By Victor Haghani and James White¹

Quick Summary (by ElmAI)

• For a typical investor, the fees paid to a leveraged long/short direct indexing (LSDI) tax-loss harvesting manager eat up most of the tax savings. In our base case, the investor pays more than half as much in fees as the taxes she was trying to avoid, and ends up with a lower post-tax expected return than if she’d just sold, paid the tax, bought an index fund and moved on.
• Even if you believe the manager can generate 0.5% per annum (pa) of stock-picking alpha, under the best of circumstances, there’s essentially only a 50/50 chance that the LSDI program leaves you better off than the simple sell-and-reinvest alternative.
• The case for LSDI improves meaningfully in specific circumstances – such as when the manager has a credible and significant source of alpha (much greater than 0.5% pa), when step-up in basis at death is around the corner, or when the investor is imminently moving from a high-tax to a low-tax state – but for many investors being pitched the strategy, the math doesn’t support the hype.

Introduction

It’s tax season, and we’ve been reading a lot about taxes – and strategies for mitigating them.² In this note, we’ll take a close look at one such strategy, known as leveraged long/short direct index tax-loss harvesting (LSDI), and explain how investors being pitched the strategy can assess whether it’s right for them.

According to research by BlackRock, over $1 trillion in U.S. household wealth is held in single-stock positions with a low cost basis, where the owner would like to diversify but is reluctant to sell due to the immediate capital gains tax hit.³ LSDI is designed to solve this problem: it allows an investor to sell a concentrated, highly appreciated stock holding and replace it with a diversified portfolio of hundreds of stocks, without paying capital gains tax along the way. We wrote (somewhat skeptically) about this a year ago in “Out of the Frying Pan and Into the Fire,” but we’ve been hearing so much about the merits of this strategy – in the press and in conversations with our clients – that we thought we should take a closer look to see if our earlier analysis missed something.

How LSDI works

Consider an investor who was an early employee of Shopify Inc. (ticker: SHOP), and now owns $10 million of SHOP with a basis close to 0.⁴ This single holding represents over 80% of her investment portfolio, and she’d love to sell it and buy a nicely diversified basket of stocks instead – but she lives in the Bay Area and would face a combined Federal and California capital gains tax rate of 37%, so she’s looking for something better.

A wealth manager proposes she enter an LSDI program. The manager will place her $10 million of Shopify stock into a separately managed account (SMA), simultaneously going long $10 million of individual stocks and short $10 million of others. Over time, the manager will realize capital losses from both the long and short portfolios. Each time a position is sold, it’s replaced with another stock to keep the long and short portfolios at their target sizes, while complying with the IRS wash sale rule.⁵

Individual stock selection will also be influenced by tracking error constraints and a stock-picking model. It’s important to realize an important distinction between long-only tax loss harvesting and LSDI; the former need not be accompanied by stock picking, while the long/short version does require individual stock selection to get the desired tax treatment.

As realized losses accumulate, the manager sells some of the investor’s Shopify holding and replaces it with a diversified basket of stocks (the “replacement portfolio”), using the harvested losses to offset the capital gains on each Shopify sale. The manager also tax-loss harvests the diversified replacement portfolio itself.

After awhile, all of the Shopify stock has been sold and the investor holds the replacement portfolio, plus an equal amount of long/short portfolio. At this point, the manager can use further harvested losses to unwind the leveraged long/short portfolio, but it will not be possible to unwind it entirely.

The investor can either continue the program indefinitely, or at some horizon, ask the manager to close out all the remaining leveraged long and short positions, generating some combination of long-term capital gains on the longs and short-term capital gains on the shorts (current tax rules require that realized gains from short positions always be treated as short-term capital gains regardless of holding period). This finally leaves the investor with what she wanted: conversion of her concentrated holding into a diversified portfolio of individual stocks with no capital gains tax paid until the long/short portfolio is unwound at her chosen horizon.

