Sentiment Spam

[danohu]

Stock trading and the like have always been at the forefront of data-mining -- though not often sharing their techniques, for obvious reasons.

The current trendy data-mining topic* is sentiment analysis based on social media -- guessing what the world thinks about a topic by searching for positive or negative opinions about it on twitter &c. Roughly, searching for "I love X" versus "I hate X", and interpreting that as a sign of general opinion.

There are surely traders basing decisions on sentiment analysis. It's anybody's guess how many, or how seriously, but it's going to grow over time.

So when is the spam coming?

Go short on company X. Spam twitter with 'X sucks' messages. Wait for other traders to use sentiment analysis, see X is unpopular, and dump their shares. Buy cheap. Profit.

You maybe couldn't affect a major company like this -- the market isn't *that* stupid. But suppose you know another trader is using sentiment analysis, and have a hunch that you can make her buy or sell by dumping enough positive or negative opinions online? Isn't that a strong incentive to spam?

[this inspired by a post suggesting that you predict layoffs by seeing whose employees are updating their CVs on linkedin -- an idea so sensible that it's probably already being used by a dozen companies]

* or rather, trendy among in the world of starry-eyed startups -- there's somewhat less academic interest. Probably because it produces results which are (a) easy to interpret, and (b) utterly unreliable.

Comments

[feanelwa.livejournal.com]

Your link about the CVs on Linkedin is broken...

[oedipamaas49.livejournal.com]

thanks, belatedly. Fixed.

[gerald_duck]

The main excitement about sentiment analysis in the financial markets isn't scouring the Internet for extra sentiment, it's analysing existing sources more quickly — especially, forming a conclusion faster than a human could react.

If a company issues a press release talking down its prospects, there's considerable advantage to being the first to sell your holding in that company.

[oedipamaas49.livejournal.com]

That makes sense; thanks for the explanation.

It's sad that so much effort comes down to an advantage of a couple of seconds, but I guess that's how the world works.

[gerald_duck]

A couple of seconds? Nowadays things move so quickly that people perform statistical arbitrage between the London and New York stock exchanges based on their being twenty light-milliseconds apart — i.e. they buy a stock in London, in the expectation that the price will be higher in New York when news of the purchase reaches NYSE.

Oh, and a certain major manufacturer of Ethernet switchgear makes a lot of money out of selling guaranteed-fair switches to stock exchanges, that will deliver a stock tick simultaneously to every port. Given that automated trading systems have wire-to-wire response times down in the hundreds of microseconds now, that matters.

[gerald_duck]

(And yes, they do have systems that in theory can pull the plug on a trading system if the market goes into meltdown, lest a stock market crash that would once have taken hours occur in a split centisecond.)

[oedipamaas49.livejournal.com]

Fascinating.

I intermittently wonder how you can set up systems to extrapolate back from the what's happening on the markets, to figure out the meaning of events in the real world.

So if you know exactly when something happened, and which stocks suddenly changed in value that second, you can make a decent estimate of which firms are going to be affected how much, and so of what the event really means. Since the traders (and their techies) are smart, well-resourced, and motivated by money rather than ideology, they'll probably give you a better understanding than you'd get by reading the pundits.

[gerald_duck]

It's tricky. The price of a stock will depend on sentiment about various groups it's in, as well as performance of the company itself and the amplifying effect of people reacting to what the stock is already doing. (If a stock falls in value, many people will sell, even if they can't see why it's falling, in case everyone else knows something they don't.)

Frankly, I scarcely understand what's happens to my employer's share price, even when I know what's really going on. (-8