THE RESEARCH
Where edge is likely: grading research ideas before testing them
Dark matter is invisible and only seen by its pull. Trading edge is similar: you rarely see it directly, only its effect on returns over many years. Before the factory tests anything, every research idea is graded on how likely it is to hold edge for a desk like this one.
The four grades
- Evidence (35%): how long the effect has been measured, in how many markets, and how often other researchers replicated it.
- Survival (25%): how much is likely left after publication and crowding. Published anomalies kept about 42% of their return on average once they were known (McLean and Pontiff, 2016), so survival is graded hard.
- Fit (25%): whether this desk can test and trade it with the data it has today.
- Cost (15%): whether it survives real spreads and fees at $2k to $25k.
The map, as of November 13, 2026
Each topic scores the average of its three best-graded ideas (out of 100):
- Factor premia (value, quality, low volatility) 87
- Momentum and trend 85
- Selling volatility (option sellers collect the insurance premium) 83
- Long-term accumulation (buy, hold, add, rebalance) 78
- Calendar and seasonality 76
- Filings and insiders 74
- Risk shields (overlays that cut size when risk rises) 73
- Event drift (prices keep moving after earnings and central-bank news) 72
- Intraday structure (the open, the close and overnight behave differently) 68
- Dealer positioning and flow 66
- Mean reversion (short, sharp moves partly reverse) 63
The bar every idea has to clear
Own the index and keep adding. Investing a lump sum has beaten spreading it out about two thirds of the time. Every strategy in the factory is measured against that, and most do not beat it after costs. A grade says where to look; only the tests say what is real. See the event horizon for those tests.
FAQ
How is a research idea graded?
On four things, 0 to 5 each: evidence (how long, in how many markets, how often replicated) weighted 35%; survival after publication 25%; fit with what a retail desk can test and trade 25%; and cost after spreads and fees 15%.
Why rank topics by their three best ideas?
A topic with one great idea and ten weak ones should not look worse than a topic with three good ones. The average of each topic’s three best-graded ideas is a fair read of where to look first.
Does a high grade mean the idea works?
No. The grade only says where edge is likely. Every idea still goes through the factory: both halves of history, a luck test corrected for how many ideas were tried, and forward trades it never saw.