Embedded Files
My thesis focuses on studying the relationship between technology and economic inequality across geographic locations with the help of network theory. In addition, I also apply machine learning on various economic problems to improve the current quantitative methods in economics.
My thesis focuses on studying the relationship between technology and economic inequality across geographic locations with the help of network theory. In addition, I also apply machine learning on various economic problems to improve the current quantitative methods in economics.
This paper examines how changes in urban industrial network structures explain the growth rates of labor productivity in cities. I formulate a multi-sector general equilibrium model with input-output networks of firms within a city and trade across cities. A key input to this framework is an Shannon entropy-style city network entropy index. It serves as a concise summary of urban industrial structures and describes the concentration level of inputs for an average firm in a given city. The major theoretical result is that the improvement of urban industrial structures, indicated by an increase in city network entropy, leads to an increase in the urban labor productivity growth rate. This is because changes in city network entropy are results of both city-specific technological shocks and the evolution of the nationwide input-output structure. When these two forces align in a way that increases city network entropy, production activities in a city become better organized, and its labor productivity grows faster. In MSA-level data from BEA, I verify the theory by showing that changes in network entropy are positively correlated with a city’s productivity growth. In two sets of counterfactuals, I demonstrate how the interaction between technology changes and urban industrial networks determine urban labor productivity. First, I find t hat the presence of urban industrial networks explains 54.7% of the variance in changes in urban labor productivity caused by local sectoral shocks. Second, I demonstrate that the variance in city network entropy can explain 45.3% of the variance in growth rates of urban labor productivity caused by shifts in the national I-O structure.
This paper demonstrates that the human capital spillover in urban industrial networks is an important factor to explain productivity variance in industries across cities. I create a human capital network index to measure the level of skilled workforce in the local environment of an industry. For a target industry, the index achieves high values not only when the city has a large quantity of educated workers, but also when this human capital is concentrated on industries that are economically close to the target industry. My analysis requires the knowledge of city-specific industrial productivity. Therefore I build a general equilibrium model with multiple industries within cities and competitive trade among firms across cities. Then I calibrate the model to the U.S. economy with MSA-industry specific data from BEA, trade flow data from CFS and individual-level labor market data from ACS. The empirical analysis shows that there are three factors that decide the influence of urban human capital spillover on the productivity of an industry: 1) the general quantity of educated workforce in the city, 2) the concentration of human capital in an industry’s input-output network 3) the ability of an industry to absorb the spillover. While the majority of sectors benefit from a more educated urban environment, certain industries experience negative human capital spillover from the rest of the city.
This paper explores the empirical evidence that traces credit risk propagation in a intersectoral input-output (IO) network. The major finding is that after ranking the supplierindustries for a specific target industry based on the weight of input shares from IO tables, the average probability for firms in an industry to default, get delisted and go bankrupt has higher and more significant correlations with more important supplier-industries than with less important supplier-industries. This relationship is robust to controls and both linear and nonlinear specifications of the model. Our conjecture for the cause of the observed financial contagion through the production network is that risk of firms passes to upstream suppliers through the heavy usage of trade credit. Once customers default or go bankrupt, net worth of suppliers will shrink and therefore the chance for insolvency for upstream firms increase. Moreover, we argue that firms have incentive to rely more on trade credit when purchasing major inputs that counts for a larger share of production costs than buying minor categories of inputs. Therefore, the credit risk contagion should be higher for major supplier-industries than for minors ones.
Work in Progress
Work in Progress
"Default Prediction of Middle Market Firms" , with Lars-Alexander Kuehn
This paper investigates the difference in default behaviors of middle market firms and big public firms by using the default records of commercial loans from a private bank and macro economic data from 2005 to 2015. I apply various machine learning technique to show that firm's own financial conditions, Macroeconomic conditions of the labor market and economic neighbors in production networks can influence the probability of default for middle market firms. The influence of these factors is different for big public firms. The paper will develop a structural model to explain why two types of firms are affected differently by these characters.
Publications
Publications
• "Regional and Sectoral Patterns and Determinants of Comparative Advantage in China" with William Charles Sawyer & Kiril Tochkov, Frontiers of Economics in China, vol. 12(2017), 7-36.
• "Regional and Sectoral Patterns and Determinants of Comparative Advantage in China" with William Charles Sawyer & Kiril Tochkov, Frontiers of Economics in China, vol. 12(2017), 7-36.
China's export performance is marked by large regional disparities which affect trade patterns at the national level. This paper uses data from input-output tables to estimate the comparative advantage of Chinese provinces in the three main economic sectors over the period 1992 to 2007. In contrast to existing studies, we include the services sector in the analysis and construct not only indices of revealed comparative advantage for overall trade, but also bilateral indices for interprovincial trade. The results indicate that West and Central China have a comparative advantage in agriculture/mining, coastal provinces in manufacturing, and metropolitan provinces in services. However, interprovincial trade exhibits a more complex pattern. Regression analysis identifies labor endowments as the key determinant of comparative advantage in total trade, while physical capital is the driving force in domestic trade. Human capital and government spending have a positive effect, whereas industrial loans and taxes, along with provincial trade barriers, impair comparative advantage.
• “Sectoral Productivity and Regional Disparities in China 1978-2006” with Kiril Tochkov, Comparative Economic Studies 55, no. 4 (2013): 582-605.
• “Sectoral Productivity and Regional Disparities in China 1978-2006” with Kiril Tochkov, Comparative Economic Studies 55, no. 4 (2013): 582-605.
This paper examines labor productivity growth in the three main sectors in China in the context of regional convergence by employing a novel methodology and new sectoral data over the period 1978–2006. The results show that all sectors experienced major shifts in productivity but with different patterns. In agriculture and services, the uniform distribution has given way to a bimodal one. The secondary sector exhibits less polarization across regions despite higher mobility. Research and development (R&D) spending, human capital, and infrastructure are found to amplify regional divergence, while physical capital and foreign trade are identified as the key drivers of convergence.
Other Projects in Machine Learning
Other Projects in Machine Learning
“Extending Fair Classification to Non-binary Attributes”, with Momin M. Malik, Shayak Sen
As algorithmic models are increasingly used for making decisions with social impacts, it becomes important to incorporate principles into their design. We consider two approaches to achieving fairness in classification along a binary protected attribute, and extend both to the multiclass case. We find that both extensions are effective, although both techniques perform better when the protected attribute is binary.
“Survival Analysis for Financial Models”
There are couple of challenges for the application of survival analysis, combined with more advanced techniques on high-dimensional financial data. First, in high dimensional datasets, classic penalty and regularization method of variable selection becomes less efficient and potentially not consistent asymptotically. Also, the time-varying nature of financial market data makes it necessary to discuss the inference of coefficient estimates and predicted results. This paper goes through recent literature, discusses and summaries the inference of survival analysis for large-scale dataset and how to combine machine learning regression technique properly into analysis with time-dependent covariates.
Google Sites
Report abuse