How Net Operating Losses Work

An honest refresher. Not a tax textbook — the items below are the ones that actually matter when reading a 10-K income-tax footnote with a practitioner's eye. We define the attribute first, then the limits on it, then the accounting that carries it.

What an NOL is — and why it is not the deferred tax asset

A net operating loss is the excess of allowable tax deductions over gross income in a taxable year, computed on the corporate tax return. The loss carries forward to offset future taxable income. For federal NOLs generated in tax years beginning after December 31, 2017 (TCJA), there is no expiration — they carry forward indefinitely. NOLs generated in years ending on or before December 31, 2017 retain the old 20-year expiration. Many large companies carry both vintages on the books.

That definition is worth reading twice, because the single most common error in this whole subject is treating the NOL and the deferred tax asset as the same thing. They are not. The NOL is a tax attribute. The DTA is an accounting estimate of what that attribute is worth. The NOL is measured in dollars of deduction and lives on the tax return, where no investor ever sees it. The DTA is measured in dollars of tax and lives on the balance sheet, where everyone does. One is converted into the other by multiplying it by a rate:

At the 21% federal statutory rate, a company disclosing a $1.0B federal NOL carryforward carries a gross NOL deferred tax asset of $210M — not $1.0B. It will not save a billion dollars of tax. It will shelter a billion dollars of income, which at 21% is $210M of tax. Read a headline NOL balance as a cash benefit and the error is the reciprocal of the tax rate: you will be wrong by roughly five times. (The worked example that runs through Sections 1.7 through 1.12 uses a 25% blended rate to approximate federal plus state; the identity is the same, only the multiplier changes.)

The two figures also move for different reasons, which is why they have to be tracked separately. The NOL balance changes only when the tax return changes — a loss year adds to it, a utilization year draws it down. The DTA changes for all of those reasons and for three more that have nothing to do with operations: a change in the enacted tax rate remeasures it overnight, a §382 ownership change reduces the recoverable portion at the source (Section 1.3), and a valuation allowance can write it to zero while the underlying NOL sits fully intact on the return (Section 1.10). A company can report a DTA of nil and still own billions of usable losses. The reverse — a DTA with no NOL behind it — is also possible, because NOLs are only one of several attributes that generate deferred tax assets.

Throughout the Institute's work, per-company tables report both: the gross NOL as the issuer discloses it, and the estimated deferred tax asset it implies. Where the two appear to disagree, the gap is the story — and it is almost always a valuation allowance, a §382 limitation, or income that never entered the U.S. tax base at all.

The TCJA 80% cap

Post-2017 NOLs can shelter no more than 80% of taxable income in any year (computed before the NOL deduction). This is a timing limit, not a haircut — the unused 20% carries forward. The CARES Act temporarily suspended the cap and re-permitted carrybacks for 2018-2020 NOLs (five-year carryback). That window is closed; the 80% cap is back in force. Pre-2018 NOLs are not subject to the 80% cap but do expire.

§382 — the ownership-change limitation

§382 is where many readers under-estimate the risk. If a loss corporation undergoes an ownership change — more than a 50-percentage-point increase in stock ownership by 5%-or-greater shareholders over a rolling three-year testing period — future use of the pre-change NOLs is capped at an annual amount equal to:

NUBIG / NUBIL adjustments

If the loss corporation has net unrealized built-in gain (NUBIG) above a threshold at the change date, the recognized portion of that gain in the five-year recognition period increases the annual §382 limit. Conversely, net unrealized built-in loss (NUBIL) recognized in the same period is itself subject to the §382 limit. NUBIG add-ons routinely double or triple the headline base limitation for asset-heavy targets — relevant to several of the examples below.

§382(l)(5) — the bankruptcy exception

If a loss corporation undergoes an ownership change in Title 11 (bankruptcy), §382(l)(5) waives the annual base limitation entirely — the NOLs survive intact — provided pre-change shareholders and qualified creditors own at least 50% of the reorganized stock. A subsequent ownership change within two years kills the NOLs entirely (so the reorganized debtor must lock the cap table down). Alternatively, the debtor can elect out of §382(l)(5) and take a special §382(l)(6) calculation that uses post-emergence equity value — usually a higher base than pre-petition would have produced.

