What you can do with 60+ years of market data

Standard stock screeners filter on the latest quarterly value or a trailing average. That works for simple questions. But finance professionals often need to screen across time, detect patterns in fundamentals, study what happened after specific events, or compare companies against custom peer groups. Those questions require checking conditions across many periods, combining multiple data sources, and summarizing results across thousands of stocks. This page walks through four categories of analysis and shows what each looks like.

For 20 complete query examples by investor type, see Examples. For a multi-session workflow walkthrough, see Use Cases.

Screen across time, not just today

Check every quarter, not just the latest

Filtering on the most recent quarter catches companies that happened to have a good three months. Screening across time finds companies with durable characteristics. The difference: one checks a single number, the other checks 20 or 40 individual data points per company and requires every one to pass.

This matters for quality investing. A company with 18% average ROE over 5 years might have achieved that through consistent 17-19% performance (durable) or through volatile swings between 5% and 30% (unreliable). The average is the same. The investment thesis is not.

Claude checks 40 quarterly values per company, rejects any where even one quarter fell below 15%, and returns the survivors ranked by market cap. The "lowest ROE quarter" column proves that every period passed - if the minimum is still above 15%, the consistency is real.

Combine current signals with historical consistency

The most actionable screens combine a current signal (a stock is technically oversold today) with a historical track record (the company has been fundamentally sound for years). This requires cross-referencing daily technical indicators with quarterly financial statements and checking consistency across multiple years.

Variations: never-negative free cash flow since a specific year, current ratio above 1.5 through every quarter including 2020, revenue growth in every single year for a decade.

Detect momentum shifts in fundamentals

Find acceleration, deceleration, and inflection points

A company growing revenue at 20% is interesting. A company where revenue growth accelerated from 8% to 12% to 18% to 25% over four consecutive quarters tells a different story - one of compounding momentum that a single growth rate number would miss.

Sequential pattern detection goes beyond magnitude to check the shape of the trend. It computes growth rates for each period, then verifies that each rate exceeds (or falls below) the prior one for a specified number of consecutive periods. This finds companies at inflection points before the headline growth numbers catch up.

Claude computes sequential growth rates and margin changes for every company, checks both patterns simultaneously over the same four quarters, and returns those where both conditions held.

Track multi-metric trends

Some of the most useful patterns involve trends in derived ratios, not just headline numbers. A company where R&D spending as a percentage of revenue has been rising for 3 years while net margins also improved is investing in growth without sacrificing profitability. That pattern is hard to spot without computing the ratio for each period and checking the direction of both series simultaneously.

Other patterns: interest coverage improving while leverage decreases, capital expenditure as a percentage of depreciation trending above 1.5 for multiple years (indicating investment in growth rather than maintenance), earnings beat streaks spanning 8 or more consecutive quarters.

Study what happened after specific events

Run event studies across the entire market

The most valuable question in investing is often not "which stocks pass my filter today?" but "what happened historically when this pattern occurred?" Event studies answer that question by identifying every historical instance of an event across all stocks, measuring what happened to prices afterward, and summarizing the results.

This is standard methodology in academic finance. It answers questions like: is buying after an earnings miss actually a good strategy? Does it depend on the size of the miss? Does it depend on the company's prior trajectory? Shibui has earnings data (actual vs. estimate, surprise percentage, report date) alongside daily prices back to 1962, making these studies possible in a single query.

Claude identifies every qualifying earnings event, joins each to the subsequent stock prices at the specified offsets, segments by market cap, and returns the summary statistics. This typically produces hundreds or thousands of data points, giving statistical weight rather than anecdotal evidence.

Backtest screening criteria

A screen that looks good today might have worked by chance. Historical backtesting validates whether the criteria you care about actually predicted future returns. Shibui can reconstruct a screen at historical points in time, identify which stocks would have qualified, and measure their forward performance.

This is directional backtesting - useful for validating whether a set of criteria has historically identified outperformers. It uses fundamentals from 1990, valuations from 1993, and daily prices from 1962. Note: Shibui does not adjust for survivorship bias or delisting returns, so treat results as directional, not definitive. See Data Sources for coverage details.

Define peer groups on the fly

Peers defined by financial characteristics, not just classification

Industry classification is a starting point, but serious comparative analysis often requires a more specific peer group. "Semiconductor companies above $20B market cap, excluding the two largest" is a different peer group than "the semiconductor industry." So is "companies in any sector with similar capex-to-revenue ratios and market cap within 2x."

In Shibui, a peer group is just a set of conditions. Any combination of sector, market cap, financial ratios, growth rates, or exclusion lists can define the group. The peer definition is as flexible as the question that motivated it.

Claude defines the peer group via the financial criteria, computes the group average operating margin per quarter, and overlays MSFT's individual trajectory against that average. The result shows whether MSFT is outperforming, converging with, or falling behind its peer group over the period.

This is a cross-sector scan with per-sector ranking - useful for building a diversified watchlist of the strongest grower in each sector, filtered by quality constraints.

Combining all three in a single query

The categories above are not isolated features. They combine in a single query to answer questions that would otherwise require stitching together multiple tools, datasets, or scripts. The following query uses screening across time, pattern detection, and event study methodology together.

This single query screens thousands of companies for a multi-year consistency pattern, detects the exact inflection point where the streak broke, and measures forward price returns at three time offsets across every historical instance. It answers a specific investment question: when a long-running quality signal breaks, how does the market react?

In one sentence of English, you get a study that spans decades of data, hundreds of individual events, and a statistical answer to a question that matters.