Selected excerpts from G Irlam's "Asset Allocation Confidence Intervals in Retirement"

With apologies for the double negative, I am never not interested in Gordon Irlam's work. Someday in the hall of retirement finance he will have his statue, the more-so for not being (as far as I can tell) a formal academic or practitioner though he has got more than enough capability for both.

I found this 2015 paper (Asset Allocation Confidence Intervals in Retirement SSRN 2675390 Oct 16 2015) interesting for it's conclusions which are more thorough, but not totally divorced from, some points I've made here, but at a much more rudimentary level: asset allocation matters but only within a pretty broad range of weights.

With almost no comment, here were some of the points I pulled out:

"For the scenario I study the 95% confidence interval for a 65 year old retiree with $400k in assets allocated between stocks and bonds is 10 to 82% stocks. Larger confidence intervals apply to retirees with small or large portfolio sizes.

When I add in the uncertainty in the retirees own situation, namely their risk aversion and mortality, I find the confidence intervals only change slightly.

Unlike asset allocation, confidence intervals for the optimal level of consumption are found to be reasonably tight except for large portfolios in the presence of uncertainty regarding mortality.
If regression to the mean exists its effects are small compared with the standard deviation of the premium of stocks over bonds.

It is unreasonable to expect a single γ value applies irrespective of the consumption level, and this may go some way to explaining the difficulty in measuring γ. In particular the γ value associated with nondiscretionary consumption is likely to be higher than the γ value for spending on items such as charitable gifts. [γ = coefficient of risk aversion]

Uncertainty as to the true value of the equity premium means that there is very little we can be confident about when it comes to determining the optimal asset allocation. For the scenario I considered, a 65 year old retiree with $400k in assets a 95% confidence interval was 10 to 82% stocks, and a 68% interval was 26 to 62%.

This uncertainty is particularly disturbing in light of a separate, well known finding that asset allocation matters⁵ . In my studies, always being 20% under the optimal asset allocation produces around a 5% reduction in portfolio derived consumption at age 65, or a 25% reduction at age 25. It is important to follow the optimal asset allocation, but it is impossible to know what it is with any degree of certainty! Even more disturbingly these performance figures are based on large numbers of simulations. The individual investor is left at the mercy of the results of a single 20-60 year sample with an even greater degree of uncertainty in the realized equity premium.

[footnote] Saying asset allocation matters is stronger than saying asset allocation plays a key role in determining portfolio performance (Brinson, 1986). It could be asset allocation is responsible for 90% of portfolio performance, but the difference in portfolio performance was inconsequential. This is not the case. I say asset allocation matters because it both plays a key role in determining portfolio performance, and its effect is substantial. [comment: the topic of consequences and payoffs are often missed when eyeing statistics directly]

In addition to the market parameters, the self parameters, risk aversion and mortality, may also be uncertain. Fortunately, confidence intervals when the self parameters are allowed to vary stochastically are found to be very similar to when they are fixed. It is uncertainty in the equity premium, not risk aversion or mortality, that is the key impediment to accurately determining asset allocations."