TL;DR. The Federal Reserve’s survey of US households, fielded in October 2025, found that 58 percent of self-employed adults had income that varied from month to month, against 28 percent of people who work for someone else, and that 22 percent of the self-employed struggled to pay bills because of it (10 percent of employees). CEOtudent’s analysis of the public microdata adds the uncomfortable part: the self-employed were no more likely than employees to hold a three-month emergency fund (56.8 percent against 57.2 percent), and among solo operators with no staff the share was 52.9 percent. The three-month rule assumes a paycheck that stops cleanly and an unemployment insurance system that softens the stop; US unemployment insurance excludes self-employment. The CEO move is to size the buffer with an explicit equation (a shock layer scaled to your volatility, plus the larger of your worst dry spell or your top-client exit) and to measure it net of tax you already owe. The student move is to recompute it from your own last 12 to 24 months every quarter instead of borrowing a number built for someone else. For most variable-income profiles in the worked examples, the answer lands between five and nine months, not three.
This piece sits inside CEOtudent’s money-management cluster for independents. The account structure that holds the buffer is in the variable income budget, the pre-quit version of this calculation is in how much runway you need to quit your job, and the client-exit scenario used below draws on the client concentration guide. This article is for people already operating: the question is not how long you can survive a launch, but how much cash an ongoing one-person business should hold against the gaps it will meet.
This article is general information about research findings and arithmetic, not financial, tax or legal advice for your situation.
Where does the three-month rule come from, and who was it built for?
The “three to six months of expenses” guideline is a convention, not a research finding, and the research that does exist cuts both ways. The JPMorgan Chase Institute, the research arm of a commercial bank, studied six million anonymized Chase deposit customers and concluded that a typical family needs roughly six weeks of take-home income in liquid assets to weather a simultaneous income dip and spending spike, which it noted is lower than the existing advice of three to six months. That finding is often read as “the rule is too conservative”. For the self-employed it points the other way, for three reasons.
First, the six-week figure describes the median family among Chase deposit customers, measured in take-home income, and it covers a one-month shock: a month in which income falls more than 25 percent below the family’s prior-year median while spending jumps. It does not describe a three-month dry spell or the loss of a client who pays half the bills.
Second, the rule quietly assumes an external floor. A salaried worker who loses a job in the United States usually files for unemployment insurance. The US Department of Labor’s comparison of state laws states the coverage principle plainly: employment for UI purposes is service of an employee for an employer, and therefore it does not include self-employment. Most states cap benefits at 26 weeks, and some replace roughly 50 percent of lost wages up to a limit.
Third, a self-employed person’s bank balance includes money that is not theirs. Income tax and self-employment tax are paid in quarterly instalments, so cash that looks like a buffer can be a tax bill that falls due in a few weeks.
How different is self-employed income, in the data?
Two independent sources, one a survey and one built on IRS and Social Security Administration records, describe the same gap.
Table 1. Income variability and emergency savings by work arrangement, US adults, 2025 (CEOtudent analysis of SHED 2025 public-use microdata)
| Group | Income varies at least occasionally | Varies quite often | Struggled to pay bills because income varied | Has a three-month rainy day fund | Could not cover three months by any means | Respondents (unweighted) |
|---|---|---|---|---|---|---|
| All adults | 29.8% | 8.4% | 10.9% | 54.8% | 30.0% | 12,934 |
| Work for someone else | 28.4% | 5.8% | 10.1% | 57.2% | 25.8% | 6,285 |
| Self-employed | 58.4% | 22.1% | 22.3% | 56.8% | 27.9% | 935 |
| Self-employed, no paid employees | 60.8% | 23.4% | 24.1% | 52.9% | 31.6% | 745 |
| Self-employed, with paid employees | 48.4% | 17.3% | 14.9% | 72.9% | 12.9% | 178 |
| Self-employed whose income varies | 100% | 37.9% | 38.1% | 50.5% | 32.8% | 550 |
Source: Board of Governors of the Federal Reserve System, Survey of Household Economics and Decisionmaking 2025, public-use data and codebook. Shares weighted with the survey’s “weight” variable, the weight used for the published report. Work arrangement is the respondent’s main job in the prior month (question D3A); “paid employees” is question D3E. “Rainy day fund” is question EF1 (funds that would cover expenses for three months); “could not cover three months by any means” means no to EF1 and no to EF2 (borrowing, using savings or selling assets). Our weighted totals reproduce the published headline figures: 58 and 28 percent with varying income, 22 and 10 percent struggling, 55 percent with a rainy day fund and 30 percent unable to cover three months. Approximate 95 percent margins are about plus or minus 3.4 points for the self-employed row, 3.8 for solo operators and 8 for employers, so read the employer row as indicative.
