1. The Actuarial Balance sheet. See Ken Steiner’s blog before mine. He covers it better.
My own BS (pun intended) has a full compendium of assets and liabilities including “flow” items like SS and future spending. Assets also include things like private direct investments valued by model, real estate, and miscellaneous hard assets. Liabilities include both hard (tax, mortgage) and soft/mezzanine (higher education exposure) items. The estimate of the present value of spending, which is ‘shaped’ to my plan, is a simple PV calc where inflation and discount rate are assumed equal for a moment (no discussion by me on discounting here) but also weighted by a conditional survival probability.
I then filter my BS as needed:
1) traditional or classic accounting net worth,
2) my version of “net monetizable assets” for retirement (NMA becomes my denominator for spending analysis), and
3) full actuarial net worth including the spend liability.
That last becomes a “feasibility” calc or “risk capacity” or test of “boundary condition for annuitization” when rendered as a ratio of net wealth to the current valuation of spending*. That’s another post, though… I update this BS monthly as part of a routine cycle. For more detailed commentary, email me.
The balance sheet gives me a periodic feasibility check (is W/spend >1? will it be?) which, as I inferred above, is essential but I have to be careful here. In some modeling I did this year I played a game where, over future years, I forced spending in a dynamic process rolling forward into the future to be whatever coerced W/spend to be =1 under some stress scenarios. That created some situations where in late years the plan crashed by flying too close to the sun by spending too much without taking a hard look at sustainability over remaining time. I once blogged that there is some math (Robinson and Tahani, 2005) that shows that a Geometric Brownian net wealth process (let’s call it simulation/sustainability for now) "going out" can be transformed in calculus to a Stochastic present value of spending "coming in," i.e., feasibility and sustainability are sorta the same thing. But that is in theory. I think in real practice one should be careful and always do feasibility (balance sheet) separately but paired closely with sustainability (Monte Carlo simulation or similar).
2. Income Statement. I keep track of income (mostly investment income but there is other stuff) and all expenses. I do this monthly and I’ve done it for 10 years. I don’t understand people that say this is too hard. Sure it’s annoying but the payoff will matter when times are hard so I do it religiously. In 2012 I went through three rounds of cost cutting to drop my risk from very very high to ok. Knowing what, exactly, I spent made this a fairly concise process over a relatively short interval. Might have to do it again. Also, if you don't know your spend rate, why are we here? Further, is your spending trending up or down? You probably have no idea and should.
Some expenses I have capitalized. For example, my education exposure is on the BS as a PV liability. In effect I have already spent it so I don’t make it hit the income statement. I know it is there, though when cash flow planning. This accounting trick is a game I play that I’m guessing others don’t. As in #1 if you are curious about something here, email me.
3. Spending Statistical Control Chart. I will admit this process is a little anal and I might've not done it if I were subjected to similar circumstances in alternative-history-world. But I was once close to the edge of the retirement abyss and now make a habit of keeping this up because it was useful to me once. Basically, I track my spending every month as a percent of net monetizable assets i.e., it's a rate, a percent. The problem is that that rate is influenced by both the numerator (spending), which I mostly control, and the denominator (net assets) which I don’t have a ton of control over. Otoh, it gives me a classic spend rate in a time series so I can see trends, variance, and jumps which I, the anal blogger that I am, like. I benchmark the rate to a couple things: 1) statistical control boundaries (high mid low, mid could be 3.4 or 5% for example, depending on age and circumstance) that are set based on my knowledge of age-adjusted risk. This age-adjusted-spend-risk is the bread and butter of 10,000 retirement finance papers and not discussed here, and 2) what I want to call Bollinger bands (+/- 2 standard deviations from a 12 month trailing SMA of the monthly spend). I annualize spending, btw, so the number makes sense to me. Note that if the markets dive, the rate will go up so it’s hard to use this method with full faith all the time. TBD
A quick note on spending. Since most advisors have a lot of control over recommendations about risk and portfolio construction, and to a lesser extent fees -- not to mention much else in my world -- you will often hear a ton about asset allocation in planning and/or investment meetings. What they don’t tell you, or not often enough maybe, is that of the four great levers over a decumulation project – asset allocation, date of retirement, income in retirement from either inside or outside of the plan, and consumption – the latter three will dominate the first. Ignoring all but the last for now, I can say that spending is the number 1 thing for me to watch in retirement. Asset allocation can injure at the extremes, of course but is otherwise is a soft tool in the middle. Spending is a make or break killer. The problem there though is sometimes not all spending choice is really in one’s own hands. That’s why caution in spending early in retirement is warranted and why shaming tactics from friends, other retirees, advisors, or academics always hit a pretty hard wall in my head.
