These are real research studies run through the Shibui Finance MCP connector. Each page shows the natural-language prompt, the SQL that Claude generated automatically, and the actual result table. Nothing is cherry-picked or hand-tuned - the SQL is raw output, and the results come straight from the database.

The point is to show what the product can do beyond simple screening: hypothesis testing, statistical backtests, and market-structure analysis that would normally require a data terminal subscription and a Python script. Here it takes a single sentence.

Connect to Claude → Backtest a strategy →

Every Nasdaq-100 drawdown since 1999: depth, duration, and recovery time

All 19 episodes of 10%+ decline. The dot-com crash took 16 years to recover. The median correction took 28 days. Episode 19 is ongoing.

Ran in <1s · 4 CTEs · 27 years of QQQ daily data

Every S&P 500 drawdown since 1993: depth, duration, and recovery time

All 12 episodes of 10%+ decline. The GFC fell 56.5% and took 4 years to recover. The median correction took 39 days.

Ran in <1s · 4 CTEs · 33 years of SPY daily data

Historical US stock market valuations: P/E, P/B, and dividend yield since 1995

31 years of market-cap-weighted valuation ratios. Trailing P/E from 12.9x (2011) to 31.9x (2020). Current valuations near 31-year highs on multiple measures.

Ran in <1s · 3 CTEs · 672 large-cap stocks

Does insider buying during market crashes predict returns?

Backtest of persistent insider buying signals during 7 crash months (2020-2024). 96% win rate in the COVID crash, mixed results in 2022.

Ran in 1.1s · 8 CTEs · Insider transactions + price data

SPY flush-and-recovery: do prior day levels actually work?

Three price-action patterns tested on 6,692 SPY trading days since 2000. Prior-day-low recovery, prior-day-high as support, and breakout - none beat the baseline.

Ran in 0.1s · 3 CTEs · 25 years of SPY daily data

Composite momentum strategy: does ranking add value?

Weekly top-5 momentum picks vs the eligible pool vs SPY, 2017-2026. The simple filter outperforms the complex ranking. More factors did not mean better returns.

Ran in 18.2s · 14 CTEs · 10,000+ stocks × 7 years weekly

How these studies work

Each study starts with a natural-language question. The Shibui MCP connector gives Claude access to 64 years of US stock market data - prices, fundamentals, technical indicators, SEC filings, and insider transactions. Claude writes the SQL, executes it, and returns the results. No Python, no data downloads, no API keys.

The SQL on each page is the raw output Claude generated. We did not edit, optimize, or simplify it. A skeptic can read the query, verify the logic, and reproduce the result by connecting Shibui and pasting the same prompt.

Related pages:

Backtest a strategy - test any stock strategy against 64 years of data
Research tab - three more research examples on the homepage
Nasdaq-100 drawdown history - every 10%+ decline since 1999
S&P 500 drawdown history - every 10%+ decline since 1993
Historical market valuations - P/E, P/B, and dividend yield since 1995
Backtesting guide - how to run your own backtests
Insider buying screener - live insider buying screen

Data note: All results are generated from a specific date and will change as new data arrives. Shibui covers NYSE + NASDAQ (10,000+ securities), daily prices since 1962 (~31M rows), quarterly financials, and SEC filing metadata. This is a data tool, not financial advice.

Run your own research

Connect Shibui to Claude in 2 minutes. Ask any question about US stocks and get real data back - no coding, no subscription, no hallucinated numbers.

Connect to Claude →