Three findings stand out. The self-employed are about twice as likely as employees to have varying income (2.06 times, by CEOtudent’s calculation) and almost four times as likely to say it varies quite often (22.1 against 5.8 percent, a ratio of 3.8). The Fed notes that self-employed adults tend to have lower incomes, but that after controlling for income they were still more likely to face hardship from income variability. And despite that, the self-employed hold three-month funds at the same rate as employees. Within the self-employed, those whose income varies are 15.3 points less likely to have a three-month fund than those whose income is steady (50.5 against 65.7 percent): the people who need the largest buffer are the least likely to hold even the employee-sized one.
The tax-record evidence comes from a working paper by Anmol Bhandari, Tobey Kass, Thomas J. May, Ellen McGrattan and Evan Schulz (NBER Working Paper 32948, October 2024, revised August 2025), built on IRS data. It finds that the 10th-to-90th percentile range of annual income changes is roughly 2.5 times larger for the self-employed than for the paid-employed, and between 2 and 3.5 times larger depending on age. The NBER summary of the paper adds that in the Great Recession, self-employed income in real estate and construction fell by 50 and 33 percent year over year. The same summary notes that 57 percent of primarily self-employed people earn less than comparable employees, while those people account for only 16 percent of self-employment income: the averages hide a very wide spread.
Which published rules and benchmarks does the method use?
The equation below is built only from components that can be checked. Table 2 lists them as published.
Table 2. Benchmarks behind the Buffer Equation (figures as published)
| Component | Figure as published | Source |
|---|---|---|
| Buffer for a simultaneous income dip and spending spike | 6.2 weeks of income (median); 65 percent of families lacked it; once every 5.5 years | JPMorgan Chase Institute, Weathering Volatility 2.0, Table 2 |
| Buffer for an income dip alone | 2.8 weeks; 48 percent lacked it; once every 9 months | Same |
| Buffer for a spending spike alone | 2.6 weeks; 46 percent lacked it; once every 4 months | Same |
| Median income volatility | Coefficient of variation 0.38, described as a 36 percent month-to-month change | Same, Finding 1 |
| Income dip month | Income more than 25 percent below the prior twelve-month median | Same, definitions |
| Self-employed vs paid-employed income risk | 10th-90th percentile range of income changes roughly 2.5 times larger | Bhandari et al., NBER WP 32948 |
| Unemployment insurance coverage | Employment for UI purposes does not include self-employment | US Department of Labor, Comparison of State UI Laws 2023 |
| UI duration and level | Most states cap benefits at 26 weeks; some replace approximately 50 percent of lost wages up to a limit | Same |
| When estimated tax is required | Expect to owe $1,000 or more when the return is filed | IRS, Estimated taxes |
| Underpayment safe harbour | 90 percent of this year’s tax or 100 percent of last year’s, whichever is smaller; 110 percent if prior-year AGI exceeded $150,000 | IRS, Estimated taxes; Publication 505 (2026) |
| Self-employment tax | 15.3 percent (12.4 Social Security, 2.9 Medicare); Social Security part on the first $184,500 for 2026 | IRS, Self-employment tax |
| Size of the group | 8.743 million unincorporated self-employed in nonagricultural industries, October 2026, seasonally adjusted; 11.9 million independent contractors (7.4 percent of employment) in July 2023 | BLS, Employment Situation Table A-8; BLS, Contingent and Alternative Employment Arrangements |
The JPMorgan Chase Institute figures are company research institute data. The NBER paper is a working paper, not yet peer reviewed. The insurance and tax rules are US rules; the method works anywhere once local equivalents are substituted.
