With 4+ years of experience in economics and data science, I specialize in using causal inference and economic principles to solve business challenges. I like using data to answer key questions and value customer focus, collaboration, and growth.

Contact: trangthao.nguyen177@gmail.com

What’s new

• [13 Dec 2024] A new blog on Medium “Understanding challenges in deploying ML” is posted.
  • In this post, I summarized the key challenges of running machine learning in practice, from handling data inconsistencies and integrating with existing systems to monitoring model performance after deployment. These insights highlight the complexities of operationalizing ML and offer strategies by other tech firms to address them effectively, ensuring that machine learning systems deliver value in real-world settings.
• [12 Dec 2024] A new blog on Medium “Understanding measuring long-term effects” is posted.
  • In this post, I summarized how to measure long-term outcomes in A/B testing. While short-term results are easy to track, they often miss lasting impacts. Three approaches can help: Long-term experiments or holdouts compare groups over time but are slow and costly; Modeling user learning tracks how users adapt to changes, offering better insights into lasting effects; and Surrogate metrics use short-term data to predict long-term outcomes, enabling faster decisions. I also summarized examples from Instacart, Netflix, and LinkedIn to show these methods’ benefits and challenges (especially in using surrogate metrics), emphasizing the need to balance speed and accuracy in decision-making.
• [12 Dec 2024] A new blog on Medium “Understanding triggered analysis” is posted.
  • In this post, I summarized triggered analysis in A/B testing, which focuses on users directly affected by the experiment (e.g., those reaching a checkout page). This method improves sensitivity by reducing noise from unaffected users but comes with challenges like smaller sample sizes and generalization issues. Key techniques like CUPED help reduce variance by leveraging pre-treatment data, making the analysis more precise and robust. For more details, see Larsen et al. (2024) and Kohavi et al. (2020).
• [12 Dec 2024] A new blog on Medium “Understanding inference” is posted.
  • In this post, I summarized interference in A/B testing, where users’ behavior influences others, breaking the assumption of independent effects (SUTVA). Common types include: Network Interference (i.e. Users affect each other through connections (e.g., LinkedIn messaging)), and Marketplace Interference (i.e. Users compete for shared resources (e.g., ride-sharing drivers). Suggested solutions include Cluster Randomization (i.e. Treat entire user clusters to limit interference), Switchback Experiments (i.e. Alternate treatment and control over time), Budget-Split Designs (i.e. Divide shared resources evenly in marketplaces), and Modeling Interference (i.e. Use network or marketplace models to account for interactions). These methods reduce interference but often trade off statistical power.

