kaisaurus/etf-synth

ETF-Synth builds reconstructed long-run monthly histories for popular Australian ETFs, such as DHHF, BGBL, NDQ, GHHF, GGBL, and GNDQ, that are actually too young to have existed through past major market events. The project stitches together data going back to January 1995 so you can see how these funds might have behaved through the dot-com crash, the global financial crisis, and COVID, periods the real funds never lived through. The synthetic histories are built entirely from free public data sources, pulling prices from Yahoo Finance and exchange rate and interest rate data from FRED. For each fund, its underlying holdings are stood in with older, low cost index funds that have a long history, using total return figures that assume dividends are reinvested. Where a fund holds overseas assets and is unhedged, the project converts those returns into Australian dollars, then layers on fees and, for the geared funds, an estimate of borrowing costs. Every synthetic series is checked against the real fund's actual performance during the period the fund has existed, and the README reports a monthly return correlation between 0.92 and 0.99 depending on the fund. The README is direct about the limits of this approach. It is reconstructed, not real, historical data, the years before 2001 rely more heavily on approximate substitutes, and the geared funds along with one fund that includes a bond component are the most sensitive to the assumptions used. Franking credits, a feature of the Australian tax system, are not included in the total return figures. The author states clearly that this is meant for research and curiosity rather than financial advice. The bulk of the README is a set of result tables showing total return, compound annual growth rate, worst drawdown, and recovery time for each fund across the full 1995 to 2026 period and again for the period after 2012 only, plus a year by year breakdown of returns and drawdowns for every fund covered. These tables illustrate how differently the funds performed depending on which historical stretch is examined, without the README telling the reader which stretch is more likely to repeat.