Why does three months of savings buy an employee more time than it buys you?
The clearest way to see the missing floor is to ask how long the same savings last. Table 3 holds expenses constant and changes only whether unemployment insurance covers part of them.
Table 3. How long three or six months of savings lasts (CEOtudent calculation; replacement shares are illustrative assumptions)
| Situation | Three months saved lasts | Six months saved lasts |
|---|---|---|
| Employee, UI covers 50% of expenses for up to 26 weeks (illustrative assumption) | 6.0 months | 9.0 months |
| Employee, UI covers 30% because of the state cap (illustrative assumption) | 4.3 months | 7.8 months |
| Self-employed, no UI | 3.0 months | 6.0 months |
Method: while benefits last, savings fund only the uncovered share of expenses; 26 weeks is 6.0 months at 52/12 weeks a month. After benefits end, savings fund all expenses.
So “three months” is shorthand for very different amounts of protection. For an employee with typical benefits it can stretch close to six months of search, which is roughly the horizon our runway guide found in US unemployment-duration data. For a self-employed person the same three months is three months. To buy the protection an employee’s three months buys, an independent needs something like six.
Why does the tax set-aside not count as buffer?
The IRS asks individuals, including sole proprietors, to make estimated payments if they expect to owe $1,000 or more. The 2026 schedule in Publication 505 splits the year unevenly, and the unevenness matters for cash.
Table 4. How long estimated tax sits in your account, US 2026 schedule (CEOtudent analysis of IRS Publication 505)
| Income earned | Months in the period | Due date | Oldest accrual held until due |
|---|---|---|---|
| 1 January to 31 March | 3 | 15 April 2026 | 3.5 months |
| 1 April to 31 May | 2 | 15 June 2026 | 2.5 months |
| 1 June to 31 August | 3 | 15 September 2026 | 3.5 months |
| 1 September to 31 December | 4 | 15 January 2027 | 4.5 months |
In mid-January, an operator who reserves correctly is holding four months of tax on autumn income. On an illustrative $9,000 of monthly revenue with a 25 percent reserve (illustrative assumptions, not a tax rate), that is $9,000 of someone else’s money by 31 December and $10,125 by mid-January, about two months of a $5,000 essential budget (2.02 months). Counting it as emergency cash produces a buffer that disappears on the due date. The rule in the method is therefore simple: the buffer is measured after the tax account, never including it. The variable income budget keeps tax in its own account for exactly this reason.
What is the Buffer Equation?
The method sizes the fund in months of essential outflow, E: the personal costs you cannot avoid plus the business costs you must keep paying (software, insurance, rent of a workspace). It has two layers, and the second is the larger of two scenarios.
Table 5. The Buffer Equation (CEOtudent editorial framework)
| Layer | Formula | What it covers | Where the inputs come from |
|---|---|---|---|
| Shock layer | 1.43 x max(1, CV / 0.38) | Ordinary bad months: a dip plus an unplanned bill | 1.43 months is the JPMorgan Chase Institute’s 6.2 weeks; 0.38 is its median CV; CV is your own coefficient of variation |
| Dry-spell scenario | L x (1 – r) | Your worst run of low months | L = longest run of months below 75% of your median income in the last 24 months; r = share of E your income still covered in those months |
| Client-exit scenario | R x max(0, 1 – (1 – s) x c) | Losing your largest client | s = top-client share of revenue; c = normal after-tax income divided by E; R = months to replace the lost work |
| Gap layer | max(dry spell, client exit) | The longest plausible gap | The larger of the two scenarios, not their sum |
| Buffer | Shock layer + gap layer, rounded up to whole months | Measured net of the tax account |
To get your CV, list your last 12 monthly after-tax incomes, divide their standard deviation by their mean. A CV of 0.38 is the bank-data median; anything above it scales the shock layer up in proportion, which is an editorial assumption, since the JPMorgan Chase Institute did not publish buffer needs by volatility level. The 75 percent threshold for a low month mirrors the Institute’s definition of an income dip.
