More Than I (Or You) Ever Wanted To Know About Spending Variance Vs. Fail Rates.

I got a really interesting result the other day when I was shaking out some of the newer spending variance features[1][2] of the simulator I recently rebuilt. But I had forgotten, for a moment, my two primary rules of financial modeling:

2016 - Systematic Risk Strategy in Annual Isolation...

2016 looks like it has been a good year so far. I've heard of some folks hittin' 30% this year (I'm always curious about their variance; I resist the impulse to ask for a time-weighted monthly series) but that, without any question whatsoever, is not my game.  But this is:

- Green: me
- 60/40 is SPY/AGG (orange)
- Benchmark is AOM (purple) , flawed index for comparison but it's easy...and free
- Hedge Fund index (pink) is Barclays Hedge, flawed but hey, it's free
- I have < 15% stuff in GMDII that acts kinda like private equity where
  the risk, like it is in the rest of life, is not so much "variance" but rather
  the permanent loss of capital. So take that x axis with a <15% weighted
  grain of salt.

As long as I was on a roll in R...

I rewrote my mean-variance mapper/plaything in R, too.  That's it for R before Xmas.

This chart is: a 2 asset (SPY AGG) EF and a 5 asset (SPY AGG EFA IYR GLD) EF from Jan. 2014 to today.  The green dot is my trading strategy (or, more accurately, systematic alt-risk "investing") over the same time frame.  A few hours of coding now saved me a few hours of monkeying around in excel each month to generate the same thing...  A fair trade-off if one assumes that generating this in the first place is a meaningful and worthwhile effort.

What Does Opening Some Floodgates On Risk Look Like In A Retirement Plan Using a Longevity-Varying Simulator

I was running some tests on some new (spend shocks) and modified (future income streams) features of my simulator as a type of programming shakeout.  Some of what I found was kind of interesting.  For example, I ran two scenarios:

Effect on Cumulative Fail Rates by Hedging Longevity Risk With a Deferred Annuity

I wasn't really expecting this chart to look like this but it totally makes sense now that I look at it again.  In this particular case I made my 'theoretical' retirement model invest into a small deferred income annuity (enough to hedge out ~90% of 85+ spending risk).  Then, in order to finance that purchase, I sucked the related capital (the stuff required to make the annuity investment) out of my "starting pile of capital" in order to buy the annuity.  My sorta-kinda-goal in this vague quasi-analysis that I was trying to do was to reduce my overall late-age spending risk by converting some of my general risk-capital from "open and unencumbered" to "closed and dedicated to after 85 spending commitments."  This chart is what it looked like in terms of the changes in fail-rate risk. I guess that one of the basic ideas here is that one can't create something from nothing. In this case, in order to mitigate late life spending risk, the money has to come from somewhere. And in this case it comes from "early" consumption risk and then that is converted here to the risks that are "late."  If I had to characterize the chart, I'd have to say that the early risk-costs might have to be worth the late-age benefits. But then again, Ill have to think about this...  

Black line: my fail rate estimate before annuity hedging
Blue line: my longevity-adjusted fail rate estimate after hedging with a deferred income annuity
Personal Data...


A sharp eye will notice that while I thought at the outset that I was hedging out 90% of my longevity risk I did not nearly achieve that goal. That means that either I have a bug in my simulator, which is always plausible, or I don't really understand how much risk there is. Either way that will be another post.  Most likely, it's an artifact of a low base case fail rate. It's like one of those annoying life lessons I monologue at my kids: you can't, in the end, at the margin, eliminate all risk.


[postscript]

Here is another result that surprised me. I ran this while my kids were eating lunch.  I changed the assumptions to generic (standard stuff: age 65, $1m, 60/40, 4% spend, etc).  I expected the hedge to work but now I remember something from Pfau about how age and capital will affect whether it makes sense to hedge.  In this case here, the 65 year old has to spend 20% of his capital (200k.  This is based on a quote from immediateannuities.com for 8k of monthly income that splits the difference on the age85+ spending.  40k today at age 85 at 3% inflation is 72k/yr and at age 105 it's 130k/yr, so the "split diff" of 100k/12 ~= 8k/month...give or take) to hedge out a major chunk of late age spending.    In this example, the hedge does not look constructive when it comes to the baseline goal of life-cycle risk management...

