Golden Butterfly Portfolio

Data: Yahoo Finance adjusted close, fetched live on each visit How this is calculated & what it can't tell you →

What is this portfolio?

The Golden Butterfly is a five-way, equal-weight portfolio designed by Tyler of Portfolio Charts — a refinement of Harry Browne's Permanent Portfolio with a tilt toward historically higher-returning assets. Each of its five 20% sleeves is built to lead in a different economic season, so the whole tends to ride steadier than any single piece. The bet isn't on prediction; it's on balance. Compare it against the other lazy portfolios we've backtested.

Five equal 20% sleeves US total market, US small-cap value, long-term Treasuries, short-term Treasuries, and gold — each at 20%. Equal weighting is the whole point: no single economic outcome gets to dominate the portfolio.

Cover every economic season Stocks carry prosperity, long Treasuries cushion recession and deflation, gold answers inflation shocks, and short Treasuries stay steady as dry powder. Something is usually working when something else isn't.

A small-cap value tilt for extra return The one real change from the Permanent Portfolio: half the equity leans into small, cheap companies (VBR), historically a higher-returning corner of the market. It's what gives the Golden Butterfly a higher ceiling than its more defensive cousin.

Rebalance, then leave it alone The low correlation between sleeves only pays off if you rebalance back to 20/20/20/20/20 on schedule — selling what ran, buying what lagged. Between rebalances, do nothing.

Risk notice: Balance has a cost — with 40% in Treasuries and 20% in gold, the Golden Butterfly gives up upside in roaring bull markets for a smoother ride. Gold can stagnate for a decade-plus (it went nowhere from 1980 to 2000), and long-term Treasuries fell more than 30% in 2022 when rates spiked. A diversified portfolio still loses money in bad years. This is educational — see the disclaimer at the bottom of the page.

The five sleeves

Equal 20% weights across five US-listed ETFs. Everything is priced in USD, so the backtest is a clean single-currency picture — but a Canadian holder also carries CAD/USD currency risk on top (see the note below). MERs are approximate; confirm the current figure on the provider's fund page (linked in Sources).

Sleeve	Ticker	~MER	Role
US total market	VTI	~0.03%	Growth / prosperity
US small-cap value	VBR	~0.07%	Return tilt (size + value)
Long-term Treasuries	TLT	~0.15%	Recession / deflation hedge
Short-term Treasuries	SHY	~0.15%	Stability / dry powder
Gold	GLD	~0.40%	Inflation / crisis hedge

No clean Canadian-listed equivalent exists for the small-cap value sleeve, which is the main reason this tracker uses the US-listed funds rather than a CAD-listed rebuild. The gold sleeve (GLD) pays no distribution; the Treasury sleeves pay interest, and VTI/VBR pay dividends.

Holding US-listed ETFs as a Canadian. These funds trade in USD, so two things matter beyond the chart:

Currency. Your real return in Canadian dollars also moves with CAD/USD, which can swing 10–15% in a year. The backtest is in USD and ignores that entirely.
Withholding tax. US dividends (from VTI and VBR) face a 15% US withholding tax that is exempt in an RRSP under the Canada-US treaty, applies and is unrecoverable in a TFSA, and is recoverable as a foreign tax credit in a non-registered account. Treasury interest and gold (which pays no distribution) are largely outside this.

This backtest uses dividend-adjusted USD prices and models neither FX nor withholding tax. Educational only — not tax advice.

Sources

Golden Butterfly design + allocation. Portfolio Charts: Golden Butterfly (Tyler).
Foreign withholding tax by account type (RRSP treaty exemption, TFSA non-coverage, non-registered foreign tax credit). Vanguard Canada: The impact of withholding taxes on Canadian ETF investors (PDF); BlackRock Canada: Understanding Foreign Withholding Tax (PDF).
ETF MERs. Provider fund pages: Vanguard (VTI, VBR), iShares (TLT, SHY), SPDR (GLD). Confirm the current MER on the fund page before relying on it.

Figures accessed 2026-05-28. Tax rules change; verify against canada.ca and the fund prospectus before acting.

Portfolio configuration

Saved presets

Portfolio assets — up to 10. Leave unused slots empty.

