Run and understand Stochastic Analysis

Compare the same five stock returns in reverse order

Open Simulation Parameters and choose an Order-of-returns comparison: Moderate, Severe, Extreme, or Custom. Review the five yearly stock returns and their exact reverse. Presets are illustrative, not forecasts; Custom preserves your entered order. Both cases use fixed bond and cash returns for five years, initially 2% and 3%.

Save & run analysis adds two matched sets of trials. Review the Sequence risk result tab for success, downside balances, drawdowns, spending cuts, and shortfall timing. Matching inflation, expense shocks, and later market paths makes return order the deliberate difference. Selected spending responses and guardrails still affect the outcomes.

Sustained early-market stress remains the separate repeated-return test that changes the primary analysis. The paired comparison uses its displayed return orders instead of that repeated-return setting. The existing Bad first 5 years sensitivity still tests a sustained downturn. Read Help on comparing the order of early retirement returns for the complete procedure and limits.

Stochastic Analysis asks how the same plan behaves under uncertainty

Stochastic Analysis runs one saved retirement plan through many generated paths for markets, inflation, and any other selected uncertainties. The paths do not predict which future will occur. Together they show how sensitive the plan is to sequence, volatility, inflation, longevity, spending flexibility, and major modeled shocks.

The analysis begins with the same household cash-flow mechanics used by Projection. Taxes, income, spending, contributions, withdrawals, and account restrictions still matter in every path. A stochastic run cannot rescue a baseline whose dates, balances, or cash flows are wrong.

For a first run, keep the standard model and understand what is already active

The ordinary starting point is Forward assumptions, YARCalc’s plain-language name for Monte Carlo, with the Standard model, pseudo-random sampling, the household’s saved lifespans, a locked scenario-specific seed, and no sustained early-market stress. Subscriber access starts at Standard precision with 10,000 primary paths. Free access starts at Fast precision with 2,500 paths. These settings are a useful baseline, not a claim that their market assumptions fit every household.

The stock, bond, cash, and inflation means and volatilities are active in every standard Forward assumptions path. Stock mean starts at 7%, bond mean at 4%, and cash mean at 3%; the stock mean does not inherit the deterministic Expected portfolio growth assumption. Healthcare shocks start with a 5% annual chance and major-expense shocks with a 3% annual chance. Bad-market response and guardrail spending cuts start off.

Subscriber access starts with eight 1,000-path sensitivities on, making the Standard default workload 18,000 paths. Model comparison, suggested-target comparison, allocation comparison, transition schedules, and paired sequence comparison start off for every access level. Free does not run sensitivities. Change one optional feature at a time when you need to explain why a result moved.

Reset to Defaults changes only the working values in the open dialog. Cancel restores the values from before the dialog opened. Apply for next run keeps the working values as the current unsaved choices. Save & run analysis saves the selected parameters and starts the background job.

Saved analyses can be renamed or deleted one at a time. Delete asks for confirmation, removes only the selected completed result, and does not change the parameters for the next run.

Use Historical baseline when you want a neutral historical starting point

Historical baseline is a named, draft-only preset. It selects Historical experience, five-year contiguous blocks, and the scenario’s saved lifespans. It keeps the selected precision and does not alter account balances, scheduled expenses, or the scenario allocation.

The preset turns off sustained early-market stress years, random healthcare and major-expense shocks, bad-market spending response, and guardrail spending cuts. It does not delete those saved values or change the scenario. Review the before-and-after draft summary, then choose Apply for next run or Save & run analysis.

Changing a preset-owned setting labels the draft Historical baseline — customized. That label is saved with the effective controls so a later result does not imply that the neutral preset remained unchanged.

Check the evidence behind the modeled allocation

The Modeled allocation card shows the market-linked stock, bond, and cash mix used by the analysis. It also reports how much portfolio value has an explicit asset class and how much is estimated, defaulted, unclassified, or modeled with contractual return rules. Classification confidence describes the evidence behind the mix; it does not judge whether the mix is suitable.

Review recommended begins when more than 10% of portfolio value is estimated or defaulted. Low confidence begins above 25%, when material value is unclassified, when a saved asset class has no recorded classification source, or when holdings materially fail to reconcile to account balances. Reviewing and saving an older holding records its explicit classification. A genuinely stock-heavy portfolio can still have High confidence while receiving a separate concentration warning.

Every newly completed result saves its allocation and classification snapshot. An older saved result without this evidence asks for one rerun instead of describing current holdings as though they were used by the older analysis. Individual bonds and fixed-rate holdings are reported separately because contractual rules replace stock, bond, or cash market returns.

