That post made us wonder: what other companies have similar off-balance sheet arrangements, and in what amounts? 

Typically that information is hard to find. Analysts need to sift through a company’s quarterly report, and the number is usually buried away at the bottom of a page somewhere in a footnote near the end of the filing. That’s how Oracle disclosed its $99.8 billion, for example. 

Thankfully, Calcbench can help you leapfrog all that work to get to the good stuff. All those off-balance sheet disclosures are also tagged in XBRL. So if you just search by XBRL tag — which Calcbench lets you do — the numbers pop up quickly and easily.

The exact tag is:

 "UnrecordedUnconditionalPurchaseObligationBalanceSheetAmount"

We searched the S&P 500 for all firms that reported numbers using that tag, and did a time-series of data for the last seven quarters (that is, since the start of 2024) to see how those off-balance sheet commitments have changed over time. The result is Figure 1, below.

Hooo boy, that’s a lot of information for a single chart. Several points jump out.

First, notice that the tech giants are racking up lots of off-balance sheet commitments, almost all of it for data centers to run artificial intelligence. We thought Oracle’s $99.8 billion was eye-popping, but the numbers for  Amazon ($AMZN) and Microsoft ($MSFT) are even larger. And while Meta’s ($META) off-balance sheet numbers are lower, those numbers are growing rapidly. 

Second, it’s interesting to see that tech companies involved in the AI arms race aren’t the only ones carrying off-balance sheet commitments, but those other companies incur off-balance sheet commitments in a very different way. 

United Airlines ($UAL), Delta Air Lines ($DAL), JetBlue ($JBLU), Alaska Air ($ALK) — they all have off-balance sheet commitments to lease aircraft sometime in the future, but the lease amounts are relatively steady. It’s much the same for Netflix ($NFLX), too: commitments for future content, but commitments that are relatively stable from one quarter to the next. 

So at least for the airlines, Netflix, and other non-tech giants, you can see a logic there. People will continue to fly and watch more seasons of “Bridgerton,” so the airlines and Netflix need to line up resources to meet that future demand. 

Amazon, Oracle, Microsoft, Meta, and other “hyperscalers” making bets on AI are different. It’s not clear that demand for AI will scale up to match those data center commitments, or that the macro-economics of AI will be sustainable over the long term. 

How to Find This Stuff

Finding the numbers for off-balance sheet disclosures is easy enough. As we said above, you can search by tag and quickly find all filers that report something using that tag.

For example, you can go to the Multi-Company search page and start typing “unrecorded” in the XBRL tag search field on your screen. A dozen possible tags show up; just select the disclosure that interests you, and you’ll see all filers who used it in the period you’re searching. See Figure 2, below.

You could run the same search in a more refined way on our Raw XBRL Data page, although this page assumes you already know the exact tag you want to search. (If you don’t know the name of the tag, one cheat code is to find that disclosure in a filing somewhere and click on the number, which should be hyper-linked. That will open a small box at the bottom of your screen that explains what the tag for that disclosure is. Copy that tag name and paste it into the XBRL search page.) 

Once you find that off-balance sheet number, there’s still the question of exactly what that number is about. That’s easy. You can use our Trace feature by clicking on the result and that will open a box to the exact footnote that explains what the number is about.

Figure 3, below, shows how it works. We searched for “Unrecorded Unconditional Purchase Obligations” (one of the common tags for off-balance sheet disclosures) among the S&P 500 in Q2 2025. Meta reported $52.56 billion worth of such obligations. We clicked on that amount, and a box opened to the footnote disclosure, which tells us that the amount is “mostly for data centers, certain network infrastructure, and colocations” — that is, data center stuff.

So what are the risks to all these off-balance sheet commitments? That’s a subject for another post, and other analysts. But Calcbench has all the data you need to ask (and hopefully answer) such questions.

Or maybe you already knew Oracle’s potential risk here because you’ve been studying those same disclosures yourself — disclosures that are all readily available in Calcbench.