Another Stab at a Longevity Hedge Analysis Using Simple DIAs

Amateur's Abstract:  Using my "hacker's" assumptions and methods on my own personal data, it is probably fair to say that buying a simple deferred annuity at my current age to hedge out a lot of late-life longevity risk vis-à-vis spending will reduce my fail rate risk by some amount that might be "worth it," it is not fair to say that it matches, dollar for dollar in present value terms, the gain in spending capacity that it (the deferred annuity purchase using my own assumptions) engenders. In other words, when it comes to buying a deferred annuity, the risk reduction benefits might have a case but spending increases might not given how I'm looking at it.

Sheer Proximity or The Purpose Of An Early Retirement

I've more often than not had a serious amount of second guesses (not the same as regret, btw) on my partially voluntary decision to retire early around age 50.  When I read about early retirement in the media and in misc blogs the sense I get, and the way it is framed verbally, is that it is all about financial independence.  An early retirement blogger that I like to read a lot, financialsamurai.com, puts it this way (indirectly because here he is talking about prestige -- but his point, I think (or am I projecting), is that prestige is a block when it comes to early retirement and that financial independence is a "good"):  

MC sim V2.201 complete and I'm done messing around with it

Ok, this sim project was fun enough but my kids are hungry and the dirty laundry is piling up. Don't ask about the sink.  In the final stretch I added a feature for a couple future streams of income like SS or annuities. On the other hand, I don't have any features for separate taxable and non-taxable portfolios or IRA start dates and the like. Anyway, I finished up. I cleaned off the desk and ran it on myself.  Before I show the image, note that: a) six or seven years ago I had fail rates that were in very high double digits, b) I started building this not because I am enamored of simulators but rather I was both ticked off at my fee-charging planner and I also wanted to learn some modern software development skills, and c) I'm not webifying this thing or making it available for commercial use.  This is just for me for the heck of it.  Now, if you were to want to send me a check or a series of checks, that's another matter altogether.  My latest version:

-fail rate under 2% for this run after a half-provision for SS was added
-30,000 sim runs; took about 7 minutes
-assumptions are private and the term. wealth axis blanked out

Here is Another Way to Look at Fail Rates

Once you let a simulator vary longevity by actuarial-type tables additional ways to view retirement fail rates become available.  Here in this post is one more way.

For example, let a simulator, given some overly simplistic and generic assumptions[1], blow out some random versions of the future  (say 10k of them) based on what little it knows, and you can get a sense of how fail rates might change with longevity expectations.  I hadn't thought of this particular one before but I've added it to the repertoire of charts I'll keep an eye on.  While personally I usually plan for the worst (worst? really?) case of 95 or 105, reality tells me that earlier ages will make more sense when push comes to shove, if you will.

This effort here, fwiw, is another way of helping me not obsess about future-risk so I can play a little harder today...rather than live like a monk fearing an uncertain future.  My SO frets that I overdo this stuff but little does she know that she will benefit if I can spend more today due to my ocd over-analysis.

Left axis - fail rate for given longevity expectation
Right axis - freq distribution of terminal age in simulation. Matches SS tables
Points - the change in fail rate expectation for increases in longevity expectation












--------------------------------------------------------------------------
[1] age 60, $1M, 4% constant spend, 60/40, some taxes and fees, no outside income, etc. etc.

Contemplating a Terminal-Age-Weighted Retirement Sim

Two things before I get to my post:

1. I'm not totally sure this advances, in any significant way, any one real person's (even my own) real retirement, and

2. Everyone and their brother -- or at least among leading personal finance researchers like Milevsky, Pfau, Blanchett, Kitces, Dirk Cotton, and a bunch of others -- have all weighed in at one point or another over the last couple years on the weaknesses and misdirects and emptiness of retirement simulators, a point of view I happen to agree with.

But...I just felt like playing around with this one more time.  In this particular case I wanted to layer in realistic probabilities for longevity into a working retirement sim and see how it goes.  I had that last year, too, but I did not have a good platform on which to play around with it. Now I do.

Weekend Links - Dec 16

QUOTE OF THE DAY

“I think everybody should get rich and famous and do everything they ever dreamed of so they can see that it’s not the answer.” – Jim Carrey

CHART OF THE DAY

RETIREMENT FINANCE AND PLANNING

Retirement Isn’t Always Fair, Boston College.  Retirement experts, including economists at this Center, urge baby boomers to hold on and work just a few more years to improve their retirement finances. But less-educated older workers often have physically demanding jobs or poorer health, making this very challenging, or even impossible. “It may well be,” the researchers conclude, “that their retirement shortfalls cannot be bridged by working longer and other solutions will be needed.”  [Comment: empathy, yes, of course, but I have to say: beware the "other solutions" if they were to be in the hands of Boston College professors]  

What Does Spending Look Like In Poor, Great, And AverageRetirement Conditions? Pfau.  

