Wednesday, December 21, 2016

Asset Allocation: Mike and the Robos

Robo-advisers are all the rage.

A robo-adviser is software – usually a website or an app on a mobile device – that provides financial advice or performs portfolio management online with minimal human intervention.

I'm not a big fan of robos because I have a finance background and I know how difficult retirement planning can be, but also because I'm a computer scientist and I know how difficult writing the algorithms can be. To develop a truly effective robo-adviser would require a great deal of financial knowledge effectively translated to a really good software developer. I don't think that happens often, at least not yet.

I'm also not a fan because my recent experience with robo-advisers has not been encouraging. Most are little more than simplistic games. The good ones, in my opinion, are better than a poor retirement planner and certainly better than not planning, at all.

The good robo-advisers are nowhere near as effective as a good, human CFP given the current state of the technology and none compares with a top financial planner. This is merely my opinion, but as I said I have a lot of training in retirement finance and I was once a damned good software developer if I do say so myself.  (I have considered writing a robo-adviser program for a long time. I haven't because I fully understand how difficult it would be to do it right.)

Several clients and readers have been stressed out by the chore of picking the “optimal” asset allocation for their retirement portfolio. I decided to try several free robo-advisers by creating a retirement planning scenario to run through each and then comparing the recommendations. I also asked retirement planner, Mike Lonier, to provide his recommendation representing the human race.

The scenario is the following. A husband age 66 and wife age 64 just retired and have saved $3.75M for retirement as shown in the table below. The husband will receive about $3,250 per month in Social Security benefits at age 70 and the wife about $2,000 per month at age 66. They have a small pension that will pay $1,000 per month beginning at the husband's age 66 and the wife will be entitled to a 50% survivors benefit. They believe they need to spend at least $180,000 a year and at most $220,000. The question I asked Lonier and the Robos (sounds like a 50's rock-and-roll band, doesn't it?) is how these funds should currently be allocated.

Desired annual spending Per Year
      "Wins the game" $220,000
      Minimum acceptable $180,000
      Taxable $1,750,000
      IRA/401(k) $1,500,000
      Roth $500,000
Social Security
      Husband at 70 $39,268
      Wife at 66 $24,703
Pension at husband's age 66 $11,000

The following table summarizes the results of the planners.

Planner/Robo-Planner Recommended Annual
Annuity Allocation Stock Allocation Bond Allocation Cash Allocation
AACalc with Annuity $260k 59% 41% N/A N/A
AACalc No Annuity $220k 0% 34% 66% N/A N/A NA 56% 44% N/A
Mike Lonier (1) $220k 8.0% 0% 86.2% 5.8%
Mike Lonier (2) $180k 8.0% 10.4% 77.4% 4.2% N/A N/A 45% 31% 24%
ESPlannerBASIC $191k N/A N/A N/A N/A
Fidelity Planning & Guidance N/A N/A 50% 40% 10% $192k N/A N/A N/A N/A
Vanguard Planning & Guidance N/A N/A 30% 70% N/A
Yahoo!Finance N/A N/A 50% 45% 5%
Elevator Ride with Annuities $150k [1] N/A 30% 70% N/A
Elevator Ride No Annuities $150k to $195k [1] N/A 40% - 60% 60% - 40% N/A

Let me explain each of the approaches., developed by Gordon Irlam is, again in my opinion, the most thorough asset allocation software available. AACalc asks several detailed, specific financial questions and explains its calculations well. (As you will see, some calculators ask generic – and frankly, sometimes plain weird – questions to generate a recommended asset allocation.)

The first two rows of the table suggest that a key differentiator among robo-planners is whether or not the client/user is willing to purchase fixed income annuities. (All of the annuities discussed in this post are simple fixed annuities without indexing. [4]) Row one shows AACalc results when fixed annuities are acceptable and row two shows recommendations when fixed annuities are not acceptable. Mosts robos don't even ask this question.

AACalc recommends in row one that a client willing to purchase annuities can generate $40,000 more lifetime income than one who won't buy them (row two). It recommends that the client invest 59% of assets in annuities and 41% in stocks. With income guaranteed by a secure floor of annuity and Social Security income, the client can be very aggressive with the remainder of the portfolio. Annuities replace bonds. This is the floor-and-upside strategy recommended by most economists.

