What If You Always Picked Next Week’s Winner?
Here is a question we keep coming back to: if you somehow knew, every Monday, which Nasdaq-100 stock would do best over the next week, and you put the whole portfolio into that one name, what would the path look like?
You cannot trade that. Nobody can. But the hindsight path is still useful. It is a ceiling. It shows how much single-name dispersion sits inside the index week to week, and how far any real strategy (including ours) sits below that ceiling.
How we ran it
Same calendar as the live model. Weekly rebalance dates from model inception () through the latest completed week (). That is 20 holding periods.
For each week we take every Nasdaq-100 name with a valid price on both the rebalance date and the next one, pick the single best simple return, and hold it at 100% weight until the next rebalance. AI score, latent rank, and buy/hold/sell bucket are the ratings that existed at the start of that week, not after the fact.
Costs: 15 basis points per rebalance, same as the model config. With one name and full turnover every week, that is a full 15 bps each time.
The numbers
Starting from $10,000:
- Hindsight oracle (gross): $831,353 (+8,214%)
- Hindsight oracle (net of costs): $811,530 (+8,015%)
- Live model top-20 portfolio: about $11,655 (+16.6%)
- Nasdaq-100 (^NDX): about +20.2% over the same window
The oracle averages roughly 26% gross per week. Compound that for twenty weeks and the curve goes vertical. That is not a strategy. It is a measurement of how wild the cross-section was.
Read the oracle as an upper bound, not a target. If a product markets weekly single-stock timing as if that bound were reachable, they are selling a story, not a process.
Did the AI ratings point at the winners?
Mostly no. Five of the twenty weeks, the best forward return belonged to a name sitting in the sell bucket at the start of the week (INTC, MSTR, ARM twice, CSCO). Several other winners were middling holds. A few weeks did land on buys (MU, INSM, AXON), but that was not the pattern.
That is not a gotcha against the ratings. Quintile research still asks a different question: do higher-rated groups beat lower-rated groups on average? A single week's maximum is almost pure noise. The oracle is the noise, stacked.
Week by week
Stock held, AI rating at rebalance, that week's return, and the running total (gross).
| Week of | Stock | Score | Bucket | Week | Cumulative |
|---|---|---|---|---|---|
| Feb 17 | SHOP | 1 | hold | +9.3% | +9.3% |
| Feb 23 | AXON | 1 | hold | +25.7% | +37.4% |
| Mar 2 | APP | −1 | hold | +22.2% | +67.9% |
| Mar 9 | MU | 2 | buy | +25.9% | +111% |
| Mar 16 | BKR | 1 | hold | +14.6% | +142% |
| Mar 23 | INSM | 2 | buy | +8.2% | +162% |
| Mar 30 | STX | 0 | hold | +22.5% | +221% |
| Apr 6 | INTC | −3 | sell | +24.7% | +301% |
| Apr 13 | MSTR | −2 | sell | +31.7% | +428% |
| Apr 20 | ARM | −2 | sell | +32.2% | +597% |
| Apr 27 | STX | 1 | hold | +24.5% | +768% |
| May 4 | QCOM | −1 | hold | +43.6% | +1,146% |
| May 11 | CSCO | −2 | sell | +20.2% | +1,398% |
| May 18 | ARM | −2 | sell | +41.4% | +2,017% |
| May 25 | MU | 1 | hold | +38.0% | +2,822% |
| Jun 1 | MRVL | 1 | hold | +29.9% | +3,696% |
| Jun 8 | SNDK | 0 | hold | +28.0% | +4,759% |
| Jun 15 | MU | 2 | buy | +13.8% | +5,428% |
| Jun 22 | AXON | 2 | buy | +25.4% | +6,834% |
| Jun 29 | AXON | 2 | buy | +19.9% | +8,214% |
Repeat names show up a lot: MU three times, AXON three times, STX and ARM twice each. High volatility names dominate a max-return path. That is expected.
What we take from it
First, the gap between "best stock each week" and anything you can actually run is enormous. Our live top-20 book and the Nasdaq-100 both sit near +17% to +20% over this stretch. The oracle is two orders of magnitude higher. That gap is the cost of not knowing the future.
Second, chasing last week's winner (or pretending you can forecast next week's) is a different game from ranking a diversified sleeve. We care about the second game: fixed universe, published rules, ratings you can inspect, portfolios you can follow.
Third, when a sell-rated name posts the best week in the index, that is a reminder that ratings are probabilistic. They are not a crystal ball for the single best ticker.
We will refresh this analysis as more weeks land. Live model performance stays on the strategy models pages. The reproducible script lives in the repo as scripts/analyze-optimal-weekly-portfolio.ts.