How the projection engine works

Ironlake's retirement and what-if projection is deterministic first. The planning frame is three explicit scenarios computed from assumptions you can inspect and change, shown as full year-by-year paths - not a single "probability of success." The goal is a defensible answer to one question: does the plan cover what you need to spend, under each set of assumptions? Alongside the three traces, the projection page offers an optional Monte Carlo stress view (described below) for seeing how wide the range around the plan is when real history is resampled.

The three scenarios

Every projection produces three traces from your model's per-class assumptions:

• Base - your assumptions exactly as entered.
• Conservative - expected returns reduced by a fixed amount and inflation raised, holding income yields constant.
• Stressed - a one-time equity-sleeve drawdown (default -30%) in an early "stress year" (default year 3); every subsequent year uses base returns. There is no modeled recovery curve - the portfolio recovers only through ordinary compounding afterward, not a special bounce.

Each trace is a year-by-year series: portfolio value, contributions and withdrawals, and the effect of any lump-sum events you add - up to the point of depletion (see the limitations below). Where a trace runs the portfolio to zero, the year of depletion is recorded as a failure marker so you can see not just whether but when a plan breaks under that scenario.

What the inputs are

The projection reads the return-related per-class assumptions on your Asset Allocation Model - the expected return, inflation expectation, and income yield - plus your scenario inputs (retirement year, spending, lump sums). Other model fields (the volatility band, tax character, and benchmark) are recorded for other surfaces and do not feed the projection, so changing them will not move the traces. Every input that does drive a result is visible on the model and the scenario, and changing one re-renders the traces.

Why deterministic first - and where Monte Carlo fits

A single "87% success" number invites false precision and a dual-trust failure mode - the user trusts a probability that is itself only as good as the return distribution assumed. Three transparent scenarios are easier to defend, easier to reason about, and align with the product's constructive, decision-support stance. The three traces are the planning frame.

Alongside them, the projection page offers an optional Monte Carlo stress view. It resamples real historical years - the same public dataset the historical replay uses (Shiller's US series plus the Bogleheads simba international series) - into one thousand alternative return sequences for your allocation, and shows the spread as percentile bands with the three deterministic traces overlaid inside the chart. Because whole years are drawn together, each path's cross-asset returns and inflation stay paired the way they actually occurred. The run is seeded and fixed-size, so the same inputs always produce the same bands, and there is still no probability-of-success gauge: the summary counts how many resampled paths sustained the plan, in plain language. Every assumption behind the view - the data range, the sampling rule, the trial count, the seed, and its limits (independent draws carry no momentum between years; non-equity classes approximate as US bonds; international data begins in 1970) - is disclosed next to the chart itself. It is a lens for seeing sequence risk and the width of the range around your plan, not a replacement for the plan.

Limitations (read these)

• It is illustrative, from your inputs - not a forecast and not a guarantee. Garbage in, garbage out: the traces are only as good as the assumptions you set.
• Returns within a trace are applied as smoothed assumptions, not a path of real historical sequences (the stressed trace adds one explicit drawdown, not a full sequence-of-returns simulation).
• Depletion is terminal within a trace. Once a trace's balance reaches zero, every remaining flow in that trace stops: no further income, contributions, or withdrawals, and a lump-sum event dated after the depletion year is not applied, so a later inflow (an inheritance, a sale) does not revive a failed trace. The engine does not model borrowing against a depleted portfolio. The projection and comparison surfaces note the lump sums this convention skips on their deterministic traces, and the historical replay and Monte Carlo stress view follow the same rule per cohort and per path.
• It is not advice. Ironlake shows the math so you can decide; it does not tell you whether to retire, save more, or change your allocation.
• Taxes on withdrawals are modeled at the planning level, not as a line-by-line return; for the precise after-tax treatment of income, see tax-character and after-tax math.

See it on your own model