Konstantin Magin of U.C. Berkeley says
"...[R]oughly half of Americans have little or no stock market investments either directly or indirectly through pensions. This is too bad, because in the long run, U.S. stocks have remarkably high returns (about 6.6 percent per year even after adjusting for inflation). And, although liberals fear that these returns come at too high a risk, I will show here that that just isn’t so. Private Social Security accounts invested in long-run diversified equity portfolios promise substantial increases in the lifetime wealth of middle- and working-class Americans, at low risk."While it's hard to argue with his math (and I've made similar arguments, with similar conclusions, in the past) a couple points are worth touching on.
First, since I favor personal accounts I agree with Magin's general conclusion. At the least, the opportunity to diversify their retirement savings portfolio has potential benefits to low earners. I half-agree with the underlying argument, which is that the equity premium is still sufficiently large that there's free money on the table for stock holders. If so, it makes sense for low earners to get their share. At the same time, the equity premium is set by the risk preferences of market participants; just because we can't explain it in the context of other risk preferences doesn't mean it comes about for no good reason.
Second, Magin is working with an equity premium of around 5.6% (assuming 6.6% real mean stock returns and a 1% bond return). Others see a more modest equity premium: The SSA actuaries project stock returns of 6.4% and a trust fund bond return of 2.9% (a higher bond return will tend to lower guarantee costs using Black-Scholes, so the overall effect is ambiguous). Trimming 50 basis points or so for a shorter-term interest rate gives you a risk premium of 4% -- still healthy, but the probability of falling short of the riskless return certainly rises. It's anyone's guess what future stock returns will look like, but many have argued for lower rather than higher than the past. (E.g., see Diamond, Shoven and Campbell here; Baker, Krugman and DeLong here; but also see Magin and DeLong here.)
Third, while Magin shows that the price of a put option guaranteeing against loss of principal is low, it's not clear why this should be the only standard. A guarantee against the lost of real principal is more expensive, and a guarantee against falling below the riskless rate more expensive still. In the policy context, it's most common to guarantee against the account purchasing an annuity smaller than current law scheduled benefits. This can get very expensive, as shown in this paper written with Kent Smetters and Clark Burdick.