Prime Harvesting in Retirement
There are many theories about asset allocation in retirement. Some say that your bond percentage should be your age. Others say it should be your age minus 20. Some even say your stock percentage should rise as you age in retirement (a so-called “rising glide-path”). In his book Living Off Your Money, Michael H. McClung recommends a strategy called “Prime Harvesting” that I examine here.
The book is very technical and covers many retirement topics, but I’m going to try to be as non-technical as possible in this article and discuss only Prime Harvesting. I’ve been thinking about the best way to handle a portfolio in retirement for some time, and this book promises new ideas and a strong evidence-based approach.
First of all, “Prime Harvesting” is a great name. If you ever get into a discussion about this subject, you’ll sound like the smartest person in the room if you say, “Well, I use Prime Harvesting.” Fortunately, Prime Harvesting can be described with a few simple rules:
Booming Stocks: If stocks boom, you end up selling off excess stocks to create cash to live on, and you keep buying bonds with what’s left over.
Lagging Stocks: If stocks give below-average returns for long enough, you end up living off your bonds until they’re gone, and then you sell stocks to live.
Having stocks boom is the happy case, but let’s think about what life is like when stocks lag. Your portfolio hasn’t been performing as well as you’d like, and all your bonds are gone. Now your work experience is getting very stale and you’ve got a risky 100% stocks portfolio that is smaller than you were hoping. How well are you sleeping?
Let’s make this a little more concrete. Suppose Liz retires today with $750,000 and plans to spend $30,000 per year (rising with inflation). She begins with $300,000 in bonds and $450,000 in stocks (a 60/40 split), and uses Prime Harvesting.
Suppose that bonds lag inflation by a modest 0.5% per year for the next 9 years because interest rates rise a little. (This isn’t a prediction, but it doesn’t seem unlikely.) Suppose as well that stocks mildly disappoint by averaging 2% above inflation for the next 9 years. (Again, not a prediction, but doesn’t seem terribly unlikely.)
In this scenario, at the start of Liz’s tenth year of retirement she will sell the last of her bonds and draw partially on her stocks. She’s left with about $530,000 (adjusted for inflation) in stocks and is very nervous. Can she still afford to spend $30,000 per year?
Then the worst happens. Stocks crash 30% in year 10, are down another 10% in year 11, and are flat in year 12. She’s now got less than $250,000 (inflation adjusted) left. Her withdrawals are completely unsustainable. She needs to cut them in half. Stocks subsequently rise steadily for years but the damage has been done to Liz’s portfolio.
To be fair, I should point out that McClung also examines strategies for variable withdrawals to adapt to disappointing portfolio returns. However, these strategies would not significantly reduce Liz’s spending until after she is 100% in stocks and gets slammed by the 30% stock market crash.
How could McClung advocate a strategy so likely to leave retirees with 100% stock portfolios? The answer is that his extensive back-testing never encountered a scenario exactly like the one I described. By chance, even milder versions of this scenario haven’t occurred in the past.
McClung says that historical market data represents “known risk” and that returns outside the historical data is “speculative risk.” He would classify the scenario I described as a speculative risk. Despite the fact that McClung demonstrates a strong understanding of the risks of data mining, I believe his recommendations suffer from data mining.
To understand data mining, think of the example of trying to raise a teenager. Good strategies involve understanding what they’re going through. But if you look to the past and develop strategies taking into account drive-in movies and sock hops, then you’re guilty of data mining. You’ve over-fitted your strategy to the past and it won’t work today.
When it comes to testing financial ideas, it’s difficult to tell when you’re guilty of data mining. I can’t prove that McClung is guilty, and he can’t prove he isn’t. However, I don’t think my example of Liz’s retirement is all that unlikely. All it takes is a period of at most modest stock growth following by a crash.