Looking under the hood

It sounds pretty good, doesn’t it? It’s also pretty complicated. It took us a while to wrap our heads around all the nuances involved, and to build a simulation model to analyze its merits (which we describe in detail in the Appendix). Before we go further, we should note that we are not tax experts and we have not had direct conversations with any of the many purveyors of this strategy, though we have reviewed with friends and clients the general contours of what is typically proposed. We should also note: if you believe the manager can pick stocks that will outperform the ones they short – while doing the tax-loss harvesting and staying on the right side of the wash sale rule – our analysis will help you decide how much outperformance you’d need to make the strategy worthwhile.

Our results depend on a few assumptions, and we want to say upfront that results depend heavily on an individual’s circumstances. For some situations, LSDI can make sense. In this note, we’re primarily going to discuss a central case with what we think are a set of typical assumptions, and most of the assumptions on markets and implementation are chosen to put the LSDI program in a favorable light. In the Appendix, we explain our simulation model in detail, and you can find the simulation results in this Google sheet.

To start, we’ll keep tax rates and rules constant over the investment horizon (37% and 54% tax rate on long-term and short-term capital gains respectively). For the stock market, we use an average single-stock volatility of 30% pa, roughly what it has been over the past five years, and average annual stock price growth of 6%.⁶ For the concentrated stock holding, we assume the same expected average annual return of 6%, but with higher volatility of 50%.⁷ We’ll assume all stocks pay no dividends. We assume the manager charges fees of 0.5% pa on the diversified long portfolio of stocks (but not on the Shopify position), and that the custodian of the SMA charges 0.75% on the long/short portfolio.⁸

Let’s start with a 20-year horizon – long enough to complete the transition out of the appreciated asset and to harvest enough losses to meaningfully reduce the size of the long/short portfolio. In our central case, selling down the appreciated takes a mean time of about 3 years. We assume that at the end of 20 years, the remaining long/short portfolio is unwound, taxes are paid, and the investor is left with a diversified holding of individual stocks, and stops paying fees to the manager.

We calculate that the investor will have earned an average after-tax, net-of-fees compound return of 2.05% pa on her initial $10 million.

What if she had simply sold her $10 million Shopify holding, paid $3.7 million in capital gains tax, and invested the remaining $6.3 million in a low-cost, broad stock market index fund? In that case, we calculate an average after-tax return of 2.16% pa on the same $10 million starting value.

The main reason the LSDI program’s expected return is 0.11% pa lower is that the fees eat up most of the tax benefit, though paying short-term capital gains tax to cover the shorts, and the volatility drag of holding SHOP for longer also hurt the LSDI program’s after-tax compound return. Over 20 years, the investor will have paid $2.7 million in fees.⁹ Compare this to the $3.7 million of capital gains tax she was trying to avoid, or at least defer. She’s paid less to Uncle Sam, but more than half as much to her LSDI manager. And at the end, she’ll still have to pay long-term and short-term capital gains tax to unwind the remaining long/short portfolio, plus she’ll have taken more risk over the horizon – and on top of all that, she’ll have $26 million in unrealized gains in her replacement portfolio. This is far more embedded capital gains liability than in the index fund she could have bought fresh 20 years earlier, which would have only $14 million of unrealized gains. This means that, if she needs to sell some of her assets for spending or wants to invest it in other assets, she’ll have to foot about double the tax bill as a result of going down the LSDI road.

Across 2,000 simulations, the difference in returns between the two strategies had a standard deviation of 2.9 percentage points. This means that, even if you think the manager can generate an extra 0.5% of stock-picking alpha (the average LSDI excess return will be about 0.39% pa) there would be only a 55% probability, hardly better than a coin-flip, that the LSDI program would give you a better compound return than selling your concentrated holding upfront, paying the taxes and investing in your desired portfolio.

