Stock Market Data Resolution and Nanosecond Timestamps

# Highest resolution of stock data?


# Highest resolution of stock data?












Out of curiosity, I'm wondering what the highest resolution of stock data there is out there. Is there stock trading data for every nanosecond, picosecond, or even lower? And how is this limit determined?

Do high frequency institutions typically use data at such a high resolution?

## Answer by databento (score 5)

https://quant.stackexchange.com/a/77809

> Is there stock trading data for every nanosecond, picosecond, or even lower?

The topic of picoseconds is quite interesting. There a few physical limits that determine this:

- Most matching engines have enough variance in their topology that the marginal engineering cost of getting faster by a few hundred picoseconds is exceeded by the marginal PnL gain, except maybe for the top 3-5 firms at the most competitive venues. (A similar marginal cost analysis exists for why people don't just litter Aurora-NJ with a lot of wireless repeaters or split the feeds 3-way to Mahwah/Secaucus/Carteret right upfront.)

- Even at venues that cater to low-latency traders, usually your messages have to travel through some minimum length of fiber and network hops that usually takes anywhere from thousands of nanoseconds to 100s of microseconds even at the speed of light.

- The fastest switches in mainstream commercial circulation have 4-5 nanoseconds of port-to-port latency with typically around 100s of picoseconds of variation between ports.

- Measurement error and uncertainty becomes dominant at sub-nanosecond time scales. Even the best hardware timestamping and time-synchronization practices with high-precision oscillators for holdover accuracy usually only get down to nanoseconds of accuracy against UTC. There are projects like White Rabbit which seek to extend this accuracy, but they're still in their infancy.

- Setting up any reliable test apparatus to measure time intervals at 4-5 nanoseconds and below takes a fair bit of work, which is usually better spent elsewhere even for a low-latency trading firm. See for example this white paper by Arista/Metamako on the lengths they went through to measure their switch latencies.

- It's inconvenient to store anything at a higher resolution than nanosecond, which can be nicely expressed in nanoseconds since UNIX epoch and fit in a 64-bit integer field. Most wire protocols, tools, libraries like Wireshark and libpcap, pandas, have built-in support for this but not picosecond-resolution timestamps.

So, single-digit nanoseconds for most firms, with less than 10 firms needing 100~ picoseconds. See also this Exablaze presentation on measurements at the picosecond level and this Cisco presentation on latency dispersion between ports.

> Do high frequency institutions typically use data at such a high resolution?

There are many ways to use nanosecond-resolution data though:

- Even if you cannot react within that time frame, it is useful for separating events so you can properly backtest and simulate.

- It's also useful for knowing which events are not "actionable" or "monetizable" in your backtest, meaning they're spaced so closely that you cannot actually trade between them.

- It's important for synchronizing events that take place at different matching engines and data centers.

- It's important for expressing certain intervals like matching engine delay, switch latencies, etc.

> Is there stock trading data for every nanosecond, picosecond, or even lower?

Anyone can build their own infrastructure to timestamp data down to the nanosecond. But practically speaking, it's cost-prohibitive and usually only tier 1-2 institutions do this properly.

Serving these tier 1-2 institutions are a handful of vendor firms that specialize in providing high-resolution market data, including stock market data, with nanosecond-resolution timestamps, but none that provide picosecond-resolution timestamps. These days I think there are really only 8 or 9 firms to consider, each with their own pros and cons:

- Refinitiv

- Pico

- BMLL

- ICE

- Celoxica

- Exegy

- Quincy Data

- Databento (for full disclosure, I work there)

## Answer by Joe (score 2)

https://quant.stackexchange.com/a/77831

@databento provided a really good answer, but missed out a very high level concept which is closer to I think the spirit of your question:

At some point tick data is what is interesting - a series of every individual trade, rather than a series of Open/High/Low/Close/Volume per timestamp. Only very liquid securities will consistently have many trades per minute.

Even for quote updates, because Level II quotes only give visibility for the oldest quote at any price, even updates there are not much more frequent than trades.

Nanosecond-resolution timestamp data is common, and I am not aware of any exchanges that offer better resolution data. Even nanosecond data is recent being commonly supported by backend systems only in 2021.