Value Averaging vs Dollar-Cost Averaging — which actually wins?

How the math works

Edleson's 1988 Value Averaging targets a portfolio value path, not a fixed contribution. The target curve is Vt = $C × t × (1+R)t where $C is calibrated so the curve hits your target M at month T. Each month you buy (or sell) just enough to land on V_t.

DCA is the apples-to-apples comparison: a fixed monthly contribution D = M × R / ((1+R)T − 1) also calibrated to hit M at the expected return. Both strategies aim at the same M; the question is which one actually gets you closer when the realized path is volatile.

We simulate both strategies on five deterministic monthly-return paths shaped to mimic real regimes: steady bull (~10%/yr, low vol), volatile bull (tech-heavy), bear market (dotcom-style), sideways volatile (the 2000-2010 lost decade), and V-shape crash + recovery (2020-style).

The honesty correction: VA's peak monthly buy is the cost most blog write-ups skip. In a sharp drawdown VA can demand 3-5× the DCA contribution in a single month. If you don't have the cash sitting idle, you can't execute the strategy — and idle cash is its own drag on returns. We surface that number directly so the comparison is honest.

Math runs locally. Inputs never leave your browser. Source on github.

The six historical windows

The "Real history" tab gives you six well-documented market windows to test the strategies on:

• S&P 500 — Lost Decade (2000–2009): two crashes bookending a flat decade. VA's flagship win case.
• S&P 500 — Post-GFC Bull (2010–2019): decade-long uptrend. DCA's flagship win case.
• S&P 500 — COVID + Recovery (2020–2024): pandemic crash, instant recovery, 2022 inflation drawdown, 2023–24 AI rally.
• S&P 500 — Late-90s Tech Bull (1995–1999): five years of double-digit gains, no losing year. Pure bull benchmark.
• NASDAQ-100 — Dotcom Crash + Recovery (2000–2009): three years of −30%+ losses then violent recovery.
• NASDAQ-100 — Tech Renaissance (2010–2019): FANG-era tech bull with regular drawdowns.

What's real, what's synthesized: each year's total return is hardcoded from public data (Shiller dataset for S&P 500; Invesco QQQ fact sheets for NASDAQ-100). The intra-year monthly path is synthesized with deterministic sinusoidal variance and rescaled so each year's 12-month compound exactly matches the documented annual return. This preserves the headline numbers while giving VA enough monthly volatility to demonstrate the strategy difference.

The year-by-year card below the picker shows you the source numbers — verify them against any public dataset.

Where this framework breaks

• Tax drag: the simulation ignores taxes. Real VA generates capital-gains events on every "sell into strength" month — in a taxable account that gap can erase the on-paper VA edge entirely.
• Cash idle cost: VA only works if you keep reserve cash for those peak months. That reserve earns roughly nothing real, dragging the all-in IRR below the simulation's figure.
• Calibrated to expected R: if your assumed return is far from the realized return, the VA target curve becomes wrong and the comparison shifts. Treat the result as conditional on R, not absolute.
• Synthetic ≠ historical: these paths are sin/cos shapes designed to look like real regimes. They don't replace backtests on actual ETF data — that's a separate scenario family.

What to actually do

1. Pick the scenario closest to what you actually fear (volatile sideways or bear) and run it.
2. Look at VA's peak monthly buy — would you have that cash, in cash, ready to deploy?
3. Compare IRR, not just terminal value. VA can win on terminal while losing on IRR if it forced you to invest more.
4. If VA wins by a small margin and the peak buy stresses your liquidity, default to DCA. Discipline you can execute beats theory you can't.
5. Re-run with a real (inflation-adjusted) target if you care about purchasing power, not future face value.