Precision, randomness, and longevity determine the comparison frame

Choose Fast for 2,500 pseudo-random paths, Standard for 10,000, or High precision for 50,000. Free includes Fast; Subscriber includes all three. Higher precision makes percentiles and small differences less noisy, but it does not improve unrealistic assumptions. High precision uses the full allowance for one configuration and cannot be combined with sensitivities, model comparison, suggested-target comparison, allocation comparison, or paired sequence comparison. The live workload summary shows the primary and added paths before you run.

The default locked seed is derived from the saved scenario, so reopening the same scenario reproduces its comparison frame without giving every plan the same seed. Keep it locked when comparing a specific change. Choose New seed every run when the purpose is to see whether a conclusion survives another sample. Seed is retained as technical run provenance, not treated as a Home-versus-What-if planning difference, and a locked seed does not make that sample more probable or more accurate.

Use saved lifespans honors each person’s death or planning-end date from Personal Details. Minimum-age stress raises every household member to at least the entered age; it does not end a two-person household when the first person reaches that age. The preview states which person is extended. Use this only for an explicit longevity question, not as an unexplained replacement for the saved household assumptions.

Background work changes waiting time, not the planning assumptions

YARCalc runs Stochastic Analysis as a saved background job so the planning pages remain usable while thousands of paths are calculated. Runtime depends on precision, plan length, selected model, optional comparisons, current demand, and available processing capacity. After Save & run analysis, YARCalc commits the queued job before calculation begins and shows waiting, preparation, and simulation stages in the existing page without reloading the document. You may navigate elsewhere and return to the same saved plan; its active progress is restored in place. Planning pages intentionally provide no job-cancellation control; operations handles exceptional cancellations separately.

Repeating Save & run analysis for the same scenario while its job is active returns to that saved status instead of creating a duplicate. A probability sweep currently reserves all analysis capacity so its many per-value simulations stay bounded and reproducible. If another analysis is queued behind one, the progress detail names that exclusive sweep instead of implying that an ordinary slot is unexpectedly unavailable. A cancelled queued sweep is removed immediately; a running sweep can continue to occupy capacity briefly while its workers reach a safe stopping point.

Variable Sweeps first shows Submitting while the browser sends the request. Once accepted, the job reports Queued or Validating saved plan and requested range. Range-dependent workload checks run inside the accepted job, so loading the saved scenario no longer leaves the page claiming that simulations have already started. If that check fails, the job shows the exact runs-per-value limit for the selected range.

The amount of processing capacity assigned to the job is not a user parameter. Changing that operational capacity should not change a seeded mathematical result; it changes how the same work is divided and may change how long it takes. If a comparison must be reproducible, preserve the seed, saved plan, model, precision preset, effective path count, and all visible assumptions rather than trying to control processing resources.

Apply for next run changes only the parameter draft. Save & run analysis saves those values with the result and starts the job. It does not copy stochastic market assumptions into Planning Assumptions or alter the deterministic Projection return. To test a plan change such as retiring later, edit a What-if or use Variable Sweeps and then run the analysis against that saved alternative.

When a job completes, Saved analyses immediately includes the result alongside up to the 50 most recent same-version runs, even when several use the same market model. Enter an optional analysis name before running or change it later without recalculating. Run times display in your browser’s timezone, while Balances as of remains the saved calendar date. An exact match reuses the existing completed calculation and identifies that reuse.

Market Assumptions describe nominal arithmetic returns and annual variation

Stock mean begins at 7% and stock volatility at 16%. Bond mean and volatility begin at 4% and 7%. Cash mean and volatility begin at 3% and 1%. Inflation mean begins with the scenario inflation estimate when present, otherwise 2.5%, and inflation volatility begins at 2%. These are starting assumptions, not forecasts calibrated to the household’s holdings. The stochastic stock mean is deliberately separate from deterministic Expected portfolio growth.

Mean is the average nominal annual arithmetic return used by the path generator. Volatility describes year-to-year dispersion around that mean. Raising a mean generally improves results; lowering volatility generally reduces dispersion. Either edit can make a plan look safer without making it more realistic, so change a value only when you can state the alternative assumption being tested.

The deterministic Projection return is separate from these stochastic asset-class means. With zero volatility, stocks, bonds, cash, and inflation each follow their own entered mean exactly. Matching only the stock mean does not make Projection and Stochastic Analysis equivalent when bond, cash, inflation, spending, shock, or lifespan assumptions differ.

Distribution and correlation controls change the shape of generated market stress

The Standard model uses correlated normal annual shocks. Fat-tail stress uses a standardized Student-t distribution so extreme annual outcomes occur more often while preserving the configured means, volatilities, and correlations. Tail severity starts at 5 degrees of freedom. Lower values create heavier tails; values near 30 increasingly resemble the Standard model. Use fat-tail stress as a comparison against the same plan and seed, not as a more certain forecast.

