Research cutoff: approximately 8:15 a.m. BST / 3:15 a.m. ET on Sunday, August 16, 2026.
Price cutoff: the Friday, August 14 U.S. regular-session close at 4:00 p.m. ET.
Target: the Monday, August 17 U.S. regular trading session close.
U.S. cash markets are closed on Sunday. There is no legitimate Sunday premarket signal to manufacture, so the forecast uses Friday's completed session plus weekend news and Monday's known event calendar.
The most important fact in today's update is that there is no new completed U.S. session since Saturday's forecast. That means there is no new out-of-sample result to score and no honest reason to pretend the model became more accurate overnight.
What has changed is the information set around Monday. The coming week begins with the Empire State Manufacturing Survey at 8:30 a.m. ET Monday, while oil and the Strait of Hormuz remain capable of moving Treasury yields before the open. Reuters' week-ahead reporting also describes a market that is increasingly asking earnings to justify AI valuations while the Federal Reserve gives little near-term policy guidance.
Friday itself remains the key cross-sectional test. The S&P 500 fell 0.17% and the Nasdaq 0.28% after July retail sales disappointed. Applied Materials fell 5.1% despite an upbeat forecast, and Broadcom lost 5.9%. AMD moved sharply the other way. That is exactly the kind of expectation-gap dispersion the old price-heavy and broad “AI demand is good” frameworks repeatedly failed to handle.
So today's report does not revert to momentum, and it does not promote Friday's winners automatically. The live framework remains a conservative relative-survival + expectation-gap filter: which stocks accepted their bullish story when the market became less forgiving, and which ones rejected it?
What the previous strategy got wrong
The older price-heavy model was already retired after producing only 31 correct active UP/DOWN calls out of 59, or 52.5%. More recent experiments also failed to establish a durable replacement edge.
Friday's catalyst/regime forecast was especially poor. Of seven calls that could be independently verified from same-day reporting at Saturday's cutoff, only AMD finished in the predicted direction. That was 1/7, or 14.3%. An always-UP rule on the same seven names would have scored 2/7, or 28.6%.
That failure matters more than a neat narrative. The market rewarded AMD while punishing a beat-and-raise from Applied Materials and heavily selling Broadcom. A strategy that mostly asks whether AI fundamentals are strong cannot distinguish those outcomes.
Candidate strategy comparison
The most complete apples-to-apples audit remains the reconstructed seven-session sample across 84 ticker-sessions. The older four sessions were used as calibration and the newest three as holdout.
| Strategy family | Full sample | Calibration | Holdout | Avg. signed return | Status |
|---|---|---|---|---|---|
| Always predict UP | 67.9% | 70.8% | 63.9% | +1.07% | Baseline to beat |
| Prior-session ticker direction | 60.7% | 64.6% | 55.6% | +0.75% | Failed baseline |
| Follow prior QQQ direction | 56.0% | 70.8% | 36.1% | +0.68% | Reject |
| One-day mean reversion | 39.3% | 35.4% | 44.4% | -0.75% | Reject |
| Fresh event / expectation-gap gate | 5/5 qualifying historical cases | 4/4 | 1/1 untouched case | Not consistently reconstructed | Useful feature, tiny sample |
| Broad catalyst + regime framework | Not fully reconstructable | — | 3/9 verified Aug. 11; 1/7 verified Aug. 14 | — | Reject as default |
| Community + options standalone | Incomplete timestamp coverage | — | No defensible holdout | — | Confirmation only |
| Relative-survival + expectation filter | New live framework | — | No completed holdout yet | — | Use conservatively |
The conclusion is uncomfortable but useful: no tested strategy currently beats the simple always-UP baseline on held-out observations. Therefore today's probabilities are forecasts from the least-bad live framework, not claims of demonstrated predictive edge.
Calibration and scoring status
The transparent scoring record remains:
- Retired price-heavy active directional accuracy: 31/59 = 52.5%.
