井上 寛康

This work aims to study and explain the business cycle correlations of the Japanese production network. We consider the supplier-customer network, which is a directed network representing the trading links between Japanese firms (links from suppliers to customers). The community structure of this network is determined by applying the Infomap algorithm. Each community is defined by its GDP and its associated business cycle. Business cycle correlations between communities are estimated based on copula theory. Then, based on firms' attributes and network topology, these correlations are explained through linear econometric models. The results show strong evidence of business cycle correlations in the Japanese production network. A significant systemic risk is found for high negative or positive shocks. These correlations are explained mainly by the sector and by geographic similarities. Moreover, our results highlight the higher vulnerability of small communities and small firms, which is explained by the disassortative mixing of the production network.

This study investigates the localization of establishment-level knowledge creation using data from a Japanese patent database. Using distance-based methods, we obtain the following results. First, Japanese knowledge-creating establishments defined by patenting experience are significantly localized at the 5% level, with a localization range of approximately 80 km. Second, localization is observed for all patent technology classes, and the extent of localization has a positive relationship with the level of technology measured by R&amp D investment. Finally, the extent of localization is stronger for establishments that are more productive in terms of both the number of patents and the number of citations received, i.e., quantitatively and qualitatively. These results indicate that geographical proximity is important for knowledge spillover, particularly for establishments that demand external knowledge intensively.

The Small Business Innovation Research (SBIR) program conducted by the Japanese government is intended to enable the rapid growth of small firms. Using comprehensive empirical data, this study examines the performances of firms that are directly affected by the program. First, to provide an outline, this study compares the changes in sales, employment, and the number of patents between SBIR awardees and matched firms. It cannot be shown that SBIR awardees have better performance using those indices. Second, this study conducts regression analyses to control for firm sizes, multiple awards, technological levels, the value of venture capital in a region, and population in regions. As a result, this study does not find that SBIR awardees have advantages after applying these detailed controls. This study shows that further detailed investigation is required to prove the benefit of the SBIR program.

Using a large firm-level panel dataset for Japan, this paper examines the effects of the structure of supply chain networks on productivity and innovation capability through knowledge diffusion. We find that ties with distant suppliers improve productivity (as measured by sales per worker) more than ties with neighboring suppliers, which is likely because distant firms’ intermediates embody more diversified knowledge than those from neighboring firms. Ties with neighboring clients improve productivity more than ties with distant clients, which is likely because neighboring clients more effectively diffuse disembodied knowledge than distant clients. By contrast, ties with distant suppliers and clients improve innovative capability (as measured by the number of registered patents), whereas ties with neighboring suppliers or clients do not affect innovative capability. In addition, the density of a firm's ego network (as measured by how densely its supply chain partners transact with one another) has a negative effect on productivity and innovative capability, implying knowledge redundancy in dense networks. These results suggest that access to diversified ties is important for improving productivity and innovation capability through knowledge diffusion.

<p>サッカーは個人（ミクロ）と集団行動（マクロ）の両側面が同時に重要なスポーツで，複雑系として学問的にも興味深い研究対象である。最近のITの進展によってサッカーに関するデータ，特にフィールド上の選手やボールの位置が時々刻々集められている。そのようなトラッキングデータに複素ヒルベルト主成分分析法を適用することにより，サッカーにおける選手同士，選手-ボール間のリード・ラグ関係を含めた集団運動を抽出する。</p>

We focused on how repeat collaborations in projects for inventions affect performance. Repeat collaborations have two contradictory aspects. A positive aspect is team development or experience, and a negative aspect is team degeneration or decline. Since both contradicting phenomena are observed, inventors have a dilemma as to whether they should keep collaborating in a team or not. The dilemma has not previously been quantitatively analyzed.We provide quantitative and extensive analyses of the dilemma in creative projects by using patent data from Japan and the United States. We confirm three predictions to quantitatively validate the existence of the dilemma. The first prediction is that the greater the patent a team achieves, the longer the team will work together. The second prediction is that the impact of consecutive patents decreases after a team makes a remarkable invention, which is measured by the impact of patents. The third prediction is that the expectation of impact with new teams is greater than that with the same teams successful in the past. We find these predictions are validated in patents published in Japan and the United States. On the basis of these three predictions, we can quantitatively validate the dilemma in creative projects. We also propose preventive strategies for degeneration. One is developing technological diversity, and another is developing inventor diversity in teams. We find the two strategies are both effective by validating with the data. Copyright:

