The goal is to treat markets not as static equilibria, but as learning systems with finite-speed information processing.
Very roughly, RIMC tries to do three things:
1. Model the market as a learning process that observes an underlying value process V(t), generated by technological recursion or factor models, with delay and noise. 2. Define the coupled dynamics between technological recursion R(t) and economic value V(t) (the “RV equations”) as a general system of differential equations. This is not limited to “tech-driven” stories — Fama–French 5 and other factor structures can be embedded as special cases of the RV system. 3. Reinterpret CAPM α not as unexplained regression residue, but as a structural drift term arising from observation delay and finite-time learning dynamics (“α-drift”).
Conceptually, the framework has three layers:
- Generative layer A value-generation engine where recursion / factors drive V(t) via an RV-type system.
- Observational layer A continuous-time CAPM-like structure where the market only sees a delayed, noisy projection of that value.
- Alpha-drift layer A structural α term α_drift(t) built as an exponentially weighted memory of the gap ε_R(t) = r_real(t) − r_market(t) over a finite window T with forgetting rate λ.
This is a working hypothesis about how structural α can emerge purely from finite-speed learning and observation delay, rather than a claim of a new “better CAPM”.
Repo (manuscript + notes, English; some Japanese commentary as well):
<GitHub repository URL> https://github.com/rimc-lab/RIMC
I would really appreciate any kind of feedback:
- Pointers to prior work I’m implicitly rediscovering - Objections to the way α is treated as a structural drift - Thoughts on whether this is a useful lens for practical quant research (factor models, RL-based strategies, macro regimes, etc.)
Even “this is obviously wrong because X” is very helpful.
Thanks for reading.