How fast prices take news, and what that asks of a stock news API

A real-time stock news API is only as fast as the clock the research actually measured. Two papers timed that clock, on different markets, and both of them are shorter than a casual poll. Louis Ederington and Jae Ha Lee, writing in the Journal of Finance in 1993, found that the bulk of the price adjustment to a scheduled macroeconomic release was inside the first minute. Jeffrey Busse and T. Clifton Green, writing in the Journal of Financial Economics in 2002, found that a positive analyst mention on CNBC was largely in the price within a minute, and that an after-cost profit on the positive midday mentions disappeared once the buy was more than 15 seconds late.

Neither paper is a trading system, and neither sample is the market of November 2026. They are measurements of how quickly a published fact became a price. A feed that arrives after that window is a record of a move that has already happened. A WebSocket is the delivery method that sits on the same side of the window as the headline.

What's in this post

• The first minute of a macro release
• Fifteen seconds on a stock mention
• Three clocks, side by side
• What a poll spends
• What to open on a stock news WebSocket

The first minute of a macro release

Ederington and Lee studied scheduled announcements in Treasury bond, Eurodollar, and deutsche mark futures. The releases include the employment report, the CPI, the PPI, durable goods orders, and the merchandise trade figures. In their sample the announcements with the largest effect on Treasury bond futures, in declining order, were employment, the PPI, the CPI, and durable goods orders. On the dollar against the deutsche mark, the largest were employment, the merchandise trade deficit, the PPI, durable goods orders, GNP, and retail sales. They date that ranking to 1988 through 1991.

The price result is sharper than the ranking. The major adjustment occurs within one minute of the release. What happens in the next minute does not follow that first minute. Of the 90 largest positive first-minute returns they inspected, the second minute had the same sign exactly 45 times and the opposite sign exactly 45 times. The first minute's return had no meaningful ability to predict the second minute's return. They describe the price at the end of that first minute as a relatively unbiased estimate of the later equilibrium.

Volatility is a different fact from the price level. Prices stay considerably more volatile than normal for about fifteen minutes, and slightly more volatile for several hours. Ederington and Lee give two reasons that can both be true. The reports are long: the employment report of that era ran about nine double-sided pages, and the popular summary is not the whole release. And traders who were not first to the number may still understand it, so later orders keep moving the price even after the first print was unbiased. Their practical line is blunt. To trade the initial reaction, you have to receive the release and execute within one minute. They doubted that anyone off the floor could do that effectively.

They also place the result against an earlier equity paper. Patell and Wolfson, in the Journal of Financial Economics in 1984, found that after dividend and earnings announcements, trading profits largely disappear within five to ten minutes, with some returns still showing up overnight and at the next open. Ederington and Lee call their own futures adjustment considerably faster, and they name three reasons: the contracts were more actively traded, the microstructure differed, and the announcements were expected down to the second, so the market was waiting.

That last reason is the one a macro news feed can use. The clock starts at a known second. category=macro on a stock and market news API is the stream for that class of headline. category=forex is the foreign-exchange feed. A request that leaves category off stays on stocks. The papers are about different instruments, and the API does not turn a CPI line into a futures trade. It does put the headline on a socket at ingest, which is the only side of Ederington and Lee's one-minute window a program can occupy.

Fifteen seconds on a stock mention

Busse and Green timed 322 stock reports on CNBC's Morning Call and Midday Call, from 12 June 2000 through 27 October 2000. The segments aired while the market was open, so they could mark the second a ticker was first said. Prices responded within seconds of that mention. Positive reports were incorporated within about one minute. Negative reports took longer, about fifteen minutes, which they suggest may reflect the higher cost of selling short. Trading intensity more than doubled in the minute after the first mention, with more buyer-initiated trades after positive reports and more seller-initiated trades after negative ones.

The midday positive reports are the clean number. The cumulative return from 15 minutes before the mention to one minute after it is 62 basis points. Forty-one of those basis points occur in the first minute after the mention. Seventy-three percent of the firms discussed positively in that segment had a positive one-minute return, and the average increase in market value was $109 million. A small reversal follows over the next three minutes.

