How to Check Your Retirement Assumptions

The following is a guest post from Monty Hothersall, co-founder of Financial Fate.

If you’ve ever projected your financial or retirement planning numbers using an online calculator, you might be surprised how much those results have changed over the past 12 months.  But this probably makes sense given the market’s free fall, right? 

Well, not exactly.  Since most of these online calculators use a statistical program called Monte Carlo Simulation (MCS), you might easily get a false sense of the future due to the problematic assumptions which drive the MCS results.  To use an example, let’s say your results from last year’s MCS planning said you had a rock-solid 98% chance of successfully meeting your retirement goals.  (But today it’s a much different story.)  So, was the 98% result a realistic number and you just had some unfortunate luck? 

Not likely.  The problem is that MCS is based on a risk model where the future market returns are projected to be “normal.”  Where in reality, market returns are far from normal because they are “uncertain.”  And since the underlying assumptions of MCS are not valid, the 98% result was not valid.  Maybe it should have been 68% or even 38%, who knows? 

Instead of using MCS based tools, you need predictive financial-modeling tools (like on my free site, FinancialFate.com, or others like ESPlanner's non-MCS solution) which allow you to challenge the rates of return by asset class and question the targeted future asset allocations.  And you must be able to realistically forecast the volatility of your free cash-flows which will be available for future investing.  This would allow you to control your scenario planning rather than "flying on auto-pilot" with MCS technology that wasn't designed for predicting market returns with all its inherent uncertainty.

Also, when developing a long-range financial plan remember that it will take 3 to 5 hours of your time.  But it usually requires only one planning session per year unless something dramatic happens, and each subsequent year should take less time than the previous year.  Tools like Fidelity's My Plan which take less and 5 minutes to generate a plan are akin to taking diet pills to lose weight instead of dieting and exercising; it's easy but it just doesn't work.

Once you allocate the appropriate amount of time, the next step is to find software or an online tool that utilizes rules-based forecasting to automatically project all your "sources and uses" of cash-flows.  These projections should be in line-item-detail and projected year-by-year into the future with total transparency.  If the output reports that support your results are not fully disclosed and available for your review, don't believe the results.  (Even if you don't plan to actually scrutinize the projections, it's nice to know they could be audited.)  And you should bear in mind these four categories that all impact each other as you plan for your financial future:

• Income Taxes:  While most solutions determine future tax bills based solely on a percentage of future income, that’s not how taxes are calculated.  This short-cut methodology neglects significant adjustments in your AGI, deductions, exemptions, tax credits, and the actual tax tables (both state and federal).  For example, if one plans to move from California to Texas in 2 years, it would have a huge impact their state tax bill.

• Expenses:  Each expense line-item will behave differently in the future.  For example, when projecting healthcare premiums, fluctuations will be due to: 1) a changing family size, 2) a future retirement, 3) future Medicare benefits for each spouse, 4) Flexible Spending Accounts, 5) tax implications, and 6) medical cost inflation.  You should be able to view these cash-flows and change underlying assumptions such as the age that Medicare kicks in, if desired.

• Income:  If one plans to retire early, go part-time, or start a company in the future, this will impact such things as Social Security and pension benefits.

• Investments:  In order to allow dynamic "what-if?" analysis to be preformed, you should be given the flexibility to select long-term expected annual returns (growth &/or income) by asset class.  You also should be able to select investment risk tolerances (asset allocations) that shift as your time horizon changes for goals like retirement and education.

Comments

[Rob Bennett]

The problem is that MCS is based on a risk model where the future market returns are projected to be “normal.” Where in reality, market returns are far from normal because they are “uncertain.”

I think you are halfway there but do not yet have it quite right.

Yes, models that assume normal returns are dangerous.

No, returns are not all that uncertain. Long-term returns are predictable enough for effective retirement planning purposes.

I've done extensive research into the historical stock-return data re this question. What the data shows is that normal returns apply only at times of normal valuations. At low valuations, long-term return are far higher than normal. At high valuations, long-term returns are far lower than normal.

There has been academic research showing that valuations affect long-term return for 28 years now. I think it is a scandal that millions of middle-class investors are going to suffer busted retirements in days to come because of the demonstrably false claims made in the conventional retirement studies and calculators.

Rob

[cmadler]

I don't think that's an accurate description of a properly designed Monte Carlo simulation. Most MCS programs assume that returns will follow a normal distribution, but that doesn't mean that returns will be "normal". On the contrary, it's the simulation of varying returns that allows MCS to say that an investor is 98% likely to reach his/her goals; if you just assume a certain return, then you will or will not reach your goal. A similar, though not quite MCS approach is used by FIREcalc, which uses historical returns (rather than random values) by asset class to make projections.

[Sam]

Oh god. This is so wrong. The author obviously has zero grasp of numerical techniques. His characterization of Monte Carlo techniques is so primitive its like watching a caveman urinating on a car battery.

I apply non-Gaussian Monte Carlo techniques all day long to finance problems. I can assure you we don't keep billions of dollars in play without more accurate techniques.

If you want to play at home then get, at the very least, something like @risk which is an excel plugin that does decent albeit basic Monte Carlo. And rest assured it offers non-normal probability distributions so that you can escape the curse of normality.

