Five components. Five identical weights. An apparently very simple structure. The Golden Butterfly is one of the most recognisable lazy portfolios precisely because of its symmetry: 40% stocks, 40% US government bonds split between long and short maturities, 20% gold. But that simplicity hides a hard question for anyone investing from Europe: which parts of the portfolio are essential to its logic, and which simply depend on the fact that it was born in the United States?
In this instalment we won’t look for the Golden Butterfly that would have returned the most in the past. We’ll do what we promised in part one: first reconstruct the original logic, then freeze a European adaptation, and only afterwards look at the results.
The original Golden Butterfly
The Golden Butterfly comes from the work of Tyler, the author of Portfolio Charts. In his 2015 article «Catching a Golden Butterfly» he presents it as an evolution of the Permanent Portfolio: it keeps the bond barbell of long and short Treasuries and the gold, but raises the equity share to 40% by adding a specific exposure to small cap value.
| Weight | Original component | Function we will try to preserve |
|---|---|---|
| 20% | US Large Cap Blend | Broad growth-oriented equity core |
| 20% | US Small Cap Value | Second source of growth with a small/value tilt |
| 20% | US Long-Term Treasury | High duration and a different response in rate/deflation regimes |
| 20% | US Short-Term Treasury | Stability, liquidity and the short end of the barbell |
| 20% | Gold | Real asset and diversifier against stocks and bonds |
Why buying five European equivalents isn’t enough
Portfolio Charts already offers an Italian localisation: 20% Europe Large Cap Blend, 20% Europe Small Cap Value, 20% Europe Long Term Treasury, 20% Europe Short Term Treasury and 20% Global Gold. It’s a perfectly legitimate translation of the domestic logic Portfolio Charts uses. But it’s not the only possible one.
Rebalix wants to answer a slightly different question: if an investor who lives and spends in euros wanted to preserve the economic role of the five components without automatically turning every US exposure into a European one, how could they do it? The distinction matters most for equities. A European investor is not obliged to treat Europe as their only domestic equity market. They can use a global developed equity core and keep the government sleeve in euros instead.
There is no inevitable translation. There is a methodological choice to declare.
The following selection is therefore frozen before the backtest. We won’t swap an ETF because an alternative would have produced a better CAGR, a smaller drawdown or a prettier curve.
The Rebalix adaptation for the euro investor
| Weight | Role | Chosen instrument |
|---|---|---|
| 20% | Developed equity core | iShares Core MSCI World UCITS ETF — SWDA — IE00B4L5Y983 |
| 10% | US small cap value | SPDR MSCI USA Small Cap Value Weighted UCITS ETF — ZPRV — IE00BSPLC413 |
| 10% | Europe small cap value | SPDR MSCI Europe Small Cap Value Weighted UCITS ETF — ZPRX — IE00BSPLC298 |
| 20% | Long-duration euro governments | Amundi Euro Government Bond 15+Y UCITS ETF Acc — LU1287023268 |
| 20% | Eurozone governments 1–3 years | Xtrackers II Eurozone Government Bond 1–3 UCITS ETF 1C — LU0290356871 |
| 20% | Physical gold | Invesco Physical Gold ETC — IE00B579F325 |
This is not the official European version of the Golden Butterfly, and it is not Tyler’s portfolio. It is a Rebalix adaptation: it keeps the 20/20/20/20/20 structure and tries to translate its functions for a euro investor.
1. The equity core: why MSCI World
For the first 20% we choose iShares Core MSCI World UCITS ETF (SWDA), tracking the MSCI World, an index of large and mid cap stocks in developed markets. The choice doesn’t come from SWDA beating an All-World index in some backtest. The reason comes before the results: the original Golden Butterfly separates a broad equity core from a second small cap value component. Adding emerging markets to the core would introduce a further strategic exposure that isn’t needed to preserve this structure.
Moreover, MSCI World belongs to the same developed universe in which we place the two small cap value components. The price of this choice is explicit: the Rebalix adaptation holds no emerging markets in its equity core.