Taking off the rose-colored glasses

To arrive at the results we described above, virtually every assumption we made was generous to an LSDI program, setting aside whether the manager can generate alpha by successfully picking stocks. Here we’ll try to get a bit more realistic:

• Our example uses the highest state tax rate in the country. Most investors face lower combined rates than California residents, and will therefore find LSDI less attractive than our example suggests. If our investor had been a Texas resident, the LSDI program would have an annual after-tax return 0.88% worse than just selling the Shopify holding, paying capital gains tax upfront and investing in a stock index fund.
• Wash sale constraints reduce harvesting in practice. We assumed daily tax-loss harvesting as if the wash sale rule didn’t bind – i.e., that every loss could be realized and the same stock repurchased immediately. This maximizes the amount of tax-loss harvesting achievable. However, in the real world, the wash sale constraint, together with limits on acceptable tracking error between the long and short portfolios, considerably reduce the losses that can be harvested. If we delay harvesting on 20% of losses each year, the expected after-tax return of the LSDI program in our example drops by 0.29% pa.
• The investor bears more risk under LSDI. The LSDI portfolio is riskier than the immediate-sale alternative for two reasons: the investor retains exposure to her concentrated Shopify holding for longer,¹⁰ and the long/short portfolio itself adds risk. We calculate that over the 20-year period, she’ll experience about 3 percentage points higher portfolio volatility than in the immediate-sale case, ignoring any risk arising from the long/short portfolio of individual stocks. This extra risk reduces her expected compound return, and for a typically risk-averse investor, the cost of bearing that risk goes beyond its drag on compounding.

• We ignored transaction costs. Bid-ask spreads, commissions, market impact, and predatory trading against a predictable rebalancing pattern all take a toll on the LSDI program’s heavy trading. With our base case assumptions, we estimate average portfolio turnover of around 400% annually on the net value of the portfolio of the LSDI program over the 20-year horizon.
• At most other horizons LSDI is worse, particularly shorter ones. For example, if we moved the horizon to five years from 20 years, LSDI becomes more than 2% pa less attractive relative to the sell-and-move-on approach. This primarily arises from paying the higher tax rate (17% higher) on short-term capital gains arising from covering the LSDI short positions at the horizon.
• Tax rates and rules may adversely change. Many economists and tax experts believe tax rates are more likely to rise than fall, and that favorable provisions like step-up in basis risk being curtailed. Any such changes would further erode the case for LSDI.
• Difficult to implement for non-public stock-holdings. Many potential use cases involve investors who own stakes in non-public companies (we hear that employees of SpaceX, Stripe, Anthropic, OpenAI, and Databricks are hearing a lot about LSDI these days) and it is difficult for them to get the leverage needed to implement the strategy as described above. Slower-moving programs that start with smaller amounts of LSDI as the investor starts to sell and diversify away from the concentrated holding in secondary-market transactions have less attractive economics than the case we set out above.
\
• Better alternatives to an immediate sale exist. For example, we assumed that the investor sold her appreciated asset and bought and held an index fund. In practice, rotating among similar index ETFs could provide effectively free tax-loss harvesting opportunities.

When LSDI can make sense

There are circumstances in which LSDI may be a reasonable choice, though none of the below considerations on their own will be conclusive:

• You expect the manager to generate significant alpha. If you believe the manager of the program can consistently and materially beat the market through stock selection, the calculus changes. A recently published article from researchers at AQR notes that “Investors should not be drawn to these strategies solely because of their tax benefits. Instead, they should evaluate them as they would any other risky investment and avoid those lacking a defensible source of pre-tax alpha.” We agree with AQR, and our analysis quantifies just how large that alpha needs to be, depending on your particular circumstances.
• You’re moving from a high-tax to a low-tax state. If you plan to relocate in the near future, the temporary tax deferral may have significant value. We model the scenario of moving from California to Florida, and with a ten-year horizon example, LSDI comes out 0.9% pa ahead of the immediate sale case.
• You expect step-up in basis and soon. If you expect to pass away within the next ten years or so, and believe your appreciated long-side assets will receive a step-up in basis at death, LSDI has an expected compound return 0.44% pa higher than the immediate-sale case. can be a material benefit adding about 1% pa of after-tax return, but to longer horizons of 20 or 30 years, step-up in basis adds about 0.15% and 0.07% pa of return to the LSDI case, respectively, hardly changing the overall picture. Note that short positions do no get stepped up.
• Charitable intentions. There may be cases where the interaction with intentions to donate appreciated assets to charity may warrant engaging in an LSDI program. However, it’s often superior to make donations sooner, rather than saving them for offsetting gains down the road. We also note that directly donating a short position to a charity is not allowed.
• Hedge fund investors. For example, you invest in hedge funds which give you short-term capital gains each year, and you want to use an LSDI program to convert short-term capital gains to long-term capital gains. We have not analyzed such a case, but it seems plausible that an LSDI program produces benefits after fees, but that will very much depend on the particular circumstances and assumptions.
• More complex structured variations. We’ve heard of some investors wrapping LSDI in complex structures. We don’t know much about them, but we suspect they are not available or cost-efficient unless the size of the program involves $50 million or more of gains.
• You’d rather pay a wealth manager than the IRS. This is a personal preference, but it’s worth being explicit about it.

Connecting the dots

We think it’s important for investors to focus on returns after-tax, after-fee, after-inflation, and after-cost-of-risk. But sometimes we worry that investors become so focused on minimizing taxes that they lose sight of this overall objective. Long/short direct index tax-loss harvesting may be a case where the promise of paying no taxes “as far as the eye can see” short-circuits a more holistic analysis – one that also accounts for fees, other costs, risk, and a realistic assessment of the odds of beating the market through stock picking.

An immediate sale of an appreciated asset and reinvestment in index funds is not the only alternative to an LSDI program, and unlikely to be the optimal action to take with a highly appreciated, concentrated holding. Other approaches exist and may do the job better than either of the two alternatives discussed in this note.

If you’re being pitched or exploring an LSDI program, we recommend asking the manager to show you results with no pre-tax alpha and ask them to use our assumptions to see if they roughly match our output, including not just the central outcome but the variability around it.

This is admittedly a more personal observation, but we also wonder: if given the choice, would more people on reflection prefer to contribute $100 in taxes – incrementing the common resources of their state and nation – than to pay $100 in fees to a wealth manager, no matter how sweet, intelligent and friendly we are?

Rather than reporting on LSDI being part of a “$1 trillion business…a new gold rush… helping wealthy Americans crush their tax bills,”¹¹ we hope the next Bloomberg or WSJ headline will tell a more complete story: the only party that is with near-certainty better off is Wall Street, while wealthy investors and the government, both individually and together, are likely worse off.

If you’d like to discuss our analysis in greater detail, or see how it applies to you or someone you know, please don’t hesitate to reach out.

Appendix

Simulation results are available in this Google sheet.

Simulation overview

Our simulation model starts with an index of 200 names, with a Pareto distribution of market-cap weights calibrated to match the S&P500. Each name follows geometric brownian motion (GBM) with a fixed expected return (6% arithmetic), fixed volatility (30%), and fixed pairwise correlation (0.3), resulting in a volatility of about 17% for the index. The simulation is run on 2000 daily paths.

The long ‘reinvestment’ portfolio invests in the index, and the long/short portfolio invests in the index on both sides – so the net PnL of the long/short portfolio is always zero, before fees and transaction costs. We also assume no wash sale rule, so positions with losses can be sold and immediately replaced. These are obviously unrealistic assumptions, but in the direction of being generous to the LSDI strategy – the simulation results should reflect an upper bound on the amount and value of harvesting available.

Harvested losses from the reinvestment portfolio and the long/short portfolio are used to absorb the embedded gain in an existing appreciated asset. The simulation runs in two phases – first we use realized losses to gradually sell the asset without realizing gains, then losses are used to partially unwind the long/short book pro-rata, also without realizing gains.