4. “Lifetime Probability of Ruin” Tool. This is similar to Monte Carlo simulation but not quite. The math of LPR is formally described in a partial differential equation developed by Kolmogorov back in the 30s. Milevsky makes it understandable in his "7 Equations" book. This PDE can be solved via a finite difference solution which to my naïve eye is not remarkably different from simulation when I look closely, but whatever. There are other ways to do it. For example, one can: a) simulate “portfolio longevity in years,” b) take it’s full likelihood distribution, and then c) weight that by a conditional survival probability (given the age of the person being evaluated). I like this approach because it takes into consideration the full scope of asset exhaustion possibilities all the way to infinity, un-bound by an arbitrary planning horizon (e.g., 30 years) and marries that distribution to the full term structure of mortality. This feels more complete to me than an MC sim to "30 years." “But, gosh, you don’t have any auto-regression/mean-reversion or volatility-of-volatility or the exact tax code, or …).” Yeah, uh, I don’t care. I am not predicting the future here, I am just doing a quick thumbnail gauge on risk. It's a dashboard metric: "oops, going too fast, better slow down." If it is super high, I pay attention. If it is changing rapidly, I pay attention. Specific numbers or minor jumps do not interest me. This tool, btw, is a custom code thing. Also, I do this evaluation very rarely -- say annually -- but I’d do it with great frequency the closer I get to the edge of the abyss.
What this tool gains in speed (a few seconds), simplicity (a few lines of code), and "depth of my conviction" -- by allowing the full constellation of asset exhaustion possibilities (phrase by Albrecht and Maurer) and the full term structure or mortality (phrase by Milevsky) -- it loses in being able to test flexible patterns, or “shapes,” of consumption. I could probably make that happen but then I’d have a MonteCarlo sim by another name and I have largely abandoned MC sim (see below) though this and the next tool are largely ‘that’ depending on how you look at it. To take the other side yet again, I do like this tool a bit because it can be disaggregated into simpler modules where one module looks only at the distribution of “portfolio longevity in years” un-constrained by planning horizons or survival probabilities. That is a pretty useful tool if you know what you are looking at on the output side.
5. Expected, Discounted Utility of Lifetime Consumption. I created a proprietary tool once that evaluates the sum of expected discounted utility of lifecycle consumption ... with an embedded feature that "snaps" spending in the model to available income when wealth depletes before death. I am not a huge fan of utility theory, not least because I have no idea what my coefficient of risk aversion is. But I was a trader for a good long while. And traders, or at least the profitable ones anyway, understand that life is not just about probabilities, it’s also about payoffs and consequences. I can have a trading system that succeeds 30% of the time (wrong 70%) and still make money because I manage the risk/return payoffs, i.e., the consequences. Same thing in retirement. There may be probabilities of such-and-such but the consequences matter and sometimes matter a lot (running out of money). The convexity of the utility calc, for all its imperfections, captures that payoff structure pretty darn well. I mean up to a point. My tongue-in-cheek opinion is that the hyper-risk-averse probably need more mental health help than financial modeling but that’s another story. Makes you wonder about risk aversion, though.
