Handling Missing Dates in Interest-Rate Autocorrelation

Summary

The discussion considers how to calculate lagged correlations when an interest-rate series has blank dates. One suggestion is to interpolate missing observations because rates often change gradually, while marking the estimated values and repeating the calculation with adjusted fills to see how sensitive the result is. This makes the assumptions behind data repair visible rather than treating estimates as observed rates.

A second approach depends on what the series represents. Overnight rates may accrue over weekends and holidays, so carrying the prior rate across those calendar days can reflect the interest actually earned. If the goal is instead to study how market information updates, the discussion recommends limiting the sample to trading days before calculating correlations. These approaches answer different questions: interpolation estimates absent observations, carry-forward represents accrual conventions, and removing non-trading dates focuses on market sessions. The document offers practical guidance but no empirical comparison, and the appropriate treatment depends on the rate definition and research objective.

Key ideas

• Choose a missing-data treatment based on what the interest-rate series represents.
• Mark interpolated observations and check how they affect the correlation result.
• Overnight interest can accrue across weekends and holidays at the prevailing rate.
• Restricting the data to trading days focuses the analysis on market information updates.

Tags

# Running an autocorrelation with blanks?


# Running an autocorrelation with blanks?












How does one run an autocorrelation when there are blanks in the dataset?

I have a dataset of interest rates and I am plotting day x vs day x-1. I’m unable to run a correlation in Excel if there are blanks. How do you recommend proceeding?

## Answer by Phil H (score 3, accepted)

https://quant.stackexchange.com/a/63124

One option is just to fill them in - interest rates don't usually jump around, so interpolating from surrounding data would be unsurprising. If you want to know what effect that is having, by all means mark those you're filling in and duplicate the analysis with a shift to all those filled in rates.

If you have an index rather than a set of term rates, then usually the last index is applied to such dates. For example an overnight index would generally have that rule.

## Answer by nbbo2 (score 1)

https://quant.stackexchange.com/a/63128

Are you dealing with overnight rates, such as Fed Funds?

In such cases the same rate continues to be paid while the markets are closed. So for example if FF is X on Friday, it means you will earn X for three days: Saturday, Sunday and Monday.

In this sense, there are no "blanks" in interest rate data, you earn some interest every day, whether a regular day, weekend or bank holiday.

However, if you are concerned with how the financial markets update information, then I would only consider trading days, that is I would delete weekends and holidays from the data before taking the correlations or scatter diagrams.

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.