8 ms·

Show HN: I made a retirement simulator: 125 years of data, 25 countries

16 points by nhootan 21d ago

nhootan 21d ago
I got tired of retirement calculators that either assume one fixed return forever, or run Monte Carlo by pulling each year independently out of a bell curve. Real markets don't behave like that. Bad years cluster, and a crash in your first three years of retirement is a completely different problem from the same crash at 85. So this one samples multi-year blocks out of actual history instead: 25 countries, 1900 to 2025, reassembled into thousands of possible futures. It's a stationary block bootstrap, which is the standard approach in the lifecycle-investing literature. When a plan fails it tells you which way it failed, whether that was weak growth, bad sequencing, or inflation at the wrong moment. Canada and the US have full tax and benefit modelling (RRSP/TFSA and CPP/OAS on one side, 401(k)/IRA and Social Security on the other). Everywhere else you get a pre-tax projection. If you're in the EU or UK you can run the Quick Check, but you can't sign up yet, because I haven't finished the compliance work. There's no account linking, and the Quick Check needs no signup at all. The methodology is written up at /methodology. It's an educational tool, not financial advice. It's free right now. Paid plans come later, and there's a founding rate if you want to lock one in early: $50/year for as long as you stay subscribed, first 50 people. https://buy.stripe.com/14A7sN6Gycny54rdXv5Rm01 https://buy.stripe.com/14A7sN6Gycny54rdXv5Rm01 Nothing is gated behind it today.
drop_star 21d ago
Why do I need to sign into everything these days? The moment I see that I close the site.
- albumen 21d ago
  You don’t need to for its quick check, which shows the core feature.
  - rconti 21d ago
    yeah, this is quite literally the first thing i saw when i opened the page.
  - bluGill 21d ago
    You need it to dig deeper. I have a pension which I can get at 65. I have options to collect social security. What if I draw my 401k first and wait until 72 to start as? Or I can get take as at 62 and the 401k grows more... Edit point is you need an account to start asking those questions.
    - nhootan 21d ago
      The timing half is there: CPP/OAS or Social Security start ages, and pension start and end ages, so claiming at 62 versus deferring is something you can run both ways today. Draw order across accounts isn't exposed yet. Right now the engine allocates by equalizing marginal tax rates across accounts each year rather than draining one at a time, so it's solving that ordering instead of letting you set it. The manual override is a fair ask and it's coming, the engine already takes the order as a parameter and it just isn't wired to the UI.
- nhootan 21d ago
  [dead]
- nhootan 21d ago
  The sign in is required for detailed plans to save the plans. Those plans also incure real comupte on the server. QC results are based on a cache in the browser and don't need sign in.
memoryleakgame 21d ago
So its a monte carlo simulation but I can't enter growth rate country time horizons or anything like that? Brother I have a better claude code vibe app than this in an hour You need to really talk to a financial planner and learn what the proper inputs are
- nhootan 21d ago
  [dead]
- nhootan 21d ago
  [dead]
bluGill 21d ago
This doesn't know the important part. Will I live to 62 as the earliest of my ancestors died, or 98 as the latest? My family history is about average for lifespan but that is a big difference. If I live to 62 I should retire at 50 - or perhaps I shouldn't have bothered saving at all since the growth isn't there. Of medical advances make me love to 150 I should save until I'm 80 which will be a nice income to really enjoy the rest of my life. I want more defined benefit accounts, but they are hard to find
- karmakurtisaani 21d ago
  So the thing is that no one knows that part.
  - Eddy_Viscosity2 21d ago
    Can't we just get AI to tell us? I'm sure it will confidently give an answer.
    - bluGill 21d ago
      The latest AI is a lot better at not being over confident. There is still an issue but not as good as it used to be
      - karmakurtisaani 21d ago
        But how can you trust anything it says if it doesn't have confidence? I tell even my children that the most important aspect of applying a job is a confident, firm handshake.
  - bluGill 21d ago
    For me no, but for a few thousand people it isn't to hard. Which is why I look for defined benefit plans that are not scams, and in frustrated that I can't fibd them
- nhootan 21d ago
betaby 21d ago
Better be safe I guess: Foresight Planner is not yet available in your region. We're launching in Canada (excluding Quebec) first. We don't yet support Quebec, because we don't yet offer the product in French. If you'd like us to let you know when we expand, email hi@foresightplanner.com. Foresight Planner n'est pas encore disponible dans votre région. Nous lançons d'abord au Canada (hors Québec). Pour être informé·e de l'élargissement de notre service, écrivez à hi@foresightplanner.com. Region: CA-QC
- nhootan 21d ago
  It's Bill 96 and Law 25. No French UI yet, so blocking beats being non compliant. It's near the top of the list.
jawns 21d ago
I got 82.4% regardless of whether I started with a portfolio value of $1M or $3M. Something is wrong with the calculations/simulations. If I can't trust the Quick Check, why would I trust anything else on the site?
- Eric714 21d ago
  Same here
- Eric714 21d ago
  That's because the withdrawals are 4%. If you have $3M, maybe lower the withdrawals to 1.25%. That would be the same as 4% of $1M.
  - jawns 21d ago
    I don't think that's accurate. If I withdraw 4% of a $1M portfolio, I only have $960K left to compound. If I withdraw 4% of a $3M portfolio, I should have $2.88M left to compound. If the question is "Will your money last?" I would not expect the simulation results to be identical. Edit: I failed here. Downvotes are right. I simulated it myself and saw the results. Leaving this comment here so others can learn from my mistake.
- nhootan 21d ago
  That result is right. The Quick Check takes a withdrawal rate, not a dollar amount, and runs with taxes off, so it's scale invariant: $1M at 4% and $3M at 4% are the same simulation times three. That's what lets it be instant and anonymous, since the whole answer space fits in a small grid I ship to your browser. It stops being true once taxes are on, which is what the full builder does. The page should say so. I'll add a line.
BretonForearm 21d ago
