QUANTGPTLEARN · RESEARCH RESOURCES · VERIFIED 2026-09-21
// 91 sources. 23 re-run in the terminal.
QUANT RESEARCH RESOURCES
From one backtest to a research process: the papers, the data, and the tape they get re-run on.
A curated list for learning systematic investing research, ordered the way the work happens: the data you test on, what makes a backtest number mean something, the factors and events the literature found, what it costs to trade them, and the markets beyond stocks.
The core list uses the paper that introduced the effect, the official documentation, or the repository that implements it. Where the terminal can re-run a paper, the entry names the data the re-run uses and links to the library study and to RUN IT. Stock re-runs use the US tape with delisted names kept and fundamentals by filing date, from 1990 at the earliest; the others run on the S&P 500 index, resolved Polymarket markets or Binance pairs. Each re-run approximates the paper rather than repeating it, and the terminal prints the approximation note with every run: what it holds, over what data, and where it departs from the original. RUN IT shows the finding as CONFIRMED, TRADE-OFF or NO EDGE, gross of costs, and says so on every output. Where the library carries the paper but has no runnable spec for it yet, the entry links to the library entry alone.
Read Start here first. After that, use it as a reference.
Returns fall after publication. The decay to expect from anything on this list.
// 1. DATA: THE TAPE YOU TEST ON
Most bad backtests are bad data. A universe without the dead, fundamentals seen before they were filed, and prices without corporate actions each add returns that never existed.
How the warehouse is built: decades of history, 42,000 tickers including delisted from 1998, fundamentals joined by filing date, gross returns.
// 2. THE BACKTEST: WHAT MAKES A NUMBER MEAN SOMETHING
Multiple testing, overfitting, decay after publication, and how to run an event study. This section is why the terminal reports a deflated Sharpe and a walk-forward split on every run.
CONFIRMED needs both axes (CAGR and Sharpe) in both windows (training and walk-forward). The methodology, stated in full.
// 3. FACTORS AND ANOMALIES
The cross-section: value, momentum, quality, low risk, size, reversal. Every entry links the original paper, the library study where there is one, and RUN IT where the terminal re-runs it on US stocks.
The bet size that maximizes growth, and why half of it is what people actually use.
// 7. ALT MARKETS: PREDICTION MARKETS, CRYPTO, SPORTS, MEMECOINS
Markets with their own literature and their own data problems. The memecoin graveyard has no paper yet because no free feed sells the dead; QuantGPT records them.
Every new pool recorded the day it appears and followed until long after it dies, because no free feed sells the dead. Survival by launch month; a token dead by the next month-end is a total loss.
LOPEZ-LIRA & TANG, 2026, JOURNAL OF FINANCIAL ECONOMICS
LLM sentiment on headlines predicts next-day returns in-sample. The multiple-testing warning applies twice.
Watchlist: AI-generated strategies
Agents that write and test thousands of rules a night are a multiple-testing machine. Any published result needs the trial count and a deflated Sharpe, or the number is omitted.
Watchlist: prediction market calibration since 2024
Volumes on Polymarket and Kalshi are a different regime from the data in the 2004 to 2013 papers. Calibration by category and time to close, on the resolved record, is the open question.
Watchlist: memecoin survival base rates
The graveyard is months old. Survival at 30, 90 and 180 days by launch month becomes a citable base rate once it holds a year of cohorts.
Watchlist: on-chain and funding factors
Funding rates, exchange flows and holder concentration as factors, pending a survivorship-complete universe to test them on.
// SOURCE POLICY
A core source must be one of the following: the paper that introduced the effect; the specification or official documentation that defines the data; the repository that implements it; or a replication that states its universe, window, survivorship treatment, costs and out-of-sample method.
Return claims need the universe, the window, the survivorship treatment, the cost assumption and the out-of-sample method. Otherwise the number is omitted.
Where an entry carries RUN IT, the terminal shows the finding (CONFIRMED, TRADE-OFF or NO EDGE) with the walk-forward split and a deflated Sharpe, gross of costs, and says so on every output. Nothing on this page is investment advice.