Research
Quant research for funds, desks and allocators, across listed equities and crypto markets, fundamentals and price structure.
Every figure is traced to its source, every measure to the paper that defines it, and every signal is counted against the search that found it. What reaches a client is the part that survives.
- desks
- 5
- measures, each cited
- 27
- asset classes
- 2
The desks
Five desks, one standard. Choose a desk to see what it measures, what each measure rests on and what a client receives.
Equity fundamentals
Every figure traced to the filing it came from.
Company and sector research built from primary filings. A retriever extracts each figure from SEC EDGAR with the document beside it, an editor writes only from those typed facts, and a checker blocks anything that cites nothing.
- Earnings qualityOperating and free cash flow against net income across periods, the balance sheet accrual ratio, DSO, DIO and DPO against revenue growthEDGAR financial statements
- Non-GAAP gapThe distance between reported and adjusted earnings, and the nature of every add-backEDGAR filings and exhibits
- Beneish M-ScoreThe eight-variable earnings manipulation screen, with every input shownBeneish, 1999
- Insider purchasesOpportunistic Form 4 buys separated from routine ones by each insider's own historyCohen, Malloy and Pomorski, 2012
- Catalyst calendarEarnings, guidance changes, M&A, rating actions, litigation and index events, about thirty classesEDGAR full text and event data
- Company deep dive
- Earnings quality screen
- Sector primer
- Market entry study
- Thirty-day company brief
How a piece of research is made
Three roles with a hard line between them, so a sentence can never say more than the evidence under it.
- 01
Retrieve
Sources are discovered, deduplicated and read. Every figure is extracted as a typed fact with the document and the place in it, and the retriever writes no prose at all.
- 02
Write
The editor turns typed facts into sentences against a fixed template, and has no tools: it cannot look anything up, so it cannot cite what was never retrieved.
- 03
Check
Every citation is resolved back to its source, the wording is checked against the compliance rules, and a failure blocks delivery rather than flagging it.
The measurement, in figures
What the house method does to a number that looks like skill. Computed by the same code that produces every client report.
Twenty variants of a strategy with no edge, the family an afternoon of tuning produces. The best one looks like skill in isolation. Measured against what the best of twenty shows by chance, most of that confidence goes. The same sample loads in one click on the free check.
- 0.9004Deflated Sharpe on the published worked exampleReproduced to four places. Bailey and López de Prado, 2014
- 12,870Overfitting splits measured in about one secondCombinatorially symmetric cross validation over a thousand trials
- 1.34Best Sharpe among 1,000 strategies with no edgeDeflated to a probability of skill of 0.36. The headline was luck
- 48,000Seeded seasons behind every scoring formulaTwelve populations of four thousand, on the public bench
- 2Independent protocols behind every on-chain closeThe median of both, refused when they disagree
- 9Invariants the ranking holds under attackFour adversarial reviews, every attack encoded and rerun
Coverage index
Every measure on every desk, with the basis it rests on.
| Desk | Measure | What it measures | Basis |
|---|---|---|---|
| EQ | Earnings quality | Operating and free cash flow against net income across periods, the balance sheet accrual ratio, DSO, DIO and DPO against revenue growth | EDGAR financial statements |
| Non-GAAP gap | The distance between reported and adjusted earnings, and the nature of every add-back | EDGAR filings and exhibits | |
| Beneish M-Score | The eight-variable earnings manipulation screen, with every input shown | Beneish, 1999 | |
| Insider purchases | Opportunistic Form 4 buys separated from routine ones by each insider's own history | Cohen, Malloy and Pomorski, 2012 | |
| Catalyst calendar | Earnings, guidance changes, M&A, rating actions, litigation and index events, about thirty classes | EDGAR full text and event data | |
| OC | Consensus close | An hour-long pool TWAP, the median of two protocols, refused when they disagree | Base pools, published method |
| Loss versus rebalancing | What liquidity providers lose to arbitrage against a rebalancing benchmark, and whether fees cover it | Milionis, Moallemi, Roughgarden and Zhang, 2022 | |
| Toxic flow share | The share of swap volume that is arbitrage against the reference price | Base swap events | |
| Price impact and spread | Exact impact of a swap of a given size from tick liquidity, and the effective spread paid | Uniswap v3 mechanics, Hasbrouck, 2009 | |
| Kyle's lambda and Amihud | Price change per unit of signed volume, and return per unit of volume, per pool | Kyle, 1985. Amihud, 2002 | |
| Extracted value | Sandwiches and back-runs detected per block, with the value taken | Qin, Zhou and Gervais, 2022 | |
| MR | Term spread | Ten-year less three-month, as a recession probability | Estrella and Mishkin, 1998 |
| Excess bond premium | Credit spreads net of expected default, a credit conditions predictor | Gilchrist and Zakrajšek, 2012 | |
| Sahm rule and NFCI | Recession onset on real-time unemployment, and weekly financial conditions | Sahm, 2019. Chicago Fed | |
| Hedging pressure | Commercial hedger positioning across commodity futures | de Roon, Nijman and Veld, 2000 | |
| Regime models | Markov switching and statistical jump models, filtered rather than smoothed | Hamilton, 1989. Shu, Yu and Mulvey | |
| Turbulence and absorption | How unusual returns are, and how fragile the system is | Kritzman and Li, 2010. Kritzman et al., 2011 | |
| TA | Time series momentum | Trend across horizons, scaled by volatility | Moskowitz, Ooi and Pedersen, 2012 |
| Realised volatility regimes | Volatility clustering and the state it implies for position sizing | GARCH family | |
| Structural breaks | CUSUM and Bai-Perron tests for the date a series changed behaviour | Bai and Perron, 1998 | |
| Liquidity structure | Where resting liquidity sits around the price, read from pool ticks on chain | Base pool state | |
| Rule deflation | Every indicator variant counted, and the best one deflated for the search | Bailey and López de Prado, 2014 | |
| VA | Deflated Sharpe | The probability of skill against what the best of N variants shows by chance | Bailey and López de Prado, 2014 |
| Probabilistic Sharpe | Confidence in the Sharpe given length, skew and fat tails | Bailey and López de Prado, 2012 | |
| Minimum backtest length | How many years a history needs before its Sharpe can be told from luck | Bailey, Borwein, López de Prado and Zhu, 2014 | |
| Overfitting probability | How often the in-sample winner falls below the median out of sample | Combinatorially symmetric cross validation | |
| Haircut Sharpe | The Sharpe left under the strictest multiple testing correction | Harvey and Liu, 2015 |
Bring the market and the question. The scope, the window and the price are agreed in writing before work starts.
Research is analysis published for professional readers, never a personal recommendation to buy or sell an instrument.