Monte Carlo or historical backtest?

A deterministic projection - one line, compounding at your expected return every single year - is the right place to start. It is legible: change a contribution and you can see exactly what it did. But markets do not return their average every year, and a plan that only works if they do is not a plan.

retplan gives you two ways to stress that line, and they disagree often enough to be interesting.

Monte Carlo: thousands of futures that never happened

Turn on Monte Carlo and every account's yearly return is drawn at random from its expected return ± volatility - thousands of times over. Instead of one line you get a band of outcomes and a success rate: the share of simulated futures in which the money lasts.

• It explores the tails. A single projection cannot show you the unlucky decade; ten thousand of them can.
• It works for any assumptions. A portfolio mix that never existed historically - or return expectations lower than the past - is no problem, because you set the inputs.
• But the futures are synthetic. Each year is an independent draw. Real markets trend, crash, and mean-revert; random draws do none of that, and the answer is only as good as the return and volatility you feed it.

Backtest: the futures that actually happened

The backtest replays your plan against every rolling window of real US market history since 1928. Retire into 1929. Retire into 1966. Retire into every other year in the record, and count how many of those retirements survive.

• The sequences are real. Crashes come with recoveries, inflation spikes persist for years, and sequence-of-returns risk - the bad decade that arrives right as you retire - is built into the data instead of assumed away.
• The worst case has a name. Instead of "the 5th percentile", the result is "your plan fails if you retire in 1966" - which is a much easier thing to reason about.
• But history is one sample. A 30-year horizon gives fewer than seventy heavily overlapping windows, all drawn from the single best-performing stock market of the twentieth century. The past is evidence, not a guarantee.

Which should you trust?

Both, and neither alone. If your plan survives ten thousand randomized futures and every depression, war, and stagflation in the historical record, it is about as robust as a spreadsheet can tell you. When the two methods disagree, that gap is information - usually a sign your expected-return assumptions are far from the historical record, and worth a closer look.

A projection is not a prediction. The point of running many of them is to find the plans that fail in most futures before you commit to one.

Both modes run on your device, on your numbers, with the planner - flip them on from the Settings menu and compare.