Structural Break Toda-Yamamoto Causality Test
Toda-Yamamoto Causality Test with Structural Breaks · Also known as: SB-TY causality, structural break modified Wald test causality, Fourier Toda-Yamamoto causality, causality with regime shifts
The structural break Toda-Yamamoto causality test extends the standard Toda-Yamamoto modified Wald (MWALD) procedure to accommodate one or more structural breaks in the time series. By identifying break dates first and then including dummy variables in the augmented VAR, the test maintains its valid asymptotic chi-squared distribution regardless of the integration or cointegration order of the variables, even in the presence of regime shifts.
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 this test when you suspect a Granger causal relationship between time series that may also have experienced structural breaks — for example, studying the causal link between energy prices and GDP across a financial crisis, or between monetary policy and inflation across a regime change. It is appropriate when the variables may be integrated of order 1 or 2 (or a mix), so that standard Granger causality testing in a cointegrated VAR or VECM would require knowing the exact cointegration rank. Do NOT use it as a first choice when there is no theoretical or empirical reason to expect structural breaks — adding unnecessary dummy variables wastes degrees of freedom. Also avoid it when the sample is very short (fewer than ~60 observations), since identifying break dates and estimating the augmented VAR reliably requires adequate data.
Strengths & limitations
- Valid asymptotic inference regardless of the integration or cointegration order of the variables, even in the presence of structural breaks.
- Does not require pre-testing for cointegration rank, avoiding specification errors that plague VECM-based causality tests.
- Accommodates single or multiple structural breaks without altering the basic MWALD distributional result.
- Applicable to both stationary and non-stationary time series, making it highly flexible for macro and financial data.
- Bootstrap critical values are available for small samples, improving finite-sample size control.
- Requires a separate, prior step to identify break dates; misidentifying the break date distorts the test.
- Each additional structural break and associated dummy variable consumes degrees of freedom, reducing power in short samples.
- Provides evidence of Granger causality (predictive precedence) but does not identify structural or economic causality.
- Power decreases as d_max and the number of breaks increase, especially with small T.
- Sensitive to lag order selection; under- or over-specifying k + d_max can inflate size or reduce power.
Frequently asked
How does this differ from the standard Toda-Yamamoto test?
The core MWALD procedure is identical. The structural break extension adds a prior step to detect break dates and then includes regime-shift dummy variables in the augmented VAR. This prevents the break from being absorbed into the error term, which would otherwise bias the test toward spurious or missed causality findings.
Do I need to know the integration order before running this test?
You need a conservative upper bound d_max on the maximum integration order. Use unit-root tests that allow for structural breaks (such as Zivot-Andrews or Perron 1997) to determine whether each variable is I(0), I(1), or I(2), and set d_max accordingly. You do not need to know the cointegration rank.
What happens if I identify the wrong break date?
The break date is treated as known once it is detected, but the detection carries its own uncertainty. Using a wrong break date introduces mis-specified dummies that either fail to capture the actual regime shift or create spurious ones, distorting both the VAR estimates and the MWALD statistic. Always report the break-detection procedure and its significance to allow replication.
Can I test causality separately within each sub-period defined by the break?
Yes. Splitting the sample at the break date and running separate Toda-Yamamoto tests in each sub-period is a complementary strategy. However, sub-samples may be short, reducing power, and the total number of observations must be sufficient to estimate the augmented VAR in each regime. The full-sample approach with dummies is generally preferred when sub-samples are small.
Which software can run this test?
EViews and Stata are the most common platforms; the procedure is implemented manually by adding lags and dummies to a VAR. R packages such as vars combined with strucchange or mbreak for break detection can replicate the procedure. StatWise streamlines the full pipeline — break detection, augmented VAR estimation, and the MWALD test — in a single workflow.
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 ↗
- Zivot, E., & Andrews, D. W. K. (1992). Further evidence on the great crash, the oil-price shock, and the unit-root hypothesis. Journal of Business and Economic Statistics, 10(3), 251-270. DOI: 10.1080/07350015.1992.10509904 ↗
How to cite this page
ScholarGate. (2026, June 3). Toda-Yamamoto Causality Test with Structural Breaks. ScholarGate. https://scholargate.app/en/econometrics/structural-break-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.
- Granger Causality TestEconometrics↔ compare
- Structural Break Granger CausalityEconometrics↔ compare
- Structural Break VAR ModelEconometrics↔ compare
- Toda-Yamamoto causality testEconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare
- Zivot-Andrews Structural Break TestEconometrics↔ compare