Skip to curriculum
Self-paced quant finance course

Curriculum

One structured quant finance course, from readiness primers in maths and Python through options pricing, portfolio construction, risk and trading. Every lesson lists its prerequisites, works the derivation, shows the Python and ends with an exercise and full solution. Start free with one topic in each primer.

Start free, no cardSee membership plans

Phase 4.1Quick Wins

European Option Price and Greeks Calculator01 / Project
Description
What is the model price of a European option, and what moves it?
Status
Completed
Data used
  • Illustrative option inputs, labelled as such
  • Deribit BTC option chain snapshot: 308 quotes from the public API, 27 Sep 2026
Output / result
An independent Black–Scholes pricer and Greeks calculator with an implied-vol solver. It prices the benchmark call at 5.4348 with parity holding to machine precision, then shows on real quotes that one expiry implies volatilities from 47.4% down to 37.0%: the constant-vol assumption fails.
Open the project →
Volatility Smile and Surface Explorer02 / Project
Description
How does implied volatility vary across strikes and expiries, and can a surface built from quotes be trusted to price options that are not quoted?
Status
Completed
Data used
  • Illustrative 5 × 3 volatility grids, including a smooth but arbitrageable one
  • Deribit BTC option chain snapshot, 27 Sep 2026
Output / result
A volatility surface tool with arbitrage checks. Smooth-looking illustrative smiles still fail nine checks; on real quotes 51 apparent mid-price violations turn out to be none executable at bid/ask, and constrained SVI and linear interpolation predict held-out strikes about equally well.
Open the project →
Bond Price and Duration Explorer03 / Project
Description
How does a yield change affect a bond, how good are the quick risk numbers, and does a DV01 hedge actually remove the risk?
Status
Completed
Data used
  • Illustrative bond terms
  • U.S. Treasury daily par yield curve, Jan–Sep 2026
Output / result
A bond price, duration, DV01 and convexity calculator. A DV01-matched 2-year hedge removes only 58% of a 10-year position’s daily P&L variance on real 2026 curve moves; the rest is almost entirely 10Y–2Y slope risk.
Open the project →
Portfolio Risk and Diversification Dashboard04 / Project
Description
Which holdings contribute to portfolio risk, and does a risk-minimising allocation deliver lower risk out of sample?
Status
Completed
Data used
  • Adjusted daily closes for 20 US stocks, 1989–2018 (survivorship caveat noted)
Output / result
A risk-contribution dashboard. An equal-weight sleeve runs at 13.2% volatility, but JPM carries 24.2% of the risk on 20% of the capital. A capped minimum-variance rebuild cut realised volatility from 21.0% to 17.9%.
Open the project →
Drawdown and Tail-Loss Monitor05 / Project
Description
How bad have losses on a US equity allocation been, and can the one-number daily risk estimate sent to the board be trusted on the days that matter?
Status
Completed
Data used
  • S&P 500 index daily prices, 1999–2018
Output / result