A caution about the most-cited example. General Motors is routinely described as a §382(l)(5) case. It is not one. GM did not reorganize and emerge — substantially all operating assets were sold under §363 in July 2009 to a new entity, and that purchaser is the public registrant today. §382(l)(5) reaches a corporation that emerges from a title 11 case, so it had no application to the buyer. Preservation there rested on Treasury Notices 2008-76 and 2009-38, which disregarded Treasury's crisis-program stock acquisitions for ownership-change testing. The economics rhymed with an (l)(5) outcome, which is why the shorthand spread — but the authority differs, and so do the consequences: no qualified-creditor test to satisfy and no §382(l)(5)(D) two-year forfeiture cliff to manage. When a filing or a write-up attributes a preserved post-bankruptcy stack to (l)(5), check whether the surviving entity was the debtor or the buyer before relying on it.

State non-conformity

Federal NOL rules do not bind the states. A few practical examples:

Translation: a $5B federal NOL stack does not imply $5B of state-equivalent shelter. State NOLs are a separate accounting exercise, and the ratio between the two is company-specific: it turns on the apportionment factors, which states the losses arose in, and whether those states conform to the federal rules at all. There is no reliable rule of thumb here. Read the state line in the deferred-tax schedule and the state valuation allowance against it, because that is the only place the answer is.

Why the GAAP loss is not the NOL

A reader who opens a 10-K, finds the operating loss on the income statement, and concludes that is the loss the company can carry forward, has missed the most important fork in financial reporting. The income statement is computed under U.S. GAAP. The NOL is computed on the federal corporate income-tax return — Form 1120, line 28 (taxable income before the NOL deduction) — and the two numbers reconcile through a long list of permanent and temporary differences. Some of those differences are small. Several are routinely in the hundreds of millions or billions of dollars for a single year for a single company.

GAAP and the Internal Revenue Code share most of their machinery but disagree on the timing — and sometimes on the existence — of many income and expense items. Those disagreements sort into two families, and keeping them apart is the single most useful organizing idea in ASC 740. Temporary differences are timing items: book and tax eventually recognize the same total amount, just in different periods. Because they reverse, they create deferred tax assets and liabilities on the balance sheet. Permanent differences never reverse. An item is income or deduction for one system and simply does not exist for the other, in this period or any period. They create no deferred tax account at all — they move the effective tax rate, and it stays moved. Section 1.8 sets out the temporary differences a practitioner sees most often; Section 1.9 does the same for the permanent ones.

Temporary differences — the timing items that reverse

Each of these items appears in the deferred-tax-asset and deferred-tax-liability schedule inside the income-tax footnote of every 10-K we have ever opened. The schedule is one of the most informative half-pages in the filing. Every item above earns a line on that schedule precisely because it reverses. The items in the next section will not appear there at all.

Permanent differences — the items that never reverse

A permanent difference is not a question of when. It is a question of whether. The item is recognized by one system and never by the other, so there is nothing to unwind in a later year and nothing to park on the balance sheet. The practical consequence is that permanent differences are invisible in the deferred-tax schedule — the disclosure most readers go to first — and visible only in the reconciliation of the federal statutory rate to the reported effective rate. General Motors, profiled in Part Two, turns on one of the items in the table below.

One caution before leaving the taxonomy. The line between the two families is drawn by the tax law, and the tax law moves. The TCJA converted a slice of research spending from an immediate deduction into a five-year amortization — a permanent-feeling change that is in fact a large temporary difference. The One Big Beautiful Bill Act of 2025 moved it again. Classification is a filing-year question, and the footnote, not memory, is the authority.

ASC 740 — DTA vs. valuation allowance

Under ASC 740 a company books a deferred-tax asset for the future tax benefit of an NOL, then evaluates whether a valuation allowance is required. A valuation allowance is recorded when it is more likely than not (>50%) that some or all of the DTA will not be realized. Three points the footnote will telegraph:

How deferred taxes for an NOL get recorded

The mechanics are best shown with a small worked example. Assume a company has a $100 GAAP loss equal to a $100 tax loss in Year 1 (no permanent or temporary differences), and a 25% blended federal-and-state tax rate.

