My research and advising: Topics

Author: Stephen J. Turnbull
Organization: Faculty of Engineering, Information, and Systems at the University of Tsukuba
Contact: Stephen J. Turnbull <>
Date: February 9, 2019
Copyright: 2019, Stephen J. Turnbull

This page is an introduction to my interests for students who want to work on topics close to my own research, past, present, or potential. There are other topics I have advised and am willing to advise. I do not restrict advising to these topics. You're welcome to propose others. These are just the obvious candidates from my point of view. (Writing a research plan is more than just choosing a topic, and I am more interested in how the plan shows how you think about research than in an exact match of your topic to my interests as described below.)

If you're thinking about working with me, you should think about your research topic and how it relates to my interests.

If you contact me directly, you must mention your own research interests. Writing only that you're interested in my research without being specific is a signal that you don't actually know what I do. It's a very bad signal. Of course you are welcome to present a research plan in more or less detail, and it is a good signal, but it's not necessary.

Note 1: Terms in italics are technical terms that I don't expect students to know to work on the related topics. If there are technical terms in roman type you don't know, expect working on those topics to require a lot of preliminary study.

Note 2: This page is a work in progress. Until April 2019, it will likely be updated frequently, after that occasionally.

Note 3: Eventually I will translate this page into Japanese and maybe Chinese.

Game theory

My principal research interest is in game theory and mechanism design. How should people plan their behavior when they know that others are adjusting their own behavior to account for their expectations of how you will behave?

Category theory and game theory

I'm currently in the second year of a three-year grant for studying category-theoretical methods for use in analyzing game-theoretic problems.

For example, if you restrict the firms in a Cournot duopoly to the quantities that result in the Cournot equilibrium or the collusive (monopoly) outcome, the resulting matrix game is a prisoner's dilemma. Is this more than a curious fact? Does this embedding tell us about something about how to understand these two games?

Another more speculative topic looks at the fact that strictly speaking, there is no special theory of mixed strategies in games. Instead, we create the "mixed extension" of a game using the probability distributions over the original "pure" strategies as strategies and the expected value of the original payoff functions as payoffs in the extension game. Then we "forget" the probability structure, and solve the extension game using ordinary Nash equilibrium. This structure ("extension and forgetting") is called a monad in category theory. Perhaps different monads can be defined that would allow (for example) prospect theory (which treats losses differently from gains) to be used instead of utility theory to define players' goals.

A final topic, more educational than analytic, is to try to clarify the relationship between moral hazard and adverse selection in markets and games, especially in view of Roger Myerson's work on these problems in extensive-form (dynamic) games.

Game theory applications

I'm thinking about game theoretic analysis of national security issues. I'm interested in comparing deterrence in cyberwar to deterrence in the conventional and nuclear spheres. The interesting aspect of cyberwar is that many of the "weapons" are exploits that allow an attacker to take over a system, at least temporarily. But the usefulness of these exploits tend to decay relatively quickly over time because of upgrades to software or firmware, improvements to overall network security, and rediscovery and publication of the vulnerability by friendly agents. This generates a "use it or lose it" incentive. On the other hand, exploits frequently leave a trace that makes it plain how they work, allowing defenders to patch the vulnerabilities. This leads to a "lose it if you use it" incentive. Finally, an exploit may open the door to "taking up residence in an adversary's systems and networks", obviously of great help in intelligence gathering, leading to the usual "sources and methods" conflicts of incentive: "let's not use this offensively at all!"

A similar application is to information security in general. You might think that this would have an analysis similar to the national security case, but they're actually rather different. In national security, the analysis is about defending specific targets against attack, often based on unique vulnerabilities in critical infrastructure. Information security in general is similar to a public health problem, where the worries are epidemics and a widespread condition of poor health and small injuries.

Urban economics

Recently I have been working with Prof. Mitsuru Ota and his student Noriyuki Hiraoka on a model of city growth, where we look at the effects on residential location choice, wages, and overall welfare of workers with different productivities as cities become larger.

Our solution concept, however, is static general equilibrium. That is, our results on effects on worker behavior and welfare are a comparative static analysis. But income and housing are the two biggest components of a worker's budget, so endowment effects and the dynamics of changing location should have large effects on our analysis. Two interesting potential projects are an economic growth model of city growth based on our static model, and a simulation study to understand the nature of endowment effects in a dynamic model.