Don’t get us wrong; we appreciate the FT’s article, which is packed with good points and insight. But consider the key metrics that the article cites:

• Segment-level revenues

• Debt levels

• Cash and short-term investments as a percent of total assets

• Free cash flow

• Leases

• Debt-to-equity ratio

Calcbench tracks all those metrics for public filers. So if you had wanted to run your own analysis of Oracle’s exposure to OpenAI, perhaps comparing that exposure to other AI “hyperscalers” such as Microsoft ($MSFT), Google ($GOOG), and Amazon ($AMZN) — well, we’ve had all that data all along, there for the picking.

Always in the Footnotes

For example, the article notes near the bottom that Oracle “has signed at least five long-term lease agreements for US data centres that will ultimately be used by OpenAI, resulting in $100 billion of off-balance-sheet lease commitments.”

That’s absolutely right: if you look at Oracle’s most recent annual report (filed last May), you’ll see $147.4 billion in total liabilities, and only $11.5 billion in operating lease liabilities. 

So where is that disclosure of $100 billion in off-balance sheet lease commitments? It’s in the footnotes! Specifically, the number is tucked away at the bottom of a note blandly labeled “Leases.” See Figure 1, below.

You’ll notice that the exact number is $99.8 billion, and that number is tagged in XBRL. This means that you can use our See Tag History feature to see how that number has changed over time. We did exactly that — and those Calcbench fans with heart conditions may want to sit down and take a few deep breaths before reading further. Ready? See Figure 2, below.

Holy poop, Oracle’s off-balance sheet leasing commitments have increased by 24,822 percent in five years — from $411 million in 2020, to $43.4 billion at the end of the company’s fiscal 2025 six months ago, to $99.8 billion as of Aug. 31.  

Oracle’s off-balance sheet commitments more than doubled in its summer quarter. Presumably all of that is due to the deal that Oracle reached with OpenAI in the same period, where OpenAI has agreed to purchase $317 billion of computing power from Oracle for years to come. 

That computing power has to come from somewhere, which means Oracle needs vastly more data center resources. Hence its leasing commitments are soaring. 

The company will file its next quarterly report in early December, and lord only knows what the off-balance sheet leasing commitments will look like then. But as soon as Oracle does file those numbers, Calcbench will have them indexed and ready for analysis within minutes. 

Other Metrics

So that’s the off-balance sheet commitments. We also mentioned a squadron of other performance metrics cited in the Financial Times article: cash, debt, free cash flow, debt ratios, and so forth. Where can you find those? 

One great place to start is our Bulk Data Query page, which lists just about every financial disclosure a company might ever make. That includes individual items on the income statement, balance sheet, and statement of cash flows; plus other non-standard disclosures and even important liquidity ratios that you’d typically need to calculate yourself. Calcbench has all that for you.

For example, if you go to the Bulk Data page and scroll to the bottom, you’ll see more than two dozen profitability, liquidity, and solvency ratios. Figure 3, below, lists them all, with debt-to-equity highlighted since that one was mentioned in the FT article.

All these metrics can be calculated and then exported to your desktop in a tidy spreadsheet. You can then fiddle with them to your heart’s content, especially if you have our Excel Add-in. Alternatively, power users can have the data piped directly into your own models using our API. If you need help with either of those, email us at us@calcbenc.com any time.

Our point is simply that you don’t need to wait for the business press to write an in-depth article exploring possible AI bubbles or any other corporate financial scenario. If the data is out there, we have it, and you can consume it as soon as you’re ready.

As always, Calcbench is on the case. 

Let’s start with Microsoft ($MSFT), which has been investing in OpenAI for years, primarily via cloud computing resources Microsoft provides to the generative AI startup. In its quarterly report from Oct. 29, Microsoft reported a $4.1 billion loss from those investments, up from a $688 million loss in the year-earlier period. See Figure 1, below.

We were intrigued. If Microsoft is reporting a $4.1 billion loss on OpenAI today and a $688 million loss one year ago, what other losses has it disclosed related to OpenAI in the past? 

Alas, we can’t really tell. Microsoft only started reporting OpenAI-specific disclosures one year ago, at the start of its fiscal 2025; that’s the $688 million loss, reported under the “Other Income, Net” line item. Prior to that period, Microsoft didn’t mention OpenAI losses at all.

Even starting in Q1 2025, however, Microsoft described the OpenAI losses in rather vague terms. Microsoft reported an Other Income loss in that period of $683 million, “primarily reflect[ing] net recognized losses on equity method investments, including OpenAI.” 