The client-exit term is deliberately harsh: it assumes the remaining clients carry on at their normal level and nothing replaces the lost one until R months have passed. If your largest client pays 30 percent and the rest of the book still covers your essentials, the term is close to zero. If one client pays 75 percent, it dominates everything else. The client concentration guide explains how to cut s over time; this equation prices what s costs you in cash while it is still high.
The two gap scenarios are not added, because a client exit usually shows up in the history as a dry spell.
What does the equation give for real-world profiles?
Five profiles show how the parts interact. Every input below is an illustrative assumption, chosen to resemble common one-person business types, and every result is reproduced in calc.py.
Table 6. Worked examples: inputs (all values are illustrative assumptions)
| Profile | CV | Longest dry spell, L (months) | Coverage in the spell, r | Top-client share, s | Coverage ratio, c | Replacement time, R (months) |
|---|---|---|---|---|---|---|
| A. Designer on three retainers | 0.25 | 2 | 50% | 30% | 1.4 | 3 |
| B. Project consultant | 0.60 | 3 | 20% | 45% | 1.6 | 5 |
| C. Seasonal course creator | 0.80 | 4 | 30% | 15% | 1.5 | 2 |
| D. Contractor anchored on one client | 0.30 | 1 | 60% | 75% | 1.3 | 6 |
| E. Large-project consultant | 1.00 | 5 | 10% | 60% | 1.5 | 6 |
Table 7. Employee rule vs volatility-adjusted months (CEOtudent analysis; illustrative inputs from Table 6, essential outflow E = $5,000 a month as an illustrative assumption)
| Profile | Shock layer | Dry spell | Client exit | Buffer, months | Buffer at E = $5,000 | Short of a three-month fund by |
|---|---|---|---|---|---|---|
| A. Retainer designer | 1.43 | 1.00 | 0.06 | 3 | $15,000 | $0 |
| B. Project consultant | 2.26 | 2.40 | 0.60 | 5 | $25,000 | $10,000 |
| C. Seasonal course creator | 3.01 | 2.80 | 0.00 | 6 | $30,000 | $15,000 |
| D. Anchored contractor | 1.43 | 0.40 | 4.05 | 6 | $30,000 | $15,000 |
| E. Large-project consultant | 3.77 | 4.50 | 2.40 | 9 | $45,000 | $30,000 |
Example for profile B: shock layer 1.43 x (0.60 / 0.38) = 2.26; dry spell 3 x (1 – 0.20) = 2.40; client exit 5 x (1 – 0.55 x 1.6) = 0.60; gap layer 2.40; total 4.66, rounded up to 5 months.
Only the steadiest profile, a designer with three retainers, a modest top client and a short history of dips, clears the bar with three months. The other four need between five and nine. Profile D is the instructive one: its month-to-month income is calm (CV 0.30), so a volatility-only rule would wave it through, yet the single dominant client makes it as exposed as the seasonal creator. Profile E shows where the employee rule fails worst: lumpy, concentrated, slow-to-replace income needs triple the conventional fund.
How do you score yourself if you do not have the data yet?
Without 24 months of clean records, use the scorecard below as a screen. It errs high by design: for all five profiles it gives the same or a larger answer than the equation.
Table 8. The emergency-fund scorecard for variable income (CEOtudent editorial framework)
| Factor | 0 points | 1 point | 2 points |
|---|---|---|---|
| 1. Income volatility (CV of last 12 months) | Below 0.38 | 0.38 to 0.75 | Above 0.75 |
| 2. Longest run of months below 75% of median, last 24 months | 1 month or less | 2 to 3 months | 4 months or more |
| 3. Largest client’s share of revenue | Under 25% | 25% to 49% | 50% or more |
| 4. Typical days from invoice to cash | 30 or fewer | 31 to 60 | More than 60 |
| 5. Months to replace a lost client or contract | 2 or fewer | 3 to 5 | 6 or more |
| 6. Share of essential outflow you cannot cut within 30 days | Under 50% | 50% to 80% | Over 80% |
| 7. Household backstop (a second stable income) | Covers half or more of E | Covers less than half | None |
Score to buffer (editorial bands): 0 to 3 points, 3 months; 4 to 6 points, 5 months; 7 to 9 points, 7 months; 10 or 11 points, 9 months; 12 to 14 points, 12 months. Always measured net of the tax account.
