Using Stationary Time-Series Regressions to Study Sovereign Spreads
Summary
The document considers how to model sovereign yield spreads for Germany, France, and Spain using unemployment and public debt. It raises choices about stationarity, whether to use absolute economic indicators or differences between countries, and how to align predictors with spreads when explaining or forecasting them.
The response recommends checking stationarity, considering yield levels as an alternative dependent variable, and including both absolute and relative measures of unemployment and debt. For forecasts, it suggests matching predictors and outcomes in time while also testing lagged specifications, since credit spreads may involve timing effects. It proposes comparing candidate models with information criteria such as AIC or BIC. These are brief suggestions rather than a complete modeling procedure: they do not specify data transformations, diagnostics, or out-of-sample validation, and the suggested significance threshold is not a substitute for those checks.
Key ideas
- Check that time-series inputs support the regression specification being used.
- Compare absolute country indicators with cross-country differences.
- Test contemporaneous and lagged predictors for explanation and forecasting.
- Use model-selection criteria such as AIC or BIC to compare specifications.
Tags
Full text
# basic econometric model on country spread methodology # basic econometric model on country spread methodology I am doing some research on Germany, France and Spain on their spread. I would like to try to 'forecast' or 'explain' the the spread on sovreign debt using OLS regression on unemployment and debt amount. I would like to ask you some suggestions on the methodology: 1) To use OLS in time series all data shall be stationary but once I have stationarized them it works fine; right? 2) Would you regress the spread between germany and france on unemployment of France or on the $\Delta$ between unemployment in france and germany? 3) To forecast I would regress spread at time $t$ on variables at $t-1$, is it fine? To explain, on the other hand, both shall be at time $t$. Do you agree? ## Answer by rrg (score 1) https://quant.stackexchange.com/a/31623 These are good questions. 1: Yes. Similarly, consider absolute yield level as a regressand. 2: Regress on unemployment absolute AND difference. You can toss out any statistically insignificant (||<2). 3: Perhaps matching in time for observable and dependent variable is best; you can test lagged models, which is significant in the context of credit spread modelling where there is some degree of "looking ahead". You may end up with multiple econometric models to test, no bad thing - here the art can be mathematicalised using AIC/BIC or similar model selection methods. e.g. `france credit spread = debt + relative debt spread + unemployment rate + relative unemployment spread`
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.