Active research line
Markets Research
Modeling non-stationary stochastic time series — the actual hard problem behind most market behavior, treated as a methodology question first and an instrument-class question second.
What this work is
The default assumption underlying most quantitative-market analysis is that the process generating returns is at least locally stationary — that recent history is informative about near-future behavior in a regime-independent way. In actual markets that assumption is false more often than it is true, and the failure is not noise: it is the structural fact that the data-generating process changes its character without warning, often without obvious cause, and frequently without leaving the kind of footprint that signals the change is happening.
We work on the methodological question of how to reason rigorously under that condition. The research is about modeling — what classes of model are honest about regime change, what inference looks like when the past is not a reliable training set for the future, how to combine domain structure with statistical learning when neither is sufficient alone.
Why methodology first, instrument second
Most published market work selects an instrument class — equity index, FX pair, commodity, derivative — and asks "what works here." That is downstream work, and it depends on having a methodology that doesn't collapse the first time the regime moves. We invert the order. The methodology question is the load-bearing one; instrument-class application follows once the methodology is honest enough to be applied.
This is not a trading research line. We do not publish signals, run a fund, or claim edge in any specific instrument. The work is about the modeling problem and how to reason about non-stationary processes in a way that survives contact with reality.
Current state
Ongoing. The methodology work is one of the longest-running research threads in the lab and predates its formal framing. Public artifacts will surface as the methodology matures into a form worth defending.
Reading the work
For substantive questions about the modeling approach, what regimes it handles well, or whether it adapts to a specific application — tim@cruxadjacent.com.