The other thing about a tool like this is that it turns out that the “shape” of spending (trends down, trends up, step functions, smile shape, hump shape, etc) matters quite a bit, certainly more than allocation does. Ruin calcs might give us some indirect insight on this but the curve of the utility math over a full lifecycle is better. It allows me to test the shapes of plans in a way other tools like feasibility or LPR can’t. I can also, even though I am a skeptic, see what happens if I am more or less afraid of the future via risk aversion. This captures psychology in a way other tools can’t. It’d be hard to explain to someone else, though. This was a custom project, btw. Note that there are some debates in the lit about reasonableness of some of the assumptions used to build things like this. Personally, I don’t care about that. This tool is not gospel or prophesy or an orbuculum, it's just another sanity double-check, to be re-run next month when things change. When I talk about triangulation, this is one of the points I try to make: many tools, checked periodically.
As a side note: as best as I can tell, this is the same as or very similar to the engine that underlies the MaxFi planner from Kotlikoff. Well, I mean it's my crayola on cardboard version to his oil painting on canvas, not least since I do not have any kind of spending pattern optimization. But maybe I could. The two tools at least speak a similar dialect which is reassuring since I consider MaxFi state of the art. But maybe that is just me projecting... ;-)
6. Monte Carlo Simulation. Ok, I don't really use this much anymore. The first thing I ever built was an MC sim in Excel. Took three hours to run and locked up my computer. I rewrote it in R probably 5 times and, while it now runs faster (minutes), it is also obsolete in the face of #4 and #5 both of which actually use simulation as an underlying engine...so they are all playing a similar game. This tool has the advantage of being very flexible when designing non-constant consumption plans and factoring in esoterica like odd tax issues. It is limited, however, by horizon myopia and the standard issue with all MC sims: proper interpretation of the results and the frequent failure to evaluate magnitude (in another context we'd call that payoffs or consequences). Fixed horizons, which I sometimes flip in my MC tool to random lifetime (this complicates interpretation), and weaknesses in the theory and meaning of fail rates, mean that I stopped using this a few years ago. It's there if I need it, though. I probably won't need it but if I do I'm more likely to re-write it from scratch than use it as is..."as is" means overloaded with research features that no are longer helpful and that I can't remember what I was trying to do.
7. Spreadsheets. 9 times out of 10 I have no idea what my code for my other tools are doing anymore. I wrote in little things here and there over time to test ideas and now, x years later, the code is obscure and a little brittle. Sometimes it’s easier to just knock out something in Excel, with which I’ve been working for decades. “Omg, that’s so deterministic, not stochastic.” Yeah, uh, so? Sometimes insight comes from simple things. This generalist facility in spreadsheets is probably more important to me than my other tools or my advisor. I can use my imagination, curiosity, and my willingness to bore myself to tears in spreadsheet-world to explore ideas, problems, changes in circumstance, etc. It's as infinite as the mind can conceive. I started in Lotus 123, btw. Anyone else remember that one?
The balance sheet gives me a periodic feasibility check (is W/spend >1? will it be?) which, as I inferred above, is essential but I have to be careful here. In some modeling I did this year I played a game where, over future years, I forced spending in a dynamic process rolling forward into the future to be whatever coerced W/spend to be =1 under some stress scenarios. That created some situations where in late years the plan crashed by flying too close to the sun by spending too much without taking a hard look at sustainability over remaining time. I once blogged that there is some math (Robinson and Tahani, 2005) that shows that a Geometric Brownian net wealth process (let’s call it simulation/sustainability for now) "going out" can be transformed in calculus to a Stochastic present value of spending "coming in," i.e., feasibility and sustainability are sorta the same thing. But that is in theory. I think in real practice one should be careful and always do feasibility (balance sheet) separately but paired closely with sustainability (Monte Carlo simulation or similar).
