David Data

Frontier LLMs are impressive at many things. Trading is not one of them. We had Fable, GPT-5.5, Gemini 3.5 Pro, and Grok 4.5 each run a portfolio for ~1.6 years on a synthetic stock market they've never seen. All of them lost money. Think you can beat the benchmark? Try it 👇

Taking the leap is hard, but there's joy in the unknown. @AdiKulkarni6 and I spent senior year pivoting through 4 ideas (including an AI hedge fund). Now we're building @davidhfdata — synthetic market data for superhuman AI trading. Excited to join @ycombinator S26!

Superhuman AI traders won't be trained on the past. They'll be trained on every possible future. That's what @sima_alexx and I are building at @davidhfdata. Excited to join @ycombinator S26

try it at davidhf.com

Try it at data.davidhf.com Docs: docs.davidhf.com

We spent $98,151.86 on stock markets that don't exist. Because your trading agent already memorized the ones that do. 2008, 2020, every real crash…it's all in the training data. David Data gives you synthetic markets your model has never seen.

@AtifHussainOG 'Worked last week, dead this week' usually means the edge was fit to recent conditions, nothing structural. Real edges have a reason — an inefficiency or a risk premium someone pays you for. If you can't name why it should keep working, the backtest only described the past.

The scariest part of trading isn’t losing money. It’s losing time. You can make money back. You cannot make back the 2 years you spent watching YouTube videos, buying courses, and sitting at your desk at midnight wondering why the strategy that worked last week stopped working this week. The traders who quit don’t quit because they ran out of money. They quit because they ran out of belief. And you know the saddest part… Most of them were closer than they thought. If you’re still here, still learning, still showing up even after the losses... You haven’t failed. You just haven’t found the right system yet. Keep going.

@MatiasScalbi Serious robustness stack. The subtle gap: block bootstrap and walk-forward still resample the one history that happened. Run thousands of strategies and a few clear the funnel on luck alone. The hardest test is whether they hold on synthetic markets that never printed.

Quería compartirles este logro de mi proyecto más importante, que vengo trabajando hace ya 3 meses QuantForge alcanzó su primer MVP, el motor completo corrió end-to-end por primera vez 🔥 ¿Qué es QuantForge? Un motor propio diseñado para generar miles de estrategias de trading vía algoritmo genético y filtrarlas con tests de robustez de nivel profesional, construido en Rust + Python Esta semana, por primera vez, la ingeniería completa corrió de punta a punta sobre data real (3 años de BTC horario): genera estrategias -> las valida out-of-sample -> las pasa por un funnel de robustez -> persiste todo de forma reproducible La infraestructura de robustez que armé (Fase 5) usa lo mejor del estado del arte: ✅ Walk-Forward Optimization ✅ Monte Carlo (block bootstrap, 5.000 simulaciones) ✅ System Parameter Permutation ✅ Deflated Sharpe Ratio (Bailey & López de Prado) Todo parado sobre tres obsesiones innegociables: - Anti-lookahead militante -> nada mira el futuro, y cada función está testeada para probarlo - Out-of-sample sagrado -> 30% de los datos jamás visto durante la optimización - Reproducibilidad bit-for-bit -> mismo seed, mismo resultado, siempre Falta mucho aún, el próximo gran paso es el paper trading (Fase 6), pero saber que voy por buen camino me motiva. Más aún de haber hecho todo esto desde lo autodidacta, leyendo y aprendiendo por horas y noches largas Mi idea sigue siendo la que les comenté al inicio: llegar a un mega portfolio de estrategias diversificadas y descorrelacionadas en PnL, al estilo del grandísimo y referente mío @IvanScherman QuantForge no tiene fin, siempre se va a actualizar y demás. Pero haber logrado todo esto me llena de alegría Pueden ir siguiendo el avance en este hilo, la idea es actualizarlo cada vez que pase un hito importante Saludos a todos y a motivarse, porque todos podemos Abrazo gente y buen finde

@L1vsun The five forces are the easy half. The trap: those loadings are estimated on one history. Funds that last pressure-test whether they hold across regimes that never actually printed. Stable in-sample, fragile out-of-sample is how a hidden force turns out to be a hidden overfit.

@ethanrkho @investingidiocy Carver's real point is statistical: your own track record is n=1. One price path, infinite ways to overfit to it. The shops that last think in distributions of outcomes, not the single timeline that happened to print...that's the whole job.

Ex-Man AHL ($70B firm) fixed-income head on breaking into quant: "You've got a much better chance of being hired by the world's best hedge fund from a non-target school with no qualifications than by trading your own money." Rob Carver (@investingidiocy) — ex-Man AHL, ran a multi-billion systematic fixed-income book | now a one-man shop across 200+ futures markets "I don't really believe I've found any inefficiency — basically all the money I make is risk premia." We cover: - The career myth that won't die — getting "noticed" by trading your own money is a one-in-a-billion event - Why he insists he's found zero market inefficiencies — it's public risk premia anyone can harvest - Skepticism as the #1 trait — every career error he's seen traces back to overconfidence in a backtest - Why "the best quants come from physics" is mostly path dependence — the Yale-historians thought experiment - His actual process: ~1 new strategy a year, a 1-in-5 strike rate — & he thinks more research would lower it - The real innovation of his last decade — running 200+ futures on a small account, not finding edge - Why he'd never join a pod shop — even though he reckons he could land an offer every couple of weeks Highlights: (00:50) The $1B loss week — why the desk stayed calm (02:45) Why he turned off his P&L email (06:40) Do the best quants really come from STEM? He pushes back (09:55) The Yale-historians trap — path dependence in quant hiring (11:35) The one trait that matters most — skepticism (13:00) The Sharpe ratio's blind spots — & the LTCM case (15:55) Geometric return vs Sharpe — the leverage catch (17:40) Avoiding overfitting — explicit vs implicit fitting (21:55) Why an honest backtest should look worse (23:05) Alpha decay by trading speed — HFT vs slow systems (25:35) Inside the portfolio — the full 10-rule suite (30:30) The career myth — trading your own money won't get you hired (31:55) Why he calls his returns risk premia, not inefficiency (33:55) His real innovation — 200+ futures on a small account (35:40) How Man AHL reviewed, vetoed & shipped new strategies (39:30) Will capital consolidate at Citadel & Millennium? (41:45) "My DMs are open" — & why he'd still never join a pod shop (42:45) The AI talent-war parallel — who you actually want to hire

Backtest on markets that never happened, agree or disagree?

Real history is one roll of the dice. We let you roll it a million times.

we want to help-max you if you're AI trading maxxing and don't have enough data maxxing or want to evaluate-max on future historically maxxed scenario maxing. We provide synthetic-maxxed financial-maxing data-maxxing API to the max (rate limit max) and we have AI cofounder Max.