Aktiesparekonto, started 20.07.2026, day 119. Numbers from the account every 15 minutes.
Strategy and status
In short (11.11.2026 12:03 UTC): the account is in the index and stays there. Eight tests today, every rule written before its run. Failed: buying after a directed share issue, the theme question on new jumps, orders and raised figures without a jump, buying a theme fund once it runs, and a sleeve that always holds the strongest theme funds (4 points a year behind the world index). Held as a filter: the compound rule's questions sort jumps on 3.177 new cases, but since 2021 the good ones only match the index. One lead with support: the questions work in the least traded shares and not in the big ones, in Sweden and, on the cases there are, in Denmark and Finland. That is now a second paper rule, small shares only, from tonight. Nothing found reaches your bar of 1 point a month; no theme or sector fund did either, even picked with hindsight.
16.11.2026 08:05 UTC: the whole account is in the index. R3 done: 140 IUSN sold at 9,25 EUR, 99 WEBN bought at 13,09 EUR (Saxo fills), 220 DKK cash left.
14.11.2026 10:12 UTC: the index only. The research now looks for one niche where the swings are big for us and too small for big players, and goes deep there.
What is in force
Default (your word, 14.11.2026): the index only. Everything not earned by a tested edge sits in WEBN (655 units since 16.11.2026).
The goal
Beat ACWI after all costs, with a model improved in versions (v1, v2 ...). A version gets real money only by the ladder 5% → 10% → 25% → more, each step on a forward record.
Target (10.11.2026): a separate account in your company, about 500.000 DKK, once a model seems to work. This account is the test. The model is built and paper-traded at that size and with that account's costs; the direction is small Nordic companies with slow signals (insiders, steady buyers) and warnings as exits.
What the research has shown
When to buy (07-09.11): no edge. Tested against random days for the same shares, after costs:
- Volume spikes, breakouts, sell-offs, HN mentions, insider buying, partner filings, Wikipedia spikes and build-ups, technology waves: nothing.
- News is priced within 5 minutes. Attention is followed by flat or worse returns. Bad news keeps falling (a sell rule for anything we hold).
Which shares to hold (10.11, first results):
- Momentum funds are the only factor funds ahead of ACWI after their costs: about +2 points a year since 10.2014 (Saxo prices to 09.11.2026). They are allowed in the account. The price: long stretches behind, 5,5 years from 2020, and -16% against ACWI since June 2026. After publication such premia shrink by about half: expect +0,5 a year on the part held, ~50 DKK a year at 15%, less than the switch costs.
- Our small-cap fund (IUSN, 15% of the account) has trailed ACWI by 2,9 points a year since 04.2018. It is an active bet nobody tested.
- Quiet beats loud. Shares that rose in many small steps kept rising; shares that rose in jumps did not: +13,3% against +1,6% a year (US, 2020-2026, ACWI +10,5%). Your "interest building", measured in the price. Our data is only 5,5 years, but the effect is published and holds in 49 markets, 1993-2020.
- Our bars were too weak to judge anything. A signal needs 70-200 scored calls for a verdict; the HN theme rule gives 4-5 a year. From now a forward test only counts if it fires 100+ times a year.
The niche, first tests (10.11.2026, Sweden, 574 shares, 2005-2026):
- Against the world index it is not enough (measured 10.11). In SEK, 2016-2026, the small-share winners made 10,4% a year after costs; the world index made 13,6%. They were far ahead in 2016-2021 (28% a year against 15%) and far behind since 2022 (-8% against +12%), with swings twice the index's. The edge is over other small Swedish shares, and that whole group trailed the world. By our own goal nothing found so far beats the index.
- Momentum works in small shares, not in large ones. In shares trading 0,5-5 million SEK a day, last year's winners beat their group by 0,65% a month in 2016-2026 (about 8 points a year before costs, about 2 points of costs). In large shares: nothing.
- Insider purchases: no edge after costs, counted from the day they are published.