The word “primarily” is doing a lot of work in that sentence. Investors at that point in time didn’t know whether “primarily” meant almost all of the $683 million loss, or 51 percent of it, or something in between. Only now, one year later, can we see that “primarily” meant essentially all of the Other Income net losses that quarter. 

If we chart Microsoft’s Other Income losses from Q1 2025 (when we get the first mention of OpenAI’s relevance to that line item) forward, we get Figure 2, below.

Again, we don’t know precisely how much OpenAI was responsible for those losses throughout the first four fiscal periods, but clearly it was a lot — and then losses soared in this most recent quarter, to the point where Microsoft started disclosing exact numbers. 

What Else Microsoft Is Saying

We also used the text search feature in our Disclosures & Footnotes Query page to see what else Microsoft has said about its exposure to OpenAI. Tucked away in the Management Discussion & Analysis of last week’s 10-Q we found this (emphasis added by us):

Current year net income and diluted EPS were negatively impacted by net losses from investments in OpenAI, which resulted in a decrease in net income and diluted EPS of $3.1 billion and $0.41, respectively. Prior year net income and diluted EPS were negatively impacted by net losses from investments in OpenAI, which resulted in a decrease in net income and diluted EPS of $523 million and $0.07, respectively.

Umm, wow. Microsoft’s total diluted EPS for its most recent period was $3.72 — so if the company didn’t have that OpenAI loss, diluted EPS would’ve been 11 percent higher. 

Even more interesting is that OpenAI losses went from a $0.07 drag on EPS one year ago to a $0.41 drag today, a six-fold increase. By comparison, total EPS rose only 12.7 percent (from $3.30 EPS one year ago to $3.72 EPS now.) 

So those OpenAI losses are growing, and growing so fast that Microsoft has decided to start reporting them; which means Microsoft believes the OpenAI relationship is material to investors. Going forward, presumably Microsoft will keep disclosing those OpenAI losses — or, in theory, any OpenAI profits, if the company ever starts making money. 

Either way, it’s a useful glimpse into the state of OpenAI and its relationship to Microsoft. You just need to know where to look, and have the right tool to help you translate data into insight.

Debt instruments can have a huge effect on stockholder equity and EPS, depending on the precise nature of the debt that a company issues. Today let’s look at two examples of convertible debt, and how companies try to manage its attendant EPS implications, from the technology firms Snap ($SNAP) and Zillow ($Z). 

First, a refresher on how convertible debt works. Convertible debt is a financing instrument that combines elements of both debt and equity. While the interest rate on convertible bonds is generally lower than the rate on traditional bond instruments, investors gain the option to turn their debt holdings into company shares at some predetermined "conversion price" set in the terms of the debt agreement. 

For the shareholders, the downside of the conversion is dilution — because the conversion of debt into equity increases the number of shares outstanding, which therefore decreases the corresponding EPS.

One way to avoid that outcome is through a type of call option. A call option gives its holder the right (but not the obligation) to buy a specific stock or other asset at a predetermined price — commonly known as the “strike price” — within a set period of time. 

“Capped call” transactions are a hedging tool that convertible bond issuers often use to protect against shareholder dilution. When a company issues a convertible bond, it can buy call options on its own stock with a strike price equal to the bond's conversion price and a cap that limits the counterparty's exposure. 

A capped call transaction effectively raises the price point at which dilution begins, and that shields shareholders from having their ownership diluted until the shares trade well above the conversion price. The trade-off is cost: capped calls typically consume 7 to 9 percent of the debt proceeds upfront, as Calcbench explained in its previous piece using Uber’s ($UBER) $141 million hedge to illustrate. 

Importantly, capped calls involve separate transactions between the issuer of the convertible and third-party underwriters, subject to a separate agreement that lays out terms and conditions. But since capped calls are used to protect against conversion-related dilution, we typically expect the companies to unwind capped calls promptly once the convertible security is redeemed or expires upon maturity of the convertible. 

The logic here is simple: capped calls are costly, and unwinding them is a liquidity-positive event that generates cash inflow. So why pay for a hedge when the dilutive convertible instrument no longer exists? 

