Manual traders stare at price action and try to mentally filter signal from noise. The problem is the human brain is spectacularly bad at this — we see patterns in randomness, ignore drift, and anchor to the last number we looked at. Algorithms don't have that luxury, or that weakness. They need a principled way to estimate where price actually is.

The Kalman filter is a recursive algorithm that does exactly that. Originally developed for aerospace navigation — yes, the same maths that guided Apollo — it produces a running estimate of a hidden state by blending a model prediction with a noisy observation. In price terms, that hidden state is the "true" underlying value the market is noisily sampling around tick by tick.

CONCEPTThe Kalman filter updates its price estimate every tick, weighting model confidence against observed noise — dynamically, not with a fixed lookback.
WARNINGA Kalman filter tuned on historical volatility can lag badly during regime shifts — your noise model becomes wrong exactly when it matters most.
KEY IDEAThe Kalman gain is the filter's self-adjusting dial: high gain trusts observations, low gain trusts the model — and it recalibrates every step.

What separates the Kalman filter from a simple moving average is adaptivity. A 20-period EMA treats every bar equally within its window. The Kalman filter maintains two variance estimates — process noise and measurement noise — and uses their ratio to compute a Kalman gain. That gain determines how much weight this tick's price gets versus the prior estimate. It's a moving average that audits its own confidence.

Raw Price vs Kalman EstimateT=0T=17PriceRawKalman

In practice, traders use the Kalman filter as an input to signal generation — the smoothed estimate drives a mean-reversion or trend-following rule rather than trading raw price. The classic challenge is parameterising the noise covariances without overfitting them to a specific historical period. Backtest performance on a well-tuned filter can look immaculate; live performance reminds you that markets didn't read your optimisation report. Robust implementations often treat the noise ratio as a slow-moving adaptive parameter rather than a constant. For foundational reading, the Wikipedia entry on the Kalman filter covers the full mathematical derivation, while Investopedia's explanation of signal contextualises why cleaner price estimates matter for any systematic strategy, and their moving average breakdown makes the contrast with fixed-lookback smoothers concrete.

The Kalman filter won't save a bad strategy — but it gives a rigorous, mathematically grounded answer to the question every algo must answer first: what is price, really, right now?

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.