Why not let people try it from abroad? Not in the USA? No luck. Why does the withdrawal rate need to be a multiple of 0.5%? It's curious that the highest chance was associated with the highest (100%) stock ratio.
- pelagic_sky 21d ago
  I'm in NZ and it worked. It gave me US figures and numbers though. And I find there are existing FIRE calculators already out there that are tried and true.
  - nhootan 21d ago
    Glad it ran, but the numbers aren't US. Once you login, "Other, rest of world" gives you a gross projection with tax and benefits off. The returns come from the same 25 country bootstrap everyone gets, with the domestic country resampled rather than pinned to the US. What's genuinely missing for you is NZ tax and NZ Super. So it worked, but you got the generic version. On the existing calculators, they're good and I use them too. Most are tried and true against US history, which is the one sample where equities did unusually well. That's the gap I was after.
- nhootan 21d ago
  Sorry, that's the EU account gate rather than a geoblock. You can browse and run the Quick Check from anywhere except Quebec, but account creation is blocked in the EU/EEA/UK/CH, so the full builder is out of reach. The reason is that GDPR has extraterritorial reach and real penalties, and doing it properly is weeks of work I haven't done for a market I'm not launching in yet. I'd rather block the signup than pretend I'm compliant. Not a satisfying answer if you're the one blocked, I know. The 0.5% step is because the Quick Check is a 66 cell precomputed grid shipped to your browser, so it runs with no account and no server. The builder takes dollar amounts. On residence, there's an "other, rest of world" mode that runs an untaxed gross projection, so it isn't US only. The 100% stocks result holds up. In the international panel, bonds get wrecked by inflation more often than equities do, so more stocks wins on "did the money last" while being much rougher along the way. Cederburg and co-authors find the same. That's why the full detailed results show the full distribution and the drawdowns, not one number.
rachelb1331 21d ago
I didn't want to put in my email address in order to access the detailed calculator... The quick check doesn't have nearly enough information to give anyone an idea of what actually to expect...
- nhootan 21d ago
  Fair, and it's the part I'm least settled on. The account exists because the full plan is a server side run that gets stored, not to build a mailing list. Being straight about it: the long-run panel is the Dimson-Marsh-Staunton dataset, which is a paid commercial licence, plus the compute. So this won't be free forever. The Quick Check will stay free and anonymous. The detailed side is where the real inputs are: mortality table life expectancy, CPP/OAS or Social Security start ages, pensions, tax, flexible spending. Four inputs can't tell you much, agreed.
bryanlarsen 21d ago
It's really nice to see a simulator that removes the home country bias. So many of these types of simulations use American data. But that's pretty extreme cherry-picking: America has had a pretty exceptional run these last 100+ years. Any retirement simulator cannot assume that America's exceptional growth will continue. Maybe it will, but you cannot plan on a maybe.
- nhootan 21d ago
  Thanks, that's most of why I built it. A US only backtest is one draw where you already know who won. This is 25 countries over 125 years, block bootstrapped with the domestic country resampled, so something like Japan after 1990 shows up.
  - bryanlarsen 21d ago
    Do you have Germany 1914 in your dataset? There's no safe withdrawal rate for that particular data point...
    - nhootan 21d ago
      The Germany's data is in since 1901, but there is a gap between 1922-1923, cause the underlying dataset, DMS, has no bonds data for that period.
fusslo 21d ago
huh, yesterday I vibe coded a similar tool. It lets me set my current savings, investment, retirement. Expected yearly savings, expected number of years of work left. After retirement, only withdrawals are allowed (investments, savings still get returns tho) Codex suggested a bunch of improvements: monte carlo, investment volatility, expense inflation, capital gains tax etc. One thing codex didn't ask about was social security. So that's another factor that could be improved. I wanted to know 'do I really need to work 30 more years?' What would be really interesting is discovering an attainable formula where my investments outpace my expenses. Or, what variables let me die with basically $0 in the bank At the end, I realized I had no idea what the calculator was REALLY doing. So it could be right... it could be really wrong too. Having historical data might help solve those issues
- nhootan 21d ago
  Your last paragraph is why I did using MC simulations based on credible data. Once you pick a return and a volatility, the output just restates your inputs and you can't tell a wrong answer from a right one. Historical data doesn't fix that, but it moves the assumption somewhere you can argue about, and shocks arrive as correlated bundles that actually happened. Dying with roughly $0 is a real mode in the app: floor and ceiling, or Yale style smoothing, where spending flexes instead of staying fixed.
rickydroll 21d ago
It will be far more interesting to show a model of how to have a successful retirement with more or less "normal" financial disasters, such as not making enough money to save in a high-COL area, losing your job in your 40s or later, followed by one to two years of unemployment, and then the replacement job pays a fraction of what you made before. There are other financial disasters, such as divorce or a major medical problem, that don't have to lead to bankruptcy, just a serious financial hit. Or even a model of what we have today, where people don't make enough money to save for retirement given the high COL.
- nhootan 20d ago
  That's a great feedback. I think I can add these scenarios as new features to the life events section. I have also thought of allowing to specify correlation between income and markets performance. If you are interested in beta testing these feature, shoot me an email at hi@foresightplanner.com
iamthemonster 21d ago
It's really sobering how much your use of international data impacts the failure rates. If you limit yourself to US historical data then I think about 3.5% withdrawal rate is bulletproof. But in your data I managed to get 2.2% failure rate when withdrawing just 2.5% p.a for 20 years!
- nhootan 20d ago
  [dead]
BrandoElFollito 20d ago
Does not work for France, it's a pity. I guess that all of EU will be out too as we have similar systems
- nhootan 19d ago
  Exactly right, it's GDPR, not specific to France. Working on it ...

