Extracting real, repeatable edge from the market

02 / Strategy

From over a hundred data streams to a single position

1. 01Raw data streams

   x_i(t), i = 1, …, N

   Over a hundred raw data streams enter the system in parallel.

2. 02Layered computation

   h^(ℓ+1) = 𝒯_ℓ(h^(ℓ), h^(<ℓ))

   Each layer performs one computation; with each layer the data gets smaller and the information denser.

3. 03Decision data

   Layer by layer, the results aggregate into decision data, which then converge to a single point.

4. 04Signals

   S_j = Φ_j(y_t)

   Decision data generates the signal, expressed as ten input dimensions.

5. 0511-dimensional space and regimes

   (S₁, …, S₉, r) ↦ q

   Ten dimensions plus the position make an 11-dimensional space. The market is first classified into one of several regimes, each with its own parameter set; within a regime, the position is set jointly by signal strength, the adaptive dimension and multiple constraints.

   Dimensions used by each regime

   	I	II	III	IV	V	VI	VII	VIII	IX
   Regime I	●	●	○			■		■
   Regime II			○	●	●	■
   Regime III						■	■		■

   ● Signal strength○ Adaptive■ Conditions and constraints

6. 06Positioning logic

   b_k = β_k(S; r_t)

   Grouped by regime, the ten dimensions feed five positioning rules, then pass through separate long and short constraints.

7. 07Optimal position

   q = π(b; Θ)

   Through the 11-dimensional algorithm and regime switching, one decision remains: how large the position should be.