If the leaves on a tree represent each historical return pattern we’ve experienced, then speculative risk comes from the possibility that your retirement will have a return pattern that is outside the tree’s boundaries. However, I believe that return patterns that cause problems for Prime Harvesting exist within the tree’s boundaries, but at places where is currently no leaf. In other words, there are return patterns that cause problems for Prime Harvesting, but are well within the character of past returns.
All this said, I’m a fan of McClung’s rigorous approach to looking for real evidence to back up retirement advice. The challenge is to define what is a reasonable range of likely future stock and bond return patterns. In my opinion, McClung is clinging too closely to historical returns. We simply have far too short a history of returns to say that we’ve seen all there is to see. Even seemingly inconsequential differences from historical returns can give big problems for Prime Harvesting.
McClung attempts to deal with the data mining problem by testing ideas on data from different countries and performing simulations where returns in 5-year groups get randomized. These efforts certainly help, but they aren’t enough. In the end, his recommendations are heavily influenced by the worst retirement period that started in 1969.
I certainly don’t know what portfolio allocation strategy is best, but I think it shouldn’t involve huge increases in portfolio risk over time. Modest adjustments to risk level could be sensible, but starting with a 60/40 allocation to stocks and bonds, and ending up 100% in stocks after a decade makes no sense to me. The fact that it has worked out reasonably well in the past brings me little comfort.
The book is very technical and covers many retirement topics, but I’m going to try to be as non-technical as possible in this article and discuss only Prime Harvesting. I’ve been thinking about the best way to handle a portfolio in retirement for some time, and this book promises new ideas and a strong evidence-based approach.
First of all, “Prime Harvesting” is a great name. If you ever get into a discussion about this subject, you’ll sound like the smartest person in the room if you say, “Well, I use Prime Harvesting.” Fortunately, Prime Harvesting can be described with a few simple rules:
1. At the start of each year, if your stocks are worth more than 20% more than they were when you started retirement (adjusted for inflation), sell off 20% of the initial value of the stocks.Even though these rules are simple enough, their implications aren’t immediately obvious. Let’s look at two extreme examples to get a feel for Prime Harvesting.
2. If there isn’t enough cash to make up your intended withdrawal for the year’s spending, sell bonds to make up the difference.
3. If you run out of bonds, sell stocks to make up the rest of your spending needs for the year.
4. If step 1 produced too much cash, buy bonds with the excess.
Booming Stocks: If stocks boom, you end up selling off excess stocks to create cash to live on, and you keep buying bonds with what’s left over.
Lagging Stocks: If stocks give below-average returns for long enough, you end up living off your bonds until they’re gone, and then you sell stocks to live.
Having stocks boom is the happy case, but let’s think about what life is like when stocks lag. Your portfolio hasn’t been performing as well as you’d like, and all your bonds are gone. Now your work experience is getting very stale and you’ve got a risky 100% stocks portfolio that is smaller than you were hoping. How well are you sleeping?
Let’s make this a little more concrete. Suppose Liz retires today with $750,000 and plans to spend $30,000 per year (rising with inflation). She begins with $300,000 in bonds and $450,000 in stocks (a 60/40 split), and uses Prime Harvesting.
Suppose that bonds lag inflation by a modest 0.5% per year for the next 9 years because interest rates rise a little. (This isn’t a prediction, but it doesn’t seem unlikely.) Suppose as well that stocks mildly disappoint by averaging 2% above inflation for the next 9 years. (Again, not a prediction, but doesn’t seem terribly unlikely.)
In this scenario, at the start of Liz’s tenth year of retirement she will sell the last of her bonds and draw partially on her stocks. She’s left with about $530,000 (adjusted for inflation) in stocks and is very nervous. Can she still afford to spend $30,000 per year?
Then the worst happens. Stocks crash 30% in year 10, are down another 10% in year 11, and are flat in year 12. She’s now got less than $250,000 (inflation adjusted) left. Her withdrawals are completely unsustainable. She needs to cut them in half. Stocks subsequently rise steadily for years but the damage has been done to Liz’s portfolio.
