Reference 0016
Factor Exposure Recipe Card
Caps bound names; walls bound themes. Exposure is e = Fᵀw — a weighted sum no per-name cap can see. One wall per factor; one dual per wall; one tax per name.
The Four Wall Forms
tilt cap: βᵀw ≤ L one tilted wall
neutrality: βᵀw = 0 equality: wall + floor welded
band: |βᵀw| ≤ τ -τ ≤ βᵀw ≤ τ
(Ref 0015 pattern 1: NO auxiliaries)
many factors: lower ≤ Fᵀw ≤ upper F is n × k; one wall per column
benchmark-relative: constrain Fᵀ(w - w_bench), not Fᵀw
The Dual Is a Factor Tax
wall binds with rent λ ⇒ fund name i while α_i > λβ_i
rank by after-toll alpha: α_i - λβ_i
one dual, n taxes: each name pays λ × (its loading)
zero-beta alpha rides free; high-beta alpha must clear a higher bar
two hats (Ref 0013): λ = tax rate that replaces the wall
λ = alpha gained per unit of exposure headroom
slack wall ⇒ λ = 0 (no rent on a wall nobody touches)
at the dual: the marginal names TIE after tax (solver indifference)
CVXPY Snippets
e = F.T @ w # exposures, one per factor constraints = [e <= upper, e >= lower] # bands constraints += [beta @ w == 0] # beta-neutral equality prob.solve() lam = constraints[0].dual_value # rent per unit exposure, per factor taxed_alpha = alpha - F @ lam # what the optimizer really ranked
Whiteboard Traps
tightening weight caps to control a theme: blind (caps don't see loadings)
and blunt (rations low-beta alpha too)
|βᵀw| ≤ τ with auxiliaries: legal but wasteful — two walls suffice
reading raw exposure on a benchmark fund: the bet is Fᵀ(w - w_bench)
equality walls: dual is sign-free (push from either side)
Use this page with Lesson 0016 (the hidden tilt, the toll-hunting lab), Reference 0014 (toll replaces wall), Reference 0013 (two hats), and Reference 0015 (the band rewrite). Sources: MOSEK Portfolio Cookbook (factor models); Boyd & Vandenberghe §4.2, §5.6; Boyd et al. multi-period trading paper (holding constraints).