25,000 USD to Day Trade?

This article makes no claim about anybody's intentions. It does not allege that a threshold was designed to exclude anyone, and it does not defend the choice as wise. I have no document describing why the figure was set where it was, and I am not going to speculate in place of one. Every argument here is arithmetic, and where I reason rather than cite, the sentence says so.

It also is not investment advice, and nothing in it should be read as a recommendation to trade in any manner whatsoever.

One further commitment, made here because the argument depends on it. Every statistical method used below is named, and where a choice of method changes the answer, all the candidates are shown. An article whose entire complaint is that nobody put a number on something cannot then produce a number without saying how.

What it is: a post-mortem. For roughly a generation, an American retail trader who wanted to day trade in a margin account had to hold twenty-five thousand dollars, and for roughly a generation people argued about whether that number meant anything. That argument can now be settled on one side at least, because the rule is being retired, and because the arithmetic was always available to anyone who cared to do it.

---

PART I. THE RULE, AND ITS END

FINRA has replaced its day trading margin provisions, including the pattern day trader requirements, with a new risk-based intraday margin regime. The change takes effect 4 June 2026, with a permitted transition period through 20 October 2027 for firms that need longer.

Under the regime now ending, a firm designated an investor a pattern day trader by counting trades, and that designation carried a minimum equity requirement of twenty-five thousand dollars. Under the regime replacing it, per FINRA's own description: there is no twenty-five thousand dollar minimum equity requirement for day trading, and there is no pattern day trader designation based on counting trades. Instead a firm monitors the account through the day to ensure equity remains adequate relative to actual open positions. An account that cannot cover its open positions has an intraday margin deficit, expected to be met promptly; repeated failure can restrict the account for up to ninety days.

Two points of precision, because both are commonly misstated. The maintenance requirement, a minimum of twenty-five percent of the current market value of long margin-eligible equity securities, must be held throughout the trading day, not merely at its end. And the two thousand dollar figure that remains is the minimum equity to trade on margin; below it, an account may still trade unleveraged against available cash. Firms may impose higher requirements than the rule sets, and many do.

So the gate is coming down. The question this article asks is the one that was never answered while it stood: was there arithmetic underneath it?

---

PART II. WHAT A TRADER IS ACTUALLY MEASURING

Set the regulation aside for a moment and consider a plain statistical problem, because the two collide later and the collision is the point.

A trader has some true win rate. Call it p. It is a real property of that person's method under prevailing conditions, and, crucially, they do not know what it is. They only ever observe a sample: the trades they have actually taken. The realized results are an estimate, and estimates carry uncertainty that shrinks as the sample grows. That shrinking is the law of large numbers, and its rate is not a matter of opinion.

Three assumptions, stated openly because everything downstream rests on them.

1. Reward and risk are one to one, which puts break-even at a fifty percent win rate. This is the article's simplification. Part V shows which way the answer moves when it does not hold.
2. The win rate is a constant. Whether a trader improves with practice is a real and interesting question and it is not this one. Here, p is whatever it is; the trader is assumed to have some experience and a settled method.
3. Ninety-five percent confidence, the conventional standard, giving z equal to 1.96.

2.1 A digression that is not a digression: where n minus one belongs

Anyone who has spent time in finance will have noticed something missing from Fig. 01. Volatility, Sharpe ratios, tracking error, every covariance matrix any risk desk has ever built: they all divide by n minus one, not by n. So why does the formula above divide by n?

Bessel's correction exists to fix a specific bias. When you compute a sample mean and then measure spread around that same sample mean, you have used the data twice, and the resulting variance comes out systematically too small. Dividing by n minus one repairs it. The correction is not a convention or a safety factor; it is the arithmetic consequence of having spent a degree of freedom.

A Bernoulli trial spends no such degree of freedom, because its variance is a known function of its mean. A win-or-lose outcome with probability p has variance exactly p(1 minus p). Nothing is estimated. So when the standard error of a sample proportion is derived, the n that appears comes from summing n independent variances and dividing by n squared, not from counting degrees of freedom. There is no second estimate, therefore nothing to correct.

Write n minus one into Fig. 01 and you have not made it more rigorous. You have made it wrong.

The correction returns the moment this analysis leaves win rates and moves to returns, which are continuous and whose mean genuinely is estimated from the sample. That is a different article and a harder one, and it is where every n minus one in finance lives.

2.2 The interval is a choice, and the choice moves the answer

There is a second decision hiding in Fig. 01, and it is the one that trips most published work on this subject.