[comment: there is a hell of a lot of rules and guidelines in this article.  My advice: constant vigilance and triangulation.  There are no rules that will save you.  Each year, you have to make a judgement about what you have, what you expect,  and what you plan to do this year and for a couple years looking forward based on what you think about future returns, inflation, spending, blah blah blah.  I'll do a post on this sometime because I think rules can be a distracting crutch.  Constant, vigilant adaptation is the only plan that works. That, and making sure you have enough when you start…which means that early retirement is fraught…]

What Would a Pessimistic View of Next-10-Year Returns do to Simulated Fail Rates?

In the process of running some tests on my rebuilt simulator, I ran a set of paired tests, each one with 10x1000runs[1]:

  1. Once with a slightly modified table of historical returns,

  2. A second time with same table as a source but returns are
      suppressed in the first 10 years of each sim by a couple pct.
   

The first test had an average fail rate of 9.3% [2].
The second test had an average fail of 15.8%.

-------------------------------------------------
[1] Short runs but I didn't want to wait for 10k runs. Assumed age 65, $1M endowment, 40k constant inflated spend, 60/40 portfolio, uncertain longevity using 2013 SS life table, factor for fees and taxes, etc.

[2] Seems high but note that there are taxes, fees, and fully dispersed longevity uncertainty in there. I wish I could figure out how to feather in some autocorrelation but I couldn't.

if anyone is nerdy enough to follow this blog...

For geeks only:
I just rewrote my archaic (let's just call it a slide rule or maybe two tin cans with a string or maybe an abacus) Excel 2002+visual-basic Monte Carlo simulator -- but now in R. Was I surprised at  the efficiency? No, not really.  The gain in time for 5,000-10,000 runs? let's call it around 80%. I'll bet if I hired a programmer I could tamp that 80% down even more...but I'm not going to do that...cuz I'm cheap. And lazy. I've got 99% of what I need.  

Longevity Uncertainty 3

I was trying to build enough rudimentary programming skills in R last week that I could maybe consider rewriting my Monte Carlo simulator into something other than an old version of Excel.  Using the 2013 SS table (male) to shake out something I was trying to do with nested loops and R "lists" I ended up with this chart which is a more granular version (data for each age vs every 10 years) of a previously posted chart on longevity uncertainty that used boxplots[1] (same visual, I just needed something for each age in my exercise) . I thought I'd throw it out there as long as I had it.  For retirees, the main point is probably to think more carefully about a planning horizon before just blindly sticking "30 years" or "to age 82" into an online retirement calculator that is asking for "duration."

(+/-)std = plus or minus 1 standard deviation.  I did not label the chart well but I'm not doing it over.

[1] Given the non-normal distributions involved, the quartile boxplots are probably a better representation of dispersion than using standard deviation but then my goal was "programming training" and not statistics. Also median is probably better than mean but...same excuse.

My Personal Residential Real Estate Planning Chart

(I live in South FL)

http://thisisindexed.com/2016/11/asset-liquidity/

Weekend Links - Dec 9

QUOTE OF THE DAY

“Risk means more things can happen than will happen.” -Elroy Dimson

"Uncertainty about the future does not necessarily equate with risk, because risk has another component: Consequences."  -Farnam Street

CHART OF THE DAY

RETIREMENT FINANCE AND PLANNING

How Much Wealth Will You Have 30 Years Into Retirement? Pfau.  Considering spending and wealth are both important—as retirees should not be narrowly focused on a singular goal to avoid financial wealth depletion—financial goals for retirement can essentially be reduced to two competing objectives: to support as much spending as feasible, and to maintain a reserve of financial assets to support risk management objectives such as protecting from spending shocks or otherwise provide a legacy.  

The 'Never-Retirement' Plan: Many Millennials Plan To KeepWorking, Survey Says.  Forbes  

Optimizing Retirement Income Solutions in DefinedContribution Retirement Plans A Framework for Building Retirement IncomePortfolios. Pfau, Tomlinson, Vernon. SoA.org.   [Individual retirees can jump to page 43 first. I might've linked this before]  

Women Face 20% Higher Health-Care Costs in Retirement, SurveyFinds, WSJ.  Longevity seen as the reason for the gap with men. 