Example 1: Snap

In February and May 2024, Snap repurchased its 2025 convertible notes. Upon completion of the repurchase of convertibles in May 2024, Snap terminated its capped call transactions, recording a $62.7 million inflow from cash from financing. Snap disclosed the transaction as follows (emphasis added):

“Our financing activities for the six months ended June 30, 2024 primarily consisted of the Note Repurchases for $859.0 million, repurchases of our Class A common stock for $311.1 million, and the purchase of the 2030 Capped Call Transactions for $68.9 million, partially offset by the issuance of the 2030 Notes for net proceeds of $740.4 million and the termination of the 2025 Capped Call Transactions for proceeds of $62.7 million.”

But not every company reports a capped call transaction this way. Sometimes you find an outlier. Zillow seems to be such an exception.

Example 2: Zillow

In September 2019, Zillow issued a $500 million convertible note with maturity in 2026, a conversion price of $43.51, and the initial cap price of $80.575. The terms of the convertible and the capped call imply that Zillow purchased an anti-dilution hedge, which protects shareholders from dilution when the stock trades between $43.51 and $80.575. 

Put differently: the hedge effectively raises the dilution threshold to $80.575. On Sept. 27, 2019, Zillow’s stock closed at roughly $29.53, more than 30 percent below the initial conversion price. 

Five years later, in December 2024, Zillow redeemed its 2026 convertible note. In contrast to Snap, Zillow decided to keep the associated capped calls outstanding (again, emphasis added):

“In connection with settling our 2026 Notes in December 2024, we elected to keep the associated capped call transactions outstanding.”

The capped calls were terminated on Aug. 25, 2025, about eight months after the redemption of the convertible note:

“On August 25, 2025, Zillow Group, Inc. (the “Company”) will enter into agreements … to unwind and terminate certain capped call transactions by and between the Company and each Counterparty (such transactions, the “Capped Calls,” and the unwind and termination of the Capped Calls, the “Unwind Transactions”)... Upon settlement of the Unwind Transactions, the Company will have no remaining capped call transactions outstanding.

The Company expects to receive from the Counterparties an aggregate of 3.1 million shares of the Company’s Class C capital stock, which will reduce the Company's Class C capital stock outstanding, and $38.2 million in cash (the “Unwind Amount”), upon the Unwind Transactions. In connection with the Unwind Transactions, the Counterparties may buy or sell shares of the Company’s Class C capital stock in secondary market transactions and/or unwind various derivative transactions with respect to such Class C capital stock.”

Why did Zillow elect to delay the termination of capped calls? We don’t know. Recall, however, that the economics of the capped call transactions is a company buying a call option on its own stock. Buyers of the call options usually believe that the price of the underlying instrument is likely to increase, sending a bullish signal to the market.

One possible explanation is that Zillow's management believed that Zillow's stock price was likely to increase, and incorporated this belief into their decision not to terminate the calls. Zillow’s stock price closed at $76.6 on Dec. 24, 2024 — about 5 percent below the initial cap of $80.575. The stock closed around $88.80 on Aug. 25, 2025, an increase of 16 percent over the eight-month period. For comparison, the S&P 500 increased about 6.6 percent over the same period.

Let’s also remember that Zillow received both cash and shares as part of consideration received for unwinding the capped call transaction. Another possible explanation is that Zillow preferred to receive the shares — effectively reducing dilution — in 2025 rather than in 2024. 

Why would a company do that? Perhaps to match the timing of accretive capped calls unwinding with dilution related to the issuance of shares upon exercise of stock options. 

According to Zillow’s Q2 2025 filing, the company issued 1.32 million Class C shares upon exercise of stock options, 2.97 million shares upon vesting of RSUs, and repurchased 4.2 million Class A and 1.4 million Class C shares in the first half of 2025:

“During the six months ended June 30, 2025, we repurchased 4.2 million shares of Class A common stock and 1.4 million shares of Class C capital stock at an average price of $70.09 and $73.19 per share, respectively, for an aggregate purchase price of $297 million and $103 million, respectively.”

The company also reported the transaction in the following table:

For comparative purposes, during the first six months of 2024 Zillow repurchased 1.1 million of Class A stock and 5.996 million of Class C stock. So the 3.1 million of Class C capital stock received by Zillow in August 2025 as part of the consideration could have potentially reduced the buybacks needed to mitigate the dilutive effect of options issuances and RSUs vesting in 2025. 