Multiplying the standard error by z and adding it either side of the estimate produces the Wald interval, which is what nearly everyone means by a confidence interval on a proportion. It is the one taught first. It is also the one with the worst coverage of the common choices; its actual confidence falls short of its stated confidence, and the shortfall is worst exactly where a bound sits near the edge of a decision.

The Wilson score interval, published in 1927, is better behaved: it is asymmetric, it never runs outside zero and one, and its coverage holds up at sample sizes where the Wald interval quietly fails. Better still is the Clopper-Pearson interval of 1934, which is computed from the binomial distribution directly and makes no normal approximation at all; it is exact, at the cost of being conservative.

This article reports the Wilson figure and shows all three. The reason will become obvious in section 2.4, where the Wald answer turns out to fail its own test.

2.3 The bar that actually matters

Now the question that decides everything, and it is not the one usually asked.

Knowing your win rate precisely is not the objective. The objective is knowing whether it is above break-even. So the requirement is this: the lower bound of the confidence interval must sit above fifty percent. Which means the margin of error has to be smaller than the edge itself.

Say that plainly, because it is the hinge of the entire article. A trader who wins fifty-five percent of the time has a five point edge. Give that trader a sample whose margin of error is five points and the interval runs from fifty to sixty. The lower bound touches break-even. The edge has not been established. The trader has done a great deal of work and learned nothing.

2.4 Worked, and then corrected

For a trader whose true win rate is fifty-seven percent: 0.57 times 0.43 is 0.2451. Multiplied by 1.96 squared, which is 3.8416, that is 0.9415762. The edge squared is 0.0049. Divide and you get 192.16, so 193 occurrences.

Now notice the trap in that derivation, because it is worth more than the number.

The closed form was obtained by setting the margin of error equal to the edge. In the Wald frame that places the lower bound exactly on fifty percent. It clears by construction, with zero margin. Test the same 193 occurrences against the Wilson interval and the lower bound comes out at 0.499458. It does not clear. It was never going to clear; the formula guaranteed a tie and a tie is not a pass.

Wilson requires 197. Clopper-Pearson, exact, requires 214. Three methods, one question, three answers, and the spread between them is larger than most of what gets argued about in this field.

2.5 The requirement across the range

† At a seventy percent win rate the normal approximation is no longer legitimate. Twenty-five occurrences at that rate implies only 7.5 expected losses, below the conventional threshold of ten that any normal-approximation method requires. The Wilson figure is shown for continuity and should not be relied on. The exact figure governs that row: 33 occurrences, and the capital column reports it, because Clopper-Pearson makes no normal approximation and carries no such condition.

Note the shape of the Wilson column. The requirement is not flat and it is not linear. It explodes toward the left. The better the trader, the smaller the sample needed, and the relationship is severe rather than gentle, because the edge enters the formula squared.

Note also what the correction cost. Moving from the textbook interval to the better one added exactly four occurrences at every win rate in the band. Small, uniform, and the difference between a figure that survives inspection and one that does not.

---

PART III. THREE ROADS TO ONE NUMBER

One derivation is a claim. What follows is why I think this one is more than that.

3.1 The second road: a different question entirely

Estimation asks how precisely a trader knows their own rate. A different discipline asks something else: can this record be distinguished from chance? That is hypothesis testing, it uses a different formula with different assumptions, and it is a fair question to put to a trading record.

Ask it of a sixty percent system against a fair coin, at ninety-five percent confidence and eighty percent power. The z terms are 1.96 and 0.8416. The square root of 0.6 times 0.4 is 0.4899. That gives 1.96 times 0.5, which is 0.98, plus 0.8416 times 0.4899, which is 0.412298. Their sum squared is 1.938494, divided by the difference squared, 0.01, giving 193.85. 194 occurrences.

These two formulas are not the same tool and neither is a rederivation of the other. One estimates a parameter and one tests a hypothesis. They answer different questions with different machinery and they land three occurrences apart.

3.2 The third road: the desk

The third figure did not come from a formula. It came from a working trader's own reflection on how many occurrences a dataset needs before its averages settle, arrived at independently, with parts of it estimated by eye rather than calculated. That figure is 180.

I am publishing it as what it is, with its provenance stated, because a number's origin is part of its evidence and hiding a rough estimate inside a table of computed ones would be dishonest. It was not derived. It was judged.

The two derived figures sit 1.5 percent apart, having been reached through different formulas resting on different assumptions. The judged figure sits about nine percent below both, which is a respectable showing for a number arrived at by eye.