A drawdown, VaR and expected-shortfall monitor. The normal model understates the 99% tail by 1.47×, and a historical VaR that passes the Kupiec test still clusters its breaches: 12.7% the day after a breach instead of 5%.
Open the project →
Minimal Strategy Backtester06 / Project
Description
Does a 50/200-day moving-average rule on the S&P 500 really beat buy-and-hold once the accounting is done properly, and does the answer survive a different start date and an honest out-of-sample test?
Status
Completed
Data used
  • S&P 500 index daily prices, 1999–2018
  • Eight days of made-up prices for a hand-worked ledger
Output / result
A minimal backtester with next-open fills and costs. A 50/200 moving-average rule beats buy-and-hold on 2008–2018, but the whole lead comes from one signal six days before the test began. Started after 2009 it lags in 98% of start months: crash insurance, not a source of return.
Open the project →
Monte Carlo Pricing and Convergence Lab07 / Project
Description
Can the desk's Monte Carlo engine be trusted? When it prints a price with a 95% interval, is the interval honest? Is variance reduction safe? For a real exotic, how much of the uncertainty is numerical and how much is the model?
Status
Completed
Data used
  • Model experiments only for the base case (no market data)
  • Deribit BTC option chain snapshot for an illustrative barrier option
Output / result
A validated Monte Carlo engine. Standard errors fall as 1/√N, but small-sample error bars under-cover and antithetics can double variance. For the BTC barrier, monitoring frequency (USD 264) and volatility choice (USD 104) matter far more than the USD 10 simulation error.
Open the project →
Order-Book and Slippage Calculator08 / Project
Description
Before a bitcoin order is sent to Coinbase, what will it cost against the mid, what drives that cost (spread, depth or fee), and would splitting it over the next minute be cheaper?
Status
Completed
Data used
  • Coinbase BTC-USD level-2 order book: ten snapshots over one minute
  • Coinbase daily candles for realised volatility
  • Coinbase published taker fee tiers
Output / result
A pre-trade slippage calculator. A 0.5 BTC market order walks 12 levels for 0.66 bp of impact, while the taker fee is 4–60 bp: 6–90× the book cost. Splitting the order only helps the part that was already small.
Open the project →
Polymarket Bitcoin Probability Explorer09 / Project
Description
For a prediction-market contract such as "BTC above $90k at 08:00 UTC on 30 Oct 2026", what exactly is being priced, how does the price compare with a transparent model and with the BTC options market, and would the model's probabilities have been reliable in the past?
Status
Completed
Data used
  • Illustrative Polymarket contract and quotes
  • Deribit BTC option chain snapshot
  • Coinbase daily candles and order book
  • Coin Metrics BTC daily price history, 2010–2026
Output / result
Three estimates of "BTC above $90k on 30 October": a lognormal model (0.297), options-implied (0.265) and the market (0.305). None of the gaps is tradeable, and a 2015–2026 calibration study shows the model cannot be told apart from a naive baseline.
Open the project →

