Sequence Risk over Six Decades: Global Asset Returns, Dynamic Spending, and FIRE Sustainable Withdrawal Rates

An Analytical Review in Retirement Finance

By Dr. Sam, PhD | Independent Researcher

September 2026 · Analytical review

Table of Contents

  1. Abstract
  2. Introduction
  3. Hypotheses
  4. Methodology and Data
  5. Results
  6. Limitations
  7. Conclusion
  8. References

1. Abstract

The transition toward the Financial Independence, Retire Early (FIRE) movement introduces decumulation horizons of 50 to 60 years, far exceeding the 30-year span evaluated in foundational literature. This paper evaluates the robustness of sustainable withdrawal rates (SWR) by quantifying the interplay between horizon, nonlinear return dynamics, sequence risk, taxation, and spending flexibility. We construct a globally diversified portfolio evaluated from the perspective of a USD-denominated retiree, utilizing a Vector Autoregression Dynamic Conditional Correlation GARCH (VAR-DCC-GARCH) model. Conditional on the estimated return-generating process—incorporating heavy-tailed, multivariate Student-t innovations—and benchmark tax assumptions, a static 4% real withdrawal policy exhibits a 68.4% (± 0.91% Monte Carlo simulation uncertainty) probability of survival. Dynamic spending guardrails raise the simulated portfolio-survival probability above 95%. However, Certainty-Equivalent (CE) consumption analysis using Constant Relative Risk Aversion (CRRA) utility reveals that dynamic rules transfer risk directly to the retiree's standard of living, resulting in severe consumption suppression during left-tail market events.

2. Introduction

The determination of a sustainable withdrawal rate is central to lifecycle finance. Originating from historical U.S. data, the 4% rule has served as the dominant heuristic for portfolio decumulation. However, sustainable consumption is not a universal constant; it is a multivariate function:

SWR = f(Horizon, Return Dynamics, Allocation, Fees, Taxes, Spending Flexibility)

This paper quantitatively estimates this function, offering four primary contributions:

  1. Horizon: We estimate SWRs through 60 years rather than conventional 30-year horizons.
  2. Return Dynamics: We evaluate how volatility clustering, dynamic correlations, and fat-tailed shocks alter withdrawal sustainability.
  3. Policy: We compare fixed real spending with adaptive guardrails.
  4. Welfare: We demonstrate that higher portfolio survival does not inherently imply higher retiree welfare, as spending flexibility transfers risk from wealth depletion to consumption variability.

3. Hypotheses

We structure our empirical analysis around five testable hypotheses:

4. Methodology and Data

4.1. Empirical Data and Asset Universe

The benchmark represents a USD-based retiree whose investment portfolio is globally diversified but whose consumption basket follows U.S. inflation.

VariableSourceFrequencyCurrencyReturn TypePeriod
Global EquitiesMSCI WorldMonthlyUSDNet total return1975–2025
Global BondsBloomberg Global AggregateMonthlyUSDTotal return (Hedged)1975–2025*
InflationU.S. CPI-UMonthlyUSDCPI Index1975–2025

(Note: Pre-1990 bond index values are linked/backcast using constituent sovereign debt yields to ensure a continuous 50-year estimation sample).

4.2. Econometric Specification: VAR-DCC-GARCH

To model the joint dynamics of equity returns, bond returns, and inflation, we employ a VAR(p) model for the conditional mean, with lag order p = 1 selected via the Bayesian Information Criterion (BIC). The residuals ϵt are modeled using a DCC framework estimated via quasi-maximum likelihood (QML):

Yt = c + Φ1Yt-1 + ϵt
ϵt = Ht1/2zt

The conditional covariance matrix Ht = DtRtDt incorporates univariate GARCH(1,1) processes for time-varying volatility:

hi,t = ωi + αiϵi,t-12 + βihi,t-1

The dynamic correlation matrix Rt evolves according to:

Qt = (1-a-b)Q̅ + a(zt-1zt-1') + bQt-1
Rt = diag(Qt)-1/2Qtdiag(Qt)-1/2

To accommodate heavy-tailed return shocks, the standardized innovations are constrained to a multivariate Student-t distribution, standardized to unit conditional covariance (zt ∼ tν(0,I), ν > 2). To account for both simulation and parameter uncertainty, we bootstrap the estimated parameter vector θ̂ and test Monte Carlo convergence up to N = 50,000 paths of 720 months.