- New short positions: not a warning. Once published, nothing is left in them.
- Good reports: a 20-day drift of about 2,5% in small shares in both halves of the data, but it failed the 60-day bar I set, so it is a hypothesis.
- The Nordic check (10.11): momentum shows in Finland's small shares too (about 0,65% a month on the longer sample) and in the larger shares of Denmark, Norway and Finland (0,4-0,6% a month). Denmark and Norway have too few small shares that trade enough, so the model's universe is mostly Swedish plus about 30 Finnish names.
- Only shares listed today are in the prices, which flatters levels. Each market alone is weak evidence; together with the published research it is enough to build a paper portfolio on.
Big moves: what we know so far (10.11.2026 13:50 UTC)
Sweden, 07.2016-08.2026, Saxo prices, exploratory.
- A jump is an alarm, not a signal. After a jump of 20%+ in a week with heavy turnover, a share doubles within 60 trading days 6,8% of the time and halves 8,1% (any share, any day: 1,3% and 2,1%).
- One thing price does tell: stay away from jumps in broken shares. A jump in a share 10% or more under its one-year high loses 6,8 points to the world index over the next 60 days (median -11,5%; holds in both halves of the data).
- Jumps near a one-year high carry all the big winners, and most still lose. Mean +5,5% against the world index, median -2,6%. A few very large winners make the mean (Sivers Semiconductors, +2.264% in 60 days from March 2026).
- 24 random cases read one by one, in two rounds (10.11.2026 14:20 UTC). I read only what was public on the day of the jump, wrote my call down, then looked. On the 21 usable cases my call was right 18 times. A fixed six-point checklist, tried in round 2, got 7 of 10: it caught every halver and missed three of five doublers.
- What halves a share after a jump is mostly a share issue. 5 of 10 halvers fell on an issue announced or delivered within weeks; all five were burning cash with little income.
- Doublers had a theme or a regulator's yes, and money in the bank. AI optics, bitcoin, space and defence, obesity. They sold shares too, but after the run, and that is often the top.
- Jumps over 150% in a day were given back within days.
- Regulator and trial dates are coins, and some doublers had no reason I could name on the day.
- It is not only Sweden. The same alarm lifts both tails in Denmark, Finland and Norway. Our price data there is thin (about 90, 120 and 70 shares against 535), so widening it is on the list.
- How often: in Sweden about 3 new jumps a week, one of them near a one-year high.
- Tested on a random sample of all jumps (10.11.2026 17:33 UTC; your idea). 40 of the 459 Swedish jumps since 01.2024, drawn whatever happened next, scored before I looked. Return over 60 trading days against the world index, bought at the next close:
- As a buy rule it fails. Most jumps are a good report in an ordinary company. Those neither double nor halve; they drift a little behind the index.
- As a warning it works. The pass group lost 14 points and held every halver. On all 457 jumps I could check, a jump with no company news behind it lost 8,8 points to the index (median -12,5%), in each of the three years.
- Both doublers were in the pass group. The big winners come out of the same broken or unfunded shares as the halvers.
- The one jump that had everything, a theme included, was the best buy: MilDef, February 2025, +59% against the index. One case, so a lead, not a result.
- Stop losses tested (10.11.2026 17:37 UTC; your question). On all 1.573 Swedish jumps since 2016 a stop does not add return: hold 60 days +0,8% against the world index, a 10% stop +1,1%, a 20% stop +0,6%. A 10% stop sells 64% of the positions and 19 of the 78 doublers before they double. Trailing stops sell the giants. A stop is insurance (fewer disasters), not an edge; it helps most in the jumps we should not buy at all.
- Can the doublers be picked out some other way? Not from prices (10.11.2026 18:05 UTC; your question). On all 1.573 jumps, 5,0% doubled and 5,5% halved. Everything that makes a doubling more likely makes a halving at least as likely: the doublers and the halvers are the same kind of share (small, cheap, volatile, under its high). A model trained on earlier years picked a group that doubled 8,9% of the time and halved 13,8%, and lost 7 points to the index. Waiting for the price to confirm did not help either.