2. Small cap value: one global ETF or two regional exposures?
This is the least trivial part of the adaptation. A global small cap value UCITS ETF exists today — Avantis Global Small Cap Value UCITS ETF — but it is an active strategy with a very short history. For this experiment we preferred an indexed construction with two State Street ETFs: 10% ZPRV (MSCI USA Small Cap Value Weighted, TER 0.30%) and 10% ZPRX (MSCI Europe Small Cap Value Weighted, TER 0.30%).
A note on the term «Value Weighted». ZPRV and ZPRX don’t simply select a subset of small caps classified as value. The MSCI Small Cap Value Weighted indexes start from the small cap universe and reweight it based on fundamental characteristics associated with value. It is therefore a different implementation of the value tilt from some US indexes that explicitly select a «value» segment.
The 50/50 split between the US and Europe is a choice of the Rebalix adaptation, not a replication of a global small cap value index. Among other things, it excludes the small cap value of Japan, Canada and the developed Pacific. It is therefore geographically less complete than a true global exposure. In exchange, it uses two UCITS indexed strategies launched in 2015 and makes the weight given to the two regions explicit. A perfect example of the series’ principle: adapting means choosing. And every choice also creates something that’s missing.
3. The bond barbell: short and long, not an average duration
The Golden Butterfly doesn’t simply hold 40% bonds. It splits them into two extremes: 20% short-term and 20% long-term Treasuries. That barbell shape is a structural feature of the portfolio. Replacing it with a single intermediate-duration bond ETF would be simpler, but it would change the recipe. In the Rebalix adaptation we therefore keep the barbell, but move the government risk from the US Treasury curve to the euro government curve.
The long leg: why 15+ years. For the long side we choose Amundi Euro Government Bond 15+Y UCITS ETF Acc. The goal is not to find the bond ETF with the best historical performance, but a duration high enough to play a role comparable to the original portfolio’s long leg. In the Rebalix dataset the current declared modified duration for this ETF is about 16 years: a snapshot, not an immutable property — duration changes over time with composition, yields and residual life.
The short leg: why 1–3 years. For the short side we choose Xtrackers II Eurozone Government Bond 1–3 UCITS ETF 1C. The 1–3 year bucket keeps rate sensitivity low and creates a sharp contrast with the 15+ block. The point is not that short bonds are risk-free: part one already showed how misleading it is to equate bonds with safety. Here they mainly serve to preserve the two-extremes duration structure.
Why euro governments and not US Treasuries? It is probably the choice that changes the original Golden Butterfly the most. Keeping US Treasuries would have preserved both the American rate risk and the dollar exposure. By choosing euro governments we instead reduce the currency mismatch between the bond sleeve and the currency in which the European investor pays most of their expenses.
But we won’t present this choice as a certain improvement. In some crises the dollar strengthened and helped the euro investor holding unhedged US assets. In a future crisis it may happen again, or not. Translating the function also means accepting that the future behaviour will not be that of the American portfolio.
4. Gold: here translation is barely needed
The fifth block is the easiest to interpret. Gold is not a US domestic exposure: Portfolio Charts already defines it as Global Gold. We choose Invesco Physical Gold ETC, physically backed by gold. Technically it is an ETC, not a UCITS ETF: the distinction is not cosmetic and we will always keep it.
What we kept — and what we changed
| Element | US original | Rebalix adaptation | Preserved? |
|---|---|---|---|
| Weight structure | 20/20/20/20/20 | 20/20/20/20/20 | Yes |
| Equity core | US large cap blend | Global developed | Function, not geography |
| Small cap value | US | 50% US / 50% Europe | Factor yes; geography no |
| Long bonds | US Treasuries | Euro governments 15+ | Duration/barbell yes; currency no |
| Short bonds | US Treasuries | Eurozone governments 1–3 | Short end yes; currency no |
| Gold | Global Gold | Physical gold via ETC | Substantially yes |
| Reference currency | USD | EUR | No |
This table matters more than the ETFs’ tickers. It says exactly where the experiment stays faithful to the original structure and where it takes new decisions.
