Kalman Filtering for Noisy Market Data and Pairs Trading
Summary
The article introduces the Kalman filter as a recursive method for estimating a changing, partly unobserved state by combining model predictions with noisy measurements and their uncertainty. It explains concepts including normal distributions, variance, measurement noise, and process noise, then surveys trading applications such as volatility surface estimation, market impact modelling, portfolio covariance estimation, and pairs trading. The article also compares the filter with other filtering methods and describes a Python implementation, although portions of the tutorial are omitted from the supplied text.
Examples cited include a reported use of an extended Kalman filter for automated FX implied volatility surface marking and research combining Kalman filtering with momentum, autoregressive models, and support vector regression. These references illustrate possible uses, but the excerpt does not provide enough detail to assess the studies’ methods or reproducibility. A Kalman filter’s estimates depend on its model and noise assumptions; the article itself notes strengths in Gaussian noise settings and does not establish that filtering alone creates a profitable strategy. Treat the applications as modelling approaches that require market-specific validation.
Key ideas
- A Kalman filter updates a state estimate by combining new measurements with model-based predictions and uncertainty.
- Measurement noise and process noise represent different sources of estimation error.
- Trading applications discussed include pairs trading, volatility estimation, market impact modelling, and covariance estimation.
- The article cites examples of Kalman filters combined with momentum, autoregressive models, and support vector regression.
- Model assumptions matter, and the cited applications do not by themselves demonstrate reliable trading profits.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.