Models of spending as retirement progresses

🇺🇸

 examines five models of retiree spending patterns from retirement until death. These models extend the Initial Spending estimates developed for the beginning of retirement so that they describe spending until death.

The Constant (Real) Spending model assumes that the real spending at the start of retirement continues unchanged until death. Although this approach is widely used, it does not correspond to a typical retiree's spending pattern. It usually (but not always) leads to an overestimate of the savings needed at retirement.

The Stages of Retirement model breaks up retirement into three or four age ranges, each of which has a different real spending pattern. Spending within each range can either be constant or varying. This model does a much better job of representing retirees' the diverse spending patterns.

Investment Returns Dependent models let retirement spending vary based on how total savings or investment returns change over time. These models are best combined with Dual Budget models of initial retirement spending. These models mimic retirees' tendency to vary their spending based on their net worth.

Flexible Spending models let you incorporate many individual spending categories into the total spending plan. Each spending category can have its own start and stop age, as well as spending amount. This model is excellent at representing large step changes in spending during retirement. Many retirement calculators will combine elements of this model with others already described. This combination is a very flexible framework for accurately modeling a diverse range of retiree spending patterns.

Life Cycle models calculate how a household should smooth their discretionary spending over their entire lifetime, not just in retirement. They give a detailed spending and savings plan to follow in pre-retirement years. Doing this will gives an optimized, real spending lifestyle for years both before and after retirement.

Retirement calculators offer all of these model types. For specific examples, see Retirement calculators and spending.

Introduction
This article is an in-depth expansion of the section Spending as Retirement Progresses from the introductory article on retirement spending models. Please also see that article for background information concerning approaches used to model Initial Retirement spending. These initial spending models supply the starting point for extrapolating spending across retirement.

As in other modeling articles of this series, this article treats retirement spending as the key independent variable in the retirement planning process. You first develop an estimate of your desired spending in retirement. From there, the planning process makes assumptions about you longevity, inflation and investment returns, and then estimates you future savings needed to maintain that retirement spending.

This article considers only the real component of retirement spending. Inflation and retirement spending covers how inflation adjustments are incorporated into the retirement planning process.

Retirement calculator models
There is a strong connection between retirement spending models and retirement planning calculators. Most software used for retirement planning, whether explicitly or implicitly, assumes some model of retirement spending. Its options reflect the choice of spending model. Unfortunately, some retirement calculators have many strengths but are needlessly weak in how they incorporate retirement expenses.

Retirement calculators and spending contains an extensive list of calculators which have been categorized in terms of the retirement spending model(s) they use. The tables in that article particularly highlight models of spending as retirement progresses. See that article for real-world examples of the spending models discussed here.

Retiree spending: key observations
Surveys of retirement spending reviewed and summarized data on retiree spending patterns collected by two ongoing, national U.S. surveys: the Bureau of Labor Statistics' Consumer Expenditure Survey, and the Health and Retirement Study. These surveys, as well as from the practical experiences reported by professional financial planners, show certain key facts:
 * The average retiree exhibits a slight drop in real spending at retirement, followed by a steady decline in real spending as they age into their late 70’s. This decline in real spending is voluntary and not a result of limited financial resources.
 * A substantial fraction of retirees (25% to 28% before the 2008 recession, rising to almost four out of ten ) enter retirement involuntarily. They exhibit a sharp drop in real spending at retirement. If involuntary retirement was health related, these retirees may subsequently exhibit a medical expense induced, real increase in total spending near their death.
 * A smaller percentage of retirees (roughly 12% ) exhibit an increase in real spending at retirement. This is often driven by a jump in travel or other leisure activities. After a certain time period these special activities end, and real spending drops to a lower level, often closer to that of "average retirees."

How well each spending model can represent these key retiree spending patterns is an important topic.

Constant (real) spending models
This is the simplest of all approaches used to model spending as retirement progresses. It assumes that your real spending at the time of retirement will continue unchanged until death. It does not ignore inflation, but as explained in the Introduction, treats it as a separate, adjustable variable in the retirement planning process. In terms of spending, this model states that spending for any year in retirement equals spending in the first year multiplied by an inflation adjustment.

The constant spending model is most often combined with a Replacement Rate  or  Single Budget model of initial retirement spending. A Dual Budget model of initial spending tends to be combined with a model that allows real spending to vary between the essential and preferred budgets as retirement progresses.

Comparison with retiree spending
Constant real spending models do not reflect the reality that the average retiree’s spending steadily drops during retirement. Many retirement planners have pointed this out as a major shortcoming. They also point out that these models overestimate the total savings needed at retirement.

