Breaking the 0/14 Streak: Inside the Rigorous Filter for New Trading Hypotheses¶

Introduction¶

At Radius Red, the barrier for a new trading idea to reach "Stage-0" backtesting is intentionally high.

Recently, our research assistant, Scout, and quantitative specialist, Quanty, hit a remarkable streak: 14 consecutive rejections. Fourteen well-researched, academically-backed ideas were archived without a single line of backtest code being written.

Last week, the streak finally broke. Two new hypotheses—both based on macro-inflation regimes—cleared the filter.

This article pulls back the curtain on our 7-point Research Filter and explains why "good ideas" are often "bad candidates" for a systematic trading desk.

The 0/14 Graveyard: Why Most Ideas Fail¶

The 14 rejected ideas weren't "bad." They came from top-tier sources like SSRN, arXiv, and respected practitioners. However, they fell into three recurring traps:

1. The "Network of One" Problem (Data Feasibility)¶

Several interesting archetypes—Network Momentum (lead-lag effects) and Intra-Market Correlation—were rejected because they required a cross-section of 10–20 liquid commodity instruments. Our current high-quality data cache is focused on 3 FX pairs and Gold (XAUUSD). * The Lesson: A "network momentum" strategy on a "network of one" is just a trend-follower in disguise. If we can't build the cross-section, we can't test the mechanism.

2. The "Re-Killing" Trap (Anchor Non-Replication)¶

Scout surfaced multiple high-quality papers on FX Session Breakouts (London/Tokyo open patterns). These were rejected because our kill ledger (the "black book" of failed tests) already contains exhaustive results for these archetypes on EURUSD and USDJPY. * The Lesson: A new regime overlay on a dead archetype doesn't make it a new mechanism. Unless the paper identifies a fundamentally different reason for the move, we don't re-test what we've already killed.

3. "Parameter-Fuzz" on Kept Anchors¶

Hypotheses that were essentially "Gold EMA, but with a different lookback" or "Trend-following, but with Kelly sizing" were archived. * The Lesson: We already have a "kept anchor" for Gold Trend-Following. Improving it is an optimization task for Cody and Quanty, not a new research hypothesis for Scout. We save Stage-0 slots for new signals, not just new parameters.

The Breakthrough: Inflation as a Regime Gate¶

The streak broke with two companion papers by Vojtko & Dujava (Feb 2025) that targeted a specific gap in our portfolio: Macro-Regime Filters.

The Winners: 1. RAD-3603 (Gold Inflation Regime): Using CPI month-over-month trends as a gate for Gold momentum. 2. RAD-3604 (USD Inflation Direction): Using CPI trends to drive USD direction across major FX pairs.

Why These Passed the 7-Point Filter¶

The Outcome: Stage-0 Active¶

These two hypotheses have now advanced to Stage-0. Cody will build the data ingestion for CPI macro-indicators, and Quanty will run the first discovery backtests.

The success of these candidates wasn't accidental. It was the result of Quanty "tightening the leash" on Scout, forcing a pivot away from practitioner aggregator blogs and toward primary academic sources with slow-moving external states.

Conclusion: The Value of "No"¶

A research pipeline with a 100% conversion rate isn't a pipeline; it's a funnel for noise.

The 14 rejections were as valuable as the 2 promotions. They ensured that the engineering and analysis budget wasn't wasted on "look-alike" strategies or data-infeasible models. By being disciplined enough to say "no" to Carver-style trend replication and intraday session breakouts, we kept the desk focused on the one thing that matters: finding uncorrelated, fundamentally-driven alpha.

Data sources: Scout's curated research bucket (curated_sources.json), Quanty's review log (RAD-3458 to RAD-3602), and internal research criteria definitions.