Research · Beta · v0.1 · 4 factors

Signal Library

Cross-asset Carry / Value / Momentum / Defensive — 4 classical factors × 10 assets. Each factor documented with formula, asset-specific definition, reliability, and known crash modes.

Method

跨资产 carry / value / momentum / defensive 因子库, 借鉴 AQR (Asness et al.) + Newfound.
每个因子先在自己的资产空间算原始值, 然后用 z-score 跨资产标准化.
4 因子等权合成 → 战术 tilt (输入 view 形成).

Carry

— Roll/yield over funding ★★★ 在 rates / credit / FX 上证据最强 (Koijen et al. 2018)

Formula

Carry = (asset_yield - funding_rate) / vol
或: 期货空间 = (front - next_contract) / vol

Asset-specific

equity: Dividend yield + buyback yield − risk-free
rates: Yield-to-maturity − short-end rate (slope)
credit: OAS − default expectation
commodities: Roll yield (front-back contango/backwardation)
cash: Short rate vs reference

Update cadence

monthly

Crash mode

Carry crash: 高 carry 资产同时垮 (2008, 2020)

Value

— Price vs intrinsic anchor ★★ 长期信号好, 月度噪音大. 不要用于 < 6m horizon

Formula

Value = (anchor − current_price) / anchor
anchor 用各资产合适的 fundamental

Asset-specific

equity: CAPE 偏离 10y mean → 高 value score = 低估
rates: Real yield vs trend
credit: Spread vs default-adjusted fair
commodities: Price vs 10y inflation-adjusted mean
cash: n/a

Update cadence

monthly

Crash mode

Value trap: 低估的不一定均值回归 (e.g. 港股 2015-2024)

Momentum

— Past performance autocorrelation ★★★ Cross-asset momentum 是 trend-following 行业基石 (Moskowitz et al. 2012)

Formula

Mom = (price_t / price_{t-12m}) - (price_t / price_{t-1m})
(12m 减去最近 1m, 标准做法去 mean-reversion)

Asset-specific

all: 同上公式, 跨资产标准化

Update cadence

monthly

Crash mode

Momentum crash: 反转时极差 (2009-03, 2022-04)

Defensive

— Low-vol / low-beta preference ★★ Frazzini-Pedersen Betting Against Beta. 长期 Sharpe 高但 Cumulative 低

Formula

Defensive = -1 × beta_to_global_equity
高 defensive score = 低 beta = risk-off 偏好

Asset-specific

equity: Low-vol stocks vs market
rates: Long duration vs short
credit: Investment grade vs HY
commodities: Gold > industrial metals
cash: Always +1

Update cadence

monthly

Crash mode

Bull market 时 cumulative drag, 测试期需要长

Current scores

v0.1 framework only — 实际 z-score 待 scripts/compute_signals.py 上线

上线计划:
- v0.1 (现在): 框架文档 + 公式定义
- v0.5 (2026-09): 4 因子 × 10 资产 = 40 格当月得分表
- v1.0 (2026-12): 累计 12 个月历史 + 因子合成 → 战术 tilt 上 portfolio 页

Composite rule (v0.5)

tactical_tilt_i = w_carry * carry_i + w_value * value_i + w_mom * mom_i + w_def * defensive_i
初始: w_* = 0.25 (等权), 后续按因子 IC 自适应

Caveats

经典因子在 cross-asset 上效力不如 cross-sectional equity (Asness 2013)
v0.1 没真值, 不要把 framework 文档当结论
Defensive 因子需要 ≥ 60 月历史才有意义
Carry / Momentum 月度 IC 在 0.05-0.10 之间, signal-to-noise 不高