Year 1 entry — recording the NOL deferred tax asset

The DTA goes on the balance sheet at $25. If management cannot conclude it is more-likely-than-not that the DTA will be realized — for example because the company has cumulative recent losses — ASC 740 requires a valuation allowance that writes the DTA down to zero (or to the portion that is realizable). The journal entry for a full valuation allowance is:

Net effect on the balance sheet: DTA $25 minus Valuation Allowance $25 equals zero. The gross NOL is still disclosed in the footnote — it has not gone away — it is simply written down for GAAP purposes pending evidence that the company will use it.

Years 2–5 — the DTA roll-forward

Per the Institute's May 2026 review: a reduction implies a beginning balance. The cleanest way to teach the DTA mechanics is a roll-forward table that shows the asset accumulating over multiple loss years and then drawing down in the profitable year. Assume the company runs additional $100 GAAP losses in Years 2 and 3 (same facts), achieves break-even in Year 4, and turns profitable in Year 5 with $200 of pre-tax book income. We continue to use a 25% blended federal-plus-state rate. To isolate the NOL mechanics, assume tax depreciation matches book, no other temporary differences, and that the company has now concluded sustained profitability supports realization — so the prior valuation allowance is fully released.

Before the balance-sheet view, the income-statement and tax-return view. The roll-forward below tracks a stock — the deferred tax asset. This table tracks the flows that produce it: what the company earned for book, what it owed on the return, and the gap between the two. Every figure in the roll-forward is derived from a line in this table.

Provision is shown gross of any valuation allowance, to match the roll-forward below. Read the last two rows together: the company reports $50 of tax expense in Year 5 and writes a $10 check. The $40 difference is not an accounting artifact — it is cash that stayed in the business, and it is the same $40 that comes off the deferred tax asset.

Activity rationale — Years 1–3: $100 tax loss each year, so $100 onto the NOL and $25 onto the DTA; Year 4: break-even, no activity; Year 5: $200 of pre-tax book income, against which the 80% cap allows $160 of NOL, so $160 comes off the carryforward and $40 comes off the asset. The two bands tie at every column: the ending NOL times 25% is the ending DTA, and $140 × 25% = $35.

For teaching simplicity we apply the post-TCJA 80% cap to the 25% blended rate; in practice the 80% cap is a federal-only rule and state NOL availability varies by jurisdiction (see Section 1.6 on state non-conformity).

Reading the roll-forward. The Year-5 reduction of $40 is not coming out of nowhere — it is drawn down against a $75 beginning balance that was built up over three years of $25 additions. That is the practical answer to a reader who asks ‘how can the company reduce a deferred tax asset by $40 if it just created the asset for $25 this period?’ The asset is a stock, not a flow; the period activity is two flows (additions and reductions) that combine to produce the ending stock.

When the NOL moves the provision, and when it only moves cash

There is a rule buried in the two tables above, and it is worth stating plainly because it is the single most common source of confusion in this whole area.

Once the deferred tax asset is carried on the balance sheet at full value, using the NOL does not change the total tax provision. It changes the composition of the provision and it changes the cash. The current tax expense falls, because the return shows less taxable income. The deferred tax expense rises by exactly the same amount, because the asset is coming off the balance sheet. The two moves cancel. Total provision stays at the statutory rate times book income, the effective tax rate looks ordinary, and net income is untouched. The benefit is real, but it is visible only in cash taxes paid — which is why the cash flow statement, not the income statement, is where an NOL shield is found.

That rule has one large exception, and it runs the other way. While a full valuation allowance is in place, the asset is already written down to zero, so there is nothing to draw down. Using the NOL still cuts the cash tax, but the deferred expense from the DTA drawdown is offset by an equal benefit from releasing the allowance that stood against it. Nothing cancels the current-expense reduction. The total provision falls, and the effective tax rate collapses.

Run Year 5 under both regimes and one number moves and one number does not. The total income tax provision moves — $50 under one regime, $10 under the other, a factor of five on identical operations. The cash tax paid does not move at all. It is $10 in both columns, because the company sent the Treasury the same check either way. The right-hand column of the table below says, line by line, which is which.

