Every feature, and what it tells you

Which company, and which earnings event. Ticker, company name, sector, industry, country, and the date of the report itself.

Every row in the data is one company reporting once, and these columns are how you know which one. EARNINGS_DATE and TICKER together identify a row uniquely - if you are joining our data to your own, that pair is the key.

What the market is paying for the company, relative to what the company earns, owns or sells. Price/earnings, price/sales, price/book, enterprise value and the analyst target price.

These are the standard yardsticks for whether a stock looks expensive, and they matter at an earnings event because expectations are priced in before the report lands. A company that beats modestly can still fall if it was priced for a large beat.

One catch worth knowing: most of these divide a figure that only updates when the company files - earnings, sales, book value - by a share price that moves every day. So the ratio drifts even in a week when nothing about the business changed.

What the company earns, what it is expected to earn, and how fast that has been growing. Earnings per share over several horizons, sales, and the surprise against consensus at the last report.

This is the centre of an earnings event. The market has a number in mind, the company reports a number, and the gap between them drives much of what happens next. The growth columns give the trend the report is landing against - a miss after four straight beats reads differently from a miss after four misses.

Note that EPS_Q/Q is year-over-year despite the name, which is the vendor’s convention rather than ours.

Whether the company actually makes money, and how much of each sales dollar it keeps. Margins at the gross, operating and net level, return on equity, return on assets, and the dividend.

Margins are the clearest read on business quality in this data. Two companies can grow sales at the same rate while one keeps twenty cents of every dollar and the other keeps two, and that difference tends to persist. Margin direction also often moves the stock more than the earnings number itself, because it speaks to whether growth is profitable.

What the company owns and owes. Cash, debt, the ratio between them, and the current ratio measuring whether short-term obligations are covered.

At an earnings event this is mostly about survivability and flexibility. A company with heavy debt has less room to absorb a bad quarter, and its stock tends to react more violently to one. A large cash position cuts the other way.

Who holds the stock, and who is betting against it. Institutional and insider ownership, recent insider transactions, short interest, and the short ratio.

Short interest is the one to watch around earnings. A heavily shorted stock that reports well can rise far more than the news alone justifies, because short sellers buying to close their positions add demand on top of ordinary buying. Crowded positions amplify surprises in both directions.

Timing caveat: these come from regulatory filings on their own schedules - institutional holdings quarterly with a lag of about 45 days, insider transactions within days, short interest twice a month. They can be several weeks stale relative to the earnings date, which is a property of the filings, not of our data.

Where the stock has been trading and how it has behaved. Current price, moving averages, distance from 52-week highs and lows, recent performance over several windows, relative strength, beta and average volume.

Two things make this useful at an event. First, momentum going in shapes the reaction: a stock already at a high has more expectation priced in than one near a low. Second, average volume tells you whether you can actually trade the name in the size you want - a thinly traded stock can move sharply on a position that would be routine elsewhere.

Simple yes/no markers. Whether the company is in the S&P 500, Nasdaq 100, Dow, Russell or is US-listed, plus a few event markers.

These exist mainly for filtering. Index membership is a reasonable proxy for size, liquidity and institutional attention, and restricting to S&P 500 members is the most common way to cut this data down to names that trade easily.

How violently this stock has moved, and how badly it has gone wrong before. Volatility measured several ways, worst-case loss estimates, and risk-adjusted return measures such as Sharpe and Sortino ratios.

We compute these from a return history ending at the earnings date, so they describe the stock you are about to trade rather than the market in general. The practical use is sizing: two stocks with the same expected move are not the same trade if one routinely swings three times as hard.

VAR_H is the loss this stock exceeded on only a small fraction of past days, measured from its own return history. C_VAR is the average loss on those worst days, and it is the more honest figure of the two: it describes how bad the bad days actually get rather than just where the threshold sits.

