3

I've been trading, in a very amateur way, in the stock market for a year now. I've noticed something pretty obvious. When some evident good news appear on a stock, the price goes up. So it goes the other way around. For example when news appear that a big investment company has changed the strategy for a certain stock from "strong buy" to "strong sell" the stock price drops.

Taking this in mind, why couldn't one be constantly checking on the volume of all stocks in the market, and when an unusual fluctuation occurs check the news for potential evident good or bad news (Such as the example, or beating by far the expected returns for the quarter) and the buy or short the stock until the end of the day?

The potential pitfall that I see is that the market doesn't respond as one expects to the news, or that one may get too late in, but when the news are so evident, and an algorithm for checking them on time is deployed, can this be avoided?

(Sorry for the English, feel free to edit!)

Edit

Considering rhaskett answer I wanted to post a specific case in which this works, note that the question is if it can be generalised to most cases.

This is today and yesterday BBRY chart (12/11/2014 - 13/11/2014)

BBRY chart

There's a big spike on volume at 12:00. If the algorithm was slow, by 12:30 a notification could be received. The news of BlackBerry-Samsung partership where posted at 12:27. It isn't the most evident good news. But my point is that the price change came after some of the volume spike. In spanish we have a saying that could translate to: "After the war, everyone is a general".

To avoid making the question too open, does this not happen in a lot of cases? Is it too hard to discern evident good or bad news?

Improve this question
2
  • 1
    While full articles hit the internet twenty minutes later. The press release was actually at 12:04pm marketwatch.com/investing/stock/bbry/news
    – rhaskett
    Nov 14, 2014 at 18:29
  • While in this case the strategy might have worked as long as you didn't continue to hold until the next day, I encourage you to study many cases as keshlam's third challenge below applies more often than you would think. Also, I encourage you to do more research into natural language processing, it remains a hard problem even to just distinguish good/bad consistently much less the problem of how good which is really what is necessary for an informed decision.
    – rhaskett
    Nov 14, 2014 at 18:45

3 Answers 3

Reset to default
5

First challenge: Creating a system which can understand written English well enough to read the news. Nothing short of IBM's Watson has proven very good at extracting meaning from unstructured text.

Second challenge: By the time it reaches "the news", the big actors already know and have responded.

Third challenge: It's not uncommon for a stock to drop on good news, or rise on bad, because the price had previously adjusted to an expectation of even better/worse news and is now correcting itself.

Basic principle: It it was simple and obvious, everyone would already be doing it.

Improve this answer
2
  • By algorithm I meant a specified set of steps to check the news on time, but not necessarily by a computer. I agree fully with point 2 though. Point 3 doesn't convince me much, but it could be backtested
    – Javier
    Nov 13, 2014 at 22:31
  • I've heard #3 happening fairly frequently on the radio's market report, and I barely pay attention to that report. May be more common with large companies than with smaller ones. But the point remains that the market just Ain't That Simple; price shifts are less rational, and less tied to the actual value or productivity of the company, than they used to be. And they weren't all that rational even then.
    – keshlam
    Nov 13, 2014 at 22:35
0

People are trying ideas like this, actually. Though they generally aren't very public about it. While keshlam ventures into hyperbole when mentioning Watson, he is certainly correct human language parsing is a extremely hard problem.

While it is not always true that the big players will know before the news (sometimes that would qualify as insider trading). The volume spike that you mention generally comes as the news arrives to the major (and minor) players. So, if you have an algorithm run after the volume spike the price will likely have adjusted significantly already.

You can try to avoid this by constantly scanning for news on a set of stocks however this becomes an even harder problem. Or maybe by becoming more specific and parsing known important and specific news sources (farm report for instance) and trying to do so faster than anyone else. These are some methods people use to not be too late.

Improve this answer
0

I wouldn't be turned off due to the difficult of parsing English, for a few reasons.

Firstly, you don't have to perfectly parse to find meaning. You can look for keywords and write some algorithms to approximate, and of course if you get enough of a statistical advantage (and can repeat it) you can make money.

Second, it probably isn't long before third-party software is made available either to do something like this or to provide a framework for it. In fact, it probably already is available somewhere. (Note the influx of Silicon Valley types to New York as more machine intelligence is applied to trading and journalism.)

Thirdly, as hinted by the mention above of journalism, there's already software using numerical data to write pretty human articles. Some are pretty robotic and you can catch them (I noticed one and searched for a key phrase to discover several very much like it, each having a different fake author name). This will mean not only a continued improvement of parsing but also more push for more data to be released in machine-readable formats, such that press releases will be increasingly parsible.

Finally, to vindicate your idea, the keyword approach has been done with some success. Try this link and note the additional links on the same topic.

If you have the time and processing resources, you might like to try your idea by training a neural network to find correlations of keywords (and phrases -- that's important, too) with trends in the market.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .