Time-Varying Parameter Quantile-on-Quantile (TVP-QQ) Regression
Time-Varying Parameter Quantile-on-Quantile Regression · Also known as: TVP-QQ regression, time-varying QQ regression, dynamic quantile-on-quantile regression, TVP quantile-on-quantile
TVP-QQ regression extends the quantile-on-quantile (QQ) framework by allowing the slope coefficients to evolve over time. It maps how the quantiles of a predictor variable affect the quantiles of an outcome differently across the joint distribution and across different time periods, uncovering dynamic, heterogeneous dependence structures that standard regression cannot detect.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use TVP-QQ regression when you suspect that the relationship between two variables varies both across their joint distribution and over time — for example when studying how extreme movements in energy prices affect tail quantiles of financial returns during crisis versus tranquil periods. It is especially suitable for long time series where structural breaks or regime shifts are possible. Avoid it when the sample is short (fewer than ~150 observations), when only conditional-mean effects are of interest, or when the computational burden of nonparametric estimation is prohibitive. It is not a substitute for linear VAR models when causal ordering and orthogonalisation matter.
Strengths & limitations
- Captures heterogeneous effects across the entire joint distribution of predictor and outcome, revealing asymmetries invisible to OLS or standard quantile regression.
- Accommodates time variation in dependence without imposing a rigid break-point, making it robust to gradual structural change.
- Produces rich visualisations (coefficient heat-maps) that communicate complex distributional dynamics intuitively.
- No parametric distributional assumption is required for either the predictor or the outcome.
- Widely applicable in financial econometrics, energy economics, and macro-finance where tail and crisis-period effects are the primary concern.
- Requires a relatively large time series to support two simultaneous nonparametric smoothing dimensions (quantile and time).
- Results are sensitive to bandwidth selection; poorly chosen bandwidths can over-smooth or under-smooth the coefficient surface.
- Inference is computationally intensive because bootstrap resampling must be repeated for every (τ, θ, t) grid point.
- The method is descriptive and correlational; it does not identify causal effects without additional assumptions.
- Software implementations are not yet standardised across major econometric packages, requiring custom code in R or Python.
Frequently asked
How does TVP-QQ differ from standard quantile-on-quantile regression?
Standard QQ regression (Sim & Zhou, 2015) estimates a fixed coefficient surface over the (τ, θ) quantile grid for the whole sample. TVP-QQ adds a second kernel along the time dimension so that the coefficients at each quantile combination are allowed to change smoothly over the sample period.
How many observations do I need to run TVP-QQ reliably?
As a rule of thumb, at least 150–200 time-series observations are recommended so that the simultaneous smoothing over quantiles and time leaves enough effective observations at each local estimation point. With fewer observations, the bandwidth must be set very large, which defeats the purpose of detecting time variation.
How do I choose the bandwidth parameters h and b?
Cross-validation over a grid of candidate values is the most principled approach. In practice, many applied studies follow Sim and Zhou (2015) and set the quantile-dimension bandwidth using Silverman's rule of thumb, then select the time bandwidth b to balance temporal resolution against estimation noise.
Can TVP-QQ be applied to panel data?
The baseline method is designed for time series. Extensions to panel data require additional assumptions about cross-sectional pooling and the treatment of heterogeneous units, and are still an active area of methodological development.
What software can I use to estimate TVP-QQ models?
There is no dedicated, well-maintained CRAN or PyPI package as of 2024. Researchers typically implement the estimator from scratch in R (using the quantreg package for the inner quantile regression step) or Python (using statsmodels), adapting code shared in replication archives of published papers.
Sources
- Sim, N., & Zhou, H. (2015). Oil prices, US stock return, and the dependence between their quantiles. Journal of Banking & Finance, 55, 1–8. DOI: 10.1016/j.jbankfin.2015.01.013 ↗
- Bouri, E., Gupta, R., & Vo, X. V. (2021). Jumps in geopolitical risk and the cryptocurrency market: The singularity of Bitcoin. Defence and Peace Economics, 33(2), 150–161. link ↗
How to cite this page
ScholarGate. (2026, June 3). Time-Varying Parameter Quantile-on-Quantile Regression. ScholarGate. https://scholargate.app/en/econometrics/time-varying-parameter-quantile-on-quantile-regression
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Quantile RegressionEconometrics↔ compare
- Quantile-on-Quantile RegressionEconometrics↔ compare