Acting on unverified data: the signal that was a lie

Graduate · enrolled learners

This lesson opens with The AP Twitter hash crash, 23 April 2013.

What happened

Hackers took over the Associated Press Twitter account and posted a false message that there had been explosions at the White House and that the president was injured. Within about three minutes the S&P 500 fell about 14 points, roughly one percent, briefly wiping out an estimated 136.5 billion dollars of value by Reuters' count, before the tweet was exposed as fake and the market recovered just as fast. Traders blamed automated trading for the speed of the fall and the rebound; Nanex's study of the tape found the first high-frequency firms reacted about 15 seconds after the tweet, so people may have started the selling and machines amplified it. Either way, nothing that sold could tell the source had been compromised: a message from a trusted account was treated as a trusted signal. Nothing about the trades was a bug. The failure was upstream, in taking a data source's authenticity for granted, so a single lie injected into a trusted feed became very real trades.

The decision point

A trading system that acts on a data source is trusting not only the content of that source but its authenticity, and if the source can be compromised, spoofed, or simply wrong, the system will act on the lie at machine speed, before anyone can intervene. The AP hash crash is the case: a hacked but trusted account's false message moved the S&P 500 by an estimated 136 billion dollars in about three minutes, with automated selling widely blamed for the speed, because nothing that sold had a way to know the trusted source was lying. This sharpens the prior lesson: it is not enough for a feed to be up and fresh, it must also be telling the truth, and a system that treats any message from a normally-reliable source as automatically trustworthy inherits every way that source can be compromised or mistaken. The speed that makes automated trading powerful is the same speed that makes an unverified signal dangerous, because there is no human in the loop to say that headline cannot be right before the orders are already sent. So the discipline is to treat a signal's provenance and authenticity as part of its validity, not just its content: to weigh how much a source can be trusted, to require corroboration for a signal extreme enough to move the book, and to build in limits and human confirmation for actions large enough that being wrong is catastrophic, because the AP hash crash showed that a system fast enough to trade a headline in milliseconds is fast enough to trade a lie in milliseconds, and the only defense is to have doubted the source before the message arrived.

Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.

It is free. We do not sell the list and there is nothing to buy at the end of it.