HFEA stands for HEDGEFUNDIE's Excellent Adventure.
Read from the start: chapter 1 The RuleLeveraged ETFs can lose a large part of their value in a short time.
Getting started
For the first weeks, before you buy anything.
A strategy from the Bogleheads forum in 2019: 55 percent in a three times leveraged S&P 500 ETF (UPRO), 45 percent in a three times leveraged ETF on long-dated US Treasuries (TMF), rebalanced to these weights every quarter.
The name comes from the author's forum name, Hedgefundie. The idea: in the decades before, stocks and long-dated government bonds often moved in opposite directions. Leveraged and rebalanced regularly, the pair was meant to deliver more than stocks alone, with declines one can still sit through.
The first proposal was 40 to 60; in the course of the discussion it became 55 to 45. The author explicitly described it as an experiment with a small, ring-fenced part of his assets.
This page computes the classic version: UPRO and TMF, 55 to 45, quarterly.
A leveraged ETF moves each day by a multiple of its index. In the original HFEA, both blocks are three times leveraged.
A leveraged product moves each day by a multiple of its market's move, up as well as down. Which multiple applies per block is stated with the product and on the rule card.
The word daily matters: the leverage applies to the daily move, not to a year. In this strategy both blocks are leveraged in the original version. In Europe the available bond blocks differ from the original in leverage or in maturity.
There is no minimum amount, but every rebalancing is two orders. With small amounts a deviation is often cheaper to live with than to correct.
The rule has no minimum amount, but it needs two positions that can be shifted in small steps. Every rebalancing is two orders, and fixed order fees fall due regardless of size.
With small amounts a deviation can therefore be cheaper to live with than to correct. What an order costs at your broker, and whether it trades fractions, decides more here than the strategy does.
The rule knows no entry point. Waiting for rate cuts, a pullback or a trend signal means following a different rule in the meantime.
The rule has no entry condition. Waiting for rate cuts, for a pullback or for a signal from a trend filter means following a different rule in the meantime.
That is permissible, but then it is not this strategy. Trend following on a moving average is a rule of its own with a page of its own.
Then what counts first is how far your shares sit from the target weights. From then on the dates of your version apply and, if set, the band.
Then the first question is how far your actual shares sit from the target weights. The rule itself knows no entry; it knows only deviations and dates.
In practice that means the dates of your version apply from the moment you run the strategy, plus a band if you set one. The curves on this page start at the beginning of the period you set, not at your purchase date.
Rebalancing and day to day
What happens on the dates and what happens between them.
In the standard version once a quarter, on the date; in between there is nothing to do.
In the standard version once a quarter, on the date. Between the dates there is nothing to do. If you have set a band, after the date one waits until a share leaves its band, and only then trades; how that works in detail is described at the trigger.
That is the big difference from a trend following rule: you do not have to look every day. LETF.io tells you when your version would trade.
Because the author set it that way after his backtest. Other dates are just as defensible; the results scatter strongly by date, without a best one being readable in advance.
The version from the original thread rebalances quarterly. The author named two reasons: the outcome that his backtest favoured most, and no better reason for any other date (post in the thread). The first reason is exactly the selection effect this page describes elsewhere: rhythm and date are chosen on the past.
What has been recomputed publicly since shows one thing above all: the date scatters. Run the same strategy on every possible date within the rhythm and the results lie far apart, without a best date being readable from it in advance (modern_football, Allocate Smartly).
In practice: more often means more trading, more spread and more tax. Less often means larger deviations in between. What your version does is on the parameter card for the rhythm.
Not an emergency: the deviation stays until you act. You can catch up at any time; afterwards the rhythm of your version applies again.
A missed date is not an emergency: the deviation stays where it is until you act or the next date arrives. If you have set a band, its breach comes on top of that. Unlike a trend following rule, nothing here hangs on a single day.
You can catch up as soon as you notice. What matters is that afterwards you follow the rhythm of your version again, rather than pulling the next date forward or pushing it back by feel.
For the streak in the tracker what counts is whether you answered in time; that is in the chapter Staying the Course.
Yes, and new money can be directed specifically into the underweight block. That moves the account towards the target weights without you selling.
Yes, and inflows even have a practical advantage here: fresh money can be directed into whichever block is underweight. That moves the portfolio towards the target weights without anything having to be sold.