- Retired predictions at 60%+ confidence: 22/39 = 56.4%.
- Retired 65%+ bucket: not consistently issued, so no honest score is available.
- Retired Brier score: 0.257.
- Event-day classification: 3/24 = 12.5% on two reconstructed event sessions.
- Ordinary-day classification: 34/60 = 56.7% on the other five sessions.
- Always-UP average signed return: about +1.07% per ticker-session over the seven-session audit, before costs.
- Prior-direction average signed return: about +0.75%.
- QQQ-direction average signed return: about +0.68%.
- One-day mean-reversion average signed return: about -0.75%.
There is no new Sunday outcome to add to those numbers. Monday will be the first completed session that can score the current relative-survival filter prospectively.
Ticker-level lesson from the old price family
One-day momentum accuracy in the same seven-session audit was highly uneven:
| Ticker | Old one-day momentum accuracy |
|---|---|
| AAPL | 71.4% |
| NVDA | 57.1% |
| MSFT | 71.4% |
| AMZN | 71.4% |
| TSLA | 71.4% |
| GOOGL | 57.1% |
| META | 57.1% |
| AVGO | 85.7% |
| AMD | 28.6% |
| ORCL | 71.4% |
| TSM | 57.1% |
| ASML | 28.6% |
That is why AMD's Friday surge is not being treated as “momentum says up.” The information is in the relative divergence: AMD survived a session in which several AI/equipment names were rejected.
The selected Sunday-to-Monday framework
The current ranking uses seven layers:
- Friday acceptance/rejection. Did the stock outperform when investors became less forgiving?
- Expectation gap. Did good fundamental news produce good price action, or did the stock fall anyway?
- Fresh company catalysts. Weekend information only gets weight if it genuinely changes Monday's setup.
- Sector confirmation. Semiconductor calls need participation beyond one ticker.
- Macro regime. Oil, Treasury yields, Hormuz headlines and Monday's Empire State survey can override company evidence.
- Analyst revisions. Fresh changes matter more than stale price targets.
- Community/options. Public sentiment is used as a crowding check only. I could not verify sufficiently fresh, contextual single-stock options data this weekend, so options receive no directional weight rather than being invented.
Public community snapshots for NVDA, AMD and AVGO are also too stale/noisy to justify a directional score. They are useful mainly as a reminder that AMD/Nvidia enthusiasm is crowded while Broadcom discussion is much thinner.
Monday forecast for all 12 stocks
Expected ranges are rough close-to-close volatility bands, not price targets. Weekend geopolitical news can create gaps beyond them.
| Ticker | UP | DOWN | Expected Monday range | Conviction | Dominant signal |
|---|---|---|---|---|---|
| AAPL | 52% | 48% | -2.0% to +2.1% | Low | Friday resilience versus recent estimate pressure; no fresh positive catalyst strong enough for more |
| NVDA | 57% | 43% | -2.8% to +3.2% | Medium | Held almost flat while Nasdaq and several AI peers weakened; Aug. 26 earnings keep expectations high |
| MSFT | 53% | 47% | -2.0% to +2.2% | Low | Azure/AI economics remain strong, but the stock has already absorbed a large earnings re-rating |
| AMZN | 48% | 52% | -2.4% to +2.5% | Low | Friday's weak retail-sales backdrop directly challenges the consumer side despite strong AWS |
| TSLA | 53% | 47% | -3.5% to +3.8% | Low | High beta can benefit from risk-on Monday, but no fresh verified weekend catalyst deserves extra weight |
| GOOGL | 51% | 49% | -2.5% to +2.5% | Coin flip | Strong core business versus capex/free-cash-flow sensitivity; no decisive Friday signal |
| META | 52% | 48% | -2.4% to +2.5% | Low | Strong ad economics versus exceptionally heavy AI infrastructure commitments |
| AVGO | 44% | 56% | -4.0% to +4.0% | Medium DOWN | Friday's 5.9% fall is a clear expectation rejection; long-term AI networking demand is the rebound risk |
| AMD | 58% | 42% | -4.0% to +4.5% | Strongest lean | Strong Friday relative survival while AI/equipment peers sold off; Monday debt settlement is a counterweight |
| ORCL | 48% | 52% | -3.5% to +3.8% | Low | AI backlog is real, but financing/lease concentration makes the stock sensitive to higher yields |
| TSM | 56% | 44% | -2.6% to +2.8% | Medium-low | July revenue rose about 45% YoY; foundry diversification offsets some single-customer risk |
| ASML | 54% | 46% | -3.2% to +3.4% | Low | Structural lithography demand remains strong, but Friday's equipment selloff says the expectation bar is dangerous |
No name reaches 60%. Under this report's own rules, a 60%+ call requires holdout evidence from a strategy that beats simple baselines. We do not have it.