Biases in technological progress

The advanced societies of Europe, North America, Australia, and Japan are experiencing a "hollowing out" of their economies in the sense that the factory jobs which were the mainstay of improving standards of living since World War II, especially for the non-college-educated working class, are now being automated or off-shored to lower-wage emerging markets such as India, Southeast Asia, and China. The more advanced second-tier economies such as South Korea, Taiwan, and in the near future China are likely to face a similar phenomenon. As we have seen in Europe and the US, aside from the loss of welfare for the affected workers, this led to populist movements threatening the stability of their governments and economies.

Empirical studies

Mark Muro of the Brookings Institution recently participated in a podcast, "How automation and AI are redefining work," based on a Brooking report authored with Robert Maxim and Jacob Whiton. Podcast Report

This report is specific to the US. I am interested in similar work on Japan, China, and perhaps other countries or regions, gather data sets and performing the analysis.

Management strategy

I am also interested in the question of how to engineer for small scale. That is, automation generally is used to substitute for labor, allowing a business to grow rapidly without the frictions associated with human resource management. It should be possible to create technology that increases the productivity of the production worker without substituting for her, thus allowing both production and wages to grow. I don't know how to do this work yet, but here's an example of the kind of idea I'm thinking about. One of my recent students looked at how social media can cheaply substitute for expensive market research using surveys for niche markets, which might allow useful market research in the hospitality industry to be done by small businesses and municipalities.

Social networks

See Chapter 1 of Kleinberg and Easley's excellent textbook Networks, Crowds, and Markets: Reasoning about a Highly Connected World. It provides a great overview of the scope of the field as well as the methods used.

Open source project management

Open source software has become an important component of modern information and communication systems. The Linux operating system is perhaps the most famous example, but most Internet functionality has open source implementations, all cryptographic functionality has open source implementations, and some of these implementation dominate their fields. Linux itself does in server markets, though Microsoft Windows is making a comeback. Almost all compilers and interpreters for programming languages are open source, although there are a few proprietary modules to handle special hardware such as video cards and parallel processing with GPUs. And the Apache webserver is still at least as popular as Microsoft's webserver.

I spent some effort studying the Python language development process a few years ago, and the project's recent transition from the benevolent dictator model to a somewhat more democratic model means it's still a very interesting topic for study.

Computational general equilibrium modeling

Computational general equilibrium (CGE) is a simulation technology used in macroeconomics, international trade economics, and development economics to provide quantitative estimates of real effects of policy changes and their welfare implications. I have the necessary software (GEMTAP and GAMS) to solve a wide variety of models, as well as the GTAP global trade model dataset, and some experience with them. You will need to access domain expertise in some other way in most cases (i.e, previous study as an undergraduate, self-study, or through coursework and appropriate selection of your advisory group).


Although my research is not in statistics, it was one of my specialties in graduate school and I have studied big data theory in recent years because it's heavily used in social network studies. As with CGE, you will need additional guidance or self-study in domain knowledge for many fields (see below).

"Big data", machine learning, and artificial intelligence

Big data is a confusing name for data sets that have high dimensionality (number of variables) and low information density. The high dimensionality means that it's hard to decide models in advance, so we'd like the data itself to tell us. Low information density means you need a lot of data to get significant results. These problems are addressed by statistical methods that are often called artificial intelligence (which they aren't!), and better called machine learning (something else that it really isn't) or automatic pattern recognition (the best name for the technology).

These methods are frequently used in social network studies and textual analysis.

Warning: Conventional statistics and Bayesian analysis

In most fields of economics and business studies, there are highly developed statistical methods adapted to those fields. The Faculty of Policy and Planning Science has many specialists, so you might be better off with one of them as an advisor.

Also, in the past I have advised many theses based on marketing themes, trying to analyze consumer behavior or demand in a particular industry. However, these topics are generally not well thought of in our department, because they're rarely very original. In these topics, the examiners (both in the entrance exam and in the final presentation of a thesis) generally demand new methods or new models, rather than simply applying well-known models to new data. I'm willing to discuss how to adjust a topic to be sufficiently original to meet this standard, but you should be aware that naive proposals of this kind are likely to be unattractive to our faculty.