Sansan Tech Blog


We Held a Global Mini-Conference!

Hi there. I'm Juan, from Sansan R&D.

As you may know if you follow us on social media, one important part of our work as Social Scientists at Sansan R&D is performing research. It is how we find new creative ways to understand and process the data coming from our services and turn it into new value for our customers. We conduct some of our research in collaboration with academics performing leading research in several fields. One of our ongoing projects is a collaboration with professor Angelo Mele. This year, we decided with professor Mele to practice our network formation skills and engage the academic community by hosting together a small conference. In this post I talk about this event.

The Concept

The idea of the event was to discuss everything related to networked data, including its many challenges, new methods, topics and applications. That's why we called it Network Economics: Applications and Challenges. Many methods employed in Economics assume that observations are independent from each other. However, network data is the exact opposite setting! It is because observations are related in some way that the data is interesting; but this makes the analysis very complicated. So we were interested in what new methods and applications other leading academics are developing to work with such data. Note that although the event includes the keyword network, we are actually interested in all sorts of settings where the relationship between the agents is the focus of the analysis.

We had the pleasure of counting with the participation of three distinguished professors as guest speakers: Eleonora Patacchini from Cornell University, Vincent Boucher from Laval University, and Ryo Nakajima from Keio University. Add professor Mele to the mix and I'd say the lineup was very impressive.

Our small conference was held on May 21st from 18:00 (EDT), as it was the best one given the schedule of the participants.

I will now provide a very brief summary of everyone's presentations.

Information, Mobile Communication, and Referral Effects

Eleonora was the first person to present. She showed results from her team's recent study on the estimation of the effect of job referrals. Their paper makes use of mobile phone data along with other micro-data from a city in China, and attempts to estimate the effect of referrals on several outcomes. They find evidence of referrals having a significant effect on the choice of the place to work, as well as in other outcomes for workers and companies, including higher incomes, regular employment, the success of recruits, etc.

Estimating the effect of referrals is difficult, first because it requires data on social interactions, which is hard to find, and because there are so many confounding factors (homophily, sorting, preference to work with friends, etc.).

They employed data on mobile communications to capture social interactions, and applied all sorts of controls to clean their estimates from other confounders. My impression was that there's a lot that we can learn from Eleonora's research in terms of methodology and regarding the effect of referrals.

Towards a General Theory of Peer Effects

The second presenter was Vincent, with a presentation about Peer Effects. It's probably easier to explain with an example: imagine that you are the manager of a company which has several sales teams. You have records of the number of visits that sales persons make to the customers. You will likely see that persons who make many visits tend to be in the same team, or interact a lot. This is a pattern that is commonly observed in Economics and can be attributed to peer effects (when the behavior of peers influences your own behavior).

Vincent's presentation deals with the problem of modeling peer effects. Going back to our example, you could imagine that the number of sales visits depends on the characteristics of the sales person (their level of experience, the item they are selling, their location, etc.), and some combination of the number of visits of its team mates. However, we don't know if the peer effect is only the result of spillovers*1 or of conformism*2, or a combination of both!

Vincent's research shows that it is possible to create an economic model that considers both spillovers and conformism, as well as nonlinearities of the peer effect, and provide the conditions for identification. He also showed that the actions that a policy-maker (let's say, a sales manager) can make to improve the outcomes will depend on whether spillovers or conformism are the leading source of peer effects.

I was impressed by Vincent's ability to explain economic theory in an easy way. He literally used two equations and a bunch of stick figures to explain his paper. I think that his team's contribution can be very useful for the purpose of understanding collaboration within firms and to optimize the performance of work teams.

Marshall Meets Bartik: Revisiting the Mysteries of the Trade

Ryo was in charge of the third presentation. In this research with professor Yasusada Murata, Ryo focuses on a famous quote by Alfred Marshall:

So great are the advantages which people following the same skilled trade get from near neighborhood to one another. The mysteries of the trade become no mysteries; but are as it were in the air, and children learn many of them unconsciously.

This quote is about the benefits of industrial agglomeration, and more specifically about knowledge spillovers. Naturally, there's some controversy about whether it is true that knowledge is in the air, and how impactful that spillover is. In Ryo's presentation he shows how they attempt to answer these questions by employing data from the United States on patents and inflows of top inventors. Their identification strategy employs a Bartik estimator and instrument the inventor's migration probability using data on income tax differences across states. They find evidence of a significant and important knowledge in the air effect, where the migration of a top inventor increases the productivity of local inventors.

Ryo's presentation is closely related to our own ongoing research project on urban agglomeration (of which I hope we can write more about soon), as it shows that urban centers provide advantages through knowledge spillovers. How this effect influences the locations of firms and the mechanics behind business encounters is an interesting research topic that we plan to follow very closely.

The Economics of Business Networks: Estimation and Applications at Scale

Finally, there was a double presentation between Angelo and Sansan, Inc. During the first part, there was an introduction of Sansan, its services and the team of economists currently working at R&D's SocSci*3 Group. We also briefly talked about Eight's data and the digitization process behind the encounters database.

During the second part, Angelo presented the preliminary results of our research project on network formation using anonymized data from Eight. Our research attempts to model the network formation process among Eight users. The complexity of traditional methods makes it difficult to work with networks larger than a few hundred nodes. Hierarchical Exponential Random Graph Models (HERGMs) make it possible to perform the estimation process in a more scalable way. For our research, we created a more efficient implementation of the HERGM algorithm*4, which makes it possible to perform estimation on networks formed of hundreds of thousands of nodes. We also made it possible to employ information on the characteristics of the nodes for the block recovery step. Our paper develops a structural model of business card exchanges and applies HERGM to estimate the structural parameters. Angelo used simulations to compare the performance of our implementation with that of other algorithms.

You can find the working paper over here:

The software library can be found on GitHub on the link below:

Final Impressions

I'm personally very satisfied with our mini-conference. Presenters were all very friendly. The discussion after each presentation was very rich and full of new insights, and in general I think we all had a very good time. We also had many valuable comments from the audience, including some suggestions for improving our own research. We realize that the time was a bit inconvenient for persons in Europe and Africa (not mentioning that we held it on an early Sunday morning in Japan Time!). Hosting a global event online is difficult because of time differences, but we will do our best to make it more convenient in the future.

We're very thankful to everyone, presenters and participants, who took some time of their weekend to join us. Also I'd like to thank professor Angelo Mele for co-hosting this event and inviting such wonderful speakers. Also thank you so much to all the persons at Sansan, Inc. who were involved in the logistics of our mini-conference. I personally believe that collaborations between industry and academia are win-win relationships that can have a massive impact in the way we provide value for our customers. I'm really excited for all the future research projects to come!


  • Eleonora Patacchini & Yves Zenou, 2012. "Juvenile Delinquency and Conformism," Journal of Law, Economics, and Organization, Oxford University Press, vol. 28(1), pages 1-31.

*1:The experience of the team mates being shared.

*2:Patacchini & Zenou (2012) summarized it very well: "Conformism is the idea that the easiest and hence best life is attained by doing one’s very best to blend in with one’s surroundings and to do nothing eccentric or out of the ordinary in any way."

*3:Social Sciences.

*4:We're very thankful for the feedback we obtained from professor Michael Schweinberger, one of the authors behind the HERGM algorithm and its original R implementation.

© Sansan, Inc.