How the analyst price-target tracker works

Where the targets come from

Each target is taken from a publicly reported news article about a research firm's price target on a stock. The record keeps the firm, the stock, the target price, the date the article was published and a link to it. Only firms on the site's list of tracked research firms are recorded.

Before a target is stored it has to pass fixed checks. It is refused if:

• the firm is not on the list, or the stock cannot be identified;
• there is no passage in the article to attribute the figure to;
• no closing price for the stock is available within 5 days of the article's publication date, so the price the firm was looking at is unknown;
• the target is more than 20 times the price, or less than 1/20 of it, which means a misread figure rather than a forecast;
• the target equals the price, so there is no call;
• its target date has already passed, which would be scoring a known outcome;
• it is the same target from the same firm on the same stock already stored from another outlet (wire stories are reprinted under several addresses, and one opinion must count once).

Every stored target starts as pending, and only approved targets are shown, scored or counted anywhere on the site. A separate verifier checks each figure against the source and against other reporting where it exists, and recommends approving it, rejecting it, or a person's review.

While automatic approval is switched on, a target is approved without a person when all of these hold: the verifier found the figure confirmed by other reporting, or reported by a single source and not contradicted; it recommended approval or a person's review (a "review" recommendation alone does not hold a target back, because of the first 151 targets decided by hand, none that met the other conditions here and carried a "review" recommendation was rejected); and none of the site's own checks raised a warning. Automatic approval can be switched off, and it stops by itself if 2 automatically approved targets are retracted within 30 days; while it is off, every target waits for a person.

Everything else waits for a person: a target the verifier found contradicted or could not find, one it recommended rejecting, one it has not checked, one more than 3 times the price or under 1/3 of it, one where the report did not make clear whether the firm set, raised, lowered or restated its target, or one restating a target the firm already has open on that stock. Targets rejected at review are never published.

How a target is scored

• Price when published. The daily close nearest the article's publication date: a note published on a Saturday is a view about Friday's close. It is recorded once and never recomputed.
• Direction. Worked out from the two numbers, not from the rating's wording: a target above the price when published is bullish, a target below it bearish.
• Target date. The date the report states, if it states one. Otherwise the publication date plus the horizon the report states, rounded to the nearest of 6, 12, 18, 24, 36, 60 or 120 months. If the report states neither, 12 months, the industry convention, and the firm page marks the target "12-month default horizon" so our assumption is never presented as the firm's.
• Judgment. Once its target date has passed, a target is judged once against the closing price on that date. When the market was shut that day, the last close before it is used, provided it is no more than 10 days earlier. If no close that recent is available, the target is not judged from missing data: it stays due and is tried again later.
• Hit or miss. A bullish target hits if the close is at or above the target; a bearish target hits if the close is at or below it. A close exactly at the target is a hit.
• Ever reached. Recorded separately: whether any daily close from the publication date to the target date, both included, reached the target. It answers a different question (was the firm right on direction but wrong on timing) and is shown, but not used for ranking, because the target had a date on it.
• Never revised. A verdict is written once. Open targets are not scored and count toward nothing.

These are the same rules, from the same code, that judge a member's scorecard call on this site, so a firm's hit rate and a member's are measured the same way.

How firms are ranked

A firm's hit rate is its settled targets that hit divided by its settled targets. Its ever-reached rate is its settled targets that reached the target at any point, divided by the same number. Only approved, unretracted targets count.

A firm is ranked only once it has 10 settled targets. Ranked firms are ordered by hit rate, highest first; at an equal rate the firm with more settled targets comes first, as the better-evidenced figure. A firm with fewer than 10 is still listed on /analysts, with its record shown, under a heading that says why it is not ranked. A firm with nothing settled yet is listed with its open targets and the date the next one comes due.

10 is not a statistically comfortable sample, and the ranking does not pretend it is: the interval around a rate measured on 10 targets is wide. That is why every rate is shown with the count it came from. Read the ordering as a record, not a verdict on which firm is better.

Corrections and retractions

When we find that a target we published misreported what the firm said, we retract it, recording the date and the reason. From that moment it counts toward no rate and no open count, and it is never judged. A target that has already been judged can be retracted too; its verdict then stops counting.

A retracted target is not deleted. It stays on the firm's page, under "Retracted", with its date and reason, so the record shows our corrections rather than quietly losing them. A retraction is about our reporting of what the firm published, never about how the stock moved afterwards: a verdict is not changed because the price moved later.

What each field means

The fields below are the ones shown on /analysts, on each firm's page and on /analysts/month. Dates are New York dates. The columns of the download are listed below.

Download the data

Download the data (CSV), saved as dalsna-analyst-price-targets.csv. One row per approved target: open, hit, miss and retracted, the same targets the pages show. Pending and rejected extractions are left out, and so are members' own scorecard calls.

The file holds no market prices: not the price when a target was published, not the close it was judged against, and nothing computed from either, such as the implied move or the direction. Those prices come from third-party data providers and are not ours to license. The file is rebuilt at most once every 60 minutes, and sooner after a target is approved, retracted or judged; a copy you download can be up to 10 minutes older than that.

Format: UTF-8, comma-separated, a header row, every cell in double quotes, lines ending CRLF. A cell that would begin with =, +, - or @ (or their full-width forms), whether or not spaces come first, or with a tab, carriage return or line feed, starts with an apostrophe instead, so a spreadsheet does not run it as a formula. The columns, in order:

Licence and reuse

Data: Dalsna Finance, licensed CC BY 4.0 — credit dalsnafin.com. The licence covers the targets, dates and verdicts we compile and the rates we derive from them. The linked source articles belong to their publishers, and the market prices we settle against come from third-party data providers; neither is covered.

Targets are reported as facts about what each firm published. They are not recommendations from Dalsna Finance, and no firm listed is affiliated with this site.

<a href="https://dalsnafin.com/analysts">Dalsna Finance analyst price-target tracker</a>