Bed Bath & Beyond paid $53.32 a share for stock it later cancelled for nothing.

SECTION 01What the shares cost

Between fiscal 2008 and fiscal 2019, Bed Bath & Beyond bought 163,780,000 of its own shares and paid $8,732,165,000 for them. That is an average of $53.32 a share.[1]

On September 29, 2023, every one of those shares was cancelled without consideration.[11]

$53.32 a share, going in. Zero, coming out. There is no later price to argue about, because the last one is the final one.

That is what makes this worth computing rather than estimating. Most arguments about a share repurchase are arguments about what the stock will do next. This company has no next. The shares ended at zero, so the average price paid is not a snapshot, it is the whole result.

Read the average price column downward. The company bought most heavily when its own stock was dearest and had nearly stopped by the time it was cheap. 78.5 percent of the money, $6,854,413,000, went out at $50 a share or more. In fiscal 2019, with the shares at $14.66, it spent $99.7 million.

None of that required an opinion about buybacks. It is a purchase ledger with a known final value, and the arithmetic has one answer.

SECTION 02The sentence that repeated

On April 30, 2013, Bed Bath & Beyond filed its annual report for fiscal 2012. Gross profit was 40.2 percent of net sales, down from 41.4 percent, and the filing said why: an increase in coupons, and a shift in the mix of what people bought.[2] It was a good year. Net earnings were $1.04 billion, the highest in the sixteen years of filings we read.

The next annual report said the margin had fallen again. So did the one after that. And the one after that. Eight annual reports in a row, fiscal 2012 through fiscal 2019, each recording a gross margin lower than the year before, and seven of them naming the coupon in the explanation.[1]

Read the gross margin column on its own. From fiscal 2011 on, nine numbers, each smaller than the last, 41.38 down to 31.74, a run of 8 falls. Nearly ten points of margin on a twelve-billion-dollar business, gone at roughly a point a year.[1]

A point a year is the most dangerous rate there is. No single year looks like a crisis. Every year has a reason, and the reason is true.

The company did not hide it. It did the opposite. Three annual reports in a row, for fiscal 2014, 2015 and 2016, carried a sentence saying that operating margin compression was likely to continue in the coming year, and listing coupon expense first among the causes.[3][16][5] The signal was not buried in a footnote. It was a forecast, in plain words, three years running.

SECTION 03The years the top line hid it

Here is why it was easy to live with. Net sales rose every one of the first six years of the run, from $10.9 billion to its peak of $12.3 billion in fiscal 2017.[1] The store count rose too, to 1,552 that same year, the most in any filing we read.[6] Comparable sales stayed positive through fiscal 2015.

So for six years the two numbers an owner looks at first, sales and store count, said the business was growing. It was. It was also keeping less of every dollar it grew, and the two facts were printed on the same page.

Net earnings peaked in fiscal 2012, the first year of the run. For the next five years, sales were higher and profit was lower than the year before, every year. By fiscal 2017, sales were up 13.14 percent on 2012 and earnings were down 59.06 percent.[1]

Comparable sales went negative in fiscal 2016, down 0.6 percent, and the annual report filed on April 25, 2017 recorded it.[5] Then 1.3 percent, then 1.1 percent, then 6.8 percent.[6][7] By the time the top line turned, the margin had been falling for five years.

SECTION 04Is eight years normal?

Retail margins move. Mix shifts, freight moves, a bad winter clears at a discount. Before an eight-year run means anything, the ordinary run has to be measured.

So we ran the same arithmetic on fifteen other large US retailers, from the same SEC data, over the same fiscal years.[12] Net sales less cost of sales, as a share of net sales, each year as originally reported. Three of the fifteen also went through Chapter 11 in the period, Sears, J.C. Penney and Pier 1, and they are in the set on purpose, so the comparison is not only against companies that did well.[14][15] Thirteen carried enough years of data to count a run.

Across 168 retailer-years, the typical move in gross margin from one year to the next was a fall of 0.05 points. One year in ten it fell by more than 1.90 points; one year in ten it rose by more than 1.18. Declines were 53.0 percent of all years. Nobody strung more than five together.

