At the time, I did not think of attending Arizona’s flagship public university as making a six-figure financial decision. In-state tuition was $2,272 for the year. Add $76 in required fees, and the published tuition bill was $2,348.

Twenty-five years later, the University of Arizona lists tuition and fees for an Arizona resident at $13,900. Add its on-campus estimate for housing and food, and one year now totals $31,670.

Meanwhile, the university is confronting a very different enrollment market. Its fall 2025 first-year class fell 19%, driven mostly by fewer out-of-state and international students. In May, UA said the incoming fall 2026 class would likely be smaller again, although the final count will not be known until the fall census.

I am not going to tell you the price increase caused the enrollment decline. The university itself points to a mix of national, geopolitical, economic and policy forces, and its Arizona-resident yield has actually improved.

I am going to show you what happened to the price—and let you decide whether the value proposition changed.

The apples-to-apples price

UA’s 2000–01 Common Data Set gives us an unusually clean starting point. It reported $2,272 in tuition, $76 in required fees, and $5,888 for on-campus room and board. Total: $8,236.

For 2026–27, UA estimates $13,900 for resident tuition and fees and $17,770 for on-campus housing and food. Total: $31,670.

That is the comparison that matters: $8,236 became $31,670—a 284% increase.

• Tuition and fees: $2,348 → $13,900, up 492%.

• Room and board/housing and food: $5,888 → $17,770, up 202%.

• Combined: $8,236 → $31,670, up 285% before rounding.

UA’s broader cost-of-attendance estimate is $37,270 after books, travel and miscellaneous expenses. I am not using that larger number in the historical comparison because the 2001 figure above does not include the same categories. The case is strong enough without mixing definitions.

If the price had merely tracked inflation

The Consumer Price Index for All Urban Consumers averaged 177.1 in 2001. In June 2026—the latest available reading as of this analysis—it stood at 333.952.

That means the general price level increased about 88.6%.

If UA’s $8,236 tuition-plus-room-and-board price had risen at that rate, it would now be about $15,530.

The actual comparable estimate is $31,670—more than twice the inflation-adjusted number.

Figure 2. What ordinary inflation would have produced versus UA’s current estimate.

Tuition alone is even more striking. The 2001 tuition-and-fee bill of $2,348 would equal about $4,428 after ordinary inflation. The current published figure is $13,900.

Put another way, using 2001 as an index of 100:

• General CPI reached 189.

• UA housing and food reached 302.

• UA’s combined price reached 385.

• UA tuition and fees reached 592.

Figure 1. Indexed price growth, with 2001 = 100.

The graph makes the argument without adjectives. Tuition did not merely keep pace with the rest of the economy. It ran far ahead of it.

The affordability test: how many weeks of Arizona wages?

Percentages can feel abstract. Weeks of work do not.

The Bureau of Labor Statistics reported average annual pay of $33,408 for covered Arizona workers in 2001. One year of UA tuition, fees, room and board therefore represented about 12.8 weeks of the average worker’s gross pay.

The latest full-year statewide figure available from the same BLS wage program is an average weekly wage of about $1,224. At that rate, the current $31,670 annual UA price consumes about 25.9 weeks of gross pay.

The wage burden roughly doubled: from about one-quarter of a working year to about one-half.

Figure 3. The annual UA price expressed in weeks of average Arizona gross pay.

This is not a perfect measure. Average wages are not median wages, the workforce mix changes, and most families do not pay an entire college bill from one year’s current earnings. But the definition is consistent enough to answer the basic question: did Arizona earnings keep up with UA’s published price?

They did not.

What was waiting on the other side of the degree?

College is not just a purchase. It is an investment in future earning power. So the fairer test is not simply what the degree cost, but what a new graduate could plausibly earn afterward.

The National Association of Colleges and Employers reported an average starting salary of $42,557 for bachelor’s degree graduates in 2001. Its latest final figure is $65,677 for the Class of 2024.

Starting pay rose about 54% in nominal dollars. UA’s combined annual price rose 285%.

Using four years at each year’s published price as a simple benchmark:

• 2001-era four-year cost: $32,944, equal to about 9.3 months of the average starting salary.

• Current four-year cost: $126,680, equal to about 23.1 months of the latest average starting salary.

Figure 4. Four annual published costs expressed in months of the average new graduate’s starting salary.

The tuition-only version tells the same story. Four years of 2001 tuition and fees equaled about 2.6 months of the average new graduate’s gross pay. Four years at today’s published tuition and fees equals about 10.2 months.

