Trinity Study Deep Dive: Origins and Limits of the 4% Rule

Origins of the 4% rule

"Withdraw 4% of your assets per year in retirement and you have a high probability of not running out over 30 years." This is the most-cited rule in retirement planning. It comes from William Bengen's original 1994 paper and a 1998 follow-up by three Trinity University professors (Cooley, Hubbard, Walz) — together known as the Trinity Study.

Bengen 1994: the original paper

William Bengen, a California-based financial planner, published "Determining Withdrawal Rates Using Historical Data" in the 1994 Journal of Financial Planning.

Methodology:

Data: 1926–1992 US stock and bond returns
Assumption: 50% stocks, 50% bonds
Withdrawal method: first-year withdrawal of X% of assets, then adjusted for inflation annually
Test: for any retirement year in that historical window, would the portfolio last 30 years at that withdrawal rate?

Findings:

4% was the "safe regardless of retirement year" rate
5%+ ran out within 30 years for some start dates (e.g., retiring in 1965–1966 right before high inflation)
3% was extremely conservative — after 30 years, residual assets were typically 2×+ the starting amount

Bengen called 4% the "SAFEMAX" (Safe Maximum Withdrawal Rate).

Trinity Study 1998: expanded research

The three Trinity professors built on Bengen. Key additions:

Extended data range to 1926–1995
Tested various stock/bond ratios (100/0 down to 0/100)
Tested multiple horizons (15, 20, 25, 30 years)
Defined "success" as ending the period with assets greater than zero

Core finding (30-year horizon):

Stock allocation	4% success rate	5% success rate
100% stocks	95%	80%
75% stocks / 25% bonds	98%	83%
50% stocks / 50% bonds	95%	75%
25% stocks / 75% bonds	71%	30%
100% bonds	20%	0%

Takeaways:

4% had 95%+ success for 50–100% stock portfolios
Moving to 5% dropped success rates noticeably, especially for lower stock allocations
The "ultra-conservative" all-bond portfolio had a high failure rate

Why all-bond portfolios fail

Intuitively, "all bonds is safest" — but in practice:

Bonds have low nominal returns (3–5%)
Real returns after inflation are very low (0–2%)
Withdrawing 4% exceeds real returns → principal erodes
In high-inflation periods, depletion accelerates

Lesson: retirement portfolios can't fully avoid equities. Stocks' long-term growth is the main defense against inflation and withdrawals.

Bengen's updates: upgrading to 4.5%

In 2006, Bengen's book Conserving Client Portfolios During Retirement proposed:

With small caps, international stocks, and REITs added
SAFEMAX could rise to 4.5%

In 2020, Bengen updated again:

With a diversified portfolio and moderate inflation
SAFEMAX could reach 4.7%

Even so, 4% remains the most-cited standard.

Limitations of the Trinity Study

1. US-only data

The Trinity Study is built on US stocks and bonds. US equities had the best returns of any major market in the 20th century. Results might differ significantly if you were in:

Japan (stock market flat for 20 years after the 1990s crash)
Taiwan (relatively volatile)
Europe (stagnant after the financial crisis)

2. 30-year assumption

Trinity assumed a 30-year retirement. If you:

Retire early (e.g., age 50, living to 90 = 40 years): 4% carries higher risk
Retire late (e.g., age 65, living 20 more years): 4% is overly conservative

3. "Strict" inflation adjustment

The original research assumed you mechanically adjust withdrawals for inflation every year. In reality, retirees often:

Naturally reduce spending during high inflation
Reduce withdrawals when markets are down

This kind of dynamic withdrawal can support higher initial rates (5–6% in some studies).

4. Impact of two consecutive crash years

Trinity did include 1929, 1973, and 1987 crashes, but a hypothetical 2 consecutive years of −30% right after retirement cuts assets by 50% before withdrawals, which is hard to recover from. This is Sequence of Returns Risk — see Monte Carlo retirement simulation.

Practical adjustments for Taiwan

1. Inflation assumption

Taiwan's inflation over the past 20 years has averaged 1.5–2% (below the US's 2.5–3%), which is relatively favorable for retirement withdrawals.

2. Investment vehicles

Holding global stock ETFs (VT, VWRA) and bond ETFs (BND, AGG) essentially replicates the Trinity research portfolio.

3. Labor Insurance and Labor Pension as a base

Taiwan's Labor Insurance (勞保) and Labor Pension (勞退) offset part of the withdrawal pressure. See how much Taiwan residents need to retire.

4. TWD vs USD

Investing in US assets means you'll need to convert to TWD for spending in retirement, taking on currency risk. This variable wasn't part of the original study.

Newer academic perspectives

Dynamic withdrawal strategies

Recent research proposes "Guardrails":

Good markets: raise the withdrawal rate (up to 5–5.5%)
Bad markets: cut to 3–3.5%

This flexible withdrawal approach performs better in long-life scenarios.

Late-retirement allocation

Academic work increasingly supports a "retirement glide path":

Early retirement: 30–40% stocks (low volatility, defending against SoR risk)
Mid-retirement: 40–60% stocks
Late retirement: 60–70% stocks (shorter remaining horizon, less stock risk)

This is the opposite of the traditional "more bonds as you age" advice, but has statistical support.

Practical guidance

When applying Trinity Study conclusions:

4% is a starting point, not the final answer
Your situation may justify 3.5–5% — use the Monte Carlo simulation to check your success rate
Labor Insurance and Labor Pension as a base reduce personal withdrawal needs, allowing a wider effective rate
Preserving flexibility (cutting spending in bad markets) is more robust than mechanically defending 4%

Tools

4% withdrawal calculator: enter your numbers to see scenarios at various rates
Monte Carlo simulation: run 1,000 scenarios to see the success rate

Disclaimer

The Trinity Study is historical back-testing research and does not guarantee future performance. Retirement planning should integrate individual circumstances, longevity expectations, income sources, and other factors.