Fourier Quantile-on-Quantile Regression
Fourier-Augmented Quantile-on-Quantile Regression · Also known as: Fourier QQ regression, Fourier-QQR, Fourier quantile regression with quantile regressors, smooth structural-break QQ regression
Fourier quantile-on-quantile regression extends the quantile-on-quantile (QQ) framework of Sim and Zhou (2015) by embedding Fourier trigonometric terms into the local linear quantile model. This allows the estimated dependence between the quantiles of one variable and the quantiles of another to vary smoothly over time, capturing gradual structural change without imposing a known break date.
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 Fourier QQ regression when you suspect the dependence between two variables is asymmetric across their respective distributions and also changes gradually over a long time series, without a single sharp break. It is well-suited to energy economics, finance, and macroeconomics where tail dependence (e.g., commodity prices and stock returns at extremes) evolves through business cycles or policy regimes. Do not use it when the sample is short (fewer than about 100 observations), as the Fourier terms and kernel estimation jointly consume many degrees of freedom. Avoid it when structural change is abrupt rather than smooth; in that case a Chow-break or Markov-switching specification is more appropriate. If cross-sectional rather than time-series variation is of primary interest, plain QQ regression without the Fourier component is simpler and preferable.
Strengths & limitations
- Simultaneously captures distributional asymmetry (via QQ grid) and smooth time-variation (via Fourier terms) in a single model.
- Does not require a pre-specified break date; the Fourier terms adapt to when the relationship shifts.
- Reveals the full (theta, tau) dependence surface, exposing whether tail-to-tail interactions differ from median-to-median ones.
- Flexible yet parsimonious: one or two Fourier harmonics often suffice, avoiding parameter proliferation.
- Applicable without strong distributional assumptions, relying on semi-parametric quantile methods.
- Requires large samples (100+ observations) for reliable kernel and Fourier estimation; small samples produce unstable surfaces.
- Computational cost is high: estimates must be computed at every (theta, tau) grid point with bootstrap inference.
- The choice of bandwidth and Fourier frequency order is not fully automatic and can affect results materially.
- Interpretation of the two-dimensional dependence surface is more complex than a single slope coefficient.
Frequently asked
How does Fourier QQ regression differ from plain QQ regression?
Plain QQ regression treats the dependence between quantiles as fixed across the entire sample. Fourier QQ regression adds sine and cosine terms to allow that dependence to evolve smoothly over time. If the Wald test for the Fourier terms is not significant, both models give the same story; when it is significant, the time-varying model reveals how the quantile relationship has shifted.
How many Fourier harmonics should I include?
Start with k = 1 (one pair of sine and cosine) and use AIC or BIC to decide whether k = 2 improves the model. In most applied work one harmonic is sufficient to capture a single-phase smooth structural change, and more than two harmonics rarely improve fit while increasing the risk of overfitting.
How do I choose the bandwidth for the kernel estimation?
A data-driven rule-of-thumb based on the sample size — such as h = 0.05 for T around 200 — is common in QQ papers. Cross-validation over the quantile grid is more principled but computationally intensive. Sensitivity checks across a range of bandwidths are good practice to ensure conclusions are not bandwidth-driven.
What software can estimate this model?
There is no single dedicated package; researchers typically code the method in R using quantreg for the local linear quantile estimation, building the Fourier regressors manually. Some applied papers provide replication code. StatWise implements a guided version that handles the grid, Fourier term construction, and bootstrap inference automatically.
Is a large sample really necessary?
Yes. At each of the many (theta, tau) grid points the estimator uses only observations in a local neighbourhood of x near F_x^{-1}(tau), and the effective local sample can be much smaller than the full T. With T < 100 the local sample may contain too few points for stable estimates, and adding Fourier regressors aggravates this further.
Sources
- Sim, N., & Zhou, H. (2015). Oil prices, US stock return, and the dependence between their quantiles. Journal of Banking and Finance, 55, 1-8. DOI: 10.1016/j.jbankfin.2015.01.013 ↗
- Gallant, A. R. (1981). On the bias in flexible functional forms and an essentially unbiased form: The Fourier flexible form. Journal of Econometrics, 15(2), 211-245. DOI: 10.1016/0304-4076(81)90115-9 ↗
How to cite this page
ScholarGate. (2026, June 3). Fourier-Augmented Quantile-on-Quantile Regression. ScholarGate. https://scholargate.app/en/econometrics/fourier-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.
- Fourier ARDL Bounds TestEconometrics↔ compare
- Fourier Granger CausalityEconometrics↔ compare
- Nonlinear ARDLEconometrics↔ compare
- Panel Quantile-on-Quantile RegressionEconometrics↔ compare
- Quantile RegressionEconometrics↔ compare
- Quantile-on-Quantile RegressionEconometrics↔ compare