The "glide path" is the schedule of how your stock/bond mix changes over retirement. Target-date funds glide equity down; a well-known line of research argues for a "bond tent" that glides equity back up. This study answers the question directly for a 30-year retirement: we hand the whole 5-period schedule to QuantCalc's portfolio optimizer, let it maximize the modeled 30-year success rate under the production forward-looking model, and then re-score its recommendation — and four preset comparison schedules — on an independent Monte-Carlo stream with paired common random numbers, so every reported difference carries a standard error.
What shape wins: under QuantCalc's production forward-looking model, the optimizer's recommended glide path is rising, roughly flat (varies by withdrawal rate) in equity, and the head-to-head comparison points to a schedule shape that depends on the withdrawal rate. The full per-persona tables below give each schedule's re-scored 30-year success rate with its standard error, and every rising-vs-declining call is reported only when the paired difference exceeds twice its standard error — where it does not, we say so.
Why this study is different
QuantCalc already publishes two companion studies on withdrawal safety. This one adds the dimension neither covers: the allocation schedule itself.
The three personas
All three retire at 65 with $1,000,000, spend from the portfolio for 30 years (to age 95), and take no further contributions. They differ only by the initial withdrawal rate — 3.5%, 4%, 4.5% of the starting balance. The glide path has five contiguous 6-year periods (years 1–6, 7–12, 13–18, 19–24, 25–30); the optimizer is free to choose any equity weight in each, and the four comparison schedules are chosen in advance.
3.5% withdrawal ($2,917/mo)
The optimizer's recommended equity glide (per 6-year period) is 33/67/67/67/67% — rising from 33% to 67% equity. Re-scored independently, the rising-equity schedule wins by +1.35 pp (rising 85.9% vs declining 84.6%), which exceeds the ±2×SE noise band (±0.14 pp). Against the best preset strategy (Static 60/40, 86.1%), the optimizer's path lands -0.05 pp — inside the noise band (a statistical tie).
| Strategy | Equity by period (%) | Success (re-scored) | ±SE (pp) | Product-reported |
|---|---|---|---|---|
| Static 60/40 | 60 / 60 / 60 / 60 / 60 | 86.1% | ±0.11 | — |
| Optimizer (recommended) | 33 / 67 / 67 / 67 / 67 | 86.1% | ±0.11 | 85.9% |
| Rising glide (30→70) | 30 / 40 / 50 / 60 / 70 | 85.9% | ±0.10 | — |
| Static 40/60 | 40 / 40 / 40 / 40 / 40 | 85.2% | ±0.12 | — |
| Declining glide (70→30) | 70 / 60 / 50 / 40 / 30 | 84.6% | ±0.10 | — |
4% withdrawal ($3,333/mo)
The optimizer's recommended equity glide (per 6-year period) is 67/67/67/100/67% — flat at 67% equity with a single-period excursion to 100% in period 4 (an optimizer artifact within the noise band). Re-scored independently, the declining-equity schedule wins by +3.83 pp (rising 69.6% vs declining 73.4%), which exceeds the ±2×SE noise band (±0.10 pp). Against the best preset strategy (Static 60/40, 74.9%), the optimizer's path lands +0.99 pp — a difference beyond the noise band.
| Strategy | Equity by period (%) | Success (re-scored) | ±SE (pp) | Product-reported |
|---|---|---|---|---|
| Optimizer (recommended) | 67 / 67 / 67 / 100 / 67 | 75.9% | ±0.10 | 75.8% |
| Static 60/40 | 60 / 60 / 60 / 60 / 60 | 74.9% | ±0.10 | — |
| Declining glide (70→30) | 70 / 60 / 50 / 40 / 30 | 73.4% | ±0.11 | — |
| Rising glide (30→70) | 30 / 40 / 50 / 60 / 70 | 69.6% | ±0.12 | — |
| Static 40/60 | 40 / 40 / 40 / 40 / 40 | 68.9% | ±0.12 | — |
4.5% withdrawal ($3,750/mo)
The optimizer's recommended equity glide (per 6-year period) is 100/100/100/100/100% — constant 100% equity throughout. Re-scored independently, the declining-equity schedule wins by +11.05 pp (rising 49.6% vs declining 60.6%), which exceeds the ±2×SE noise band (±0.18 pp). Against the best preset strategy (Static 60/40, 61.7%), the optimizer's path lands +5.72 pp — a difference beyond the noise band.