16 points by nhootan 21d ago

I got tired of retirement calculators that either assume one fixed return forever, or run Monte Carlo by pulling each year independently out of a bell curve. Real markets don't behave like that. Bad years cluster, and a crash in your first three years of retirement is a completely different problem from the same crash at 85. So this one samples multi-year blocks out of actual history instead: 25 countries, 1900 to 2025, reassembled into thousands of possible futures. It's a stationary block bootstrap, which is the standard approach in the lifecycle-investing literature. When a plan fails it tells you which way it failed, whether that was weak growth, bad sequencing, or inflation at the wrong moment. Canada and the US have full tax and benefit modelling (RRSP/TFSA and CPP/OAS on one side, 401(k)/IRA and Social Security on the other). Everywhere else you get a pre-tax projection. If you're in the EU or UK you can run the Quick Check, but you can't sign up yet, because I haven't finished the compliance work. There's no account linking, and the Quick Check needs no signup at all. The methodology is written up at /methodology. It's an educational tool, not financial advice. It's free right now. Paid plans come later, and there's a founding rate if you want to lock one in early: $50/year for as long as you stay subscribed, first 50 people. https://buy.stripe.com/14A7sN6Gycny54rdXv5Rm01 https://buy.stripe.com/14A7sN6Gycny54rdXv5Rm01 Nothing is gated behind it today.

drop_star 21d ago

Why do I need to sign into everything these days? The moment I see that I close the site.

albumen 21d ago
You don’t need to for its quick check, which shows the core feature.
- rconti 21d ago
  yeah, this is quite literally the first thing i saw when i opened the page.
- bluGill 21d ago
  You need it to dig deeper. I have a pension which I can get at 65. I have options to collect social security. What if I draw my 401k first and wait until 72 to start as? Or I can get take as at 62 and the 401k grows more... Edit point is you need an account to start asking those questions.
  - nhootan 21d ago
    The timing half is there: CPP/OAS or Social Security start ages, and pension start and end ages, so claiming at 62 versus deferring is something you can run both ways today. Draw order across accounts isn't exposed yet. Right now the engine allocates by equalizing marginal tax rates across accounts each year rather than draining one at a time, so it's solving that ordering instead of letting you set it. The manual override is a fair ask and it's coming, the engine already takes the order as a parameter and it just isn't wired to the UI.
nhootan 21d ago
[dead]
nhootan 21d ago
The sign in is required for detailed plans to save the plans. Those plans also incure real comupte on the server. QC results are based on a cache in the browser and don't need sign in.

