A Critical Assessment of a Critical Assessment of Lifecycle Investment Advice

A few weeks ago I listened to a great episode of the Rational Reminder podcast about Professor Cederburg’s latest research. A research that seems to disrupt traditional thinking in the field of Lifecycle Investment Advice. Two main things are considered common knowledge in the field that the paper disrupts: one is to diversify across stocks and bonds and the other is that the young should invest more heavily in stocks than the old.

Here is the link to the paper and here is the link to the podcast episode. If you have the time (and the courage, considering the boring topic), I suggest you listen at least to the podcast first.

I read the paper and I want to comment on some passages…and its conclusions.

First, some aspects I strongly agree with:

• Focus on real returns: how impactful inflation can be on bonds is, around here, well known. Meb Faber talks about it A LOT. Cederburg provides a great additional point: bond relatively low returns make it difficult for the asset class to recover after periods of high inflation; once they lost 30%, getting back in the green is quite challenging. More so if you are also selling part of the investment to fund your expenses.

• Focus on long horizons: again, something we already discussed. Traditional retirement (and FIRE) studies look at 30-year time frames but we know the real target is to survive longer than that. To be consistent with this objective, Cederburg’s model looks at long-term (10 years) means, standard deviations and correlations of stocks and bonds.

The Paper

“We evaluate strategies based on four retiree outcomes: wealth at retirement, retirement income, conservation of savings, and bequest at death.“

The paper considers several investment strategies but the “juice” is the comparison between Target Date Funds (TDF) and a 100% stock portfolio.

It is obvious that the longer the period, the better it is to stay invested in stocks. Retirees need stock’s high returns to fund their lifestyle. The strategy behind standard TDF, to progressively reduce the stock portion to buy bonds and then stay there is already surpassed: a rising glidepath after retirement, if allowed by circumstances, is demonstrated to work. Unfortunately, this is not considered in the paper (but I understand why: if I buy a Vanguard/Blackrock TDF, that’s what they do).

While strategies are evaluated based on retirement income, the focus is not on finding a safe withdrawal rate. All the strategies considered fail in some scenarios. The model considers only one withdrawal strategy, the classic 4% rule.

“As some people’s lifetime investment periods stretch for 75 years or longer, the relatively short history of US financial markets presents an extreme small sample problem for long-horizon returns.”

In order to correctly model the long-term properties of stocks and bonds, the paper employs a block bootstrap simulation approach with 10-year blocks. Financial data on US securities goes back a hundred years but that does not provide enough independent blocks to build a relevant simulation. For this reason, and to avoid survivorship bias, Cederburg integrates the data set with other developed markets.

Here is the issue I have with this process: in a simulation built this way, a 10-year block based on French financial data is equally probable as one based on the US. While conceptually correct, I do not think that this reflects how we invest (or should invest) today: there’s too much home-country bias. Using only US data is cherry-picking but that’s also the market we are all exposed to, at least to a certain extent; any market cap-weighted index leans towards the biggest, and therefore the best performing, market by construction, so giving more prominence to that market, in a way or another, is less of a stretch than otherwise. Thinking that investors would be able to invest 100% in the best opportunity in hindsight is wrong, but avoiding acknowledging a bit of momentum is not correct either.

Why does this matter? Look at the data:

to me it is pretty clear that adding markets different from the US means adding worse-performing bonds. The issue is not that stocks and bonds might be more correlated than past research suggested but the fact that bonds themselves, based on these series, are rather bad long-term investments.

Just an eye test, more on this in the conclusion of the piece.

“TDFs shift away from equity and towards fixed income as investors age to preserve wealth, whereas the Stocks and Stocks/I strategies maintain constant allocations.“

In my opinion, the paper doesn’t reinvigorate how good 100% stock portfolios are in retirement, they simply highlight how TDR built for shorter horizons are ill-suited when we increase the time they have to work for.

This glidepath is very conservative:

• doesn’t start at 100% stocks
• begin to de-risk 22 years before retirement
• continue to de-risk for many years after retirement and land on a very bond-heavy portfolio

Obviously the paper focuses on what is used today in retirement plans, not known improvements in those strategies. But here we can talk about them 😉

“Despite the fact that Stocks/I dominates the TDF in each retirement outcome, Stocks/I often has a larger intermediate drawdown (i.e., a larger peak-to-trough decline in asset value) than the TDF. Concentrating on the largest real drawdown during the couple’s retirement period, Stocks/I averages 50% whereas the TDF strategy averages 38%.”

Despite their bootstrapping strategy that tries to diminish diversification benefits, a combination of stocks and bonds reduces drawdowns.

In the long run, stocks and bonds have a positive correlation because (duh?) they both offer real, positive returns. Under this POV, there is no point in stashing bonds because they simply offer lower returns; also, the longer the horizon, the less risky stocks become:

Less risky if you look at what you have now compared to 20 years ago. But who does that? At best, we compare ourselves to last year; investors make decisions based on what happened yesterday, this unfortunate behavioural component is as real as the existence of markets outside the US (the paper author admits that simulated couples never get spooked by drawdowns).

Even if we stick with the 100% stocks portfolio, there is a lot of stress involved. Is this stress worth the extra million (on top of the other millions) when we are 80? For our heirs sure, but for us? Plus, the 100% stock strategy does not enable anyone to transfer earlier their wealth, something their kids would enjoy, because investors consider their gains temporary, given stocks’ volatility.

