True, companies may use earnings to repair balance sheets or to fund future acquisitions. "The cost of retained earnings capital appears to be near zero because managements can use it for anything they want," Bernstein told the audience at the JP Morgan Fleming meeting. If he's right, then the return on retained earnings capital also would approximate zero.

It certainly is true that many companies have only a few really good ideas that cost surprisingly little, and lots of mediocre projects that get expensive fast. Take Microsoft, the model Information Age business. Twenty-five years ago Microsoft paid consultants $10,000 to create the DOS operating system, which gave it a huge competitive advantage over all other PC software makers. These days it will spend tens or hundreds of millions to produce a marginally better spell-check program.

Then there is the managed earnings issue. Anyone who read Barbarians At The Gate was stunned learning about KKR's Henry Kravis meeting with the head of the Nabisco unit of RJR Nabisco. The executive told Kravis his biggest challenge was coming up with ways to spend about $400 million to depress earnings, so RJR Nabisco could give shareholders nice smooth 15% annual earnings growth. That's a different kind of burn rate than the one that Internet companies made famous at the turn of the millennium, but it's squandering capital nonetheless.

Bernstein cited the 1980s swashbuckler, Boone Pickens, who urged companies to borrow and leverage up their balance sheet because the cost of capital "is real." Debt, as Michael Milken and other junk bond aficionados proclaimed in that era, forced corporate managements to discipline their operations.

And any constraint that forces businesses to accept a higher hurdle rate for capital spending projects inevitably improves the rate of return on invested capital. This is the underlying thesis of a bold prediction Bernstein made that night in Oak Brook. "Companies with higher [dividend] payout ratios will have higher earnings growth rates in the next five years," he said. "The evidence is very powerful." In fact, he noted that Cliff Asness and Robert Arnott won the Graham and Dodd award for an article in the Financial Analysts Journal proving that lower payout ratios lead to lower earnings growth rates.

Two companies Bernstein cited that have displayed impressive growth despite their size and generous payouts are GE and IBM. GE's payout ratio has averaged 40% for decades and it raised the payout in the 1990s. In the 1950s, IBM managed to pay out more than 20% of net income and grow at 20% even though its business was extremely capital intensive, because it leased its computers instead of selling them. By 1981, its dividend amounted to 81% of its 1956 share price.

Searching for an answer to the dividend paradox almost certainly finds its origins with the growth stock boom that started in the late 1950s and 1960s. Bernstein quotes his wife, Barbara, as saying that today's generation doesn't respect dividends for the same reason many people no longer prefer fresh-squeezed orange juice-because they're only familiar with the frozen kind. "Shareholders prefer that companies use earnings to lift share prices rather than pay dividends," Bernstein said.

That's all very nice, but he reminded attendees at the JP Morgan Fleming dinner that companies can't use earnings to raise share prices. In fact, no one can make the market do anything on a sustained basis. It marches to the beat of its own drummer. Bernstein recalled that between 1955 and 1961, earnings for the Standard & Poor's 500 Index fell 12% while the index rose 59%. Fast forward to the 1969 to 1974 period, and earnings rose 54% while the index slipped 26%.

What a 20-year bull market has caused this generation of investors to forget is that, in Bernstein's view, the difference in risk between dividends and capital gains is enormous. "Dividends are known, and earnings per share are not known," he remarked, with a eulogy for Arthur Andersen. "What you see [with earnings] is not necessarily what you get."

Bernstein argued that this explains why John Burr Williams created his famous valuation model, the dividend discount model, around dividends rather than earnings. Between 1871 and the late 1950s, dividend yields typically fluctuated between 4% and 6%. In 1949, the S&P 500 yielded 6%; during the next eight years, when earnings rose 80%, the yield still stood at 4.4%. Going back a long way in time, Bernstein said it's hard to find a sustained bull market that began with the S&P yielding less than 4%.