For clients unwilling to purchase annuities, AACalc recommends a 34% stock allocation and 66% bond allocation. is perhaps the best-known robo-investment adviser. The website asked me just three questions, my age, my annual income and whether I was already retired, before advising me that I should invest 56% of my portfolio in stocks and 44% in bonds. It calculated this without knowing the size of my portfolio, whether I was married, or even how much I needed to spend annually from the portfolio. I clicked on "Details" to find a very precise breakdown of twelve sub-asset classes I should invest in – 4.8% in emerging markets, for example. Who knew retirement planning was so easy?

It isn't. And any planner, living or robo, who doesn't ask the right questions is subject to one of the oldest truisms of computers: garbage in, garbage out.

Mike Lonier also advocates the floor-and-upside strategy, unsurprisingly since he is associated with the Retirement Income Industry Association (RIIA). Lonier provided two recommendations, the first based on $220,000 annual spending and the second on $180,000 spending. For planned spending of $220,000 per year, he recommends an 8% annuity allocation (specifically, to a type of annuities called QLACs that are held in retirement accounts – see note in References below) with the remainder of the portfolio invested in bonds. He also notes a reasonably high concern that this level of spending may not be attainable.

For $180,000 annual spending, a level at which Lonier is more comfortable, he still recommends an allocation of 8% of assets to QLACs, but a 10.4% stock allocation with the remainder in bonds and cash.

Lonier elaborates on his approach:
"I don’t usually recommend “bonds” meaning bond funds. I recommend a lifestyle (floor) allocation built as a liability matched risk-free or guaranteed ladder of CDs, individual Treasury bonds held to maturity, and/or single premium income annuities (SPIAs) and deferred income annuities (DIAs). The SPIA/DIAs would be as floor ladder components, in addition to any DIAs that are part of longevity.  
With very-well funded clients, some percentage of the Lifestyle (floor) portfolio might be exposed to credit and interest rate risk in a partial allocation to bond funds, meaning that the very well-funded might follow, to some degree, an investment-based (MPT) portfolio rather than a strict goals-based approach, especially with funds above those needed to fund expenses in the ladder.  
My own algorithm, which analyzes the present value of assets and liabilities (current and future) on the balance sheet, indicates 0% and 10.4% for the Growth and Legacy (Upside) allocation for the two spending scenarios. My “human” adjustment changes that based on the balance sheet math but increased by the planner to 20% for the $220k scenario and 30% for the $180k scenario. The rationale is that there are sizable assets, so the concerning risk is spending risk, not necessarily asset loss (sequence of return risk), so the plan would manage spending risk while providing modest exposure to growth."
I should note that all of the robo's services I studied are free, though Mr. Lonier's services are not. Still, I believe he provides much better value. The above quote shows the additional value you get from human tweaking.

Next, I looked at Bankrate requests more relevant input than most robo-advisers but less than asked by Mr. Lonier or AACalc. In this scenario, Bankrate recommended 45% stocks, 31% bonds, and 24% cash. Still, if someone ran this scenario by me in an elevator, I would probably suggest a 40% to 50% equity allocation and be in the ballpark most of the time.

ESPlannerBASIC doesn't offer a recommended asset allocation that I could find, but it did suggest annual spending of $191,000 that I found useful.

Finding the calculator isn't always easy. After I had given up on Fidelity Investments offering an asset allocation calculator, I discovered the link that would provide one. Go to GUIDANCE & PLANNING > RETIREMENT PLANNING > PLANNING & GUIDANCE CENTER . . . You know what? Can you just trust me that there is an asset allocator in there somewhere if you look long enough?

Along the way, Fidelity posts a warning that their “advice” is for educational purposes only and shouldn't be used for actual planning. I feel like that warning belongs on all robo-advisers. In the end, Fidelity recommends 50% stocks, 40% bonds, and 10% cash.

My favorite robo-adviser so far is I've been working with founder Stephen Chen on this website and I find it closer to a real adviser than any of the alternatives. Unfortunately, the site doesn't provide asset allocation recommendations, at least not yet, but it did recommend annual spending of $192,000. I recommend you try out this website. It gets better all the time and I think it's already as good as the typical human planner.

