Correlating Irregularly Timed Market Data

Summary

The document raises two practical problems in estimating correlation from market time series: missing observations within a trading day and trades that arrive at different times for each security. It points readers toward methods designed for irregularly spaced observations, citing research that discusses covariance and correlation estimation and compares interpolation approaches. These methods address the timing mismatch without assuming that every security records a trade at the same instant.

For a simpler data-handling approach, the responses suggest choosing bid, ask, or midpoint prices, or producing a regular time series by carrying forward the last trade when no new trade occurs. That approach depends on liquid markets and a chosen conflation interval. The document does not explain or evaluate the cited estimators, give a worked example, or establish which price convention is preferable. Missing-data treatment and price selection therefore remain dependent on the instruments, data, and research question.

Key ideas

• Correlation estimation must account for missing observations and asynchronous trades.
• Methods for irregularly spaced data can estimate covariance and correlation without requiring simultaneous trades.
• Bid, ask, or midpoint prices are possible inputs, depending on the chosen convention.
• Carrying forward the last trade creates regularly spaced observations but relies on assumptions about liquidity and timing.

Tags

# Normalization of Market Data in Time Series Correlation


# Normalization of Market Data in Time Series Correlation












Suppose we have 2 time series of market data, one for each security and we want to correlate between these 2 securities. My question is

- How do we handle gaps of missing data in the time series? Imagine the time series is one day tick data of a stock price and we have a 10 mins gap of missing data sometime during the day.

- How do we correlate the tick-by-tick market data of these 2 securities that do not happen at the same time for each tick? If we correlate them in the same time intervals, what price do we use?

## Answer by ViK (score 2)

https://quant.stackexchange.com/a/15397

Looking for the same issue, I found an article by de Jong(1997). In section 2 you can find a method for estimation of covariances and correlations between irregularly spaced data. Also look at the article by Jonas Andersson where some interpolation methods and method form de Jong are presented and compared together. Hope it helps.

## Answer by Nin Sute (score 0)

https://quant.stackexchange.com/a/27521

One assumption is that both (or more) instruments are liquid enough to offer a market (both sides).

You can

- use the bid/ask/mid (your choice) or

- "conflate" (implemented by the big boys on their data feeds). i.e. 1 second conflation: if no trade, send out last trade price (or assume so in your application).

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.