To be fair, I should point out that McClung also examines strategies for variable withdrawals to adapt to disappointing portfolio returns. However, these strategies would not significantly reduce Liz’s spending until after she is 100% in stocks and gets slammed by the 30% stock market crash.
How could McClung advocate a strategy so likely to leave retirees with 100% stock portfolios? The answer is that his extensive back-testing never encountered a scenario exactly like the one I described. By chance, even milder versions of this scenario haven’t occurred in the past.
McClung says that historical market data represents “known risk” and that returns outside the historical data is “speculative risk.” He would classify the scenario I described as a speculative risk. Despite the fact that McClung demonstrates a strong understanding of the risks of data mining, I believe his recommendations suffer from data mining.
To understand data mining, think of the example of trying to raise a teenager. Good strategies involve understanding what they’re going through. But if you look to the past and develop strategies taking into account drive-in movies and sock hops, then you’re guilty of data mining. You’ve over-fitted your strategy to the past and it won’t work today.
When it comes to testing financial ideas, it’s difficult to tell when you’re guilty of data mining. I can’t prove that McClung is guilty, and he can’t prove he isn’t. However, I don’t think my example of Liz’s retirement is all that unlikely. All it takes is a period of at most modest stock growth following by a crash.
If the leaves on a tree represent each historical return pattern we’ve experienced, then speculative risk comes from the possibility that your retirement will have a return pattern that is outside the tree’s boundaries. However, I believe that return patterns that cause problems for Prime Harvesting exist within the tree’s boundaries, but at places where is currently no leaf. In other words, there are return patterns that cause problems for Prime Harvesting, but are well within the character of past returns.
All this said, I’m a fan of McClung’s rigorous approach to looking for real evidence to back up retirement advice. The challenge is to define what is a reasonable range of likely future stock and bond return patterns. In my opinion, McClung is clinging too closely to historical returns. We simply have far too short a history of returns to say that we’ve seen all there is to see. Even seemingly inconsequential differences from historical returns can give big problems for Prime Harvesting.
McClung attempts to deal with the data mining problem by testing ideas on data from different countries and performing simulations where returns in 5-year groups get randomized. These efforts certainly help, but they aren’t enough. In the end, his recommendations are heavily influenced by the worst retirement period that started in 1969.
I certainly don’t know what portfolio allocation strategy is best, but I think it shouldn’t involve huge increases in portfolio risk over time. Modest adjustments to risk level could be sensible, but starting with a 60/40 allocation to stocks and bonds, and ending up 100% in stocks after a decade makes no sense to me. The fact that it has worked out reasonably well in the past brings me little comfort.
I've been looking forward to hearing more of your thoughts on this :). Good point that Prime Harvesting is an impressive sounding but not very meaningful name (what is "prime" about it, I guess because you harvest income form stocks when they are in their prime?).
ReplyDeleteYou've got me thinking, but I don't think I'm convinced about this weakness of Prime Harvesting relative to other strategies. Your counter example with Liz might seem reasonable, but I suspect it is a lot more contrived than you might think. Wouldn't 5000 simulated markets created via resampling with replacement fill out the tree enough to convince us that the Liz scenario is more a speculative risk? And Prime Harvesting has higher minimum bond averages than all other strategies, sometimes much higher. Could that mean all other strategies are also unduly tuned to 1969? I think data mining bias is pretty carefully considered and avoided in the analysis.
I suspect that strategies that handle the Liz scenario would be way too conservative, providing way less retirement income than is reasonably possible.
I also noticed that the bucket strategies, which are similar to the strategy you propose, do fairly poorly if you look at lowest bond averages and bond averages. It would be interesting to see how your strategy would do in the Liz scenario and in simulated markets.
Good post, as always interesting and gets me thinking.
@Greg: The Liz scenario is contrived to show Prime Harvesting at its worst to illustrate the problem. In general, the problem is any return pattern that leads to the retiree's portfolio 100% in stocks. This happens an alarmingly high fraction of the time with PH.