Before the results: we freeze the analysis period too
Selecting the instruments is not enough. To prevent the historical period from being chosen after seeing the results, we now fix a second rule: the investable backtest starts only when all the selected ISINs actually exist.
The decisive case is Amundi Euro Government Bond 15+Y UCITS ETF Acc. The Amundi feed in the Rebalix database contains NAVs going back to 2 October 2012, while the inception of ISIN LU1287023268 is 21 July 2016. For the main backtest we will not use those pre-inception NAVs to artificially backdate the instrument’s birth.
The price of this choice is a shorter rear-view mirror. The advantage is knowing exactly what we are looking at. The period includes, among other things, the 2020 pandemic and the 2022 equity-and-bond shock, but not 2008, the dot-com bubble or the inflation shocks of the Seventies.
Rebalancing: the rule is frozen now
The Golden Butterfly is brought back to its target weights once a year. We keep the same logic in the Rebalix adaptation: annual rebalancing to the 20/20/20/20/20 weights, with the small cap value block kept at 20% and split 10% ZPRV / 10% ZPRX. Rebalancing happens on the last available trading day of the year, using the valuations available on that date. No rebalancing bands, no tactical interventions.
This rule too is fixed before observing the results: we will not change frequency or date to retrospectively improve the portfolio’s behaviour.
One currency of analysis: the euro
All Rebalix metrics in the series are computed in euros. When a series is denominated in a currency other than the euro, every observation is converted using the ECB reference rate of the same date. On days without a new ECB fixing, the last available previous fixing is used. The methodology is frozen as ECB_REFERENCE_RATE_V1.
Converting the series into euros is not a currency hedge and doesn’t change the instrument’s economic exposure. The exchange rate serves to measure the result from the viewpoint of an investor who accounts in euros; an unhedged global equity ETF remains exposed to the currencies of its underlying holdings. For the real CAGR we use the HICP EA20 (2015=100): in the run frozen on 4 September 2026 the latest available month was July 2026, with a July 2016 → July 2026 deflator of 1.3255.
Nominal and real are not the same measure
Portfolio Charts presents many of its historical metrics in real terms, i.e. net of inflation, and its simulations don’t embed the costs of investable ETFs in the same way. Rebalix NAVs embed each instrument’s TER; transaction costs are not simulated and are declared separately. Currency, period, inflation and cost treatment must therefore stay visible when comparing different levels of analysis.
In the main comparison we use the same period, currency and calendar for both portfolios. The frozen benchmark is a plain 60/40: 60% SWDA and 40% Xtrackers Global Government Bond UCITS ETF 1C EUR Hedged (LU0378818131), rebalanced annually with the same convention as the Golden Butterfly. It is not the portfolio to beat: it is a control, to make the trade-off delivered by the Golden Butterfly readable.
| Level | Question |
|---|---|
| Original Golden Butterfly / asset classes | How did the portfolio’s logic work historically in its own context? |
| Rebalix adaptation / investable instruments | How did the European instruments we chose ex ante behave? |
| Common benchmark | What trade-off emerges against a simple reference over the same period and currency? |
Now we can open the backtest
Only at this point do we look at the results. Instruments, weights, period, currency, inflation, benchmark and rebalancing had already been frozen. Run v1 was closed on 4 September 2026: any different method will have to become a later, documented run — never a silent rewrite of the past.
In run v1 the observable common period went from 21 July 2016 to 1 September 2026. The end date is the common minimum of the available series — in that run it was set by the latest available NAV of the Invesco Physical Gold ETC — and from here on it keeps advancing on its own: the table and charts below are recomputed every day under the same frozen rules, and always declare up to when they are measuring. The calendar uses only dates present for all instruments, with no forward-filling of NAVs.