A constant real spending model could be appropriate in if your retirement is involuntary and your financial resources are limited. This situation would force you to immediately drop down to an essentials only spending budget. If you were living at your minimum acceptable level, you would expect your nominal spending to grow at around the rate of inflation (constant real spending) as you are forced to pay ever increasing market prices for essential goods.

However, this would also motivate you to substitute, when possible, less expensive items for your essential budget spending needs. This Substitution Effect causes spending to somewhat lag behind constant real spending.

Relationship to withdrawal studies
Some of the earliest research on Safe Withdrawal Rates used constant real withdrawals in their calculations. Constant real withdrawals is a close proxy for constant real spending that avoid complication from taxes. The constant withdrawals constraint let the researchers focus on their main issue of interest: how Sequence of Returns Risk impacts retiree savings survival. But because so much subsequent research uses this constant withdrawals constraint, some have mistakenly concluded that this is a realistic approximation to how retirees withdraw and spend money. As noted, the average retiree steadily reduces real spending as they age.

Use in retirement calculators
Free online retirement calculators most commonly use a Constant spending model. This is especially true for calculators that are deterministic, less so for those that use Monte Carlo or Historical Returns approaches.

Some free online retirement calculators let you to choose models other than constant real spending. But these will often include a constant spending model, for comparison with simpler calculators.

Retirement calculators that you pay for are much less likely to rely solely on a constant spending model. But they often include this model for comparison.

Although Life Cycle spending models sometimes produce a constant real spending in retirement, they do this from an entirely different Economics Perspective. For more, see Life Cycle models below.

See Retirement calculators and spending for examples of retirement calculators that incorporate a constant real spending model.

Stages of retirement models
Studies have shown that a typical retirement splits into three to four Stages or Phases. Within each stage, retirees tend to exhibit similar patterns of physical activity and spending. As a specific example, consider the following four Stage proposal by Robert Carlson:

Not all retirees pass through Carlson's Stage 1 of increased real spending early in retirement. The average retiree does not, but some retirees do. Typically retirees pass though Stages 2, 3 and 4. (These are the only stages considered in most 3 Stage models.) Some retirees in very poor health might have increased real spending in Stage 4, but this depends on whether their insurance is able to cover the associated higher medical costs.

Many authors besides Carlson have chosen age 75 as one of their stage boundaries. There are probably several reasons for this. For retirees in reasonable health, this age is approximately when natural aging processes lead to a slower pace of living (this is Michael Stein's Slow-Go stage ). It is also possible that they chose this age to match the BLS Consumer Expenditure tables, which break down their retiree data into age groupings of 65-74 and 75+.

Other authors believe that variations in retiree life experiences and spending prevent fixing ages to the stage boundaries. An common example is involuntary retirement. For people entering retirement involuntarily with insufficient financial resources, their spending could very quickly drop down to a low, Stage 3-like level even though they could be in their early 60's. Keeping this variation in mind, a good Stages of Retirement model should let you adjust boundary ages.

Stages having constant real spending
The very simplest Stages of Retirement model assume that the real spending within each Stage is constant. All changes in real spending occur between the stages. For the average retiree spending drops at these boundaries. If it uses a budgeting approach to estimate the real spending within each stage, you can think of these as Step Change Budgets. But a changing (often dropping) replacement rate could just as easily be used to model each successive stage.

This type of model can be made more realistic by further breaking down spending into budget categories within each stage. In his Age Banding model, Somnath Basu incorporates a separate budget for each of four broad expense categories: taxes, basic living, healthcare and leisure. At each of the stage boundaries his model allows a step change in real spending, either upwards, downwards or no change. For example, basic living expenses could be given step drops at each stage boundary, but healthcare given step increases at each boundary. Further, his model allows for a different inflation rate to be applied to each spending category within each stage.

Stages having variable real spending
Studies have also shown that retirees steadily reduce their spending as they age, rather than as a series of sharp drops. Following on from this, several retirement planners have suggested that Stages of Retirement models can be made even more realistic by allowing real spending to vary gradually within one or more of the stages.

Tacchino & Saltzman. A very widely cited paper that suggested incorporating a steady spending drop into retirement planning calculations was published in 1999 by Kenn Tacchino and Cynthia Satlzman. Arguing from the U.S. Consumer Expenditure survey data, they concluded:

"... all available data and research indicate that there is a gradual reduction in spending starting shortly after retirement. Spending decreases are natural, voluntary and acceptable, and should be reflected in the client's accumulation model. The downward adjustment in spending by age 75 is approximately 20 percent of the initial spending levels during retirement that starts at age 65."