Vanguard's Advice & Guidance tool recommended 30% stocks/70% bonds based on generic input. Still not much different than my elevator advice.

Lastly, Yahoo!Finance offered the most detailed asset allocation but again based on a few generic questions. With so little input data these calculators are giving the same advice to a huge range of clients. Yahoo! recommended 10% US large cap growth stocks, 15% US large cap value stocks, 10% US small/mid-cap stocks, 15% international equities, 45% bonds and 5% cash.

How do we compare such a wide range of recommendations? First, let's separate recommendations with annuity allocations from those without because these represent two very different strategies. The former is floor-and-upside while the latter is essentially a sustainable withdrawal rate (probabilistic) strategy. The two strategies provided by Lonier have low stock allocations because the required spending is high (4.8%) and because his strategy is to guarantee one's standard of living first and invest anything left over aggressively. (That's my approach, as well.) Also, note that the highest annual spending comes from these strategies and not from large stock market bets.

AACalc's no-annuity recommendation proposes a similar strategy to Lonier's, but invests in a bond ladder instead of annuities – secure the floor, then invest aggressively. Bottom line, these are quite similar strategies when annuities are allowed.

Looking at the remaining strategies, recommended annual spending ranges from $191,000 to $213,000. That's only an 11.5% difference.

The strategies without annuities recommend stock allocations between 30% and 50%. Personally, I would recommend more weight be given to calculators that require a lot of financial input instead of generic lifestyle questions. Those were AACalc and Fidelity with recommendations of 35% and 50% equities, respectively.

The goal of my readers and clients is presumably to narrow down an “optimal” asset allocation. But as I explained in The Whoosh! of Exponential Retirement, Irlam showed that the 95th percentile optimal asset allocation could range from 10% to 82% equities. The problem is that we simply can't predict future stock returns accurately enough to determine the optimal allocation so any of these allocations is mathematically defensible. The best strategy is to determine not the optimal asset allocation, but one that works well across a broad range of potential future outcomes.

My own approach is quite different and doesn't require a computer. It is based on the advice of William Bernstein and the idea that an asset allocation has two main purposes: to manage short-term volatility (my risk tolerance [2]) and to optimize long-term risk-adjusted return within my risk capacity.

Managing short-term volatility means implementing a stock allocation that won't lose so much money in a severe bear market that you might be tempted to bail out of equities. I believe I can personally tolerate about a 15% short term loss of my investments (though my risk capacity [3] is greater than that) and that has historically meant choosing about a 40% equity allocation. Bernstein suggests we begin there.

I can tolerate a
short term loss of
Equity Allocation
35% 80%
30% 70%
25% 60%
20% 50%
15% 40%
10% 30%
5% 20%
0% 10%

For investors who can ignore bear market losses, including the more than 50% loss during the Great Recession, an equity allocation can be much higher. Floor-and-upside strategies can recommend up to 100% equity positions because, with a floor, you're not betting your standard of living.

I personally find that my short-term volatility limit of a 15% loss is lower than any long-term asset allocation I calculate, so the latter becomes irrelevant for me. My advice is that investors who can't stomach large losses in a bear market go with the Bernstein max-loss allocation. For many, that will mean somewhere between 40% to 60% equities.

For those retirees who can tolerate short-term losses well, use AACalc to determine an allocation, remembering that its developer notes that an optimal asset allocation is probably unknowable. (In other words, don't sweat a 5% change in your allocation.)

Lastly, retirees with a secure floor can be as aggressive with equities as they like because even in the extremely unlikely event that they lose their entire portfolio, they won't lose their standard of living.

I'm a computer geek and have been since the early seventies. I'm also a statistics and modeling geek. I have found, though, that with many retirement planning calculations using extensive computer models don't give us dramatically new usable information. As you can see in the last two lines of the table, the recommendations I would give you for an asset allocation during a two-minute elevator ride don't vary dramatically from what robos recommend and I would be providing this advice with precious little information about your financial situation, as do most of the robos.

Importantly, the optimal asset allocation is unknowable so there's no sense laboring over a precise allocation, tweaking yours 5% one way or the other, or rebalancing frequently.

Worse, the computer calculations to four decimal places leave people who don't understand computers and models overconfident in their predictions.

