There are so many factors that it is probably hard to come up with a right answer for a group scheme if you go to extremes, 100% in equity would be fairly controversial. Ultimately I think you would want to avoid extremes for a group scheme.
If you are optimistic, have a long life expectancy (most people don't know, but we are moving closer to having Gattaca style data on everyone) and don't mind working longer if needed, 100% equities is less risky for you.
If you hate your job and there is a family history of lower life expectancy, you probably want to avoid crash risk, literally like the plague.
Worst case you have a short underfunded retirement or zero retirement.
Admittedly I'm no expert on AE, I'm in a DB scheme, so apologies in advance if I might be missing a piece of the puzzle in that the risk of delayed retirement is mitigated somehow.
Reducing risk exposure closer to retirement seems to be established practice and academic research seemed to back it up, John Campbell did a lot of relevant work:
Academic finance has had a remarkable impact on many financial services. Yet long-term investors have received curiously little guidance from academic financial economists.Mean-variance analysis, developed almost fifty years ago, has provided a basic paradigm for portfolio choice. This approach...
books.google.ie
Maybe the move away from annuities to ARF's due to low interest rates has changed the picture a bit since then, who knows if that is changing back but there are a lot of arguments for lower interest rates being a systemic change. But even with that maximising your ARF return seems like a very optimistic strategy ignoring mortality.
I think there is something to having some downside insurance for people with a very short investment horizon being funded by giving up some of the upside for people with a much longer investment horizon and Colm's scheme worked a bit like an insurance pool diversified over investment horizon.
Receiving say a 100k boost/insurance into your fund value when the market is down and you are about to retire would be valued a lot more by most people than when your fund is up 100k and you have 20 years to retirement. Some approach that allows the group to exchange the two in a group scheme (granted it might be difficult or impossible to implement) seems interesting.
As Brendan said 'I would be prepared to give up some of my gains in exchange for knowing that my fund was not going to fall by 50% all of a sudden.'
So by taking it off of future investors (who effectively pay it now but it is repaid into their fund when the market is up so really they are using their future '100k when the market is up money'), saves the insurance premium for that guarantee for the whole group. Of course that is all in average and seems to have dependency on the market being cyclical and the periodicity of the crashes and recoverys etc.
Talking about a fund set aside to do the same, there you would lose the equity gains on your reserve fund. But it all would need a lot of analysis on relative utility under various market scenarios and be subject to a large number of assumptions that could turn out to be wrong if there is some big regime shift affecting market behaviour, like the introduction of QE, impact of any UBI on saving and investment etc.
The issue I see with Colm's smoothing approach, apart from possible data mining on the various filter parameters that Coyote points out is that there is no-one profiting from implementing the scheme. Being slightly cynical, people are more likely to get on board with things when they see future fees coming their way, this is a move in the opposite direction.
Based on Coyote's observations, looking at the smoothed index for the UK in Figure 2 in Colm's paper and evaluating the amount of time that the market is above it, that seems most of the time in Figure 2 except for the two main crashes and the amount of the drop insured seems very sensitive to the ERP parameter in Figure 7 and less so to the p parameter in Figure 8.
The question then is it worth it overall depends on the value of insuring the drops versus the net lost utility when the market is above the smoothed index evaluated for the whole group and the drop mitigation and the forfeited returns look to be very sensitive to the ERP.
Aswath Damadoran publishes historical ERP figures for the US and they vary quite a bit (file attached, but again that would depend on the method used to estimate ERP, he is one of the leading researchers in that area in academia but maybe an average of a bunch of estimators would be better practice to use). Using the previous year one for the following year and updating yearly and re-running the evaluations in the paper would be a good start to shed more light on the data fitting concern. It seems less sensitive to p but again you'd need to show that for more than just 2 values, in particular show the optimal p over a bunch of markets and periods to show the variation and to get the optimal one you need some scheme utility score that you are optimising. I think it needs to address group utility and show that for a representative investor it results in overall utility gains most of the time.
I think the conclusion of the paper should be clearer and succinct if the message is to be brought across, e.g targeting something like the below:
I show that using a smoothed index target for an AE scheme, the average utility gain to a representative investor over investment in the raw index is x% , and show this gain to be statistically significant. Investors (across a range of commonly used investor utility functions and risk aversion parameters) benefit from the scheme in (90%, 95%, 100%?) of the rolling 30 year time frames examined in our study across 3 countries (use 20 countries?). This is robust to the selected 30 year investment period, using over 150 years of data. All evaluation is completely out of sample with all model parameters estimated using prior historical data point in time in a rolling window.
On the general cost question, surely some simple, infrequently rebalanced SAA approach using index ETF's would be very low cost.
I get that 100% equities is simpler but there is so much money involved in this the economy of scale on implementing that must be huge.
In terms of the fund composition maybe there needs to be some process on that such as a risk questionnaire sent out and a democratic average setting used, people tend to be more risk averse than is good for them so the framing of the questions would be important.
A range of options: you are willing to lose x amount of the fund on retirement for a y% return overall.
Then you have some constraints for the SAA , you have to chose the assets allowed and the optimiser and the rebalancing frequency, it could all be mostly automated without any committee decisions.
).