Guest mode: your portfolio is kept in this browser until you connect.
Sys online·Guest·--:-- local
> research
Market ideas, tested on real data.
Out of sample, costs included, sources linked. Each study asks one question and publishes the answer, including the ideas that fail.
25 ideas tested: 13 clear answers, 8 that depend on conditions and 4 with no edge.
29 of 50weekends, Nov. 2025 to Oct. 2026, on which Tesla’s Hyperliquid perp sat on Sunday evening on the side of Friday’s close where Monday opened; 42 at noon UTC Monday
data: Hyperliquid xyz 4-hour and hourly candles; Yahoo Finance daily opens and closes · Nov. 2025 to Oct. 2026
On Sunday evening, Tesla’s perp missed Monday’s open by a median 1.03%, against 1.12% for Friday’s close. Across 85 US stocks and funds, the perp was nearer the open than Friday’s close in 45% of weekends until 6 p.m. Sunday, and 79% at 9 a.m. Monday, once futures and overnight trading had reopened.
Not on Sunday; by Monday morning, yes.29 of 50 weekends, Nov. 2025 to Oct. 2026, on which Tesla’s Hyperliquid perp sat on Sunday evening on the side of Friday’s close where Monday opened; 42 at noon UTC Monday
On Sunday evening, Tesla’s perp missed Monday’s open by a median 1.03%, against 1.12% for Friday’s close. Across 85 US stocks and funds, the perp was nearer the open than Friday’s close in 45% of weekends until 6 p.m. Sunday, and 79% at 9 a.m. Monday, once futures and overnight trading had reopened.
29 of 50weekends, Nov. 2025 to Oct. 2026, on which Tesla’s Hyperliquid perp sat on Sunday evening on the side of Friday’s close where Monday opened; 42 at noon UTC Monday
Less than one fund twice.41.3% of their money is the same, on Vanguard’s VGT file of September 30 and VanEck’s SMH file of October 29, 2026
79.9% of SMH sits in companies VGT holds, and 20.0% in five chip companies based abroad that VGT doesn’t hold. Across ten tech and growth ETFs, VGT and FTEC are 95.1% alike and SOXX and IGV share no company.
Higher a month later, mostly in its first weeks.6 of 9 days HYPE closed 15% or more higher, Nov. 2024 to Oct. 2026, after which it was higher 30 days later; median 30-day change across all 9 +28.2%, but 5 of the 9 came in its first month of trading
Hyperliquid (HYPE) fell the next day after only 2 of its 9 15% days and was higher a month later after 6; the median 30-day change across all 9 was +28.2%. Five came in the weeks after the November 2024 airdrop; of the 4 since 2025, 3 were higher 30 days later.
6 of 9days HYPE closed 15% or more higher, Nov. 2024 to Oct. 2026, after which it was higher 30 days later; median 30-day change across all 9 +28.2%, but 5 of the 9 came in its first month of trading
Less than feared both ways, more one way.374 of 1,431 ETF pairs under 50% overlap both ways in which one fund has 75% or more of its money in the other’s companies, issuer files of September 25 to October 18, 2026
The median pair of 54 ETFs shares 4.3% of its money and 172 pairs share half or more. Read one way, 633 pairs have one fund half or more inside the other. Top-10 lists show 908 pairs sharing nothing, against 448 on the full files.
374 of 1,431ETF pairs under 50% overlap both ways in which one fund has 75% or more of its money in the other’s companies, issuer files of September 25 to October 18, 2026
In a falling market, yes; alone, no.6 of 6 S&P 500 declines of 19% or more since 2000 in which Hermès (RMS.PA) fell less than the index; in 2026 it is −39.2% through Oct. 22, 2026 while the S&P 500 is +11.9%
Hermès fell a median −15.6% in the six declines, against −29.7% for the S&P 500, and LVMH −22.4%. 2026 is the first year since at least 2001 that both fell 30% or more while the S&P 500 rose; Hermès’s sales grew and its price/earnings went from 65.1 to 30.1.
6 of 6S&P 500 declines of 19% or more since 2000 in which Hermès (RMS.PA) fell less than the index; in 2026 it is −39.2% through Oct. 22, 2026 while the S&P 500 is +11.9%
Payout is not return.$91.85 left from $100 in TLT, 5% a year out, Oct. 2023 to Sept. 2026
The payout covered 90.4% of the withdrawals, but TLT’s price fell far enough that $100 with 5% taken out each year was $91.85 after 35 months, against $105.33 left alone; the 10-year yield was 4.88% at the start.
Same direction, not the same amount.20.3% of 261 weeks to Oct. 16, 2026 in which SMH and SOXX were more than 1 point apart, at a 0.987 correlation; over the last year SMH returned 13.82 points less
Of 32 ETF pairs that share half their holdings, the four on the same index never had a week more than 1 point apart. The 28 on different indexes correlated 0.930 to 0.996 but ended the last year a median 5.37 points apart. Whether a pair is substantially identical for the wash-sale rule is a legal question the data does not answer.
20.3%of 261 weeks to Oct. 16, 2026 in which SMH and SOXX were more than 1 point apart, at a 0.987 correlation; over the last year SMH returned 13.82 points less
A coin flip on direction.32 of 60 results reports since 2012 after which Tesla (TSLA) fell the next session; the median move was 6.58% either way
Tesla fell after 32 of its 60 results reports and rose after 28, so the direction was close to a coin flip. The median move in either direction was 6.58%, about 3.7 times an ordinary session, and 16 reports moved the stock 10% or more.