memoryleakgame 21d ago

So its a monte carlo simulation but I can't enter growth rate country time horizons or anything like that? Brother I have a better claude code vibe app than this in an hour You need to really talk to a financial planner and learn what the proper inputs are

nhootan 21d ago
[dead]
nhootan 21d ago
[dead]

bluGill 21d ago

This doesn't know the important part. Will I live to 62 as the earliest of my ancestors died, or 98 as the latest? My family history is about average for lifespan but that is a big difference. If I live to 62 I should retire at 50 - or perhaps I shouldn't have bothered saving at all since the growth isn't there. Of medical advances make me love to 150 I should save until I'm 80 which will be a nice income to really enjoy the rest of my life. I want more defined benefit accounts, but they are hard to find

karmakurtisaani 21d ago
So the thing is that no one knows that part.
- Eddy_Viscosity2 21d ago
  Can't we just get AI to tell us? I'm sure it will confidently give an answer.
  - bluGill 21d ago
    The latest AI is a lot better at not being over confident. There is still an issue but not as good as it used to be
    - karmakurtisaani 21d ago
      But how can you trust anything it says if it doesn't have confidence? I tell even my children that the most important aspect of applying a job is a confident, firm handshake.
- bluGill 21d ago
  For me no, but for a few thousand people it isn't to hard. Which is why I look for defined benefit plans that are not scams, and in frustrated that I can't fibd them
nhootan 21d ago

betaby 21d ago

Better be safe I guess: Foresight Planner is not yet available in your region. We're launching in Canada (excluding Quebec) first. We don't yet support Quebec, because we don't yet offer the product in French. If you'd like us to let you know when we expand, email hi@foresightplanner.com. Foresight Planner n'est pas encore disponible dans votre région. Nous lançons d'abord au Canada (hors Québec). Pour être informé·e de l'élargissement de notre service, écrivez à hi@foresightplanner.com. Region: CA-QC

nhootan 21d ago
It's Bill 96 and Law 25. No French UI yet, so blocking beats being non compliant. It's near the top of the list.

jawns 21d ago

I got 82.4% regardless of whether I started with a portfolio value of $1M or $3M. Something is wrong with the calculations/simulations. If I can't trust the Quick Check, why would I trust anything else on the site?

Eric714 21d ago
Same here
Eric714 21d ago
That's because the withdrawals are 4%. If you have $3M, maybe lower the withdrawals to 1.25%. That would be the same as 4% of $1M.
- jawns 21d ago
  I don't think that's accurate. If I withdraw 4% of a $1M portfolio, I only have $960K left to compound. If I withdraw 4% of a $3M portfolio, I should have $2.88M left to compound. If the question is "Will your money last?" I would not expect the simulation results to be identical. Edit: I failed here. Downvotes are right. I simulated it myself and saw the results. Leaving this comment here so others can learn from my mistake.
nhootan 21d ago
That result is right. The Quick Check takes a withdrawal rate, not a dollar amount, and runs with taxes off, so it's scale invariant: $1M at 4% and $3M at 4% are the same simulation times three. That's what lets it be instant and anonymous, since the whole answer space fits in a small grid I ship to your browser. It stops being true once taxes are on, which is what the full builder does. The page should say so. I'll add a line.

BretonForearm 21d ago

Why not let people try it from abroad? Not in the USA? No luck. Why does the withdrawal rate need to be a multiple of 0.5%? It's curious that the highest chance was associated with the highest (100%) stock ratio.

pelagic_sky 21d ago
I'm in NZ and it worked. It gave me US figures and numbers though. And I find there are existing FIRE calculators already out there that are tried and true.
- nhootan 21d ago
  Glad it ran, but the numbers aren't US. Once you login, "Other, rest of world" gives you a gross projection with tax and benefits off. The returns come from the same 25 country bootstrap everyone gets, with the domestic country resampled rather than pinned to the US. What's genuinely missing for you is NZ tax and NZ Super. So it worked, but you got the generic version. On the existing calculators, they're good and I use them too. Most are tried and true against US history, which is the one sample where equities did unusually well. That's the gap I was after.
nhootan 21d ago
Sorry, that's the EU account gate rather than a geoblock. You can browse and run the Quick Check from anywhere except Quebec, but account creation is blocked in the EU/EEA/UK/CH, so the full builder is out of reach. The reason is that GDPR has extraterritorial reach and real penalties, and doing it properly is weeks of work I haven't done for a market I'm not launching in yet. I'd rather block the signup than pretend I'm compliant. Not a satisfying answer if you're the one blocked, I know. The 0.5% step is because the Quick Check is a 66 cell precomputed grid shipped to your browser, so it runs with no account and no server. The builder takes dollar amounts. On residence, there's an "other, rest of world" mode that runs an untaxed gross projection, so it isn't US only. The 100% stocks result holds up. In the international panel, bonds get wrecked by inflation more often than equities do, so more stocks wins on "did the money last" while being much rougher along the way. Cederburg and co-authors find the same. That's why the full detailed results show the full distribution and the drawdowns, not one number.

