Search
× Search
Tuesday, April 23, 2024

Archived Discussions

Recent member discussions

The Algorithmic Traders' Association prides itself on providing a forum for the publication and dissemination of its members' white papers, research, reflections, works in progress, and other contributions. Please Note that archive searches and some of our members' publications are reserved for members only, so please log in or sign up to gain the most from our members' contributions.

Mini flash crashes caused by sloppiness, not machines

photo

 Trevor Neil MSTA MCSI, Technical Analyst and Portfolio Manager @BETAfinancial

 Wednesday, June 19, 2013

By John McCrank NEW YORK, June 18 (Reuters) - Unexplained rapid price drops in single stocks have generally been triggered by human error, not nefarious trading activity or high-speed trading algorithms gone wild, an official at the U.S. Securities...


Print

5 comments on article "Mini flash crashes caused by sloppiness, not machines"

photo

 Trevor Neil MSTA MCSI, Technical Analyst and Portfolio Manager @BETAfinancial

 Monday, June 24, 2013



The Volkswagon/Porche run up and fall was not a 'flash crash'. Although both parties used computers to execute their business it was a titanic squeeze battle between two parties with tragic consequences. More sloppiness than computer driven.


photo

 private private,

 Monday, June 24, 2013



@Raphael, @Don

I meant the chaos with respect to the problem of behaviour of a multithreaded software.

@Raphael

Cant you consider an ecosystem "just" a complex chaotic system? "Simply" put ^-^ .

@Don

I thought one can find great parallels (pun intended ^_-) between HW and SW multithread design. In IC design with growing chip area the clock distribution poses risk of race cond. so there are tendencies of also using asynchronous design that eliminates the problem. But this is maybe the biggest difference, SW does not utilize any Clock construct (unless u code ur parallel code in assembler(?)) and the race is due something else there.

Another difference is probably (? )that in HW it is easier to test for all 100(0..) of thousands of interleavings to check for disaster. Since you can monitor the state exactly (the current interleaving), probably even 'inject' a desired interleaving and you can repeat the test trial billions of times a second.

I did only some simple multithreading and even there some interesting challenges arose. So I can appreciate your work.


photo

 private private,

 Monday, June 24, 2013



@Antonio @Trevor

Porsche did their homework. And when their 70%+ holdings of VW came to light a killer short squeeze was born. The question is if the sneaky accumulation was kosher, it must have been really cleverly executed or just another dirty Swiss game. IMHO the Swiss cheese has way too many holes in it.

On the other hand Adolf and the others should not have been too greedy, overconfident and mean (= sloppy as Trevor said) and shoulda limit their risk instead of just playing with outright stock short (or as my grandma used to say:one should play only with the billions one can afford to lose)

... and BAM right in the kisser.

The funny part is the shorts were dealing with mostly one entity, Porsche. Borrowing from and selling the stock to Porsche and later on covering with the merciful 5% of stocks Porsche decided to put back on the market to calm it down. How nice.


photo

 Trevor Neil MSTA MCSI, Technical Analyst and Portfolio Manager @BETAfinancial

 Monday, June 24, 2013



I think that poor Aldolf probably did his homework too before he bet his sizable fortune on that trade. VW just had a card up its sleeve he did not know about.

The big point is, with regard to mini-crashes (and this was surely one), it was not caused by machines but by people. They may have used machines to execute their business but its was not the machines which caused the convulsion. This is essentially the point of the report. The trades count in the '70%' of all trades done by machines number. Machines get the blame but its most often people behind the mini-crashes.


photo

 Don Sannella, Professor of Computer Science at University of Edinburgh and Founder/CEO of Contemplate Ltd

 Monday, June 24, 2013



@Ondrej, like you say, a big difference between HW and SW is the absence in SW of a clock. In SW, you have no control over the interleaving chosen by the scheduler, or over the speed of propagation of changes in per-core hardware caches to main memory. Also the number of interleavings rises steeply with the number of threads and the number of events per thread. (For n threads with m events each, there are (nm)!/(m!)^n interleavings.) So exhaustive testing of interleavings is not realistic.

Please login or register to post comments.

TRADING FUTURES AND OPTIONS INVOLVES SUBSTANTIAL RISK OF LOSS AND IS NOT SUITABLE FOR ALL INVESTORS
Terms Of UsePrivacy StatementCopyright 2018 Algorithmic Traders Association