If there’s one thing clients love to fret about, it’s whether their nest eggs will last for the rest of their lives. Of course, as most advisors will say, setting a specific figure as a retirement target is an imprecise process because the lasting power of assets depends on so many random assumptions and events.

To track these variables more accurately, many advisors have been using Monte Carlo simulations available on various Web sites and software packages. Although a useful exercise, the results of these calculations are all over the map, potentially creating as many problems as they purport to solve, says Moshe Milevsky, a professor of finance at York University’s Schulich School of Business.

Earlier this year, he set out to investigate the extent to which these simulators, particularly those widely available to the public, would provide different answers to the “will my money last?” question.

In his simulations, he looked at a 55-year-old single male in average health who was contemplating early retirement with a nest egg of $500,000 and no other source of retirement income, and who would like to withdraw $25,000 annually, adjusted for inflation each year. The subject’s entire nest egg was assumed to be invested and rebalanced in a portfolio of diversified equities that was projected to earn an average of 7% after inflation each year, equivalent to a geometric average of 5% with a standard deviation of 20%.

In his analysis, Milevsky found the sustainability of income for this individual to be 75% — meaning, according to his figures, that the probability of ruin is actually 25%. Yet his preliminary analysis of several calculators revealed a wide range of results for this 55-year-old retiree. In fact, the lowest sustainability number was 48% and the highest was 88%.

And, were these simulations to be used for a couple, the results would be even more diverse, he warns.

Although Monte Carlo generators are based on random returns, results should only differ by a statistical margin of error, unless assumptions differ.

Milevsky traced the range of results to a number of factors, including whether the sponsor uses:

> the S&P 500 or the Russell 3000 as a proxy, and their differing views of the dynamics of equity markets and future stock prices;

> retirement horizons based on population or annuity mortality tables, U.S. Social Security Ad-ministration projections, or arbitrary 25 or 35 years of retirement;

> sophisticated models for calculating the yield curve or assuming a basic “random walk” process similar to the one used for equity prices;

n the CPI as a means of calculating cost of living in retirement, given that retirees’ annual expenditures may only be loosely correlated. Some programs explicitly model the evolution of inflation, then increase withdrawal rates by this amount each year; others implicitly model the real after-inflation evolution of portfolio returns;

> an adequate number of scenarios. Running the 100,000 scenarios needed to provide a minimal margin of error is not feasible for most real-time engines, but running a few hundred can be inadequate.

The retirement income planning industry cannot afford to ignore the divergence of the Monte Carlo results, Milevsky says. If it does, outsiders (including regulators) unfamiliar with the subtleties of simulations may incorrectly interpret the dispersion of results as a flaw in the methodology instead of being a powerful technique for measuring and explaining risk.

One solution may be to develop criteria for which all Monte Carlo retirement income programs would generate the same results, within a statistically tolerable margin of error, he says.

Advisors using such simulations need to familiarize themselves with the underlying assumptions and be prepared to defend them. IE