Year 5 journal entry — using the NOL against taxable income

In Year 5, GAAP pre-tax book income of $200 is fully taxable. The post-2018 NOL can shelter 80% × $200 = $160. We carry the 25% blended rate through the journal entry (federal-only equivalent on the residual $40 at 21% would be $8.40): GAAP tax provision = $50; cash tax = $10; the remaining $40 of provision is the NOL coming off the balance sheet.

Tie-out: the P&L still carries the full $50 tax expense at the blended rate (so EPS looks ‘normal’). But only $10 of that is cash — the other $40 is the NOL coming off the balance sheet. The deferred tax asset has just paid for itself.

The check to the Treasury — where cash actually moves

The provision entry above never touches cash. It debits an expense and credits two balance-sheet accounts. Cash moves in a second, separate entry, when the company settles the liability it just recorded:

Ten dollars leaves the building. Not fifty. To see what the deferred tax asset was worth, run the same year twice — once with the NOL and once without it, changing nothing else.

How NOL utilization runs through the cash flow statement

The drawdown is invisible on the income statement, but it is the headline event in the operating section of the cash flow statement.

Section 1.11 recorded two separate entries. The provision entry credited Income Taxes Payable $10 — the liability. The settlement entry debited that payable and credited Cash $10 — the check. Whether that check clears before the balance sheet date or after it changes the operating section, so the table below runs the same year twice. Note what does not change between the two columns.

Tie-out — paid column: $200 of pre-tax income less the $10 check to the Treasury = $190. Unpaid column: $200 of pre-tax income and no check yet = $200. In both columns the company kept $40 it would otherwise have paid; the NOL did that, and the payable line did not.

Three practical notes. First: the ‘deferred income taxes’ line in the operating section of the cash flow statement is the single most useful field for spotting NOL drawdown in real time. A large decrease in the DTA (a positive number on this line under the indirect method) during a year of GAAP profitability almost always indicates the company is sheltering cash taxes with prior-year NOLs. Second: the disclosure of ‘cash income taxes paid’ in the supplemental schedule at the bottom of the cash flow statement (or in the income-tax footnote) is the cleanest read on what the company is actually sending to the Treasury — and it is what tells you which of the two columns above you are looking at. Third: no issuer prints those two columns. In a real filing the payable movement and the deferred movement sit side by side in the same list of adjustments, with nothing on the page marking one as a durable attribute and the other as a timing wiggle. Separating them is the reader's job.

Credits are not NOLs — and they have their own limits

The 80% cap in Section 1.2 is a net-operating-loss rule. It lives in IRC §172(a)(2) and it constrains the NOL deduction. It does not reach tax credits, and a reader who assumes credits are capped at 80% of anything is applying the wrong statute.

That is not the same as saying credits are unlimited. They are limited — by different provisions, on a different base, with different carryover periods. The table sets the three regimes side by side.

The ordering matters, and it is the part most models get backwards. The two attributes do not compete for the same space on the return; they operate at different stages of it. Losses come first: the NOL deduction reduces taxable income, subject to the 80% cap. The statutory rate is then applied to what survives. Credits come last, reducing the resulting tax dollar for dollar, subject to their own limits. A company can therefore be constrained by the 80% cap and by §38(c) in the same year, on different bases, and the two constraints do not net against each other.

One modern wrinkle worth flagging. Several of the Inflation Reduction Act energy and manufacturing credits are transferable under IRC §6418 or eligible for elective payment under IRC §6417. Where those provisions apply, a credit a company cannot currently use is not stranded — it can be sold to an unrelated buyer for cash, typically at a discount to face. That converts a limitation problem into a pricing problem, and it means the §38(c) cap is a weaker constraint on the transferable subset of the credit stack than it is on ordinary research credits. It also means the credit carryforward line in a footnote may understate what the company has already monetized.

The corporate alternative minimum tax — a floor under the whole framework

Everything in this Reference up to this point describes the regular corporate tax. Since 2023 there has been a second calculation running alongside it for the largest companies, and it is built on book income rather than taxable income — which makes it the one provision capable of reaching straight past an NOL and a credit stack alike.