So the answer to the article's question is yes, in part. Twenty-five thousand dollars supported roughly 180 to 200 occurrences at about a hundred and thirty dollars apiece, and a sample of that size establishes an edge for a trader at approximately fifty-seven percent and above. Fifty-seven percent is a defensible place to draw a line between a trader who has an edge and one who does not.

Whether anyone ever performed that arithmetic before choosing the figure, I do not know and do not assert. What the arithmetic shows is that the number lands somewhere a careful person could have chosen on purpose. That is a narrower claim than the ones usually made about it, in both directions, and it is the only one the evidence supports.

---

PART IV. WHERE IT STOPS WORKING

Now read the same table one row up, and the picture changes completely.

4.1 The marginal trader

A trader at fifty-five percent needs 385 occurrences on the Wilson interval, funding each with $64.94. But clearing break-even by a hair is a thin standard; the interval's lower bound sits barely off the line. Ask instead that the edge be twice the noise, which is a modest requirement and closer to how a professional would want to see it. For a five point edge that means the lower bound must clear 52.5 percent.

1,533 occurrences. Divide the threshold and each one carries $16.31. The textbook interval would have told you 1,522 and $16.43; it understates the requirement here as it does everywhere else.

Ask the obvious question and let it sit: what does sixteen dollars and thirty-one cents buy?

Not enough shares of anything liquid for the bid-ask spread to be a rounding error rather than a material cost. In the options market it buys a contract far enough out of the money that its delta is too low to capture an ordinary move, while theta erodes what was paid for it during the wait. Cheap contracts are cheap for a reason, and the reason is that they usually expire worthless. The budget selects the instrument, and the instrument the budget selects has the worst expectancy available.

There is a second cost, less often counted. At five trades a day, 1,533 occurrences is 307 trading days, which is more than a full trading year. At two a day it is over three years. The 197 occurrences a competent trader needs takes about 39 trading days at the same pace. The constraint was never only capital. It was also time, and the two multiply.

4.2 And below that, the mathematics simply refuses

The edge sits in the denominator, squared. Push it toward zero and watch.

At fifty percent the required sample is infinite, and it is infinite for a reason that has nothing to do with regulation: there is nothing there to measure. No sample size establishes an edge that does not exist. A trader without one is not being held back by a capital requirement; they are in the wrong game, and the arithmetic will not rescue them at any account size, in any jurisdiction, under any rule.

That cuts the other way too, and honesty requires saying so. The same symmetry means a trader who is genuinely losing at forty-five percent needs the same sample to establish that. Denied it, they never found out. Whether the threshold protected such a trader's capital or merely withheld the verdict is a fair question, and this article does not pretend to answer it.

---

PART V. THE ASSUMPTION, AND WHAT IT ALL COMES TO

5.1 Where the one-to-one assumption bends

Everything above fixed reward and risk at one to one, which put break-even at fifty percent. In general it is not.

Break-even is one divided by one plus R, where R is reward over risk. At two to one it falls to 33.3 percent; at one to two it rises to 66.7 percent. Every figure in this article moves with that, because the edge is measured from break-even and the edge is what sits in the denominator. A trader working at two to one and winning forty percent of the time holds a 6.7 point edge, and lands in the same neighbourhood by a different route.

The structure does not change. Only the coordinates do.

5.2 What it all comes to

The old rule set a capital floor, and by doing that it set a sample ceiling. It has been removed. So ask the question that actually matters now: did removing it solve the problem, or only stop enforcing it?

Take the 197 occurrences that certify a competent trader, and run them against accounts of different sizes:

That last figure is worth a moment. Ten dollars and fifteen cents, at the smallest account the new rules permit on margin, sits within a quarter of the $10.41 the arithmetic produced for a fifty-two percent trader holding the full twenty-five thousand. Two entirely different starting points, one destination, and that destination is a position too small to be worth executing.

The arithmetic never cared about the rule. It cares about capital divided by occurrences, and it binds a two thousand dollar account far harder than it ever bound a twenty-five thousand dollar one. The gate has come down. The mathematics that made the gate meaningful is exactly where it was, and it now applies to a larger group of people who will meet it without a number posted on the door to warn them.

For twenty years the argument about that threshold was conducted as a dispute about fairness, and the participants divided into people who thought it protected the naive and people who thought it protected the incumbent. Both sides were arguing about a barrier. Neither side, so far as I can find, sat down and worked out what the barrier bought.

