Model 02 · Ranking
A forecast should know how uncertain it is.
A walk-forward model ranks liquid US stocks by expected return relative to the market over the next month. It is a sorting signal—not a crystal ball.
01
Rank the field, do not predict the tape
Predicting the market’s direction and ranking stocks against one another are different problems. Phidox targets next month’s return minus the equal-weighted market return. That keeps the model focused on relative opportunity rather than learning the unhelpful fact that equities tend to drift upward over long periods.
Its inputs combine momentum, volatility, drawdown, liquidity, beta, profitability, growth, leverage, valuation, size, sector, and the age of the latest fundamentals. Every numeric feature is ranked cross-sectionally at the date it would have been known.
02
The test is the product
The model is evaluated with an expanding, walk-forward window and retrained every twelve months. Across 197 unseen months, the ordering had a positive rank correlation more often than not, with a modest average signal. That is the honest shape of financial prediction: weak, noisy, and potentially useful in aggregate.
Phidox can use the ranking when it helps a question, ignore it when it does not, and show the metric alongside the output. A single percentile is never presented as certainty.
03
Where the model stops
- It ranks stocks relative to one another; it does not forecast the market’s direction.
- The historical universe is based on current listings and therefore has survivorship bias.
- Any individual month can—and often will—disagree with the average signal.
Try it in the analyst
“Which large US stocks rank highest for next-month relative return, and why?”Ask this question