DeleteSimulating 5000 markets with resampling sounds impressive, but McClung just reorders returns within 5-year groups. This means that graphs of stock prices retain their general shapes when viewed over decades. Retirement outcomes are generally not terribly sensitive to short-term results, but they are sensitive to bad outcomes over many years. But the resampling method guarantees that all simulations have the same total return every 5 years. Combining this with the importance of just a few historical retirement years, there is no reason to believe that the simulations have examined cases similar to the one I created for Liz. The boundaries of the tree have spaces densely filled with leaves as well as empty spaces.
PH's high minimum bond averages comes from averaging in the historical retirement starting years where stocks boomed and the bond percentage stayed high. It could be that some of the other strategies were unduly tuned to historical data. Looking over the notes I took while reading, I jotted down many cases where I suspected data mining. I agree that McClung considered data mining carefully, but for the reasons I described, I think the measures he took were ineffective at preventing data mining.
Retirement income strategies involve balancing risk and income. We need to define what "reasonably possible" means. I don't think it can be defined solely in terms of averages. There has to be some cap on risk in every scenario. For example, it could be that PH performs even better if we allow it to use up to 25% leverage in some circumstances (i.e., effectively a negative bond percentage). But this seems too risky, even if it would have worked well on historical data. I'd be interested in redoing the entire "competition" with PH constrained with some minimum bond percentage. Perhaps it would still beat all the other strategies (with similar restrictions).
There are a great many bucket strategies, most of which I don't like. But judging them on minimum bond averages isn't the right measure because the average gets skewed by irrelevant cases where stocks boomed. Constraining the strategies to a minimum bond (or other fixed income) percentage is a better way to compare them. It could be that a constrained version of PH would beat bucket strategies including my own, but I'd have to do some work to test this.
In the end, I don't believe that most retirees who choose to begin with 40% or 50% bonds would knowingly allow their portfolios to go to 100% stocks. So, we might as well simulate a strategy that retirees might actually follow.
I'd not remembered well how McClung did resampling, you are right, he limited the reordering in order to preserve the character of the return data, which would rule out the Liz scenario unless it is already there in the historical data. So I guess it comes down to whether you believe it is alarmingly likely that there will be a return pattern which is bad for PH.
DeleteThere are a several answers in the book for this concern:
1. Prime Harvesting with a 30% bond floor also works fairly well, though McClung's recommendation is the extra safety isn't worth the sacrifice in retirement income (a withdrawal rate 0.4% less on average, in other words 6-10% less income).
2. Scaling back your withdrawal rate and spending is a more effective way to deal with this risk than enforcing a minimum bond level.
3. Conservative investors should also first consider an annuity to cover their bare minimum expenses (and in Canada we could count CPP, OAS, and DB pension plans toward covering bare minimum expenses).
Has the book got you interested and motivated to test your planned strategy using similar methodologies? I'd be very interested to see how it does.
@Greg: 1. I'd like to see how PH with a bond floor fares against other strategies.
Delete2. This may be true for historical returns, but I wouldn't want to get crushed 10 years into retirement while holding 100% stocks.
3. Annuities are definitely something to think about, but I don't think it's being conservative to not want to be 100% in stocks after getting too old to work and make up for losses.
The reason I haven't done any such testing so far is that I don't know how to characterize stock market returns. Historical returns are too small a sample. Lognormally-distributed returns are just wrong. In fact, any return pattern without correlations in time is wrong. The best I have so far is using separate statistics for increases in company profitability and changes in P/E, but it's far from obvious what distributions to use.
@ Michael
ReplyDeleteIf you like to geek out with spreadsheets, you have to check out the site below. US centric but very cool info...
https://portfoliocharts.com/
Good post MJ. Strange and unexpected occurrences do happen. Witness the preferred shares meltdown earlier this year when the BoC dropped interest rates. I had bought $60K in BMO LADDERED PREF SHARE IDX ETF(TSE:ZPR) in May 2015 that dropped 27%! Yikes. It's recovered somewhat now so I bought another $25K 2 days ago. Still, no one expected or understood the consequences of the rate decrease to preferred shares.