If you use an asset allocation calculator, use a good one like AACalc that asks all the important financial questions instead of a pretty app with a lot of marketing pizazz and "ease-of-use" that doesn't. At this point, however, as a planner and a programmer, I prefer skilled humans.


 SWR spending rates and annuity payouts were taken from the Dashboard at Wade Pfau's Retirement Researcher blog.

[2] Risk tolerance describes an investor's emotional ability to deal with market volatility.

[3] Risk capacity describes a retiree's ability to risk investment losses and still achieve important financial goals.

[4] Fixed annuity, variable annuity and indexed annuity, defined at Investopedia.

The Intelligent Asset Allocator, William J. Bernstein, Table 8-1, page 144., robo-investment adviser.

Mike Lonier, Retirement Planner, email

AACalc, asset allocation calculator by Gordon Irlam.

RIIA, Retirement Income Industry Association.

QLAC, qualified longevity annuity contract., asset allocation calculator.

ESPlannerBasic, retirement planning website.

Fidelity Investments retirement planner.

Vanguard Investments retirement planner.


  1. Hi Dirk-Long time reader, and always look forward to your posts, but this is the first time I have commented. I had never used AACalc and thought I would try it based on this post. I'm not sure if I did something wrong but I was surprised-OK, I was stunned-when the recommendation came back to be 100% stocks with my investments. I'm 60, my wife is 57 and we plan to retire in a year or two. We will have pensions totally about 30K annually pre tax and both of plan to delay SS (myself to 70, her to 67) so that will be about 70K annually pre-tax (estimated). I suppose I may have inputted other data incorrectly, but I was surprised-OK, I was stunned...when the recommendation came back to invest 100% of taxable, deferred and Roth retirement plans in stocks. Basically the totals were about 1.6M lifetime in pensions/SS and 1.4M in investments. Does this make sense? Is this telling me the floor is covered but I am not taking enough upside risk? Our investments now are about 35% stock, 65% bonds to manage sequence risk, with the plan to slowly increase risk through our 60s while spending some of the bonds before SS.

    1. Rob, thanks for writing. It could make sense, but I would need to see the complete AACalc report to give a complete answer. Email it to me at if you'd like.

      If your floor is adequately covered then losing your entire portfolio (highly unlikely) wouldn't change your standard of living. So from a financial perspective, a 100% stock portfolio is defensible with a secured floor.

      Personally, though, I wouldn't do it. I have enough money to fund the retirement I want without losing sleep over market losses. (I have the risk capacity to do so, but not the risk tolerance.) That's where the Bernstein recommendation comes in. It sounds like you're in a similar boat.

      From my perspective, this is another argument for a skilled planner.

    2. Thanks for sending the report.

      As I suspected, you asked for annual spending levels that your pensions and Social Security benefits alone guarantee. Therefore, you have the risk capacity to invest 100% in equities.

      On the other hand, you don’t sound like you have the risk tolerance to invest 100% in equities.

      You used the default risk aversion factor of 3, which is "typical." Maybe your risk aversion is higher than typical?

      You might rerun with your coefficient of relative risk aversion (γ) set to a higher degree of risk aversion, like 5 or 6, to reflect your lower risk tolerance and see if you get an answer you’re happier with.

  2. I think it's best to use robos for accumulators, not retirees or near retirees. I also don't think people think of robos as really financial planners, but rather asset managers. And hat they do very well. Lastly I think the real reason for high portfolio value people to favor them over a human is the ability to direct index into 1000 individual stock positions and daily tax loss harvest. Currently only offered by wealthfront but what I believe is the future.

    1. Agree with your first sentence, but mostly because accumulation is a simpler problem.

      I don't see the independent evidence that they do asset management that well, however. If you have some independent academic research showing that, please send me a link.

      I would need to also see evidence that indexing 1,000 stocks and tax-loss harvesting outperforms using existing index funds. Diversification benefits decline with the number of equities purchased. There are diminishing returns. I think the jury is still out.

      Lastly, regardless of how people think of them, there are robos that present themselves as planners and robos that present themselves as asset managers. And robos that present themselves as other things, as well.

      Thanks for writing!

