The Risk Museum: I test trading-strategy ideas the way a quant desk would, on an engine I built, and I publish the ones that fail my gates. New here? Start at the entrance →
Research Notes
short essays from Vasilios Dimopoulos
Each note is a case study in what I tested, retired, and learned, written to be read cold by a practitioner, with every number traceable to the same gated code that runs the rest of the museum.
The Caravan and the Cluster
V. Dimopoulos · Working Paper No. 18 · Jul 28, 2026 · a banker who never left Florence wrote the best guide to the Silk Road in 1340. Read as an underwriting manual it maps seam for seam onto the AI buildout, and six of its lessons were already priced in my forecast book: the relay splits the markup, the license beats the cargo, and the fine print decides who eats the loss. The camels never made anyone rich. The paperwork did.
Where the Edge Was
V. Dimopoulos · Working Paper No. 17 · Jul 27, 2026 · the 620 World Cup forecasts of note #5, cut by category. Rescored against a benchmark that is actually written down, the two books that lost most ground to the crowd turn positive, so the deficit was the competition and not the forecasts. 42.2% of questions resolved yes, and hedging to 50% left that on the table. Withdraws three claims the earlier paper made, including its own baseline.
Lazy Prices Meets Survivorship
V. Dimopoulos · Working Paper No. 15 · Jul 26, 2026 · 14,155 EDGAR filings mined for the disclosure-text anomaly (when a firm rewrites its footnotes, it underperforms): it does not replicate in a survivor-only sample. It inverts (|t|=2.55, wrong sign) and falls below the bar because survivorship deletes the informative short leg. A published edge, disqualified by the data I have.
Priced Wrong on Purpose
V. Dimopoulos · Working Paper No. 16 · Jul 26, 2026 · 250,858 resolved Kalshi markets: contracts priced 0.30–0.40 resolve YES 17% of the time and 0.60–0.70 resolve 88%. Pre-registered, and two of my three predictions were falsified, one of them in sign.
Measuring the Short Side: crowding as risk, not signal
V. Dimopoulos · Working Paper No. 12 · Jul 26, 2026 · a fresh FINRA daily short-volume panel (2,006 days, 406 names) through the gates: zero of four cross-sectional signals pass, but the most-shorted names diverge +220 bps in the nine worst months (t=5.34). This is a survivorship-biased squeeze risk, not an edge.
The Factor Zoo Meets an Honest Bar
V. Dimopoulos · Working Paper No. 13 · Jul 26, 2026 · all 158 of Qlib’s Alpha158 technical alphas as one family: 29 look significant, zero survive the deflated (t≥4.11) and cost-stressed bar. The zoo is short-term reversal in disguise.
The Uncorrelated Book
V. Dimopoulos · Working Paper No. 14 · Jul 26, 2026 · what diversification actually buys a multi-manager platform: ten desks whose 18.0% average standalone volatility collapses to a 7.7% book (a 2.35× diversification ratio), with betting-against-beta −0.66 to beta as the built-in hedge.
Estimate-Revision Momentum: a view against consensus
V. Dimopoulos · Research View · Jul 26, 2026 · the most crowded published anomaly, and why I would trade the disagreement and the forced flow around revisions rather than the revision sign, stated as a view and not a backtest because the data to test it at this bar is not free.
Reading a Book: a risk teardown
V. Dimopoulos · Research note #11 · Jul 25, 2026 · a hypothetical long/short book, taken apart the way a risk desk does it, a hidden +0.56 market beta under +30% net, a −0.50 growth tilt it never intended, one name driving 41% of the variance, then VaR, stress, and the fixes.
The Light Was Green
V. Dimopoulos · Research note #10 · Jul 25, 2026 · three standard 1-day VaR models on 21 years of the S&P, all under-covered the tail (Kupiec fail), yet Basel’s traffic light glowed green. A model-validation case study.
The AI-CDO Is Already Being Built
V. Dimopoulos · Research note #9 · Jul 24, 2026 · structured finance reassembles around AI infrastructure, the four failures of 2008, each observable in the AI-credit complex today.
What the Swarm Knows and What It Does Not Know
V. Dimopoulos · Research note #8 · Jul 23, 2026 · the swarm’s provable blind spots, the regions it structurally cannot cover, mapped and published on purpose.
The Edge Was Never the Hard Part
V. Dimopoulos · Research note #7 · Jul 23, 2026 · 61 finance-trained players, a coin guaranteed to win, 28% went broke. Kelly, volatility drag, and how the museum sizes.
Measuring the Crowd
V. Dimopoulos · Research note #6 · Jul 23, 2026 · the 13F crowding study: two real effects (t=4.1, t=3.9) that still didn’t clear the bar to become a tradable signal. Crowding is a risk measure, not an edge.
620 Probabilities
V. Dimopoulos · Research note #5 · Jul 20, 2026 · top 6% in Jump Trading’s Brier-scored World Cup Probability Cup, with no hero calls, and the calibration flaws published on purpose.
How to Read a PM’s Book
V. Dimopoulos · Research note #4 · Jul 17, 2026 · factor dissection (intended vs unintended) and the drawdown protocol, what to do when a PM is losing money.
“I tightened my own gates. The gates took my model first.”
V. Dimopoulos · Research note #3 · Jul 16, 2026 · the Dimopoulos Chain (a dealer-positioning strategy), v1→v3, a case study in iterating a research process.
The Propulsion Room
The BlueShip Research Engine · Jul 13, 2026 · flight mechanics for portfolios, thrust, drag, gravity, and the price of steering, with the pipeline’s real friction ratios.
“I brought a neural network. The 1993 model won again.”
V. Dimopoulos · Research note #2 · Jul 13, 2026 · a neural network priced the same bonds at daily frequency and still lost to a 1993 model, more capacity is not more information.
“I named a model after myself. A 1993 model beat it.”
V. Dimopoulos · Research note #1 · Jul 12, 2026 · The Dimopoulos Chain, v1, dealer positioning tested against a ladder of simple econometric baselines it had to beat.