This is not a payback schedule. It ignores taxes, scholarships, annual price changes, time out of the workforce and the fact that students must eat and live somewhere whether they attend college or not. It also does not count the lifetime earnings premium that a bachelor’s degree can still provide.

It is a scale test. The price of obtaining the credential grew much faster than the salary attached to newly using it.

The degree on the diploma matters more now

UA does not charge engineering students one tuition rate and education or communications students an entirely different base price. Yet the labor market places very different values on the degrees.

NACE’s final Class of 2024 data show average starting salaries of $80,482 for engineering graduates, $68,644 for business graduates and $63,608 for health-professions graduates. Electrical engineering averaged $90,526, while the overall bachelor’s average was $65,677.

These are national figures rather than UA-specific outcomes, and they should be read as broad signals, not promises.

The point is not that one major is “good” and another is “bad.” Society needs teachers, nurses, engineers, artists and communicators. The point is that when the price of nearly every degree rises, the financial risk of choosing a lower-paying field rises with it.

A $2,348 annual tuition bill allowed more room for uncertainty. A $13,900 tuition bill narrows the margin for error.

Two caveats that matter

First, sticker price is not net price. Grants, scholarships and institutional aid reduce the amount many students actually pay. A student who lives with family or off campus with roomates faces a much lower housing cost; UA’s current estimate for a resident living with a parent is $22,640 including books, travel and miscellaneous expenses.

Second, housing and food are not entirely “college costs.” Students need both whether they enroll or not. But room and board are still part of the real decision for an Arizona student who must relocate to Tucson. That is why I have separated tuition from living costs throughout rather than hiding them inside one total.

Neither caveat erases the published-price trend. It tells us not to confuse a university’s price sheet with every family’s final bill.

So did the value proposition change?

The data does not say college has no value. A bachelor’s degree can still expand career options and lift lifetime earnings. UA remains a major research university with programs that create genuine economic and public value.

But “valuable” and “the same value proposition” are not the same claim.

In 2001, one year of UA tuition, fees, room and board cost $8,236. It equaled about 13 weeks of average Arizona pay and roughly one-fifth of the average college graduate’s starting salary.

Today, the comparable price is $31,670. It equals about 26 weeks of average Arizona pay and nearly half of the latest average starting salary.

Ordinary inflation cannot explain the gap. Arizona wages cannot explain it. New-graduate salaries cannot explain it.

Something fundamental changed in the price side of the transaction.

The numbers cannot tell us why students are not coming. They can tell us what those students are being asked to pay—and what the jobs waiting for them are likely to pay.

Where did the extra money go?

That is the obvious next question, but it deserves its own analysis rather than an easy accusation.

The Arizona Auditor General reports that UA had $3.23 billion in total revenue and $3.05 billion in total expenses in fiscal 2025. Net tuition and fees supplied $978.2 million, or 30.3% of revenue. Instruction accounted for $732.8 million, research for $631.1 million, and academic plus institutional support for $650.5 million.

The UA Foundation manages a $1.2 billion endowment with a 4.25% payout rate. That is substantial, but it is not a checking account: endowment gifts are generally restricted to donor-designated purposes, and a sustainable annual payout is only a fraction of principal. Research grants must largely fund the research for which they were awarded. Athletics and auxiliary operations must be measured against their own expenses.

None of that answers whether the institution’s cost structure changed in ways that benefited undergraduate students—or whether tuition absorbed growth elsewhere.

It frames the next investigation:

Where Does a University of Arizona Tuition Dollar Actually Go?

The right test will compare 2001 with today on an inflation-adjusted, per-student basis: instruction, administration, state funding, staffing, financial aid, research, facilities, debt and athletics.

Do not assume “administrative bloat” is the answer.

Test it. As always, The Math Doesn’t Lie.

Sources and methodology

• University of Arizona, Common Data Set 2000–01.

• University of Arizona Office of Scholarships and Financial Aid, 2026–27 Cost of Attendance.

• U.S. Bureau of Labor Statistics, CPI-U and Quarterly Census of Employment and Wages.

• National Association of Colleges and Employers, historical starting-salary series and Summer 2025 Salary Survey.

• University of Arizona Office of the Provost, Update on Enrollment for Fall 2026.

• Arizona Auditor General, University of Arizona FY2025 Highlights.

• University of Arizona Foundation, Asset Management Highlights.