- The cases are an illustration and some were stories I half remembered. The clean test is forward: every new jump gets the checklist on the day and is scored later. No orders.
What I am watching, and why
Forward data recorded from 09.11 by timers (backup nightly):
- Avanza owner counts (2.666 shares) and Reddit mentions: ordinary investors moving in. Scored both ways from day one: as a buy (H1) and as a warning (H1b), because the published evidence points down (Robinhood: -4,7% in 20 days after the heaviest inflows) and nobody has tested Nordic owner counts. First signals mid-December; no verdict before 100 signals.
- The Swedish short register (from 10.11, daily; history from 2010): who is betting against which company. Today 336 companies are shorted, by 74 funds led by Citadel, JPMorgan, Two Sigma and D. E. Shaw (Finansinspektionen, 10.11.2026). Two tests with rules written first: do the shorts rise when Avanza owners pile in, and does a new short position warn of a fall.
- HN front page, specialist tech press, Nordic announcements, every 2 hours.
- GitHub (40 technologies), Wikipedia (3.600 companies, 421 technologies), app charts, Polymarket, Hugging Face, arXiv, daily.
The crowd gauge (10.11.2026 19:33 UTC)
How much Avanza's customers trade (trades a day per 1.000 customers, from Avanza's monthly statistics), against its own three-year median.
- Today: OFF. October 2026: 68, median 80. It was ON through 2020-2021 (143 and 155), OFF in 2022-2024, ON for most of 2025.
- With the gauge ON, jumps in shares at their highs beat the world index by 14% in 60 days in 2016-2020 and by 1% in 2021-2026. With it OFF: +3% and -7%.
- As a switch on the simulated account it kept most of the gain (+13,1 points a year against +14,9), with half the trades and half the worst year (-10,5 against -22,5).
- One boom in the data, so a lead, not proof. It is now read every month and shown here.
- Checked on our neighbours (11.11.2026 08:40 UTC): not confirmed. With small companies added for Denmark and Finland (46 Finnish ones from Yahoo, 10 Danish from Saxo), jumps there did no better with the Swedish crowd in than out. In Sweden the effect is real in both halves and strongest in small shares (under 1 million SEK a day: +20% and +6% with the crowd in, +5% and -8% with it out). So it is a Swedish finding so far. I also gave each country its own crowd figure from Nordnet (11.11.2026 08:50 UTC): still not confirmed, on thin data. Three tries abroad, none works; I leave the neighbours until the paper record shows the Swedish gauge is real.
- On new jumps the gauge is weaker (11.11.2026 10:17 UTC). On 3.177 smaller jumps it made 4 points of difference in 2016-2020 and under 1 point since 2021. It did sort the directed issues (+4% with the crowd in, -8% with it out). Still a lead, weaker than it looked yesterday.
- Second witness for today's reading: Nordnet's customers also trade less than usual (98 trades a day per 1.000 customers in October against a three-year median of 110).
- Also tested today: seven gauges from prices alone (none works as a switch), buying on the second piece of news (no), theme waves among Swedish companies (no; a jump after its peers jumped is late). A rights issue after a jump is a sell (-12% in the next 60 days).
The compound rule, on paper from 11.11.2026 (10.11.2026 20:09 UTC)
Your idea: stack several things. Each new jump gets four yes-or-no questions:
- Is the crowd in? (the gauge above)
- Is it a sound jump? Real company news behind it, no share issue, the share at its high.
- Is there a booster? An order, raised figures, or an insider who bought.
- Is its theme running? A big Hacker News story on the theme in the last 45 days.
What jumps did on history by the number of yes answers to the first three (60 trading days against the world index, 2016-2026):
- The more yes answers, the better, in both halves of the ten years. It avoids the losers; it does not find the doublers.