The honest part of the control is what it does not say. The size of the fall is not unique. Over its own worst eight-year window Macy's lost 11.09 points, from fiscal 2012 to the pandemic year of 2020, and Pier 1 lost 10.74 from 2010 to 2018, before its own Chapter 11 filing in February 2020.[13] Both fell further than Bed Bath & Beyond's 9.64. Neither fell every single year.

Big drops happen. What does not happen is eight in a row. The finding is the run, not the depth.

SECTION 05Where the cash went

Over the sixteen fiscal years from 2007 to 2022, the company's operations produced $11.81 billion of cash. It spent $10.43 billion of that buying back its own shares, 88.4 percent, and $332 million more on dividends.[1]

The concentration is the striking part. In fiscal 2012 through 2016, the first five years of the falling-margin run, buybacks came to $6.18 billion against $5.83 billion of operating cash flow. That is 106.1 percent. The business bought back more than it made.

We are not here to argue about buybacks. Home Depot spent 61.7 percent of sixteen years of operating cash flow on them and is fine. Dollar General spent 62.7 percent and is fine. What none of the fifteen did, in any five-year window, was cross 100 percent. The closest was Pier 1 at 99.6 percent in fiscal 2011 to 2015.[12]

Then there is the year the bars are tallest. In July 2014, year three of the run, the company sold $1.5 billion of notes: $300 million due in 2024, $300 million due in 2034, and $900 million due in 2044, at 3.749, 4.915 and 5.165 percent.[4] Its annual report for the year says what the money was for: share repurchases. The same day the notes were issued, it entered a $1.1 billion accelerated repurchase and eventually received 16.8 million shares at an average price of $65.41.[3] Buybacks for the fiscal year came to $2.25 billion, the most in any of the sixteen years.

In fiscal 2014, when the borrowing happened, the margin had fallen for three years and the annual report filed the following April said it would fall again.[3] The information and the decision were in the same document.

SECTION 06The clock

So here is the shape of the thing, measured in days from the first annual report that recorded the margin falling.

Ten years, less a week. And the first row is not the earliest the signal existed. It is the earliest the company wrote it down for the public, in a document filed once a year. The filing is annual. The register is not.

Detection was never the constraint. The number was known, printed and forecast. What the run measures is the distance between a fact being available and a fact changing what happens next, and that distance is not a data problem at all.

SECTION 07Two numbers, one document

Here is the thing worth sitting with. Both halves of this story were printed in the same annual report, in the same spring, by the same company.

The fiscal 2014 filing says gross margin fell again, names the coupon, and forecasts more compression next year. The same filing says the company spent $2.25 billion buying its own shares, the largest of any year, at a blended $68.30 each.[4] Two facts. One document. A reader could reach both in the same sitting without turning on a computer.

Nothing was hidden and nothing was missing. The two numbers simply never appeared in the same sentence.

That is the specific thing a data team does, and it is narrower and duller than the job description suggests. Not finding a secret. Putting two visible things side by side and asking whether they can both be true at once.

They usually can, for a while. A business can have a thinning margin and still be worth investing in. What the pairing does is set a clock on that argument, and the clock was readable from the day of the filing.

The version of this in a business with no floor of analysts

It is never a buyback. It is the pair that lives in two different systems and therefore belongs to nobody.

• What you invoiced against what you actually collected, one in the job system and one at the bank.
• What a job quotes at against what it costs by the time it closes, one in the estimate and one across four suppliers.
• What a customer is worth against what you paid to get them, one in the till and one on a card statement.
• Headcount against the work that arrived, one on the payroll and one on the schedule.

Each half is visible to somebody. Neither half is wrong. They are in different rooms, so the question that needs both of them is the one nobody is holding.

SECTION 08The point

$8.73 billion at an average of $53.32 a share, for stock that ended at zero, while the same filings said the business was keeping less of every dollar it sold, eight years running.

The expensive part was not that anyone lacked the numbers. It was that the two numbers that had to be read together were being read apart, by different people, for different reasons, in a company with more analysts than most businesses have employees.

Which is worth knowing if you are the one holding all of it yourself. The pairing does not take a floor of people. It takes somebody whose actual job is to ask what two of your own numbers say when you put them next to each other, and then to keep asking every month.