Retirement Spending Guardrails: How the Risk-Based Approach Works

This analysis draws on Derek Tharp and Justin Fitzpatrick’s work at Kitces.com, where their research on risk-based guardrails is published. We’d recommend reading it in full.

Why a Fixed Spending Number Falls Short

Before retirement, the test of living within your means is simple: spend less than you earn. Retirement removes the paycheck that defined the means. What replaces it is a portfolio of uncertain future value, income sources that start at different ages, and a time horizon nobody knows. The question becomes how much a household can spend, and give, without spending through money it will need later.

The classic answer is a fixed number. William Bengen’s 1994 study in the Journal of Financial Planning tested rolling historical periods for portfolios with between half and three-quarters in stocks. A retiree could withdraw about 4% of the starting portfolio, raise the dollar amount with inflation every year, and still last 30 years in every period he examined, including the worst. The Trinity study, published in 1998 by three professors at Trinity University, reached a similar conclusion. The 4% rule was built around the worst case on purpose: the retiree of the mid-1960s, who met poor returns and high inflation at the same time, barely made it.

That is also its weakness. A number set by the worst case is, in most other cases, far too low. In Michael Kitces’s analysis of rolling 30-year periods back to the 1870s, a retiree following the 4% rule with a portfolio of 60% stocks and 40% bonds finished with more than the starting balance in about 90% of periods. The median retiree finished with almost 2.8 times the starting balance in nominal dollars, and even after adjusting for inflation, the retiree finished with more purchasing power than at the start about 60% of the time. For that median retiree, the rule did not so much protect a retirement as postpone much of it. Money that could have paid for the family trips, a grandchild’s tuition, or the causes the retiree cared about went unspent and ungiven, and much of it became an unplanned inheritance that arrived decades after it could have helped. We’ve written about why a rule of thumb is the wrong place to start in how much you need to retire.

A single number has to choose which future to be wrong about: too cautious for the ordinary decades, or too exposed for the bad one. A guardrail sidesteps the choice by agreeing in advance on how the number will change. Jonathan Guyton, whose work with William Klinger brought guardrails into the mainstream, also proposed writing those rules into a withdrawal policy statement, the retirement income counterpart to an investment policy statement, so that the decision about what to do in a bad market is made while nobody is frightened. We lean toward the same idea: the guardrails are in the plan document before they are needed.

The First Generation: Withdrawal-Rate Guardrails

Guyton and Klinger’s 2006 paper in the Journal of Financial Planning turned the idea into rules simple enough to run by hand. Pick an initial withdrawal rate. If the current withdrawal rate (this year’s spending divided by today’s portfolio) falls 20% below the starting rate, raise spending 10%. If it rises 20% above the starting rate, cut spending 10%. Skip the inflation raise after a year in which the portfolio lost money, if the withdrawal rate is already above where it started, and stop making guardrail cuts in the last 15 years of the plan.

In dollars, the rules are easy to explain, which is a large part of their appeal. A retiree with $1,000,000 who starts at a 5% withdrawal rate spends $50,000 a year. The raise comes if the portfolio grows to $1,250,000, where $50,000 is a 4% withdrawal rate, and lifts spending to $55,000. The cut comes if the portfolio falls to about $833,000, where $50,000 is 6%, and drops spending to $45,000. Instead of watching the balance tick down and wondering when to worry, the retiree knows where the line is and what would change there. That clarity was a real step forward, and nearly every guardrails approach since, ours included, keeps it.

The trouble is the yardstick. A withdrawal rate is a ratio of two numbers on a single date, and it knows nothing about the rest of the plan.

Where Withdrawal Rates Mislead

The same rate carries different risk at different ages. A 6% withdrawal rate is a strain at 65 and modest at 90, because the money has fewer years left to cover. Tharp and Fitzpatrick showed how far this can go. For a couple who began at 65 with a 5% withdrawal rate from a portfolio of 60% stocks and 40% bonds, a 6.1% rate at 75 could be more sustainable than the 5% they started with. A static 6% guardrail would call for a cut at the very moment a raise might be the more prudent move. Guyton and Klinger saw the issue, which is why their rules stop cuts in the final 15 years, but a single switch at year 15 is a rough stand-in for risk that changes every year.

