Research Recipes¶
A recipe is a focused, reproducible research task: a hypothesis, the data needed to test it, and a starting point for the notebook. Each recipe can be completed in a few hours when the needed data pipeline is already available.
Macro & Policy¶
Fed Policy Probability vs. Equity Sectors¶
Hypothesis: Polymarket probabilities for Fed rate decisions lead sector ETF returns.
Data: Polymarket (Fed FOMC markets) + yfinance (XLK, XLF, XLE, XLU, XLY sector ETFs)
Research questions: - Do rising rate-cut probabilities predict utility and REIT outperformance? - Is there a lead/lag structure between prediction market prices and ETF returns? - Does the effect strengthen in the 5 days before an FOMC meeting?
Status: Recipe pattern only — use the Election & Industry Returns notebook and the Golden Path as the template for converting prediction-market probabilities into ETF sensitivity research.
Treasury Yield Curve vs. Bank Stocks¶
Hypothesis: Yield curve steepening predicts bank stock outperformance.
Data: Treasury.gov rate data + WRDS or yfinance bank-stock returns (JPM, BAC, GS, MS)
Research questions: - What is the rolling correlation between the 10Y-2Y spread and bank stock returns? - Does the effect hold after controlling for market beta? - Is the signal stronger at curve inversions?
Status: Experimental data pipeline exists (treasury_gov_rates), but this recipe has not yet been implemented as a notebook.
Crypto & Prediction Markets¶
Crypto Returns vs. Election Prediction Markets¶
Hypothesis: Crypto prices are sensitive to political and policy event probabilities.
Data: Coinbase/Kraken (BTC, ETH, SOL) + Polymarket election or policy markets + yfinance (COIN)
Research questions: - Do crypto returns correlate with changes in election-market probabilities? - Does the correlation change between early and late campaign periods? - Is COIN a better proxy than BTC for political-market sensitivity?
Status: Recipe — build it with the Crypto and Polymarket pipelines. A worked election × ETF example ships as the example notebook, canonical workflow, and golden path.
Crypto Volatility Regimes¶
Hypothesis: Crypto volatility clusters and is predictable short-term.
Data: Coinbase/Kraken OHLCV (BTC, ETH, SOL)
Research questions: - Do GARCH-family models outperform realized volatility as a one-day-ahead forecast? - Do volatility regimes cluster by asset class (BTC vs. ETH vs. SOL)? - Does the correlation between BTC and ETH increase during high-volatility regimes?
Status: Ready to build — data available from crypto pipeline
Equities & Fundamentals¶
Piotroski F-Score Cross-Sectional Strategy¶
Hypothesis: High F-score stocks outperform low F-score stocks over 12-month holding periods.
Data: WRDS/CRSP (prices) + SEC EDGAR (Piotroski F-scores)
Research questions: - Does the classic Piotroski (2000) result hold in the 30-stock universe? - What is the Sharpe ratio of a long/short F-score portfolio? - Is the effect concentrated in small-cap or value stocks?
Status: Ready to build — both pipelines available
Analyst Earnings Surprises and Drift¶
Hypothesis: Stocks that beat analyst EPS estimates drift upward over the following 30 days (PEAD).
Data: IBES analyst estimates (WRDS additional entitlements) + WRDS/CRSP prices
Research questions: - Is post-earnings announcement drift present in the 30-stock universe? - Does the drift magnitude correlate with earnings surprise magnitude? - Does the effect decay faster in large-cap stocks?
Status: IBES data available with additional WRDS entitlements
ETF Constituent Crowding¶
Hypothesis: Stocks with high ETF ownership show lower idiosyncratic volatility (crowding discount) and larger drawdowns during market stress.
Data: WRDS ETF constituents + WRDS/CRSP prices
Research questions: - Do high-ETF-ownership stocks have lower residual volatility relative to their factors? - Do they show larger drawdowns during VIX spikes? - Is the effect stronger in passive vs. active ETFs?
Status: Coming soon — ETF constituent pipeline not yet built
Backtesting Diagnostics¶
Rolling Sharpe Stability¶
Hypothesis: Strategies with unstable rolling Sharpe ratios are overfit.
Data: LEAN backtest ObjectStore output
Research questions: - How does the 12-month rolling Sharpe ratio evolve over the backtest period? - Is performance stable across market regimes (2008, 2020, 2022)? - Is the out-of-sample Sharpe within one standard deviation of the in-sample Sharpe?
Status: Ready to build for any backtest — use ObjectStore results
Trade Attribution¶
Hypothesis: A small number of trades drive the majority of backtest P&L.
Data: LEAN backtest trade log
Research questions: - What fraction of trades account for 80% of gross P&L? - Is performance driven by a few outlier trades or distributed across many small winners? - Does trade concentration increase or decrease after parameter tuning?
Status: Ready to build for any backtest — use ObjectStore results
How to use a recipe¶
- Pick a recipe that interests you
- Make sure the required pipelines are running locally (see Data Pipelines)
- Create a new Marimo notebook in
infrastructure/marimo/notebooks/ - Work through the research questions, documenting findings as you go
- If it's interesting, open a PR to share it
Good research recipes make good contributions.