Calculations use the 2001 annual-average CPI-U and the June 2026 CPI-U, the latest available reading when this draft was prepared. Four-year cost comparisons multiply one published annual price by four; they are scale comparisons, not estimates of an individual student’s actual bill or loan repayment.

About This Analysis

This article was researched with the assistance of artificial intelligence. AI was used to help locate public records, analyze historical and current financial data, perform calculations, and assist with editing. The underlying figures were checked against cited primary or authoritative sources. The analysis, conclusions, editorial judgment, and any errors remain the responsibility of The Math Doesn’t Lie.

That figure comes from Northwestern Mutual’s 2026 Planning & Progress Study, which asked 4,375 U.S. adults how much they believed they would need to retire comfortably. The average answer rose from $1.26 million in 2025 to $1.46 million in 2026—back to the same level reported in 2024.

It is an interesting measure of retirement anxiety. It is not a calculation of what any particular household needs.

Northwestern Mutual makes that distinction. Its release says there is no universal retirement number and points readers toward several rules of thumb. One is the familiar 25-times-spending formula. Another is the 4% Rule: withdraw about 4% in the first year, increase that dollar amount with inflation, and expect the portfolio to last roughly 30 years.

Four percent of $1.46 million is $58,400.

That gives us an experiment.

My previous article, “The 4% Rule Was Never Really a Rule,” examined why a historical success rate is not a law of retirement. This follow-up asks a narrower question:

What happens when the new $1.46 million retirement number meets the market events of the 21st century?

Not an average return. Not a smooth projection. The Dot-com Bust. The 2008 Financial Crisis. COVID. The 2022 stock-and-bond decline. And the Lost Decade that forced a retiree to experience two major collapses before the first one had fully faded from memory.

The answer is not simply yes or no.

The portfolio survives most of the tests. But the result depends enormously on when retirement begins—and on whether the same $1.46 million sits in a Roth account or a traditional IRA or 401(k).

The number is the same. The spendable retirement wealth is not.

What is Retirement Decision Lab?

Retirement Decision Lab, or RDL, is a project in development for The Math Doesn’t Lie website. The website is also tracking a live covered-call experiment using a portion of my own portfolio, with results updated monthly. It is being built as a transparent financial scenario-testing tool: users define the household, accounts, spending, income, taxes, healthcare and market assumptions, and the model calculates what follows.

RDL does not predict markets, select investments or tell anyone what they should do. It is not a financial advisor. Its purpose is to make the assumptions visible, apply the same rules consistently and show how a retirement plan behaves when conditions change.

That makes it useful for experiments like this one. Instead of declaring that $1.46 million is—or is not—enough, we can define a household, expose the assumptions and run the same plan through different 21st-century events.

The experiment

I constructed a hypothetical married couple in Retirement Decision Lab 1.6. The couple retires at age 67 and the model follows them for 30 years, through age 97.

The starting assumptions are intentionally ordinary:

• $1.46 million invested at retirement

• 60% stocks and 40% 10-year U.S. Treasury bonds

• Annual rebalancing

• 7% nominal return in ordinary modeled years

• $58,400 of spendable portfolio support in the first year

• Spending increased by 2.5% annually

• $38,496 in annual Social Security, based on the 2026 average for an aged couple with both spouses receiving benefits

• Social Security receives a 2.5% cost-of-living adjustment in the model

• Both spouses are enrolled in Medicare

• Medicare Part B, supplements, Part D and an out-of-pocket allowance are included

• Federal and Arizona income taxes are modeled

• No pension, employment income, inheritance or spending reduction

The household begins with $96,896 of gross annual resources: $58,400 from the portfolio and $38,496 from Social Security. Medicare costs are included within that budget rather than added after the fact.

This is not a recommendation and it is not a probability forecast. It is a deterministic stress test: hold the household constant, change the market sequence, and see what breaks.

A necessary correction to the stress data

The historical templates in RDL use S&P 500 total-return sequences. That is appropriate for an equity stress proxy, but this experiment assumes a 60/40 portfolio.

Applying the S&P 500 decline to the entire account would describe a 100% stock portfolio while calling it balanced.

For this analysis, I blended each historical S&P 500 return with the corresponding return on 10-year U.S. Treasury bonds:

60/40 return = 60% × stock return + 40% × Treasury return

The underlying annual return data come from New York University professor Aswath Damodaran’s historical series. After each event sequence ends, the model returns to the ordinary 7% assumption.

This adjustment matters. In 2008, the S&P 500 lost 36.55% while 10-year Treasuries gained 20.10%. A rebalanced 60/40 portfolio lost approximately 13.89%. Bonds did what retirees hope they will do.

