Nice posts. You see 0 value to the latency arms race, and I think you are probably right.
However, I think it's not hard to argue that some speed increases have value. For example, if the world's only stock market only allowed trades to happen n times a year, we would expect a pretty inefficient market and this mechanism for capital allocation would be pretty poor. And as we increase n, market efficiency would increase, and this has value.
So are we at a point right now where an increase in n has no value (say market information gets processed on the order of nanoseconds instead of the current microseconds)? What's the cutoff point?
This is a good question and one I don't have a good answer to. I don't actually know the optimal points for most of these values and I'm generally content to let the market decide.
This is why I think eliminating the subpenny rule is a good policy change - it eliminates a market distortion while not requiring you to perfectly tune a parameter (e.g., the size of a transaction tax or a minimal trade latency). All it does is allows price competition.
Of course as a real policy proposal it's dead in the water, since it doesn't provide much opportunity for moralizing or posturing.
> You see 0 value to the latency arms race, and I think you are probably right.
How do you draw the line between investments in reduced latency which are socially beneficial, and those which aren't? It's not like these are cheap investments that HFT firms are making. If they're truly pointless, we should expect them to lead to overcapacity and ultimately the exit of some players (unless, of course, the arms race is being unnaturally subsidized in some way, e.g. by the lack of a sub-penny rule).
When Paul Reuter was using carrier pigeons and telegraphs to move tradeable information between stock exchanges ahead of his competitors, I wonder if people considered this an arms race of negligible social value. After all, the world had gotten by just fine without telegraphs up until then.
You've conflated "value to society" with "potential to make money". The two are unrelated. Dumping toxic waste in the river may make money, but has negative societal value.
Whatever societal value there was to faster trading was exceeded when trades could execute by the time your browser refreshed. Humans make financial decisions on the scale of minutes at fastest.
At this point HFTs are still pursuing "making money" (and there may still be money to be made) but certainly not providing any societal value whatsoever; since they are stealing from the general public their societal value is strongly negative.
Meaningful stuff could happen in the elapsed time, so motivating people to get info there sooner was more valuable. As someone who's working in HFT, I think that the surface defined by a given expenditure in (expenditure on price accuracy)x(expenditure on speed)x(profitability) space, the gradient points too far toward speed on too much of the surface.
Saying something that is such a zero-sum waste is great because as a side effect it brought us the telegraph is like saying government bureaucracy is great because firms have had to develop new technologies that happen to be useful in other domains in order to deal with it.
> So are we at a point right now where an increase in n has no value (say market information gets processed on the order of nanoseconds instead of the current microseconds)? What's the cutoff point?
My intuitions want to say the point of diminishing returns is around the limits of human cognition. Stocks are ownership of some human activity, so once you start valuing them faster than human activity you're really redefining what stocks are. Now stocks become predictions on ownership of some human activity.
Since the first formalization of Sport (some might argue earlier), it's been a human tendency to go Faster, Stronger, Better. Moving from ownership of some human activity to predictions on ownership of some human activity is a manifestation of this Faster, Stronger, Better (since the prediction is itself a human activity... we're owning the human activity of predicting human activity).
All that is not to say any of this is inherently good or bad (I disagree at the loaded term 'rigged' being used by Lewis). There's no single cut-off point in additional value earned. It's more of a transition zone where you're not sure when exactly you left it, but at some point you realize on the other side is not the world you started in.
I don't see why the speed of human cognition should have anything to do both it. After all, the HFT algorithms are the result of human cognition. It's just computers "doing human cognition" much faster than human brains can, which is what all computers do.
However, I think it's not hard to argue that some speed increases have value. For example, if the world's only stock market only allowed trades to happen n times a year, we would expect a pretty inefficient market and this mechanism for capital allocation would be pretty poor. And as we increase n, market efficiency would increase, and this has value.
So are we at a point right now where an increase in n has no value (say market information gets processed on the order of nanoseconds instead of the current microseconds)? What's the cutoff point?