The Conservation Law of Tax Alpha
A simple minded trader and theoretical physicist deconstructs Wall Street's current hottest product.
On April 20th, the inimitable Matt Levine brought this one to my attention, as he so often does. What follows is my own take on it, in the physicist register, which is a different angle from his lawyer-and-markets one. You should read him too.
“It’s the hottest product in Greenwich,” one Wall Street person told the Financial Times in mid-April about tax-aware long-short hedge funds. The FT reporter added that fathers at a local sports game were fixated on AQR’s Delphi Plus fund. I live in Greenwich. I have not personally been cornered at a sideline — yet — but the day is young, and the ambient enthusiasm is real. So is the product category: $90 billion of net new money since the start of last year, almost a third of AQR’s assets now in tax-aware funds, and new entrants from Two Sigma, Quantinno, JPMorgan, WorldQuant, and Millennium piling in behind.
I kept reading the tax-aware long-short pitch decks and something felt off, but I couldn’t articulate it. Then I remembered I’m a physicist. Just a simple-minded trader and theoretical physicist. I ought to be able to articulate it. And so, here you are.
Three things about taxes, before we start
Before we can even begin to talk about what these funds do, a few things are worth stating plainly, because most of what comes later follows from them.
First: tax authorities are very good at collecting taxes. They have the entire mechanism of the monopoly on force behind them. No one pays taxes because they want to. People pay taxes because they are forced to — at the point of a gun, or loss of freedom. This is a useful thing to keep in your head whenever someone shows you a product that claims to reduce your taxes. The IRS has been doing this for a long time. They’re reasonably competent, and the ones who aren’t competent are backed up by people who are.
Second: if something seems amazingly good from a tax point of view, there is almost certainly a catch. Sometimes the catch is visible in the small print. Sometimes it’s visible only to someone willing to do the terminal accounting. Sometimes it’s not visible at all until the government gets around to looking at it.
Third: the catch usually arrives late. This is the part everyone forgets. You may run a structure for a decade and collect annual statements showing enormous savings before the bill shows up. The IRS is not a market-maker providing instant feedback. They are a collection agency with infinite patience and a monopoly on force. If something was wrong, you will find out — but possibly not for ten years.
One more contemporary note before we start. On April 17th, Bloomberg reported that a short seller called Orso Partners has taken a position against Affiliated Managers Group, the $800 billion investment firm that owns a stake in AQR. The specific thesis: concerns about the tax-aware long-short ecosystem. Real capital, real position, real bet against the boom. When someone is willing to put money against the hottest category on Wall Street, it is worth asking why. What follows is my version of the answer.
RenTech, or: how the smartest people in the room lost
If you want a case study for the third point, consider Renaissance Technologies.
RenTech — arguably the most quantitatively sophisticated hedge fund in the history of the industry — used a structure called a “basket option” from 2005 to 2015. The mechanics: Medallion (the insider fund, run on employees’ own money) bought options from Deutsche Bank and Barclays whose value tracked baskets of stocks that Medallion itself directed the banks to trade. Stocks held for days or seconds inside the basket; gains claimed as long-term because the option was held for more than a year. Presto: short-term gains converted to long-term.
In 2014, Senator Carl Levin’s subcommittee called the structure out publicly. RenTech fought it through IRS Appeals for seven more years. In September 2021 — sixteen years after they started — the firm’s current and former executives settled for approximately $7 billion. Largest tax settlement in U.S. history. Twice the prior record.
Notice the pattern. Ten years of running the structure. Seven years fighting it. A decade-plus of annual statements that looked great. And then the bill.
Notice also who lost. It wasn’t some marginal boutique. It was the people with presumably the best tax counsel in the world, running the structure on their own capital, in the fund with the highest Sharpe ratio in history. If those guys couldn’t make aggressive tax structuring stick — couldn’t, in the end — that should tell you something about the base rate.
Put aside the risk of that here
But let’s be generous. Let’s assume the tax-aware long-short category has constructed enough of a fig leaf that the strategy really does have economic substance independent of its tax benefits. Assume it survives any IRS challenge. Assume the factor bets are real, the wash-sale engine is immaculate, the short book actually holds up, and the whole apparatus does exactly what the pitch deck says it does.