Phases

The LSDI simulation runs in two sequential phases, plus a final horizon settlement.

Phase 1 – Sell-down.

The goal of this phase is to move out of the appreciated asset entirely without paying tax. Daily loss harvesting runs on the reinvestment portfolio and the long/short book, building up ST and LT loss pools. At each year-end, the model computes the maximum number of asset units that can be sold such that the realized LT gain from the sale is fully absorbed by the available loss pools. Those units are sold, the corresponding losses are consumed from the pools, and the cash proceeds are immediately reinvested in the index as a new tranche. That new tranche joins the daily harvesting loop for the remainder of the simulation, generating additional losses going forward. Phase 1 continues year after year until either the appreciated asset is fully exhausted or the simulation reaches the horizon.

Phase 2 – Wind-down.

Once Phase 1 ends, the goal shifts to unwinding the long/short book itself – closing long and short positions without realizing net gains. Daily harvesting still runs on the reinvestment and long/short portfolios. At each year-end, we solve for the largest fraction f such that closing f of every remaining long and short position produces zero net tax, given the current loss pools and the composition of the unrealized gains. That fraction is closed across all active positions, and the loss pools are drawn down accordingly. The wind-down repeats annually; each year’s harvesting restocks the pools, enabling another close-out slice. The portfolio is thus progressively drained until either it’s fully closed or the horizon arrives.

Horizon

At the horizon date, we do the following:

• Remaining appreciated asset (if any): taxed on its full embedded LT gain at the terminal LT rate (0% if `stepup = True`).
• Reinvestment portfolio (tranches): not taxed – the modeling assumption is that these positions receive their own step-up or are held indefinitely past horizon. Their unrealized gains are reported as an output but don’t hit the terminal tax bill.
• L/S book (if not fully wound down): long unrealized gains taxed at the terminal LT rate (0 if `stepup`), short unrealized gains taxed at the terminal ST rate (never stepped up – shorts don’t get step-up treatment). The long/short net market value is ≈ 0 by construction (market-neutral), so only the tax on the unrealized side matters.
• CA/FL relocation: if the state is `CA/FL`, the terminal settlement uses non-CA rates (the investor has moved to FL by horizon), while the immediate-sale portfolio uses CA rates.

Portfolios

Reference portfolio: At \(t = 0\) the appreciated asset is sold outright; the investor pays LT tax on the embedded gain. There’s no harvesting, no fees, no turnover. This is the counterfactual against which LSDI IRR is measured.

LSDI – four moving pieces:

1. Appreciated asset: Assumed to have zero basis, and follows its own correlated GBM path with the same expected return as each stock in the index. It is held intact until Phase 1 year-ends, when units are gradually sold – each year only as many units as can have their realized LT gain fully offset by available ST+LT loss pools. Sale proceeds fund the reinvestment portfolio. Any residual asset still held at horizon is marked to market and taxed at the terminal LT rate (0% if `step_up`).

2. Reinvestment portfolio (tranches): Each annual asset sale purchases a new tranche. Tranches are bought at the current stock prices using the same Pareto weight template, and each tranche thereafter participates in daily loss harvesting on its own basis (so each tranche has its own ST/LT clock and its own unrealized P&L). The reinvestment portfolio is the sum of all live tranches. It is not taxed at the horizon – the implicit assumption is that tranches receive a step-up or are held indefinitely past horizon. Its unrealized gains are reported but don’t hit the terminal tax line.

3. Long-side L/S portfolio: The long index portfolio which is harvested daily as available, whatever’s left from the wind-down phase is unwound at horizon at the prevailing LT rate (0 if step-up).

4. Short-side L/S portfolio: The short index portfolio which is harvested daily as available, whatever’s left from the wind-down phase is unwound at horizon at the prevailing ST rate.

