Tax Loss Harvesting in a Taxable Account: The Year-End Checklist, 2026 Wash-Sale Rules, and the December Math

A $5,000 loss in a taxable brokerage account is worth $750, $360, or $0.00 this year, and which one it is gets decided by three things: the row your gain sits in, whether you keep your hands off the same holding for 31 days, and whether you remember that the rule counts the 30 days before the sale as carefully as the 30 after. The loss itself is the easy part. The figure is what the mechanics recover — and the mechanics are where the figure leaks out.

I have spent 16 years in portfolio tax work — an adviser’s tax room first, then a tax-efficient-investing desk — and since 2022 I have written The Tax Sort. This page is the harvest lane of that series: the rule, the dated window, the counted December math, and the checklist. Every figure below is dated and situation-specific; it is arithmetic on the rule, not a ruling on your return.

What the harvest is, and where it lives

Tax-loss harvesting is selling a holding in a taxable account at a loss on purpose, so the loss offsets capital gains realized elsewhere in the same year and, if losses exceed gains, up to $3,000 of ordinary income — with the excess carried forward. That is the whole mechanism. The word “harvest” oversells it: nothing is created. A tax bill you would otherwise pay this year is what you are actually collecting, and the tax bill is the return you did not keep.

Which is why the harvest is a taxable-account play only, and this belongs to the sort before it belongs to any calendar. Sell a loser inside a traditional IRA, a Roth IRA, or a 401(k) and nothing happens: the account has no taxable events, so there is no loss to claim and no deduction to recover. The holding in the right account does not need to be harvested. The holding in the taxable account does, and this page is its dated rules.

The wash-sale rule, from the documents

The rule that decides whether the loss survives the sale is not folklore; it is a statute with a regulation and a publication sitting on top of it. IRC §1091(a), as published in the US Code, disallows the deduction where, within a period beginning 30 days before the date of the sale and ending 30 days after it, the taxpayer “has acquired … or has entered into a contract or option so to acquire, substantially identical stock or securities.” Publication 550 (2025 edition, the wash-sale pages I read from the PDF on October 4, 2026) states the same period and prints the four ways the window gets tripped. You have a wash sale when you sell stock or securities at a loss and within 30 days before or after the sale you:

1. buy substantially identical stock or securities,
2. acquire substantially identical stock or securities in a fully taxable trade,
3. acquire a contract or option to buy substantially identical stock or securities, or
4. acquire substantially identical stock for your IRA or Roth IRA.

Two more sentences from the same pages, because they catch real portfolios: “If you sell stock and your spouse or a corporation you control buys substantially identical stock, you also have a wash sale.” And the option prong is symmetric with the statute — the wash-sale rules apply to contracts and options to acquire or sell stock or securities; they do not apply to losses on commodity futures contracts and foreign currencies.

What a disallowed loss does is move, not vanish — with one exception. Add the disallowed loss to the cost of the replacement shares, and the publication is explicit that the adjustment “postpones the loss deduction until the disposition of the new stock or securities,” with the new holding period tacking on the old one. Except prong (4): a loss killed by a repurchase inside an IRA gets no basis adjustment anywhere. That figure is gone for good. The taxable-account loss that re-buys the same holding in the Roth is not a harvest with a delay; it is a donation.

The 2026 window, counted on the calendar

The rule is stable; what makes any given December different is the calendar. Anchor the sale to the last day of the tax year, Thursday, December 31, 2026, and the §1091 period runs exactly 30 days each side of it:

• window opens: December 1, 2026 (30 days before the sale — a buy on this date is inside it),
• window closes: January 30, 2027 (30 days after the sale — a buy on this date is still inside it),
• safe re-buy for the same holding: January 31, 2027 — the 31st day.

The part sellers get wrong is that the window is half built before the sale. Every purchase of the same holding from December 1 onward is already waiting to disallow a sale you have not made yet — including purchases you did not consciously make, because the statute asks only whether you acquired, not whether you chose: an automatic investment plan pointed at the same fund, or the same-account dividend reinvestment, is a purchase on the day it executes. Here is the December harvest checked against specific dates, with the arithmetic stated once so you can redo it for your own sale date:

A December 31 sale is the maximally disciplined date, not the laziest one: it puts the whole of January inside your wash window, so the harvest does not survive the winter unless the re-buy discipline does. Flip the lever and the window moves with it — sell the position on, say, November 15, 2026 and the window is October 16 through December 15, 2026, already closed long before the year ends. The harvest does not have to wait for December; it waits for the loss and your bracket forecast, and December is simply the last day the tax year can still be changed.