The three strongest leans
These are the strongest available signals, but none qualifies as a true high-conviction call under the historical thresholds.
1. AMD — 58% UP / 42% DOWN
AMD still has the cleanest positive cross-sectional signal from Friday. The key is not that the stock went up; it is that it rose strongly while Broadcom and Applied Materials were being repriced lower.
That suggests investors were differentiating among AI beneficiaries rather than simply abandoning semiconductor exposure. Recent Bank of America commentary has also highlighted AMD as a preferred chip name as analysts expect the AI-server semiconductor market to keep expanding.
The contradiction is financing and expectations. Reuters reported that AMD launched a $4–5 billion, four-tranche debt offering, with settlement expected Monday. The company also still carries the early-August expectation reset that followed otherwise strong results. Friday's relative strength improves the setup; it does not erase the risk.
What would invalidate the lean: AMD gives back most of Friday's relative advantage while NVDA/TSM remain firm, or a macro shock sends oil and yields sharply higher before the open.
2. Nvidia — 57% UP / 43% DOWN
Nvidia is the cleaner large-cap survivor. Friday's stock move was roughly flat while the Nasdaq fell and several AI-linked peers were hit much harder. In a market that punished crowded stories, not breaking was information.
The fundamental support remains obvious but should not be over-counted. Strong cloud and AI-infrastructure earnings continue to support accelerator demand, and Nvidia reports fiscal second-quarter results on August 26. The same earnings date also raises the expectation bar and can make positioning more crowded as it approaches.
What would invalidate the lean: NVDA loses its relative advantage while AMD and TSM hold up, or the semiconductor group opens broadly weak after a rise in yields.
3. Broadcom — 56% DOWN / 44% UP
Broadcom is the best negative expectation-gap setup. The long-term AI networking/custom-silicon thesis did not disappear Friday, yet the stock fell 5.9%. That kind of rejection matters precisely because investors already know the bullish story.
The wrong response is automatic mean reversion. The historical one-day mean-reversion strategy scored only 39.3% overall and 44.4% on holdout. A large Friday decline is not, by itself, a Monday buy signal.
What would invalidate the bearish lean: AVGO opens strong and holds the recovery while AMD, Nvidia and broader semiconductor breadth also remain positive. That would suggest Friday was exhaustion rather than a continued expectation reset.
TSMC and ASML: good businesses, dangerous expectation bars
TSMC's official July revenue was approximately NT$467.6 billion, up about 44.7% year over year. That supports the physical AI-capacity thesis and gives TSM one of the strongest fundamental demand signals in the watchlist.
ASML has the same structural support farther upstream. But Friday's Applied Materials reaction is a warning for the entire equipment complex: a strong quarter and upbeat outlook can still be insufficient when valuation already assumes a lot of AI spending.
That is why TSM stays at 56% while ASML is only 54%. The businesses remain strong; the short-horizon expectation risk is different.
The low-conviction group
AAPL, MSFT, TSLA, GOOGL and META are all between 51% and 53% UP. These are effectively low-conviction calls.