This 20% reduction (2.2% annually) is an average across all retiree income levels. They also show that real spending drops for every income range within the BLS survey data, although the percentage decrease varies in a somewhat random fashion.

Tacchino and Saltzman did not particularly emphasize using a continual (annual) real spending reduction until age 75. They spent more time presenting ways in which stage models having constant spending could be modified to approximate the steady spending decreases. But they did state that a retirement calculation incorporating a real spending drop each year would lead to a better estimate for the total savings needed at retirement.

Ty Bernicke's ''Reality Retirement Planning. Ty Bernicke published a widely cited paper Reality Retirement Planning: A New Paradigm for an Old Science in 2005. In it he emphasized using a steady, real spending reduction in each successive year of retirement until age 75. Real spending was assumed to remain constant in stages after age 75. His arguments were (again) based on data from the BLS Consumer Expenditure Surveys.

In his paper Bernicke recommends using a 50% spending drop between the 55-59 age group and age 75. A sample calculation presented used an annual real spending drop of 3.7%. This is certainly too large. Referring back to Table 1 in this paper, it can be seen that the raw Total Consumer Expenditure dollar amounts from the 2002 CE Survey were used. These dollar amounts included savings dollars (Social Security payroll taxes and personal retirement savings) which should have been subtracted out before estimating the percentage spending drop. Also, there was no correction for the different average household sizes between the age groups (i.e. 2.1 persons for age 55-59 versus 1.5 persons for age 75+). Incorporating these corrections would have reduced the predicted spending drop by age 75.

Recommended Annual Spending Drop. Bernicke's annual percentage drop of 3.7% until age 75 is too large. The equivalent annual spending drop from Tacchino and Saltzman, 2.2%, is much lower and probably closer to the truth. Fisher and others have analyzed the same Consumer Expenditure Survey data and recommended using a 1% annual real spending drop until the late 70's.

The BLS Consumer Expenditure Survey does not supply the best database for analyzing drops in real spending as retirement progresses. This is because the BLS survey does not follow the same households in each successive year. Changes in spending patterns with aging can only be derived by comparing shifting averages in large populations of households. It would be much better to perform retiree spending studies using the Health and Retirement Study survey database. In this biennial survey the same households are followed from age 51-56 until death. Unfortunately there appear to be no no such studies of retiree spending patterns.

William Bengen’s Prosperous Retirement. Published in 2001, Bengen’s Prosperous Retirement model< and William Bengen’s Prosperous Retirement model is aimed at retirees whose spending early in retirement is higher than average. This is in contrast to the previous two models that targeted a more "average" retiree. This model has three Phases: Active, Transition and Passive. In the Active Phase (until age 75), the model assumes a constant real spending. In the Transition Phase the model has a strongly declining real spending each year until age 85. (The example in the paper used 4% below inflation annually.) In the Passive Phase the model has a weakly declining real spending each year until death. (The example in the paper used 2% below inflation annually, which gave a slightly increasing nominal spending each year.)

Comparison with retiree spending
Using three (or four) Stages of Retirement, each of which can have a different real spending, models retiree spending patterns much better. All three of the basic patterns discussed earlier in the Key Observations section can be mimicked. And if the first one or two stages allow a gradual annual spending drop rather than just a constant real spending, then an even better match to actual retiree spending is possible.

Use in retirement calculators
Calculators with a Stages of Retirement spending model are much less common than those with constant real spending model. Only a few of the calculators in the Retirement calculators and spending article explicitly include this model. Some of these include Ty Bernicke’s Reality Retirement Planning approach.

Although the documentation for only one of calculators explains how (the Flexible Retirement Planner), it is usually possible to set up a Stages of Retirement model with any calculator having a Flexible Spending capability. Each spending Stage (start age, stop age and constant real spending amount) is modeled using one of the custom spending inputs. To make this calculate properly, you may need to set the initial spending budget to $0.

Investment returns-dependent models
It unsurprising that retirees would cut back on their spending during times when they do not feel financially secure. For someone who is retired and who has stock market investments, stock bear markets would motivate them to reduce spending. Conversely, if they feel financially secure (for example, during a stock bull market), they would probably increase their real spending. These periods of increased and decreased spending can be modeled by Withdrawal Methods that depend on on either the previous year’s total savings or its investment return. An extensive summary of these methods is given on the Variable Withdrawals in Retirement (archived page from the Bob's Financial Website).

A Dual Budget model is a natural complement to variable withdrawal methods. The Essential budget spending level would set a lower limit on spending as retirement progressed. The Preferred budget spending level would set a corresponding upper limit on spending. These constraints combine with those of the withdrawal model, providing a better approximation of the retirement savings needed. But since the Dual budget is an initial spending model, some method of extrapolating the spending across the retiree’s lifetime must be used.

It seems reasonable to extrapolate the Essential budget component of a Dual budget model using constant real spending. If you were living at your minimum acceptable level, you would expect your nominal to grow at around the rate of inflation (constant real spending) as you are forced to pay ever increasing market prices for essential goods.

Extrapolating the Preferred budget component using a constant spending model is not reasonable, for all the reasons previously discussed. Instead, the Preferred budget should be extrapolated using either a Stages of Retirement model or a  Flexible Spending model. Both allow a more realistic description of how actual retirees alter their real spending as they age.

Comparison with retiree spending
By itself an Investment Returns Dependent spending model cannot adequately describe the key observations on retiree spending discussed in the Introduction. It cannot match the steady decline in real spending as retirees age. But it does supply an element of reality that other models lack: the ability to reflect how retirees alter their spending in response to changes in their net worth. A combination of these two spending tendencies would be very beneficial.

A useful analogy is to think of waves on the ocean as the tide goes out. The waves would represent the investment return-dependent spending and the tide going out would represent the steady decline in spending as the retiree ages. The waves raising and lower the ocean surface level, but do not change the more powerful effect of the lowering tide. But without including both effects, the water level at any instant cannot be measured accurately. Similarly, retiree spending in any year is better described by combining these spending models.

Use in retirement calculators
Calculators having an Investment Returns Dependent spending model are much less common than those using a constant real spending model. However, referring to the Calculators and Spending article shows that more calculators incorporate some variation of this model than those explicitly incorporating a Retirement Stages model.

Flexible spending models
A Flexible Spending model allows a total spending plan to incorporate multiple individual spending categories. Usually there is no need to enter every conceivable budget category as a separate spending item. Rather, these models usually let you first enter a base spending amount from another model (Constant Spending, Stages of Retirement, and so on). Then you can enter any spending that still is not properly represented as a separate, individual spending item. Typically the only information required is the starting age, ending age and spending amount for each category. This type of modeling flexibility is particularly important if you anticiapte one or more large expenses that occur only over a limited age range. The following situations are examples of where a Flexible Spending model would allow a closer match between actual and modeled spending:
 * You plan to purchase expensive, individual medical insurance until they become eligible for a less expensive Medicare + Medigap combination at age 65.
 * You want to explore how possibly needing to pay for several years of her mother’s long-term medical care might affect how long her own savings would last.
 * You are close to retirement with mortgage payments lasting another 13 years, and want to incorporate this into your proposed retirement spending plan.
 * You are 10 years from retirement and want to accumulate enough savings to enable you to buy a winter home in Arizona several years after retiring at age 65. You would keep your current home until age 75, then sell it.

A Flexible Spending model can incorporate these types of category-specific, step changes in real spending. At the same time other spending categories, for example basic living expenses, could be described using other models that provide a better match to fairly smooth spending changes.

Often you can use a Flexible Spending model to mimic the spending pattern of a Stages of Retirement spending model. You do this using one flexible spending category to contain all the spending within one of the retirement stages. Most Flexible Spending models assume constant real spending within each category, so the resulting Stages model would also have constant spending. But there are a few calculators that allow variable real spending within a flexible spending category. If a retirement calculator provided a very large number of separate flexible spending entries, then you could even create a separate spending entry for each year of retirement. This would provide enough planning flexibility to cover almost any conceivable situation.

Robert Carlson has discussed the advantages to being able to accurately model large, periodic expenses in retirement. Examples of such periodic expenses are car purchases and major home appliance replacements. The simple approach is to use a "sinking fund", where the total cost of the item and its expected lifetime are used to calculate an effective annual cost.

But a more precise treatment would use a specially designed flexible spending category. This special category would accept the following data: the expense year, total cost, and expected lifetime. The spending model would then calculate all subsequent years in which a repeated purchase of that item would be anticipated, as well as the inflated cost in each of those years. This approach gives a more realistic projection of variable retirement spending. This level of modeling detail would let you incorporate tax bracket management into your retirement planning.

Comparison with retiree spending
Provided there are enough flexible spending categories/entries available, this model can do an excellent job of representing the diversity of retiree spending patterns discussed earlier. But if it offers only a small number of categories or entries available, it cannot model a steady decline in retiree real spending through their late 70’s or early 80’s. At best a Stages of Retirement model equivalent with constant real spending is possible. But even in such a limiting situation the final spending estimate would be better than that from a constant real spending model.

