ψ Psi Quant Research

Signal Report · US

UPST HOLD

Composite decision report grounded in 1-XH strategy signals and a completed multi-agent debate (agents-3node-v1), generated at 2026-10-28 10:38.

Final Agent Decision

Action
HOLD
Weight
0%
Confidence
35%
Variant
xiaohan_gogo_v2

Reasons

Bull Case (Analyst)

- Technical signal is a clean 1-XH "buy" (score 0.70, xiaohan_gogo_v2): price 24.15 sits above both the fast EMA (23.55) and slow EMA (23.83), and the gogo_detail shows synchronized bottoming across 1h/2h/3h/4h plus the 1h blue-upper-above-yellow-upper condition — a multi-timeframe turn rather than a single-bar blip.
- Cross-sectional composite z = +0.37 (rank 59/252, top 23.4%) with the strongest single lifter being xh_gogo_score at +2.44 — the same factor family driving the technical signal, so the factor rank and the timing signal corroborate rather than contradict.
- Secondary lifters beta_60d (+1.01) and bollinger_b_20d (+0.57) point to a high-beta name sitting in the lower part of its 20-day Bollinger range — favorable positioning if the regime stays risk-on.
- Fundamental/news backdrop is not bearish: Q2 2026 revenue of $364.7M beat expectations, the Upstart Macro Index ticked down to 1.49 (stock jumped 6% on that print), and consensus Q3 EPS is $0.60, +15.4% YoY, with estimates unchanged.

Bear Case (Analyst)

- The cross-section did not endorse the name: UPST is not in the production basket and ranks only 59/252 — the composite is positive but far from top-decile, so this is a mid-pack factor profile, not a conviction long.
- The draggers are structural, not noise: close_to_high_60d (-0.90), sharpe_60d (-0.77), and momentum_6m (-0.67) all say the stock is far below its 60-day high with poor risk-adjusted and medium-term momentum — the technical buy is fighting a downtrend.
- Event risk is acute and imminent: earnings on Nov 3, 2026, with options pricing a ~12% move — a 0.70-score technical entry three to four trading days ahead of a binary event is poor risk/reward.
- Analyst sentiment is negative on structure: Citizens cut its price target to $18 on bank charter concerns, well below the 24.15 close, and recent coverage shows sharp drawdowns (e.g., -19.77% in a week in one cited episode).
- The bull case rests almost entirely on the gogo timing factor (+2.44); with momentum and drawdown factors negative, there is no independent cross-sectional confirmation — if the multi-timeframe bottom fails, little else in the bundle supports the long.

Devil's Advocate

COUNTER TO BULL:
- "Clean 1-XH buy" is a single-factor timing signal; score 0.70 is not high conviction, and synchronized multi-TF bottoming is a lagging description of a bounce that may already be priced. EMA stack (24.15 > 23.83 > 23.55) is thin — a one-day reversal flips it.
- Corroboration is circular: xh_gogo_score (+2.44) and the technical signal are the *same factor family*. That's one bet counted twice, not two independent confirmations.
- Composite z = +0.37 at rank 59/252 is mid-pack. Calling that "corroboration" overstates it; top-quartile ≠ signal. Not in production basket = the model itself doesn't want it.
- beta_60d +1.01 and low Bollinger %B are only "favorable" *if* regime stays risk-on — that's a conditional, not a reason to buy. High beta into an event is leverage on the outcome, not edge.
- Fundamentals are backward-looking: Q2 beat and a 6% pop on the Macro Index are already in the price. "Estimates unchanged" cuts both ways — no upward revision momentum. Q3 EPS +15.4% YoY is consensus, i.e., the bar, not a catalyst.
- Hidden assumption: that the multi-TF bottom holds through a binary event. Nothing in the bull case addresses gap risk.

COUNTER TO BEAR:
- "Not in production basket / rank 59" is a threshold argument, not evidence of downside. Mid-pack names can still work; the bear treats a screen cutoff as a thesis.
- Draggers (close_to_high_60d, sharpe_60d, momentum_6m) are all *trailing* and highly collinear — they describe the same past drawdown three ways. Backward-looking factors are exactly what a bottoming/reversal setup is supposed to fade. Citing them as "structural" assumes mean-reversion won't happen, which is the very thing being bet against.
- Event risk cuts both ways: ~12% implied move is symmetric. A 0.70 entry pre-earnings is only "poor risk/reward" if you assume direction; the bear smuggles in a directional view while claiming it's about variance.
- Citizens' $18 PT is one analyst, pre-dated to bank-charter concerns, and PTs lag price. Using a stale target below spot as bearish evidence is weak — the market already disagrees at 24.15.
- "-19.77% in a week in one cited episode" is anecdotal, undated, and unfalsifiable as stated. Not a structural argument.
- Hidden assumption: that the gogo factor *must* fail. The bear's whole case is "if the timing signal fails, nothing supports it" — but that's true of every timing trade and isn't evidence it will fail.

Net: bull double-counts one factor and leans on stale fundamentals; bear dresses trailing momentum up as structure and treats a screen cutoff and one PT as proof. Both are overconfident.

Technical Signal (1-XH)

action
buy
score
0.70
variant
xiaohan_gogo_v2
timestamp
2026-10-28T15:30:00-04:00
close
24.15
ema_low_fast
23.55
ema_high_slow
23.83
gogo_score
0.7
gogo_detail
1h_bottom,2h_bottom,3h_bottom,4h_bottom,1h_blue_upper_above_yellow_upper

📈 Backtest Track Record xiaohan_gogo_v2 · signals simulated over available history (~3-month 30-min window, accumulating daily)

Trades
2
Win rate
100%
Avg / trade
+6.7%
Total (compounded)
+14%
best / worst
+8.8% / +4.6%
avg holding
35.9 days
open position
yes — marked-to-market +4.6%
Entry Exit Entry Exit Return Exit reason
2026-10-01 open 23.10 24.15 +4.6% open_position_mark_to_market
2026-03-16 2026-04-30 28.68 31.20 +8.8% adaptive_breakdown_1h

复盘 = each past xiaohan_gogo_v2 buy signal simulated: enter next bar, manage adds / adaptive sells / −25% hard stop. Demonstration only, not investment advice.

News & social (8 items)