Apple's recent operating quarter was strong, but estimate and supply concerns still matter. Microsoft has perhaps the cleanest enterprise AI economics in the group, yet the stock already experienced a huge post-earnings re-rating. Alphabet and Meta continue to produce strong advertising/cloud evidence while carrying very large AI-capex obligations.
Tesla remains a high-beta stock with a wide Monday range, but I did not find a sufficiently strong new weekend catalyst to justify the higher confidence assigned in Saturday's first-pass report, so the probability is trimmed to 53%.
AMZN and ORCL lean slightly DOWN at 52%. Amazon's AWS story remains constructive, but weak retail-sales data is directly relevant to its consumer business. Oracle remains unusually sensitive to financing conditions because the AI capacity buildout requires large capital and lease commitments.
What can invalidate Monday's forecast
1. Empire State Manufacturing at 8:30 a.m. ET
The New York Fed's August calendar schedules the Empire State Manufacturing Survey before the U.S. open Monday. A large upside growth/inflation surprise can push yields higher; a very weak reading can turn “rate relief” into a growth scare.
2. Oil and the Strait of Hormuz
Friday's weak consumer data did not push oil lower because shipping through Hormuz remained severely disrupted. Another weekend escalation could lift crude and long yields, which is especially uncomfortable for long-duration growth stocks.
3. Semiconductor breadth
AMD/NVDA bullish leans are much stronger if TSM, ASML and the broader chip complex participate. If only one ticker is green, the signal becomes company-specific and less durable.
4. Broadcom repair
AVGO is the clearest expectation-rejection stock. A strong opening recovery that holds would be evidence against the bearish lean and, more broadly, evidence that Friday's AI selloff was an exhaustion event.
5. The week's earnings calendar
Reuters' week-ahead preview says investors are leaning heavily on earnings to validate AI and equity valuations. Retail earnings later in the week, including Walmart and Target, can also change the interpretation of Friday's weak sales data even though those reports arrive after Monday's target session.
Transparent audit: what Monday will actually teach us
The useful test Monday is not whether every one of 12 arrows is correct. It is whether the three strongest cross-sectional calls—AMD UP, NVDA UP and AVGO DOWN—outperform the near-coin-flip group.
If those three fail together, the relative-survival framework should not be defended with another explanation. It needs another material strategy change.
If they perform better while the low-conviction names remain mixed, that is at least evidence that selectivity is improving—even though one session would still be far too small to claim a validated edge.
For now, the honest hierarchy is:
- AMD UP 58% — strongest positive relative-survival signal.
- NVDA UP 57% — cleanest large-cap AI survivor.
- AVGO DOWN 56% — clearest expectation rejection.
Everything else is closer to a watch than a forecast worth trading aggressively.
Educational disclaimer: This article is for informational and educational purposes only. It is not financial advice, investment research, or a recommendation to buy or sell any security. Short-horizon stock direction is highly uncertain, probabilities are estimates rather than guarantees, and weekend geopolitical or macro news can create Monday gaps beyond the ranges discussed.
Sources
Friday market and week-ahead context
- Reuters — S&P 500 ends lower as investors weigh data and Middle East tensions, Aug. 14
- Reuters — Wall Street Week Ahead: investors look to earnings to keep stocks afloat, Aug. 14
- New York Fed — August 2026 economic indicators calendar
Semiconductors / AI infrastructure
- Reuters — AMD launches $4–5 billion debt offering, Aug. 13
- TSMC — 2026 monthly revenue
- Investopedia — chip-sector rebound and Bank of America analyst view
- Nvidia Investor Relations — Q2 FY2027 results scheduled for Aug. 26
- Reuters — ASML's AI-chip demand, valuation and execution debate
Current company context
- Apple — fiscal third-quarter 2026 results
- Microsoft Investor Relations — FY2026 Q4 results
- Amazon Investor Relations — quarterly results
- Meta Investor Relations — quarterly results and events
- Oracle — fiscal Q4 and full-year 2026 results
Cover image