Golden Butterfly and 60/40: almost the same final money, not the same journey
In run v1, starting from €10,000, the Rebalix Golden Butterfly reached €19,439; the 60/40 benchmark €19,958. Nominal CAGR: +6.79% versus +7.07%. In real terms, using the HICP EA20: +3.86% versus +4.13%. The return difference is therefore small. More visible is the difference in how the journey was made: annualised volatility of 8.25% for the Golden Butterfly against 9.53% for the 60/40; maximum observed drawdown −18.80% against −20.29%.
2022: when stocks and bonds fall together
2022 is the most instructive episode of the observed period. The Golden Butterfly loses 11.49%, against −13.91% for the 60/40. But the aggregate number hides what makes the portfolio interesting: the components did not all play the role a textbook description might have assigned them.
The long leg in euro governments loses 35.92% in 2022 alone. The global EUR-hedged government fund used in the benchmark also closes at −15.34%. The euro short end limits the loss to −4.69%. Gold is the only one of the big defensive components to close the year in positive territory: +5.61%.
The result doesn’t prove that gold «always protects», nor that long duration «doesn’t work». It shows something more useful: an inflation shock accompanied by sharp rate rises can hit stocks and long-dated bonds at the same time. In the period we can observe, the Golden Butterfly went through exactly that regime.
COVID tells an even less linear story
2020 also warns against overly simple labels. Over the full year the long leg closes at +13.03% and gold at +13.47%. But in the acute phase of the crash, from 21 February to 18 March 2020, the Amundi 15+ loses about 8.4% and gold about 10.0%; in that window it is mostly the short leg that holds, at around −1.1%.
The same component can therefore look defensive over the full year and fail to be so during the most violent weeks of the crisis. Roles change not only across economic regimes, but even across phases of the same episode.
The worst component stays in the portfolio
The harshest test of the frozen rule comes from very long duration euro governments. In run v1, from 21 July 2016 to 1 September 2026, €10,000 invested in the Amundi 15+ component became €7,253. Nominal CAGR: −3.13%; maximum observed drawdown: about −46.0%, from December 2020 to October 2023, and at the close of the run it had not yet been recovered.
It is exactly the kind of result that invites you to rewrite the portfolio after seeing the past. We could replace that component today with something we know did better. We won’t.
Who did what
One last snapshot reminds us why there is no permanent hero. Gold, the most visible diversifier of 2022, recorded in 2026 its deepest drawdown of the whole sample: about −23.4% between March and July, not yet recovered at the close of run v1. The component that helped yesterday can be the one that weighs today.
What did the Golden Butterfly actually buy?
In the observed period, the European Golden Butterfly did not produce anti-crisis magic. It produced something much more concrete: a return very close to the 60/40’s, with slightly smaller swings and with different components absorbing different shocks — but none of them protecting in every regime.
The observed trade-off was, in short, a bit less return in exchange for a bit less volatility and slightly shallower maximum losses. That’s not enough to declare the Golden Butterfly «better» than the 60/40, and the slightly lower return is not enough to declare it «worse».
Above all, ten years don’t contain all the regimes the structure was born to weather. The sample includes COVID and the 2022 inflation shock; it doesn’t include 2008, the dot-com bubble or the Seventies. The long leg suffered precisely in the regime most hostile to it. We don’t know which will come next.
That is why this page doesn’t really end here. The rules stay fixed, while the data will keep updating. In a year, five or ten, we will be able to observe what happened without having rewritten the experiment in the meantime.
The data goes on. The rules don’t.
Next instalment: the Permanent Portfolio
The Golden Butterfly was born by modifying an even more radical structure: Harry Browne’s Permanent Portfolio. Four components, four 25% weights and a precise idea: build a portfolio able to cross prosperity, recession, inflation and deflation without having to predict which regime comes first.
In the next instalment we will do the same operation: start from the function of the four components, build the European adaptation and freeze the choices before looking at the results.