Whether your version provides for that is stated in the inflow rule. This page only describes the mechanism.
The split and the bond leg
What the strategy is made of, and why.
There is no right one; which looks strongest in hindsight depends on the interest rate environment of the period. More stocks make the result more equity-like, more bonds make the bet on the counter-movement larger.
Your version fixes one, and the range of usable splits is wider than the debate suggests. Which one looks strongest in a backtest depends heavily on the interest rate environment of that backtest.
The higher the equity share, the more equity-like the result. The higher the bond share, the larger the bet on the counter-movement.
On the origin: the first proposal of 40 to 60 was meant to give both legs about the same risk, because stocks swung about one and a half times as much as long-dated bonds. In August 2019 the author switched to 55 to 45 after a backtest from 1955. This value, too, was chosen on the past.
Because only for the classic version do blocks with the leverage of the original exist. The rule is the same everywhere; what differs is the product in the bond role.
The classic version is the one for which blocks with the leverage of the original exist. European pairs are versions of their own; the page does not compute them, they can be set up in the Tracker.
The rule is the same in all versions. What differs are the products in both roles: three times leveraged equities exist in Europe only as ETPs, funds go up to two times.
Both are a different version. In public recalculations cash did worse than bonds, and gold was tested there only as an admixture to the bond leg.
Both make a different version, and both have been recomputed publicly (ZahlGraf, part 6).
Cash comes out consistently worse there than the bond leg. It does not fall in equity crises, but it does not rise either, and rebalancing lives off the second leg rising.
Gold is not tested there as a replacement for the bonds but as part of the bond leg: as an admixture in the hedging part beside the bonds. What is computed and shown on this page holds for the versions with a bond leg.
Then the core of the strategy does not work, as in 2022 with rising rates. HFEA is therefore also a silent assumption about the interest rate environment.
Then the core of the strategy does not work in that phase. 2022 is the example: both sides fell sharply because rates rose.
That is why HFEA is not only an allocation but implicitly an assumption about the interest rate environment. No backtest carries that assumption as a warning, but every one of them contains it.
That cannot be said in advance. In 2022 TMF lost about 73 percent, UPRO about 57 percent and the 55/45 mix about 64 percent: exactly the case the strategy has no answer for.
2022 hit exactly the assumption the strategy rests on: stocks and long-dated bonds fell at the same time, and the bond leg fell harder. That was not a fault in the execution but the case the strategy has no answer for. For the leveraged bond leg 2022 is not the only year of that kind: in backtests across the phase of rising rates from the 1950s to the early 1980s a leveraged long-maturity leg lost over decades, with a decline near total loss.
What has changed since is the level of interest rates: long-dated bonds pay a coupon again. The leveraged bond leg pays the short rate for its leverage. The part of the return that the leverage contributes is therefore negative with an inverted yield curve.
Whether the counter-movement returns because of that, nobody knows. Through history it was there at times and absent at others, and this page treats it as a property of certain decades rather than a fixed quantity.
They make leverage more expensive: a three times leveraged ETF finances about twice its assets at the short rate. Rising rates also push down the price of long-dated bonds, and that hits the bond leg three times over.
For a sense of scale: at a short dollar rate of 3.9 percent (SOFR, as of November 2026) the financing costs about 7.8 percentage points a year per block, before fees and the swap counterparty's markup.
A higher coupon on long-dated bonds helps the bond leg. From the interest alone, though, it only carries itself as long as the long yield stays above about two thirds of the short rate.
The equity leg pays the same short rate for its leverage. High short rates therefore make both blocks more expensive, regardless of the direction of prices.
Very strongly. The underlying index of long-dated US Treasuries has a duration of about 16 to 17 years, so about 50 at three times leverage: one percentage point more yield costs about half, to a first approximation.
Duration measures by how many percent the price of a bond falls when its yield rises by one percentage point. Long maturities have a high duration, and leverage multiplies it.
In 2022 the yield on 20-year US Treasuries rose by a good 2 percentage points, and TMF lost about 73 percent. The same mechanism works the other way round when yields fall.
Duration changes with the level of interest rates; the current value is given in the index factsheet.
