What does accepting risk change?
Compare your starting capital with the amount needed without investment risk for your setup's . Learn about .
The mortality fit is not ready. Check Investor & mortality if fitting fails.
Compare withdrawals, risks and investment strategies.
Checking for a saved setup…
Find a balance that feels right for your retirement.
The mortality fit is not ready. Check Investor & mortality if fitting fails.
Compare your starting capital with the amount needed without investment risk for your setup's . Learn about .
The mortality fit is not ready. Check Investor & mortality if fitting fails.
Compare the CE-model reference with the estimated risk under the fitted .
When projection is selected, a paired policy comparison will also appear here. Estimates depend on the model and are not known future risks.
See how the selected policy allocates your portfolio.
by wealth between the and the .
The mortality fit is not ready. Check Investor & mortality if fitting fails.
Explore possible outcomes and the borrowing exposure of the unconstrained policy.
Explore the chance that your portfolio and how long borrowing may last.
The mortality fit is not ready. Check Investor & mortality if fitting fails.
Where the model assumptions apply, > 1 indicates that the relative bandwidth exceeds the global no-borrowing threshold. Possibility does not imply borrowing on a particular path. The last possible time is a model boundary, not a predicted borrowing time.
Monte Carlo intervals describe sampling uncertainty only; they do not cover numerical approximation or horizon effects.
Borrowing time has not been calculated. The global feasibility diagnostic is analytical when available. The probability of ever borrowing is not calculated yet.
Set the market and investor assumptions behind your analysis.
Your inputs are saved locally in this browser. Fits and results are not saved automatically.
Define up to 10 assets and how they move together.
Market assumptions can affect the fitted . The fit updates automatically after changes; the underlying cohort data do not change.
Enter expected annual and as percentages. Growth is an annualized compounded rate. Import or edit assumptions directly; enter the and separately above.
The real risk-free rate is positive, the is positive definite, ρ > 0, and all components are non-negative.
These checks use the current market inputs. Mortality fit and policy-specific conditions are checked separately.Set the person and retirement context for your setup.
Weights the female and male survival curves; 0.50 gives an equal mixture.
Explore from your expected retirement age and the fitted . The compatible CE fit depends on market assumptions and updates automatically when valid inputs change. The cohort data themselves do not change.
Loading cohort data…
Compare variations of your setup and save your work.
Compare variations using the same market and investor settings. Export only the current table.
Compare withdrawals and risk across unconstrained and projected policies. All other settings are shared.
The mortality fit is not ready. Check Investor & mortality if fitting fails.
The CE value uses the unconstrained policy, including for projected rows. GM risk is for the selected policy, through the CE horizon.
| VARIATION | RISK OR WITHDRAWAL | POLICY | ANNUAL WITHDRAWAL | ESTIMATED GM RISK | GM STATUS | |
|---|---|---|---|---|---|---|
| — | — | NOT AVAILABLE | ||||
| — | — | NOT AVAILABLE |
Keep your setup and your results separately.
Save or restore your inputs, table variations and run settings. Fits and results are not included. Learn about .
Save the computed tables and diagnostics, linked to the setup and run that produced them.
Plain-language explanations, with technical detail when you want it.
Retirement can last many years, but no one knows exactly how long. At the same time, investments fluctuate and inflation changes what money can buy. A fixed rule for everyone cannot reflect each person's savings, spending and circumstances.
The question is how much income your savings might support while limiting the chance of falling below a level you consider unacceptable. That level need not be zero: you may want a reserve for essential or unexpected expenses, or hope to leave something to your loved ones. The threshold measures wealth falling too low during your lifetime; it does not guarantee an inheritance.
The simulator compares choices under a simplified model of markets and longevity. It can help you understand trade-offs, but it cannot predict your lifespan or future returns, or guarantee that your savings will last.
Open Market & investor and select Market. The example starts with two assets. For each, enter a name, expected annual growth and volatility; volatility describes how much returns fluctuate. Enter the nominal risk-free rate and your inflation assumption as annual percentages.
Under the asset list, open the correlation matrix to describe how assets move together. You may use up to ten assets. Alternatively, import a CSV with dates and adjusted closing prices already expressed in euros. The app estimates the market inputs and shows uncertainty for the imported data.
These are assumptions, not promises. Historical estimates depend on the data and period chosen and do not predict future returns. The market checks flag inputs that cannot be used by the model.
Select Investor & mortality. Enter your birth year and expected retirement year, then choose the mortality table and longevity scenario. The cohort information describes a population, not your personal lifespan.
