*Work in progress.

This project extends the IMF Working Paper by Liu et al. (2024) on machine-learning based inflation forecasting. In their original study, the authors implement four classical supervised learning models, engineered for time series forecasting, to predict inflation in Japan. Japan provides a particularly informative testing ground, having experienced nearly three decades of near-zero inflation followed by an unusual inflation surge in recent years. This environment offers a natural setting to evaluate whether machine learning methods outperform traditional econometric time series models under conditions of high uncertainty. I extend their analysis in two directions. First, I implement a more extensively tuned XGBoost model. Second, I introduce two deep learning approaches: a standard LSTM model and a custom implemented LSTNet architecture. For better comparability, I also implement LASSO, the best-performing model in the original study.

This paper examines the effect of YCC on aligning market's policy expectations using data from Japan. Japan has the longest history of implementing this novel policy. Using the high frequency identification method, I captured the monetary policy surprises around the Bank of Japan’s policy meetings in the past two decades. I implemented a customized Bayesian vector autoregression (VAR) model and decomposed the raw surprises into policy specific and macroeconomic information elements via sign restrictions. My analysis on the decomposed surprises shows that YCC reduces market uncertainty when the control range is sufficiently narrow. This effect extends from the target asset to the shorter end of the yield curve and to adjacent asset markets, albeit with diminishing magnitude, depending on their distance from the targeted asset in terms of both maturity and asset type. The effect weakens and uncertainty increases as the range widens. The effect is monetary specific. It doesn’t affect market’s expectation about macroeconomic conditions. My findings suggest that YCC has the potential to be a viable policy option, with the set up of target control range being a crucial consideration in policy design.

In this paper, I address the divergent behavior of labor force participation rates in the US and Japan during 2010s. I propose a New Keynesian model with endogenous labor force participation decisions and non-separable preference to illustrate a potential role for monetary policy in this process. I show that in the model, the divergence in participation can be driven by either or both factors: the intratemporal substitution elasticity between market goods and home production, as well as the elasticity of labor supply. I then estimate the model for the US and Japan using Bayesian technique. The estimation results indicate that while the two countries have similar intratemporal substitution elasticities, they differ significantly in terms of labor supply elasticity, and consequently, labor participation shows different movements responding to similar policy shocks. A subsequent calibration experiment confirms the findings.

This paper examines the effect of monetary easing (QE) on corporate loans in Japan using sector level data. Comparing an aggregated level VAR model with a sectoral panel VAR, I show that while QE stimulates overall lending, its effects are highly heterogeneous across sectors. Most of the responses are from construction, retail, wholesale, finance and real estate sectors. When I further split the loan data by firm size, I find that small and medium-sized firms (SMEs) across all non-manufacturing sectors tend to borrow more following a QE shock. In contrast, while large firms in the previously identified sectors also increase their borrowing, those in other sectors show a mild decline. Overall, the distribution of the stimulative effect across sectors appears to be related to the allocation of zombie firms. However, this paper does not establish a causal relationship between the two. A formal examination would require analysis based on firm-level data.