← Research logMomentum & Time-Series
KILLED · memoryless
#

Discretize returns into states (e.g. quintiles) and use the empirical transition matrix to forecast the next state’s expected return.

$$ \pi_{t+1} = \pi_t\,P,\qquad P_{ij}=\hat P(s_{t+1}=j\mid s_t=i) $$

Estimate the transition matrix on rolling windows, trade the expected next-state return, costs applied.

Transition matrix vs uniformbarely distinguishable
Forecast edge≈ 0
KILLED
Discretized returns are very nearly memoryless — the estimated transition matrix is almost row-uniform, so next-state forecasts carry no edge. A first-order Markov chain cannot find structure that is not there. Killed.
If the transition matrix is nearly uniform, the process is nearly memoryless. Discretizing a random walk does not create predictability.

We publish the failures too.

This is one of 100+ documented hypotheses. Browse the full lab notebook, or see the strategies that survived into production.