Ask a systematic trader what keeps them up at night and half-life estimation lands near the top of the list. It sounds like a physics problem — because it basically is. You're asking: if this spread drifts away from its average, how long does it typically take to snap back? Get that number wrong and your entire position-sizing framework is built on sand.
The honest answer is that half-life estimation is genuinely hard, especially in Australian fixed income markets where liquidity is thinner, the yield curve has its own personality, and RBA policy shifts can invalidate months of historical calibration overnight. The Ornstein-Uhlenbeck (OU) process is the most rigorous tool most systematic traders reach for — but it comes with sharp edges.
The OU process models a spread as something constantly being pulled back toward a long-run mean, with random noise layered on top. Think of it like a dog on a retractable lead — the dog wanders, but the lead keeps yanking it back. The speed of that yank is your θ parameter, and from θ you derive half-life as ln(2)/θ. In practice, traders estimate θ by regressing the daily change in the spread against yesterday's spread level. The slope coefficient — usually negative — gives you θ directly.
In Australian government bond spreads — think the 3s10s curve or the spread between ACGBs and semi-government paper — typical estimated half-lives range from a few days to several weeks depending on the regime. A half-life under five days suits an active systematic strategy; anything beyond 30 days starts demanding serious capital patience and carry analysis. For deeper reading on the mathematics, the Ornstein-Uhlenbeck process Wikipedia entry is surprisingly thorough, and Investopedia's explainer on mean reversion in financial markets grounds the concept practically. For the statistical underpinning — specifically, why you must confirm cointegration before running any OU calibration — Investopedia's coverage of cointegration is worth bookmarking.
Today's practical takeaway: fit your OU regression on a rolling 90-day window, not a fixed historical sample. Plot the θ estimate over time — if it's swinging wildly, your spread isn't stationary enough to trade systematically.
Half-life tells you how long to hold the trade; ignoring it tells you how long to hold the loss.
This content is for educational purposes only and does not constitute financial product advice. Past performance is not indicative of future results. Profit Logic Ltd (ACN 688 669 936) accepts no responsibility for errors or omissions in this content or anywhere on this website. Always seek advice from a licensed financial adviser before making investment decisions.