How to Size a Position Around an Accounting Red Flag

An accounting red flag is a disclosed fact that widens the range of outcomes you can reasonably expect from a holding. That's the definition I work with, and it's deliberately narrow. A red flag is not proof of fraud, and it's not a forecast that the stock will fall. It's a signal that the numbers you used to build conviction are now less reliable than they were last quarter — receivables growing faster than sales, earnings that stop turning into cash, an auditor who walks away, a non-GAAP figure drifting further from the GAAP one every year.

Defined that way, the portfolio question changes shape. If the flag widens the range of outcomes, then the job isn't to guess which end of the range you'll land on. The job is to make sure that the bad end of the range is survivable at the size you hold. That's a position sizing decision, and it has more than two answers.

Most investors treat it as a binary: sell now or ignore it. I think both reflexes are expensive. So this post compares the three real responses side by side — trim, hold, exit — and then gives a framework for choosing among them. I'm going to treat the flag itself as a given input. Diagnosing whether a flag is real is a separate skill, and our guide to spotting accounting red flags covers it far better than I could here.

Trim, Hold or Exit: Three Responses Compared

Each response answers a different question. Trimming says: I still believe the thesis, but I no longer believe it at this weight. Holding says: the flag doesn't change what I'd pay or how much I'd own. Exiting says: I can no longer underwrite the numbers at all, and a business I can't underwrite has no place in the portfolio at any size.

Trimming

Trimming is the response most people skip, and it's the one I'd reach for most often. It caps the damage if the flag turns out to be the first sign of something worse, while keeping you exposed if it turns out to be noise. The arithmetic is simple. Take a hypothetical portfolio where one holding is 8% of capital. If that stock falls 50% after a restatement, the portfolio loses 4 percentage points. Trim it to 4% when the flag appears and the same event costs 2 points. You've halved the loss without having to be right about the outcome.

The cost of trimming is the mirror image: if the flag clears and the stock recovers, you participate at half weight. That's a real cost. But it's a bounded one, and bounded costs are the kind a disciplined process should prefer.

Holding

Holding is right when the flag has an innocent explanation you can verify from the filings themselves — a receivables jump that traces to one large, creditworthy customer, say, or a cash flow dip that matches a disclosed inventory build for a product launch. Holding is wrong when the explanation comes only from management commentary, because the reliability of management's numbers is precisely what's in question.

Exiting

Exiting is right when the flag attacks the foundation of the thesis rather than one input to it. An auditor resignation, a delayed annual report, or a restatement of revenue all fall in that category: they don't tell you a number is a bit off, they tell you that you may not know the numbers at all. The cost of exiting is that you crystallize a loss and, often, a tax bill, and you lose the option to benefit if the company clears its name. Sometimes that cost is worth paying. Not always.

Trim when the flag lowers your confidence. Hold when the filings explain it. Exit when the flag means you can no longer trust the filings.

A Case Study in Escalating Flags: Super Micro, 2023 to 2025

Super Micro Computer is a useful example because its flags arrived in stages, and each stage called for a different response. Its fiscal year ends in June, so fiscal 2024 is the year ended June 2024. None of what follows is a view on the stock today; it's a study of how the evidence accumulated.

Start with the working-capital picture. Super Micro's 10-K for fiscal 2023 reported revenue of $7.12 billion, net income of $640 million and operating cash flow of $664 million — cash conversion of about 1.04, computed as operating cash flow divided by net income. Earnings and cash agreed. Its 10-K for fiscal 2024 told a different story: revenue of $14.99 billion and net income of $1.15 billion, but operating cash flow of negative $2.49 billion. Cash conversion fell to roughly −2.16.

The balance sheet showed where the cash went. Accounts receivable rose from $1.15 billion at June 2023 to $2.74 billion at June 2024, up about 138% against revenue growth of about 110%. Days of sales in receivables, computed as receivables ÷ revenue × 365, moved from about 59 days to about 67. Inventory rose from $1.45 billion to $4.33 billion — roughly tripling. In a company doubling its sales into an AI server boom, some of that is explainable. But a $3.6 billion gap between reported profit and operating cash in one year is the kind of divergence between cash flow and earnings that should, at a minimum, cost a position some weight.

Then the flags escalated. In late August 2024 a short seller published a report alleging accounting problems, and the next day the company said it would delay its annual report. On October 30, 2024, an 8-K disclosed that Ernst & Young had resigned as auditor. That is the foundation-level event described above. The 10-K for fiscal 2024 was eventually filed in late February 2025 with a new auditor, and it disclosed material weaknesses in internal control.