- The theme question now has proper data (11.11.2026 09:13 UTC). With fresh prices for 80 theme leaders, European ones included, a jump in a company whose theme is running did about 5 points better than other jumps with the same score (100 jumps; +11 points in 2020-2022, +2 in 2023-2026). It caught the defence waves of 2022 and 2025 and this year's two giants, Sivers and Unibap. It still does not find doublers reliably: 7% doubled, 6% halved.
- Tested on jumps the rule had never seen (11.11.2026 10:17 UTC). 3.177 smaller jumps (10% in a week on heavy turnover, about 6 a week), 60 trading days against the world index, before costs of about 2%:
- The questions sort. About 5 points between the bottom and the top, in both halves. That part is real, not fitted.
- The level is not there. Since 2021 the top group only matches the index. Small Swedish shares as a group have trailed the world index since 2021, and the rule does not lift them over it.
- Orders and raised figures are the steady question (5 points in both halves). The crowd question is weaker here.
- The theme question failed on those new jumps (11.11.2026 10:19 UTC). 62 smaller jumps in a running theme lost 6,6% to the index (median -12%). On the big jumps it was found on, a tighter way to tag themes (from the company's own announcements) looks better still, but that is a handful of giants like Sivers. So the theme question is unproven; the paper record will show.
- Buying after a directed share issue: no (11.11.2026 10:17 UTC). Yesterday's lead (+12% on 42 cases) was a small sample's luck: on 104 new cases the share lost 5 points to the index. The sell rule holds: after a rights issue that follows a jump the share lost 15 points in 60 days (71 new cases).
- A lead with support: the questions work where little money can go (11.11.2026 12:00 UTC). 60 trading days against the world index, before costs:
- Jumps without the yes answers in the same small shares lost 6 to 11% since 2021. None of the 198 Swedish cases with the yes answers halved.
- I found the pattern after the fact in Sweden, then tested it on Denmark and Finland with the rule written first. There the half that can be judged agrees; the other half has too few cases. So: support, not proof.
- The typical case is near zero in Sweden (a few winners carry the average) and a trade in such a share costs about 4%.
- It is now paper rule version 2 (small shares only), fixed before its first trade; about 20 buys a year on history, fewer while the crowd gauge is off. It fits this account's size; a 500.000 DKK account could only trade the upper end of these shares.
- Orders and raised figures without a jump: not a signal (11.11.2026 10:32 UTC). 558 cases: +4% against the index in 2016-2020, -5% since 2021.
- Themes running today: vaccines and obesity drugs. Quantum, space, nuclear and hydrogen leaders are down 37-40% over three months.
- This is history the rule was built on, so it flatters. From 11.11.2026 every new jump is scored and bought on paper: 5%, 10% or 15% of a paper account for 2, 3 or 4 yes answers. The record is in
research/PAPER.md. - More triggers scanned (10.11.2026 20:15 UTC): fifteen more ideas tested on whether they add to the three questions above. One passed: an insider buying for 1 million SEK or more in the two months before the jump (+3 points, in both halves). Close: an all-time high, the company buying another business, last year's winner. No use: how the jump day closed, short sellers, big against small shares. Nearly every trigger is better while the crowd is in, so the crowd gauge is the master switch.
- Real money is unchanged: only a theme call under R2, with your yes.
Theme funds as the trade (11.11.2026 11:27 UTC)
Your plan is the index plus big opportunities. A big theme can be bought in this account as a fund. I took every theme and sector fund on the list of Xetra, the German exchange, that Saxo offers: 282 funds (chips, defence, banks, uranium, robotics and so on), prices from 2012, against the world index.
- Buying a theme fund once it runs is a coin flip. 274 times a theme started running (up 20% in three months, at its high). Over the next three months the fund beat the world index by 0,1% on average; 48% were ahead. Over six months 12% ended 20 points or more ahead and 7% as far behind.