The client-share bands match those in the concentration guide so the two tools can be read together. Applied to the five profiles (illustrative scoring in calc.py), the scorecard gives 5, 7, 7, 7 and 12 months, against the equation’s 3, 5, 6, 6 and 9. The gap is the price of not having data: once you have 12 to 24 months of records, the equation should replace the scorecard.
How do you build the fund once you know the number?
Knowing the target is the easy half. The data in Table 1 says most solo operators are below even the three-month mark, so the order of operations matters more than the destination.
- Separate the tax account first. A buffer that includes owed tax is an illusion that ends on a due date. Sweep the reserve on the day each payment arrives.
- Fund the shock layer before anything optional. It is the part you will use most often: in the Institute’s data, an income dip alone hit about once every nine months for the typical family, and a self-employed income is more volatile than that.
- Then fund the gap layer. This is the slow part. Routing the surplus from above-median months, rather than a fixed monthly amount, matches saving to the months when it is affordable; the owner-pay waterfall in the paying-yourself guide sets the order in which surplus is split.
- Shrink the target as you de-risk. Every input in the equation can be moved by operating decisions: a retainer lowers CV, a second large client lowers s, shorter payment terms shorten dry spells. Each change is worth recomputing, because a smaller required buffer frees cash for reinvestment.
- Recompute quarterly. Use the same day you review the one-person P&L, so the buffer and profitability are read from the same numbers.
How does this fit with the runway and buffer pieces in this cluster?
The quit-runway calculation covers the one-time ramp from a salary to break-even; the buffer in the paying-yourself guide protects the owner’s salary through uneven months; this fund is the reserve an operating business holds against shocks and gaps once it is steady. In the first year the runway is the binding number. Holding both at full size usually double counts.
What this does not mean
- It is not a forecast. The equation turns your history into a size. A history of 24 months cannot show a recession or an industry shift; the NBER summary’s real estate and construction declines show how large such shocks can be.
- The anchors are imported. The 6.2-week shock figure and the 0.38 median CV come from Chase checking accounts in 2013 to 2018 of households of all kinds, measured in take-home income, not from a sample of the self-employed. Scaling them by your CV is an editorial assumption, labelled as such.
- The worked examples are not people. Profiles A to E are illustrative assumptions built to show how the terms interact. They are not survey respondents and not typical values for any occupation.
- The survey cannot show causality. Table 1 shows that self-employed adults with varying income are less likely to hold a three-month fund. It does not show why: lower average income, younger businesses and recent shocks could all contribute.
- US rules are examples. Some countries offer the self-employed unemployment schemes or voluntary insurance, and tax calendars differ. Substitute local rules first.
- It is not a reason to hoard. Cash above the computed buffer has an opportunity cost; the method exists to make the target finite.
The CEO and the student in this decision
A CEO does not accept a cash policy because it is popular; they ask what risk it covers and whether that risk is theirs. The three-month rule covers a salaried worker’s job loss in a system that pays partial benefits. A one-person business carries different risks: volatility, dry spells, a dominant client, tax timing, and no insurer behind any of them. Writing those risks into an equation, and sizing the reserve to them, is ordinary treasury management applied to a company of one.
The student side is measurement. Most independents can quote their best and average month; few have computed their coefficient of variation or replacement time, and those inputs decide the answer. Recomputing them every quarter, and letting the target fall as the business de-risks, turns an open-ended savings goal into a number with a reason behind it.
Frequently asked questions
Is three months ever enough for a self-employed person?
Sometimes. In the worked examples, a designer with three retainers, no client above 30 percent and short dips needed three months. That profile has low volatility, a diversified book and a fast replacement time. Without all three, the equation lands above three months.
