During the first year of my PhD I worked on algorithms to process high frequency financial data (with our aim being to use asynchronous tick-level data). The idea behind the work was that by applying a special form of Bayesian filter, the variable rate particle filter, that could react quickly to jumps in asset prices and trends, we could develop a successful momentum strategy. Transitions from one trend to the next are where momentum strategies based on smoothing really lose out, since they necessarily involve some degree of lag, so the introduction of trend jumps in particular has the potential to reduce some of their downside.
This work led to our paper Forecasting high-frequency futures returns using online Langevin dynamics in the IEEE Journal of Selected Special Topics in Signal Processing (download below).
My first year PhD report, which also contains much relevant background information in this area is available below.
This is still an active area of my research, with my current focus on improving parameter estimation for the models used.
PhD Thesis: Chapters 3 and 4 from my PhD thesis are relevant. It is available here.
Paper: Bayesian Parameter Estimation of Jump-Langevin Systems for Trend following in Finance (ICASSP paper on parameter estimation for such systems)
Report: Bayesian Methods For HF Financial Time Series Analysis (pdf) (PhD first year report)