The paper includes a utility function, a combination of utility from consumption during retirement and wealth at death (bequest), that tries to answer the above issue. I cannot really comment on how this utility function is built, especially because they are trying to model “general” preferences. But the result sustains my idea:

The highlighted number basically says that a couple would get the same utility either saving 10%/year and employing the TDF strategy or saving 10.5%/year while having a 100% stock portfolio. In other words, TDF allows one to consume more today while maintaining the same utility in retirement.

Why do I compare TDF with Stocks and not Stocks/I?

My understanding is that TDF is built only with Domestic stocks&bonds, there is no International exposure. Why they didn’t build a TDF/I? I wish I knew, it would have been a great strategy, especially considering the results on the paper but…we do not have it.

Wealth at Retirement

For each asset allocation strategy, the table reports the mean, standard deviation, and distribution percentiles of real wealth at retirement in millions of 2022 USD.

• Cash is shit
• International, non FX-hedged, exposure pays!
• …again, why they didn’t do a TDF/I?!?

The risk-adjusted return of TDF is not that far from the one of Stocks. But by employing TDF, the investor meaningfully reduces the risk of going bankrupt straight after retirement (the glidepath is there for this reason). Unfortunately, this TDF doesn’t meaningfully decrease the overall ruin probability (as shown later) because the strategy shifts heavily to bonds and then stays there. On a long horizon, this is why the strategy fails. It would have been interesting to see how a rising glidepath after retirement would have fared.

Working-Period Drawdown

The reported drawdowns are the largest working-period peak-to-trough declines in asset values for a given strategy, and they are expressed in decimal form.

If you are wondering how Bills can have such a drawdown, I would like to introduce you to the concept of inflation 😉 Did I already say that cash sucks?

I am not 100% sure I am reading this table correctly. Here’s my doubt: TDF spends c50% of the working period invested 90% in stocks. If a massive stock drawdown happens in that timeframe, TDF and Stocks get hammered almost the same way. If this is what is reported in the above table, the reader is missing a relevant point: TDF suffers those drawdowns more likely when the amount saved is relatively small.

If your savings go from 10k to 5k or from 1M to 500k, in both cases you lost 50%. But your ass is not burning the same way.

Sure, the drawdowns in QDIAs (the first group of strategies in the Table) might look comparable to the 100% stock portfolios but…they are? (again, I might read the table in the wrong way so, thoughts are welcome).

Income Replacement Rate

For each asset allocation strategy, the table reports the mean, standard deviation, and distribution percentiles of the real income replacement rate. The last column in the table shows the proportion of simulations in which the real income replacement rate is less than one.

If I read these results correctly, the main conclusion here is that the glidepath strategy simply does not work. I am not necessarily surprised that TDF E[1π<1] is lower than the Stocks one but if the difference between the 5th percentile of TDF and Stocks is so low, why bother with the glidepath at all?

There are scenarios where catastrophic drawdowns happen shortly after retirement but such a high mean for Stocks denotes that more often than not the portfolio manages to recover even with constant 4% withdrawals. On the other side, bad scenarios for Stocks might be masked by the use of mean replacement rates: situations where the retiree gets a stream of rather high cash flows that then go to 0 after stocks experience a nasty bear market. Here, the total cash flows generated by an investment strategy are as important as their distribution (sure, Social Security represent a floor for the retiree in any case).

Wealth at Death

Maybe the “catastrophe shortly after retirement” scenario is mixed in the results below?

What I mean is that, even if TDF and Stocks low percentile results look similar, is possible that those “0.00” are born out of different circumstances:

• for Stocks, ruin arrives straight after retirement (bad sequence of returns)
• for TDF, ruin arrives later because of the strategy overexposure to bonds (high inflation)

Conclusions

This is a very interesting paper, especially in the way it challenges the protection power offered by stock-bond correlation. That said, I am still convinced that part of their conclusions is more driven by the particular model choices they took than anything else.

Inflation represents a constant struggle. If everyone gets a raise, no one really gets it because prices simply go up. It is also a struggle between you as a consumer and you as a shareholder. Profits you get as a shareholder come from the exploitation of you as a consumer. That’s why you cannot really take a risk-off approach and expect to receive anything above inflation; someone else who’s willing to take risks will eat all your profits.

International, non-FX-hedged diversification works. The 50 / 50 split between domestic and International more or less reflects the current situation for a US-based investor. I would not go 50 / 50 if I continued to live in Switzerland thought. 50 / 50 might be a good split from an FX risk point of view but it is a lousy allocation from an asset class point of view, especially in a world where I can express the currency risk differently.

We need stocks to hedge inflation and longevity risks. I see proper diversification as a way to sacrifice part of the right tail to have a smoother ride, something that this research piece is not interested in testing. Benefits of lower portfolio drawdowns and volatility are the reduction in sequence of return risk, higher intertemporal consumption flexibility and ultimately…easier behavioural decisions. But it is legit to ask, as the paper does, if this proper diversification is actually achievable or if it is just a mirage.

I would have delved more into the ruin situations, identifying the pain points for TDF and Stocks. I suspect there are different situations in which they do not perform, therefore opening the door for alternative, better solutions.

“When domestic equity does poorly, bonds and bills also tend to do poorly“. This drives two conclusions for me:

• that’s why we should allocate on a market-cap basis
• if done the way the paper does it, there is no diversification benefit

I was surprised to find that the difference in block lengths significantly reduced the diversification benefit (see pic below). Corey Hoffstein demonstrated the diversification benefit of Trend Following strategy using 3-month blocks in a recent paper Would be interesting to see more research on this aspect.

Ultimately, I have dedicated just a few hours to this analysis, it was more driven by my curiosity on the paper than building something more structured and informed. Caveat lector 😉

What I am reading now:

Follow me on Twitter @nprotasoni