Should the fund be measured in months of income or months of expenses?
Expenses. The method uses E, the essential personal and business outflow you cannot avoid in a lean month. The JPMorgan Chase Institute measured weeks of take-home income; for a household whose spending roughly matches its income the two are close, which is why the conversion to months of E is used as an approximation.
How is the coefficient of variation calculated?
Take the last 12 monthly after-tax incomes, compute their standard deviation and divide it by their average. Use income after business costs and tax reserves, because that is what funds living costs.
Does money in a tax savings account count toward the buffer?
No. Under the US 2026 schedule, income earned from September to December is not due until 15 January 2027, so the tax account can hold over four months of reserve. That money is already owed. Measure the buffer after it.
What if a partner has a steady salary?
That is factor 7 in the scorecard. A second stable income that covers half or more of essential outflow reduces the share of E your own business must protect. In the equation, it raises r and c, which shrinks both gap scenarios.
Where should the emergency fund be kept?
The research cited here measures liquid cash in checking and savings accounts, and the method assumes the money can be reached within days. Which specific accounts or products suit you is a question for a qualified adviser in your country.
How often should the number be recomputed?
Quarterly is a workable rhythm, and after any change to the inputs: a new anchor client, the loss of one, a change in payment terms or a new fixed cost.
Sources
- Board of Governors of the Federal Reserve System. Economic Well-Being of U.S. Households in 2025. May 2026.
- Board of Governors of the Federal Reserve System. Survey of Household Economics and Decisionmaking 2025, public-use data file (CSV) and codebook. Accessed 15 November 2026.
- Diana Farrell, Fiona Greig and Chenxi Yu. Weathering Volatility 2.0: A Monthly Stress Test to Guide Savings. JPMorgan Chase Institute, October 2019. Company research institute report.
- Anmol Bhandari, Tobey Kass, Thomas J. May, Ellen McGrattan and Evan Schulz. On the Nature of Entrepreneurship. NBER Working Paper 32948, October 2024, revised August 2025.
- National Bureau of Economic Research. The Earnings of Self-Employed Workers. The Bulletin on Entrepreneurship, summary of NBER Working Paper 32948.
- US Department of Labor, Employment and Training Administration. Comparison of State Unemployment Insurance Laws 2023, Chapter 1: Coverage.
- US Department of Labor, Employment and Training Administration. Comparison of State Unemployment Insurance Laws 2023, Chapter 3: Monetary Entitlement.
- Internal Revenue Service. Estimated taxes. irs.gov, accessed 15 November 2026.
- Internal Revenue Service. Publication 505 (2026), Tax Withholding and Estimated Tax.
- Internal Revenue Service. Form 1040-ES (2026), Estimated Tax for Individuals.
- Internal Revenue Service. Self-employment tax (Social Security and Medicare taxes). irs.gov, accessed 15 November 2026.
- US Bureau of Labor Statistics. The Employment Situation, October 2026, Table A-8: Employed persons by class of worker and part-time status.
- US Bureau of Labor Statistics. Contingent and Alternative Employment Arrangements, July 2023 (USDL-24-2267), 8 November 2024.
Note on method: Table 1 is a CEOtudent analysis of the SHED 2025 public-use microdata, weighted as in the published report and checked against the published headline figures; margins are approximate and ignore the survey’s design strata. Tables 3, 4 and 7 are CEOtudent calculations from the sources named under them, reproduced in calc.py. Tables 5 and 8, the score bands and the client-exit and dry-spell formulas are CEOtudent editorial frameworks; scaling the shock layer by CV is an editorial assumption. Every input in Tables 6 and 7 and in the tax example is an illustrative assumption. Table 2 reproduces figures as published. The widely repeated claim that a fixed share of freelancers have “no savings at all” was not used because no primary source for it was opened; the JPMorgan Chase Institute’s 36 percent median income change is already reported in our hub guide and is cited here only as the definition of its 0.38 coefficient of variation.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:














