Leverage Lab

How it is calculated

One dataset, one cost calibration and the same engine for the simulator, the asset pages and the study.

Study based on data up to October 2026. The simulator and the asset pages are updated on the 1st of every month.

Data

A universe of 28 indices and 25 sectors, each with its daily total return in its own currency and its local short-term interest rate. Total return is preferred everywhere; where an index has no official total-return series, dividends are added from the ETF that tracks it, and the page marks the stretch where dividends are approximated. The Nasdaq-100 starts in October 1985 and the S&P 500 in 1928.

Simulating a leveraged ETF

Daily simple return with a daily leverage reset: r_L = L·x − (L − 1)·(m·rf + s)/252 − ER/252, where x is the index daily total return, rf the previous day’s annual short rate, m and s the multiplier and spread of the swap financing cost, and ER the expense ratio.

  • One calibration for every asset: s = 0.50% and m = 1.2, fitted on a grid against QLD, TQQQ, SSO, UPRO and SPXL. Over the full period all five stay within 0.3 pp a year.
  • Expense ratio: 0.90% a year for the 2x and 3x model.
  • Validation against 46 real leveraged ETFs worldwide (34 with reliable data): median gap of −0.04 pp a year. Each real ETF gets its own effective cost and currency mode so the simulator can replicate it.
  • The simulated series are not spliced with real ETF prices: before a fund existed, the result is a replica.

Currencies

Results can be shown in USD, EUR, GBP or the index’s own currency. The conversion is that of an investor who buys without hedging: leverage multiplies the index, not the currency. Exchange rates start in 1971, so outside the index currency the series starts no earlier than that. The edge over 1x in percent does not depend on the currency.

Metrics

  • Annual return (CAGR), or IRR when simulating with money and periodic contributions.
  • Max drawdown: the largest fall from a previous high.
  • Worst calendar year.
  • Time underwater: the longest stretch without recovering the previous high.
  • Versus 1x: value of the leveraged product divided by the value of 1x, minus one; with the final edge, the worst point and the time spent behind.
  • Volatility, Sharpe and Sortino (with the short rate of the currency shown), Calmar and Ulcer.

Historical windows and Monte Carlo

Rolling windows of 5 to 20 years with monthly starts. Block bootstrap of historical days and a Markov-switching multifractal model (Calvet and Fisher, k = 8) fitted to the S&P 500 since 1928; the Nasdaq-100 is generated from the S&P with β = 1.19 and a residual scaled by the volatility regime.

Known limits

  • The MSM understates kurtosis (5.8 against 17.3 real) and long memory, and uses a constant risk-free rate.
  • The same financing cost for every asset and currency; real ETFs show biases on the FTSE 100 (dividend) and Taiwan (US rate as a proxy).
  • No taxes or rebalancing costs.
  • Selection bias: choosing the Nasdaq-100 in 1985 is choosing a known winner.