It bought a sample. Adequate for a good trader, hopeless for a marginal one, and irrelevant to someone with no edge at all. That was true the whole time, it was computable the whole time from statistics any undergraduate holds, and it was the one thing nobody said.

Which is the only lesson here worth carrying anywhere else: an argument that runs for twenty years without anybody putting a number on the thing is not a hard problem. It is an unattempted one.

---

DISCLAIMERS

1. This article is not investment, financial, legal, or tax advice, and nothing in it is a recommendation to trade, to adopt any strategy, or to take any position. It analyses a statistic and a regulation.
2. No claim is made about anyone's intent in setting, maintaining, or retiring the threshold discussed. I have no document stating a rationale for the figure and I do not assert one existed. The article's claim is functional and nothing more: this is what the number bought, whoever chose it and for whatever reason.
3. The authorship of the original rule is not established here. I could not confirm from any primary record who proposed the twenty-five thousand dollar figure or on what basis, and I will not name a person on a claim I cannot support.
4. Headline figures are computed on the Wilson score interval, with the Wald and Clopper-Pearson results shown alongside wherever they differ. An earlier draft of this analysis used the Wald interval alone; those figures were four occurrences short at every win rate and are corrected here. The method is named at every point because the method changes the answer.
5. The seventy percent row is reported on the exact method only. Any normal-approximation interval is unreliable at that sample size, and the article says so rather than presenting a number it does not trust.
6. Every figure is computed in the text from the formulas shown, under the assumptions stated in Part II. Where a number is an estimate rather than a calculation, notably the 180 in section 3.2, it is labelled as an estimate at the point of use.
7. The regulatory description in Part I is taken from FINRA's own published guidance, cited below. Firms may impose requirements stricter than the rule, and the transition period means practice may differ between firms until October 2027. Readers should confirm current requirements with their own broker rather than with this article.
8. The distinction between a rule as written and a rule as enforced is real, and this article addresses only the former, because the former is documented and the latter is not.

---

FAIR USE NOTICE: Under Section 107 of the Copyright Act of 1976, allowance is made for fair use for purposes such as criticism, comment, news reporting, teaching, scholarship and research.

Source Citations

Ordered by authority: regulators first, then the statistical literature. The arithmetic in this article is derived in the text; the citations below establish the interval methods it rests on.

I. Regulators

• Financial Industry Regulatory Authority. "Intraday Margin Requirements." FINRA Investor Insights. The source for every regulatory statement in Part I: the 4 June 2026 effective date, the transition period through 20 October 2027, the removal of the $25,000 minimum equity requirement and of the pattern day trader designation by trade count, the intraday maintenance requirement of 25 percent held throughout the trading day, and the $2,000 minimum for leveraged trading. finra.org/investors/insights/intraday-margin-requirements

II. Interval methods

• Wilson, E. B. "Probable Inference, the Law of Succession, and Statistical Inference." Journal of the American Statistical Association 22, no. 158 (June 1927): 209-212. The score interval used for every headline figure in this article. tandfonline.com/doi/abs/10.1080/01621459.1927.10502953
• Clopper, C. J., and E. S. Pearson. "The Use of Confidence or Fiducial Limits Illustrated in the Case of the Binomial." Biometrika 26, no. 4 (December 1934): 404-413. The exact interval, computed from the binomial distribution with no normal approximation; the governing method for the seventy percent row. academic.oup.com/biomet/article-abstract/26/4/404/291538

III. Standard results used without citation

• The variance of a Bernoulli trial, the standard error of a sample proportion, and Bessel's correction (Figs. 01 and 02). Elementary; found in any introductory mathematical statistics text.
• The two-variance sample size formula for a one-sample test of a proportion at specified significance and power (section 3.1), used here at 95 percent confidence and 80 percent power.

Note on sources not cited

I looked for a published rationale for the twenty-five thousand dollar figure and did not establish one from any primary record. That absence is reported rather than filled. A number of accounts of the rule's origin circulate in secondary trading commentary, including named attributions of authorship; none is cited here, because I could not verify any of them, and a citation I have not seen is not a citation. If a reader can point me to the primary document, I will publish the correction and credit them.

Assumptions in force throughout: reward to risk of one to one, break-even at fifty percent, the win rate treated as constant, 95 percent confidence, and trials treated as independent. Real trading returns are not perfectly independent, and dependence would increase the sample required rather than reduce it; the figures here are therefore the optimistic case.

Figures set by the author. Formula plates are original work.