Then what?
Then we can ask the real question: Are you actually saving any taxes?
And beneath that, the even better question: What is really going on?
What’s being sold
Briefly, for those who haven’t been pitched one of these: a tax-aware long-short fund is a factor strategy run at gross leverage — 150/50, 200/100, 250/150 — wrapped around a tax-loss-harvesting engine. The idea is that by going long and short a few hundred to a few thousand names, you generate a much richer stream of tax lots than a long-only direct-indexing book would. Dispersion creates losers. Losers create harvestable losses. Harvested losses offset gains you have elsewhere. The annualized savings — “tax alpha” — is typically quoted somewhere in the range of 1.5% to 2%.
AQR has been the most visible marketer of the category. According to the FT, they have added $47 billion of tax-aware assets since March of last year — almost a third of the firm’s total book. Quantinno, an independent firm founded by ex-AQR traders in 2018, has pulled in another $39 billion since January 2025. Two Sigma’s Beacon Fund, $950 million. JPMorgan, Merrill, WorldQuant, Millennium, Morgan Stanley’s Parametric — everybody is in. The underlying academic work by Sosner, Liberman, Sialm, and others is careful and honest. The pitch decks built on top of it, less so. And the scale — $90 billion of net new money in eighteen months — is precisely the thing that should make anyone who has watched cycles before start paying attention.
Books must balance
Here is the physicist’s reflex. If you trace the entire arc of a strategy — dollar in at the beginning, dollar out at the end, including every tax interaction in between — the accounting has to close. The IRS does not forget. They have not agreed to give up their claim on your gains. They have agreed to let you sequence those claims across time, within certain rules.
Call the investor’s total lifetime tax bill on the round trip T. Then the question to ask of any “tax alpha” strategy is whether it reduces T, or merely redistributes T across time.
Those are very different things. And the difference is where this post is going.
What the long-short structure actually does
To be fair, the long-short structure genuinely does something that a long-only direct-index book cannot do, and it is worth crediting honestly.
In a rising market, a long-only direct index runs out of harvestable losses. Everything is above cost. The harvesting yield decays to zero. This is arithmetic.
A long-short book has two structural refresh channels that a long-only book lacks. First, the short leg generates losses on winners — when a name you’re short appreciates, the short position has an unrealized loss that can be harvested by covering and re-shorting. In a rising market, the short book is the harvesting engine. Second, the factor tilts rebalance continuously — value and momentum and quality rotate against each other, positions open and close, and each rebalance establishes fresh cost bases. Every new lot at today’s price is, functionally, a new contribution from the harvesting engine’s point of view, even with no cash added.
Plus: gross leverage of 2-4x scales the lot surface area roughly linearly. More lots, more dispersion, more opportunities.
So the honest summary is: tax alpha decays, but in a long-short structure it decays more slowly than in long-only because the short book and factor rebalancing keep reseeding the loss inventory.
Fine. That’s real. But none of it escapes the conservation law.
The conservation law
Harvesting does not make taxes disappear. Harvesting realizes a loss today, which lowers your tax bill today, but it also lowers your cost basis (or for shorts, establishes a low basis) on whatever you reopened. When you eventually close that position at a gain, the gain is larger by exactly the amount of the loss you harvested. The IRS gets the same money. They just get it later.
If the strategy works and the factor book grinds up over decades, the aggregate book will drift to a progressively lower and lower basis relative to its market value. The deferred tax liability on the book grows. You can see the number on the statement — look at the “unrealized gains” line and multiply by your tax rate. That is what you owe, eventually, to the IRS. It is a liability. It is real. It is growing.
This is unavoidable. It is conservation. Harvesting shifts T forward in time. It does not reduce T.
What actually reduces your lifetime tax bill
Now — to be honest — there are ways the strategy can reduce T rather than merely shift it. They are narrower than the marketing suggests, and they are the entire real case for the product. Four of them:
1. Character conversion. If you have short-term gains elsewhere — trading profits, comp, carried interest in a short-term fund — taxed at ordinary rates (roughly 40% federal plus state), and the short leg throws off short-term losses, you are converting ordinary-rate tax into deferred long-term-capital-gains-rate tax. That is real new money: approximately the rate spread (say, 20 percentage points) times the volume of offsets. This is the strongest part of the case. It also means the product is genuinely well-suited to a specific kind of client — someone with large ordinary-rate income streams elsewhere — and materially weaker for everyone else.
