Does AI Stock Picking Work? The ClaritX Forward-Tested Record
ClaritX pre-registers every AI stock-selection strategy with frozen rules and judges it only on forward data — wins and losses stay public. Strategies are simulated with next-close execution and 10bps/side trading costs against a total-return S&P 500 benchmark. Nothing is backfilled: each arm has a registered inception date and its curve starts there.
As of , ClaritX tracks 16 registered strategy arms. The strongest live arm is AI band × momentum Top-30 at -1.0pp vs the S&P 500 over its window (total return +1.0% vs SPY +2.1%). The weakest arm, Top 10 by AI score (control), is -18.3pp — published, not hidden: negative results stay on this page permanently.
| Strategy arm | Live since | Days | Return | S&P 500 | Alpha | IR | Status |
|---|---|---|---|---|---|---|---|
| AI band + valuation room | 117 | +9.6% | +2.8% | +6.8pp | 1.24 | Passed Gate 1 on 2026-08-21; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| Live Ranking (dynamic) | 103 | +8.4% | +2.8% | +5.6pp | 0.82 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| Lab: Live Ranking, momentum + value only | 28 | +3.7% | -1.6% | +5.3pp | 2.94 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| Lab: Live Ranking, value ranked within sector | 28 | +1.1% | -1.6% | +2.7pp | 1.78 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| Lab: Live Ranking, exact top 30 daily | 28 | +0.3% | -1.6% | +1.9pp | 1.23 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| AI band + hysteresis | 117 | +3.2% | +2.8% | +0.5pp | 0.10 | Retired 2026-09-23: failed its pre-registered kill rule | |
| AI score band 70–89 | 117 | +2.9% | +2.8% | +0.2pp | 0.03 | Retired 2026-09-23: failed its pre-registered kill rule | |
| Lab: Live Ranking, max 6 per sector | 28 | -1.7% | -1.6% | -0.1pp | -0.04 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| AI band × momentum Top-30 | 68 | +1.0% | +2.1% | -1.0pp | -0.12 | Testing | |
| Momentum 12-1/vol Top-30 (SPMO recipe) | 70 | +0.8% | +2.4% | -1.6pp | 0.06 | Retired 2026-10-22: failed its pre-registered kill rule | |
| Site Ranking (composite v3, quarterly) | 116 | +1.1% | +2.8% | -1.7pp | -0.29 | Retired 2026-10-20: failed its pre-registered kill rule | |
| Live Ranking (PMHM, forward test) | 40 | -3.6% | -1.9% | -1.7pp | -0.77 | Testing; AI inputs stale since 2026-10-15, later numbers are not evidence | |
| AI band percentile | 117 | +0.2% | +2.8% | -2.5pp | -0.50 | Retired 2026-10-12: failed its pre-registered kill rule | |
| Composite v2 (control) | 116 | -0.4% | +2.8% | -3.3pp | -0.55 | Retired 2026-10-16: failed its pre-registered kill rule | |
| Top 30 by AI score (control) | 116 | -8.1% | +2.8% | -10.9pp | -1.63 | Retired 2026-09-23: failed its pre-registered kill rule | |
| Top 10 by AI score (control) | 116 | -15.5% | +2.8% | -18.3pp | -2.43 | Retired 2026-09-23: failed its pre-registered kill rule |
Do higher AI scores do better?
As of , across 1,085 forward-tested analyses: stocks scored 90–100 ran -3.9pp against the S&P 500 with a 45.6% win rate; stocks scored 0–49 ran -12.0pp with a 32.8% win rate. No band beat the S&P 500 on this snapshot: bands scored 70 and above ran -3.9pp to +0.4pp against it. The score still separated losers from the rest — bands below 70 ran -12.0pp to -8.3pp — so on this data the signal works as a loser filter, not as a source of alpha. Published as measured.
| AI score band | Analyses | Win rate | Mean return | Alpha vs S&P 500 |
|---|---|---|---|---|
| 90–100 | 57 | 45.6% | -1.1% | -3.9pp |
| 85–89 | 206 | 60.7% | +3.2% | +0.4pp |
| 80–84 | 194 | 55.7% | +0.2% | -2.2pp |
| 70–79 | 300 | 57.0% | +2.3% | +0.1pp |
| 50–69 | 209 | 39.2% | -6.3% | -8.3pp |
| 0–49 | 119 | 32.8% | -10.0% | -12.0pp |
Each analysis is scored 0–100 when published, then judged only on what the stock did afterwards, against a total-return S&P 500 over the same window. No backfilling, no survivorship pruning; the losing bands are in the table.
How to read this
Every arm follows pre-registered rules (entry, exit, rebalancing) frozen before its clock starts. Changing a rule restarts the clock. Kill criteria and graduation gates (8-week, 6-month, 12-month) are defined in advance. Past performance does not guarantee future results — this is educational research, not investment advice.
Methodology & evidence — Research Hub · Monthly AI Rank Report · Free AI Stock Rankings · Disclaimer