[anonDataUser]

To expand on this a bit:

Within the US there are a number of ECNs/exchanges. Each one publishes there own order book including prices and sizes on each side of the book. To get the most accurate information possible on market prices, you need to have a direct connection to each exchange and usually, you want to be co-located within the same data center as them. Unless you're a market maker, high-frequency trader, or run an institutional electronic trading platform, this level of data is not necessary.

For most human traders, a consolidated feed is sufficient; that data comes from an authority called the CTA. From wikipedia: "Since the late 1970s, all SEC-registered exchanges and market centers that trade NYSE or AMEX-listed securities send their trades and quotes to a central consolidator where the Consolidated Tape System (CTS) and Consolidated Quotation System (CQS) data streams are produced and distributed worldwide." The consolidation process delays the data by ~100 ms. This is still considered real-time by many people including professional traders.

Most HN readers are getting a conflated version of the consolidated feed. Google's "real-time" feed falls into this category and adding 5 seconds doesn't really matter at this point.

[jsavimbi]

> For most human traders

You're living in the past.

> The consolidation process delays the data by ~100 ms. This is still considered real-time by many people including professional traders.

A couple of months ago we achieved round-trip [cleared] trades between Tokyo and New York of under 50ms. That's your barrier to entry.

You're not in business if you're looking at an average of 100ms.

[cluda01]

I'm not sure if this is possible. According to Wolfram Alpha the distance between NYC and Tokyo is ~10879 Kilometers (as the crow flies, not as the cable lies). The speed of light in Kilometers per second is 299,792.458

If we evaluate (10879 / 299792.458) * 1000 = 36.28. This is for a single leg. So we get 72.57 milliseconds just for as the crow flies transport. This does not include other factors that contribute to latency (network, processing, etc.)

For "round trip" as you say then this number should be multiplied by 2 as it has to go NYC -> Tokyo, Tokyo -> NYC.

Have you developed some sort of faster than the speed of light communication mechanism we haven't heard of? Inquiring minds demand to know.

Edit: One possible explanation is that you are using multiple time sources that are not synchronized. E.g. if you send a FIX message to an ECN and use the clock available on your machine and then use the ECN's timestamps somewhere in your calculation this renders the measurement invalid because you can't ensure that your clocks are synchronized.

[anonDataUser]

Human traders = past? You should see the floor that I work on. There are plenty, and they're doing very well. No algo has ivy league connections.

> You're not in business if you're looking at an average of 100ms.

You're not in business if your business is latency arb. Buffett doesn't care about your 50 ms.