Income that starts later makes early withdrawals look dangerous. Many households retire in their early or mid 60s and claim Social Security later, sometimes at 70. Until benefits begin, the portfolio carries the whole budget, so the withdrawal rate starts high; once benefits arrive, it drops sharply. Pensions, rental income, a few years of part-time work, and a planned home sale do the same thing. Tharp and Fitzpatrick named the resulting shape the retirement distribution hatchet: a tall blade of heavy withdrawals early, then a long, low handle. A withdrawal-rate guardrail reads the blade as danger, when the plan was built that way on purpose. Whether delaying Social Security is right for a particular household is its own decision, which we walk through in our article on when to claim Social Security.

Spending itself changes with age. Most retirees do not spend the same inflation-adjusted amount every year. David Blanchett’s 2014 research found that real spending tends to fall gently early in retirement, faster through the middle years, and then levels off and can tick back up near the end, a pattern he called the retirement spending smile. His 2026 follow-up found that for the typical retiree the late upturn is missing, more of a smirk than a smile. In his modeled example, planning for a decline of either shape supported an initial portfolio withdrawal rate about 20% higher than planning for flat real spending: 6.2% for the smile and 6.4% for the smirk, against 5.2%. A withdrawal-rate guardrail cannot tell a planned decline from an unplanned one: when the rate falls, it cannot know whether markets did well or the plan simply reached a year when spending was meant to be lower.

Our default follows the household rather than a formula. When a budget is all essentials, we lean toward modeling flat real spending, because there is little that naturally falls away with age. When there is a discretionary layer, we lean toward modeling real spending that declines through the slower middle and later years of retirement, with an uptick in the last few years for medical costs, the more cautious of Blanchett’s two shapes. Either way, the shape is a starting assumption we revisit with the household, not a rule.

Risk-Based Guardrails: Measuring the Whole Plan

Tharp’s 2021 research, which introduced what he called probability-of-success-driven guardrails, proposed keeping the guardrails and replacing the yardstick. Instead of a withdrawal rate, the trigger is the risk of the whole plan, measured by running the household’s actual plan, with its Social Security dates, its pension, its spending path, its legacy goal, and its planning horizon, through a large set of market scenarios. The plan then works backward from that risk to dollars: the portfolio value at which spending would rise, the value at which it would fall, and the new spending level at each. The client sees dollars. The modeling, with all its assumptions, stays underneath.

Because the risk measure reads the entire plan, it handles what withdrawal rates cannot. It knows Social Security starts in year five. It knows about the roof replacement in year eight and the gift toward a grandchild’s wedding in year twelve. It knows the household is a year older at every review, which by itself lowers the risk of a given spending level. And it knows what the household wants to leave behind. We lean toward setting that two ways at once: an explicit legacy goal where there is one, and the risk setting itself. Tharp’s research found that when a plan keeps adjusting to hold a chosen risk level, the median, best, and worst spending paths look surprisingly similar whether that level is cautious or aggressive. What the chosen level mostly changes is how much is left at the end, though a household that starts higher has further to fall to the same low point. So the dial is, in large part, a choice between spending and giving now and leaving more later.

How long and for whom. Two more inputs shape the risk number. For plan length, we lean toward setting an age the money has to last to, adjusted for health and family history, and revisiting it as health changes. For couples, we model the first death every time, because the survivor typically loses one of the two Social Security checks and, usually from the year after the death, files under single brackets, while much of the household’s spending continues. We cover that tax shift in the widow’s tax penalty.

Where the scenarios come from. We lean toward running the analysis on historical returns rather than Monte Carlo simulation, which generates thousands of random market paths from assumed average returns and volatility. Fitzpatrick and Tharp’s 2022 comparison, which built its Monte Carlo runs from the same 150 years of returns, found that at the risk levels most plans use, Monte Carlo estimated about 5% to 10% more sustainable income than the historical record supported. Their explanation is that Monte Carlo treats each year as independent of the last, while real markets have tended to string bad years together into multi-year slumps, the kind that strain a retirement plan most. The counterargument deserves a fair hearing. U.S. market history is a single path, drawn from a period that includes America’s rise to global economic power, and nothing guarantees the next 30 years will resemble any 30 years before. At the far tails, Monte Carlo imagines worse sequences than history recorded. We treat history as a record of what has happened, not a promise of what will. Inflation belongs in the same conversation: the 1965 retiree was hurt by rising prices as well as by weak markets, and we lean toward weighting inflation risk more heavily than many plans do, rather than assuming it will settle at a steady target. Reasonable people disagree, pointing to the Federal Reserve’s 2% target and to long stretches when inflation stayed low.

Two Risks, Not One: Overspending and Underspending

Fitzpatrick’s 2024 research offers the vocabulary we use with clients. Every spending level carries two risks. Overspending means that at some point, spending will need to come down. Underspending means living more frugally than the household’s resources allow, which shows up as experiences not had, gifts not made while the giver could see them land, and an estate that grows well past what the household meant to leave.

The familiar probability-of-success score counts only the first. It labels every scenario where the money lasted a success, including the ones where the household could have spent or given far more. As Fitzpatrick puts it, a 100% probability of success is a 100% probability of underspending. The balance point, where overspending and underspending are equally likely, is the plan’s best guess at sustainable spending. In success language it is a 50% probability of success, which sounds alarming and is simply the median. Many households, and most of the plans we build, start below the best guess, in the underspending zone, to hold a buffer. How far below is the real decision, and it cuts in both directions.

The Settings: Where Spending Starts and What Moves It

These are the settings we lean toward starting from, adjusted plan by plan where a household needs something different.

Where spending starts. Spending begins at a level with about a 20% risk of overspending and an 80% risk of underspending. In plain terms, if that spending level were held steady through every historical scenario, about four in five would have lasted with money to spare, and about one in five would have run short. Starting there puts the plan in the underspending zone, with a cushion, and it means the household should expect more good news than bad.

When spending rises. The raise comes when the risk of underspending reaches 100%, meaning no historical scenario would have run short at the current spending level. Spending then rises all the way back to the starting risk level. Because the move is a full one, a raise is large enough to plan around: a trip, a project, a gift.

When spending falls. The cut comes when the risk of overspending reaches 75%, meaning three in four historical scenarios say the current spending is too high. That can sound late. Fitzpatrick’s framing explains why it is not. At a 50% risk of overspending the plan sits at its best guess, and below that it is still more likely to be underspending than overspending, so a cut there would be an overreaction, even though a 30% or 40% risk of overspending (a 70% or 60% probability of success) is often where a plan starts to look worrying. Waiting until 75% means the plan is fairly confident it is overspending before anything changes. In Tharp and Fitzpatrick’s machine-learning test of many possible settings, triggers like these fared best, though the cuts in that test were larger than the small ones we describe here. When the cut does come, it is small: spending is trimmed just enough to move the plan a tenth of the way back toward its starting risk level, often a few percent of spending.

Why small cuts. Small cuts are easier to live with, and the research on them is encouraging, with one important difference from our approach. Tharp’s 2017 study used annual market data back to 1871 and a portfolio of 60% stocks and 40% short-term bonds over 30-year periods. A retiree who trimmed real spending by 3% in each year the stock market fell could historically have started at a 4.56% withdrawal rate, against 4.08% for a retiree who never adjusted. Cutting 10% for one to three years after each down year lifted the starting rate only to between 4.33% and 4.40%. Those small cuts were permanent, and the study credits that permanence for much of their effect: over a long retirement, a string of small permanent cuts adds up to more than a large temporary one. Ours lean temporary, in a specific sense: a cut stays in place until the plan recovers enough to earn a raise, and the raise resets spending all the way to the starting risk level at that point. Temporary can still mean several years in a hard decade, and a long decline can bring a second small cut, or a third. Fitzpatrick’s advice to planners is the way we explain it to clients: it is better to make several adjustments than to adjust too much at once, and every adjustment can be revisited.

When the cut happens. We lean toward the waiting period Matthew Jarvis described in his 2023 article on running guardrails in practice: once the portfolio falls below the lower guardrail, wait two calendar quarters before cutting. A sharp drop can reverse within months, as it did in 2020, and a cut made in the middle of it would be undone almost as soon as it was made. We also tell clients when they are getting close to the lower guardrail, well before any cut, so the conversation about where they would trim happens calmly. We typically update retired clients’ plans about twice a year.

Customized. These are starting points, not a formula. A household with little room to adjust might start further into the underspending zone, or forgo raises entirely and let good years build a larger reserve, an option Tharp describes. A household with a large discretionary budget might accept more risk of a cut in exchange for more spending and giving now.

What the Stress Tests Show

Rules are easy to like in good markets. As Tharp and Fitzpatrick point out, the long bull market after the 2008 financial crisis is the stretch most people have watched withdrawal-rate guardrails run through, and it made them look gentle. Sequences that start badly are where sequence-of-returns risk does its damage, so their 2024 study ran both approaches through four well-known hard starting points in U.S. history, for a household drawing from a $1,000,000 portfolio of 60% stocks and 40% bonds. The Guyton-Klinger household started at $43,000 a year, a level that would have lasted in 90% of historical periods, so the test did not stack the deck with an aggressive start. The risk-based household started at a level with a 20% risk of overspending, raised spending when that risk fell to zero, and cut when it reached 75%, moving its risk of overspending 20 percentage points back toward the start (from 75% to 55%). Those trigger levels match the ones we generally start from, though the study’s cuts were larger than our default, which trims only a tenth of the way back, and its summary describes its risk measure as Monte Carlo based, where we measure ours against history.

Retiring in 2007, before the financial crisis. Guyton-Klinger rules called for three cuts in quick succession, leaving real spending 28% below plan, and it took more than a decade to climb back. Risk-based guardrails cut 3%.

Retiring in 1999, before the dot-com crash. Guyton-Klinger rules hit the lower guardrail three times and bottomed out 36% below plan. Risk-based guardrails never cut at all.

Retiring in 1936, in the Great Depression. Two cuts and skipped inflation raises drove Guyton-Klinger spending 45% below plan, nearly halving it in the most active years of retirement. Risk-based guardrails bottomed out 8% below plan.

Retiring in 1965, before the stagflation era. Guyton-Klinger spending fell 54% below plan and stayed down for more than a decade. Risk-based guardrails fell 32% below plan, for several years.

The gap comes from overcorrection. Because a withdrawal rate cannot see the rest of the plan, the rules keep cutting after the plan has already done enough, preserving capital at the expense of the life it was meant to fund. In the dot-com test, the Guyton-Klinger portfolio, which began at $1,000,000, still held $1,590,000 in 2023, while the risk-based portfolio held $710,000. At its low point, the Guyton-Klinger household had cut spending by more than a third to protect money it turned out not to need, money that could have been spent or given along the way. (That test included no legacy goal; a plan with one would build it in from the start.)

Fairness cuts the other way too. The study skipped the inflation raise after any losing year, without the original rule’s withdrawal-rate condition, so on that point its test was somewhat stricter than the 2006 paper, most of all in the 1936 result, where skipped raises did much of the damage. A 32% cut is also still a deep one, even for a few years, and for a household whose budget is mostly essentials it may simply not be possible. Tharp and Fitzpatrick note that in more realistic plans, with Social Security and declining spending built in, cuts in the stagflation period are typically smaller than in their simplified test. And because the risk-based approach is the authors’ own, it is worth knowing that their Guyton-Klinger findings line up with independent work they cite. Karsten Jeske’s 2017 analysis found that a retiree who started in 1966 with a 4% withdrawal rate under Guyton-Klinger rules would have seen withdrawals fall 59%. Wade Pfau’s 2015 Monte Carlo study found that a retiree starting at 4.8% faced a 20% pay cut by year 10, even in the median scenario.

The Floor Under the Plan: Core and Adaptive Spending

Guardrails assume spending can move. Not all of it can, and not all of it should. We lean toward building every plan in two layers. Households start by sorting needs from wants, and then we turn that into core spending and adaptive spending. Those labels come from Michael Kitces’s 2019 work at Kitces.com (building on Jonathan Guyton’s approach), which splits each spending category into the core of a household’s lifestyle and the part that can flex. (We use “adaptive,” “discretionary,” and “non-core” interchangeably; the conversation matters more than the label.)

The core number is a hard floor: the plan does not cut below it, and guardrail cuts come out of the adaptive layer. Core is not the bare minimum. Technically, core could be groceries and shelter, but a retirement built only on that is not the retirement anyone saved for. For many of the families we work with, in Texas and beyond, an annual trip with the children and grandchildren is core: it happens every year. What is adaptive is the size of it, a rented house at the lake or a week abroad, four nights or ten. Giving can work the same way, with a household’s regular giving treated as core and larger gifts as the adaptive part.

The layers do two jobs. First, they make a cut concrete: when we tell a household it is getting close to the lower guardrail, we already know together which part of the budget would give. Second, they shape where spending can start. The larger the core share, the less room adjustments have to work with, and the more conservatively the plan has to begin. Tharp and Fitzpatrick illustrate the range with two retirees: one who spends half the budget or more on discretionary items might be comfortable with the risk of a deep cut, while one spending 90% on essentials would want a plan whose worst case is a small one. Reliable income, Social Security above all, often covers much of the core layer by itself, which is part of why claiming decisions and spending decisions get made together. One cost fits neither layer neatly: long-term care, which we lean toward planning case by case, often as its own named and modeled reserve, as we describe in funding long-term care.

What We Show Clients Instead of a Probability-of-Success Score

We lean toward not handing clients a probability-of-success score at all. “An 80% probability of success” sounds like a 20% chance of failure, and failure sounds like running out of money. In a plan that adjusts, that is not what it means. The 20% are the scenarios where the starting spending, held steady, would have been too high, and where the guardrails would have trimmed it along the way. To put round numbers on it, as an illustration rather than a result from any plan: across those scenarios, spending might run something like 5% below plan on average, and perhaps 20% below at the worst, with the very hardest historical starts, like the ones in the stress tests above, going deeper. That is a very different conversation from a one-in-five chance of going broke.

What we show instead comes in three forms.

The decision tree, in dollars. Say, hypothetically, a couple with a $2,000,000 portfolio spends $11,000 a month, Social Security included. Their guardrails might read: if the portfolio falls to about $1,450,000 and stays below that line for two quarters, spending moves to about $10,600 a month; if it grows to about $2,500,000, spending moves to about $12,300 a month. (These are round numbers to show the shape; where the lines fall depends entirely on the plan.) The most useful part is what this says about everything in between. A drop to $1,700,000 is unpleasant to watch, but it changes nothing about this year’s trip. The lines shift a little at every review as the couple ages and the plan updates, so they are approximate by design.

How often the plan would have adjusted. We show the percentage of historical scenarios in which spending ended up above plan and below it, and how often, historically, a raise or a cut would have come along over a retirement. In one of Tharp’s case studies, run with more conservative settings than ours, the couple would have spent more than planned in 76% of historical periods. Numbers like that set expectations better than any score: the household learns that raises are the more likely news, and that a cut, if it comes, is a known and survivable event.

The history, on request. Individual historical paths, such as what this plan would have done for someone retiring in 1966 or in 2000, are available to any client who wants to see them.

All of this lives in the plan document, and most of it lives in conversation. The numbers matter, but what makes them useful is talking through what each line would mean for the trips, the projects, and the giving the household actually has in mind.

Guardrails Work Within a Retirement; the Date Is Its Own Lever

Guardrails respond to whatever market a retiree happens to enter. They cannot change which market that is. Georgios Argyris’s 2026 research, looking at 30-year retirements across U.S. market history, estimated that about three-quarters of the variation in outcomes came from the overall return environment a retiree entered, and only about a quarter from the order of returns within it. In his opening example, two otherwise identical retirees each started by withdrawing 4% of a $1,000,000 portfolio of 60% stocks and 40% bonds and raised that withdrawal with inflation each year. One retired at the end of 1973 and the other at the end of 1975. Thirty years later, the first had $278,000 and the second $3.36 million.

