Methodology
How the numbers are made
We'd rather show our work. Here's where the data comes from, how backtests are calculated, and, just as important, what we assume and don't.
Data coverage & source
Market data is provided by EODHD. Prices are end-of-day, updated after each market closes, and US stock prices during market hours are delayed by 15 minutes. Coverage spans about 30,000 stocks and ETFs listed in the US, Canada, the UK, Germany, France, the Netherlands, Belgium, Portugal, Switzerland and Spain, including international listings via their exchange suffixes.
Backtest methodology
This is how every portfolio backtest in Indices Lab is calculated, including the free one.
- Total return. Backtests are computed on a total-return basis from historical prices.
- Common window. A backtest starts on the earliest date for which every holding and benchmark has data, or your chosen start date, whichever is later.
- Rebalancing. Your chosen rule is simulated over time, never, monthly, quarterly, semi-annual, annual, or a drift threshold that rebalances when a weight strays too far.
- Weighting. Equal, market-cap or inverse-volatility, or weights you set by hand. Market-cap weights can use each company's market value today (the weights you'd buy now) or on the backtest's start date (a fair test of how a market-cap strategy would have done). Applied to the past, today's values give the companies that grew the most their full weight from day one, which can make past returns look better. Start-date values are estimated from today's market value and the share price's move since then, so share buybacks and new shares issued in between aren't reflected. Inverse-volatility weights use each stock's volatility over the backtest window.
- Contributions (DCA). Optional regular contributions are simulated, with a money-weighted (IRR) return alongside the strategy's own return.
- Currency. Holdings in other currencies are converted to your chosen base currency using historical rates.
Metric definitions
- CAGR, annualized growth rate over the period.
- Volatility, annualized standard deviation of returns.
- Sharpe / Sortino, return per unit of total / downside risk, above the risk-free rate.
- Calmar, annual return divided by maximum drawdown.
- Max drawdown, the largest peak-to-trough fall.
- Beta / alpha, sensitivity to, and excess return over, the reference benchmark.
- Up / down capture, how much of the benchmark's up and down moves the basket captured.
Stock screener
Every figure uses the same rule for every stock. If the inputs aren't available, you'll see "–" rather than an estimate. "n/m" means the ratio isn't meaningful for that company. Returns and risk are in US dollars with dividends reinvested, so stocks from any market compare like for like.
How has it behaved?
- Return 1Y / 6M: Total return in US dollars, dividends reinvested, from the last close on or before the date 1 year (or 6 months) ago to the latest close. Shown only if the stock has traded for the whole period.
- Return 3Y (a year): The 3-year total return in US dollars, expressed as a yearly rate (compound annual growth).
- Volatility: How much the price swings: the standard deviation of weekly returns (Friday to Friday, US dollars) over the last 3 years, multiplied by √52 to make it yearly. Uses the time since listing if shorter; needs at least 52 weeks.
- Max drawdown: The largest fall from a previous high to a later low, using daily closes over the last 3 years (or since listing; needs at least 1 year).
- 52-week range: Where the latest close sits between the lowest and highest closes of the last 52 weeks, in the stock's own currency, adjusted for splits.
- Trend line: The last 52 weeks of total return in US dollars.
Good business, fair price?
- P/E: Market value ÷ net income to shareholders over the last 12 months. Market value is updated with each closing price. n/m when net income is zero or negative.
- Dividend yield: Dividends paid per share over the last 12 months ÷ the latest share price.
- Revenue growth: Revenue over the last 12 months vs the 12 months before. When the earlier quarters aren't available, the last two financial years are compared.
- Net margin: Net income to shareholders ÷ revenue, over the last 12 months.
- Net debt / EBITDA: (Total debt − cash and short-term investments) ÷ EBITDA over the last 12 months. "Net cash" when cash exceeds debt; n/m when EBITDA is zero or negative; not shown for banks and insurers, where borrowing is part of the business.
- Last 12 months: The four most recent quarters, when they are consecutive and the latest ended within about 6 months. Companies that report half-yearly or yearly use their latest financial year instead (marked FY). Figures more than 15 months old are not shown.
What does it do to my basket?
- Correlation: How closely the stock's weekly returns have moved with your basket's weekly returns at its current weights, over up to 3 years (at least 52 weeks). 1 = in lockstep, 0 = unrelated, below 0 = opposite.
- Badge: Diversifier: correlation below 0.4, it tends to spread your risk. Partial diversifier: 0.4 to 0.7. Reinforces theme: 0.7 or more, it strengthens the theme you're building. Neither is better; it depends on your goal.
- Volatility change: Your basket's volatility if you add the stock as one equal share (1 of N+1, the rest scaled down), minus its volatility now, over the same weeks. Negative means the stock calms the basket.
Prices and financial statements from our market-data provider. Financial statements refresh weekly, prices nightly.
What we assume, and don't
- Main results are before trading costs and taxes, for your index and its benchmarks alike, so ideas compare like for like. ETF prices already include their annual fee.