rachelb1331 21d ago

I didn't want to put in my email address in order to access the detailed calculator... The quick check doesn't have nearly enough information to give anyone an idea of what actually to expect...

nhootan 21d ago
Fair, and it's the part I'm least settled on. The account exists because the full plan is a server side run that gets stored, not to build a mailing list. Being straight about it: the long-run panel is the Dimson-Marsh-Staunton dataset, which is a paid commercial licence, plus the compute. So this won't be free forever. The Quick Check will stay free and anonymous. The detailed side is where the real inputs are: mortality table life expectancy, CPP/OAS or Social Security start ages, pensions, tax, flexible spending. Four inputs can't tell you much, agreed.

bryanlarsen 21d ago

It's really nice to see a simulator that removes the home country bias. So many of these types of simulations use American data. But that's pretty extreme cherry-picking: America has had a pretty exceptional run these last 100+ years. Any retirement simulator cannot assume that America's exceptional growth will continue. Maybe it will, but you cannot plan on a maybe.

nhootan 21d ago
Thanks, that's most of why I built it. A US only backtest is one draw where you already know who won. This is 25 countries over 125 years, block bootstrapped with the domestic country resampled, so something like Japan after 1990 shows up.
- bryanlarsen 21d ago
  Do you have Germany 1914 in your dataset? There's no safe withdrawal rate for that particular data point...
  - nhootan 21d ago
    The Germany's data is in since 1901, but there is a gap between 1922-1923, cause the underlying dataset, DMS, has no bonds data for that period.

fusslo 21d ago

huh, yesterday I vibe coded a similar tool. It lets me set my current savings, investment, retirement. Expected yearly savings, expected number of years of work left. After retirement, only withdrawals are allowed (investments, savings still get returns tho) Codex suggested a bunch of improvements: monte carlo, investment volatility, expense inflation, capital gains tax etc. One thing codex didn't ask about was social security. So that's another factor that could be improved. I wanted to know 'do I really need to work 30 more years?' What would be really interesting is discovering an attainable formula where my investments outpace my expenses. Or, what variables let me die with basically $0 in the bank At the end, I realized I had no idea what the calculator was REALLY doing. So it could be right... it could be really wrong too. Having historical data might help solve those issues

nhootan 21d ago
Your last paragraph is why I did using MC simulations based on credible data. Once you pick a return and a volatility, the output just restates your inputs and you can't tell a wrong answer from a right one. Historical data doesn't fix that, but it moves the assumption somewhere you can argue about, and shocks arrive as correlated bundles that actually happened. Dying with roughly $0 is a real mode in the app: floor and ceiling, or Yale style smoothing, where spending flexes instead of staying fixed.

rickydroll 21d ago

It will be far more interesting to show a model of how to have a successful retirement with more or less "normal" financial disasters, such as not making enough money to save in a high-COL area, losing your job in your 40s or later, followed by one to two years of unemployment, and then the replacement job pays a fraction of what you made before. There are other financial disasters, such as divorce or a major medical problem, that don't have to lead to bankruptcy, just a serious financial hit. Or even a model of what we have today, where people don't make enough money to save for retirement given the high COL.

nhootan 20d ago
That's a great feedback. I think I can add these scenarios as new features to the life events section. I have also thought of allowing to specify correlation between income and markets performance. If you are interested in beta testing these feature, shoot me an email at hi@foresightplanner.com

iamthemonster 21d ago

It's really sobering how much your use of international data impacts the failure rates. If you limit yourself to US historical data then I think about 3.5% withdrawal rate is bulletproof. But in your data I managed to get 2.2% failure rate when withdrawing just 2.5% p.a for 20 years!

nhootan 20d ago
[dead]

BrandoElFollito 20d ago

Does not work for France, it's a pity. I guess that all of EU will be out too as we have similar systems

nhootan 19d ago
Exactly right, it's GDPR, not specific to France. Working on it ...