Substantially identical: what actually trips it

The phrase doing the work is “substantially identical,” and the publication refuses to define it away: “you must consider all the facts and circumstances in your particular case.” What it does print is the ordinary rule and the edges. Ordinarily, the stock of one corporation is not substantially identical to the stock of another — the reorganization predecessor-successor pair is its example of when even that holds. Bonds or preferred of a company are ordinarily not identical to that company’s common — unless the preferred is convertible into the common, carries the same voting rights, is subject to the same dividend restrictions, trades at prices that do not vary significantly from the conversion ratio, and is unrestricted as to convertibility. Five conditions; meet them and it can be identical. Sell a loser in the common and buy warrants for the same corporation inside the window: wash. The same-day block rule is nastier than it looks: a wash-disallowed loss on one block cannot be used to reduce the gain on an identical block you sold the same day — the disallowed loss is not allowed to buy its way back in through the side door.

For funds, read the two sentences in order instead of the one you want: different issuers are the ordinary case for “not identical,” and “all facts and circumstances” is what governs when they are not. Two index products tracking the same index from different providers sit precisely in that gray — which is a mechanics observation, not a hedge, and it is why I will not tell you which fund to swap into; picking the holding is not my lane and never has been. What I can state is the shape of the safe end of the line: the further the substitute’s index and issuer are from the one you sold, the more defensible “not substantially identical” is — and the cost of running two similar funds is counted, fund by fund, in the index fund vs. ETF comparison, which is the page for the wrapper decision.

Then the account-crossing traps, which are where the figure actually dies: the rule has no account boundary. Publication 550 prints the broker-side convenience — Form 1099-B box 1g shows the disallowed wash amount only where the sold shares were covered securities and the replacement had the same CUSIP in the same account — and then removes the comfort: “you cannot deduct a loss from a wash sale even if it is not reported on Form 1099-B.” A buy in your second brokerage, your spouse’s account, or your IRA trips the rule even though no form will net it for you at year end. What you file anyway: the wash row goes on Form 8949 in Part I or II, “W” in column (f), and the disallowed amount entered as a positive number in column (g).

The $3,000 that is not a limit on the loss — it is a limit on the year

The number most sellers carry around is “you can deduct $3,000.” That is the row that misleads people, because it describes the year’s allowance, not the loss’s size. Publication 550 states the limit precisely: your allowable capital loss deduction is the lesser of $3,000 ($1,500 if married filing separately) or your total net loss on Schedule D line 16, and “you can use your total net loss to reduce your income dollar for dollar, up to the $3,000 limit.” The same $3,000 / $1,500 pair is the figure already dated on this site in the 2026 capital gains rate table; that page is the rates, this page is the mechanism, and the two do not overlap.

Everything above $3,000 is not lost — it is queued. The unused part carries to the next year and is treated as incurred there, and it carries again until it is used up. Two properties of the queue decide whether your harvest is worth the transaction costs, and both are worth more than the $3,000 headline:

• The carryover keeps its character. “When you carry over a loss, it remains long-term or short-term,” and a long-term loss carried over “will reduce that year’s long-term capital gains before it reduces that year’s short-term capital gains.” Long-term losses get first call on long-term gains — the gains that carry the highest rate you could avoid — for as many years as the queue runs.
• Short-term losses are spent first, inside the year. When the carryover is figured, short-term losses are used first “even if you incurred them after a long-term capital loss,” and only then long-term. So a short-term loss harvested in December is the most spendable inventory you can hold: it is applied to the $3,000 before the long-term loss is touched.

The publication’s own example, which is the arithmetic to keep in your head: a $7,000 loss with $26,000 of taxable income, married filing jointly — deduct $3,000, carry $4,000 into 2026. And the floor case that no one expects: had the loss been $2,000 instead, the deduction is $2,000 and the carryover is zero. A loss under the limit is not banked for later; it is simply used.

One more property of the queue, and it is the reason the deadline is December rather than April: the carryover is computed from the year’s allowable deduction “whether or not you claimed it and whether or not you filed a return for the current year.” Not filing does not preserve anything. The deduction you did not take is a deduction the queue already assumes you took.

