Every Prometheus run is assembled from the models below, each calibrated to decades of real history and built to capture the skew and fat tails that ordinary bell-curve models miss. Here is the full roster, in plain English. For the end-to-end method, see the methodology.
Each tradable or economic driver gets its own model, calibrated to decades of history and built to capture the skew and fat tails that a plain bell curve misses.
A year of stock-market returns for the index a household holds.
An AR-GARCH process reproduces real markets, volatility that clusters, and fat, skewed crash-and-rally tails. The shock distribution is chosen as the best fit among several fat-tailed distributions using statistical tests (e.g. BIC). How deep the bad years go is learned from the long record of 18 countries’ stock markets.
In the methodology →How the government-bond yield curve can shift over a year, and what a household’s bonds return: their interest plus the price change.
Principal-component analysis distils the whole curve into three interpretable moves - overall level, steepness, and curvature - each driven by a fat-tailed distribution fit to history. The reconstructed curve shock revalues the household’s bonds as rates move, and the engine credits the interest they earn at the market yield for their maturity.
In the methodology →How market-implied inflation can move, and what it does to a household’s cost of living.
Two factors, level and slope, on the market-implied inflation curve, each with a fat-tailed distribution. The simulated one-year inflation level scales the household’s running costs in every path.
In the methodology →What a home could be worth a year from now.
A flexible fat-tail-aware distribution (EGB2) is fit to decades of regional house-price history, separating property’s excess return from the prevailing risk-free rate so the two move together as they do in reality. How far prices can fall is learned from many countries’ housing busts.
In the methodology →How a private-equity holding moves over a year, beyond what listed markets do.
PE is projected onto the engine’s simulated equity factor with a higher beta in down markets than up markets, plus an idiosyncratic skewed shock for write-down risk, then recentred so its long-run return matches published private-equity benchmarks.
In the methodology →How a foreign-currency exposure can swing over a year.
The same AR-GARCH machinery as equities, applied to exchange-rate moves, capturing the volatility clustering and fat, sometimes one-sided tails of currency swings, then simulating thousands of month-by-month paths on the household’s foreign holdings.
In the methodology →Two building blocks sit under the factor models: the distributions that give them their fat tails, and the historical record that teaches them how to move together.
The lopsided, fat-tailed shock distributions behind rates, inflation and property.
Rather than assume a bell curve, shocks are drawn from a fat-tailed distribution, for example an EGB2 or a Student’s t, with the family chosen per factor by how well it fits. It is why extreme moves aren’t quietly underestimated.
In the methodology →How markets, the economy and a household’s life events move together in a single run.
The engine is trained on more than fifty years of economic history across 18 countries, calm years and crisis years alike. Every simulated year learns from that record how shares, bonds, house prices, jobs, inflation and currencies behave together, so a market crash and a job loss can strike together about as often as history says, while leaving each factor’s own distribution exactly intact. For anyone who prefers to set the links by hand, a correlation matrix is available instead; if it is hand-edited, a PSD repair quietly snaps it to the nearest mathematically valid one so the simulation never breaks.
In the methodology →The operational layer turns plain-language risk estimates into loss distributions; the wealth engine runs everything together a million times.
One-off shocks - job loss, fraud, a major bill - and the insurance that offsets them.
Each risk takes two plain-language estimates, a typical bad event and a really bad one, and how often it happens. Those become a Monte-Carlo loss distribution (compound Poisson or Bernoulli for how often, lognormal or normal for how large). The household’s real policy limits and deductibles are then netted off, path by path, to leave the loss it would actually bear.
In the methodology →A household’s whole balance sheet, played out a million times over the year ahead.
On top of every factor model sits the wealth engine: one million one-year paths of the household’s finances - salary in (invested month by month via dollar-cost averaging), markets up or down, a possible shock - producing its full outcome distribution, its chance of ruin, its chance of simply ending behind, and its average worst case (expected shortfall).
In the methodology →