high-frequency-trading

Yesterday’s AP Twitter account hack raises some interesting questions and some potentially easy answers.

This has nothing to do with Twitter

In reality it’s not a Twitter thing, an account got hacked and a piece of untrue information was posted. Also this can’t be put down to Associated Press.

It’s the 84%

“Only about 16 percent of stock market transactions consist of what most people think of as buying or selling of company or mutual fund shares (“real” investors, interacting with actual brokers). The rest, according to analysis by Morgan Stanley’s Quantitative and Derivative Strategies group and covering October to December 2011, were performed by computers acting automatically” (FT – Link)

You can’t even call this event a “black swan”, well not unless the swan was 60ft tall wearing a tinsel necklace and hotpants. The simple deduction is this: without any form of verification, checking, correlation against another source a high frequency trading network reacted on a single tweet. And that’s the real issue here.

Keeping in mind that it’s fair to say that 60% of tweets are utter junk this raises some serious questions.

First off, one simple check. Confirmation. The story appears on a tweet on the @ap account (now no more) but doesn’t show up on AP’s actual news feed. This doesn’t need big data, complex algos or data scientists. All it requires is a small amount of common sense.

Secondly, getting more complex. Events in time have a certain amount of huh factor that can affect the markets, I accept that. But these fall under several grades of “little stock sees 25% increase in profits” to “an explosion at the White House that injured the President”.  Sentiment is easy to check, structure of sentence against other tweets from the same account is much harder.

As HFT networks buy and sell within the blink of an eye there’s little you can do. Add to that the potential for a Lorenz moment and anything could have happened. And like previous episodes of spiralling algos the only thing you can do is halt, reboot and hope the system corrects itself.

A few ideas….

Every person has some ranking of influence, putting Klout and Peerindex to one side for a minute, so it would be easy to weight each user’s twitter account. Some carry more weight that others. With this comes an amount of verification and responsibility.

So for example:

@jasebell has an influence ranking of 50/100

@cimota has an influence ranking of 70/100

@mmarymckenna has an influence ranking of 70/100

@david_crozier has an influence ranking of 70/100

and so on and so on.

I’m assuming that HFT systems are just acquiring data from the firehose, so there’s no idea who is really posting what and nor does it care.  As long as the data is flowing and the sentiment works out then it’ll make trades regardless.

Weighting is important though. I can’t really tell anyone to go and seek funding for example, I’d redirect them to Matt (@cimota) or Mary (@mmarymckenna) as they know far more about what’s going on than I do.

If someone wanted to know about cyber security things then I’d certainly rank David’s information (@david_crozier) far better than mine. He’s in that domain and I’m not.

And for the other 16% here’s a snowball

Well it was too late. Stop losses were triggered before anyone knew anything about it hence the drop. The humans didn’t stand a chance as the reaction time is much slower. In a Bloomberg report Barry Schwartz with Baskin Financial Services and others found the whole episode “quite scary”.

“Don’t let computers rule your investments,” Schwartz said.

It’s a bit late for that sir, that’s all I can say.

I agree with the sentiment though. Once the trigger starts then the snowball will start. This time we got off lightly.

 

Advertisements