Argyris draws the conclusion that the retirement date deserves a larger role in planning, and we agree with part of that. We lean toward letting the household’s life, not a market call, set the date, and toward keeping valuations out of the withdrawal rate a plan starts from; at most, they inform some investment decisions at the margins, never the spending plan. What we do instead is show a household, before the date is set, what flexibility is worth: a year later, a year of part-time work, a phased step back from a business. The point is not only whether the plan works. It is how much room the plan has for the things the household is retiring toward, and that is their decision to make with the numbers in front of them.

How We Think About It

We lean toward risk-based guardrails over withdrawal-rate rules, because the risk measure reads the whole household: when Social Security starts, how spending changes with age, what the plan has to leave behind, and how long it has to last. We lean toward running those plans on historical returns, and toward translating them into dollars before anything happens, so that a falling market brings a known number rather than a guess.

On the settings, we lean toward starting in the underspending zone, raising spending fully when the plan clearly allows it, and cutting in small steps, typically only after the plan has stayed past the lower line for two quarters, with warning beforehand. Underneath, we lean toward a hard floor set at a core level of spending that reflects how the household actually wants to live, not the bare minimum.

And we lean toward remembering what the raises are for. In our experience the more common problem is not spending too much but spending and giving too little. We see households that could afford the trip to Italy they have talked about for twenty years, or could help a child with a first home now rather than through an estate later, and do not, because a fixed number told them to be careful forever. A raise is room to spend or give more, and what it goes toward should point back to the household’s Statement of Financial Purpose. For many families that means more time together, more giving while they can see what it does, or simply more margin. For some, it means the raise is declined and the margin kept, and that is a fine answer too, as long as it is a decision rather than a default.

Common Questions

How are risk-based guardrails different from Guyton-Klinger guardrails?

Guyton-Klinger guardrails use the withdrawal rate as the trigger: spending rises 10% when the current withdrawal rate falls 20% below the starting rate and falls 10% when it rises 20% above it. Risk-based guardrails use the risk of the whole plan, measured across market scenarios with the household’s Social Security timing, spending path, legacy goal, and planning horizon built in, and then translate that risk into portfolio values and spending levels in dollars. In historical stress tests by Derek Tharp and Justin Fitzpatrick, risk-based guardrails called for much smaller cuts in the hardest periods: 32% versus 54% for a 1965 retiree, and none versus 36% for a 1999 retiree.

How big are the spending cuts with retirement guardrails?

It depends on the settings and on the market. Under the settings we generally start from, a single cut trims spending only enough to move the plan a tenth of the way back toward its starting risk level, often a few percent, and it typically comes only after the portfolio has stayed below the lower guardrail for two quarters. A long decline can bring more than one cut. In the hardest historical periods, total cuts can be large under any approach. In Derek Tharp and Justin Fitzpatrick’s tests, which used larger cut steps than our default, risk-based guardrails cut 3% for a 2007 retiree, 8% for a 1936 retiree, and 32% for several years for a 1965 retiree, against 28%, 45%, and 54% under Guyton-Klinger rules.

What withdrawal rate do retirement guardrails allow?

There is no single rate, because a risk-based plan sets spending from the whole household picture rather than from a percentage. When someone asks for a ballpark before any modeling, the rough sketch we lean toward is about 4% of the portfolio for a plan that will never adjust and about 5.5% for one that will follow guardrails. That sketch starts a conversation; it is not a rule of thumb and never a plan. The higher figure is bought with a willingness to adjust: it assumes spending and giving will come down in a bad decade and go up in a good one. The real number for a household depends on its other income, its spending path, its legacy goals, and how much of its budget is core.

What if part of our spending can never be cut?

That part becomes the core layer of the plan, and the plan treats it as a hard floor: guardrail cuts come only from the adaptive layer above it. The larger the core layer relative to the whole budget, the more conservatively spending has to start, since there is less room to adjust. Reliable income such as Social Security often covers much of the core layer, and core spending can include more than necessities, such as an annual family trip whose size and cost are the adaptive part.