Of course, there could also be other reasons besides stock price expectations and buyback considerations that we did not discuss here. But Zillow’s delay in unwinding the hedge looked unusual, certainly interesting enough to ponder a few ideas. 

For the readers interested in conducting similar research in their own companies that they follow, convertible debt, capped calls, and note hedges are all easy to find using Calcbench’s Interactive Disclosure and Multi-Company pages.

For example, we searched “capped call” in the 2024 disclosures of S&P 500 companies and found more than 75 results, including NRG Energy, Super Micro Computer, Las Vegas Sands, and more. From there you can conduct your own analysis much as we did here with Snap and Zillow.

Editor’s note: Olga Usvyatsky is an author of Deep Quarry newsletter and occasional contributor to the Calcbench blog. Usvyatsky enjoys raising interesting questions about financial disclosures, and can be reached at olga@deepquarry.com.

Then, as always, Calcbench squinted more closely at that $28.2 billion in pretax income. Only $17.4 billion of that number came from Amazon’s operating income; another $10.2 billion came from the ever-mysterious line item known as “Other” income. What was that about? 

So we dove into Amazon’s footnote disclosures, because you can do that in Calcbench thanks to our Tracing feature and our Disclosures & Footnotes Query page. Soon enough, we found this explanation on Page 30 of the 10-Q, tucked away in the Management Discussion & Analysis:

The net gain of $10.2 billion in Q3 2025 and $14.1 billion for the nine months ended September 30, 2025 is primarily from an upward adjustment for observable changes in price relating to our nonvoting preferred stock in Anthropic, and the reclassification adjustments for the gains on available-for-sale debt securities from the portions of our convertible notes investments in Anthropic that were converted to nonvoting preferred stock during Q3 2025 and for the nine months ended September 30, 2025.

In other words, more than one-third of Amazon’s pretax income for the quarter ($10.2 billion of the $28.2 billion total) comes from its holdings in AI darling Anthropic — a nonpublic company that reportedly doesn’t yet turn a profit.

Don’t get us wrong. Amazon still had $17.4 billion in operating income, and that’s not nothing. Moreover, even if Anthropic isn’t turning a profit, the company is growing like weeds along the Amazon River. Whatever Amazon’s equity or debt holdings in the AI might be, clearly those holdings are worth more today than they were three, six, or nine months ago. Regardless, that’s still a material amount of Amazon’s pretax income tied to financial engineering rather than people buying and selling stuff. 

So we got to wondering: Are any other companies enjoying similar bumps to the bottom line from “other” income?

Spoiler: the answer is yes.

Exploring Other Income

Researching this question was easy. We simply went to our Multi-Company search page and then looked up operating income and “Other non-operating income” for the S&P 500 to see what we’d find.

For Q3 2025 filings so far, we found 178 firms that filed both numbers. Then we expressed Other Income as a percentage of Operating Income, and sorted the answers from highest to lowest. Figure 1, below, shows the 10 firms with the largest Other Income amounts relative to Operating Income.

That’s the list; then you can use the Disclosures & Footnotes Query page to investigate specific companies and why they had such bumper crops of other income.

For example, AT&T ($T) reported $6.25 billion in other income for the quarter, more than all of the telecom giant’s operating income. Most of that $6.25 billion came from the sale of AT&T’s stake in DirecTV to a private equity group:

On July 2, 2025, we sold our interest in DIRECTV to TPG Capital (TPG) and recorded a current note receivable of approximately $3,600, which we expect to receive the majority of by the end of 2025, and a long-term receivable of $500. The disposition of DIRECTV also resulted in the release of approximately $2,900 of historical deferred tax liabilities. We recorded a gain on the sale of DIRECTV of approximately $5,500, which includes the impact of the transfer of deferred tax liabilities, indemnification liabilities and unfavorable contracts…

AT&T reported net income of $9.67 billion for the quarter. If not for that $5.5 billion booked thanks to the DirecTV sale, net income would’ve been less than half of that.