What it is. The Inflation Reduction Act of 2022 added a corporate alternative minimum tax at IRC §55(b)(2): 15% of adjusted financial statement income, reduced by the CAMT foreign tax credit, imposed as an add-on to the extent it exceeds regular tax plus BEAT. It applies to tax years beginning after December 31, 2022. An applicable corporation is one with average annual AFSI above $1B over the three prior tax years; a foreign-parented group faces a $1B group test plus a separate $100M U.S. test. Status is sticky — once a corporation is an applicable corporation it generally remains one, even if income later falls below the threshold.

Why it was enacted. The political motivation is not in dispute and is worth stating plainly, because it explains the design. Legislators objected to large, plainly profitable companies reporting substantial income to shareholders while reporting little or no federal tax expense. The remedy was to tax the number those companies show investors rather than the number they show the Treasury. Adjusted financial statement income starts from net income on the applicable financial statement — the GAAP figure — and adjusts from there.

Where the 80% cap reappears — and it is a different 80%. CAMT has its own loss carryforward, and it is not the §172 NOL. IRC §56A(d) allows AFSI to be reduced by a financial statement net operating loss carryover — a book loss, not a tax loss — limited to 80% of AFSI computed before the deduction. Only book losses from tax years ending after December 31, 2019 create FSNOL, so the attribute has a hard start date and no pre-2020 history. It carries forward indefinitely. A company can therefore hold billions of §172 NOLs and have no FSNOL at all, or the reverse. They are separate accounts, tracked separately, and a footnote that discloses one says nothing about the other.

Credits against CAMT, and CAMT credits against regular tax. General business credits remain available against the CAMT liability, and the CAMT foreign tax credit is built into the rate calculation, so a credit-heavy attribute stack is markedly more CAMT-durable than a loss-heavy one. Running the other direction, CAMT paid in a year generates a minimum tax credit carried forward indefinitely and usable against regular tax in later years. For a company that expects to return to a regular-tax position, CAMT is therefore closer to a prepayment than to a permanent increase — a timing cost and a cash-flow cost, not necessarily a rate cost. That distinction is invisible in the effective tax rate and visible only in the cash-tax line and the deferred-tax rollforward.

The interaction most models miss. The general business credit limitation at §38(c) caps the credit at net income tax less the greater of the tentative minimum tax or 25% of net regular tax liability above $25,000. From 2018 through 2022 that first prong was dormant, because a corporation was deemed to have a tentative minimum tax of zero. The Inflation Reduction Act narrowed that deeming rule so that it now reaches only corporations that are not applicable corporations. For an applicable corporation, CAMT is the tentative minimum tax, and the first prong of the §38(c) limitation is live again for the first time since the TCJA. The practical effect is that CAMT does not merely add a floor — for the largest credit-rich taxpayers it can also compress how much of the credit stack is usable in the year. This is a genuinely technical corner and it is worth naming rather than assuming away.

Where the law actually stands as of this Reference. CAMT has been administered largely through interim guidance rather than final regulations. Treasury and the IRS issued Notices 2023-7, 2023-20, 2023-64 and 2024-10, published proposed regulations in September 2024, and have continued issuing interim guidance since, including Notice 2026-7, released February 18, 2026. Notices are numbered in the order they are assigned, not the order they are released, so a higher number can carry an earlier date — Notice 2026-15, cited in Part Two, is dated February 12, 2026. Final regulations have not been issued as of August 2026. A practitioner should treat the specifics below the statutory level as provisional and check the current guidance before relying on any particular adjustment. This is the least settled area covered anywhere in this Reference.

Where this framework gets applied

Part Two takes everything on this page and reads it against four live income-tax footnotes: Lyft, Cleveland-Cliffs, Snap, and General Motors — four companies, four industries, four different reasons the tax footnote matters. Every figure is traced to a filed 10-K, and every point-in-time figure carries its own measurement date rather than borrowing the document’s.

It is the perishable half of the set, and it is written on the assumption that you have read this one.

Position disclosure. The author owns LYFT and CLF, two of the four companies discussed in Part Two. No compensation from any issuer named, no material non-public information, and a 30-day trading blackout around publication. This reference analyzes no company and names no position, but the disclosure travels with the set.

Related free references and tools