2. Rate or jurisdiction arbitrage. Harvest at 37% federal plus 13% California today. Realize at 20% federal in Florida or Puerto Rico in retirement. Real, situational, not structural.
3. Step-up at death, or charitable transfer. The deferred liability actually evaporates. This is the one that does all the work in the long-run IRRs the marketing implies. The product is implicitly an estate-planning vehicle whenever this is the assumed exit.
4. The time value of the deferral itself. Delaying a dollar of tax is worth something — roughly the discount rate times the duration. At a 5% real discount rate over a 20-year horizon, deferring $1 of tax is worth about 60 cents in present value terms. Annualize that against the gain base and you get something like 1% a year — not zero, but also not the 1.5-2% in the pitch book. The deck annualizes the year-one harvesting bonanza; the reality is a front-loaded decay.
That’s it. Everything beyond those four items violates the conservation law and should be treated as rhetorical flourish.
Run the numbers
Let’s make this concrete with a round example.
Start with a $10 million portfolio. The marketing quotes a 2% tax alpha — so $200,000 in year one. The management fee on a tax-aware long-short wrapper is typically around 100 bps, call it $100,000. The financing spread on the short book (stock borrow, margin, the cost of running a 150/50 or wider structure) is around 50 bps of NAV on the modest end, call it another $50,000. Subtract: $50,000 of net benefit in year one. A quarter of the headline.
That’s the best year.
The harvesting yield decays as the embedded gain grows — the published academic curves show it roughly halving by year ten. But the fees don’t decay. By year ten you’re still paying $150,000 in fees and financing, but you’re generating maybe $100,000 of gross tax alpha. You are now losing $50,000 a year, net. By year fifteen, on any realistic factor book, the product is inside the fee. You are paying the manager to reduce your after-tax return.
And that’s the generous case. At the more aggressive 250/150 leverage common in the wrappers actually sold to retail, with retail pricing around 125 bps, the financing spread alone runs ~150 bps of NAV. Year one becomes $200,000 gross tax alpha minus $125,000 management minus $150,000 financing — negative $75,000, before any decay at all. The product is underwater from day one. The conservation law never even gets to assert itself, because the fee structure got there first.
Meanwhile, the 75/25 SPY/QQQ buy-and-hold investor paid around $12,000 a year in blended expense ratio. They realized nothing. They generated zero tax drag. The difference in net outcome over fifteen years is not marginal — it is orders of magnitude in the wrong direction.
This is before you account for the benchmark problem, which is where it gets much worse.
The benchmark that isn’t in the pitch deck
The product’s entire “tax alpha” number is quoted against a benchmark that itself generates continuous realizations. Of course a tax-aware strategy beats an actively managed one. But the relevant comparison for a long-term investor is not tax-aware-long-short versus actively-managed-churn. It is tax-aware-long-short versus two ETFs and a long nap. And on that comparison — which is the one the pitch deck is structured to avoid — the two ETFs win for almost any client who doesn’t have large ordinary-rate income to offset. Because the conservation law is trivially satisfied when you never realize anything to begin with. Deferral is free. You don’t need to pay anyone 100-150 basis points a year to give it to you.
At this point the sponsor’s fallback is always the same: forget the tax alpha, we also generate pre-tax alpha from the factor book — real returns, above market, net of fees. Fine, except: (a) if the factor book really did earn 200+ basis points of persistent net-of-fee pre-tax alpha, the manager would not need to lead with the tax story at all, and would not be marketing primarily to high-net-worth taxable clients; (b) the two decades of out-of-sample performance on most published equity factors have not been kind to that claim — value, momentum, quality, low-volatility have all gone through long stretches where net-of-fee returns trailed the index (I have written about this problem before); and (c) if you actually want factor exposure, you can buy a factor ETF with an expense ratio in the 15-30 bps range and be done with it. Wrapping the factor bet inside a leveraged tax-optimization engine does not make it more likely to earn alpha. It just makes the fee structure bigger. The pre-tax alpha defense is an argument for some factor product, maybe, for some investor. It is not an argument that specifically rescues the tax-aware long-short wrapper.