Research

• “Bank Expansion, Firm Performance and Gender Gaps: Evidence from Vietnam” - Paper, Slide
  • Expansion of bank branches has played a pivotal role in enhancing financial development and access to finance. Using firm-level and district-level data from Vietnam during 2007 - 2019, we document that (1) increased bank expansion has a positive impact on a firm’s revenue and number of employees, but not on labor productivity; and (2) this result is driven by men-led firms, while for women-led firms, we find no effects on firms’ sales, employment and labor productivity. This might be because men-led firms have a higher probability of getting loans compared to women-led firms with increased exposure to banks. Another interesting result is that if firms get loans, women-led firms have a higher probability of adopting modern technologies, such as cloud computing technology, advanced robots, and additive manufacturing technology (3D printing), with an increase in exposure to banks. Our results highlight the importance of bank expansion in increasing firms’ performance in developing countries, but suggest more policies in supporting women-led firms in accessing to banks, apart from increasing the number of bank branches.
  • 
• “Special Economic Zones and Firm Performance: Evidence from Vietnam” with Karsten Mau and Tania Treibich - Paper, Slide
  • This paper examines the impact of Special Economic Zones (SEZs) on firm performance in Vietnam, using firm-level dataset and employing cancelled SEZs as a control group to address endogeneity concerns. We find that SEZ entry leads to signifcant improvements in firm outcomes, with direct effects including a 18.3% increase in employment, a 55.3% rise in sales, and a 25.9% boost in labor productivity. Indirect spillover effects are also observed within communes hosting SEZs, particularly through improved labor productivity and sales of non-SEZ firms. Heterogeneity analyses reveal that foreign firms, large firms, science-based, and supplier-dominated firms beneft the most. Firms located in industrial SEZs seem to drive our results. We also provide insights into the mechanisms driving these effects, including enhanced credit access to explain why direct effects are stronger than indirect ones. Input-output linkage might explain why we have substantial effects for employment and sales, especially for spillover effects. However, the technology gap still remains a challenge for domestic firms as they experience labor productivity improvement when learning from SEZ firms with foreign direct investment originating from developing countries but show no signifcant gains when FDI originates from developed countries.
  • 
• “Automation Adoption and Export Performance: Evidence from French firms” with Giacomo Domini, Marco Grazzi, Daniele Moschella, and Tania Treibich - Paper, Slide
  • The debate about the impact of new automation technologies (industry 4.0 revolution) has been intense for a while. The topics mainly focus on the aspect of “stealing” the jobs of new automation technologies. However, this debate seems to be biased towards the aspect of replaceability of new automation technologies that it seems to ignore the advantages of new automation technologies on the export activity of firms and countries. In this paper, we aim to fill these gaps. We use French firm-level data from 2002 – 2017 and examine the question how do automation technologies affect firms’ export activity, i.e. export quality, quality-adjusted price, export values, number of exported products and number of exported countries. We propose three potential underlying channels that explain the effects of automation technologies on export activities, including product innovation, process innovation and sunk cost channel. Our assumption is that automation technologies have the aspect of “automatic” – which produce fewer errors compared to humans, provide a consistency of accuracy in producing products. In our preliminary results, we find that firms adopting automation technologies have higher number of exported countries and number of exported products compared to firms not adopting automation technologies. Similarly, firms adopting automation technologies also have higher export values and the top value, only for max share of exports, firms not adopting automation technologies have higher value compared to non-adopters. In terms of quality, adopters also have higher quality compared to non- adopters, higher quality-adjusted price, and higher unit value which is consistent with our expectations in conceptual framework. We also note that there is an increase in log exports value, quality, quality-adjusted-price, log unit price, number of exported countries, and number of exported products. However, one year after the time of the automation event, quality and quality-adjusted price decreases, so the evolution around an automation event of quality and quality-adjusted price is ambiguous in this case. Similarly, two and three year after the time of the automation event, number of exported countries decreases, so to say about the effects of automation on automating firms around an automation event, we still need to run more careful analysis to have a more accurate picture.
  • 
• “Automation Adoption and Reshoring: Evidence from Asian Input-Output Table” - Paper, Slide
  • The introduction of technologies such as computers and ICTs to better coordinate production organization and the opening of lower labour cost countries have contributed to an international fragmentation of production in the 1990s and 2000s. However, the recent rise of new automation technology in production and service has raised concerns about disrupting global value chains. In this paper, we examine the role of automation adoption as a driver of reshoring in the period 2008-2019, using a new measure that takes into account both intermediate and final imports, considers reshoring as a flow process, and includes direct and indirect effects. We find a negative relationship between automation adoption and reshoring, indicating that automation adoption reduces reshoring. We also find that this negative relationship is more pronounced for high-income and lower-middle-income countries, and for adoption of ICT and 3D printing technologies. We examine different time periods and find that the negative relationship between automation adoption and reshoring was strongest in the period 2008-2013, with a magnitude of around 0.28 percent if automation adoption increased by 1 percent. We find that automation adoption reduces reshoring in both manufacturing and service sector, but service sector drives this relationship. Our results suggest that he view that automation technology can replace offshore tasks and promote reshoring is not yet complete.
  • 
• “Social Network and Economic Development” - Paper
  • I use anonymized data from Facebook to construct social diversity and social connectedness weight matrix of counties in the U.S to evaluate the relationship between social network and economic development. I find that there is a negative relationship between social diversity and MDI rate, which further confirms the theoretical understanding that diverse contact helps improving socio-economic performance at individual and community levels. However, once I estimate this relationship for only the 10 percent counties having highest social diversity, this relationship disappears which implies that the relationship between having high social diversity does not hold true for better economic development performance for counties with highest social diversity level. I further ask the question of social and spatial spillovers in economic development between counties and try to answer by estimating an econometric specification for geographical and social connection weight matrix. My findings suggest that there are spillover effects among counties in terms of geographical and social connection. However, I cannot conclude these effects as causal relationships. Nevertheless, this paper can be considered as another step in understanding the population-level relationship between social network and economic development based on online network data to proxy for real life relationships.
  •