- Costs & taxes (optional). Add your broker's fee per trade (fixed and/or %), a tax rate on gains and a tax rate on dividends, and a separate run shows what you'd keep, for your index and each benchmark under the same assumptions. Fees apply to every buy and sell, including the first purchase. Tax on gains is charged only when you sell: gains and losses are netted each calendar year using each stock's average purchase price, a net gain is taxed at year end, and a net loss is carried forward. Dividends are cut by the dividend tax rate before being reinvested. Benchmarks are bought once (and with each contribution) and held. "If you sold everything at the end" deducts selling fees and tax on gains not yet taxed. Simplifications: one rate for all gains, no yearly allowance. An estimate, not tax advice.
- Skip tiny trades ignores trades smaller than 1% of the index, or where the fixed fee would exceed 1% of the trade. Rebalance with contributions puts new money into the most underweight stocks first, so less has to be sold.
- How the optimizer searches. It scores thousands of random weightings of your stocks, then refines around the best one, using daily returns over your chosen dates, any per-stock limits you set, and your own rebalancing rule and currency, so its figures match the backtest for the same weights (before costs and tax). When you rebalance on drift, the search uses quarterly rebalancing first, then checks its best candidates under your exact drift rule. Every free account gets one optimizer run, on the portfolio optimizer page or in the app, using the best risk-adjusted goal. On that page it covers the last 3 years with a 50% cap per stock (75% when you pick only two), holding the weights at target every day.
- Optimized weights are fitted to the past (in-sample), which flatters historical results. Trust them most for the protect and spread goals (minimum volatility, drawdown, worst-day losses, risk parity) and least for maximizing return.
- Survivorship and data quality. Results depend on the underlying data; gaps or errors in a provider's history can affect a backtest.
- Everything is hypothetical. A backtest shows what would have happened for a basket you defined, it is not a prediction.
Future simulation: simulated futures
The Future simulation runs 10,000 simulated futures of your index (a Monte Carlo simulation), built from the same total-return prices as the backtest. Every figure below the chart says what it assumes; this section explains the method.
- Building the futures. We take each holding's real monthly returns (US dollars, dividends reinvested) and the benchmark's, over the months they all have in common. Each simulated future is a chain of random 12-month blocks drawn from that history, so a crash and the recovery after it stay together, and every holding and the benchmark move through the same months. The index is run through each future with your weights, your rebalancing rule, and your fees and dividend tax if set. Monthly additions buy at target weights; withdrawals are sold across the holdings after each month's move.
- Stocks with short histories (factor filling). The futures need at least 15 years of shared history to include 2008. When a stock listed later, its months before listing are estimated from how it has moved with its home market (SPY for US stocks, the country ETF for others) and its sector (the US sector SPDR), plus a random month of its own stock-specific noise. No stock-specific edge is carried back. Real months are never changed, and the page names every stock treated this way. Basis: Stambaugh (1997).
- Cautious (the default). Each stock's average return is pulled towards its equilibrium return: the long-run T-bill rate of 3.36% plus its beta times the long-run equity premium of 6.58% a year (S&P 500 and T-bills, 1928 to 2024, Damodaran), the neutral starting point of Black and Litterman (1992). How far it is pulled follows Jorion's (1986) Bayes-Stein weight: short or noisy histories move most. Volatility, crashes and correlations are unchanged.
- Historically. Each stock keeps its own historical average, with no adjustment. The page shows which years were used and what your index averaged over them, and flags an average above 20% a year.
- My assumptions (Portfolio Manager). You type an expected return and volatility per stock. A volatility rescales each month's distance from the average; a return shifts every month so the long-run growth equals your figure. The month-to-month pattern and the correlations stay as they were.
- Drawdowns. Measured at month ends on the index's value per dollar, from its highest value so far, so monthly additions don't hide a fall. A fall that recovers within a month isn't counted.
- Crisis replays. Not simulations: today's basket is run through four real periods on daily prices (dot-com crash from 24 Mar 2000, financial crisis from 9 Oct 2007, Covid from 19 Feb 2020, the 2022 rate shock from 3 Jan 2022). Stocks that didn't exist yet are estimated from market and sector moves without random noise, and the card says so.
- Inflation. On by default, at a fixed 2.5% a year: monthly additions and withdrawals rise with it to keep their buying power, and every amount on the page is shown in today's money. Off, nothing is adjusted and amounts are in future dollars.
- Living off it. Withdrawals are one twelfth of the yearly amount each month. A pot that can't pay a withdrawal counts as run out. The safe yearly amount is the figure 9 futures in 10 can afford for the whole horizon, rounded down to $100. Tax, other income and fees paid elsewhere are not included.
- Not included. Tax on gains, companies failing, and your own future changes to the basket.
- Reproducible. The same inputs always give the same 10,000 futures, and changing the amount, mode or costs reuses the same simulated months, so comparisons are like for like.
Sources: Jorion (1986), Bayes-Stein Estimation for Portfolio Analysis, Journal of Financial and Quantitative Analysis 21(3) · Black and Litterman (1992), Global Portfolio Optimization, Financial Analysts Journal 48(5) · Stambaugh (1997), Analyzing Investments Whose Histories Differ in Length, Journal of Financial Economics 45(3) · Damodaran, Historical Returns on Stocks, Bonds and Bills: 1928–2024, NYU Stern.
Tracking
When you save an index, its starting prices are locked and performance is tracked from that day. Tracking is event-sourced: contributions, rebalances and weight changes are journalled, so editing an index never erases its history.
Indices Lab is a research and educational tool and does not provide investment advice. Past performance does not guarantee future results.
Build your index →