*Cross-sectional equity signals for WorldQuant BRAIN, and an automated harness that runs thousands of backtests a day.*
**Summary:** As a research consultant at WorldQuant BRAIN I design systematic equity signals and test them with a cloud harness that runs 2,000 to 4,000 backtests a day. More than 7,700 signals tested so far.
**Status:** Ongoing. **Year:** Since July 2026. **Role:** Research consultant, WorldQuant BRAIN. **Tags:** Alpha research, Backtesting, Python, Cloud automation
---
```math {w_{i,t} = \frac{\tilde s_{i,t} - \langle \tilde s \rangle_{S(i),t}}{\sum_j \big|\tilde s_{j,t} - \langle \tilde s \rangle_{S(j),t}\big|}},\qquad{\tilde s_{i,t} = \frac{s_{i,t} - \bar s_t}{\sigma_t}} ``` *Score, standardize across stocks, subtract the sector mean, scale to unit gross exposure.*
```math {\mathrm{PnL}_t = \sum_i w_{i,t-1}\, r_{i,t}},\qquad{\mathrm{SR} = \sqrt{252}\;\frac{\mathbb E[\mathrm{PnL}_t]}{\sigma(\mathrm{PnL}_t)}},\qquad{\mathrm{TO} = \Big\langle \sum_i |w_{i,t} - w_{i,t-1}| \Big\rangle_t} ``` *Daily P&L of yesterday’s weights, its annualized Sharpe ratio, and turnover: how much of the book trades each day.*
```math \max_k\ \mathrm{corr}\big(\mathrm{PnL}^{\text{new}},\ \mathrm{PnL}^{(k)}\big) < \rho^{*} ``` *A new signal must also stay below a correlation bar against every signal already in the pool, so it adds something new.*
```math \mathbb E\Big[\max_{k \le N} \widehat{\mathrm{SR}}_k\Big] \approx \sigma_{\widehat{\mathrm{SR}}}\,\sqrt{2 \ln N} ``` *Background: for N variants with no real edge, the expected best backtest Sharpe still grows like \sqrt{2\ln N}. That is why ideas are judged against their whole family of variants, not their luckiest run.*