Why the last big loophole is made of losses, and how it closes
This summer Bloomberg profiled the world's largest hedge fund complex and the product behind its growth: tax-loss harvesting at industrial scale. The mechanics are simple. Run a portfolio long and short at two or three times its capital. Sell the losers constantly; let the winners ride. The portfolio makes money — and the tax return shows a loss, which offsets income from anywhere else. The IRS bill shrinks while the wealth grows.
The Accord closed the gain side of the tax code completely. Every disposition of an investment asset now realizes gain — the swap-and-defer doors are deleted. Basis step-up at death is gone. The deferral wrappers end for new money. And that thoroughness creates the one opening left: when every gain must eventually be taken, a manufactured loss becomes the most valuable product in finance. Close the gain side without the loss side, and the harvesting machines simply absorb what the closures collect.
There is a further consequence, and it is worse than a tax break.
The harvested losses shelter income earned somewhere else — a private equity exit, a bonus, carried interest. That income leaves the account and is spent. The gain that offsets it stays inside the account, unrealized, compounding. Nothing links the size of the deduction to the size of the asset that will eventually have to pay for it.
Run the arrangement's own advertised case. A hundred million dollars invested, roughly tripling over a decade, generating well over five hundred million dollars of losses along the way. At the end the account is worth about three hundred million and carries an embedded gain of roughly one and three tenths billion. Taxed at death with no basis step-up, that is a bill of about five hundred and thirty million against an estate of three hundred million.
The liability is roughly one and three quarter times the asset that produced it. An estate cannot pay a hundred and seventy-five percent of itself. The tax is not deferred. It is uncollectible, and the public absorbs the difference.
So the loss side closes too, with three rules.
First: losses are limited to losses. Over the life of a position, deductible losses may not exceed the amount actually lost. A portfolio that went up produces no net loss for the tax return, no matter how artfully its losers were sold.
The rule was first written year by year, which left one window open: a losing year inside a profitable decade still produced a deduction, even where the position was far ahead over its life. Measured across the whole life of the position instead, that window closes. The excess isn't confiscated; it waits, and releases when the account is closed out or genuinely goes into loss, because at that point the loss is real.
This closes the collectibility problem by construction. If the deduction can never exceed the economic loss, the embedded gain can never exceed the account's value, and the bill is always payable out of the thing that owes it.
This is not mark-to-market. Rising value is never taxed as income — nobody here pays tax on a gain they have not taken. The mark is how the cap is measured, and both numbers already sit on the standard year-end brokerage form.
Second: the wash-sale rule joins this century. The current rule blocks re-buying a "substantially identical" stock within thirty days — written for a world where you'd re-buy the same ticker. A modern optimizer sells the loser and buys a statistical twin: a different name with the same behavior. The rule tightens to substantially similar, extends to short positions, and counts all accounts under common ownership as one.
Third: both sides of the ledger are reported. A vehicle claiming a loss reports the position it came from — current value, long side and short side, stated separately.
This is a reporting rule rather than a tax rule. It imposes no liability and changes no rate, and nobody whose position is what they say it is pays anything more. It is also what makes the first rule work, because a limit on losses cannot be applied by an agency that can only see the losses.
Ordinary investors are outside all three rules twice over. A per-taxpayer allowance leaves everyday harvesting in a down market untouched, and a portfolio that genuinely lost money deducts its real loss exactly as before. The nurse who sells a losing fund in a bad year sees no change. The rule finds the structure sold on "tax alpha," because that structure has a signature: losses flowing out of an account that is, in fact, winning.
One more piece of design worth naming. Every closure in the Accord now states, on its page, the loophole it closes and the interlocking piers that must travel with it — the other rules that keep it sealed. A legislator who wants to lift one closure into a separate bill can see exactly what else the seal requires. The loss-side rule lists its own: universal realization (the gain side it protects), the ownership registry (so splitting into trusts and shells changes nothing), and the modernized wash sale (so replacement trading can't leak around the cap).
The plan expected this. The Accord holds that any mechanism which is tax-favored will be abused, and that is a design assumption rather than a prediction about particular people. It is why the framework taxes value where it appears instead of legislating against each arrangement as it emerges. Finding this one and closing it is the assumption operating as intended.
With this closure, every leg of the oldest strategy in the book — buy, borrow, die, and harvest — is priced or closed. Gains when they are taken. Losses only when they are real.
The full mechanism, with who pays, who is protected, and the interlocking piers: [Loss-side integrity](/engines/revenue/loopholes/loss-side-integrity). The complete closure set: [Closing the Escape Routes](/closing-loopholes).