Phase 4.2Capstones

TR — Treasury curve risk and hedge validation10 / Project
Description
This project asks a rates risk manager's question: should a long-duration bond book be hedged with a parallel (10-year DV01-matched) hedge or with a constrained key-rate hedge, and how often should the hedge be reset? The book and hedge bonds are hypothetical, valued on the Federal Reserve's fitted GSW Treasury curve (daily, 1961–2026). The study replicates GSW (2007), Litterman–Scheinkman (1991) and Ho (1992). It then replays 94 monthly hedge reviews with daily revaluation, P&L attribution, an append-only decision ledger, stress scenarios and labelled injected failures.
Status
Completed
Data used
  • Federal Reserve Gürkaynak–Sack–Wright fitted Treasury curve, 1961–2026
  • FRED effective fed funds rate and 2- and 10-year constant-maturity yields
Output / result
A hedge validation for a USD 1.1bn bond book. A key-rate hedge in 2/5/10/30-year bonds, reset only when net DV01 drifts beyond the 5% mandate, cut daily residual P&L volatility to USD 876k against USD 1,003k for a parallel duration hedge over 2019–2026.
Open the project →
VR — Volatility forecasting and risk-limit calibration11 / Project
Description
Next-day variance forecasts for BTCUSDT, with ETHUSDT as external validation, built from Binance 1- and 5-minute bars (2022-01 to 2026-08). The forecasts feed a replayed daily risk-limit service for a hypothetical long BTC position: USD 1m 97.5% ES budget, USD 30m cap, costs assumed at 10 bp.
Status
Completed
Data used
  • Binance BTCUSDT one-minute bars, 2022–2026 (2.45m bars, checksum-verified)
  • Binance ETHUSDT five-minute bars for validation
Output / result
A next-day volatility forecast and exposure-limit rule. HARQ on 5-minute realised variance had the lowest forecast loss in the replay year, sized through filtered historical simulation. The report shows where it fails: jump days, multi-leg crashes, and a hysteresis band that left risk above budget on 190 of 365 days.
Open the project →
MV: Real-time macro forecasting (US industrial production)12 / Project
Description
Does a richer model improve a one-month forecast of US industrial production growth when it may only use data published on the forecast date? This project answers that with the Philadelphia Fed Real-Time Data Set (vintage values) and ALFRED release dates.
Status
Completed
Data used
  • Philadelphia Fed Real-Time Data Set for Macroeconomists (historical data vintages)
  • ALFRED vintage release dates
  • FRED term spread, credit spread and fed funds series
Output / result
A real-time forecast comparison for US industrial production. The factor model won in development but lost in 2012–2026, with RMSE 28% above the autoregressive baseline. Recommendation: keep the simple model and never switch on a development-sample win.
Open the project →
AL — Deposit betas, repricing risk and funding stress (KeyBank N.A.; SVB stress case)13 / Project
Description
A treasury/ALM workflow built only from public data. It replicates Drechsler, Savov & Schnabl (2017) deposit betas on every FDIC-insured bank. It reconstructs KeyBank N.A.'s balance sheet and repricing ladder from Call Reports and replays 18 quarterly report arrivals (2022Q1–2026Q2) with monthly rate updates, planning NII, measuring EVE and outflow cover, and recording decisions in an append-only ledger. SVB at end-2022 is the stress case.
Status
Completed
Data used
  • FDIC BankFind Call Report panel: 246,917 bank-quarters, 2014–2026
  • FDIC Summary of Deposits branch data
  • FRED effective fed funds rate and the Federal Reserve fitted Treasury curve
Output / result
A deposit-beta, EVE and funding-stress workflow for KeyBank with an SVB stress case. A beta model fitted on the bank’s own history under-predicted 2022–23 deposit costs by up to 57%, and EVE loss on the same balance sheet ranges from 5% to 69% of equity depending on the assumed deposit life.
Open the project →
PR — Portfolio risk forecasting and rebalancing control (Select Sector SPDR book)14 / Project
Description
Replicates Engle & Colacito (2006) — comparing covariance forecasts by the realised variance of the portfolios they build — on the Fama–French 10 industries (daily, 1995–2026), with Ledoit–Wolf shrinkage, Jagannathan–Ma constraints, the DeMiguel et al. 1/N benchmark and a cost break-even. The model chosen on 1995–2018 then runs a monthly-rebalanced, long-only book of the 11 Select Sector SPDR ETFs from 2019-01 to 2026-09 with a holdings and cash ledger (distributions, splits, costs), pre-trade controls, approvals and three labelled drills.
Status
Completed
Data used
  • Kenneth French daily industry portfolios and factors, 1926–2026
  • SSGA NAV, distribution and premium history for the 11 Select Sector SPDR ETFs