Further Reading and References

• Haghani, V. and White, J. (2025).“Out of the Frying Pan and Into the Fire: Selling A Highly Appreciated Stock Without Paying Taxes?” SSRN.
• Haghani, V. and White, J. (2024). “Direct Indexed Tax Loss Harvesting: Is the Juice Worth the Squeeze?” Elm Wealth.
• Haghani, V. and White, J. (2020). “To Realize, or Not to Realize.” Elm Wealth.
• Haghani, V. and White, J. (2023). “Chapter 17: Tax Matters.” The Missing Billionaires: A Guide to Better Financial Decisions. Wiley.
• Haghani, V. and White, J. (2017). “How Much Should the Tax Tail Wag the Asset Allocation Dog?” Elm Wealth.
• Haghani, V. and White, J. (2018). “US Tax Reform Leaves Even Less of the Pie for Individual Investors in Alternatives.” Elm Wealth.
• Haghani, V., Hilibrand, L. and White, J. (2019). “When it Pays to Pay Capital Gains.” Elm Wealth.
• Liberman, J., Sosner, N. and Freitas, P. (March 2026). “The Tax Benefits of Pre-Tax Alpha.” SSRN.
• Liberman, J., Sialm C., Sosner, N. and Wang, L. (2020). “The Tax Benefits of Separating Alpha from Beta.” Financial Analysts Journal 76:1.
• Moehle, N., Kochenderfer, M., Boyd, S., and Ang, A. (2021). “Tax-Aware Portfolio Construction via Convex Optimization.” BlackRock AI Labs.
• Sialm, C. and Sosner, N. (2018). “Taxes, Shorting, and Active Management.” Financial Analysts Journal 74:1.
• Quantitative Investment Strategies. (Feb 2018). “The Benefits of Tax Loss Harvesting Strategies in Investment Portfolios.” Goldman Sachs Asset Management.

1. This is not an offer or solicitation of investment services or financial advice. It reflects the views of the authors, subject to change, and not necessarily the views of Elm Wealth, where the authors work. Although this article discusses taxation, the authors are not tax experts and nothing herein should be construed as tax advice. Past returns are not indicative of future performance. We thank our partner Jerry Bell, and our friends Mark Grinblatt, Larry Hilibrand, Bill Montgomery, and Rob Stavis for their helpful comments.
2. Such as this one from Bloomberg: “Slashing Tax Bills for Rich Investors Is a $1 Trillion Business” by Justina Lee and Denitsa Tsekova (March 16, 2026).
3. BlackRock. “Strategies for Managing a Concentrated Stock Position.” (2024).
4. We chose Shopify for this example to liven things up a bit versus the typical use of Apple and Nvidia for these examples.
5. See Section 1091(e) of the Internal Revenue Code. The wash sale rule imposes a 31-day window before or after a loss realization during which you cannot replace your position with a “substantially identical” security.
6. In more technical terms, 6% is the arithmetic expected return, net of dividends.
7. Shopify’s annualized volatility of daily returns over the past two years was 58%. Many other high-flying companies which might have become concentrated public market single stock positions – such as Broadcom, Crowdstrike, Tesla, Palantir, Coinbase, Datadog, Twilio and Pinterest, to name a few – have realized volatility of over 50% over the past two years.
8. We believe these are typical fees. However, we have recently heard that Fidelity and Schwab are no longer offering SMA custody for new LSDI programs, or if they do, that they are charging a long/short fee in excess of the 0.75% we used for our analysis.
9. We don’t actually present value the fees in the analysis – they’re just counted as part of the cash-flows when calculating the IRRs, but if you want to get a sense for what the fees are in present value terms, they’re about $1.9 million discounted at 4%.
10. This risk can be partly mitigated by structuring the long/short portfolio to account for the single stock holding risk – which some LSDI programs try to do – but for most single stocks, it’s challenging to reduce the single stock idiosyncratic risk significantly.
11. Justina Lee and Denitsa Tsekova, “Slashing Tax Bills for Rich Investors Is a $1 Trillion Business,” Bloomberg, March 2026.