Both come in several flavours. VAR_P assumes a normal distribution, FHS_VAR and FHS_CVAR use filtered historical simulation, which handles fat tails and volatility clustering better. If you only look at one, look at FHS_CVAR.

Several columns in this family come as buckets from 1 to 5 or 1 to 10, ranking each stock against its peers. Order is meaningful; arithmetic is not. A stock in bucket 4 is riskier than one in bucket 2, but not “twice as risky”.

Where this earnings event sits among all the others happening around it. How crowded the reporting calendar is that week, how the event ranks against peers, and our composite event score.

Context changes how an event trades. A report landing on a quiet day gets the market’s full attention; the same report on a day when two hundred other companies also report competes for it. Crowding also affects how quickly the price settles after the news.

What actually happened after the event. Returns at 3, 30 and 60 days, abnormal returns with the market’s own move stripped out, and realized direction labels.

The abnormal returns are the ones that answer “did this company’s news matter”, because a stock up 4% on a day the whole market rose 4% has told you nothing. These are what you would measure a strategy against.

Two things to know. These are only populated for events already far enough in the past for the window to have completed - a 60-day figure needs 60 days to have passed - so they are empty for recent and upcoming events. That is a real answer, not missing data. And because these are the outcomes our own models train to predict, they are gold tier.

Model output for upcoming events: our estimate of the expected move and its direction, produced by the silver-tier models.

These are forward-looking, computed for events that have not happened yet, and they update as new information arrives before the report.

Our highest-confidence model output for upcoming events, including the conviction score and sector-relative ranking.

The distinction from silver is selectivity rather than volume. These models decline to score most events, and GOLD_CONVICTION is empty for the majority of rows by design - a model that had an opinion about everything would be less useful, not more. When it is populated, it reflects a setup the model recognises.

Which company, and which earnings event. Ticker, company name, sector, industry, country, and the date of the report itself.

Every row in the data is one company reporting once, and these columns are how you know which one. EARNINGS_DATE and TICKER together identify a row uniquely - if you are joining our data to your own, that pair is the key.

What the market is paying for the company, relative to what the company earns, owns or sells. Price/earnings, price/sales, price/book, enterprise value and the analyst target price.

These are the standard yardsticks for whether a stock looks expensive, and they matter at an earnings event because expectations are priced in before the report lands. A company that beats modestly can still fall if it was priced for a large beat.

One catch worth knowing: most of these divide a figure that only updates when the company files - earnings, sales, book value - by a share price that moves every day. So the ratio drifts even in a week when nothing about the business changed.

What the company earns, what it is expected to earn, and how fast that has been growing. Earnings per share over several horizons, sales, and the surprise against consensus at the last report.

This is the centre of an earnings event. The market has a number in mind, the company reports a number, and the gap between them drives much of what happens next. The growth columns give the trend the report is landing against - a miss after four straight beats reads differently from a miss after four misses.

Note that EPS_Q/Q is year-over-year despite the name, which is the vendor’s convention rather than ours.

Whether the company actually makes money, and how much of each sales dollar it keeps. Margins at the gross, operating and net level, return on equity, return on assets, and the dividend.

Margins are the clearest read on business quality in this data. Two companies can grow sales at the same rate while one keeps twenty cents of every dollar and the other keeps two, and that difference tends to persist. Margin direction also often moves the stock more than the earnings number itself, because it speaks to whether growth is profitable.

What the company owns and owes. Cash, debt, the ratio between them, and the current ratio measuring whether short-term obligations are covered.

At an earnings event this is mostly about survivability and flexibility. A company with heavy debt has less room to absorb a bad quarter, and its stock tends to react more violently to one. A large cash position cuts the other way.

Who holds the stock, and who is betting against it. Institutional and insider ownership, recent insider transactions, short interest, and the short ratio.

Short interest is the one to watch around earnings. A heavily shorted stock that reports well can rise far more than the news alone justifies, because short sellers buying to close their positions add demand on top of ordinary buying. Crowded positions amplify surprises in both directions.

