New Instrument: Ranking Models
Ranking Models live in a new menu on the left toolbar. Think of a ranking model as the last column of a stock screener — the thing that actually sorts the universe, not just filters it.
A screener decides which tickers are eligible.
A ranking model decides which tickers rank first.
That’s how strategies like Millennium Alpha, Vision, and Low Beta are built: screener + ranking model, not a screener alone.
How ranking actually works
We start with a stock universe from a screener (example: top 1,500 US stocks by dollar volume, the same filter used in Millennium Alpha).
Add metrics to the model (revenue, gross profit, sharpe ratio, GEX, etc.).
Assign each metric a weight score.
For each metric, the universe is sorted by that metric. Points are awarded by rank × weight:
Top ticker gets weight × number of stocks in the universe
Second ticker gets weight × (universe size − 1)
And so on down the list
Example
Universe = 100 tickers.
Sharpe ratio weight = 10. The universe gets sorted by sharpe ratio, then:
#1 by Sharpe → 10 × 100 = 1,000
#2 by Sharpe → 10 × 99 = 990
#3 by Sharpe → 10 × 98 = 980
If a metric has weight 1, the top name simply gets 100, the second 99, etc.
Then every metric’s points are added together. That total is the model’s final score. The highest combined score is rank #1.
Higher weights matter more. A metric scored 10 moves the ranking 10× as much as a metric scored 1.
Best Practices
You always need to attach a stock screener before ranking. The idea is not to mix common stocks with ETFs, or USD names with EUR names, in the same universe. Build a clean screener first (asset type, exchange, volume, min price, etc.), then rank that list.
One Ranking Model can run on several screeners. Switching tabs switches the universe, but the model stays the same.
You can download a CSV template with all available metrics and default weights set to 1. You can edit the weights yourself or by employing an LLM, then upload the model. Launch a backtest on the top N names (e.g. top 15 or top 20) to see how the ranking would have performed.
Ranking results are cached while the strategy stays in your strategy tab. Don’t delete it if you want to reuse that ranking for other tests, as the ranking process itself is computationally expensive (read: slow). Once a ranking is completed, you can easily switch parameters like rebalancing frequency, number of stocks included or a stop loss logic.
A note on shared models (Millennium / Vision)
Those models appear in your menu as shared objects, but the internal weights are hidden.
You can add your own overlay: type +1, +2, +5 (or a negative number) in the score column to nudge that model’s ranking without seeing the original table.
You now control the same type of ranking engine behind the platform’s flagship strategies. Screen a universe, weight the metrics, sum the rank scores, backtest the top names, iterate.
That’s the new instrument — a full ranking layer on top of the screener.
As always,
Happy Investing!
Andrei & Signal Sigma Team