testfolio

What it is

testfolio — the site is testfol.io, a domain hack for the same word — takes tickers, weights, a rebalancing rule and a cashflow schedule and returns CAGR, drawdowns, rolling returns and withdrawal statistics. Two choices separate it from the rest of this category.

The first is daily data. Portfolio Visualizer computes from month-end balances and Portfolio Charts from annual ones; testfolio runs daily, which is why it can model a leveraged fund's daily reset at all.

The second is the preset series: of 121 with a dated coverage range, 105 stitch real fund returns onto a modelled segment covering the years before the fund existed — which is what lets a Nasdaq-100 backtest start in 1986 and a US market backtest in 1885. The rest are cash, rate, inflation and index series, beside 39 undated research factor datasets.

Around that sits Monte Carlo, a safe-withdrawal calculator, an optimiser and efficient frontier, factor regression against Fama-French and AQR factors, PCA, tactical allocation with signals and a set of leveraged-ETF tools. It is run by testfolio LLC, whose terms put governing law and venue in Illinois; no individual is named anywhere on the site.

Pricing

Four tiers, and the free one is the story. It gives full available history — no ten-year truncation — along with the backtester, asset analyzer, tactical allocation, factor regression and PCA. It allows 5 portfolios of 20 tickers, caps Monte Carlo at 15 years and 500 runs, factor regression at 6 factors and PCA at 3 tickers, and grants zero saved runs, zero uploads and no downloads. All of that needs an account: signed out, factor regression and PCA are unavailable.

Pro is 15 dollars a month, Pro+ 30 and Max 100. The pricing page opens on the annual toggle, where those read 12.50, 25 and 80 — two months free. Pro buys downloads, uploads, saved runs and a 70-year Monte Carlo; Pro+ raises every numeric limit to its ceiling; Max adds PDF reports, the Tactical Strategy Optimizer and MCP access. The trial is 24 hours. The site labels current Pro and Pro+ rates "early member" pricing and says they will rise for new members.

Data & coverage

US series to 1885, built from Schwert's historical stock data, S&P 500 index data and Shiller dividends and CPI, with Ken French's library, AQR, MSCI, LBMA gold and FirstRate Data behind others. Beyond US listings, thirteen exchange suffixes cover Canada, the UK, Germany, France, the Netherlands, Italy, Spain, Switzerland, Australia and Japan, with results convertible into seven reporting currencies — USD, CAD, EUR, GBP, JPY, AUD and CHF — and inflation-adjusted against each one's own CPI.

The preset series are the thing to understand before trusting an output. Each carries a stated source and a stated assumption: REITSIM is the Fama-French real estate portfolio less 0.13% a year, then STMDX and DFREX, before VNQ; BNDSIM is a VBMFX backfill less 0.12%; TLTSIM is a synthetic bond series built from 20- and 30-year Treasury yields. The leveraged proxies are modelled outright — UPROSIM is 3x daily SPYSIM with an assumed financing rate, 1.115x swap exposure, a 0.42% spread and a 0.95% expense ratio. The vendor says so itself, and the size of the difference is easy to see: requested on 14 September 2026, TQQQ over its real life from February 2010 returned 42.3% CAGR, while TQQQSIM back to 1986 returned 13.3% and a maximum drawdown of −99.98%, peaking in March 2000 and never recovering. That near-total loss is a property of the model, not an observed outcome.

Integrations

Thin by design. Allocations import from CSV, results and charts download as CSV on Pro and above, and Max produces PDF reports. There is no public API — the app calls its own undocumented JSON endpoints, which carry no support or versioning commitment, so anything built on them can break without notice. The one supported machine interface is the MCP server sold with Max, connecting to Codex and Claude Code today with ChatGPT and Claude listed as coming; the vendor is explicit that testfolio runs the analysis and the assistant only reads the results back.

Limitations

• Pre-inception history is modelled, not observed. Every SIM ticker mixes two different kinds of claim in one line on the chart, and nothing in the result marks where the join is.
• The leveraged proxies are the sharpest version of that. Financing spread and swap exposure are assumptions, and over a 40-year modelled window they decide the answer.
• No screener, no tracker, no live prices. It has no idea what you actually hold, and daily bars land a few days back — a 14 September run ended 11 September.
• The free tier withholds persistence, not analysis. Nothing saves, uploads or downloads, so every session starts from scratch or from a shared URL.
• Anonymity cuts both ways. testfolio LLC publishes no named principal, no methodology paper and no data licence terms; the series definitions are one-line summaries, not a reproducible specification, and there is nobody to escalate to.
• Not a registered adviser, by its own terms, and everything it prints is in-sample.

Alternatives

Portfolio Visualizer keeps Black-Litterman and rolling optimisation, which testfolio does not offer, but charges 360 dollars a year and truncates free ticker backtests to ten years — see what to use instead. Portfolio Charts is free and annual, models asset classes rather than tickers, and covers twelve home currencies. For the withdrawal question specifically, FI Calc and cFIREsim go deeper. If the point is a strategy's entry and exit rules rather than a static mix, backtesting frameworks is the right shelf.