So I did some queries on Google Scholar, and the term of art academics seem to use is target date fund. I notice divided opinions among academics on the matter. W. Pfau gave a nice set of citations of papers with which he disagrees, so I'll start with them.
In 1969, Paul Sameulson published the paper Lifetime Portfolio Selection By Dynamic Stochaistic Programming, which found that there's no mathematical foundation for an age based risk tolerance. There seems to be a fundamental quibble relating to present value of future wages; if they are stable and uncorrelated with the market, one analysis suggests the optimal lifecycle investment should start at roughly 300 percent of your portfolio in stocks (via crazy borrowing). Other people point out that if your wages are correlated with stock returns, allocations to stock as low as 20 percent might be optimal.
So theory isn't helping much. Perhaps with the advent of computers we can find some kind of empirical data. Robert Shiller authored a study on lifecycle funds when they were proposed for personal Social Security accounts. Lifecycle strategies fare poorly in his historical simulation:
Moreover, with these life cycle portfolios, relatively little is contributed when the allocation to stocks is high, since earnings are relatively low in the younger years. Workers contribute only a little to stocks, and do not enjoy a strong effect of
compounding, since the proceeds of the early investments are taken out of the stock
market as time goes on.
Basu and Drew follow up on that assertion with a set of lifecycle strategies and their contrarian counterparts: whereas a the lifecycle plan starts high stock exposure and trails off near retirement, the contrarian ones will invest in bonds and cash early in life and move to stocks after a few years. They show that contrarian strategies have higher average returns, even at the low 25th percentile of returns. It's only at the bottom 5 or 10 percent where this is reversed.
One problem with these empirical studies is isolating the effect of the glide path from rebalancing. It could be that a simple fixed allocation works plenty fine, and that selling winners and doubling down on losers is the fundamental driver of returns. Schleef and Eisinger compare lifecycle strategy with a number of fixed asset allocation schemes in Monte Carlo simulations and conclude that a 70% equity, 30% long term corp bonds does as well as all of the lifecycle funds.
Finally, the earlier W Pfau paper offers a Monte Carlo simulation similar to Schleef and Eisinger, and runs final portfolio values through a utility function designed to calculate diminishing returns to more money. This seems like a good point, as the risk of your portfolio isn't all or nothing, but your first dollar is more valuable than your millionth. Pfau finds that for some risk-aversion coefficients, lifecycles offer greater utility than portfolios with fixed allocations.
And Pfau does note that applying their strategies to the historical record makes a strong recommendation for 100 percent stocks in all but 5 years from 1940-2011. So maybe the best retirement allocation is good old low cost S&P index funds!