Generally speaking, the tools that advisors have at their disposal to help clients plan for retirement are much better than they were a few short years ago. With the advent of faster processors and better programming, PCs and Web-based software can now perform complex calculations that would have only been possible on mainframe computers ten years ago. Better yet, most of those calculations can be hidden behind the scenes, so clients do not have to see the raw data unless they want to. This allows the advisor and the client to focus on what is important: the plan itself.

In spite of the advances we've made, retirement planning software is still evolving. Surprisingly, there is no one universal standard for performing retirement planning calculations within our industry. As a result, the advisory firm, or in some cases the individual advisor, must decide what methodologies and tools are the most appropriate to use when counseling clients.

Deterministic Models
Ten to 15 years ago, the deterministic or "straight line" model was the norm in the industry. This approach assumes that there will be a constant, steady pretax rate of return. In its simplest form, it also assumes a constant tax rate (or constant tax schedule) and a constant inflation rate. For example, you might create a model that shows a 65-year-old client retiring at the end of this year with $1 million in an IRA. If you assume that he will withdraw $50,000 per year beginning at age 66, that he will enjoy a 6% annual return and that he will encounter a 3% inflation rate with a tax rate of 28%, the client will have exhausted the account by age 84, at least according to a simple calculator on the FINRA Web site.

The beauty of the deterministic model is that it is simple to explain and simple to use. This makes it a good place to start a retirement planning discussion with clients. It also makes it a starting point for a do-it-yourself investor trying to figure out how much to save for retirement. There are, however, some severe limitations to this approach. The first is that advisors' projections for the average rate of return, inflation and taxes are likely to be wrong. The odds of a do-it-yourselfer getting the projections right are much lower.

Even if you do get all the projections right, however, the averages can be misleading. After all, if you put one arm in the freezer and the other in an oven, your average body temperature may be "normal," but you are not going to feel very comfortable. The same holds true for investment portfolios. Assuming you have invested this hypothetical IRA in a diversified portfolio, the returns for some years will be higher than average and for others they will be lower. If the IRA holder experiences the worst annual returns in the first year or two of retirement, he'll see his account depleted well before he's age 84. Or, in a more extreme example, if the portfolio declines 50% in year one, even if it returns 6% over the next ten years, with all the best returns projected for the final years, the account will still be depleted by the time the account holder reaches age 75.

Monte Carlo Simulations
In an effort to improve upon the deterministic model, many advisors now use Monte Carlo simulations. The basic underlying idea is that you have the computer simulate portfolio returns over time and then you run the simulation hundreds or thousands of times.

In a basic Monte Carlo simulation, you might start with a portfolio value (again, let's say $1 million), a tax rate, an inflation rate, an average return, a withdraw amount and a life expectancy (the entire period in which you'll be taking annual withdrawals). If you are looking at a 30-year retirement period, you'd have the computer select 30 different annual returns, withdraw $50,000 from the portfolio each year, and then see how long the portfolio lasts. If it retains some value over the simulated 30 years, that series of returns is deemed a success. If not, it is a failure. If you run this scenario 1,000 times, you can then determine what percentage of the trials succeeded.

Advisors are often puzzled when they enter the same data into two different Monte Carlo retirement programs and arrive at very different answers. The reason is that different software packages use different assumptions. One popular assumption is that returns will follow a "normal" pattern-that is, the distribution of the returns will resemble a bell-shaped curve. So if one program assumes there will be a normal distribution and another assumes there will be some variation of the normal distribution, the results will differ. In fact, even if both programs use a normal distribution, the results could differ because they may be using a different standard deviation of returns. All other things being equal, the higher the standard deviation, the less likely it is that the plan will succeed.

Monte Carlo And Risk
One of the criticisms leveled at advisors' Monte Carlo applications, particularly since the recent market meltdown, is that they underweight the likelihood of a significant market decline. In a 2009 article entitled "Déjà vu All Over Again," Paul Kaplan, Morningstar's vice president of quantitative research, looked at "standard models." These models use standard deviation as a measure of risk and they assume returns follow a normal, bell-shaped distribution. According to Kaplan, "If returns follow a normal distribution, the chance that a return would be more than three standard deviations below average would be a trivial 0.135%. Since January 1926, we have 996 months of stock market data; 0.135% of 996 is 1.34-that is, there should be only one or two occurrences of such an event."