In 2022, stocks lost 18.04% and 10-year Treasuries lost 17.83%. The 60/40 portfolio lost approximately 17.96%. Bonds joined the decline.

The allocation was the same. The protection was not.

Figure 1. Thirty-year ending balances under identical household spending.

The smooth projection hides the experiment

Under a perfectly smooth 7% return, the Roth portfolio grows from $1.46 million to approximately $4.09 million after 30 years, despite the inflation-adjusted withdrawals. The traditional account ends with approximately $2.76 million after funding taxes along the way.

That is the retirement-calculator result people are accustomed to seeing. Enter a balance, a withdrawal and an assumed return. Watch the line rise.

The calculation is not wrong. It is incomplete.

Returns do not arrive in a straight line, and withdrawals make the order of those returns matter. A loss near the beginning of retirement damages more than the current balance. It removes capital that can no longer participate in the recovery while the retiree continues selling assets to fund spending.

This is sequence-of-returns risk. The post-2000 tests show its scale.

The Dot-com retiree

The Dot-com sequence begins with three consecutive stock-market losses. Bonds reduce the damage substantially, but the balanced portfolio still experiences a difficult opening:

• 2000: +1.25%

• 2001: –4.88%

• 2002: –7.13%

• 2003: +17.17%

The Roth version survives and finishes with approximately $2.00 million.

The traditional version also survives, but finishes with only about $888,000. That is a $1.11 million gap created without changing the starting balance, portfolio allocation, Social Security or lifestyle.

The difference is taxation.

The 2008 retiree

The Financial Crisis produces the most dramatic first-year stock loss in the historical calendar-year sequences, but Treasury bonds provide powerful protection. The 60/40 portfolio falls approximately 13.89%, then participates in the recovery.

After 30 years:

• Roth account: approximately $3.01 million

• Traditional account: approximately $1.79 million

This is the traditional 60/40 argument working as intended. The bond allocation does not eliminate risk. It prevents a 36.55% stock collapse from becoming a 36.55% portfolio collapse.

The COVID retiree

RDL treats COVID as a peak-to-trough crash proxy compressed into one annual model step, followed by a recovery step. It is intentionally harsher than using the positive 2020 calendar-year stock return.

The modeled 60/40 portfolio falls 20.34% in the crash step, then rebounds sharply. Both account types survive comfortably:

• Roth account: approximately $3.83 million

• Traditional account: approximately $2.51 million

COVID is frightening but brief in this model. A fast recovery allows the portfolio to regain the capital lost near the beginning.

The 2022 retiree

The 2022 sequence is different because the usual diversifier fails. Stocks and Treasury bonds fall together.

The first-year return is almost identical whether the portfolio holds 60% stocks or 70% stocks:

• 60/40: –17.96%

• 70/30: –17.98%

The subsequent stock recovery helps the more aggressive allocation, but the opening year exposes a weakness in treating “bonds” as a synonym for safety. Bonds carry interest-rate and inflation risk. They do not promise positive returns when stocks decline.

The 60/40 results still survive:

• Roth account: approximately $2.85 million

• Traditional account: approximately $1.65 million

The Lost Decade is the real test

The Lost Decade combines the Dot-com Bust with the 2008 Financial Crisis. The retiree absorbs three early down years, continues withdrawing through the recovery, and then encounters another major collapse before the decade ends.

Across 2000–2009, the annually rebalanced 60/40 portfolio still earns a positive cumulative market return of approximately 33.5%.

Yet the traditional retirement account fails.

That apparent contradiction is the entire point of sequence risk. A positive ten-year return does not guarantee that a retiree can spend safely through the path that produced it.

Figure 2. Lost Decade portfolio paths; the 70/30 line is an independent extrapolation.

The Roth portfolio finishes the 30-year period with approximately $991,000. In starting purchasing-power terms, that is about $473,000. It survives, but it is not a comfortable pass: the ending balance is also the lowest balance, meaning the portfolio is still declining at age 97.

The traditional 60/40 account enters the final year with only $81,016. It is depleted during that year and leaves approximately $39,161 of spending unfunded.

An independently extrapolated traditional 70/30 portfolio performs worse. Using the same estimated cash draws, it is depleted around year 26—approximately four years earlier than the 60/40 portfolio.

More stocks do not rescue this retirement because the additional equity exposure increases the early losses. The stronger recoveries arrive after the household has already withdrawn from a more damaged account.