Enter the savings available at retirement and a desired reserve below that amount. This is the wealth threshold used to assess your risk; for example, it could represent a buffer for essential or unexpected expenses. The simulator models the chance of reaching it while you are alive. It is not set aside separately or guaranteed. The smallest supported threshold is €1.
The mortality models fit automatically as relevant inputs change. You can review the survival chart and fit diagnostics on this tab; if a fit fails, check the inputs and retry.
Open Evaluation and Simulation settings. Choose a target risk to calculate an annual withdrawal, or enter a withdrawal to see its corresponding model risk. The amounts are in today's purchasing power.
The analytical CE relationship uses the unconstrained strategy. That strategy can involve borrowing or short selling. Turn on projection to compare a strategy that prevents both; the projection is a feasible comparison, not a claim of constrained optimality.
Select Run simulation to estimate risk under GM mortality and explore the available charts. Simulation settings control paths and time step. Monte Carlo intervals describe sampling uncertainty, not uncertainty in assumptions or all numerical errors; displayed GM results cover the simulation horizon.
Open Data & tables to compare variations using shared market, investor and simulation settings. Add rows, choose each row's inputs and compute them. Changing relevant settings can make earlier results need updating.
Export a setup to save editable inputs and table configurations. Export results separately; table and chart menus also offer downloads. Browser storage normally restores your inputs, but not previous simulation results. Imported historical price files and their diagnostics are not included in the setup export.
These labels explain the state of a result. The examples do not reflect an actual run.
Additional labels describe input validation and captured settings:
Fit and imported-data diagnostics also indicate whether their results match the current inputs. The feasibility box separately reports whether the strategy is feasible.
Showing 36 terms
Equations are rendered in the technical details of selected entries.
A wealth threshold you choose for assessing the risk of your savings running low.
The simulator estimates the risk of wealth reaching or falling below this amount while you are alive. It is not set aside separately or guaranteed; it could represent a buffer for expenses or an amount you hope to leave to others.
The paper calls this threshold the impoverishment level. The first-passage time is the first time wealth reaches or falls below it; simulation results cover the stated run horizon.
X is wealth; xᵢₘₚ is the threshold. The event of interest is reaching it before death.
In the research paper
Starting capital is the wealth available at retirement. A withdrawal is the amount taken from it to fund spending.
The model uses a constant annual withdrawal in today's purchasing power. Where regular income is included, the modeled withdrawal is net of that income.
The continuous-time model deducts a constant real withdrawal rate C from wealth; the displayed amount is in euros per year.
In the research paper
The model's wealth threshold for funding the chosen withdrawal without investment risk until the mortality endpoint.
It provides a benchmark for comparing capital when investment risk is allowed. This is a model result, not a guarantee about an actual product.
For the non-degenerate finite-horizon model with r > 0 and C > r xᵢₘₚ, the safe boundary is:
C is annual withdrawal, xᵢₘₚ the threshold, r the real risk-free rate and T the remaining-lifetime endpoint.
In the research paper
Capital relief compares the capital needed with risk-taking against the safe boundary for the same annual withdrawal.
At zero risk, capital above the safe minimum is shown as a surplus. The current comparison uses the unconstrained CE model.
The comparison holds C fixed. For positive selected risk, relief is the safe boundary at retirement minus the selected starting capital; nearby risk levels are varied relative to α.
In the research paper
Nominal amounts use money at the time; real amounts are adjusted for the chosen inflation assumption.
Inflation reduces what a fixed amount of money can buy. The app takes a nominal cash rate and displays the derived real model rate.
The model's real rate r uses effective nominal rate iᴺ and inflation ι:
The result is continuously compounded; the same inflation convention is applied consistently to asset inputs.
In the research paper
The model's assumed return on its risk-free asset, before or after inflation as labeled.
This is a model assumption. A real cash product may change its rate or fail to preserve purchasing power.
The current CE model requires a strictly positive continuously compounded real risk-free rate r.
In the research paper
An assumption about an asset's annual compounded nominal growth, before the selected inflation adjustment.
Historical estimates depend on the dates and data supplied and are not forecasts. The app's convention is not the same as the expected one-year percentage price return.
For annualized log growth a, the entered compounded rate is g = exp(a) − 1. The corresponding model's arithmetic drift also includes half the log-return variance.
In the research paper
A measure of how much an asset's modeled returns fluctuate.
A higher value means wider modeled fluctuations; it does not by itself predict a gain or loss.
The entered annual volatility is the annualized standard deviation of log returns, conventionally measured per square root of a year.
In the research paper
Correlation describes how two assets' modeled returns move together in a linear sense.
Positive values indicate a tendency to move together; negative values indicate opposite movements. Zero does not generally guarantee independence.