2. Income Statement. I keep track of income (mostly investment income but there is other stuff) and all expenses. I do this monthly and I’ve done it for 10 years. I don’t understand people that say this is too hard. Sure it’s annoying but the payoff will matter when times are hard so I do it religiously. In 2012 I went through three rounds of cost cutting to drop my risk from very very high to ok. Knowing what, exactly, I spent made this a fairly concise process over a relatively short interval. Might have to do it again. Also, if you don't know your spend rate, why are we here? Further, is your spending trending up or down? You probably have no idea and should.
Some expenses I have capitalized. For example, my education exposure is on the BS as a PV liability. In effect I have already spent it so I don’t make it hit the income statement. I know it is there, though when cash flow planning. This accounting trick is a game I play that I’m guessing others don’t. As in #1 if you are curious about something here, email me.
3. Spending Statistical Control Chart. I will admit this process is a little anal and I might've not done it if I were subjected to similar circumstances in alternative-history-world. But I was once close to the edge of the retirement abyss and now make a habit of keeping this up because it was useful to me once. Basically, I track my spending every month as a percent of net monetizable assets i.e., it's a rate, a percent. The problem is that that rate is influenced by both the numerator (spending), which I mostly control, and the denominator (net assets) which I don’t have a ton of control over. Otoh, it gives me a classic spend rate in a time series so I can see trends, variance, and jumps which I, the anal blogger that I am, like. I benchmark the rate to a couple things: 1) statistical control boundaries (high mid low, mid could be 3.4 or 5% for example, depending on age and circumstance) that are set based on my knowledge of age-adjusted risk. This age-adjusted-spend-risk is the bread and butter of 10,000 retirement finance papers and not discussed here, and 2) what I want to call Bollinger bands (+/- 2 standard deviations from a 12 month trailing SMA of the monthly spend). I annualize spending, btw, so the number makes sense to me. Note that if the markets dive, the rate will go up so it’s hard to use this method with full faith all the time. TBD
A quick note on spending. Since most advisors have a lot of control over recommendations about risk and portfolio construction, and to a lesser extent fees -- not to mention much else in my world -- you will often hear a ton about asset allocation in planning and/or investment meetings. What they don’t tell you, or not often enough maybe, is that of the four great levers over a decumulation project – asset allocation, date of retirement, income in retirement from either inside or outside of the plan, and consumption – the latter three will dominate the first. Ignoring all but the last for now, I can say that spending is the number 1 thing for me to watch in retirement. Asset allocation can injure at the extremes, of course but is otherwise is a soft tool in the middle. Spending is a make or break killer. The problem there though is sometimes not all spending choice is really in one’s own hands. That’s why caution in spending early in retirement is warranted and why shaming tactics from friends, other retirees, advisors, or academics always hit a pretty hard wall in my head.
4. “Lifetime Probability of Ruin” Tool. This is similar to Monte Carlo simulation but not quite. The math of LPR is formally described in a partial differential equation developed by Kolmogorov back in the 30s. Milevsky makes it understandable in his "7 Equations" book. This PDE can be solved via a finite difference solution which to my naïve eye is not remarkably different from simulation when I look closely, but whatever. There are other ways to do it. For example, one can: a) simulate “portfolio longevity in years,” b) take it’s full likelihood distribution, and then c) weight that by a conditional survival probability (given the age of the person being evaluated). I like this approach because it takes into consideration the full scope of asset exhaustion possibilities all the way to infinity, un-bound by an arbitrary planning horizon (e.g., 30 years) and marries that distribution to the full term structure of mortality. This feels more complete to me than an MC sim to "30 years." “But, gosh, you don’t have any auto-regression/mean-reversion or volatility-of-volatility or the exact tax code, or …).” Yeah, uh, I don’t care. I am not predicting the future here, I am just doing a quick thumbnail gauge on risk. It's a dashboard metric: "oops, going too fast, better slow down." If it is super high, I pay attention. If it is changing rapidly, I pay attention. Specific numbers or minor jumps do not interest me. This tool, btw, is a custom code thing. Also, I do this evaluation very rarely -- say annually -- but I’d do it with great frequency the closer I get to the edge of the abyss.