Another adventure in footnote disclosures is Honeywell International ($HON). The company reported $822 million in Other Income, compared to $1.75 billion in operating income. First we traced that $822 million to an Other Income footnote that Honeywell included in its Q3 filing, which included a table (see below) that said most of that Other Income number came from a settlement related to Resideo, a maker of home temperature and security controls that Honeywell spun out in 2018. 

The Other Income footnote directs people to look at the Commitments and Contingencies footnote for more detail about the Resideo issue. We went to that footnote. Long story short, the $802 million is a one-time payout from Resideo to Honeywell to end a long-term indemnification arrangement where Resideo had been paying Honeywell for the costs of environmental matters at old manufacturing sites. 

For the record: yes, Honeywell did report this one-time Other Income gain as an adjustment to net income. The company only reported the number as part of adjusted EPS, rather than adjusted earnings; but the adjustment is in there. 

We could keep going, and perhaps in future posts we will. Our point is simply that the impressive net income and pretax income numbers we’re seeing these days can often be driven by Other Income numbers. There’s nothing inherently wrong with that, but shrewd analysts will keep that question in mind as you assess the long-term earnings quality of the companies you follow. 

All the data is there; Calcbench just helps you find it.

The template, available to Calcbench subscribers on DropBox, tracks numerous airline-industry performance metrics — everything from revenue per available seat mile (RASM) to passenger revenue, load factor, average fuel cost, and more.

We track six U.S. airlines:

• Alaska Airlines ($ALK)

• American Airlines ($AAL)

• Delta Air Lines ($DAL)

• Jet Blue ($JBLU)

• Southwest Air ($LUV)

• United Airlines ($UAL)

Figure 1, below, tracks their quarterly EPS (along the bottom) and quarterly RASM (along the top). 

Or for those of you who prefer to consume your financial data in table format, here’s the RASM data for the last five quarters in Figure 2, below.

Why did we include Q4 with a row of “N/A”? To remind you that our earnings templates automatically capture the latest data as firms file! So when the airlines start filing Q4 data in three months’ time, that information will be whisked from the filings to the templates within minutes, ready for your analysis.

Two caveats. First, for the template to populate updates automatically, you must be a premium-level Calcbench subscriber. If you want a Premium subscription or aren’t sure whether the subscription you have qualifies, email us at us@calcbench.com and we’ll get you squared away. 

Second, you must have the Calcbench Excel Add-in installed and running. 

Once you have those items done, the template will run automatically. Onward and upward!

All banks must set aside some portion of capital to cover loans they’ve extended which might subsequently default. Those are loan loss provisions, disclosed in absolute dollars. You can then calculate a bank’s loan loss ratio by dividing the loan loss provisions into total loans outstanding. 

That ratio gives investors a sense of how confident a bank is about its loan portfolio, and really about the economy overall. A declining loan loss ratio means banks aren’t as worried about the risk of default among their loan recipients; a rising one means the banks are.

So what does that picture look like today? We pulled the results for five major banks: Bank of America ($BAC), Citigroup ($C), JPMorgan Chase ($JPM), US Bank ($USB), and Wells Fargo ($WFC). Figure 1, below, tells the tale.

As one can see, the banks’ loan loss ratios for Q3 2025 (in red) were all lower than their loan loss ratios in the year-earlier period (in blue). Some were only marginally lower, others materially lower — but across all five, lower.

For those curious about the precise amounts set aside for loan loss provisions, we present Figure 2, below.

Again, these numbers are easy to find and analyze in Calcbench. We used our Multi-Company page, where you can create peer groups of companies to study and then pull up any of scores of standard disclosures we track — including numerous bank-specific disclosures, such as loan loss provisions and total loans. 

Hardcore Calcbench subscribers can also use our API to mainline earnings data directly into your own models; or you can use our Disclosures and Footnotes Query page to drill down into a single company’s detailed disclosures. 

After all, these loan loss ratios are only one performance metric among many that banks report about the performance of their loan portfolios. Banks also report loans by category (consumer credit card loans, other consumer loans, commercial loans, real estate loans, mortgage loans, auto loans, and so forth), deposits, and a host of other metrics. 

For example, Figure 3 below, is one small snippet of the disclosures from JPMorgan’s Q3 earnings release.

It’s a lot of data that banks publish. Calcbench captures it all and has it ready for your analysis.