Asset 1

Asset 2

Asset 3

Asset 4

Asset 5

Asset 6

Asset 7

Asset 8

Asset 9

Asset 10

Filled weights must sum to 100%. Current total: 100% (4 assets)

Backtest mechanics

Starting capital ($)

Start date

Rebalance frequency

Rebalance tolerance (%)

Commission + slippage (%)

Risk-free rate (%, annual)

Benchmark 1 (buy & hold)

Benchmark 2 (buy & hold)

Cash flow — contributions & withdrawals

Amount per period ($)

Direction

Frequency

Cash goes into

VTIVanguard US Total Market · VBRVanguard US Small-Cap Value · TLT20+ Year Treasuries · SHY1-3 Year Treasuries · GLDSPDR Gold Shares Data through: —

Return

Risk

Quick reality check before you get excited. This is a learning tool, not financial advice. Every number on this page came straight from the inputs you set above — change the start date or growth target and the whole picture shifts. Play with the dials, see what moves, but don't mistake a backtest for a guarantee.

Heads up for Canadian readers. All returns shown are in USD. Holding US-listed ETFs from Canada adds currency risk (CAD/USD can swing 10–15% in a year), foreign withholding tax on distributions (0% in RRSP, 15% in TFSA on US dividends, treaty in non-registered), and every rebalance is a taxable event in a non-registered account. The numbers above ignore all of that. Real after-tax, after-FX returns in CAD will be lower.

Next rebalance signal

Portfolio vs. benchmarks

Start-date sensitivity — rolling CAGR by start date

Drawdown from peak

Annual returns

Allocation over time

Cumulative contributions vs. portfolio value

What follows are simulations, not predictions. The market doesn't know we ran them — and neither do taxes, fees, or how you'll actually feel when your account is down 50%. Use these to get a sense of the risk, not to plan around a specific outcome.

Monte Carlo — 10-year forward projection (2,000 paths)

What is this telling me?

We ran 2,000 imaginary versions of the next 10 years by replaying real historical daily return patterns in random order. Each "path" is one possible future. Together they show the full range — from lucky outcomes to unlucky ones — so you can see what this portfolio might realistically do going forward.

Green line (median path)

Half of all 2,000 futures ended above this line, half below. Think of it as the single most representative outcome — not a guarantee, just the middle of the range.

Orange bands

The darker band shows where the middle 50% of futures landed. The lighter outer band covers 90% of all simulated futures. A wider band means more uncertainty.

Median CAGR

The typical yearly growth rate across all simulations — like a compound interest rate. A 30% CAGR means your money roughly triples every ~4 years in the median scenario.

Risk of Ruin

The % of futures where your portfolio fell below 50% of what you started with. This is the worst-case catastrophic loss risk — you should be comfortable with this number before investing.

Probability of Loss

The % of futures where you ended 10 years with less money than you started. Even over a long horizon, some paths go negative — this tells you how often that happened.

Mean Max Drawdown

On average across all futures, how far did the portfolio fall from its peak before recovering? For example, -60% means at some point you'd be watching your account drop by more than half. This is the emotional pain number.

⚠️ Important: This is based on historical patterns and does not predict the future. Diversification reduces risk but never eliminates it — in a severe market crash, correlations rise and most assets can fall together. Past performance does not guarantee future results.

Percentile fan chart — portfolio value (block-bootstrap, not assuming normality)

50th pct (median) 25th–75th pct band 5th–95th pct band

Stress tests — portfolio performance during market crises

What is this telling me?

Unlike the Monte Carlo which simulates imaginary futures, stress tests look at what actually happened during real historical market crises. Because this portfolio spreads across assets that don't all move together and rebalances back to target weights, crashes affect it differently than holding any single asset alone.

Return %

How much the total portfolio gained or lost from the start to the end of that crisis window. A -39% return means if you had $100,000 at the start, you'd have about $61,000 at the end of that period.

Max Drawdown

The worst single drop from a peak during that period — the lowest your account would have looked. This is deeper than the overall return because it captures the scariest moment before any recovery.

Days

How many trading days the period covered. Shorter crashes (like COVID in 2020) recovered faster. Longer downturns (like 2022) dragged on and gave the portfolio less opportunity to average down.

Why some periods are missing

A crisis period can only be shown when every asset in the portfolio has real price data covering it. If one of your chosen ETFs launched after a given crisis (for example, the 2008 Financial Crisis), that period is hidden — there's no real data to backtest it.

💡 Tip: Large negative returns during crashes are expected with this portfolio. The key is whether the portfolio recovers — look at the Post-COVID Recovery row to see how dramatically the portfolio can rebound after a crash.