Timing caveat: these come from regulatory filings on their own schedules - institutional holdings quarterly with a lag of about 45 days, insider transactions within days, short interest twice a month. They can be several weeks stale relative to the earnings date, which is a property of the filings, not of our data.

Where the stock has been trading and how it has behaved. Current price, moving averages, distance from 52-week highs and lows, recent performance over several windows, relative strength, beta and average volume.

Two things make this useful at an event. First, momentum going in shapes the reaction: a stock already at a high has more expectation priced in than one near a low. Second, average volume tells you whether you can actually trade the name in the size you want - a thinly traded stock can move sharply on a position that would be routine elsewhere.

Simple yes/no markers. Whether the company is in the S&P 500, Nasdaq 100, Dow, Russell or is US-listed, plus a few event markers.

These exist mainly for filtering. Index membership is a reasonable proxy for size, liquidity and institutional attention, and restricting to S&P 500 members is the most common way to cut this data down to names that trade easily.

How violently this stock has moved, and how badly it has gone wrong before. Volatility measured several ways, worst-case loss estimates, and risk-adjusted return measures such as Sharpe and Sortino ratios.

We compute these from a return history ending at the earnings date, so they describe the stock you are about to trade rather than the market in general. The practical use is sizing: two stocks with the same expected move are not the same trade if one routinely swings three times as hard.

VAR_H is the loss this stock exceeded on only a small fraction of past days, measured from its own return history. C_VAR is the average loss on those worst days, and it is the more honest figure of the two: it describes how bad the bad days actually get rather than just where the threshold sits.

Both come in several flavours. VAR_P assumes a normal distribution, FHS_VAR and FHS_CVAR use filtered historical simulation, which handles fat tails and volatility clustering better. If you only look at one, look at FHS_CVAR.

Several columns in this family come as buckets from 1 to 5 or 1 to 10, ranking each stock against its peers. Order is meaningful; arithmetic is not. A stock in bucket 4 is riskier than one in bucket 2, but not “twice as risky”.

Where this earnings event sits among all the others happening around it. How crowded the reporting calendar is that week, how the event ranks against peers, and our composite event score.

Context changes how an event trades. A report landing on a quiet day gets the market’s full attention; the same report on a day when two hundred other companies also report competes for it. Crowding also affects how quickly the price settles after the news.

What actually happened after the event. Returns at 3, 30 and 60 days, abnormal returns with the market’s own move stripped out, and realized direction labels.

The abnormal returns are the ones that answer “did this company’s news matter”, because a stock up 4% on a day the whole market rose 4% has told you nothing. These are what you would measure a strategy against.

Two things to know. These are only populated for events already far enough in the past for the window to have completed - a 60-day figure needs 60 days to have passed - so they are empty for recent and upcoming events. That is a real answer, not missing data. And because these are the outcomes our own models train to predict, they are gold tier.

Valuation in the style of Benjamin Graham: an estimate of what the business is worth based on its assets and earnings, and how far the current price sits below or above it.

This is a deliberately conservative, balance-sheet-driven view of value, and it disagrees with the market often - that is the point of it. It answers a different question from the standard ratios: not “is this cheap relative to similar companies” but “what would this business be worth on its numbers alone”.

What our models predicted for this event, recorded before the outcome was known.

These sit in the same row as what actually happened, which is what makes this data a record of prediction beside result rather than a set of scores you have to take on faith. You can measure our hit rate yourself over any period you like.

Frozen at prediction time and never revised afterwards. Gold tier.

How each figure moved since the company’s previous report. For a fundamental like earnings per share or operating margin, this is the same measure one quarter earlier subtracted from the current one.

Direction is often more informative than level. A 12% margin means little on its own; a 12% margin that was 9% last quarter means something specific, and the two together are what a reader actually wants to see. Each change column sits beside the figure it describes.