Stock/bond correlation starts at 0.20. Stock/inflation starts at negative 0.10, bond/inflation at negative 0.30, and cash/inflation at 0.40. Positive correlation means shocks tend to move together; negative correlation means they tend to offset. Correlations can change during crises, so avoid tuning the matrix to create a preferred result. Invalid combinations that cannot form a usable correlation structure are rejected.

Spending responses and shocks can materially change what “success” requires

Non-discretionary and discretionary totals are read-only summaries of the saved scenario categories. The simulation continues to use the scenario’s scheduled spending in each modeled year rather than replacing it with two flat stochastic amounts. Those classifications matter because bad-market response and guardrails cut only discretionary scenario spending. Correct the scenario when an amount or classification is wrong; an incorrect classification can make resilience appear stronger than the household could actually tolerate.

Bad-market response and guardrails start off. When bad-market response is enabled, its negative 10% stock-return threshold triggers a 25% discretionary cut in the following year. When guardrails are enabled, the starting settings trigger at a 20% portfolio drawdown, cut discretionary spending by 15%, and cap the cumulative discretionary reduction at 50%. A plan that succeeds only through frequent cuts may be mathematically successful but behaviorally unrealistic.

Healthcare shocks start with a 5% annual chance and a $10,000 to $50,000 amount range. Major-expense shocks start with a 3% annual chance and a $50,000 amount. Each annual draw is independent, so a shock can occur more than once over a long plan. Set a chance to zero to turn that shock off, and avoid counting the same known expense both as a scheduled scenario cost and as a random shock.

Sustained early-market stress starts off because Years of repeated weak returns is zero. When enabled, it overrides stock, bond, and cash returns beginning at retirement for the chosen number of years. The starting stress values are negative 8% stocks, 2% bonds, and 3% cash. This is a deterministic sequence-risk stress layered onto each run, so label it clearly and compare it with the ordinary baseline.

Optional comparisons answer separate questions and add substantial workload

Analysis scope is a primary choice: Run one market model uses the selected method, while Compare market models runs Forward assumptions and Historical experience as peers. The setup shows paths per model, total comparison workload, and the historical block size before the run. Run sensitivities adds eight 1,000-path stress cases. Test suggested planning targets runs another comparison using the displayed suggested settings. Sensitivities start on for Subscriber access; the other comparisons start off. High precision cannot be combined with any of these workload multipliers.

The Overview places both success probabilities, conservative and median ending balances, typical maximum drawdown, and median first-shortfall timing side by side. The detailed Model comparison tab remains available. YARCalc retains a compatibility fingerprint covering the saved plan version, balance date, strategy, horizon, path count, seed policy, spending and market stresses, response rules, allocation evidence, historical endpoint, and block size. If a matching counterpart is not retained, the page asks for a comparison run instead of combining unlike saved results.

Below the comparison, YARCalc identifies three evidence-backed factors that likely contributed: configured returns versus the available history, independent synthetic years versus contiguous historical blocks, and the plan’s exposure to return order through allocation, withdrawals, taxes, inflation, and active response rules. These are sensitivity explanations, not claims that one factor caused the result or that either method predicts the future.

Current versus suggested policy shows success probability for each policy. It omits Probability of ruin because that value is exactly the complement of success and would repeat the same information.

Retirement allocation comparison starts off. For Free access, first enabling it compares Current with 60/40, for 5,000 primary paths at Fast precision. For Subscriber access, it compares Current with 70/30, 60/40, and 50/50, for 40,000 primary paths at Standard precision. The 80/20 and 40/60 presets, a custom static allocation, and a transition schedule remain optional. Uncheck strategies that do not represent a useful decision because every selected strategy adds another full set of paths.

A custom static allocation uses the entered stock percentage and treats fixed income as the remainder. A transition schedule can move immediately, at retirement, gradually between years, through a bond-tent or rising-equity path, or through a custom YEAR:STOCK% schedule. The ordinary transition target starts at 60% stocks; the bond-tent example starts at 50% stocks at retirement, 60% five years later, and 70% ten years later. These are examples to edit, not recommended allocations.

Historical experience and Economic regimes answer different advanced questions

Historical experience is the plain-language name for historical bootstrap. It samples contiguous blocks from the supported historical stock, bond, cash, and inflation series through the plan’s saved historical end year. Block size starts at five years. Larger blocks preserve longer historical sequences; smaller blocks reshuffle history more aggressively. The model can reproduce relationships that occurred in the available common source period, but it cannot create a future event absent from that history or prove the source period is representative. The Stochastic dialog does not provide a separate start-year selector.

Economic regimes is the plain-language name for the two-state Markov chain. It begins in Expansion and allows Expansion and Recession conditions to continue from one year to the next. Expansion starts with 8% mean return, 8% return volatility, scenario inflation or 2.5% mean inflation, and 0.8% inflation volatility. Recession starts with negative 10% mean return, 12% return volatility, 1.5% mean inflation, and 1.2% inflation volatility. The annual transition chances start at 18% from Expansion to Recession and 55% from Recession to Expansion.