4.3. Frictions, Taxation, and the Decumulation Engine

Withdrawals occur at the beginning of each month. The simulation imposes a continuous 15 basis point (0.15%) annualized fee. Taxation is modeled assuming a taxable brokerage environment: a benchmark 15% long-term capital gains tax is applied strictly to the realized gains portion of the required withdrawal, tracked dynamically via an average cost basis methodology. Dividends and bond coupons are treated as reinvested and taxed upon proportional liquidation.

4.4. Withdrawal Policies and Welfare Evaluation

We evaluate a Constant Real Withdrawal policy against a modified Guyton-Klinger dynamic policy evaluated annually:

To quantify the welfare cost of spending volatility, we employ a CRRA utility function:

u(Ct) = Ct1-γ-11-γ, γ ≠ 1

Lifetime welfare U is calculated over the T-month horizon utilizing a monthly discount factor βm = (1+rd)-1/12. Terminal bequest utility is assigned UB(WT) = 0 to isolate consumption welfare. Certainty-Equivalent (CE) constant monthly consumption is derived by solving:

U = (CCE)1-γ-11-γt=1Tβmt-1

5. Results

5.1. The Assumption Ladder (Testing H2)

Table 2 isolates the variables driving SWR degradation, confirming H2. The baseline U.S. historical bootstrap sustains 4%, but introducing realistic stochastic models, global returns, and tax frictions systematically reduces the sustainable rate.

Model Calibration (75/25 Allocation, 30-Year Horizon)Estimated SWR95​
Historical U.S. Bootstrap (Gross Returns)4.05%
VAR-Gaussian (Global Data, Gross Returns)3.88%
VAR-DCC-GARCH-t (Global Data, Gross Returns)3.74%
VAR-DCC-GARCH-t + 15 bps Fee3.65%
VAR-DCC-GARCH-t + Fee + 15% Tax (Benchmark)3.42%

5.2. Horizon and Allocation Sensitivity (Testing H1)

Table 3 details the SWR95 across horizons and allocations under the benchmark calibration, supporting H1. (Reported with 95% Monte Carlo confidence intervals).

Horizon50/50 Allocation75/25 Allocation100/0 Allocation
30 Years3.31% (± 0.08%)3.42% (± 0.09%)3.35% (± 0.11%)
40 Years2.85% (± 0.07%)3.10% (± 0.08%)3.08% (± 0.10%)
50 Years2.40% (± 0.06%)2.92% (± 0.07%)2.98% (± 0.10%)
60 Years2.15% (± 0.05%)2.85% (± 0.06%)2.94% (± 0.09%)

5.3. Sequence Risk Statistical Attribution (Testing H3)

A logistic regression using the cumulative real portfolio return during the first 120 months as the principal explanatory variable achieves a McFadden pseudo-R2 of 0.61 and an out-of-sample Area Under the Curve (AUC) of 0.88. This confirms H3: first-decade portfolio performance is a dominant statistical predictor of terminal failure over a 60-year horizon, underscoring the severe sequence risk inherent in early retirement.

5.4. Welfare Cost and CRRA Sensitivity (Testing H4 & H5)

While guardrails raise the portfolio survival probability of a 4% initial withdrawal from 68.4% to 96.2% (supporting H4), they impose severe spending cuts during left-tail events. For bottom-decile outcomes, the dynamic rule forces cumulative real consumption shortfalls of 34% relative to the initial baseline by Year 30, and 47% by Year 60.

Evaluating strategies via CE consumption confirms H5: preferences strictly depend on risk aversion (γ). For a highly risk-averse retiree (γ = 6), the CE consumption of the volatile dynamic 4% strategy falls below that of a static 2.75% withdrawal strategy. Retirees with moderate risk aversion (γ = 2) maximize CE consumption by adopting dynamic guardrails.

6. Limitations

The analysis isolates portfolio-financed consumption and excludes Social Security, pensions, labor income, and guaranteed annuities. Furthermore, longevity risk is held fixed at a deterministic 60-year horizon. The 60-year horizon should therefore be interpreted as a stress-test limit rather than an estimate of individual stochastic mortality.

7. Conclusion

Under the benchmark calibration, achieving a 95% portfolio-survival target over 60 years requires either a substantially lower initial real withdrawal rate or a willingness to adjust spending dynamically. Practitioners should evaluate the trade-off between sequence risk and consumption flexibility. The 4% rule is not a universal constant; when incorporating heavy-tailed asset shocks, dynamic correlations, and realistic tax frictions, high portfolio survival probabilities achieved via dynamic spending often mask profound declines in utility for highly risk-averse retirees.

8. References

← All research articles How we build & check these tools