Why We Open-Source Our Quant Research

The Quant Industry Has a Transparency Problem

Every week, a new “quant trading course” promises to teach you the secrets of systematic trading for $2,000. Every month, a new fintwit guru posts an equity curve with no methodology. Every year, retail traders collectively lose billions chasing strategies they cannot verify, cannot reproduce, and cannot trust.

We started Curupira because we were tired of it.

Not tired of losing — we’ve done plenty of that too. Tired of the asymmetry. The hedge funds publish nothing. The gurus publish screenshots. The academics publish papers two years after the alpha has decayed. And retail traders are left picking through the wreckage, trying to figure out what actually works.

The Thesis: Transparency Is Alpha

Here’s the counterintuitive argument: sharing your research makes you a better trader.

Not because of some vague open-source philosophy. Because of three concrete mechanisms:

1. Survivorship bias dies in public. When you publish every strategy — including the 27 out of 31 that failed — you can’t lie to yourself about your hit rate. You can’t cherry-pick the one backtest that worked and forget the rest. We tested 31 systematic strategies. Four survived. That 12.9% survival rate is the most important number we’ve ever published, because it keeps us honest.

2. Peer review finds bugs faster. Our entropy collapse volatility timing strategy had a look-ahead bias in the first version. We caught it because someone on our Discord pointed out that we were using the close price to calculate the signal for the same bar. That’s a $50,000 mistake in production. It cost us nothing because we published the code.

3. Compounding knowledge beats compounding capital. A single trader can test maybe 5 strategies per month if they’re disciplined. A community of 200 traders, sharing methodology and results, can test 50. The strategies that survive community scrutiny are battle-tested in a way that private research never is.

What We Actually Publish

Let’s be specific about what “open-source quant research” means at Curupira:

Full strategy specifications. Not just “we use RSI and MACD.” The exact entry rules, exit rules, position sizing, and risk parameters. Every parameter, every threshold, every filter.
Complete backtesting code. Python notebooks you can run yourself. Our data sources, our preprocessing steps, our execution assumptions. If you can’t reproduce our results, the research is worthless.
Honest performance metrics. Profit factor, Sharpe ratio, max drawdown, win rate — but also the metrics people hide: number of trades, average hold time, sensitivity to parameter changes, and out-of-sample degradation.
The failures. This might be the most important part. We publish strategies that didn’t work, with analysis of why they didn’t work. A well-documented failure is more valuable than a suspiciously perfect backtest.

This is the question we get most often, and it deserves a serious answer.

Yes — if your edge is a single exploitable inefficiency with limited capacity, sharing it will kill it. If you’ve found a way to arbitrage a mispriced option on a low-liquidity exchange, don’t post it on Twitter.

But that’s not what most retail quant research produces. Most of what we research falls into two categories:

Risk premia harvesting. Strategies that profit from bearing risk that others don’t want — volatility selling, momentum, carry. These edges don’t disappear when more people know about them. They’re structural features of markets. The academic literature has published hundreds of papers on momentum, and it still works.

Statistical regime detection. Strategies that identify when market conditions favor certain approaches — our entropy collapse work, Hurst exponent analysis, regime classification. Knowing that entropy collapse predicts volatility expansion doesn’t help you if you don’t have the discipline to execute the strategy mechanically over 500 trades.

The real edge in systematic trading isn’t the signal. It’s the infrastructure, the risk management, the execution, and above all, the discipline. Those can’t be copied from a GitHub repo.

What We’ve Built So Far

In two months of public research, we’ve published:

Entropy collapse volatility timing — a strategy that uses Shannon entropy of price returns to time volatility regime changes, achieving a 1.44 profit factor on EURUSD.
FVG magnetism analysis — the first rigorous statistical test of whether fair value gaps actually “fill” as price action traders claim (spoiler: it’s more nuanced than the influencers suggest).
Hurst exponent mean reversion detection — using the Hurst exponent to dynamically switch between trend-following and mean-reversion strategies in forex.
A $1.30/day LLM trading agent — proof that you don’t need a $10,000/month infrastructure budget to run AI-assisted trading on crypto markets.
31 strategy test results — the full dataset of what we tested, what survived, and what we learned from the failures.

The Vision: A Quant Research Cooperative

Curupira isn’t a hedge fund. We’re not managing capital. We’re not selling signals or subscriptions.

We’re building a research cooperative — a community of traders who believe that transparent, reproducible, honestly-reported research is the foundation of sustainable edge in markets.

Our code is on GitHub. Our strategies are documented to the parameter level. Our failures are published alongside our successes.

If you think that’s naive, you haven’t been paying attention to what open source has done to every other industry it’s touched.

If you think that’s interesting, read our research. Run our code. Break our backtests. Tell us what we got wrong.

That’s how we all get better.

This is the first post on the Curupira blog. Start with our entropy collapse research for a taste of what our strategy posts look like, or read 31 Strategies Tested, 4 Survived for the big picture.