Map that onto the three responses. The cash flow gap alone was a trim signal — a working-capital anomaly in a fast-growing business, worth sizing down but not, on its own, worth abandoning. The delayed filing raised the stakes. The auditor resignation was an exit signal for anyone who couldn't independently verify the numbers, because the firm paid to check them had declined to keep doing so. An investor who trimmed on the first flag and exited on the third took two smaller, deliberate decisions instead of one forced one.

And here's the part that makes this hard. The flags didn't end in a collapse of the business. Super Micro's 10-K for fiscal 2025 reported revenue of $21.97 billion, net income of $1.05 billion and operating cash flow of $1.66 billion, cash conversion back above 1. An investor who exited at the auditor resignation gave up that recovery. I don't think that makes the exit wrong. It was the correct response to the information available at the time, and a process that only looks right in hindsight isn't a process.

Persistence matters too, and the story isn't finished. Super Micro's 10-K for fiscal 2026 — the year ended June 2026 — reported net income of $2.23 billion against operating cash flow of negative $6.81 billion, a conversion ratio of about −3.05, with inventory at $12.90 billion. Whether that's a growth company funding a backlog or a recurrence of the old pattern is exactly the kind of question the flag can't answer for you. What it can do is tell you how much to own while you find out.

A Framework for Deciding How Much to Cut

I size the response along two dimensions: how severe the flag is, and how persistent it has been. Conviction in the underlying business then adjusts the result, but only in one direction.

1. Classify severity. Tier 1 is a single-metric anomaly with a plausible operating explanation (a receivables spike, one year of weak cash conversion). Tier 2 is a pattern or a change in method (a revenue-recognition policy change, a non-GAAP gap widening for three years running). Tier 3 is anything that questions whether the numbers can be relied on at all (auditor resignation, delayed 10-K, restatement, disclosed material weakness).
2. Count persistence. A flag that appears once is weaker than one that shows up in two or three consecutive annual reports. Move up one tier if a Tier 1 or Tier 2 flag has persisted for two or more years.
3. Set the response by tier. Tier 1: trim to roughly two-thirds of the current weight, or hold if the filings themselves explain the anomaly. Tier 2: cut to about half, and cap the position at your starter size until the pattern reverses. Tier 3: exit, or reduce to a size whose total loss you could absorb without changing anything else in the portfolio.
4. Apply conviction as a brake, never an accelerator. High conviction in the business can justify the gentler end of each range. It can't justify adding to a position while a flag is open, and it can't move a Tier 3 flag down to Tier 1.
5. Write down the exit condition for the flag. Decide in advance what would clear it — two quarters of cash conversion above 1, a clean audit opinion from a new firm — and what would escalate it. Then you're reviewing evidence against a rule instead of reacting to a headline.

The exact fractions are my defaults, not a formula. Your numbers should depend on how concentrated the portfolio is to begin with: in a 12-stock portfolio a single holding can easily be 10% of capital, and the case for cutting early is stronger than in a 40-stock one. Our piece on concentration vs. diversification goes into how that starting point changes the math.

I'll be honest about the weakest part of this framework. The tier boundaries are judgment calls, and the data on how often a Tier 1 flag escalates into a Tier 3 event is thin — most studies of restatements look backwards from the restatement, which tells you little about the base rate of flags that came to nothing. I've moved some flags between tiers over the years, and I expect I'll keep doing so.

Mistakes That Turn a Flag Into a Permanent Loss

The sizing framework fails mostly through behavior, not arithmetic. Four patterns show up again and again.

• Averaging down while the flag is open. A lower price feels like a better deal, but you're buying more of a set of numbers you've just decided to trust less. Wait for the exit condition before adding.
• Letting the purchase price set the decision. Whether you're up or down on the position has nothing to do with what the flag means. Size against the current weight and the current evidence.
• Treating a trim as a failure. A trim isn't half a sale you lacked the courage to finish. It's a full decision in its own right: lower confidence, lower weight.
• Never revisiting. A flag that cleared two years ago should stop costing the position weight. If the filings now show the problem resolved, rebuild toward your conviction-based target deliberately, as you would with any new idea.

That last point gets missed most. Decision discipline cuts both ways: the process that shrinks a position when the evidence weakens should be the same one that rebuilds it when the evidence recovers. If you need a refresher on setting those targets in the first place, start with what position sizing is and then sizing by conviction.

Key Takeaways

• An accounting red flag widens the range of outcomes. The right response is to make the bad end survivable at your size, which usually means trimming rather than an automatic sale.
• Hold only when the filings themselves explain the flag; exit when the flag calls the reliability of the filings into question, such as an auditor resignation or a delayed 10-K.
• Size the response by severity and persistence, and let conviction in the business soften the cut but never reverse it.
• Write down what would clear or escalate the flag, and rebuild the position when the evidence recovers. Super Micro's fiscal 2023 to 2025 filings show why both halves of that rule matter.