Inflation Dynamics and Labor Market Specifications: A Bayesian Dynamic Stochastic General Equilibrium Approach for Japan's Economy
Ichiue, Hibiki, Kurozumi, Takushi, Sunakawa, Takeki, Economic Inquiry
Inflation dynamics have been investigated using dynamic stochastic general equilibrium (DSGE) models with sticky prices. In the models, inflation is determined by the new Keynesian Phillips curve (NKPC), where real marginal cost of producing output is the main driving force and varies with changes in labor cost. Therefore, the specification of the labor market plays a crucial role in describing inflation dynamics.
Many previous studies use sticky wage models (e.g., Christiano, Eichenbaum, and Evans 2005; Levin et al. 2006; Smets and Wouters 2003, 2007). In this specification of the labor market, workers set wages on a staggered basis la Calvo (1983) while firms employ all workers and adjust labor input by changing hours per worker (i.e., intensive margin of labor). Consequently, the cost involved to adjust employment is absent in sticky wage models.
Recently, there has been a surge of interest in the role of the extensive margin of labor (i.e., employment) for inflation dynamics. Walsh (2005), Krause and Lubik (2007), and Trigari (2009), for instance, incorporate sticky prices in a labor market search model that has been analyzed in the literature starting from Mortensen and Pissarides (1994), Merz (1995), and Andolfatto (1996). (1) In this labor market specification, firms adjust labor input by changing employment while wages are determined through bargaining between workers and firms. These characteristics of the labor market differ radically from those in sticky wage models. Despite this fundamental difference, existing literature lacks the formal comparison of sticky wage models and labor market search models.
The present article fills this gap. Specifically, we estimate a sticky wage model and a labor market search model with Japan's data using a Bayesian likelihood approach, and compare these two models in terms of marginal likelihood and out-of-sample forecast performance.
The estimation results show that the labor market search model is superior to the sticky wage model in terms of both marginal likelihood and out-of sample forecast performance, particularly regarding inflation. Why is the former model better able to fit the data and forecast inflation? Because the non-labor market part is identical between these two models, the key difference is the relationship between real wages and inflation or output. In the sticky wage model, real wages are highly correlated with output in the presence of monopolistically competitive labor markets, and unit labor cost under full employment is identical with real marginal cost and thus drives inflation dynamics. However, the cross-correlation between real wages and output in the data is not so high as in the sticky wage model. Besides, the data of real wages and output determine the unit labor cost, which lags far behind inflation by more than 3 years whereas a high contemporaneous correlation between inflation and real marginal cost is represented in the NKPC. In the labor market search model, labor bargaining generates a mild cross-correlation between real wages and output. Moreover, real marginal cost is determined by both hiring cost and unit labor cost that varies with employment fluctuations, which gives rise to a high contemporaneous correlation between inflation and real marginal cost. Therefore, the labor market search model is better able to fit the data and forecast inflation.
In related literature, Rabanal and Rubio-Ramirez (2005) show that a sticky wage model matches U.S. data far better in terms of marginal likelihood than a flexible wage model. Christoffel, Linzert, and Linzert (2006), Gertler, Sala, and Trigari (2008), and Krause, Lopez-Salido, and Lubik (2008) estimate a labor market search model with U.S. or Euro area data. To our knowledge, the present article is the first to compare a sticky wage model and a labor market search model in terms of marginal likelihood and out-of-sample forecast performance. …