ReplyDelete@Marko: Thanks. When it comes to retail investors, I agree that most probably didn't understand that a drop in interest rates would hurt their preferred share investments, particularly because the conversion rules on preferred shares used to be different and dropping interest rates actually increased the value of the preferred shares in many cases. Conversions now seem to be from a period of fixed payments to floating payments. However, you can count on the fact that professional investors understand the effect of interest rate changes on preferred shares very well.
Delete@Michael
ReplyDeleteThere is a really interesting (and long) discussion about this book going on over at the "bogleheads" forum. Here is a link..
https://www.bogleheads.org/forum/viewtopic.php?f=10&t=192105
@Garth: Thanks for the pointer. I'm pretty good at wading through large amounts of material, but this thread is a little much for me. The wheat/chaff ratio is low. If you found a particular part that is worth considering, please point me to it. I didn't see much in the parts I read.
DeleteI've read the book and I don't believe this is correct;
ReplyDelete1. At the start of each year, if your stocks are worth more than 20% more than they were when you started retirement (adjusted for inflation), sell off the excess above the 20% gain.
I believe you sell the 20% gain and buy bonds. Any gain above the 20% you stay in stocks
That's a minor point though, and I agree, my thought in reading that is that you could blow through your bonds at some point and be left with only stocks as you get older. I'm put off about that possibility like you are.
@Anonymous: The book's author chose to describe the strategy in a simple way that might lead to extra pointless transactions like selling stocks, buying bonds, and then selling bonds to make up the year's withdrawal. I chose to describe the strategy the way you'd actually implement it to avoid possible extra transactions. If you think through the possibilities, what I described is the same as what's in the book, except that I avoid the extra transactions.
DeleteI agree that a 100% stock allocation is concerning.
@Michael James I'm pretty sure what you describe is different from prime harvesting. With prime harvesting when you sell stocks you sell 20% and keep the excess above 20% invested in stocks. So if stocks were up 124% you'd sell 20% and have 104% of your initial stock value (adjusted for inflation) remaining invested in stocks. You seem to be saying you'd sell 4% of the stocks in this case.
Delete@Anonymous and @Greg: I see now that you were right. @Greg also pointed this out to me. I will update the post. This didn't affect the Liz scenario because stocks never hit the 20% gain mark.
Delete"How could McClung advocate a strategy so likely to leave retirees with 100% stock portfolios? The answer is that his extensive back-testing never encountered a scenario exactly like the one I described."
ReplyDeleteThat isn't what McClung said. McClung's answer is: caring about being 100% stocks is the wrong thing to care about.
You can agree or disagree with that argument -- but that's the argument he's making. He's not claiming "it never happened in my backtests". He doesn't claim that because it would be nonsensical: it DID happen in his backtests, in chapter 11 when he talks about the "Failure-Market Benchmark".
@Anonymous: No scenario like the one I described occurred in McClung's back-tests. If you think getting to 100% invested in stocks is the only important feature of the scenario I described, then you didn't understand my argument.
DeleteMr. James, I'm wondering if there are *any* withdrawal strategies that *work* under the Liz scenario you provided. Because any other (not resulting in 100% stocks), would presumably mean selling stocks at a loss at some point before the 100%-stocks situation has occurred.
ReplyDeleteFrankly, as best I can tell, any scenario looks grim under the parameters you laid out, no?
Hi Edward, There are degrees of grimness. Any scenario that has a prime harvester's stocks crash after getting to 100% in stocks is going to be devastating. In the same scenario, other withdrawal strategies would be much better, if still somewhat grim. I'd rather have my withdrawals drop from $30k to $20k than to drop all the way to $10k.
Delete