  3. Does the term "robo advisor" really apply to what is essentially just a free asset allocation calculator you found on the internet?

    I thought the term was only applied to investing platforms where a human or computer runs you through a questionnaire, and then proposes an asset allocation, but not just for educational purposes. It then actually invests your money, manages the allocation automatically, and charges you a fee for the service.

    1. The term is commonly used in several different ways. I'm not sure who gets to decide which definition is "correct."

      I prefer the broader definition – "Robo-advisors are a class of financial adviser that provides financial advice or portfolio management online with minimal human intervention." Others use a definition similar to yours.

      This is largely a semantic issue in my opinion and focusing on it distracts from the point of my post: how does software without human intervention compare to software with human intervention specifically regarding asset allocation and nothing more. The robos represent the former and Mike is the latter.

      Ultimately, I think you can call them whatever you like so long as everyone knows what you're talking about but, yes, I believe AACalc meets the broader definition of robo-adviser, for what that's worth.

      Thanks for writing, Steve!

  4. Dirk what I like about RoboAdvisors is the automatic Tax Loss Harvesting. I use Betterment and while I am perfectly capable of maintaining an asset allocation in Vanguard I get an extra bonus from the robo advisor of the tax loss harvesting.

    1. Andy, thanks for writing!

      This post is solely about asset allocation, but I have other concerns with robos. They've only been around for a few years, so there's no real evidence that they will outperform index funds, or even tax-managed index funds, over the long term.

      A lot of new investment ideas sound really great, but after a few market cycles most underperform simple indexing. The evidence just isn't there, yet.

      Back in May, Michael Kitces wrote that "Robo-advisors appear to be spending more to get clients than they can ever earn by serving them." His post on the topic from last May is a good one.

      I just don't see the evidence – yet – that robo-advisors outperform, but I hope they work for you!

  5. Nice post looking at a wide variety of sites Dirk.

    The elephant in the room for decumulation (retirement income planning) is that they all presently use a fixed single period over which to do the calculation they use (all are slightly different in that regard).

    Our latest research published in JFP ( ) found that age matters! Why age? Because the time frame associated with age changes due to less time remaining. It makes little sense to use a 30 year period when you're 75 for example. The fixed single time frame perspective used results in cash flow errors and thus capital balance errors over time.

    This perspective is a carry over from the slide rule days, that carried over to calculator days, that carried over to computer programming days, etc. What hasn't been incorporated into software programming is software capabilities to much more than doing a single, single period calculation.

    Multi-casting should be on the horizon for software programming so that actual practice of annual reviews align with the fact that aging makes a difference in what income is prudent and feasible at any given age. Rules of thumbs with other rules of thumbs aren't a practical approach since those don't provide enlightenment as to when or how to change cash flow (income) as one ages.

    Programming has come a long way and the profession needs to move with those capabilities. Much like medicine improved from the late 1800's to today, software programming needs to catch up to use the capabilities that exist now. So software and programming are capable ... but the thought processes for programming are still analog based.

    Merry Christmas and Happy New Year Dirk!

  6. I have concern whenever I read about the potential benefits of fixed income annuities that some readers will not be aware of the huge difference between low cost, simple, fixed income annuities ( which are only lightly marketed due to their relatively low commissions) and high cost, painfully complex variable annuities ( which are highly marketed due to their higher commissions.) So I wish that those discussing fixed income annuities explicitly distinguish them each and every time from their more expensive and less desirable cousins - the variable annuities.

    1. Good point and I'll try to remember that. Since fixed income annuities are the only kind I ever talk about (I personally never met a variable annuity that I liked) I tend not to dwell on specifically calling out fixed annuities.

      But feel free to remind us if and when you catch me doing it again.

      Thanks, Barry!

  7. Thanks, Larry, and a Merry Christmas to you and yours, as well!

    Thanks for this perspective. (I have actually intended to contact you to discuss multi-casting and hope to soon.)

    Your comments here demonstrate a serious limitation of the "software without human intervention" model. It isn't enough that the software capabilities improve. Someone still has to understand the model to use the software effectively.

    There is an example above in which a reader had difficulty understanding the output from AACalc. If the user has no clue whether his coefficient of risk aversion is one or ten (and who does?) and that coefficient can dramatically change the outcome, then running this guess through software doesn't improve the process. Garbage in, garbage out.

