Nonlinear Toda-Yamamoto Causality Test
Nonlinear Toda-Yamamoto Granger Causality Test · Also known as: nonlinear TY causality, rank-based Toda-Yamamoto test, modified Wald nonlinear causality, NTY causality test
The Nonlinear Toda-Yamamoto causality test extends the classic Toda-Yamamoto (1995) modified Wald procedure to detect causal linkages that are hidden in the means of series but manifest through nonlinear dynamics such as asymmetries, threshold effects, or volatility transmission. It fits an augmented VAR on rank-transformed or otherwise nonlinearly mapped series and applies a chi-squared Wald test on the extra-lag coefficients.
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 Nonlinear Toda-Yamamoto causality when series may be nonstationary (unit roots) and you suspect that causal relationships are asymmetric, regime-dependent, or transmitted through higher moments rather than the conditional mean. It is especially appropriate for energy, commodity, and financial market data where volatility spillovers and threshold dynamics are common. Do not use it when the theoretical mechanism is plausibly linear and well-captured by the standard Toda-Yamamoto test, or when sample size is very small (fewer than about 60 observations), as asymptotic chi-squared inference becomes unreliable. Also avoid it if a structural VAR or cointegrated VECM is required to recover structural parameters.
Strengths & limitations
- Detects causal relationships hidden in nonlinear dynamics — asymmetries, threshold effects, and volatility transmission — that standard linear causality tests miss.
- Inherits the Toda-Yamamoto advantage of not requiring pre-testing for cointegration; the augmented VAR delivers valid chi-squared inference regardless of the integration order.
- Straightforward to implement: requires only a rank or squared transformation of the series before applying the standard modified Wald procedure.
- Applicable to both stationary and nonstationary series without differencing, preserving long-run information.
- Widely accepted in energy economics, finance, and environmental econometrics literature.
- The choice of nonlinear transformation (ranks, squares, BDS filter) is ad hoc and different transformations may yield conflicting results.
- Asymptotic chi-squared distribution may perform poorly in small samples (n < 60); bootstrap critical values are advisable but add computational cost.
- Does not identify the direction or economic nature of the nonlinear channel — a significant test only confirms predictive dependence, not structural causation.
- Sensitive to lag-order selection; misspecification of k can inflate size or reduce power.
- Cannot distinguish between different types of nonlinearity (asymmetry vs. volatility spillover vs. threshold switching).
Frequently asked
How does this differ from the standard Toda-Yamamoto test?
The standard Toda-Yamamoto test detects linear predictive causality in the conditional mean. The nonlinear version adds a transformation step — typically converting the series to ranks or squared values — before fitting the augmented VAR, so it can detect causal links transmitted through higher moments, asymmetries, or threshold effects that the linear test would miss entirely.
Why use ranks as the nonlinear transformation?
Rank transformation converts the data to a uniform distribution, removing the influence of the original marginal distributions and making the test robust to outliers and heavy tails. Because ranks preserve the ordinal dependence structure while discarding parametric assumptions about the mean, they are effective at revealing monotone nonlinear relationships.
Do I need to test for nonlinearity before applying this method?
It is good practice to first apply a nonlinearity test such as the BDS test to the VAR residuals. If the residuals show no significant nonlinearity, the standard Toda-Yamamoto test is sufficient and the nonlinear extension adds complexity without power gain.
What sample size is adequate for reliable inference?
As a rough guideline, at least 60 to 80 observations are recommended for asymptotic chi-squared inference to be reliable. With smaller samples, bootstrap or Monte Carlo critical values should replace the asymptotic cut-offs to control size distortion.
Can I use this test in a multivariate system with more than two variables?
Yes. The augmented VAR can include multiple variables. Each pairwise or block-exclusion hypothesis is tested separately with a modified Wald statistic on the respective coefficient block. Multiplicity of tests should be acknowledged and corrections such as Benjamini-Hochberg applied if many hypotheses are tested.
Sources
- Toda, H. Y., & Yamamoto, T. (1995). Statistical inference in vector autoregressions with possibly integrated processes. Journal of Econometrics, 66(1-2), 225-250. DOI: 10.1016/0304-4076(94)01616-8 ↗
- Sims, C. A., Stock, J. H., & Watson, M. W. (1990). Inference in linear time series models with some unit roots. Econometrica, 58(1), 113-144. DOI: 10.2307/2938337 ↗
How to cite this page
ScholarGate. (2026, June 3). Nonlinear Toda-Yamamoto Granger Causality Test. ScholarGate. https://scholargate.app/en/econometrics/nonlinear-toda-yamamoto-causality
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.
- Cointegration TestEconometrics↔ compare
- Granger CausalityEconometrics↔ compare
- Nonlinear Granger CausalityEconometrics↔ compare
- Toda-Yamamoto CausalityEconometrics↔ compare
- VAR ModelEconometrics↔ compare