XAnalytics@UBS · 51 verified calls backtested
@UBS's calls have trailed their sector by 1.7% per month, winning 50% of the time.
Even their cleanest calls trailed the market.
What “return vs sector” means
Across 20 bets, @UBS's average call trailed its sector by 1.7% — t = -1.3, not enough to tell skill from chance.
t = -1.3 — the strength of the evidence. Above 2, there's under a 5% chance it's luck.
How the t-value works
Each call is one bet, scored by how much it beat its own sector (fairer than the whole market). The t-value is the average beat divided by its uncertainty:
Same-week, same-sector calls ride one move, so they prove less than fully independent wins — the 20 bets count as 20 effective ones. That's why the honest t is -1.3, not the raw -1.3. Full breakdown on the analysis page.
Performance over time
| After | vs sector | vs S&P 500 | Hit rate | Calls |
|---|---|---|---|---|
| 1 day | -0.2% | -0.1% | 40% | 20 |
| 1 month | -1.7% | -0.0% | 50% | 20 |
| 3 months | -1.9% | +0.2% | 40% | 20 |
| 6 months | -5.2% | -2.6% | 25% | 20 |
“vs sector” compares each call to the stock's own industry — the fairest test of skill.
What drives the signal
Does conviction matter?
Return by post type
Long vs short
The evidence — best & worst calls
Biggest wins (after 1 month)
Full post
Full post
Full post
Biggest misses (after 1 month)
Full post
Full post
Full post
Their strongest setups — best combination at each horizon
- 1 day long news-reaction posts +0.0% vs -0.2% overall · 4 bets
- 1 month long news-reaction posts +1.4% vs -1.7% overall · 4 bets
- 3 months long news-reaction posts +1.4% vs -1.9% overall · 4 bets
Step behind the counter — every call, every chart, every time horizon. Filter by conviction or direction and chart any single trade.
Open the data room →Help shape XAnalytics
Was this report useful?
How we measured this · download the raw data
We pull @UBS's posts, use AI to extract each call (ticker, direction, conviction), and backtest it against real prices. A call wins only if it beat its own sector over the month after — not just if the stock rose. Headline numbers use deduplicated bets.
Download the full backtest data (JSONL)
Backtested results are computed from historical price data and do not predict future performance. Every signal here is extracted automatically from public posts and may misread intent, sarcasm, or context. This is not investment advice — verify everything independently before acting on it.