Economic regimes uses a single regime return rather than separate stock and fixed-income return processes, so allocation comparison is unavailable in that mode. Change regime values only for a documented stress question. A more complicated model can expose multi-year market conditions; it does not become a calibrated forecast merely because it has more parameters.

Read the distribution, not only the success percentage

Success probability is the share of generated paths that meet the selected success definition. It is conditional on the model, assumptions, horizon, and spending rules. It is not a confidence score and it is not the real-world probability that retirement will succeed.

P50 is the middle outcome. P10 is a downside percentile: ten percent of paths finish below it and ninety percent above it. P90 or P95 is an unusually favorable outcome, not a conservative estimate. Percentile lines can join values from different paths at different ages, so they are distribution summaries rather than biographies of one simulated household.

Today’s dollars is the starting display for a newly completed result and expresses amounts in the purchasing power of the analysis start year. YARCalc divides each path’s value in each year by that same path’s cumulative inflation through that year, then calculates the displayed percentiles. It does not divide a nominal percentile by one average inflation rate. Future dollars shows the original nominal amounts.

The dollar switch updates ending-balance cards, annual paths, confidence ranges, detailed tables, allocation and model comparisons, sensitivities, suggested-policy results, sequence comparisons, and the exported report without rerunning the analysis. Success probability and percentage drawdowns do not change because their definitions are independent of the display-dollar basis. Older saved results that did not retain path-specific inflation-adjusted values remain in Future dollars and request one rerun rather than estimating missing results.

Use saved lifespans preserves every person’s scenario death/planning age. Minimum-age stress raises every household member to at least the chosen age, while Annual Projection rows identify a person as deceased after the saved lifespan rather than continuing to increase that age.

Read first-shortfall years, depletion timing, drawdowns, spending cuts, and sensitivities beside ending balances. Early failures often point toward sequence risk; late failures may point toward longevity, inflation, or an ambitious legacy target. Investigate whether an apparent improvement comes from a realistic tradeoff or merely a more favorable assumption.

On screen, result tabs separate the base Overview from Allocation, Model comparison, Sensitivities, Planning actions, and Run details. Only completed result families appear. Export PDF ignores the selected tab and prints every available result section in one ordered report.

A valid comparison preserves the plan, valuation date, and assumptions

Before attributing a result change to a planning decision, confirm the saved plan, valuation date, strategy, lifespan mode, precision, effective path count, model, and readable assumption comparison. YARCalc retains seed and fingerprints internally for reproducibility and matching, but does not present those technical identifiers as planning differences.

Long analyses run in the background. Progress and the completed result belong to the saved plan submitted with the analysis. You may leave and return, but a later plan edit does not update the saved result. Rerun after a material plan change. If operations stops the work, the planning page reports that the analysis ended without exposing a cancellation control.

When a difference is small, repeat the comparison with the same assumptions and another seed or a higher available precision. If the ranking changes easily, the evidence is not strong enough to support a precise conclusion.

Some familiar retirement-simulation terms are not controls on this page

YARCalc does not label a result as a personal “ruin age.” It reports shortfall and depletion timing among generated paths. Those dates describe when modeled portfolio funding became insufficient under the entered assumptions; they do not predict a person’s death, health, or inevitable financial outcome. The result also summarizes percentile values rather than exposing a complete year-by-year diary for every failed path.

Named market-conditioned episodes such as “1970s stagflation” or “2008 crisis” are not selectable scenarios here. Supported alternatives are Historical experience, Economic regimes, fat-tail stress, and sustained early-market stress with explicit return inputs. Run Details retains the technical model names. Use the closest supported method only when it answers the intended question; do not treat a custom return sequence as a verified reconstruction of a named event.

The 4% rule is a withdrawal-rule reference, not a Stochastic parameter or a YARCalc recommendation. The result may display an initial withdrawal rate, but success still depends on the household’s actual spending path, taxes, income, horizon, allocation, inflation, and responses to poor markets. Test a spending policy in a separate What-if instead of assuming one historical rule applies unchanged.

Access limits can change the available precision and comparison workload shown to an account. The ordinary baseline remains understandable at the available precision; a higher limit adds sampling or comparison capacity, not a more favorable model.

Generated paths are structured stress tests, not calibrated certainty

Every stochastic model simplifies markets, inflation, behavior, law, health, and household decisions. Historical data may not describe the future, correlations can change during crises, and rare events cannot be assigned a trustworthy probability merely by choosing a more complicated distribution.

Use the analysis to reveal sensitivity and compare resilience. Do not interpret a high success percentage as a guarantee, a low percentage as a verdict, or a tiny difference as meaningful without checking sampling stability and assumptions.