What this tool gains in speed (a few seconds), simplicity (a few lines of code), and "depth of my conviction" -- by allowing the full constellation of asset exhaustion possibilities (phrase by Albrecht and Maurer) and the full term structure or mortality (phrase by Milevsky) -- it loses in being able to test flexible patterns, or “shapes,” of consumption. I could probably make that happen but then I’d have a MonteCarlo sim by another name and I have largely abandoned MC sim (see below) though this and the next tool are largely ‘that’ depending on how you look at it. To take the other side yet again, I do like this tool a bit because it can be disaggregated into simpler modules where one module looks only at the distribution of “portfolio longevity in years” un-constrained by planning horizons or survival probabilities. That is a pretty useful tool if you know what you are looking at on the output side.
5. Expected, Discounted Utility of Lifetime Consumption. I created a proprietary tool once that evaluates the sum of expected discounted utility of lifecycle consumption ... with an embedded feature that "snaps" spending in the model to available income when wealth depletes before death. I am not a huge fan of utility theory, not least because I have no idea what my coefficient of risk aversion is. But I was a trader for a good long while. And traders, or at least the profitable ones anyway, understand that life is not just about probabilities, it’s also about payoffs and consequences. I can have a trading system that succeeds 30% of the time (wrong 70%) and still make money because I manage the risk/return payoffs, i.e., the consequences. Same thing in retirement. There may be probabilities of such-and-such but the consequences matter and sometimes matter a lot (running out of money). The convexity of the utility calc, for all its imperfections, captures that payoff structure pretty darn well. I mean up to a point. My tongue-in-cheek opinion is that the hyper-risk-averse probably need more mental health help than financial modeling but that’s another story. Makes you wonder about risk aversion, though.
The other thing about a tool like this is that it turns out that the “shape” of spending (trends down, trends up, step functions, smile shape, hump shape, etc) matters quite a bit, certainly more than allocation does. Ruin calcs might give us some indirect insight on this but the curve of the utility math over a full lifecycle is better. It allows me to test the shapes of plans in a way other tools like feasibility or LPR can’t. I can also, even though I am a skeptic, see what happens if I am more or less afraid of the future via risk aversion. This captures psychology in a way other tools can’t. It’d be hard to explain to someone else, though. This was a custom project, btw. Note that there are some debates in the lit about reasonableness of some of the assumptions used to build things like this. Personally, I don’t care about that. This tool is not gospel or prophesy or an orbuculum, it's just another sanity double-check, to be re-run next month when things change. When I talk about triangulation, this is one of the points I try to make: many tools, checked periodically.
As a side note: as best as I can tell, this is the same as or very similar to the engine that underlies the MaxFi planner from Kotlikoff. Well, I mean it's my crayola on cardboard version to his oil painting on canvas, not least since I do not have any kind of spending pattern optimization. But maybe I could. The two tools at least speak a similar dialect which is reassuring since I consider MaxFi state of the art. But maybe that is just me projecting... ;-)
6. Monte Carlo Simulation. Ok, I don't really use this much anymore. The first thing I ever built was an MC sim in Excel. Took three hours to run and locked up my computer. I rewrote it in R probably 5 times and, while it now runs faster (minutes), it is also obsolete in the face of #4 and #5 both of which actually use simulation as an underlying engine...so they are all playing a similar game. This tool has the advantage of being very flexible when designing non-constant consumption plans and factoring in esoterica like odd tax issues. It is limited, however, by horizon myopia and the standard issue with all MC sims: proper interpretation of the results and the frequent failure to evaluate magnitude (in another context we'd call that payoffs or consequences). Fixed horizons, which I sometimes flip in my MC tool to random lifetime (this complicates interpretation), and weaknesses in the theory and meaning of fail rates, mean that I stopped using this a few years ago. It's there if I need it, though. I probably won't need it but if I do I'm more likely to re-write it from scratch than use it as is..."as is" means overloaded with research features that no are longer helpful and that I can't remember what I was trying to do.
