Risk
Beta stock definition: 5.04-year window
A beta's stock definition is its window. Ours is 5.04 years, 2021-08-12 to 2026-08-26, so no risk figure here can contain 2020 or 2008.
By Javier SanzPublished Updated : The staleness section is re-dated and re-measured. Every price in the system was more than a week old on 2026-09-07; the refresh has since run, and the heading said otherwise in the present tense.
Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.
Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)
The price archive here runs from 2021-08-12 to 2026-08-26, a span of 5.04 years, so every risk figure computed from it is measured over a window that cannot contain early 2020, or 2008, or anything before them.
That matters most to beta. The definition of a stock's beta is the slope of a regression of its returns on the market's, and the only thing that makes one beta different from another is the window it was measured over. So the useful question about any published beta is not what it means. It is how far back the prices go, and here the answer is a decision by the data plan rather than an accident. The endpoint that reports it says so without being asked.
What a five-year window can hold in 2026
The window opens on 2021-08-12. What it contains is whatever happened between that date and 2026-08-26, and the useful half of that sentence is what it cannot contain: nothing before August 2021 is in it. Not the selloff of early 2020, not 2008, not 2000.
A volatility figure, a drawdown figure or a beta computed on this archive is therefore measured over a period that begins after the last general market break. That is not a reason to ignore the figures. It is a reason to state what they are measured over every time they are used, which is what the archive span is for.
Coverage falls as the window lengthens
On 2026-09-07 the screener held a price for all 5,257 listings in its pool. What it held beyond that depends on how much history the measure needs.
| Figure | Listings holding it | Share of the 5,257 | History it needs |
|---|---|---|---|
| Share price | 5,257 | 100.0% | one quote |
| Maximum drawdown, one year | 4,949 | 94.1% | one year of daily prices |
| Price volatility, 52 weeks | 4,948 | 94.1% | one year of daily prices |
| Total return, one year | 4,953 | 94.2% | one year |
| Beta | 4,577 | 87.1% | two years of weekly returns against SPY |
| Total return, three years | 4,369 | 83.1% | three full years |
Three hundred and four listings have a price and no one-year total return. Another 584 have a one-year total return and no three-year one, because a company without three full years of price history is refused rather than measured over a shorter window and labelled as though it were three years. That refusal is the reason the column falls, and it is the right way round: a number computed over eleven months and printed as a three-year figure is worse than a blank.
Beta is ours, and its window is this one
Beta used to arrive ready-made on the company profile, with its window, index and return frequency unstated. Since 11 September 2026 it is regressed here, over two years of weekly returns against SPY, out of the same archive as every other row in that table. SPY is an exchange traded fund tracking the S&P 500, not the index and not the whole market, and every beta's source trail says so along with the window it used. It is held for 4,577 listings, 376 fewer than hold a one-year total return.
That matters more than it looks, because beta is not usually consumed directly. It goes into the cost of equity through the capital asset pricing model, the cost of equity goes into the weighted average cost of capital, and the cost of capital is the discount rate that does most of the work in a discounted cash flow. A figure whose measurement window is unstated sits three steps upstream of the answer.
On 2026-09-07 every price here was older than a week
The market data stamp on 2026-09-07 read 2026-08-26. All 5,257 listings in the universe carried a price older than seven days, which is every one of them, and the oldest quote in the set dated from 2026-04-14.
So on the day these counts were taken, the newest price in the system was twelve days old. Anything divided by a price was twelve days old too: every multiple, every yield, every enterprise value.
That was a stalled refresh, not the design. Re-read on 2026-09-13, the stamp is 2026-09-11 and 8 of the 5,518 listings carry a price older than seven days. The point of the section survives the repair: a price that is two days old is still a price that has to be dated on the page, because the freshness of a quote and the time a ratio was recomputed are two different facts.
Filed fundamentals are not affected by this. Companies report quarterly, the statements do not change after they are filed, and the filed history behind the pool goes back ten years for 2,788 of its listings. The asymmetry is worth internalising: the accounting half of this data is deep and slow, and the price half is shallow, and only as current as the night's last refresh.
Four things the window rules out
A through-cycle beta. Two years of weekly returns out of this archive cannot reach a period containing a crash, so a beta that would have been 1.6 in March 2020 and 0.9 since is indistinguishable from a beta that has always been 0.9.
A drawdown that means anything about the worst case. The one-year maximum drawdown says what the worst peak-to-trough fall was in the last year. It is a fact about the last year and it is not an estimate of anything.
Any statement about long-run returns. Three years is the longest return window here, and three years of price is a fact about sentiment as much as about a business.
Backtesting. There is no backtest in this product and this is one of the reasons. Five years of prices, beginning after the last general market break, is not a sample you can learn a strategy from, and a backtest run on it would produce a number that looks like evidence and is not.
What the window does support
It supports the honest version of the same measures: how volatile has this been recently, how far did it fall in the last year, how has it moved against SPY over two years, all stated with the window attached.
And it supports the observation that price history and filing history are different assets with different depths. A rule written on filed fundamentals, which is what a thesis guardrail is here, can be tested against ten years for 2,788 listings. A rule written on price cannot be tested against anything much at all.
What this leaves out
The archive span, the coverage counts and the market data stamp are single readings taken on 2026-09-07 from the public endpoints. They change. The live figures are on /coverage.
The coverage column above says how many listings hold a figure, not how long each one's own history is. A listing can hold a three-year return and still have less history than another listing that holds the same figure.
Beta is regressed over two years of weekly returns against SPY, inside this archive, so the claim that it cannot be checked against a crash is a claim about the window it actually uses.
The statement that the window begins after the last general market break is an observation about the archive's start date, not a claim that the period since has been calm. Individual listings had their own crises inside it, and the drawdown figure will show them.
Nothing here is a view on any company or any market, and none of it forecasts volatility or return.