Toda-Yamamoto Causality Test
Toda-Yamamoto Modified Wald Causality Test · Also known as: Toda-Yamamoto test, TY causality test, modified Wald test for Granger causality, TY-MWALD
The Toda-Yamamoto (TY) causality test is a modified Wald procedure for testing Granger causality in vector autoregressions (VARs) estimated in levels, even when variables are nonstationary or cointegrated. By intentionally over-fitting the VAR with extra lags equal to the maximum integration order, it restores the standard chi-squared asymptotic distribution of the Wald statistic without requiring prior unit-root or cointegration pretesting.
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.
+2 more
When to use it
Use the Toda-Yamamoto test when you want to test Granger causality among time series that may be integrated of order 1 or 2, or cointegrated, and you wish to avoid the risk of pretesting bias from unit-root and cointegration tests. It is particularly valuable when the integration or cointegration properties are uncertain. Do not use it as a substitute for structural causal analysis — TY tests only Granger predictability, not structural or policy-invariant causation. Avoid it with very short samples (T < 50), as the chi-squared approximation can be poor. When series are clearly stationary, a standard VAR Granger test is simpler and equally valid.
Strengths & limitations
- Does not require pretesting for unit roots or cointegration, reducing the risk of specification errors that cascade from pretesting.
- Applicable to systems with mixed orders of integration (I(0), I(1), I(2)) or unknown cointegration rank.
- Uses standard chi-squared critical values, making results easy to interpret and report.
- Works in levels, preserving the long-run information in the data that differencing would discard.
- Straightforward to implement in any VAR-capable software package.
- Tests only Granger predictability (temporal precedence plus incremental forecasting power), not structural or counterfactual causation.
- Over-fitting with extra d_max lags reduces degrees of freedom and loses power, especially in small samples.
- The choice of d_max must be made a priori; underestimating it can invalidate the asymptotic chi-squared result.
- Like all Granger-type tests, it is sensitive to omitted variables that are correlated with both X and Y.
- Does not identify the direction or sign of the causal effect — complementary impulse-response analysis is needed.
Frequently asked
How does the Toda-Yamamoto test differ from the standard Granger causality test?
The standard Granger causality test is valid only when all variables in the VAR are stationary. The Toda-Yamamoto test estimates the VAR in levels with k + d_max lags and applies the Wald restriction only to the first k lags, restoring a standard chi-squared distribution even when series are integrated or cointegrated.
How do I choose d_max in practice?
Run unit-root tests (ADF, PP, KPSS) on each variable. If all series are at most I(1), set d_max = 1; if some may be I(2), set d_max = 2. It is safer to over-estimate d_max slightly than to under-estimate, though every extra lag costs degrees of freedom.
Can I use Toda-Yamamoto when variables are cointegrated?
Yes — this is one of the test's main advantages. You do not need to identify the cointegration rank or estimate a VECM. The augmented VAR in levels implicitly captures the cointegrating relationship through the extra lags.
Does a significant Toda-Yamamoto result prove economic causation?
No. The test only establishes Granger causality: that past values of X improve the prediction of Y beyond Y's own history. This is a necessary but not sufficient condition for structural causation, and the relationship can be spurious if relevant third variables are omitted.
What sample size is adequate for the Toda-Yamamoto test?
The chi-squared approximation is asymptotic, and the test loses power as k + d_max grows relative to T. A rough minimum is T = 50 to 60 observations; T > 100 is preferable when k is large or d_max = 2.
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 ↗
- Dolado, J. J., & Lütkepohl, H. (1996). Making Wald tests work for cointegrated VAR systems. Econometric Reviews, 15(4), 369-386. DOI: 10.1080/07474939608800362 ↗
How to cite this page
ScholarGate. (2026, June 3). Toda-Yamamoto Modified Wald Causality Test. ScholarGate. https://scholargate.app/en/econometrics/toda-yamamoto-causality-test
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.
- ARIMA modelEconometrics↔ compare
- Augmented Dickey-Fuller unit root testEconometrics↔ compare
- Granger Causality TestEconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare
- Vector Error Correction ModelEconometrics↔ compare