No more than an ordinary weekend.+0.06% median Friday-to-Monday move of the CAC 40 across 405 weekends since October 2018; the three protest weekends of 2018 ranged from −1.47% to +1.00%
Three protest weekends, three moves inside the range of ordinary weekends. Too few to say anything about protests in general.
4 min read
+0.06%median Friday-to-Monday move of the CAC 40 across 405 weekends since October 2018; the three protest weekends of 2018 ranged from −1.47% to +1.00%
On the first day, usually.13 of 17 evening product events since 2012 after which Tesla (TSLA) fell the next session; the median first session was −4.54%
The first session after a Tesla show moved a median 4.54%, about 2.4 times an ordinary day, and the 7 times it opened higher it closed under the open. Twenty sessions later the stock was higher 9 times out of 17.
Not across issuers.90.8% of the Vanguard Mid-Cap ETF (VO) in S&P 500 companies, October 2026; it shares one company with the iShares Core S&P Mid-Cap ETF (IJH)
Vanguard and S&P draw the size bands in different places. The Vanguard Mid-Cap ETF is mostly S&P 500 companies, the Vanguard Small-Cap ETF holds only 18.6% in S&P SmallCap 600 names, and three small-cap funds overlap 18.3% to 47.2% by weight. Same-index funds such as VOOG and SPYG are 97.7% alike.
Big gains, deeper falls.−53.5% median worst fall of the eight US-listed neoclouds from March 28, 2025 to Oct. 12, 2026; Nvidia −20.2%
An equal-weight basket of CoreWeave, Nebius, IREN, Applied Digital, Cipher Digital, TeraWulf, Hut 8 and Core Scientific turned $1,000 into $5,989, against $2,069 in Nvidia, with a median volatility of 106.5% a year. Index funds hold little of them: $41.47 in $10,000 of QQQ, none in VOO.
Only when yields climbed fast.−9.3% median S&P 500 change a year after the 56 records set since 1963 with the 10-year at 5%+ and up a point on the year; +14.9% when it was not rising
Until 2008, 80.6% of S&P 500 records came with the 10-year at 5% or more, and when the yield was not climbing fast the index gained a median 14.9% the next year. The 56 records set after a 1-point jump fall into six runs: 4 of 6 were higher a year after their first record, but three ended at the tops of 1980, 1987 and 2000.
−9.3%median S&P 500 change a year after the 56 records set since 1963 with the 10-year at 5%+ and up a point on the year; +14.9% when it was not rising
Only if record profit returns.80.2% under Nike’s record close on Oct. 1, 2026, its deepest fall since it listed in 1980
Nike recovered from three earlier falls of 50% or more and beat the S&P 500 in the decade after each. This time the profit fell too: in eight scenarios from its own numbers, $500 beats the index’s median decade ($1,329) only if Nike earns its 2022 record again.
Through inflation, not directly.5 of 8 oil spikes to a 5-year high since 1970 that came with a falling S&P 500; 0 of 5 under it
US crude rose 50% or more in a year 13 times since 1970 and inflation rose every time. Stocks fell only when oil reached a 5-year high, and commodity producers are now 3.7% of the US market, against 26.6% in 1980.
Close to a coin flip.12 of 21 US buybacks of $50B or more that beat the S&P 500 over the next 12 months
Across every US buyback authorization of $50 billion or more since 2018, the stock beat the S&P 500 a year later 12 times out of 21, by a median 2.6 points. Nvidia’s record $150 billion is 2.8% of its value, 19th of 25 by size.
Cash, not bonds.7 of 7 yield spikes that sank stocks where T-bills still rose; 10-year Treasuries rose in none
In the 7 yield spikes since 1963 when stocks fell, a 60/40 with 10-year Treasuries lost almost as much as stocks alone, a median 6.6% against 7.4%. Holding T-bills instead cut the loss to 2.5%.
Only with inflation.7 of 20 yield spikes since 1963 came with a falling S&P 500, all with inflation at 3%+
A 1-point jump in the 10-year yield came with stocks rising in every low-inflation episode and falling in half of the high-inflation ones. The yield alone tells you little; the inflation behind it decides.
Often not.88.6% of VOO and VTI is the same money, 435 pairs measured
Measured on every holding, VOO and VTI are nearly one fund and QQQ is half of VOO; SCHD and VOO share under 8%. Calculators that read only the top ten holdings miss most of the overlap.
Most of it.85.5¢ median drop at the open per $1 of dividend, 14,012 ex-dates
Prices fall by most of the dividend, not all of it, and by more of it as the payout grows: 67 cents per $1 on the smallest dividends, 97 cents on the largest. Most are back above the old close within a session or two.
A coin flip.47.3% of 1,250 beats closed lower the next session
Companies beat the consensus five times in six, so a beat is priced in. Below a 5% beat the stock fell more often than it rose; a 1% beat did almost as badly as a miss.
No edge.0 of 3 models that beat “no change”, 429 forecasts
Ridge, random forest and gradient boosting all forecast the 10-year yield worse than assuming it stays put. What protects a bond portfolio is shorter maturities, not a prediction.
No edge.−10.6% weather model, 20 months out of sample
Arabica moves on weather forecasts, days before the thermometer confirms anything — and the stocks that buy the beans feel it a year later, through margins.
Research, not recommendations. Not investment advice.
> analytics?Visits and masked replays, no ads. Visit counts (Google Analytics) and screen replays with every figure hidden (PostHog). No ads, no data sold. Why