The same network, earlier in the day, did not do this. Positive Morning Call reports had an average one-minute return of 6.8 basis points, and the cumulative half-hour return was negative. Busse and Green read that as information that was slight, or already known. A mention is not a fact. The segment that moved prices was the one that still had something to add while the market was open.

They then priced a trade, not a return. Buys are buyer-initiated prints and sells are seller-initiated prints, so the spread is in the number. Buying within five seconds of a positive Midday Call mention and selling one to two minutes later earned about 40 basis points, around $300 on the average trade size in the sample. Buying within 15 seconds earned about 16 basis points, around $123. After 15 seconds the average was no longer statistically significant. After 25 seconds it was negative. Knowing the headline before it aired was worth more: a buy four to five minutes before the segment earned about 60 basis points after costs. That is the leak, not the public mention.

Busse and Green set this against Barber, Lehavy, McNichols, and Trueman in the Journal of Finance in 2001, who found that strategies built on consensus analyst recommendations lost the abnormal return to transaction costs when the trade was done at the close. The difference in Busse and Green is the clock. The close is hours after a mention that was finished in a minute.

Three clocks, side by side

Read the table as three measurements, not as a ranking of which market is "more efficient" in 2026. The futures were waiting for a clock they already knew. The CNBC sample is analyst opinion on television, twenty-six years ago, in a market that traded less often than a liquid stock trades now. Busse and Green themselves say the old "five to fifteen minutes" had already become seconds by 2000. A modern open has only gotten faster. The useful leftover is the shape: the first slice of the reaction is the one with the information, and the slices after it are volatility, detail, and people who were late.

What a poll spends

A program that asks a news API every 30 seconds learns of a headline 15 seconds late, on average, and 30 seconds late in the worst case. Fifteen seconds is the entire window in which Busse and Green still measured a significant after-cost return on positive Midday Call mentions. A poll every minute spends the whole of Ederington and Lee's adjustment. By the time the row is in your process, their first minute is over, and the second minute is a coin flip relative to the first.

That is an argument about delivery, not about alpha you can invoice. Ederington and Lee did not think someone off the floor could trade the first minute even in 1993. Busse and Green's 16 basis points is a sample average from television in 2000, after a spread, on one segment, and it is not a forecast. The point for a builder is narrower. If the product is "tell me when the headline exists," a poll inserts a delay as long as the event the journals measured. If the product is a archive, a page, or a model that reads the release after the close, REST is the right call and the delay does not matter. The REST and WebSocket guide is the build choice. This post is the reason the choice has a clock.

What to open on a stock news WebSocket

On MarketNewsAPI the live socket is wss://ws.marketnewsapi.com/v1/stream. The article is sent when it is ingested. The summary is written after that, and it is null until enrichment finishes. A second socket, /v1/stream/summaries, sends the article once the summary exists. If the job is the clock in this post, use the first socket. Waiting for the summary spends the minute Ederington and Lee measured, on top of whatever the wire itself took.

Channels are OR. A macro screen and one equity watchlist can share a connection:

{"action":"subscribe","channels":["category:macro","category:forex","ticker:AAPL"]}

Stocks are the default category on REST. Pass category=macro or category=forex when you want those feeds, and leave it off when you do not. Pro includes the WebSocket, up to 3 connections and 200 channels on a connection. Free is REST only, delayed 15 minutes, which is the wrong side of every number in the table. Paid Pro is $15 a month, or $162 a year. The 7-day trial is the live feed. None of that is a commercial license. A product you sell needs its own agreement.

The headline is still not the release. A morning post on this blog, what a stock news API is for once the release is public, is about reading the figures after the verb. This one is about when the verb arrives. Open the WebSocket docs for the connection, and the channel list for the prefixes. The 7-day Pro trial is the live socket if you want to time it against a release you already know is coming.

Frequently asked questions