Use in retirement calculators
Calculators with a Flexible Spending model are much less common than those with a constant real spending model. However, the Retirement calculators and spending article offers a handful of free calculators that do include variations of this model. Most paid-for retirement calculators have this spending model.

Life Cycle models
Life Cycle models calculate how a household should smooth their discretionary spending over their entire lifetime, not just in retirement. Because they typically make a distinction between essential spending and discretionary spending, they are conceptually connected to Dual Budget models of retirement spending. Life Cycle models provide a detailed spending and savings plan to followin in the pre-retirement years. Doing this gives an optimized, real spending lifestyle for both the pre-retirement as well as the post-retirement years.

Spending predictions from Life Cycle models may sometimes be similar to spending in Constant Real Spending models, because in a few cases both will recommend a constant real spending in retirement. But further investigation reveals a deep difference in their fundamental approach. A Life Cycle model predicts how a household should smooth their discretionary spending over their entire lifetime, not just in retirement. Utility theory predicts that people would be happiest if they neither over-spend nor under-spend each year as they age, whether retired or not. Starving now to ensure a comfortable retirement makes no more sense than splurging now and starving in retirement.

It is difficult to estimate how much you would need to save each year before retirement so that you could maintain a comparable lifestyle each year after retirement, especially if you have to account for uncertain future investment return, life span, and health care expenses. Calculators offering the Life Cycle approach overcome these difficulties using Dynamic programming. The following table summarizes the main differences between the Life Cycle model and other spending models:

A Life Cycle model sometimes predicts a constant real spending during retirement, for example when longevity risk aversion is ignored and a fixed (deterministic) retirement lifespan is assumed. But even in this restrictive case you should not think of that spending level as an unchanging quantity to be spent each year regardless of how retirement investments perform. Rather, the Life Cycle model calculation should be run every year, so that its optimal spending prediction adjusts for investment performance. In this sense a Life Cycle model is different from a constant spending or withdrawal model set up using a Safe Withdrawal Rates approach.

When a Life Cycle model is run using more realistic assumptions (for example, a probabilistic retiree lifespan combined with longevity risk aversion), it predicts that the retiree should reduce their real spending as they age. The lower the retiree's longevity risk aversion (that is, their fear of outliving their savings), the more they will pull real spending forward into the earlier years of retirement. Of course, this retiree lived an unexpectedly long time, they would need to significantly reduce their real spending late in retirement. But stating that the retiree's longevity risk aversion is low means that they are not particularly bothered by this possibility.

Life Cycle models of retirement spending can readily incorporate pension-like sources of retirement income that canot be out-lived, such as U.S. Social Security benefits,  Defined Benefit pensions, and  immediate fixed annuity payments (SPIA's). When such sources of income are included, it is observed that the optimal Portfolio Withdrawal Rate (PWR) from the retiree's personal savings depends on their longevity risk avesion and on their level of pension-like income. The larger the amount of the preexisting pension income, the preater the PWR is and thus the greater is the optimal consumption rate.

Comparison with retiree spending
If they use a fixed retiree lifespan and no longevity risk aversion, Life Cycle models predict a constant (real) discretionary spending target throughout retirement. This is clearly different from the average retiree spending patterns discussed earlier. But when Life Cycle models use more realistic assumptions (for example, longevity risk aversion and probabilistic lifespans), they correctly predict that real spending decreases with age.

Economic theory predicts that people should attempt to maintain a constant consumption rather than a constant spending. For example, spending on food tends to drop sharply after retirement. However, careful studies have shown no change in food consumption; that is the amount and quality of the food they eat remains the same. Because retirees choose to spend more time preparing their food, they therefore need to spend less money. Similar trade-offs doubtlessly affect other budget categories, especially as retirees make use of the ubiquitous "Senior Discount." Ideally, a Life Cycle model should recognize this difference between smoothed lifetime consumption (the actual goal) and smoothed lifetime spending (a simple proxy).

As for retirees who plan on sharply increasing real spending after retirement for extensive travel and other expensive leisure activities, the Life Cycle model simply says: ''Why wait? Enjoy some of these activities while you’re young!'' The Life Cycle model will help you determine how much of the leisure activity expense can be pulled ahead before retirement.

Use in retirement calculators
Calculators include Life Cycle models less frequently than other spending models. This is unsurprising, given the complex programming required to implement the model. Some retirement calculators partially implement the Life Cycle concepts, which simplifies their programming. Retirement calculators and spending shows only a few calculators that either fully or partially incorporate the Life Cycle approach. ESPlanner and ESPlannerPLUS implements this fully. Optimal Retirement Planner, being a free program, not surprisingly has only a partial implementation of these characteristics. But it is still a very powerful calculator.