Sequential Regressions with Structural Breaks and Residuals

Summary

The document asks whether a two-stage regression is appropriate for local government bond yields and their potential determinants when explanatory variables experience structural breaks at different times. The proposed approach groups variables with similar break dates, estimates models on corresponding data segments, combines their residuals, and then regresses another set of variables on those residuals.

It also asks whether non-normal residuals from the first stage violate assumptions. The document gives no answer, estimation results, or diagnostic evidence, so it does not establish that this procedure is valid. Assessing it would require specifying the model and identifying the target inference; break handling, generated-regressor uncertainty, and the assumptions required for second-stage inference all matter. Non-normality alone does not automatically invalidate ordinary least squares estimates, but it can affect inference under some conditions. The entry is a methodological question, not a tested recommendation.

Key ideas

• The proposed method groups explanatory variables by structural break timing before sequential regressions.
• Residuals from the first stage would become the response variable in a later regression.
• The document does not establish whether combining residuals across segments is valid.
• Residual normality and its effect depend on the estimation assumptions and intended inference.

Tags

# Regressing on Residuals


# Regressing on Residuals












I have a time series dataset (Local Gov. bond yields and probable determinants). Due to structural breaks in different exogenous variables in different time points, I have an idea of regressing a set of parameters with similar structural break times first (dividing dataset and estimating models, merging residuals), and then regressing other part of variables on residuals of the first estimation. Is it a correct way to do? Do I break any assumptions if residuals of the first model are not normally distributed?

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.