Pooling Bond Auction Data Across Maturities for Yield Forecasting

Summary

The document frames a modeling question about predicting the difference between a bond auction’s yield and its when-issued yield. The data are described as sparse: auctions occur monthly, and the author reports 850 results across maturities over a twenty-year period. Splitting observations into separate maturity groups would leave fewer than 200 examples per group, motivating a comparison between separate maturity-specific models and a single model that includes maturity as a categorical feature.

No answer, modeling results, or recommendation is provided. The text therefore serves as a problem setup rather than evidence that pooling improves forecasts. It identifies the tradeoff between using maturity-specific patterns and preserving sample size, but leaves open how to encode maturity, validate out-of-sample performance, account for time effects, or handle differences among bond types. Any choice would need evaluation on suitable held-out data.

Key ideas

• The target is the difference between auction yield and the when-issued yield forecast.
• Monthly auction frequency limits the available observations in the described dataset.
• Separate maturity models would have fewer than 200 examples per group, according to the author.
• The proposed alternative is one pooled model with bond maturity encoded as a feature.
• The document poses this comparison but supplies no results or final modeling recommendation.

Tags

# Predicting bond auction result. Should I train separate models for different maturity in face of Data deficiency?


# Predicting bond auction result. Should I train separate models for different maturity in face of Data deficiency?












#### Problem Statement

Trying to predict how bond auction result ( in terms of yield ) is different from its forecast (the when-issued yield ).

More info:http://www.mortgagenewsdaily.com/mortgage_rates/blog/242898.aspx

#### Data deficiency:

Auction only happens monthly. Therefore there are only 850 auctions result for bond of all maturity during the last 20 year interval.

If dividing the auction result by maturity term, it would leave each category with less than 200 results.

#### Plot

Plot of auction yield and when-issued yield difference , separated by maturity

Plot of auction yield and when-issued yield difference , not separated by maturity

#### Questions:

Should I stay away from fitting a different predictive model for each bond maturity due to data deficiency ?

Instead, Should I do one-hot encoding and use bond maturity as a feature and then fit a single model for all 850 result ?

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.