日本株式市場における資産評価モデルのGRS検定に関する一考察

摘要

資産評価モデルの妥当性をGRS検定から検証する際，米国株式市場では企業特性に基づくポートフオリォをモデルのファクターで回帰分析すると切片項が小さくなりモデルが採択されやすいバイアスが生じるため，ポートフォリオをクラスター分析で構築すべきとの指摘がある.日本株式市場では，クラスター分析を採用したポートフォリオの方が，決定係数が低くなる影響からモデルは,されやすくなる結果を得た.この背景をシミュレーション実験により検討したところ，GRS検定は切片項と決定係数の相対的な程度に依存するため，モデルの妥当性を検証するには切片項の平均や決定係数の大きさを個別に確認するなど，GRS検定を補う視点の必要性が示唆された.%The GRS test is a popular statistical method for the evaluation of asset pricing models. It evaluates the asset pricing models focusing on the existence of the unexplained excess return of the test portfolio (whether alpha do or do not exist). Due to the bias making the explanatory power of the model higher when the test portfolios are highly correlated with the factors of the model, it is recommended to adopt the test portfolios constructed by the cluster analysis in the GRS test for asset pricing models. In this research, we perform the GRS test for asset pricing models using the Japanese equity market data and surprisingly find that the asset pricing models tend not to be rejected when we adopt the test portfolios constricted by the cluster analysis. This result is totally different from that of the preceding research using the US equity market data. We examine the reasons behind our empirical result based on the simulation analysis. Our simulation analysis suggests that the asset pricing models sometimes may not be correctly evaluated when we rely only on the result of the GRS test and the magnitude of the average intercept and the coefficient of determinant should also be used to supplement the evaluation.