HFT has connections to both and can see the cross. Everyone is better off.
This might be different in the US due to NMS, but HFT strategies are all pretty similar: they can provide liquidity better if they are fast, and other people can take advantage of that.
I built a trading platform ~6 years ago that had a 100 microsecond tick to trade (a term that means from the time a market update hits your nic to the time an order leaves your nic).
We knew then that we were too slow for large swaths of the HFT space & wouldn’t even attempt those trades.
I haven’t kept up with the industry but I’d guess tick to trade times at the speed edge are in the low hundreds of nanos now. In an environment like that a 1ms time improvement is out of the capabilities of your software/hardware engineers and can only happen at the network.
https://www.youtube.com/watch?v=L5cZaIZ5bWc
And that was more than 6 years ago, one ticker symbol only. You can imagine now.
That gave me a much better sense of scale than hearing "a trade takes 200ns" or whatever. The engineering that goes into making this possible must be just insane.
In trading you're not calculating nearly as much stuff.
In trading it's probably very sharp, the line between making and losing money. You want to be under 100us for sure. That's micros. Depends on what you're doing of course but 10ms is OOM to slow.
There's a similar article at When The Speed Of Light Is Too Slow: Trading at the Edge ( http://www.kurzweilai.net/when-the-speed-of-light-is-too-slo... )
The paper that both of these articles are based on: Relativistic statistical arbitrage https://www.alexwg.org/publications/PhysRevE_82-056104.pdf which opens with:
> Recent advances in high-frequency financial trading have brought typical trading latencies below 500 µs, at which point light propagation delays due to geographically separated information sources become relevant for trading strategies and coordination e.g., it takes 67 ms, over 100 times longer, for light to travel between antipodal points along the Earth’s surface. Moreover, as trading times continue to decrease in coming years e.g., latencies in the microseconds are already being targeted by traders, this feature will become even more pronounced.
You don't get any benefit from putting your trading system half way between the two exchanges, you colocate it with one or the other (or both). Likewise if there are three exchanges to trade on, you don't build a datacentre at the circumcentre of the triangle or whatever voodoo geometry nonsense. You colocate at each of the three and buy/rent/build a line for each of the three point-to-point connections. Source: worked in HFT for more than 15 minutes.
And are the reactions automated based on some NLP? Milliseconds wouldn’t matter if a human has to still parse it, correct?
Acting on M&A news by taking a more aggressive position generally requires more confidence than, say, backing off based on ‘bad’ news. In the first case, you want to be pretty sure of the headline before taking action. In the latter, the mere presence of words like ‘fraud’ and ‘investigation’ might be enough to play it safe and say, “we’re going to back off trading this underlying until a human has a chance to look at this.”
Now let’s say there’s negative news on the Chinese economy. Someone could be running an algorithm that sells down companies with Chinese exposure. But, inherently, there likely is a much lower probability that this news is uniform or material across companies with China exposure. A strategy like this likely would get put on small positions and get in and out quickly as it could get run over depending on how humans process the information and make bigger discretionary bets.
There definitely is a risk too all of this. Roughly 10 years ago, somehow an old united Airlines bankruptcy article hit the wire and the stock sold off hard. Eventually, United released a statement saying the article was innacurate. This isn’t typically the type of strategy you would make large, single bets with, or hold the positions over a long period.
One tends to carry more serious consequences when you’re wrong. Accordingly, the ‘take’ path generally has more checks, and more opportunities to abort.
Yes, you might misread here or there. But this is a game of statistics and the real win/loss is whether you read correctly more often this read incorrectly. You want your net wins to be positive.
I am still not clear why microwave is faster than fiber though. Is it simply that the fiber lines are multi-tenant, so there is not a direct hop?
It's not pure HF -- there's a fibre link alongside the HF link where stuff like protocol negotiations, encryption keys, channel sounding, feedback on data received and RF conditions, and agreement on what kind of modulation/coding schemes to use for upcoming data (and what transmitted bits are to mean in terms of higher-layer protocol (something about price informations? I don't know much about HFT)).
Likewise, the availability of GPS makes time/frequency synchronisation easier (and combined with sounding signals and feedback over fibre), which very likely allows for eliminating latency-inducing complexity in the receiver.
It's a far cry from the traditional data-over-HF scenario, which is assumed only to be relevant when other (higher bandwidth) links such as fibre or satellite or point-to-point microwave are unavailable, so everything must be carried out over the RF channel (look at ALE).
https://micro.magnet.fsu.edu/primer/java/speedoflight/index....
A cable is not going to be direct. If it's above ground, it needs some slack on the line to account for pole movement (you don't want your fibre holding two poles together...).
You'll need more slack for your splices to be done at surface level during construction. Splice trays every x meters will add more distance.
Sometimes the poles are overloaded and you need to run down and bury it for a while.
Your tunnels are never really straight. You need to follow existing easements, or get your own. When you do bury, you need to go around various physical obstructions, like rivers, buildings, cities, parks.
And if you want redundancy against back-hoe fade, your latency in those situations will shut you down, perhaps erasing your profits when you're down and your competitors are up.
The cost of running your own line point-to-point would be be monstrous, though I guess you could build in some latency and sell access to non-HFTers.
Oh, and diamond has a large birefringence due to its crystal structure, which would cause temporal smear in signals.
This is true of multimode fibre, but not single mode. Long haul communications cables are single mode.
AT&T was going to build out a cross country, single mode, circular waveguide network in the 60’s, but was obsoleted due to fiber. Maybe that’s what the HFT ought to due, or maybe air-line coax.
By the way, the uncertainty principle doesn't say we can't have a narrow pulse. Frequency versus time can be used as an analogy to help explain the uncertainty principle. If we know a precise frequency, the signal is poorly localized in the time domain. If a signal is narrowly localized in the time domain, it is poorly localized in the frequency domain. The uncertainty principle is something else, though: it revolves around not being able to make certain related measurements simultaneously, at the quantum level. We can be quite certain about the frequency content of a transient pulse and its exact temporal shape, if it's not something from the quantum domain.
The US market has this interesting phenomenon that the futures trade in Chicago, but the single stocks trade in New York. For those who are not familiar, this means you can make a bet on the S&P 500 index as a whole in Chicago by buying or selling futures. But if you wanted to replicate the basket, you would go to New York.
Naturally there's a relationship between the two. If the futures are going down, the prices of all the stocks needs to be adjusted. And if you look at stock prices, it turns out quite a lot of any stock price is simply exposure to the market at large. So you really want to know as soon as possible if the market is moving.
Now you might have thought that the index is the index of shares, which is in NYC, so why does it matter what Chicago thinks? Well actually the tail wags the dog in this case. The futures are how people express a view on which way the index is going, and you want to know that view immediately.
How does it actually work? Variants on "if the market goes down, pull all your bid orders". Basically if things are fine you are happy to look at queue position and imbalance as an indicator of whether you want to leave orders to capture spread with. But the moment there's an indication from Chicago that the market is dipping, you forget that and pull your orders. Or if you're really fast dump a sell order on the guy behind you in the queue.
There's a huge amount of stock trading in NY each day, so it's worth your while to have this line that you use to send the warnings down.