A cautionary tale for quants and HFT designers

This blog has always been an advocate of using human intelligence as oversight over quantitative investment systems. In particular, I have been concerned about the takeover of trading the HFT robots.

On this day that America decides who gets to have his finger on the nuclear button for next four years, consider this story (from Business Insider) about how close we came to disaster in 1995:

In the 67 years since the detonation of the world's first nuclear weapon there is only one time the so-called nuclear briefcases were broken out and opened up. On January 25, 1995 they not only opened, they nearly launched Russia's nuclear arsenal at the United States.

As the story goes, the Soviets mis-interpreted the launch of a Norwegian test rocket as an ICBM attack:

When Norwegian Kolbjørn Adolfsen gave the nod to send a Black Brant rocket from the Andøya Rocket Range off the northwest coast of Norway to study the aurora borealis, he wasn't concerned at all.

Sure the Brant is a large, four-stage rocket that would fly to 930 miles above the earth near Russia, but he'd contacted the proper Kremlin authorities and hadn't given the flight a second thought.

What Adolfsen didn't know when he left the rocket base shortly after the missile was launched, is that the Brant's radar signature looks just like a U.S. sub-launched Trident missile.

Uh-oh...

The radar operators at Russia's Olenegorsk early warning station promptly reported the incoming missile to their superiors, but not a soul on duty within the military had been notified of Adolfsen's plans.

The officers at Olenegork believed it could be the first leg of a U.S. nuclear attack.

Four years after the Berlin Wall came down and Russia was in the throes of change, stable systems had been demolished and replacements had yet to fall into place. One thing that had gotten only more developed since 1991, however, was the Kremlin's mistrust of the United States.

So as the Brant streaked its way near Russian airspace, military officers had to decide if this was an electro-magnetic pulse attack that would disable their radar and allow for a full on American attack, and what they should do about it.

The matter was decided when the Brant separated, dropped one of its engines, and fired up another. The radar signature now looked so much like a multiple re-entry vehicle (MRV), a missile carrying multiple nuclear warheads, that military officers no longer had any doubt.

Time was getting short to respond:

There were now five minutes during which the missile's trajectory would be un-tracked by Russian radar, and when it could strike Moscow; a slice of time that was devoted to deciding whether to launch a counterattack.

Fortunately, the Soviets didn't have a hair trigger launch on warning policy, but actually had adults in charge:

Boris Yeltsin was alerted, and immediately given the Cheget, the "nuclear briefcase" that connects senior officials while they decide whether or not to launch Russia's nuclear weapons. Nuclear submarine commanders were ordered to full battle alert and told to stand by.

Apparently Yeltsin doubted the U.S. would launch a surreptitious attack and within five minutes, Russian radar came back confirming the missile was heading harmlessly out to sea.

Now consider what might have happened if algorithms were in charge? The missile signature looked like it was carrying a multiple warhead. What if the algos had decided that it was a hostile launch. What would it have done?

This a cautionary tale for HFT algo designers, who claim that their systems have numerous safeguards. Safeguards can and do fail. We found how an unexpected perturbation during the flash crash created enormous systemic problems. Fortunately, the only thing that people lost during the flash crash was money.

If algos had been in charge of the Soviet defense system in 1995, we all could have lost our lives.



Cam Hui is a portfolio manager at Qwest Investment Fund Management Ltd. ("Qwest"). This article is prepared by Mr. Hui as an outside business activity. As such, Qwest does not review or approve materials presented herein. The opinions and any recommendations expressed in this blog are those of the author and do not reflect the opinions or recommendations of Qwest.

None of the information or opinions expressed in this blog constitutes a solicitation for the purchase or sale of any security or other instrument. Nothing in this article constitutes investment advice and any recommendations that may be contained herein have not been based upon a consideration of the investment objectives, financial situation or particular needs of any specific recipient. Any purchase or sale activity in any securities or other instrument should be based upon your own analysis and conclusions. Past performance is not indicative of future results. Either Qwest or Mr. Hui may hold or control long or short positions in the securities or instruments mentioned.