  • Daily ETF prices, 2016–2026
Output / result
A month-end rebalancing workflow for a USD 10m sector book. Minimum variance cut volatility from 18.0% to 15.8% and drawdown from −36% to −32%, but earned 1.8 points a year less than equal weight, mostly by missing technology. Shrinkage and dynamic covariance models added nothing measurable.
Open the project →
FD — Financial filing change and deterioration monitor (50 large U.S. companies)15 / Project
Description
Replicates Cohen, Malloy & Nguyen (2020, Lazy Prices) on SEC filings — cosine, Jaccard and diff similarity of each 10-K/10-Q against its year-earlier counterpart, by document, MD&A and risk-factor section — and tests both channels: next-month returns of changers vs non-changers, and whether the next report deteriorates. Baselines: ratio-only (Piotroski-style) and text-only. An analyst workbench replays 2023–2026 in EDGAR acceptance-time order with a five-review quarterly budget, ranked alerts with excerpts, review states (open / investigate / escalate / dismiss / expired), amendments, late filings, non-reliance events and firm exits.
Status
Completed
Data used
  • SEC EDGAR filings for 50 large US companies: 1,946 10-Qs and 10-Ks, 2017–2026
  • As-filed XBRL company facts
  • Daily prices for the 50 stocks and SPY
Output / result
A filing review queue for a research team. The ratio model put deteriorating firms at the top 61% of the time against a 23% base rate (lift 2.7), but text-change screens did worse than random and the Lazy Prices return effect did not appear.
Open the project →
IV — Independent model validation and change control (VR limit model)16 / Project
Description
Validates the VR project's BTC exposure-limit model the way a model risk function would under SR 26-2 principles: a model inventory, immutable content-addressed candidate artefacts, an independent benchmark engine written from the protocol text, a regression and stress suite (Kupiec, Christoffersen, Basel traffic light, ES residuals, FZ0, look-ahead invariance, a Hull–Suo simulation, a Danielsson et al. risk ratio, Green–Figlewski estimation error), an approval workflow recorded in an append-only ledger, a controlled version change with an injected defect, twelve monthly monitoring checkpoints, and a late-data restatement that cannot rewrite approvals.
Status
Completed
Data used
  • Binance BTCUSDT one-minute bars from the VR project
  • The VR limit model and its frozen protocol
Output / result
An independent validation and change-control record for the VR limit model. Its numbers were reproduced to 10⁻¹¹ and its 99% VaR is well calibrated, but losses on breach days were 18% worse than forecast ES. One change request was rejected for look-ahead before a fixed version was approved.
Open the project →
EX — Execution replay and transaction-cost diagnostics (Coinbase BTC-USD)17 / Project
Description
This project replicates Cont, Kukanov & Stoikov (2014) and tests Almgren & Chriss (2001) on 69 full sessions of Coinbase BTC-USD.
Status
Completed
Data used
  • Tardis.dev Coinbase BTC-USD level-2 book updates, trades and quotes: 69 monthly sessions, 2021–2026
Output / result
An execution replay comparing immediate, TWAP, POV and Almgren–Chriss schedules. For orders up to 10 BTC, sending at once costs 0.56–3.36 bp with far less variance than TWAP, and the assumed 5 bp taker fee is larger than the impact of every order size.
Open the project →
MM — Inventory-aware market-making research (Coinbase BTC-USD)18 / Project
Description
This project replicates Avellaneda & Stoikov (2008) on 69 full sessions of Coinbase BTC-USD: the exponential execution-intensity law; the claim that inventory-aware quotes cut inventory and P&L dispersion relative to symmetric quotes.
Status
Completed
Data used
  • Tardis.dev Coinbase BTC-USD level-2 book updates, trades and quotes: 69 monthly sessions, 2021–2026
Output / result
A market-making replay comparing symmetric, Avellaneda–Stoikov and GLFT quoting. Every policy lost money at a zero maker fee because of adverse selection, while a 1 bp rebate would turn almost all of them positive: the fee tier decides the business.
Open the project →
OV — Option quote validation and pricing control19 / Project
Description
Hourly independent price verification for Deribit BTC inverse (coin-margined) options, with ETH as an out-of-family check. The data are every Deribit option quote and trade on the first UTC day of each month, January 2021 to August 2026 (68 sessions; Tardis.dev options_chain, reduced to 5-minute snapshots, plus Deribit's registry, delivery prices and trade tape). Development 2021–2022 fixed every threshold; 2023-01 to 2026-08 (44 sessions) was evaluated once.
Status
Completed
Open the project →