Taxes quietly change the withdrawal rate

The clean Roth experiment withdraws exactly $58,400 in the first year. That is 4% of $1.46 million.

The traditional account must support the same lifestyle and pay the taxes created by its own distributions. RDL calculates approximately $9,255 of first-year federal and Arizona taxes.

The portfolio therefore distributes approximately $67,654:

$58,400 for spendable portfolio support

+ $9,255 for taxes

= $67,654 total distribution

That is a 4.63% withdrawal from the first year—not 4%.

Figure 3. Taxes increase the traditional account’s first-year distribution above 4%.

This distinction is easy to miss in retirement discussions. A traditional IRA may display $1.46 million on the statement, but every dollar is not available for spending. Some of the balance represents a future tax liability.

The tax cost also compounds. In the traditional Lost Decade run, the household pays approximately $388,000 of modeled lifetime taxes before the portfolio is depleted.

Taxes do not merely reduce the inheritance at the end. They force larger withdrawals throughout the retirement, including the early years when selling additional assets is most damaging.

RMDs add another layer. Beginning at age 73, the traditional account must distribute money according to the IRS schedule whether the household needs the full amount or not. Unspent RMD proceeds are moved into a taxable account in the model, so they are not assumed to vanish. But the distribution still creates taxable income and can increase Medicare premiums through IRMAA.

The Roth account has neither ordinary-income tax on qualified withdrawals nor owner RMDs under current law.

That does not make Roth money free. The tax was paid before the money entered the account or when it was converted. This experiment does not compare the lifetime cost of accumulating the two portfolios. It begins at retirement with the same displayed balance and asks what each balance can support from that point forward.

That is precisely why the comparison matters.

What does 70/30 change?

A higher stock allocation raises expected long-term return, but it also increases exposure to an early equity collapse.

In the Dot-com opening:

• 60/40 returns +1.25%, –4.88% and –7.13% in the first three years

• 70/30 returns –1.32%, –6.62% and –10.84%

During 2008:

• 60/40 loses approximately 13.89%

• 70/30 loses approximately 19.56%

Across the complete 2000–2009 sequence:

• 60/40 cumulative market return: approximately +33.5%

• 70/30 cumulative market return: approximately +23.2%

The more aggressive portfolio does better when the shock is brief and the rebound immediate, as in the COVID proxy and the 2022–2024 sequence. It performs worse when retirement begins before prolonged or repeated losses.

Expected return is not the retiree’s only problem. The path determines how much capital remains available to earn that return.

What the $1.46 million figure can—and cannot—tell us

Northwestern Mutual’s survey captures a widespread belief about the amount needed for a comfortable retirement. The same study finds that 48% of Americans believe they are somewhat or very likely to outlive their savings.

The RDL experiment helps explain why both beliefs can coexist.

$1.46 million sounds precise, but it leaves the most important questions unanswered:

• At what age does retirement begin?

• How much Social Security or pension income is available?

• Is the portfolio Roth, tax-deferred or taxable?

• What does “comfortable” cost?

• Does spending increase with inflation?

• How are healthcare and Medicare funded?

• What happens if retirement begins at the edge of a Lost Decade?

• Is there any ability to reduce spending after a shock?

The survey number is not wrong. It is incomplete by design. It measures what Americans believe they need, not whether a particular retirement plan survives a particular market sequence.

The number is not the plan

The controlled experiment produces three conclusions.

First, $1.46 million can support a 4% inflation-adjusted withdrawal plus average Social Security through most post-2000 market shocks under the assumptions tested.

Second, a Roth portfolio and a traditional retirement portfolio with the same displayed balance are not economically identical. Under the Lost Decade stress, the Roth account survives with $991,000 while the traditional account is depleted.

Third, increasing stock exposure from 60/40 to 70/30 does not automatically improve retirement durability. It helps during rapid recoveries and hurts when large losses arrive early or repeat before the portfolio has healed.

The original 4% research never promised certainty, and the $1.46 million survey result does not create it.

A retirement number is a starting input.

The retirement plan is the interaction between spending, taxes, Social Security, healthcare, asset allocation, inflation and time.

The balance may be $1.46 million.

The outcome depends on everything that happens next.

Methodology and limitations

This analysis used Retirement Decision Lab 1.6, a deterministic financial modeling tool. It is not a financial advisor and produces no investment recommendation.

Historical stock returns are S&P 500 total returns. Bond returns are annual returns on 10-year U.S. Treasury bonds from Aswath Damodaran’s historical data series. The 60/40 and 70/30 event returns are calculated from those two series with annual rebalancing.