The diagonal is one and the matrix is symmetric. The app requires a positive-definite, sufficiently well-conditioned matrix; pairwise estimates can fail that requirement.
In the research paper
A table that combines each asset's volatility with how its return moves with the others.
It is used to model combined market fluctuations and construct portfolio weights.
With volatilities σᵢ and correlations Rᵢⱼ, the annual covariance is:
In the research paper
Historical closing prices adjusted for events such as splits and distributions.
Import prices that are already adjusted and expressed in euros. The app does not adjust, convert currencies or fill missing prices.
The CSV has ascending dates and one column per asset. Return intervals use available observations and elapsed calendar time.
A range describing uncertainty when estimating a market parameter from the supplied historical sample.
These intervals depend on the sample and model assumptions. They are not forecasts of future returns or Monte Carlo intervals for the simulation.
A 95% confidence procedure is designed to cover its fixed parameter in approximately 95% of repeated samples under its assumptions. Pairwise intervals use matching observation endpoints.
These are quantities calculated from the market assumptions; they help define the model's investment strategy.
The Merton vector is a direction, not a portfolio by itself. A negative component means the unconstrained policy sells that asset short in its risky region.
For excess-return vector θ and covariance matrix Σ, the direction m and risk premium ρ are:
The factor one-half is part of the model's definition of ρ. The compatible CE model requires ρ > 0.
In the research paper
A cohort groups people by birth year; a longevity scenario describes an assumed population mortality pattern.
The packaged German tables include observed and projected mortality. Low, medium and high scenarios are not predictions for an individual.
The app constructs a remaining-lifetime survival curve conditional on reaching the selected retirement age, using the chosen cohort and scenario.
In the research paper
Survival is the modeled chance of being alive at a later age, given survival to the selected retirement age.
The custom slider combines the female and male survival curves: 0 selects female, 1 selects male. It is a chosen mixture, not automatically a population-weighted series.
For remaining lifetime D, S(t) = P(D > t). Mixtures combine the two conditional survival curves after each has been constructed.
In the research paper
Fitting chooses model parameters so a survival model approximates the selected cohort data.
The app fits GM and market-compatible CE models automatically when relevant inputs change. Market assumptions can affect the CE fit; the underlying cohort data do not change.
The objective compares annual death-probability bins and the final survival tail using KL divergence. Optimizer convergence does not itself measure fit quality.
In the research paper
CE and GM are two mathematical descriptions of remaining lifetime used for different parts of the app.
The compactified-exponential (CE) model is fitted to remain compatible with the analytical policy. Gompertz–Makeham (GM) is a reference model used to evaluate simulated risk.
CE has a finite remaining-lifetime endpoint. GM has an age-increasing hazard plus a constant term and may extend beyond the CE simulation horizon.
In the research paper
Fit diagnostics describe how the fitted curves compare with the selected cohort data and how the optimizer behaved.
A smaller KL divergence is a closer match under that criterion, not a guarantee of predictive accuracy. Convergence means the optimizer stopped according to its criterion; it is not a quality grade.
KL compares annual death-probability bins and the final survival tail. Largest difference is the maximum absolute survival difference on the annual grid.
In the research paper
The CE endpoint and simulation horizon are model limits measured in years after retirement, not a known date of death.
GM mortality may extend past the CE horizon. GM risk reported by the simulation covers events only through the simulated horizon.
T denotes the deterministic right endpoint of the CE remaining-lifetime law; the random time of death is a different quantity. Cohort observations end at age 100.
In the research paper
The chance of wealth reaching the desired reserve before death that you choose when the CE model derives a withdrawal.
The analytical target applies to the unconstrained CE policy under its assumptions. GM mortality or the projected policy can produce a different risk.
For a positive target under the compatible unconstrained CE model, the derived policy satisfies:
α is the target probability, τᵢₘₚ the impoverishment time, and 𝒯 the random time of death. This α is not the GM hazard parameter also called α.
In the research paper
The estimated chance of wealth reaching the desired reserve before death, within the simulation horizon, under fitted Gompertz–Makeham mortality.
It can differ from the CE target because the mortality model or policy differs. It is a model-based estimate, not a known future risk.
Simulated first-hit times are weighted by GM survival through the CE run horizon. GM survival may extend beyond that horizon.
In the research paper
Percentage points measure the direct difference between two percentages.
A change from 5% to 6% is an increase of 1 percentage point, or 20% relative to 5%.
A 5% relative increase from 5% gives 5.25%, a change of 0.25 percentage points. The capital comparison uses relative changes in α.
A Monte Carlo estimate uses simulated paths; repeating the run with different random draws can change the result.
The standard error and approximate 95% interval describe sampling uncertainty, not all uncertainty about the future.