Phase 4.3JD to Projects

VT — Systematic Volatility Trading and Delta-Hedged Options Book20 / Project
Description
When and how much BTC volatility to sell or buy on Deribit, in which tenors, and how to hedge it.
Status
Completed
Open the project →
XV — Counterparty Exposure, XVA and Exposure-Model Backtesting21 / Project
Description
Is the exposure model fit for limits and pre-deal CVA, at which horizons, with what add-ons? Answer: fit up to about three months with a 1.2–1.5× dispersion add-on, not validated beyond one year (details in the memo).
Status
In progress
Open the project →
BS — Grid Battery Valuation, Dispatch and Hedging22 / Project
Description
What is the battery worth, how should it be dispatched and hedged, should it be tolled?
Status
Completed
Open the project →
PM — Prediction-Market Pricing, Calibration and Market-Making23 / Project
Description
Which event contracts to quote or take, at what fair value and size?
Status
Brief
Open the project →
OB — Machine-Learning Signals on Limit-Order-Book Data24 / Project
Description
Is a short-horizon order-book signal real, robust and worth trading after costs?
Status
Completed
Open the project →
LS — Queue-Position and Latency-Aware Order-Book Simulator (C++)25 / Project
Description
Which passive-fill model and latency assumption should the research simulator certify for Coinbase BTC-USD, and which upstream claims (MM, EX, OB, OM) survive it?
Status
Completed
Open the project →
AV — AI/ML Credit-Model Validation, Monitoring and Change Control26 / Project
Description
Approve, restrict or retire an ML credit model and its GenAI reason generator?
Status
Brief
Open the project →
CT — Systematic Futures Trend and Carry (Managed-Futures Book)27 / Project
Description
How to weight trend and carry across a futures book, at what risk, and when to de-risk?
Status
Brief
Open the project →
XS — Exotic and Structured-Product Pricing, Hedging and Model Reserve (C++ engine)28 / Project
Description
Which model marks and hedges an autocallable book, and how big is the model reserve?
Status
Brief
Open the project →
NL — Language-Model Signals from Corporate Disclosures, with Look-Ahead Control29 / Project
Description
Do LLM signals from 8-K filings survive look-ahead control and costs?
Status
Brief
Open the project →
NG — Natural-Gas Storage Valuation, Hedging and LNG Cargo Optionality30 / Project
Description
What is a storage contract worth, how to run and hedge it, and should the desk buy it?
Status
Completed
Open the project →
OM — Options Market Making: Quote Engine, Greek Inventory and Adverse Selection31 / Project
Description
Where to quote every option, how wide, how skewed by Greek inventory, and when to pull?
Status
In progress
Open the project →
IS — Intraday Cross-Sectional Statistical Arbitrage: From Signal to Monetised P&L32 / Project
Description
Do intraday cross-sectional signals survive costs, and how much does a cost-aware optimiser add?
Status
Completed
Open the project →
TX — Transformer and Deep Sequence Models for Market Data, with Honest Baselines33 / Project
Description
Does a from-scratch Transformer earn a place in the forecast stack, and which design choices matter?
Status
Brief
Open the project →
RP — Research and Backtesting Platform Engineering for a Systematic Desk34 / Project
Description
What research platform should the desk build, and does it prove research–production parity?
Status
Brief
Open the project →
SW — Swap Curves, Evaluated Pricing and Automated Quoting from Public Swap Trades35 / Project
Description
Which public swap prints to trust, what marks to publish with what grade, and how to auto-quote.
Status
Completed
Open the project →
FI — Systematic Credit and Fixed-Income Factor Portfolios36 / Project
Description
Which credit factors to hold and how to build a bond portfolio that beats the index after costs?
Status
Brief
Open the project →
SP — Sports-Market Pricing, In-Play Trading and Correlated Parlays37 / Project
Description
What are fair sports prices, including correlated parlays, and which bets have real edge?
Status
Brief
Open the project →
CR — Electronic Credit Market Making: RFQ Pricing, Bond Inventory and Hedging38 / Project
Description
How to price RFQs in illiquid bonds, skew by inventory and hedge with proxies?
Status
Brief
Open the project →
ER — Energy Trading Risk: Forward-Curve VaR, Hedge Effectiveness and Price Verification39 / Project
Description
Is the VaR/ES model fit for limits on an energy book, are curves verified, are hedges effective?
Status
Brief
Open the project →
RX — Rates Volatility and Exotics: SABR, LGM and Cheyette Models for Bermudans, CMS and Structured Notes40 / Project
Description
Which rates model prices and hedges Bermudans and CMS products, and how big is the model reserve?
Status
Brief
Open the project →
ST — Bank Stress Testing and Credit-Loss Models: Develop, Then Validate41 / Project
Description
Are the stress-loss and CECL models fit for capital planning and reserving, and with what overlays?
Status
Brief
Open the project →
AR — Alternatives Replication and Quantitative Manager Due Diligence42 / Project
Description
Replicate or allocate, and does a manager's track record show skill or hidden risk?
Status
Brief
Open the project →
IX — Index Rebalance Trading, Index Arbitrage and Delta One43 / Project
Description
Which index events to trade, how to size from passive demand, and how to execute at the close?
Status
Brief
Open the project →
EF — ETF Fair Value, Market Making and Creation/Redemption Arbitrage44 / Project
Description
What is an ETF worth right now, how to quote and hedge it, and when to create or redeem?
Status
Brief
Open the project →
CB — Crypto Basis and Funding: Perpetuals, Futures and Cross-Venue Capital Allocation45 / Project
Description
Which funding and basis trades to hold, with how much capital on which venue?
Status
Brief
Open the project →

