A SIMPLE PROPOSAL TO KILL HIGH FREQUENCY TRADING IN THE STOCK MARKET
Due to technological possiblities, success in stock market trading
has increasingly become dependent on being at the front of a rat
race in software, computing capacity and fast optical cable
connections. While the firms whose stock is traded obviously do not
change their prospects over time horizons shorter than days or even
months, to win in the stock selling and buying game, today you have
to act and react on a time scale of fractions of milliseconds.
Automated high frequency trading (HFT) enables this. This race has
now become so absurd that even the
business press and financial regulators and pundits have become
critical to it. One technical measure against HFT that has
been implemented by a new stock exchange, IEX, is that all
traffic go through a roll of cable (!) that delays the signals
so much that HFT trading cannot profit parasitically from orders
given to that exchange.
Here follows a simple proposal which does not depend on any changes
to physical infrastructure, can be mandated by regulators and
implemented easily at any exchange, and which enables not only the
blocking of trading on millisecond scales, but can remove trading on
any time scale that is considered too short. Being implemented as
software, it also has the advantage that its parameters can be
easily adjusted based on how the system performs. This solution does
not presuppose any transaction fee, and it impacts small and big
trade(r)s in the same way.
The proposal is inspired by the signal processing literature, more
specifically the concept of a digital finite impulse response
("FIR") filter, known in the wider community as a moving average
filter. The FIR filter outputs a weighted average of a finite number
of earlier inputs. The output becomes a time sequence that has the
high frequency components attenuated, and it is slightly time
delayed, where we define the delay as the half of time interval back
to the oldest input used in the filter. The filter can be seen as a
moving "window" that smooths the input signal. The time breadth T of
this window is essential for how the filter works. The larger T is,
the stronger smoothing of the input signal, equivalently: the
stronger removal also of lower frequency components.
Let us construct an example to explain how the filter works. We
choose the time window T = 4 minutes (firms' prospects of course do
not change much during that time interval either, but we don't need
to make the interval larger and ignite unnecessary quarrels with the
regulation-hostile financial community, because T = 4 minutes is
sufficient to eradicate HFT).
Running time is called t. At t = 0 agent A buys 4 shares from B at
100$ each. Time goes until t = T/2 = 2 minutes. The server then
executes the final settlement of A's and B's trade, which occured 2
minutes earlier. This is done by checking all trades of the same
stock within the filter's time window, from t = -2 to t = 2. Assume
for simplicity that within this interval only two other events
occured (in the real world it would be a lot more, but this is no
problem for a computer): agent C bought 8 shares for 90$ at t =
-1.5, and agent D bought 2 shares for 110$ at t = 1. We don't need
to consider the exact times, only that the two events were inside
the time window that has A/B's trade in the center. The filter
outputs an adjusted price for A and B:
(8 x 90 + 4 x 100 + 2 x 110) / 14 = 95.71 $
At t slightly more than 2, the stock exchange server settles the
difference with A, in this case A is paid back 4 x (100 - 95.71) =
17.14$. And B has to pay the stock exchange the same amount.
The stock exchange had to pay out a sum, and it thus needs a buffer
for such operations. But this buffer will all the time be
replenished; every payment out is mirrored by a payment in.
Such a moving average filter should make HFT - which is technically
very expensive - unprofitable, and it will die out. It is also, as
already mentioned, a tool that can be additionally used to
discourage day traders who operate on a minutes time scale, simply
by making the time window T larger.
Essentially, one is forcing a slight bit of solidarity on the agents
doing the trading. The advantage of being first or an insider is
slightly reduced, and those close after is correspondingly better
off. Buying and selling the same stocks within short intervals
become much less attractive. But long-term investors will be
negligibly impacted by such a system. Therefore, one could say that
the above described filter has a second effect: to reduce
speculative trading without damaging other activity.