- Always holding the three strongest theme funds: 4 points a year behind the index (+7,7% a year against +11,7%, 05.2013-10.2026). Behind in both halves, with swings from +24% to -29% in a year. At this account's trading costs: 9,5 behind.
- Even with hindsight no fund reached your bar. Of the 55 theme and sector funds that have existed since 2013, 6 beat the world index, one of them by more than 3 points a year (world technology, +9,6), none by 12. The typical one trailed by 4 a year. Since 2020 the best were US technology and euro banks, about +11 a year each.
- So the moves that beat the index by 1 point a month were in single shares, not in themes as a fund holds them. And single shares are where the tests say they cannot be picked on the day.
Removing the losers: screened index funds (14.11.2026 09:58 UTC)
You asked whether research that removes the many losing companies could beat the index a little. I tested the version anyone can buy: every world fund at Saxo that screens out losers (quality, value, momentum, multifactor, low volatility), against the plain MSCI World fund, as held, after fees, in EUR. The rules were fixed before any price was fetched.
- 15 screened funds with 5 years or more of history: the typical one trailed the plain index by 1,1 points a year. 4 were ahead at all. Quality was level (0,0 and -0,1 a year since 2014-2015). Momentum was +1,1 to +1,4, but only in the first half. Low volatility was about -3,3.
- One passed the switch bar (+1 a year, ahead in both halves): iShares World Value Factor Advanced, +3,8 a year, but over exactly 5 years. Its two older siblings with the same idea trailed by about 1 a year over 11-12 years. With 15 tries one pass is what luck gives.
- Small caps: S&P 600, which admits only profitable companies, was level with the plain Russell 2000 (-0,3 a year since 2015). IUSN itself trailed the world index by 3,0 a year since 2018. Avantis Global Small Cap Value is +4,2 a year ahead of IUSN, but over only 2 years.
- So the index already does most of the removing, because a falling company shrinks in it by itself. What is left for a screen is small and comes and goes. Nothing came near your bar of 12 a year.
Research log, last runs (UTC)
- 16.11 18:20 nightly scans, Swedish small shares to today's close: 1 new alarm (Hansa Biopharma) and 1 small-share trigger (BIMobject), neither with any layer, so no paper buy.
- 15.11 08:10 construction briefs started: price facts for 13 listed Danish and Swedish names (1-year moves from -40% Lindab to +91% Bravida, to Friday 13.11 close, Saxo). Briefs and your call log next.
- 14.11 14:16 first scans since the pause, Swedish small shares to Friday's close: 4 new alarms and 2 small-share triggers, none with 2 layers, so no paper buy.
- 14.11 09:58 screened index funds (quality, value, momentum and others) against the plain index, on your go: no better on average (-1,1 a year); one pass of 15, over only 5 years.
- 11.11 12:02 second paper rule fixed (small shares only) and its nightly scan written; first triggers tonight.
- 11.11 12:00 the lead tested on 1.744 Danish and Finnish jumps: too few cases in 2016-2020 to judge; since 2021 it holds (+10% on 33 jumps in small shares, nothing in big shares).
- 11.11 12:00 Danish and Finnish announcement headlines fetched (the archive now holds 454.604 headlines).
- 11.11 11:29 jump results split by how much the share trades: the questions work in the least traded shares, not in the big ones (a lead, read after the fact).
- 11.11 11:27 a sleeve always holding the three strongest theme funds: 4 points a year behind the world index, in both halves.
- 11.11 11:27 buying a theme fund once it runs (274 cases, 282 funds): a coin flip against the world index.
- 11.11 11:18 catalogue of the 1.130 funds on Xetra that Saxo offers; prices fetched for the 282 theme and sector funds.
- 11.11 10:32 orders and raised figures on days without a jump (558 cases): not a signal by themselves.
- 11.11 10:19 theme question tested on 2.241 smaller jumps it had never seen: failed (a running theme took 4 points off).
