Where We Agree
We totally agree with Dirk that:
- “determining how much you can safely spend this year [and making other personal financial decisions] requires a good model of the future”,
- A model is not a plan. A plan is an intention or decision about what to do based on the results from a model,
- “Of course, you can plan when outcomes are uncertain…”,
- A good model should not ignore sequence of return risk, and
- A stochastic model that uses expected rates of investment return for various asset classes but adjusts for investment risk is a better model for determining safe spending levels than a deterministic model that simply utilizes expected rates of return without adjusting for investment risk, all other things being equal.
While we agree that it is important to employ a “good model” of the future when developing a retirement plan, we are not convinced that it is absolutely necessary to use a model that employs simulations. Therefore, we will push back in this post on Dirk’s recommendation that “If you're using an online calculator, make sure it incorporates simulation”.
In our opinion, a “good model” for determining how much you can safely spend in the current year, or for making other personal financial decisions, is one that:
- Does a reasonably good job of forecasting future experience,
- Adequately addresses your retirement risks, and
- Helps you to make informed financial decisions with some degree of confidence.
We believe that such a model should:
- Consider your personal financial situation and goals,
- Be relatively transparent,
- Allow you to do “what-if” scenario testing,
- Reflect all current assets and expected amounts and timing of future assets• Reflect all expected amounts and timing of current and future expected expenses,
- Employ reasonable, or relatively conservative, assumptions with respect to:o Future investment returnso Future increases in expected expenses (including before and after the first death within a couple)o Your (and your spouse’s) future lifetime, ando Other relevant future experience
- Develop a total spending budget for the current year, not just an amount to be withdrawn from invested assets/accumulated savings
Actuarial Approach with Recommended Assumptions (Actuarial Budget Benchmark)
We believe that the generalized individual actuarial model (The Actuarial Balance Equation) used in our Actuarial Budget Calculators combined with our recommended assumptions (to develop what we call the Actuarial Budget Benchmark (ABB)) will satisfy most of the above criteria to be considered a “good model” for developing retirement plans and spending budgets.
Just like most pension plan actuarial valuations, our simple Actuarial Budget Calculators (ABCs) use deterministic, not stochastic investment return assumptions. When using these Excel workbooks, we recommend using assumptions consistent with assumptions used by insurance companies to price inflation-adjusted annuities to develop a market value (or market-priced) valuation of future spending liabilities. While equities and other risky investments may be expected to generate higher returns than such low-risk investments over an individual’s (or couple’s) lifetime planning period, such investments carry more investment risk. Following basic financial economics principles, the “risk-adjusted” expected returns on these more-risky investments should be approximately the same as expected returns on low-risk investments available in the market. Thus, in applying these principles, we believe that using these low-investment risk assumptions are reasonable for the purpose of determining how much you can safely spend, and the use of stochastic modeling is not an absolute requirement for a “good model”.
For a more detailed defense of our simple spreadsheet model, see our post of October 8, 2017, and for more discussion of the implications of using basic financial economic principles to determine the cost of retirement and thoughts from Dr. Moshe Milevsky on this subject, see our post of July 10, 2017.
Our Concerns with Monte Carlo (Stochastic) Modeling
Our concerns with Monte Carlo (stochastic) modeling may be summarized as follows:
- Stochastic Model results are highly dependent on assumptions made for expected returns and variances for various asset classes (which are frequently built into the model, based on historical experience)
- Stochastic Models are generally not transparent
- Stochastic Models may not reflect all assets and spending liabilities, and
- Stochastic Model results may not facilitate the making of good financial decisions (the primary purpose of using a model)
We discuss these concerns in more detail below.
Model Assumptions. Even if you are (or your financial advisor is) using a stochastic model that meets most of the above criteria to be considered to be a “good model,” you are pretty much forced into accepting the model designer’s “built-in” assumptions for expected future real investment returns and standard deviations for various classes of assets. Therefore, the reasonableness of the results of such a model will be very dependent on the reasonableness of these built-in assumptions about the future. You can accept these assumptions on faith, or you can question whether they are reasonable. And unlike assumptions that can be relatively easily gleaned from current annuity market pricings, it may not be so easy to tell how reasonable these assumptions are. You can, of course, compare them with historical results, but everyone knows that historical results don’t necessarily reflect current economic conditions and therefore are not necessarily great predictors of the future. Alternatively, you can compare the model results of using such assumptions with the model results of using annuity-based (low-investment risk) pricing assumptions. As discussed in our post of February 4, 2018, “Should Increasing Your Investment Risk Increase Your Current Spending Budget?,” we become concerned when we see models that suggest that you can increase current spending with little or no perceived additional risk by investing in riskier assets. This is an indication to us that the investment return assumptions for equity investments (or other risky investments) built into a model may be too aggressive in the current market environment.
We understand that Monte Carlo modeling is fairly standard practice among financial advisors. We also understand that most financial advisors make their living by increasing AUM (assets under management). And while most financial advisors undoubtedly believe they are using reasonable assumptions for future investment returns and variances for various classes of assets in their models, it may be prudent for you to try to independently assess how reasonable these assumptions might be. For example, you might want to ask your financial advisor how he or she has adjusted these assumptions for variations in the Shiller Cape 10 index (which at over 30 today would suggest lower than historical real expected rates of return in the future, all things being equal.)
Transparency and Reflection of All Assets and Spending Liabilities. In general, stochastic models tend to be somewhat “black boxy” in nature. Model results tend to be something like, “if you invest as follows, you will have a X% probability of being able to spend $Y per year in real dollars as long as you or your spouse is expected to live. When using such a model, you will need to determine if all future expected expenses (such as medical expenses that are expected to increase faster than inflation, unexpected expenses, long-term care expenses, other non-recurring expenses, etc.) have been adequately reflected in the model.
Using the Model to Make Decisions. We like Dirk’s weather forecast analogy in his argument for using a model that develops probabilities. If there is a 5% chance of rain today, you may decide not to bring an umbrella. A good model is supposed to help you make more informed decisions. Unfortunately, results of stochastic modeling tend to provide probabilities of success applicable to long periods of time and may not facilitate good short-term decision making. For example, a couple may become complacent after they are told that they have a 95% probability of being able to spend $X per year irrespective of actual future investment experience. Unlike the Actuarial Approach, which encourages annual valuations, periodic scenario testing and annual thinking about the safe amount to spend (and Rainy- Day Funds to establish or use to mitigate fluctuations in spending), the results of a Monte Carlo model can encourage more of a “set-it-and-forget-it” type of behavior that could result in either significant over-spending or under-spending as time goes by.
We are also concerned with Monte Carlo models that may lead users to invest too aggressively. For example, if overly aggressive future real rates of investment return (or overly conservative standard deviations) are assumed for equities, some individuals who wish to increase their current spending levels may be persuaded to increase their equity holdings beyond their tolerance for risk.
You Aren’t Required to Use Just One Model/Conclusion
There is no law that says that you have to use just one model in your retirement planning. You should feel free to look at the results produced by several models. We don’t think you should reject models just because they use deterministic assumptions if these models adequately address retirement risks. We also think it is perfectly reasonable to be somewhat skeptical of models developed by model designers who have a financial stake in the decisions you may make as a result of using their model. If you want to use a stochastic model, or your financial advisor uses a stochastic model, we encourage you to compare the results of that model with the results of our ABC model (developed by retired actuaries who have absolutely no financial stake in the decisions you make as a result of using our models) or other models. If the different models produce significantly different results, you should try to figure out why. This exercise will provide you with more “data points” to help you make better informed financial/spending decisions, which is, after all, the primary purpose of modeling the future.