Methodology

How the rolling-window engine works

One backtest engine drives DCA-vs-lump-sum win-rates, Coast success rates, and safe-withdrawal success rates. It steps through every historical start year in a 1928–2025 annual series — Damodaran's (NYU Stern) stock and 3-month T-bill returns with BLS CPI-U for inflation — compounding total return with dividends reinvested. Monthly granularity is not used. Honesty layers are toggles, not assumptions baked in silently: idle cash earns a T-bill credit, taxable accounts pay dividend and long-term-gain tax, expense ratios and any advice fee are deducted, and every path is deflated by CPI to show real dollars. The output is always a distribution — a win-rate plus a percentile band — never a single point. No calculator on this site runs Monte Carlo simulation; every percentile shown is computed from actual historical windows.

The ticker projection model — editorial parameters, disclosed

The ticker dividend calculators are forward projections, not backtests, so alongside each fund's sourced yield, expense ratio, and dividend-growth anchor they rest on a small set of editorial parameters — numbers chosen as assumptions, not numbers measured. Contract 2 requires saying so plainly. They are:

• Total-return ceiling — 8%/yr. For option-income funds (covered-call and index-option distributions), the model caps the assumed base-path total return (distribution yield + price growth − expense) at 8%. Any surplus is booked as an annual NAV drag, and the distribution is modeled as NAV-linked, shrinking if NAV shrinks. This is an ex-ante conservative assumption — calibrated so a full-overwrite fund like QYLD models about −3.8%/yr against a realized ≈ −2.7 to −3.5%/yr since 2013 — not a measured NAV forecast, and funds whose realized NAV has held up better than the cap will beat this model. Return-of-capital fractions are tax inputs only (they lower cost basis) and play no role in the NAV path.
• Scenario band spread — ±3pp price growth, ±2pp dividend growth. The low/base/high scenario band shifts both growth inputs by these editorial deltas; it is an illustrative spread sized to bracket realized multi-year variation, not a fitted distribution.
• Erosion band gain — 3.5×. Funds carrying a modeled NAV drag get a proportionally wider scenario band (their distributions are volatility-driven), scaled by this editorial factor.
• Dividend-growth taper — 10 years + 5-year decay. A fund's per-share dividend-growth anchor holds for 10 years, then decays linearly over 5 years to its price-growth rate, where the yield stabilizes. Without this, any above-price-growth anchor would compound into an impossible yield; with it, no projection assumes a payout can outgrow its own share price forever.

All four live as named constants in the open engine (assets/js/calc/ticker-dividend.js) and are asserted by the regression tests, so changing one changes this page's obligations too.

Data freshness and expiry

Year-dependent datasets (tax brackets, contribution limits, ACA figures, COLA) carry a validUntil date. Continuous-integration checks fail the build when today is past that date, so a figure never silently outlives its source. Each result page also stamps the as-of date of the data it used and the date the page was last reviewed. Where legislation is uncertain — for example the enhanced ACA subsidies scheduled to lapse after 2025 — the affected inputs are flagged rather than presented as settled.

Corrections

The contact page promises that substantive corrections are noted with a date rather than fixed silently. This is that record. It covers changes that moved a published number materially, not routine copy edits.

All three entries below are in the forward-looking fund projection model — the ticker calculators. The rolling-window backtest engine described above, which drives the withdrawal-rate, Coast, savings-rate and lump-sum tools, was not involved in any of them and its outputs did not change. The same commit did touch those backtest files, so it is worth being precise: those edits were a button label and a strategy name (“VPW” became “PMT (VPW-style)”), not a formula, a parameter, or a data file. Their regression tests assert the same hand-computed values before and after.

2026-07-22 — covered-call distribution model

The engine treated the return-of-capital portion of a covered-call ETF distribution as if it were destroying principal. Those are different things: a distribution can be classified as return of capital for tax purposes while the fund's net asset value holds. Because the two were conflated, forward projections applied an erosion drag that the funds' own measured history did not show — SPYI was being projected at roughly −6.9%/yr of NAV decay against a record that showed no such impairment. The erosion model is now separate from the tax treatment and is calibrated per fund. Every covered-call projection on the site changed.

2026-07-22 — dividend reinvestment compounding

A defect in the reinvestment loop compounded distributions on a balance that already included them. It was most visible at long horizons. On $25,000 up front plus $500 a month for 25 years — $175,000 paid in — the QQQI base projection came out at about $188.7M, roughly 1,078× the contributions, a figure the slider made visible to anyone who moved it. The same inputs now return about $553,000. The loop is fixed, and the regression suite asserts an implied internal rate of return at or below the model's own ceiling, so this class of runaway cannot ship again.

2026-07-22 — band labels and a phantom method claim

Forward projections were labelled p5/p95, which claims measured percentiles. They are three deterministic low/base/high paths, so they are now labelled as a scenario band. The historical backtesting engines do produce real percentiles and keep the percentile labels. Separately, this page previously described a Monte Carlo step that the engine does not perform; that sentence was removed.

These were found by an adversarial audit I ran against my own engine — simulated specialist critiques generated and then checked against the code, the datasets, and the funds' published records. It was not a review by an outside expert, and the site claims none.