- 11.11 10:18 themes re-tagged from each company's own announcements: better on the big jumps (47 cases), in sample only.
- 11.11 10:17 compound questions tested on 3.177 smaller jumps: they sort (5 points, both halves), but the best group only matches the index since 2021.
- 11.11 10:17 directed share issue after a jump tested on 104 new cases: not a buy. Rights issue as a sell: confirmed on 311 cases.
- 11.11 10:17 fetched every announcement headline of Nasdaq Stockholm and First North Sweden since 2016 (255.188 headlines); collected daily from today.
- 11.11 09:13 fresh prices fetched from Saxo for 80 theme leaders (US and European); theme test rerun: the theme layer now adds about 5 points; the paper rule uses it from tonight.
- 11.11 09:05 theme-leader prices made current (US gap 30.03-02.11 filled) and 30 European leaders added, now collected daily. Today only vaccines reads hot; most themes sit 15-70% under their one-year highs.
Where this might land (10.11.2026 18:55 UTC)
A simulation on 2016-2026: the account in the world index, with 5% positions in the jumps that pass today's leads (order news in a share at its high, a trading update, insider buying), 60 days each, after costs.
- On history: +5,8 points a year against the index. That is a ceiling: most of the rule was chosen with the answers visible.
- Almost all of it came from the small-share boom of 2019-2021 (+18,5 a year). The other eight years: +0,6 a year. Since 2022: +0,3. Worst year -9,8.
- My estimate going forward: about +1 point a year against the index, somewhere between -2 and +4. Level with the index in ordinary years, well ahead in a boom, about 4 chances in 10 to be behind after three years.
- Your bar is 1 point a month (10.11.2026 19:00 UTC). On history that shows up only with big positions in a boom: 15% positions gave +14,9 a year, made of +52 a year in 2019-2021 and +2 a year otherwise, with a worst year of -22. No signal we have delivers it in ordinary years.
- After today's eight tests (11.11.2026 12:03 UTC): the rule's questions are real as a filter and seem to carry an edge only in the least traded shares: on history, after 4% costs, about +3% a trade since 2021 and +10% in 2016-2020. At about 20 trades a year and 5% positions that is 3 points a year on the account on recent history; I would halve it for luck. For everything else the level against the world index is about zero since 2021. Your bar is not in sight, and theme funds do not offer it either.
- Shorting the losers was tested and loses money in every version (-5 to -19 points a year): a short earns the share's own fall, not its lag to a rising index, and the few that double cost more than all the rest earn. This account cannot short anyway.
- Paid data (Börsdata, EUR 59 a month) could add about half a point. One month is enough for the research. Its terms are for private use only, so it cannot run the company account. Open ask A12.
Next
No fixed date (your word, 10.11: research first). The attack list (11.11.2026 12:03 UTC):
- Tonight: the first scored scan of both paper rules, after the Stockholm close.
- Denmark and Finland every day. Next I would set up the daily scan for Danish and Finnish small shares, so the small-share rule covers them, and read the 48 Danish and Finnish cases one by one.
- Avanza owner inflow per share, once a few weeks of our own data exist.
Closed today: the directed issue as a buy, orders and raised figures without a jump, theme funds bought on strength. On hold: paid data (ask A12), the crowd gauge abroad, shorting. The 15% in small caps stays, and moves into the index only if the research ends without a model (your word, 10.11).
Last decision
14.11: IUSN into the index and the index only from now on; the research hunts for one niche where big swings are too small for big players (your idea). IUSN trailed the world index by 3,0 a year since 2018 (test T30).
09.11: account moved into the index (sold Equinor, Danske Bank, BNP Paribas; bought 198 WEBN). Next decision: none dated; a proposal comes when the research supports one.
Written by the trading agent, last changed 16.11.2026 18:20 UTC.
Load is the 1 / 5 / 15-minute average; on this box anything under the core count is comfortable. Measured 2026-11-17 17:30 UTC.