Taku Yamamoto

Overview

Taku Yamamoto is a Japanese econometrician known for two distinct contributions to time series inference. His 1981 Biometrika paper derives closed-form MMSE predictors for multivariate ARMA models via a companion-form recursion and establishes the frequentist asymptotic PMSE inflation result that parallels Chow (1973)'s Bayesian framework. His 1995 paper with Toda (Toda-Yamamoto 1995) introduces the lag-augmented VAR procedure for testing Granger causality in possibly integrated and cointegrated systems without requiring pre-testing for unit roots.

Key Contributions / Features

• Yamamoto (1981) — Closed-form h-step MMSE predictor $\hat{x}_{n,h} = H'A^{h-1}PX_n$ for multivariate ARMA(p,q); asymptotic PMSE = $\Omega_h + n^{-1}E(Y_n'U_h'VU_hY_n)$; connects to Chow (1973) Bayesian multiperiod prediction.
• Toda-Yamamoto (1995) — Lag-augmented VAR: estimate VAR(p+d_max) where d_max is the maximum suspected integration order; apply standard Wald tests on first p lag coefficients only; valid χ² asymptotics without pre-testing for cointegration rank.

Related

• Bayesian Multiperiod Prediction
• Vector Autoregression
• Granger Causality
• Hiro Y. Toda
• Gregory C. Chow