We studied the Japan and U.S. patent records of several decades to demonstrate the effect of collaboration on innovation. We found that statistically inventor teams slightly outperform solo inventors while company teams perform equally well as solo companies. By tracking the performance record of individual teams, we found that inventor teams' performance generally degrades with more repeat collaborations. Though company teams' performance displays strongly bursty behavior, long-term collaboration does not significantly help innovation. To systematically study the effect of repeat collaboration, we defined the repeat collaboration number of a team as the average number of collaborations over all the teammate pairs. We found that mild repeat collaboration improves the performance of Japanese inventor teams and U.S. company teams. Yet, excessive repeat collaboration does not significantly help innovation at both the inventor and company levels in both countries. To control for unobserved heterogeneity, we performed a detailed regression analysis and the results were consistent with our simple observations. The presented results revealed the intricate effect of collaboration on innovation, which may also be observed in other creative projects.

Companies are exposed to rigid competition, so they seek how best to improve the capabilities of their innovations. One strategy is to collaborate with other companies in order to speed up their own innovations. Such inter-company collaborations are conducted by inventors belonging to the companies. At the same time, the inventors also seem to be affected by past collaborations between companies. Therefore, interdependency of two networks, namely inventor and company networks, exists. This paper discusses a model that replicates two-layer networks extracted from patent data of Japan and the United States in terms of degree distributions. The model replicates two-layer networks with the interdependency. Moreover it is the only model that uses local information, while other models have to use overall information, which is unrealistic. In addition, the proposed model replicates empirical data better than other models. © 2014 The Author. Published by Elsevier B.V. This is an open access article under the CC.

Products recently need better and more complex technologies than they used to. In addition, companies can not afford to research and develop new technologies or scientific theories because companies face fierce competition. Therefore they need to collaborate with other organizations, including not only other companies but research institutes. However, the collaborations have costs to take over boundaries of organizations. This paper reveals which categories of technologies Japanese companies tend to collaborate in, and the productivity of inventors in organizations. The following facts were obtained. (1) The highest ratios of joint applications between companiess and research institutes are in "chemistry and metallurgy", "commodities" and "physics" in that order. (2) Companies tends to choose research institutes as joint-applicants instead of other companies over collaborating with other companies. (3) The experience of inventors' joint-applications between companies does not affect inventors' productivity, but the experience between companies and research institutes does affect their productivity. (4) In a field where inter-organizational joint-inventions provide some advantages to inventors, inventors' experience of inter-organizational joint-inventions tends to their acquisition of experience.

To sustain economic growth, countries have to manage systems in order to create technological innovation. To meet this goal, they are developing policies that organically connect companies, national laboratories, and universities into innovation networks. However, the whole structures of these connections have been little investigated because of the difficulty of obtaining such data. We use Japanese patent data and create a network of jointly applying organizations. This network can be considered as one representation of an innovation network because patents are seeds of innovation and joint applications are strong evidence of connections between organizations. We investigated the structure of the network, especially whether or not the degree distribution follows a power law. After that, we also propose a model that generates the actual network, not only degree distribution, but also link distance distribution. © 2009 Elsevier Ltd. All rights reserved.

Firms have to provide for international competition more than ever before. To do this, they have to exploit resources around them so that they can set off deficiency of ability. Therefore, they try to build relations with other firms. There are various types of relations, such as transactions, shareholdings, executive dispatching, cooperative researches, developments, and alliances. In addition, the relations express complex networks. This paper treated transaction networks as one of such networks and selected the Japanese software industry as an example. This was because it has been pointed out that the industry has some structural defect and a whole picture of the structure has not been revealed. I used a p* model to analyze the structure.

Technological innovation has extensively been studied to make firms sustainable and more competitive. Within this context, the most important recent issue has been the dynamics of collaborative innovation among firms. We therefore investigated a patent network, especially focusing on its spatial characteristics. The results can be summarized as follows. (1) The degree distribution in a patent network follows a power law. A firm can then be connected to many firms via hubs connected to the firm. (2) The neighbors' average degree has a null correlation, but the clustering coefficient has a negative correlation. The latter means that there is a hierarchical structure and bridging different modules may shorten the paths between the nodes in them. (3) The distance of links not only indicates the regional accumulations of firms, but the importance of time it takes to travel, which plays a key role in creating links. (4) The ratio of internal links in cities indicates that we have to consider the existing links firms have to facilitate the creation of new links. © 2007 Elsevier B.V. All rights reserved.