Cage Match: PMT function, RMD formula, and Divide-by-10 Rule

The other day a former colleague posed a hypothetical question related to teasing out ideas for a trust rule that would provide for an orderly, sustainable and mostly full disgorgement of assets later in the life of a beneficiary if anything sizable enough were to be left at that age for him or her.

Crude Futures Options Volatility Surface

This is another one of those unnecessary visualizations.  I was reading about vol surfaces and decided to see if I could create one for something I was trying to trade (near dated short crude calls) to see what it looked like.  This is what it looked like; first chart uses delta while the second uses strike price:

A Quick Visual of the Anatomy of a 3D 5-Asset Mean-Variance Map

I'll assume you know the principles here and some of the math. I just wanted to run through a step-wise visual of the internal structure of a 3D 5-asset mean variance map the way I've been looking at it lately. I have no idea how easily this might be discredited.  I did this for myself to make sure I understood a previous post and I'm putting it here just for the heck of it;  I'm not so sure there is any practical purpose here, this is more just-for-fun.  The charts use AGG SPY EFA IYR and GLD for the asset classes and the time frame is 34 months in 2014-2016. Returns are annualized total returns. Std dev is annualized from a monthly series. This timeframe (and data) is (are) pretty arbitrary; I happened to have done some other study earlier this year that used this same data and I was too lazy to build something different.

3D Efficient Frontier

I realized I probably got a little ahead of myself in a previous post .  Kicking out a 3D scatter-mass for 5 asset classes in return-deviation-diversification space is a bit much.  I went back and looked at just bonds and stocks (aggregate bond or AGG, and US large cap in SPY form) over 34 months in 2014-2016 to create a standard efficient frontier for two asset classes.  The allocation ranges from 100% AGG to 100% SPY.  This is similar to what you see in  most finance textbooks.  Allocating between the two adds efficiency in return and standard deviation which can be seen in how the curve bends up and left as one adds stocks to a 100% bond portfolio.  Nothing new here, I just added a third dimension that gives some depth to the allocation dimension.  The start of the GIF is what it looks like in 2D. There is not much to say here, just showing what it looks like...

Weekend Links

QUOTE OF THE DAY

"...the world is not a predictable casino game…All we can do is make intelligent estimations, work to get the rough order of magnitude right, understand the consequences if we’re wrong, and always be sure to never fool ourselves after the fact."  Shane Parrish farnamstreetblog.com

"You should fret less about getting any particular decision precisely right, and instead worry about whether you’re tackling the right range of issues." jonathanclements.com

CHART OF THE DAY

3D visualization of mean-variance-diversification for ~4500 out of >4.6M combinations of 5 asset classes (large cap US, aggregate bond, international developed, gold, and real estate) over 34 months in 2014 to 2016.  A diversification function with respect to the relative concentration of the allocations is used for the z axis.  I had the main post here. Had to give myself COTD, right?

RETIREMENT FINANCE AND PLANNING

The Five Types of Retirement, MoneyBoss.com.  You see, while the idea of retirement might be relatively young, it’s achieved a level of complexity that I find fascinating. Retirement is no longer one thing. It’s many things. Or many possibilities. I thought it might be fun to visualize what I consider the five most common kinds of retirement in our current economy. 

How to Understand the ‘Probability of Success’ Metric forRetirement, David Blanchett in the WSJ.  The simplicity of “success” in this metric belies the complexity of “probability.” Because no one can predict the future, this approach requires a bunch of random projections that can be used to estimate a probability.  A better name for the metric would probably be “educated guess,” but that doesn’t sound as mathy.  

Visualizing A Missing Piece of Monte Carlo Simulation in 3D

Running out of money at 75 and living to 95 is much more catastrophic than depleting savings at 84 and dying at 85, but both cases are treated the same in determining failure rate. - Joe Tomlinson

One of the many and manifest shortcomings of doing MonteCarlo simulation -- along with: a) the opacity of the meaning of a fail rate, b) the insufficient acknowledgement of likely behavioral changes in the face of an early fail warning, and c) the usual lack of a proper "triangulating context" ... among other things -- is the failure to quantify or visualize the magnitude of the "fail:" what age, how long, how much, etc.  This is often commented upon but I have never seen it visualized.  It may be out there, I just haven't seen it.  So with my new playthings (beginner-level R and scatterplot3D) I thought I'd throw a visualization out there just to see what it looks like.