The December math: what one $5,000 harvest is worth

Same loss, three sellers, three different answers. All arithmetic below is shown so it can be redone; the bracket ceilings are the dated 2026 rows already published on this site, from the 2026 capital gains rates table (Rev. Proc. 2025-32, in effect October 9, 2025), not re-derived here.

Seller A — a gain in the 15% row. A single filer realizes $5,000 of long-term gain from a rebalance and harvests a $5,000 long-term loss in the taxable account against it. The loss cancels the gain; at the 15% row the tax saved is

$5,000 × 15% = $750.00

This is the clean case: the harvest converts to cash this year, at the rate of the gain it erased. The same page’s worked seller confirms it from the other side. Her single filer at $45,000 of taxable income with a $10,000 long-term gain pays $832.50 on the gain, because $4,450 of it fills the 0% row and only $5,550 reaches 15%. Drop this $5,000 harvest onto that same seller and it cancels the gain from the top down — the $5,000 comes off the $5,550 that was paying 15%, and what is left of the gain is $4,450 at 0% plus $550 at 15%:

$832.50 − $82.50 = $750.00 saved — the harvest recovered the 15% layer and touched no 0% dollars

That is the reconciliation between this page and hers, stated once: a harvest only saves what the canceled gain would actually have cost, which is why it prices at the top of the gain’s layers and not its average.

Seller B — no gains at all, ordinary income in the 22% row. Same $5,000 loss, nothing to offset. The $3,000 applies against ordinary income, and the ordinary rate is what prices it:

$3,000 × 22% = $660.00 this year, with $2,000 carried forward

At a 12% marginal rate the same $3,000 is worth $360.00. The harvest is priced by your marginal row, not by the size of the loss, and the $2,000 in the queue is worth whatever row you are standing in when it arrives — possibly a different one.

Seller C — the 0% bracket. The harvest is worth $0.00. A single filer whose 2026 taxable income lands under the $49,450 ceiling in that dated table pays 0% on long-term gain. The rate table on this site says it in the row itself: “A long-term gain that fills only space below your status’s first ceiling is taxed at zero. Not deferred — zero.” A loss that cancels a gain which would have been taxed at zero saves exactly that:

$5,000 × 0% = $0.00

And it is worse than worthless in one respect: if the loss is long-term and there is no other gain, only $3,000 can touch ordinary income, so the harvest burns $3,000 of a queue you may not need, at transaction cost, for a tax of $0.00. The 0% row is where harvesting stops. Before any December sale, find the row your own gain lands in — the ceilings by filing status are in the linked table, and the standard deduction that gets you from pay to taxable income is on the same page.

So the same $5,000 loss is $750.00, $660.00, or $0.00. Three sellers, one security, one year. The tax bill is the return you did not keep, and the bracket decides whether you keep it.

The year-end checklist, dated

The checklist is the article. The window arithmetic above is what each line is checking. Work it in order; every date is for the 2026 tax year.

The four dates in Q4

• Now (early October 2026). The gains-to-date number and the bracket forecast are already knowable; the December row is guesswork unless you start from realized numbers. This is the week the checklist has a purpose.
• By December 1, 2026. The wash window for a year-end sale opens. From this date every purchase of the same holding is a future disallowance. Auto-plans and DRIP get switched off before the sale, not after it.
• December 31, 2026, Thursday. The last day on which the 2026 tax year can still be changed. A loss sale on this date puts the entire January inside its window.
• January 30–31, 2027. The window closes on the 30th; the 31st is the earliest clean re-buy of the holding you sold. The tax year is over by then and the harvest is already locked or already dead — this is the date that decides whether the December trade survived it.

Where the harvest sits in the sort

The harvest is a repair, not a strategy — and it only applies to the account that has taxable events. Before any of it: every dollar inside a 401(k), an IRA or a Roth sits outside this mechanism entirely, and the room those vehicles have is dated in the 2026 contribution limits table. After the loss is banked and the substitute is chosen: the 2026 capital gains rates table is the row your harvest is priced against, and it is the page that says plainly where the value of a harvest goes to zero.

The wash-sale rule is not the obstacle people think it is; it is a 61-day calendar with a definition attached, and it is the same 16 years later that the same three sentences decide the year: what did you buy, in which account, on what date. Sort it by account, not by return — the harvest is worth what the row says it is worth, and the figure is dated: $750.00, $360.00, or $0.00 on the same piece of paper.