jcowan_at_REUTERSHEALTH.COM wrote on 2003-12-22 22:22 UTC:
> Consider a Dutch auction: the price is lowered in a predictable way, and whoever
> bids first buys at the current price. In order to handle this in a distributed
> environment, we must be able to determine who bid first accurately.
I have yet to see even the slightest plausible argument, for why
financial transactions need to be timestamped in a distributed way with
an accuracy better than a few seconds. Even in the markets where demand
and supply change fastest, namely electricity and telecommunications,
supply and price adjustments can be negotiated comfortably on a minute
per minute basis, because demand over time is well characterised by a
low-frequency signal (main spectral energy below 100 millihertz) plus
high-frequency noise with well-understood autocorrelation.
What I'd like to see is an economic argument that mankind at large
actually gets some real benefit (improved productivity, increased
stability of markets, etc.) out of a global subsecond legal and trading
infrastructure. Legal contracts and trades in the end reflect changes in
human expectations, plans and desires. The human brain can't refocus its
attention from one object to another faster than about 200 ms (a timing
tolerance that was referred to in the introduction of DUT1), which gives
us an absolute lower hardware bound for how fast human desires could
possibly change over time. Concious decisions involve evaluating a
series of options and take at least several seconds. Then there is an at
least 300 ms delay from any concious decidion to a measureable reaction
of a human actuator (the famous P300 EEG signal before the finger
presses the decision button) in the best case, which extends to several
seconds at least in more complex practical situations.
I know more about distributed computing and neurobiology than economics
and "real-time" arbitrage trading, but my instrincts tell me that there
is no real need for subsecond timestamping in trading. Accellerating
markets to enable them to make major supply and price changes
automatically with subsecond resolution just shortens the timeconstants
of the control loop equations involved and removes the ability for
humans to oversee what is actually going on. That sounds like calling
for instability and increasing the noise amplitude, which I doubt can be
justified with any actual economic benefits. Rules such that all
computers involved shall be synchronized to UTC within about 1 second
accuracy and all contracts timestamped within 3 seconds count as legally
simultaneous sound perfectly suited to me to fullfill actual
requirements.
I suspect the actual requirements in trading are more for exact and
unambiguous rules and regulations for how to deal with and interpret
timestamps, rather than for actual subsecond accuracy. I don't expect to
see caesium fountains being installed in Wall Street offices any time
soon.
Markus
--
Markus Kuhn, Computer Lab, Univ of Cambridge, GB
http://www.cl.cam.ac.uk/~mgk25/ | __oo_O..O_oo__
Received on Tue Dec 23 2003 - 03:16:07 PST