Beancount says I can retire at 4% — so why can't I stop thinking about a year-one crash?

General
1

I’ve been tracking everything in Beancount for over four years now, personal spending plus two rentals, and last month I finally ran the full retirement math. On paper I’m fine retiring around a 4% withdrawal rate. But I can’t stop thinking about timing. What if year one looks like 2008 and I’m pulling money out while everything is down? I read Kitces found the first ten years of real returns explain something like 80% of how it all turns out, with 1966 as the nightmare start into stagflation. Has anyone here actually modeled different return sequences in Beancount, and did it change anything you do?

2

What if year one looks like 2008 and I’m pulling money out while everything is down?

That crash is already in the number. Bengen calibrated 4% on the 1966 retiree eating stagflation for a decade. If your ledger says 4% works, you’ve already survived it on paper.

3

Sequence of Returns Risk: Why Your Retirement Year Matters More Than Your Average Return | Summitward walks through Kitces on why the first decade drives outcomes, with 1966 as worst case.

4

ok this might be dumb but how do you model different return sequences in beancount, like is there a monte carlo plugin or do you just export to python?

5

Not dumb at all — I haven’t found a monte carlo plugin I actually trust yet, so I just export positions with bean-query and run them through a little python script. I shuffle historical yearly returns so the average stays the same but the order changes, then watch which sequences break.

6

I created with my own tool for retirement portfolio projections with Claude. It can use Monte Carlo to generate return sequences, or use actual return sequences (with different starting years). I can also tell it it simulate different historical crashes (the Great Depression, 1973-74 oil-crisis bear market, dot-com crash, etc.) in whatever starting year I like. I have found it helpful to quantify this type of concern rather than just having it as a vague concern keeping me up at night, and hoping the Monte Carlo simulation adequately acocunts for it.

I leveraged a different open-source tool so that I didn’t have Claude creating the entire thing from scratch, figuring it would be more difficult for me to find mistakes that way.