OK, cool cool, we have the data. So how can Calcbench users actually find it? Today we’ll give you three easy-to-follow examples. 

For a Single Company

Let’s say you are researching a single company’s disclosures. Typically, the place to start is the Disclosure & Footnotes Query page. You can find that by scanning your choices along the top of the screen and finding the one that says “Disclosures.” See Figure 1, below (and look for the red arrow).

Once at the Disclosure & Footnotes page, you’ll next need to select the executive compensation disclosures as the specific data you want to see. Those disclosures are filed annually in the proxy statement, not in the quarterly financial statements or the annual 10-K. 

To find them, simply look to the left-hand side of your screen. You’ll see the list of disclosures that Calcbench offers. See Figure 2, below.

Skim all the way to the bottom of that list of disclosures, and you’ll see the compensation disclosures at the bottom under “Related Documents.” See Figure 3, below. 

Click on the choice you’d like, and that data will appear on your screen.

For Groups of Companies

Calcbench subscribers can also search larger groups of companies for compensation disclosures too. First, select the peer group you want to research. (We have a separate post explaining how to select a peer group, or you can use one of our pre-designed groups such as the S&P 500 or the Dow Jones Industrials.) 

For our purposes today, we’ll use the Dow Jones Industrials. Then you go to our XBRL Data page, which lets you search for disclosures by their individual XBRL tag (since all compensation disclosures are tagged). 

Step 2 is to select the fiscal year and period you want to research. You can select any fiscal year you’d like (although the pay versus performance disclosures are new, so if you search prior years you’ll get no results because none exist). For the fiscal period, however, you must select “Y” for the whole year because compensation disclosures only exist in the annual proxy statement.

Step 3 is to select the “statement type” you want from the drop-down menu provided. Scroll down until you reach “proxy statement pay versus performance” and select that choice. 

Then just press search, and you’ll get the results. In our example above, searching the Dow Jones Industrials for their 2024 proxies, we had 866 facts or data points. From here, you can export the data into Excel for your own modeling. That’s all there is to it. 

As a warm-up exercise for data analytics superpowers, the Calcbench analytics team pulled together a quick comparison of return on equity (one of the most important performance metrics banks report) for all three banks for the last eight quarters. See Figure 1, below. 

As you can see, JPMorgan has the best performing ROE overall, but Goldman has the best-performing trendline even though its ROE is starting from a lower base. And we also have Wells Fargo, finally unshackled from years of tight regulatory oversight and trying to put some pep back in its ROE step. Looks like Wells still has a ways to go on that front.

We built the above chart in about three minutes. Start from the Recent Filings page (which typically captures filing data within minutes of the information hitting the Securities and Exchange Commission), and then pull up individual firms’ earnings data on the Disclosures & Footnotes Query page. From there you can find specific ROE disclosures, and then use the Show Tag History feature to see prior periods’ disclosures for the same metric. Paste all that data into Excel and you’re done!

Of course, Calcbench subscribers can also mainline earnings data directly into your pre-existing models using our API or the Calcbench Excel Add-in; or you can search for disclosures across multiple companies all at once using our Multi-Company page. 

You get the idea: Calcbench has the data, and lots of different ways to find it and study it within a few keystrokes. Now get back to work studying more filings!

We can start with the top line numbers on the income statement, which all looked good. Revenue was $16.67 billion, up 6.35 percent from the year-ago period. Operating expenses were up only 4.96 percent, pretax income was up 13.8 percent, and net income popped 11.4 percent. See Figure 1, taken from our Company-in-Detail page. 

Cool stuff, but we love airline earnings releases for all the non-GAAP metrics they also report — and which Calcbench captures and indexes too.

Most notably, airlines report total revenue per available seat mile (TRASM) and cost per available seat mile (CASM) to keep investors informed about how efficiently the business is getting paying customers into seats. Figure 2, below, charts Delta’s performance on both metrics since the start of 2021.

In fact, let’s have some fun and calculate “profit per available seat mile,” which presumably would be known as PASM. This isn’t a metric Delta or other airlines actually report, but you can calculate it easily enough by subtracting CASM from TRASM. Hence we arrive at Figure 3, below.