Return % per crisis period

Stress period detail

Period	Return	Max DD	Days

Date	Action	Pre total	Max drift	Trade $	Post total	Allocation after

Date	Type	Amount	Into	Before	After	Allocation after

Metric	Portfolio

Year	Inflation	Portfolio
Year	Inflation	Return	Balance

How these numbers are calculated

Data source

Prices come from Yahoo Finance via yfinance with auto_adjust=True, so closes are dividend-adjusted (every fund's distribution stream is included). All default tickers are US-listed, so prices and every figure on this page are in USD. A Canadian holder's real return would also move with the CAD/USD exchange rate, which the backtest does not model.

CAGR

(end_value / start_value)^(1/years) − 1, where years = trading_days / 252.

Annualized volatility

std(daily_returns) × √252. Population std (divisor = n). Assumes daily returns are independent and identically distributed; in practice returns show mild autocorrelation and fat tails, so treat the annualized figure as an approximation.

Sharpe ratio

(mean_daily − rfr_daily) / std_daily × √252, where rfr_daily = (1 + rfr_annual)^(1/252) − 1. The risk-free rate is whatever you set in the config above. Default 2% is a rough time-weighted average for 2015–2022 (T-bill yield was near 0% in 2015–2021, ~5% in 2023+). For period-accurate ratios, override to match your chosen backtest window. Also reported with a standard error band: SE(Sharpe) = √((1 + 0.5·Sharpe²) / T_years) (Lo 2002, IID approximation).

Sortino ratio

(mean_daily − rfr_daily) × 252 / downside_dev_annualized, where downside deviation only counts days with returns below rfr_daily, summed-of-squares divided by total n (Sortino & Price 1994 convention).

Maximum drawdown

Largest peak-to-trough percentage loss using daily close prices only. Intraday lows are not captured, so actual lived drawdowns may be a few percent worse.

Monte Carlo

Block-bootstrap resampling of historical daily portfolio returns. Default: 2,000 paths, 10-year horizon, 5-day blocks, RNG seed = 42 on first render (click "Re-run with new seed" to see how stable the bands are). Block-bootstrapping preserves short-term autocorrelation and volatility clustering — it does not assume normal/Gaussian returns. Risk of Ruin = % of paths ending below 50% of the anchor value; Probability of Loss = % ending below the anchor value.

The MC projection anchor is configurable via the toggle above the section. Default: initial capital (matches a fresh investor's mental model: "if I start today with $X, what range of outcomes might I see in 10 years?"). The alternative is anchoring on the backtest's ending value (useful when asking "if I'm already at $X, what's next?").

Limitation: 5-day blocks preserve weekly autocorrelation but not regime/crisis dependencies. Heavier-tailed events than what's in the historical sample won't appear.

Stress tests

Hardcoded historical windows (2018 vol, COVID 2020, post-COVID recovery, 2022 rate hikes, 2022 tech crash, plus GFC 2008). Only windows with ≥5 trading days of overlapping data are shown. With the default tickers the backtest starts in early 2005 (capped by GLD's late-2004 launch), so the 2008 financial crisis window does render — a useful stress test for a portfolio built to weather exactly that kind of event.

Transaction frictions

Commission + slippage is applied symmetrically — the sell side receives price × (1 − f), the buy side pays price × (1 + f). Default 0.10% per side ≈ 0.20% round-trip, realistic for discount brokers on highly-liquid ETFs.

What's NOT modeled

Taxes — every rebalance is a taxable event in a non-registered account
Currency / FX — all returns are USD, ignoring CAD/USD volatility
Foreign withholding tax on US ETF distributions (treatment depends on account type)
Inflation — returns are nominal, not real (subtract ~2–3% per year for real CAGR)
Settlement delays — assumes T+0 cash availability
Bid-ask spread variability — uses closing price, not actual fill price

Source code

All calculations are open and inspectable in the public repository — Python tools (Monte Carlo, stress tests) and JS in this page itself.

The bottom line: this is a learning tool, full stop. Every number depends on the inputs you chose, so different choices give you different results. Even diversified portfolios can lose 20–40% in a bad year, and no backtest changes that. Foreign withholding tax, currency drag, and tax-wrapper choice can all shift the real-world outcome meaningfully from what the chart shows. Before you put real money on the line, sit down with a registered Canadian advisor. That's what they're there for.