This paper reviews the path of historical development of statistics and statistical science in Japan from the Meiji Restoration (1868) until the early Showa Era (mid 1930's) from a viewpoint of contribution of official statistics. It discusses the interaction of official statistics with other fields of statistics, and considers the future prospects of the relationship of official statistics with overall statistics and statistical science. Since the introduction of statistics to Japan in the early Meiji era, official statistics contributed to the development of statistics at large by providing materials and resources for statistical analysis and applications of statistical methodology. It also took roles in statistical education and training, which contributed to dissemination of statistical knowledge to the society at large and academia, as well as government institutions. Further, it gave impetus to the foundation of the statistical society. The roles of official statistics in the development of statistics and statistical science remain to be essentially the same in the present day. Those who are engaged in official statistics should continue their efforts to contribute to promotion of statistical knowledge in the society and the development of statistical science.

Banks employ attribute information such as income, family status, work status to screen borrowing applicants for personal card loans in general. Personal behavior characteristics may also affect defaults but are not taken into account. In this paper, we analyze them using deposit-withdrawal data of bank account, and construct a model of evaluating default for the purpose of screening the applicants for card loans. We prepare some variables related to personal behavior characteristics such as number of commission payments, average deposit balance, peak balance ratio, using about 7.6 million data, and analyze the relationship with default. In addition, we construct a logit model with these variables, and examine the model using accuracy ratio (AR). The result shows that the AR exceeds 50%, and we find the model is effective in practice. We also confirm the robustness of the results through out-of-sample test and cross-validation.

The covariance of the log-return of financial assets is a fundamental element in a wide range from asset allocation to risk management. In this paper, we execute an empirical analysis using a high-dimensional realized covariance estimation method proposed by Brownlees et al. (2018) which performs lasso regularization. Specifically, we first estimate the sparse inverse integrated covariance matrix of some assets listed on Nikkei stock average in 2016 and create networks of the market. Then we analyze it from the network perspective and compare the network structure of each industrial sector. As a result, we conclude that the Japanese stock market does not depend on one company, that is, it is not a scale-free network. Furthermore, we confirm that the network structure moves to correspond to the Nikkei stock average.

This paper reviews Continuous time Auto-Regressive Moving Average (CARMA) models.We define CARMA models as a natural extension of discrete time ARMA models through state space representations. After the continuous time extension to CARMA models, we introduce the causal stationary conditions, derive the explicit forms of covariance and spectral density functions, show the joint distributions and examine the second order properties of regularly sampled CARMA processes. Finally, we review two empirical applications to high frequency data of exchange rates and Brookhaven turbulence data. The contents of the paper are based on the presentation slides of Professor Peter Brockwell for his plenary session in Japanese Joint Statistical Meeting at Shiga University in 2019.