, but simply that the Agency model best fits the stylized facts of the market.  Their paper is very theoretical and they make some assumptions of which I am uncomfortable.  They assume that firms can acquire other firms as a more efficient way to increase the share price than paying dividends which are tax disadvantaged.  But merging with another firm involves many transaction costs (lawyers, integration costs, etc.), and is likely less efficient for investors unless a smooth match is made.  I am curious on how dividend payments of firms in the same sector affect the decision of an single firm to pay dividends. More interesting papers regarding taxation are available at Roger Gordon’s homepage.