[Monty Hothersall]

These simulation technologies work in the market until they don't. Just ask the Wall Street risk managers who used all types of probability distributions when evaluating mortgage related investments. They were thunderstruck and stunned.

Monty

[Sam]

Monty, you are off base. To someone who works in the business what you are saying sounds like "That long division stuff. Never did anyone no good. Them folks are using fancy talk.".

Many securities these days cannot be accurately priced without the use of techniques like Monte Carlo. We don't have closed form solutions for the vast majority of interesting options for example. No one really uses an untweaked Black Scholes as anything more than a convenient lowest common denominator for discussion these days.

One of the many things that went wrong with the mortgage backed securities, and more particularly the CDO and CDS instruments, was that essentially some fairly straightforward math was used to pull the wool over unsophisticated investors eyes. It was a bubble very much like the housing bubble itself. Everyone was buying and selling these things because everyone knew things just go up in price, right?

People were writing bonds with amazing sensitivity to housing price appreciation. Then using copula based models to claim they were triple-A bonds - the copula models were using CDS to calculate correlation. CDS data was available for about a decade, and was over a decade when housing prices were zooming up which meant the models spat out that default correlations were tiny. So the CDOs were worth a lot. There were plenty of people who knew this was dangerous, but, shockingly enough, there are lots of short-sighted greedy people playing in the markets. Heads of banks among them. And you'd be surprised at the lack of mathematical sophistication which lead people to actually believe some of the triple A nonsense.

The simulation techniques work just fine. They are simple and reliable. But, like long division, you put garbage in then you get garbage out. (And in fact, looking at your description, our in-house MC and MHMC based software would have no problems churning out simulations that match what you are talking about if we wanted to tweak parameters to match yours). If you don't want to use the techniques, then don't, but what you are claiming about them is just incorrect.

[Monty Hothersall]

It seems to me that a widespread miscalculation of risk among sophisticated buyers is a more accurate description of what occurred. After all, the players who contributed to the buying frenzy were not only the Wall Street risk managers, but also academics (including Nobel Laureates), financial journalists, foreign governments, highly-regulated foreign banks, Freddie and Fannie, and so on. This long list of participants, with very few dissenters, tells me that the many sophisticated parties who do their due diligence were just wrong.

But going back to retirement planning and online calculators, I refer people to Dr. David Nawrocki’s white paper in the Journal of Financial Planning: http://spwfe.fpanet.org:10005/public/Unclassified%20Records/FPA%20Journal%20November%202001%20-%20The%20Problems%20with%20Monte%20Carlo%20Simulation.pdf

Also, a great book is Dr. Nassim Taleb’s NY Times best-seller “The Black Swan.” http://www.fooledbyrandomness.com/

See

[Sam]

Monty,

You are confusing 2 things. Use of the Monte Carlo technique, and the creation of bad models which are evaluated using Monte Carlo techniques. As I pointed out in my previous reply, there were absolutely some bad models being used to 1) get a false sense of security and 2) pull the wool over people's eyes when it came to the soundness of certain investments. For example, the CDO/CDS fiasco was mathematical charlatanism to use enough greek letters to make people think the levels of risk were low. Monte Carlo simulation techniques were used as part of building the models but this was a case of feeding the techniques unrealistic data and not a problem with using such techniques. Monte Carlo is simply a mathematical technique that recalculates many different possible scenarios – but only within boundaries set by the user. I completely agree people mispriced the risk but that was an MC problem.

So for example, you can perfectly well use fat tail distributions to model extreme events - I know Nassim Taleb and have listned to many of his rants against the Gaussians - and I take objection to painting Monte Carlo techniques as bad just because some people use them with normal distributions and low correlations. I pointed to the @risk tool which uses MC. It lets you use a wide variety of distributions besides normal. Nawrocki is off base.

Do people build bad MC based models? Absolutely - I make my living taking their money. Can we build good MC based models? Yes - I make my living with them taking people's money. BTW, mostly we use pseudo-MC using low-discrepnecy sequences to speed convergence in fast trading models rather than pure MC but that's simply a runtime optimization to a first approximation.

[Sam]

A colleague pointed out to me that I should mention a couple of the books we have our college grads read to get them up to speed

Peter Jaekel's "monte carlo methods in finance" and Paul Glasserman's "monte carlo methods in financial engineering". Flipping thru the latter I see a chapter where he uses non-normal distrbutions.

[Monty Hothersall]

Sam,

My number one concern is the tens of millions of baby boomers using online retirement calculators for long-range financial planning. After completing a questionnaire that takes 5 or 10 minutes of their time and having the calculator run scenarios using MCS, they receive a sunny sky or cloudy sky picture with a number that reflects their likelihood of success. This method could wreak havoc on their retirement years and our economy.

People flock to these tools because they're quick and easy to use. And these people are sold on MCS being a panacea (e.g., they don't not worry about projecting things like their health-care premiums if they retire early because MCS shows that they are 98% likely to retire successfully). This is the way average Americans use this technology to plan for their future.

Thanks for your feedback and the book recommendations. I'll check them out.

Monty