HFEA began as a simplified, leveraged form of it: 40 to 60 was meant to give both blocks about the same risk, while the 55 to 45 version leans towards stocks. Risk parity weights continuously by measured risk, HFEA holds fixed weights.
The original thread carries risk parity in its title. With fixed weights, however, the risk contribution of both legs shifts as soon as their volatility changes, for example when bonds become more turbulent than before.
Ready-made funds that bundle stocks and bonds with leverage in one product exist mainly in the United States. They are products of their own with rules of their own and are not computed on this page.
That is a version of its own that has been recomputed publicly. Running both strategies means following two rules with different dates.
That is a version of its own, and it exists: part 9 of ZahlGraf's series computes the combination and arrives at a lower risk than HFEA on its own.
What LETF.io carries are the two strategies separately. The curves on this page hold for HFEA without a filter, the curves on the trend following page for the filter without a bond leg.
Running both at once means following two rules at the same time, and the dates of one do not coincide with the signals of the other. That is the practical price of the combination.
Products and leverage
What you fill the two legs with.
Because of the declines: leveraged stocks on their own lost the largest part of the amount committed at times in the past. The bond leg is meant to dampen that and holds you back in pure equity rallies in exchange.
Because of the declines and because of the time that follows them. A leveraged equity holding on its own has shown declines in the past that cost the largest part of the amount committed, and afterwards it took a very long time until the old level was reached again.
The bond leg is the attempt to dampen those declines, and rebalancing is meant to turn the difference between the two legs into an additional contribution.
The price stands next to it: in phases where stocks simply rise, the bond leg holds you back. And when both fall together, it does not help. That trade-off is the strategy.
That is a different strategy. Without leverage the reason for the pairing falls away; a mix of leveraged and unleveraged holdings is the benchmark HFEA has to measure up against.
That is a different strategy, not the same one with other blocks. Without leverage the reason for the pairing falls away, because the counter-movement of the bond side is meant to cushion the volatility of a leveraged portfolio.
Part of what HFEA aims for can also be approximated with a plain mix of leveraged and unleveraged holdings. That is the benchmark the strategy has to measure up against.
A version of its own. A narrower index swings more, and with leverage that counts twice, in the price and in the volatility drag.
A version of its own, recomputed publicly in part 7 of ZahlGraf's series. There, though, the Nasdaq-100 does not stand in place of the broad index but as an admixture in the growth-oriented part.
The difference is one of character: a narrower index swings more, and with leverage every additional swing counts twice, once in the price and once in the volatility drag.
For HFEA, LETF.io computes the classic version with a broad US index. Versions on the Nasdaq-100 exist on the trend following page, but without a bond leg there.
An ETP is a debt security: if the issuer goes under, you are a creditor. The usual papers are collateralised but do not have the legal position of a fund.
An ETP is not a fund but a debt security, that is a claim against the issuer. If the issuer goes under, you are a creditor and not the owner of ring-fenced assets.
The usual papers are collateralised: securities sit behind them that would be realised in that case. That reduces the risk but does not create the same legal position as a fund.
ETPs on long-dated bonds are debt securities, whatever their leverage. Unleveraged long-maturity ETFs are funds and do not carry this risk, but they lack the leverage.
The provider sells and pays out, and that realises gains or losses regardless of the rhythm. The affected role then needs a new product.
It happens: providers discontinue leveraged products. With a fund the assets are sold and paid out; with a debt security it is redeemed under the procedure laid down for it.
Either way that realises gains or losses, regardless of the rule's rhythm. For the strategy it means the affected role needs a new product and therefore a new version, because every version depends on one specific bond or equity block.
Backtests and risk
What the curves say and what can go wrong.
Very deep: in our backtest from 2009-06-25 the largest decline was about 71 percent. That is why the original thread already describes a ring-fenced portion of a portfolio, not the whole account.
Very large. Both blocks are leveraged in the original version, and in backtests the interim declines regularly reach a large part of the amount committed.
That is why the original thread already describes a ring-fenced portion of a portfolio rather than the whole account. Committing everything changes the starting position, not just the amount.
In our backtest (55 to 45, quarterly, with distributions) the largest decline was about 71 percent with the trough on 2023-10-27, after the high on 2021-12-27; the calendar year 2022 ended at about minus 64 percent.