But the actual monthly returns Kaplan found in the S&P 500 dating back to January 1926 told a different story. He discovered that the monthly returns of the S&P 500 have fallen below the three-standard-deviation average a remarkable ten times in the period under consideration. So the evidence suggests that a significant monthly decline is about eight times more likely than the standard models would lead you to believe.

More recently, in an article called "Nailing Down Risk," James Xiong says that a log-TLF (Truncated Lévy Flight ) model might offer results that more accurately match historical experience. This research looks promising, but I suspect that many advisors will have trouble understanding concepts like log-TLF models, let alone explaining them to clients.

The Money Tree Fat Tail Method
Money Tree Software has come up with an interesting alternative approach that allows advisors to run Monte Carlo simulations approximating the historical distributions of returns. The Money Tree method has at least one major advantage over the log-TLF method: Any advisor who has a basic understanding of Monte Carlo simulations should be able to grasp it in about five minutes.

Money Tree's program uses a two-stage process to show more likelihood of a major disaster-the fatter tail on the standard deviation chart. In step one, the program follows a standard Monte Carlo simulation approach. The program generates random returns using a normal distribution. The user defines the mean return. The standard deviation of returns can be defined by the user, or the program can select one. Ten thousand simulations are then run for the plan. Since this step is already familiar to most advisors, there is nothing new to learn here.

But then, in step two, the program resamples the lowest 2% of returns generated during step one and makes them even worse. What's the point? In a normal distribution, almost all of the worst returns will fall within the negative 2.5 to negative 3.0 standard deviation range. Through resampling, the program distributes those returns in a range of negative 2.5 to negative 5.0. The net effect is that you end up with many more negative returns exceeding three standard deviations.

How many more? According to Mark Snodgrass, president of Money Tree Software, those events exceeding three standard deviations using the Money Tree fat tail method are about 12.5 times more prevalent than those in the "standard model"-and about 1.5 times more prevalent than they are in Kaplan's monthly historical observations of the S&P 500.

Does the fact that Money Tree's fat tail model offers more extreme outcomes limit its validity and usefulness? Not necessarily. Remember, these are simulations, not facts. The whole idea is to try and model what might happen under a given set of circumstances. If we are trying to model an unknowable future, might it not make sense to err on the side of too many bad years as opposed to too few?

Although it's up to individual firms to decide whether the Money Tree fat tail method works for them, I find the method appealing for a number of reasons. First, as stated earlier, if you have a basic understanding of Monte Carlo simulations, you should be able to easily understand this methodology-and it's something that you should be able to explain to your clients easily as well. All you really have to tell the client is that standard Monte Carlo simulation understates the probability of a serious market downturn, and that's why you are using this new tool in addition to adjust for the somewhat higher probability that a bad market event will occur. If you or your client requires a detailed explanation of how the actual calculations are computed, Money Tree will be happy to supply you with documentation of all the particulars.

The other thing I really like about this fat tail option is that it is, in fact, simply that-an option. You don't have to use it if you do not want to. Money Tree makes the fat tail option available in both its Silver Planner and its more comprehensive TOTAL Planning Suite. If you fail to check the fat tails checkbox, Money Tree will run its standard Monte Carlo simulation routine. If you do check the box, it will run the fat tail calculation instead.

What is the impact of choosing the fat tail method as opposed to the standard method? Obviously, the fat tail method, with a more extreme number of bad events, will lower the plan's overall probability of success. The magnitude of the difference between the standard method and the fat tail method will depend on a number of factors, including the length of the plan. To give you some frame of reference, Snodgrass says that a 65-year-old retiring today with a 30- to 40-year life expectancy and a probability of success of 90% using the standard method might see that probability drop to 82% or so using the fat tail method.

By offering the fat tail option, Money Tree gives advisors the option of easily increasing the probability of some bad market events over time if the advisor deems it appropriate. In addition, Money Tree Silver Online, the application I used to test the fat tail method, offers advisors options to soften the impact that market declines will have with adjustments. For example, with the push of a button, you can view an alternative scenario that pushes retirement out a year or two. Or you can look at a scenario that marginally reduces variable expenses on the fly.

While it is likely that some purist mathematicians will find fault with the Money Tree model, it would seem that in this case, the ends justify the means. If you believe, as many do, that the bell-shaped curve accurately represents most market activity, and that the standard method's main shortcoming is that it underestimates the severity of the outliers, the Money Tree fat tail method elegantly balances the need for those severe outliers with the need for a methodology that advisors can understand and explain to their clients.

I doubt that this method is the final word in improving our retirement risk modeling, but it appears to be a step in the right direction.