| Strategy | Equity by period (%) | Success (re-scored) | ±SE (pp) | Product-reported |
|---|---|---|---|---|
| Optimizer (recommended) | 100 / 100 / 100 / 100 / 100 | 67.5% | ±0.09 | 67.4% |
| Static 60/40 | 60 / 60 / 60 / 60 / 60 | 61.7% | ±0.10 | — |
| Declining glide (70→30) | 70 / 60 / 50 / 40 / 30 | 60.6% | ±0.08 | — |
| Static 40/60 | 40 / 40 / 40 / 40 / 40 | 49.7% | ±0.09 | — |
| Rising glide (30→70) | 30 / 40 / 50 / 60 / 70 | 49.6% | ±0.10 | — |
How to read the tables. "Success (re-scored)" is the share of 160,000 independent Monte-Carlo paths (8 disjoint seeds × 20k paths) in which the plan funded all 30 years. "±SE" is the empirical standard error of that mean across the 8 seeds, in percentage points. "Product-reported" is the number QuantCalc's app shows for the optimizer's own path (its unbiased multi-seed out-of-sample estimate); the comparison strategies show a dash because the app does not report them. Because every candidate in a persona is scored on the same seeds and paths, the difference between any two rows is a paired estimate whose noise largely cancels — that is why a +0.20 pp-scale gap can still be inside the noise band.
What it means
- The schedule is model-dependent, not a law. These recommendations hold under QuantCalc's production forward-looking model (a blended capital-market expectation). A different set of return assumptions can move the preferred shape; this is a conditional result, not a guarantee.
- Why the model leans harder into equity as withdrawals rise. Success here is a binary funded/not-funded outcome, so once the baseline funded rate is low, adding return variance can raise it — a strained plan needs a strong draw to survive the full horizon, and equity supplies that dispersion. The model also draws returns from a Gaussian without fat tails or regime shifts, which understates deep-drawdown risk. Both effects push the recommended equity share up, which is why the schedule reaches full equity at the 4.5% draw: it is a property of this model — recommended under it, not a prescription for a real portfolio, where sequence and tail risk argue for holding some bonds.
- Differences are reported with their noise. Where the rising and declining schedules separate by more than twice their paired standard error, we call a winner; where they do not, we report a statistical tie rather than dress up noise as a finding.
- The weights are exactly what the product applies. The optimizer's per-period weights are rounded to integer percents before scoring, so the success rate shown is the rate of the allocation a user would actually run — not an un-implementable continuous optimum.
CC-BY-4.0 — free for any use including republication and journalism, with attribution to QuantCalc Research.
Find your own glide path
The optimizer in this study is the same one built into QuantCalc PRO. Enter your own balance, horizon, and spending, and it will search the full allocation schedule for the plan that maximizes your modeled success — then show you the year-by-year weights to apply.
Open the planner →Methodology
Model. Success rates are computed under QuantCalc's production forward-looking model — a 2-asset equity/bond blend of published capital-market expectations (equity 6.8% return / 17.0% vol, bonds 2.8% / 5.5%, correlation 0.15), the same forward-looking lens the hardened optimizer was validated against. Results are model-dependent and are not a guarantee of any future outcome.
Optimizer + cross-check. The recommended schedule comes from QuantCalc's glide-path optimizer (nlopt SLSQP over a Monte-Carlo objective, evaluated on a quasi-Monte-Carlo Sobol sequence with common random numbers across candidates). Its output was cross-checked against a brute-force grid search and independent re-simulation by the verification rig backend/tools/glide_verify.c, which enforces four invariants: weights sum to one and respect their box, the reported rate does not exceed an independent re-score (winner's-curse guard), the optimizer is not beaten by a coarse grid, and identical inputs give identical output.
Independent re-scoring. Every success rate on this page is an out-of-sample re-score from backend/tools/glide_study_dump.c: each candidate schedule is re-simulated on 8 disjoint SplitMix-derived pseudo-random seeds at 20k paths each — a generator entirely separate from the optimizer's QMC evaluation, so the scoring randomness is independent of the randomness the optimizer trained on. All candidates within a persona use the identical seeds and paths (common random numbers), so their pairwise differences are computed per-seed and reported with the standard error of the paired mean. A difference is called significant only when it exceeds twice that standard error.
Reproducibility & determinism. Persona 1's optimizer was run twice and returned bit-identical results (max weight difference 0, success-rate difference 0). The C dump is deterministic given the committed Sobol cache (QMC active, 10,000×4,096); re-running it reproduces the same JSONL, and re-running scripts/gen_glide_path_study.mjs over that JSONL reproduces this page byte-for-byte. As an in-generator guard, the generator recomputes each of persona 1's candidate means and standard errors directly from the raw per-seed rates and aborts on any mismatch, and confirms the optimizer's re-score is not beaten beyond twice the standard error by any preset schedule.
Not advice. This is reference research, not personalized investment advice. A glide path that maximizes a modeled success rate is not automatically right for your taxes, liquidity, or risk tolerance. Consult a qualified advisor before changing your allocation.
Changelog
- v2026.1 — initial release (2026-07-22). 3 personas × 5 strategies, 5-period glide path, independent 8-seed re-scoring at 20k paths, generated from
backend/tools/glide_study_dump.cviascripts/gen_glide_path_study.mjs.
Dataset license: CC-BY-4.0. QuantCalc is an independent retirement-planning research project. Model-dependent results under forward-looking capital-market assumptions; not financial, tax, or legal advice.