The COVID test is RDL’s compressed peak-to-trough proxy rather than the positive 2020 calendar-year return. The historical 2022 test applies the market return sequence while preserving the scenario’s 2.5% spending-inflation assumption; it is not a complete replay of the actual inflation path.

The primary 70/30 outcome is an independent extrapolation rather than a full RDL rerun. It holds the estimated traditional-account cash draws constant while changing the historical return path.

The tax model uses married-filing-jointly federal rules and Arizona’s 2.5% income-tax rate. Future tax brackets are planning approximations. Results are rounded and apply only to the stated assumptions.

No long-term-care event, pension, employment income, advisory fee, spending adjustment or Social Security reduction was modeled. Adding or changing any of these assumptions could materially alter the results.

Sources

Northwestern Mutual, 2026 Planning & Progress Study:

https://news.northwesternmutual.com/2026-04-01-Americans-Believe-They-Will-Need-1-46-Million-to-Retire-Comfortably,-Up-More-Than-15-Since-Last-Year,-According-to-Northwestern-Mutual-2026-Planning-Progress-Study

Social Security Administration, 2026 COLA fact sheet:

https://www.ssa.gov/news/en/cola/factsheets/2026.html

Medicare, 2026 costs:

https://www.medicare.gov/basics/costs/medicare-costs

Aswath Damodaran, NYU Stern, historical returns on stocks, bonds and bills:

https://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile/histretSP.html

The Math Doesn’t Lie, “The 4% Rule Was Never Really a Rule”:

https://themathdoesntlie.substack.com/p/the-4-rule-was-never-really-a-rule

When I launched The Math Doesn’t Lie, I began with water, agriculture, and infrastructure. Those subjects gave me the right place to establish the publication’s method: start with the evidence, make the assumptions visible, and show the math.

But finance—especially retirement, investing, and the decisions that determine whether a financial plan actually works—is where my strongest personal interest has always been.

This article marks the beginning of that next chapter. The subject is changing, but the standard is not.

Within days of each other, I came across two retirement articles that reached very different conclusions. One argued that the traditional 4% rule was effectively dead and that retirees should plan to withdraw significantly less. Another described how William Bengen—the financial planner whose research helped create the rule—had suggested that a higher starting withdrawal rate might be reasonable under a broader set of assumptions and a longer historical record.

Both articles sounded plausible. Taken together, however, they created an obvious problem. How could two serious discussions of the same retirement rule point in opposite directions?

As an engineer, that is the kind of disagreement I find useful. When two credible analyses produce different answers, the first question is not which headline is right. It is what assumptions caused the answers to diverge.

That question led me back to the original research. What I found is that the public conversation has become overly focused on the number and not nearly focused enough on the method that produced it.

The Number We Remembered

The phrase “the 4% rule” appears everywhere: financial books, retirement calculators, podcasts, news articles, and planning conversations. It is often presented as though it were a settled principle: withdraw 4% of a portfolio in the first year of retirement, increase that dollar amount with inflation, and the money should last for thirty years.

That summary is useful, but incomplete. It compresses a complex historical experiment into one memorable percentage. In the process, it hides the most important parts of the research: the portfolio assumptions, the retirement horizon, the inflation convention, the sequence of market returns, and the definition of success.

The number became famous because people wanted a simple answer to a difficult question. But William Bengen did not begin with four percent. He began with the question every retiree eventually asks:

How much can I spend each year without exhausting the portfolio that must support the rest of my life?

Before Bengen’s work, retirement planning often relied heavily on average returns. An advisor could estimate a long-term investment return, estimate inflation, and calculate a spending rate that appeared sustainable on paper.

The averages might be mathematically correct. The retirement plan could still fail.

A retiree does not receive thirty years of average returns. A retiree receives thirty individual years in a particular order. Some retire into long expansions. Others retire immediately before a market crash, an inflation surge, or a period when both stocks and bonds struggle.

Two retirees can own the same portfolio, withdraw the same amount, and earn the same long-term average return—yet experience radically different results because the returns arrived in a different sequence.

Bengen’s Real Breakthrough

In October 1994, Bengen published “Determining Withdrawal Rates Using Historical Data” in the Journal of Financial Planning. His breakthrough was not forecasting future returns more accurately. It was removing the forecast from the center of the experiment.

Bengen reconstructed retirement after retirement using actual historical market returns and inflation. Each simulation began in a different calendar year. The retiree took an initial withdrawal, increased future withdrawals with inflation, and experienced the market sequence that actually followed.