The interval does not include market-parameter or mortality-model uncertainty, time-step or bridge error, or events beyond the simulation horizon.
In the research paper
Paths are alternative simulated financial histories; the time step sets how often the model advances them.
More paths can reduce sampling noise. A smaller time step can change approximation error. Batch size groups paths for performance and progress; it is not a risk preference.
Cancellation and ETA updates occur between batches. Smaller time steps do not by themselves prove that all discretization error is negligible.
A random seed initializes the stream used to generate simulated draws.
The same seed alone does not guarantee identical results after changing path count, batch size, algorithm or numerical backend.
Table rows use separate derived streams. Paired policy simulations share market draws so their difference can be estimated from matched paths.
First passage is the first time a path reaches a chosen boundary.
A path can cross a boundary between saved time steps. The simulator uses a bridge approximation to estimate crossings that a grid alone could miss.
The bridge is approximate for the modeled moving boundaries. First borrowing and borrowing occupation use different approximations; their errors are not included in Monte Carlo intervals.
In the research paper
A portfolio policy describes how wealth is divided among assets as conditions change.
The allocation chart shows fractions of wealth at a chosen time and wealth level, not a forecast of one person's future holdings. Negative risky fractions mean short positions; a negative risk-free fraction means borrowing.
The unconstrained CE feedback depends on the safe gap, wealth and Merton direction. Its weights need not satisfy the no-short/no-borrow simplex.
In the research paper
A way of adjusting portfolio weights to satisfy the selected investment constraints.
Here projection prevents short positions and borrowing. It is a feasible comparison strategy, not a proven optimal constrained strategy.
The covariance-weighted projection of the unconstrained risky allocation π is:
p contains risky-asset weights, and Σ is the covariance matrix.
In the research paper
Global feasibility asks whether the policy avoids short positions and borrowing throughout its relevant risky region.
It is different from the chance that a particular path borrows. Under the stated Merton-vector assumptions, ζ > 1 means borrowing is possible somewhere, not inevitable.
For non-negative Merton components with M = Σᵢmᵢ > 0, the unconstrained CE policy is feasible when ζ ≤ 1:
The action at ζ = 1 changes the desired reserve while keeping the selected target input fixed. The projected policy is feasible by construction.
In the research paper
Borrowing occurs when the strategy needs a negative holding in the risk-free asset to fund its risky positions.
Ever-borrowing probability concerns borrowing while alive before wealth reaches the reserve or the simulation ends. Expected borrowing time averages over everyone, including those who never borrow.
The measures use GM survival weighting. First-borrowing detection and occupation time use different numerical approximations and can differ path by path.
In the research paper
The timing chart shows the risk of wealth reaching the desired reserve through the simulation horizon.
Forecast assesses that risk at retirement. Conditional update assumes the person is alive and has not reached the threshold by the selected age. It does not use observed current wealth.
At a selected time a and horizon H, the forecast is P(a < τ ≤ H, τ < D); the update conditions on τ > a and D > a. Later points retain that same conditioning time.
In the research paper
Remaining risk is the modeled chance of wealth reaching the reserve while alive during the remaining horizon.
The curve falls toward zero at the horizon. Density bars show risk per year; an area across a range of ages represents a probability contribution.
The displayed survival-weighted risk is not normalized to 100%. A density at one exact age is not itself the probability of an event at that age.
In the research paper
A mathematical approximation to when wealth reaches the reserve under the unconstrained policy.
The optional overlay is additional technical information; it is not an exact simulation result.
The displayed CE total-variation bound concerns timing conditional on reaching the reserve before death under CE mortality. It is not a bound for the GM chart or the age-conditional update. The empirical CDF distance has sampling and numerical errors.
In the research paper
Sample paths show possible modeled wealth histories, not forecasts of a particular person's future.
They assume the person remains alive and stop when wealth reaches the reserve or at the policy horizon. Shuffle changes the displayed sample without rerunning the simulation.
Only a bounded sample of paths and a reduced plotting grid are retained; short fluctuations can be omitted from the display.
In the research paper
Each table row changes the listed inputs while using shared market, mortality and numerical settings.
A run captures those settings. Changing an input afterward can leave an earlier result needing an update; the row name is only a label.
The withdrawal table uses unconstrained CE values and simulates GM risk for the selected policy. The feasibility table concerns unconstrained borrowing under its applicable assumptions.
A setup contains inputs; results are the estimates obtained by running calculations.
Editable inputs are normally saved in this browser. Fits and simulation results are not restored after reload. Importing a setup replaces inputs and clears old results.
Setup JSON, results JSON, table CSV and chart files serve different purposes. Original historical price files and all simulated paths are not included in those summaries.