Re: Distributed stock trading system

[Bill Stewart <bill.stewart@pobox.com>]

At 04:38 PM 1/19/98 -0600, Igor Chudov @ home wrote, but not in this order:
>I would like to hear your feedback on a project that I have just
>thought about.
>It is called a Distributed Asset Trading Network. It consists of
>Traders and Nodes. They trade Products (asset classes). A Product may
>be an Intel share, or a bond, or an option on S&P 500, or whatever. Who
>and how defines what the Product unit is is beyond the scope of this

Interesting gedanken-project.  Some related work is Bob Hettinga's 
geodesic networks and (Mark Miller?)'s agorics.

>The technical issues that are open, at the moment, are 
>1) clock synchronization, 
Long since solved - use NTP.  Assuming the nodes don't need to be
stealthy, they can just clock off each other; if they need stealth,
they should clock off some trusted worldwide internet providers.
If they need to be blazingly invisible, use a GPS receiver clock.
(Using NTP with each other makes it easy to estimate distances,
but that's a risk any time you've got uniform clocking and
reasonably fast-turnaround communication.)

>2) the need for a central authority to clear liability issues and 
You've described the Node as a "clearing corporation", with bonding
to cover liability.  The corporation and central authority both work
against your goal of "a) independent of borders and national laws."
Also, assuming that each Node posts bonds in about the same amount
as it accepts bonds posted with it, the bonds still don't provide
an incentive not to abscond with the money, since it steals about
as much as it forfeits in bond; expected positive future cash flow 
is what discourages the Node from absconding.  Posting bonds with
a central authority avoids this problem, but a central authority
is both a target for governments and for freelance thieves,
and a temptation for its operators to abscond.

>3) potential denial of service attacks through acquiring too
>many locks (again, maybe with the proper liability agreement, 
>it is not an issue).
That's a job for careful implementation; you don't want customers
taking down nodes either.

You didn't address a major problem:
0) Scalability
Talking to 10 other nodes in realtime is easy.  10000 is hard.
Not only do you need the bandwidth, but you've got to worry about
nodes that are down, or that you temporarily can't communicate with,
and you've got to be fast enough to respond to all their demand,
maintain state for your crypto sessions with them, recover from
your own outages, and deal with the queuing if they've all decided
that you've got the best price for something this second.

The more nodes you talk to, the harder it is to scale.
But the fewer nodes you talk to directly, the slower it is to
propagate information about prices, bids, and offers, and therefore
the harder it is to get prices to converge - since you discuss
eliminating arbitrage, bid-ask spreads, and expensive intermediaries,
this is a problem.  Also, assuming the system has convenient room
for negotiation, you'll still have time delays as people are deciding
what to do, so there's still arbitrage to be made, and depending
on your liability structure, there's probably room for low-balling
and high-balling on bids.


				Thanks! 
					Bill
Bill Stewart, bill.stewart@pobox.com
PGP Fingerprint D454 E202 CBC8 40BF  3C85 B884 0ABE 4639