PASM might seem volatile from quarter to quarter, but that trendline in red has a nice upward slope. 

For Calcbench subscribers who are diehard airline analysts, we also have our Airlines Earnings Template. The template is a spreadsheet available on DropBox that tracks numerous disclosures in the airline sector, including:

• TRASM, or revenue per available seat mile

• CASM, costs per available seat mile

• Load factor, which is the percentage of seating capacity filled by customers

• Fuel consumed

• Average fuel cost per gallon

• Percentage of revenue coming from passengers

• EPS

The template populates automatically with the latest data as airlines file their numbers; we should have Q3 data from all the big players before Halloween. The template only works if you are (a) a Calcbench professional-level subscriber; and (b) have installed our Excel Add-in — but once you do that, you’ll have all the latest data at your fingertips. (If you need help with any of that, contact us at us@calcbench.com.) 

Congrats to Delta. Many more Q3 releases will arrive soon!

By Stephen F. O’Byrne

Pay versus performance (PvP) disclosures give investors the ability to measure key pay dimensions for public companies and the executives leading those businesses. Investors can do this first by calculating relative cumulative pay and relative cumulative Total Shareholder Return (TSR); and then by calculating the regression trendline relating the natural log of relative pay to the natural log of relative TSR.

Such a regression analysis provides insight into four “dimensions” of executive pay. 

• The slope of the trendline provides a measure of incentive strength, what I call “pay leverage.” Pay leverage is the percentage change in relative pay associated with a 1 percent change in relative shareholder wealth. 

• The correlation between the two provides a measure of alignment. 

• The intercept provides a measure of performance-adjusted cost — that is, the pay premium at peer group average performance. 

• The ratio of pay leverage to pay alignment (that is, the slope divided by the correlation) provides a measure of relative pay risk.

All that might sound a bit complicated. Let’s consider an example. First is the pay leverage trendline for Pfizer CEO Albert Bourla, in Figure 1, below. 

Figure 1: Bourla at Pfizer 

Next is the same for Verizon CEO Hans Vestberg (Figure 2, below). 

Figure 2: Vestberg at Verizon

The peer group used to compute relative TSR for Pfizer is the NYSE ARCA Pharmaceutical Index. The peer group used to compute relative TSR for Verizon is the S&P 500 Telecommunications Services Index. 

In both examples, the solid line is the regression trendline. The dashed line is a line with a slope of 1.0 and an intercept of 0.0. It’s included in the graph to show how the company differs from a hypothetical “Perfect Correlation Pay Plan” where relative pay always equals relative performance.

What the Numbers Tell Us

When we perform a regression analysis of log relative pay on log relative performance, that gives us a window into the company’s success in achieving the three objectives of executive pay: (a) providing strong incentives to increase shareholder value, (b) retaining key talent; and (c) limiting shareholder cost. 

For example, the slope of the trendline for Bourla at Pfizer is 1.45. This means that a 1 percent increase in relative shareholder wealth increases the CEO’s relative cumulative pay by 1.45 percent. This provides a very strong incentive to increase shareholder value. 

In contrast, the slope of the trendline for Vestberg at Verizon is only 0.23. This means that a 1 percent increase in shareholder wealth increases the CEO’s pay by only 0.23 percent. Comparing the two graphs shows that Bourla’s incentive to increase shareholder wealth is roughly six times greater than the incentive for Vestberg.

The intercept is the pay premium at peer group average performance. It provides a negative measure of retention risk, and a positive measure of shareholder cost. The pay premium for Pfizer is 0.70. It’s stated in natural logarithms. We can convert it to a percentage premium by taking the anti-log. The percentage premium is 101 percent (= exp(0.70) – 1). 

This tells us that Pfizer pays 101 percent above market at average performance. This pay premium limits retention risk because Pfizer pays above average for average performance, so the CEO would have more reason to stick around. It also raises shareholder cost for the same reason. The pay premium shows that Pfizer strikes a balance between limiting retention risk and limiting shareholder cost heavily in favor of limiting retention risk. 

The natural log pay premium for Verizon is -0.30. The percentage pay “premium” is -26 percent. This pay premium shows that Verizon strikes a balance between limiting retention risk and limiting shareholder cost modestly in favor of limiting shareholder cost.