For the early decades they rest on constructed series with assumptions about financing and fees, and they flatter when financing costs are missing. They show how the rule behaves, not future results.
For the early decades they rest on constructed series, because the leveraged products did not exist then. The series are built from index data plus assumptions about financing costs and fees.
The results therefore hang on those assumptions and flatter when financing costs are missing. Our backtest computes on real prices only and with distributions in both legs. It begins on 2009-06-25, the first real UPRO price; the TMF prices go back to 2009-04-16. Take the curves as illustration of how the rule behaves, not of future results.
That depends on the period: in the years of falling rates up to 2021 HFEA was ahead in backtests, in 2022 far behind. Two benchmarks make sense: the same blocks without rebalancing and an unleveraged portfolio.
In the chart the reference is the same two blocks, held without rebalancing. That comparison answers the question of what the rebalancing achieved in this period.
The second sensible reference lies outside the chart: an unleveraged portfolio of stocks and bonds. It answers the question of whether leverage and effort achieved anything.
The two belong together. Comparing only against the leveraged holding shows the dampening but not its price; comparing only against the unleveraged portfolio shows the price but not what it bought.
Long enough for several interest rate environments, not a few months. Anyone who exits after a bad year has left the rule.
The strategy is built for long periods, and single years say little about it. That is not reassurance but a property: judging the rule takes a period covering several interest rate environments, not several months.
Anyone who exits after a bad year has left the rule. From that point you no longer know what you are actually testing.
Taxes and region
What your account and your residence make of it. Tax information as of November 2026; not tax advice.
Rebalancing means selling, and selling realises gains. US backtests from tax-advantaged accounts understate this burden in a regular taxable account.
That depends on your region and is in the chapter Implementation and Taxes of this page. The core is the same everywhere: rebalancing means selling, and selling realises gains or losses.
In the United States the original thread ran the strategy inside a tax-advantaged account, where this effect does not arise. Backtests from that environment therefore understate the burden in a taxable account.
Only the gain in the share actually sold is taxed, and that share is usually small. In exchange the tax falls due regularly; inflows into the underweight block replace sales.
In public backtests the tax on rebalancing costs roughly one to two percentage points of return a year. The important point: what is taxed is not your account and not the whole block you sell, but only the gain in the share you actually sell.
Because only the deviation is closed on the date, that share is usually small against the position. In exchange the tax falls due regularly instead of once at the end, and the amount paid out no longer compounds.
Inflows reduce this burden: directing new money into the underweight block replaces part of the sales. The chapter Implementation and Taxes works that through for your case.
In Germany the order of purchases applies: the oldest units are sold first, and they usually carry the highest gains (§ 20 Abs. 4 Satz 7 EStG).
Inside the wrapper the rebalancing triggers no tax for you. Against that stand fees, often a share of the gain, no partial exemption and the issuer risk over many years.
Inside such a wrapper the rebalancing triggers no tax for you, because it is not you who buys and sells but the wrapper. Tax falls due only when you sell the paper itself. For a strategy that sells several times a year that sounds like a lot.
Against it stand three things: an ongoing fee, often a share of the gain, and no partial exemption, because a certificate is not a fund. On top of that comes the issuer risk, which is no small item for a holding kept over many years.
This page does not compute such wrappers. It shows the strategy in products of its own, the way the chapter Implementation and Taxes describes them.
As a rule, no, because the US funds have no key information document. European blocks exist, but for the bond leg only with a different maturity or different leverage than TMF.
As a rule, no. Products sold to retail investors in the EU need a key information document in an official language, and the US funds do not provide one; most brokers therefore do not allow the purchase.
Individual routes exist; they bring their own questions: whether the partial exemption applies to a US fund is not clear in every case, and a paper your broker stops carrying tomorrow may have to be sold.
With European blocks these questions do not arise, but they differ from the original; how is in the chapter Implementation and Taxes.
Both blocks depend on the US market; in a euro account the exchange rate moves your result along with it. Whether a product hedges the currency is a matter of product choice, not a property of the strategy.
Both blocks depend on the US market. If your account settles in euro, the exchange rate moves your result along with it, in both directions.
Individual European products hedge the currency, others do not. That is a difference in product choice, not a property of the strategy.