Quant Insider.io

We're honoured to support the 4th Women in Quantitative Finance Americas Conference as Official Fintech Partner — a non-profit gathering of the brightest minds in quant, hosted jointly with Columbia University. 📍 Columbia University, New York 📅 Thursday, 21 May 2026  👥 A day of cutting-edge research, panels, and serious networking If you're attending, come find us at the Quant Insider kiosk — we'd love to hear what you're working on. Full details → wbstraining.com/events/wqfa/

The deeper point: convexity in trend-following is partly structural, partly empirical. The IID Gaussian baseline gives you skew ~2.4 at the right horizon, for free, just from the bilinear product structure. Real markets add fat tails on top of this.

The practical consequence: the same strategy reported at three different horizons tells three different stories. • 3-month CTA, monthly returns: skew ≈ 1.7 • Same fund, quarterly returns: skew ≈ 2.4 • Same fund, annual returns: skew ≈ 2.2 All correct. All the same strategy.

First thing to check: is Var(G_T) at least normal? Yes. Cov(R_t, R_u) = 0 for t≠u because future returns are independent of past signals. So Var(G_T) = T exactly. The strategy returns are uncorrelated. Where's the asymmetry hiding then?

Setup is the simplest possible: • X_t ~ N(0,1), independent across time • Signal s_t = Σ a_j · X_{t-j} (causal linear filter) • Strategy return R_t = s_{t-1} · X_t • Horizon return G_T = ΣR_t The X's have zero skew. The filter is linear. Yet G_T is skewed.

Why Momentum Has Positive Skew Even on IID Gaussian Returns ? 🧵 A momentum strategy on IID Gaussian returns, zero skew, no autocorrelation, no fat tails, produces a return distribution with strong positive skewness. The peak exceeds 2. Let me walk through why. @ObjRandom

Linear regression is the most used tool in quantitative alpha research. Here's how quant firms actually use it to build trading signals: 5 ways regression builds alphas: 1. Residual Alpha Extraction Regress stock returns on market factors. The residual is the idiosyncratic return - unexplained by known risks. Cumulate residuals over 12 months and you get "residual momentum," which produces ~2x the alpha of raw momentum because factor reversals are stripped out. 2. Rolling Beta for Risk Signals Estimate CAPM beta over 252 days. Go long low-beta, short high-beta. Works because investors overpay for high-beta "lottery tickets." Blume's shrinkage (2/3 × OLS + 1/3 × 1.0) corrects for mean reversion. Vasicek's Bayesian approach adapts shrinkage to estimation uncertainty - noisy betas shrink more. 3. Cross-Sectional Neutralization Regress your signal on size and sector at each date. The residual is your "pure" signal, orthogonal to known factors. Without this, most alphas are secretly industry bets. Fama-MacBeth formalizes this: cross-sectional regressions at each t, averaged across time. Even strong signals only predict correctly ~52 out of 100 months. 4. Cointegration for Pairs Trading Regress Price_A on Price_B for the hedge ratio. If the residual is stationary (Engle-Granger ADF test), trade it as mean reversion. Static OLS ratios are unstable , Kalman filter versions that model the ratio as a time-varying state produce ~5x cumulative return. Estimate reversion speed via OU process: half-life = -ln(2)/b, where b comes from regressing spread changes on lagged levels. 5. Elastic Net for Feature Selection With 100+ candidate signals, raw OLS overfits catastrophically. LASSO zeros out noise but arbitrarily drops correlated features. Ridge shrinks everything but keeps all. Elastic Net combines both, selecting correlated signal groups together. Gu, Kelly & Xiu (2020) showed penalized regression matches or beats neural networks for return prediction. At finance's signal-to-noise (~0.05:1), aggressive shrinkage is non-negotiable. ——— The practical checklist: → Start with an economic hypothesis (WHY should this predict returns?) → Express it as a regression, regularize aggressively → Walk-forward validation - never standard k-fold on time series → Newey-West HAC standard errors for time-series regressions → Target: Sharpe > 1.5, turnover < 30%, robust across universes → 100 backtests inflates best Sharpe by ~3.0 SDs -adjust accordingly If it doesn't survive these checks, it's noise. Want to build and test regression-based alphas on real market data - for free? The IQC 2026 by WorldQuant BRAIN is the world's largest quant competition, free, open to anyone 18+, $100K prize pool, Global Finals in Singapore. Register: platform.worldquantbrain.com/sign-up/IQC202…

We are live with Aditya Chaturvedi - Regional Strategy & Operations Head at WorldQuant BRAIN, for our International Quant Championship 2026 Webinar The IQC 2026 by WorldQuant BRAIN is the world's largest quant competition, free, open to anyone 18+, $100K prize pool, Global Finals in Singapore. Register now for the competition: platform.worldquantbrain.com/sign-up/IQC202…

2025 was "The Year of the Glut" but paper (virtual) barrels kept oil prices from falling. 2026 is "The Year of the Squeeze" but paper barrels keep prices from rising. See below why financial oil flows are still more balanced than physical. youtube.com/watch?v=IdAMLK…

What do Olympic champions and top quants have in common? More than you think. Discipline. Consistency. Process over outcomes. And the ability to perform when it matters the most. Saina Nehwal (Olympic Medalist) is now an official partner for the International Quant Championship 2026, alongside other world-class Olympians globally. Why does this matter for you as an aspiring quant? Because quant careers are not just about math or coding. They’re about: • Handling uncertainty under pressure • Staying consistent through drawdowns • Competing at a global level • Building an edge over time This is exactly what world-class athletes have mastered. International Quant Championship (IQC) 2026 is designed to bring that same mindset into quant finance. This is not another theoretical competition. This is where you: Think like a quant researcher Build and test real models Compete with top students globally Get noticed by industry professionals What’s on the line? • $100,000+ prize pool • Global Finals in Singapore • Internship & job opportunities • Strong signal for your quant career Reality check: Most students are consuming content. Very few are competing. And in quant finance — Competition is where careers are built. 🎯 Who should apply? • College students interested in: Quant Trading Machine Learning in Finance Derivatives & Options Data Science / Math / CS If you’re even slightly serious about a quant career, this is for you. 🔥 Don’t wait. Registrations are now open. 👇 Secure your spot here: platform.worldquantbrain.com/sign-up/IQC202… This is your chance to step into that arena.

2 days workshop - "Momentum Trading with Transformers" by @ObjRandom Date 18 April-19thApril Time - 7:00 PM IST/8:30 AM EST/1:30 PM GMT Register Now - topmate.io/quant_insider/… (Use Coupon code "20DNN" for 20% off)

The Mathematics of Option Pricing from a Quant by SIG youtu.be/u_eEW7N1xjE?si… We made notes from the lecture you can access it from here - drive.google.com/file/d/1GY2OvB…