  8. Good point! Continuing the medical analogy ... self diagnosis or treatment of a medical condition is fraught with risks; the same with finances related to retirement income. I don't think retirement income is a place for hobbyists or do-it-yourself-ers anymore.

  9. Hi Dirk,

    I'm very impressed with the work you have done on this. It's one of the best pieces of analysis and writing I've seen in a long time.

    I just had a couple of specific questions. I like the points you made with the first person who commented about the surprise over the 100% stock allocation. One of the ways I deal with this with clients is to translate assets into retirement income terms and try to consider all sources of income. For example, someone with a $500,000 investment portfolio might be able to generate, say, $20,000 per year from that. But if they were also receiving $20,000 per year in SS, a 100% stock portfolio would be more like a 50/50 portfolio with SS considered as bond-like income.

    It's hard to get any of these programs to deal with actual risk tolerance, which I consider to be akin to "freak out" potential. Classic utility-based risk aversion is more like tolerance for income bouncing around from year to year. And I don't think risk tolerance questionnaires or having a human advisor necessarily provides the answer--a lot of people with advisors did what turned out to be dumb things in 2008. Perhaps some of the answer is helping clients to view things in a broad frame, e.g., being aware of all sources of income rather than focusing narrowly on investment portfolios.

    And just one comment on humans vs. robos--I'm only partly joking when I tell people: "If you're knowledgeable enough to select a good advisor, you're probably knowledgeable enough to do the planning yourself." Unfortunately, it's not like the good advisor to turkey ratio is 95/5, so there is a high likelihood of picking a good advisor; I think the odds are quite a bit worse.

    Anyway--great job on this and keep up the good work.


    1. Thanks, Joe, and "amen" to all of the above. Maybe worse than 5/95.

      Interesting that you mention the "bond-like" nature of Social Security benefits. While those benefits are not bonds, I agree that they certainly should be considered "bond-like" when choosing an asset allocation. Same with fixed annuities. If you have enough guaranteed income, you have the capacity to take more risk and you're less likely to freak out.

      The AACalc question from a reader struck me, too. I'm beginning to think that the best way to use things like a "coefficient of relative risk aversion" is "backwards", as it is something no one can know with any accuracy and equally clearly changes with market conditions.

      Maybe the best approach is to use AACalc (or similar) with several coefficients (say, 1, 5, and 10), see what bond allocation is recommended for each and pick the largest one that wouldn't have made you throw up in 2008.

      Then you have to ask if you weren't better off just taking Bernstein's advice, picking an allocation that will likely work across a broad range of possible futures and limiting it to one you can stomach in a severe downturn.

      A general point I am trying to make is that taking mediocre and/or incomplete advice and turning it into a computer application, then feeding that software with guesses for input may look to the non-computer scientist like a better answer.

      It really isn't.

      Thanks for the input and have a great 2017!

  10. This comment has been removed by the author.

  11. Regarding your exchange with Joe about viewing other non-investment "secure" income as a bond-like asset within one's household wealth - I understand the value of this concept, but in my own case, I can't decide whether my defined benefit state pension is bond-like or stock-like. I actually think it's a vague bit of both; that is, bond-like because it is contractually guaranteed, but also stock-like because the pension's long term stability is highly sensitive to the long term returns of its substantial stock and real estate allocations. So working within a floor and ceiling model, how should one think about guaranteed pensions of uncertain security? My personal answer, which I find unsatisfyingly vague, is to assume that worst case my pension might only deliver 70% of its long term income promise. But don't ask me how I came up with 70%.

    1. Barry, I actually view your DB plan as far more bond-like than stock-like, but I see your point. (Bonds aren't contractually guaranteed, though Treasury bonds are backed by the full faith of the U.S. government, for example.) I'm not sure that there is a better answer, though, than the one you came up with on your own.

      I remember reading somewhere that Wade Pfau discounts his Social Security benefits by 20% to be on the safe side should (or should I say when?) conservatives reduce Social Security retirement benefits. I discount mine quite a bit less because I am older than Wade and expect my benefits would be grandfathered.

      I imagine that Wade picked 80% the same way you picked 70% and I picked my discount – just a feeling.