Phase 4.4Papers to Projects

Agentic AI Screening7 projects / Family
Agentic AI Screening for Portfolio Investment: Replicate, Audit, Extend46 / Parent project
Description
Should an LLM agent team choose the investable universe before a precision-matrix optimiser weights it, and does the paper's Sharpe advantage survive honest selection, a longer post-cutoff window and real news coverage?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Point-in-Time Agent Harness47 / Child project
Description
Which (model, data, prompt) combinations are admissible at each decision date, and does an agent pipeline earn a look-ahead certificate before its results are reported?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Agent Investment Committee48 / Child project
Description
Which agents sit on the committee, how are their votes combined, and when is an agent benched, so that the committee beats the best single agent and the unscreened book after costs?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Agentic Quant Researcher with Guardrails49 / Child project
Description
Which agent-proposed screening rules are promoted into the screen, and does a disciplined agentic search loop beat the paper's fixed three-characteristic LLM-S and an equal-budget random search?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Many Funds, Same Agents: Crowding Simulation50 / Child project
Description
How much crowding and crash risk does a fund take on when its screening agents resemble everyone else's, and which diversity or oversight policy limits it?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Red-Teaming the Agent Team51 / Child project
Description
Which agent-team topology and guardrails should be deployed, given that an adversary can reach the news feed and the documents and tool outputs the agents read?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
From Agentic Screen to Tradable Book52 / Child project
Description
How fast should the book trade toward the agents' target portfolio, and does the agentic edge survive realistic costs, impact and capacity?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Volume-Imbalance Market Making7 projects / Family
Volume-Imbalance Market Making: Replicate, Audit, Extend53 / Parent project
Description
Should a passive market maker in a one-tick stock post or withdraw each side according to top-of-book volume imbalance, and does the paper's gain survive queue-aware fills, latency, fees and one pre-registered specification?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Queue-Aware Replay for Passive Strategies in Large-Tick Stocks54 / Child project
Description
Which fill model, and which data level (order-level messages, displayed depth, or none), must a backtest of passive quoting in one-tick Nasdaq stocks use, and how far off is the P&L under each?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Fill-Conditioned Quoting with Order-Book Imbalance55 / Child project
Description
Should a passive quoter price the adverse selection of the fills it actually gets (back of the queue, mostly when the queue empties) rather than of the average market order, and when should it withdraw?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Tick-Size Transfer: Imbalance Quoting on Small-Tick Books56 / Child project
Description
Does VI's rule (withdraw the side the imbalance points at) carry over to a small-tick book such as Coinbase BTC-USD, or should a maker there trade against the imbalance, as a live crypto experiment suggests?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Spoofing Robustness of Imbalance-Driven Quoting57 / Child project
Description
How much does an imbalance-driven quoter lose to an adversary who posts and cancels fake depth, and which guardrail (age-weighted depth, cancel-rate filter, detection model, position limits) limits the loss at what cost in normal times?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Order-Book Signal Search for Quoting under Multiple-Testing Control58 / Child project
Description
Which order-book state variable (top-of-book imbalance, multi-level imbalance, queue-depletion hazard, micro-price, order-flow imbalance) should drive the post/withdraw decision, chosen so that the winner survives a deflated-Sharpe and SPA test?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Liquidity Withdrawal When Many Makers Use the Same Imbalance Signal59 / Child project
Description
If many makers withdraw on the same imbalance signal, how much does touch liquidity thin, do depletion cascades appear, and which policy (minimum quote life, staggered thresholds, signal diversity) limits the damage?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Anonymised News for LLM Backtests6 projects / Family
Anonymised News for LLM Backtests: Look-Ahead versus Distraction60 / Parent project
Description
Can an LLM news-sentiment backtest that runs inside the model's training window be reported, what correction does it need, and should company identifiers be masked in production?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Re-identification Audit for Masked Financial Text61 / Child project
Description
Is a masked text actually anonymous to a given model, and which masking level (name, abbreviation, products, tickers, people, partner firms, numbers) is enough before masked scores may be called de-biased?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
Dual-Read News Agent: Named, Masked or Abstain62 / Child project
Description
For each news item, should the agent trade on the named reading, the masked reading, or abstain, and does a disagreement-aware rule beat both single reads after the model's cutoff and after costs?
Status
In progress
Access
Free for signed-in users · No paid membership needed
Open the project →
News Timing, Latency and Costs for LLM Headline Strategies63 / Child project
Description
How late can an LLM news signal arrive and how much can trading cost before the headline long-short edge disappears, and which timing convention should a backtest use?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Model-Vintage Panel: Look-Ahead and Distraction by Model64 / Child project
Description
Which model (family, size, vintage) should score news, given its measured look-ahead and distraction, and is that ranking stable across model releases?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →
Long Text: Look-Ahead and Distraction in Filings and Earnings Calls65 / Child project
Description
Which text granularity (headline, article, 8-K, earnings-call transcript) should an LLM agent read, given that longer texts carry more signal but also more identifying detail?
Status
Planned
Access
Free for signed-in users · No paid membership needed
Open the project →