The test was straightforward: did the portfolio remain above zero for the full retirement period?

This changed the nature of the problem. Retirement planning was no longer only a debate about what returns might occur. It became a repeatable test of what had occurred under many different historical sequences.

Bengen’s lasting contribution was the methodology, not merely the percentage.

The assumptions were visible. The experiment could be repeated. The variables could be changed. The analysis could be extended when new history became available.

That is the real legacy of the original paper.

The number became famous. The methodology changed the profession.

The Trinity Study Expanded the Experiment

Four years later, Philip L. Cooley, Carl M. Hubbard, and Daniel T. Walz expanded the framework in the research commonly known as the Trinity Study. Their work did not replace Bengen’s method. It widened the design space.

Instead of evaluating one narrow set of assumptions, the Trinity researchers examined multiple withdrawal rates, stock-and-bond allocations, retirement horizons, inflation assumptions, and historical starting periods.

The result was not one universal answer. It was a matrix of outcomes.

Readers could see the tradeoffs. Higher starting withdrawals increased the risk of depletion. Longer retirements were harder to support. Stock exposure provided the growth needed for long horizons but introduced volatility. Bond exposure reduced some forms of risk but limited long-term growth.

The Trinity Study turned one historical test into a matrix of retirement assumptions.

One of the study’s most useful contributions was the portfolio success rate: the percentage of tested historical retirement periods in which the portfolio survived the full horizon.

That metric is valuable, but it is frequently misunderstood. A 100% historical success rate is not a guarantee that a strategy will succeed in the future. It means the strategy survived every historical period tested under the stated assumptions. Change the assumptions, the data, or the definition of success, and the result can change.

The Trinity Study did not prove the 4% rule in the way a theorem is proved. It showed how a particular withdrawal strategy performed across a defined historical record.

That distinction matters because the historical record did not stop in 1998.

Thirty More Years Changed the Evidence

When the Trinity Study was published, researchers had roughly seven decades of United States market history available for analysis. Today, the dataset is nearly three decades larger.

Those additional years were not quiet. They included the end of the 1990s technology boom, the dot-com collapse, the Global Financial Crisis, a prolonged period of unusually low interest rates, the COVID-19 pandemic, the sharpest inflation surge in decades, and one of the most difficult modern periods for bonds.

Each episode added a different kind of stress test. The early 2000s challenged equity-heavy portfolios after a long expansion. The 2008 crisis tested portfolios during a systemic banking shock. The years after 2008 tested retirement planning in an unusually low-rate environment. The pandemic combined a sudden decline with a rapid recovery. The 2022 inflation and bond shock challenged the assumption that high-quality bonds would reliably offset equity weakness.

The expanded record includes market, inflation, interest-rate, and bond stresses that were unavailable to the original researchers.

None of this automatically invalidates the original research. Recent experience should not be treated as inherently more important than older experience simply because it is familiar. But new evidence should not be ignored because an older rule became conventional wisdom.

The correct response is not to declare the 4% rule dead or to defend it on principle. The correct response is to rerun the experiment.

Use the expanded data. State the assumptions. Keep the methodology transparent. Then ask what changed.

Why Sequence Risk Matters More Than the Average

The expanded history leads directly back to the insight that made Bengen’s work important: the order of returns matters.

For an investor still accumulating wealth, a market decline can create an opportunity. Regular contributions buy more shares at lower prices, and time remains available for recovery.

Retirement reverses that relationship. Once withdrawals begin, a portfolio must absorb market losses while also funding spending. Shares sold during a downturn are no longer available to participate in the recovery.

Imagine two retirees with identical starting portfolios and identical long-term average returns. One experiences strong early gains followed by a downturn. The other experiences the downturn first and the gains later. Without withdrawals, the ending values may be similar. With ongoing withdrawals, the second portfolio can suffer permanent damage before the recovery arrives.

The same average return can produce very different retirement outcomes when the sequence changes.

This is why retirement portfolios are rarely broken by average years. They are tested by bad years arriving at the wrong time.

The dot-com collapse is an instructive example. Investors retiring around 2000 encountered a prolonged equity decline near the beginning of retirement. Markets eventually recovered, but retirees were withdrawing throughout the decline. The timing—not merely the average return—created the danger.

The Global Financial Crisis offered another version of the same test. The 2022 inflation and bond shock added yet another: what happens when inflation raises spending needs while both major portfolio components experience losses?

These episodes do not all behave alike. That is precisely why they are useful. A robust retirement framework should survive more than one type of stress.

The Next Question Is Not Another Magic Percentage

Once retirement is viewed as a system rather than a single withdrawal rate, the research agenda becomes broader.

The classic model intentionally isolates one problem: how much a retiree can withdraw from an investment portfolio under a defined set of assumptions. Real retirements include much more. Social Security and pensions reduce the amount that must come from investments. Taxes and account rules affect which dollars are available. Healthcare costs can change dramatically before and after Medicare. Spending is rarely perfectly fixed. Some retirees maintain cash reserves, work part time, delay large purchases, or reduce withdrawals during severe declines.

Each addition may change the outcome, but adding everything at once creates a different problem: no one can tell which variable caused the result.

The better approach is the one established by the original research. Ask one clear question. State the assumptions. Change one major variable. Test it against history. Then move to the next question.

Cash Reserves

If sequence risk is partly caused by forced selling during a downturn, can a temporary cash reserve improve the outcome by changing the timing of withdrawals?

Cash is unlikely to outperform equities over long periods, and it may lose purchasing power to inflation. But maximizing return is not the only function in a retirement system. Cash may serve as a timing and liquidity tool—giving markets time to recover before depressed assets must be sold.

That hypothesis can be tested. Hold the portfolio allocation, withdrawal rate, inflation convention, and market history constant. Change only the source and timing of withdrawals. If the reserve improves outcomes in specific stress periods, the result is useful. If it does not, that finding is useful too.

Bonds

The same discipline should be applied to bonds. Bonds have historically provided income, stability, and a source of withdrawals when equities decline. But the interest-rate and inflation environment has changed substantially since much of the original record.

A difficult bond market does not prove that bonds are obsolete. It does mean their role should be evaluated in context: starting yields, duration, inflation, credit quality, liquidity needs, and the presence of other income sources all matter.

The better question is not whether bonds are always safe. No investment is. The question is what role bonds should play inside a modern retirement system.

What This Article Does—and Does Not—Claim

This article does not produce a new universal safe withdrawal rate. It does not claim that four percent is correct for every retiree, or that it is obsolete. It does not claim that cash reserves or bonds automatically solve sequence risk.

It makes a narrower argument: the enduring contribution of the original retirement research was methodological. Bengen replaced average-return planning with historical sequence testing. The Trinity researchers expanded the test across more variables. The decades since then have added evidence worth examining.

A historical success rate remains descriptive, not predictive. It tells us how a strategy performed under the tested assumptions and available history. It cannot guarantee future returns, predict the next crisis, or identify one withdrawal rate for every household.

What it can do is make the assumptions visible. It can reveal recurring risks. It can show how different retirement decisions behaved across many market environments. And it can replace unsupported confidence with a framework that readers can inspect, challenge, and improve.

The mathematics has not changed. The evidence has grown.

The Experiment Continues

The purpose of returning to the 4% rule is not to defend a famous number or bury it. It is to recover the experiment that the number came to represent.

The original researchers worked with the evidence available to them. We now have additional market cycles, financial crises, inflation shocks, bond-market stress, a pandemic, and new retirement tools that did not fit inside the original model.

That expanded record creates an opportunity. We can rebuild the historical test, make every assumption explicit, and examine modern retirement strategies one variable at a time.

Can a one- or two-year cash reserve improve resilience during severe drawdowns? How should pensions and Social Security alter portfolio withdrawals? Can flexible spending outperform a rigid inflation-adjusted rule? How do taxes, healthcare, required distributions, Roth conversions, and account-access rules change the real-world problem?

Those questions will not produce one slogan that applies to everyone. They can produce something more valuable: a transparent body of evidence showing why the results change.

That is the standard for the financial work that follows at The Math Doesn’t Lie.

Read the original sources. State the assumptions. Extend the data. Show the math.

Sources and Further Reading

William P. Bengen, “Determining Withdrawal Rates Using Historical Data,” Journal of Financial Planning, October 1994.

Philip L. Cooley, Carl M. Hubbard, and Daniel T. Walz, “Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable,” AAII Journal, 1998.

The Globe and Mail, “The 4% withdrawal rule for retirees is dead.” https://www.theglobeandmail.com/investing/personal-finance/retirement/article-4-withdrawal-rule-retirees-dead-fire/

Yahoo Finance, “Suze Orman explains why people should withdraw less in retirement.” https://finance.yahoo.com/news/suze-orman-explains-why-people-150218592.html