It gets worse for the product. The step-up-at-death case — item 3 above, which does most of the heavy lifting in the long-run IRR the marketing implies — is available to the buy-and-hold investor without the product. Hold the two ETFs until death, and the heirs inherit at stepped-up basis. Zero tax on the entire appreciation. Permanently. The long-short investor also gets step-up on their final embedded gain, but they paid 100-150 bps a year along the way for a structure whose largest tax benefit was going to be delivered by the step-up regardless. You paid fees for twenty years to receive, at the end, exactly the outcome you would have received for a dozen basis points a year.
And it gets worse still. The buy-and-hold investor has a second tool available that the long-short investor structurally does not: borrow against the portfolio.
This is not exotic. Post your SPY/QQQ as collateral against a securities-based line of credit. Rates inside of a mortgage, because the lender’s collateral is liquid and daily-marked. Spend the proceeds on whatever you want — house, tuition, boat, consumption, whatever consumption means to you. Loan proceeds are not income. You owe no tax on them, because you have not sold anything. The portfolio keeps compounding. The loan interest accrues. You die. Your heirs inherit at stepped-up basis, sell a piece of the position to retire the loan at zero tax (because basis equals market), and keep the rest clean.
Lifetime tax on the appreciation: zero. Consumption utility along the way: whatever you wanted. This is not a loophole. This is buy, borrow, die, and it is how essentially every large long-term holder of appreciated stock in the country actually manages their affairs.
The long-short product cannot do this efficiently. The short book consumes collateral capacity that would otherwise be available to borrow against. The realized short-term character generated by the structure has to be netted somewhere. The complexity of the positions impairs their use as clean collateral. The product is, at a structural level, at war with the single most powerful lifetime-tax-minimization strategy available to a long-term holder of appreciated securities.
So the full picture is not “tax-aware long-short beats buy-and-hold on tax alpha.” It is the opposite:
Buy-and-hold beats the long-short product on realized tax drag, for anyone without large ordinary-rate income to offset, because it has no realized tax drag.
Buy-and-hold delivers step-up at death for a dozen basis points a year instead of 100-150.
Buy-and-hold lets you finance consumption against the portfolio without realizing, and the long-short structure actively interferes with this.
Three different mechanisms. All pointing the same direction. The product is not merely dominated by the naive alternative — it obstructs the two most powerful tax-minimization tools the naive investor already has.
The product is optimized for someone who never unwinds
Here is the quiet structural thing worth saying out loud.
The sponsor charges fees on the full NAV of the fund. The NAV includes the embedded, deferred tax liability — the portion of the “value” that the investor does not actually own, because it is owed to the IRS. The manager earns fees on that liability, forever, or at least until the client unwinds. If the client unwinds, fees stop, and the conservation law asserts itself: the back taxes come due in one brutal lump.
Therefore: the product is optimized for a client who never unwinds. Which is to say, for a client whose exit is death, or charitable giving, or a migration to a lower-tax jurisdiction. None of those are investment outcomes. They are estate-planning or lifestyle assumptions.
That is not an accusation. The math is clean, the academic papers are careful, and the more honest sponsor decks will say something like this in the footnotes. It is simply that the product is being sold as investing, and it largely works, to the extent it works, as estate planning with an investment wrapper. Those are different things. The investor should know which one they’re buying.
The general test
The conservation argument applies, with different constants, to most structures sold primarily on tax benefits — 351 ETF conversions, exchange funds, opportunity zones, insurance wrappers. The question to ask is always the same: is this reducing my lifetime tax bill, or is it shifting it forward and hoping for a terminal event that extinguishes it? If the product only delivers on its pitch assuming a life plan you could have executed without it, then what you’re buying is the reassurance of having bought something.
The IRS has the monopoly on force. The books must balance. Anyone offering you a shortcut around either of those is offering you something else.
For more on applying physicist-style reasoning to other domains where it tends to dissolve more than it’s supposed to — including theology, political science, law, and the question of free will itself — see my book, The Science of Free Will.


This is excellent.