7. Spreadsheets. 9 times out of 10 I have no idea what my code for my other tools are doing anymore. I wrote in little things here and there over time to test ideas and now, x years later, the code is obscure and a little brittle. Sometimes it’s easier to just knock out something in Excel, with which I’ve been working for decades. “Omg, that’s so deterministic, not stochastic.” Yeah, uh, so? Sometimes insight comes from simple things. This generalist facility in spreadsheets is probably more important to me than my other tools or my advisor. I can use my imagination, curiosity, and my willingness to bore myself to tears in spreadsheet-world to explore ideas, problems, changes in circumstance, etc. It's as infinite as the mind can conceive. I started in Lotus 123, btw. Anyone else remember that one?
I realize I come from a finance background and am a bit of a sperg on this stuff but I've also come to discover that very few retirees know or like or use spreadsheets. Their loss, I guess. If you are a spreadsheet avoider, better hope you have a good advisor who is selling him or herself at a reasonable price and has some fair amount of attention to spare you.
8. Common Sense and Triangulation. I’ve come to appreciate, kind of late in my cycle, common sense and the use of many tools to attack the phenomenology of retirement finance as opposed to large, elegant, expensive, single-point and highly integrated solutions and tools. I’ve had many opportunities to talk retirement with normal people after I mention my blog. To 99/100 of them, it appears that I have over engineered things and they walk away knowing that your avg person on the street has, by hook or crook, figured out a way to navigate things just fine. That’s not to say they could not be in a better place now if they had pursued an alternative history years ago. They could have. And that’s where common sense + many-tools comes in for me today. I have the opportunity now to pick better forks in the road with a little help and a little normal judgement.
9. Other. Since I play around in this field, there are always other tools and other methods that I run into. The blog and the paper I did discuss some of these ideas in more depth. I don't list them here because I do not use them or have not used them consistently or recently. Simple formulas and spreadsheets can sometimes absolutely destroy more complex "products and solutions" due entirely to transparency, ease of use, and simple interpretations. Never disdain a simple formula until you've tried it.
8. Common Sense and Triangulation. I’ve come to appreciate, kind of late in my cycle, common sense and the use of many tools to attack the phenomenology of retirement finance as opposed to large, elegant, expensive, single-point and highly integrated solutions and tools. I’ve had many opportunities to talk retirement with normal people after I mention my blog. To 99/100 of them, it appears that I have over engineered things and they walk away knowing that your avg person on the street has, by hook or crook, figured out a way to navigate things just fine. That’s not to say they could not be in a better place now if they had pursued an alternative history years ago. They could have. And that’s where common sense + many-tools comes in for me today. I have the opportunity now to pick better forks in the road with a little help and a little normal judgement.
9. Other. Since I play around in this field, there are always other tools and other methods that I run into. The blog and the paper I did discuss some of these ideas in more depth. I don't list them here because I do not use them or have not used them consistently or recently. Simple formulas and spreadsheets can sometimes absolutely destroy more complex "products and solutions" due entirely to transparency, ease of use, and simple interpretations. Never disdain a simple formula until you've tried it.
However, I will say that one pretty complex area area where I am very very very interested in is in AI or reinforcement learning as applied to personal finance. Gordon Irlam is in front of this wave and I am still trying to absorb what is going on here. But it's like learning Chinese at 61 and not 6...a bit of an effort to do and what sinks in sometimes doesn't stick. Stay tuned on that. There is a stub or prototype at aiplanner.com that is worth a preliminary look. TBD.
--------------------------------------------
* one could substitute an annuity price for the spending PV is one were to be comfortable that the annuity would actually defease variable and inflating spending. That's another story.
No comments:
Post a Comment