Handling Missing Interbank Rate Observations with Interpolation

Summary

The document discusses how to handle missing observations in short-term interbank rate data, motivated by a goal of predicting customer behavior. Its suggested first step is to seek a more complete, higher-quality source and use it to repair gaps where possible. If observations remain missing, it proposes simple linear interpolation: divide the difference between the observations bracketing a gap across the missing dates and fill the values in increments.

The response cautions that interpolated points should be disclosed and their presence reflected when assessing model accuracy. It recommends flagging cases where the rate changes sharply across the gap, since interpolation over a large move could distort analysis. The advice is a practical starting point, not a general statistical treatment: it does not compare interpolation with filtering methods or explain how missingness patterns affect forecasts. The appropriate handling depends on the intended model use, and the proposed method may smooth real movements between observed dates.

Key ideas

• Look for a more complete source before estimating missing market-rate observations.
• Linear interpolation fills a gap by distributing the change between surrounding observations across missing dates.
• Mark interpolated values and account for them when assessing model accuracy.
• Flag large changes across missing dates because interpolation may misrepresent rate movements.
• The suggested method is a simple baseline and may be unsuitable when missing intervals contain abrupt changes.

Tags

# interbank market rates - missing data


# interbank market rates - missing data












I took EU market rates here: https://www.banque-france.fr/en/economics-statistics/rates/main-euro-area-interbank-market-rates.html

As you can see (Eonia 05/13/1999 for example) there is non defined and non available data. I know that for prices you can't easily interpolate data because you will modify volatility in doing so. You need to use more complex techniques (kalman filter ?).

My question is: how can I handle missing data for markets rates ? Does it depends on what you want to do with your data ? (I want to predict clients comportement based on short terms rates).

## Answer by chjortlund (score 1)

https://quant.stackexchange.com/a/16008

Of cause, the best solution is to fill out the missing holes, in this case head over to the ECB Statistical Bank, and get a higher quality dataset or fill the holes, in your existing one.

Taking your EONIA example, I would simply take the difference between the rate before and after the missing spot(s), and divide it by the missing spots and increment the before rate with the found coefficient.

However it is important the you disclose that some of the data you/your model rely on have interpolated data points, and this should be incorporated into the model accuracy. I would recommend to trigger a warning if the change between the missing points is larger than a fixed amount e.g. if 2009-01-21 where to be missing, the difference between 2009-01-20 and 2009-01-22 would be 0.899 which is quite a drop, and might severely affect your model.

BTW: Take look at the one-size-fits-all interpolation methods of missing data from time series in Matlabs Financial Toolbox.

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.