On Wall Street, which is one of several places on the leading edge of financial engineering, it is said that robots (programs) are already playing a significant role in trading. Trading via programs, which is known as algorithm trading, is unlike trading through buying and selling by people, as in this world programs compete in trades that take place on a time scale of 1/1,000,000 of a second. Underlying the programs’ decision-making process on whether to buy or sell is big data. The questions that require answers include how to numerically express phenomena that are occurring in the world, how to combine and analyze these numbers, and how to connect them to information that can be used to read the economic future. We talked to Professor Hiroshi Tsuda of Doshisha University about joining the research commons Social Communication project and tackling the theme of tourism from the perspective of “finance.”

No world is so utterly expressed by numbers as finance

Although people meet face to face at bank counters, other than this we can describe finance as “a world of numbers” (laughs). Financial institutions are a mass of data and, in actuality, the flow of money is completely expressed by numbers. Moreover, we encounter major problems when the numbers are just a little unusual. Therefore, by analyzing this sort of data, we can connect it to various evaluations. For example, when a bank provides financing to a company, it estimates that company’s bankruptcy risk based on data analysis using tools from financial engineering and statistical financing. In addition, under Japan’s public pension management, since April 2014 the country’s stance has been to actively incorporate management based on a “smart” equity index that offers some sort of added value, such as higher returns and lower risks, compared to equity indexes used in the past, while the stock markets are also paying attention to trends in smart-beta management. However, this sort of development in Japan’s financial institutions is actually occurring late compared to what we have seen worldwide. On the other hand, tourism will become an increasingly important industry in the future. Package tours have been the mainstream up until now, but Free Independent Travel, in which individuals are able to reserve their own hotels and design their own trips via the Internet, is growing. Nevertheless, to date the actual situation is still not well understood. Therefore, as a new global attempt in which this development can be seen, Professor Sonehara and researcher Hiroshi Ichifuji are using data that exists on the Internet and that is accumulated in large quantities each day regarding the preferences and needs of independent travelers to ascertain which hotel they will choose.

A “logit model” that analyzes the effects on the levels of popularity of accommodation plans

As the database contains information acquired from an online reservation system for about 200 lodging facilities in Kyoto City, we can see that vacancies gradually decrease over time. Incidentally, under this sort of system, in practically all cases the accommodation plans will show a total of around 200 rooms, even when there are actually only 100 rooms. Thus, it identifies which plans are duplicating the same rooms and estimates the actual availability. Next, the various options attached to the plan are categorized, the “logit model” is used, and the factors that influence the sales speed are investigated. The logit model is frequently used to estimate, for instance, company bankruptcy risk in financial engineering, or to identify drug efficacy in the medical world. One of its major advantages is that it makes events easier to understand as it guides the researcher toward the probability of occurrence. For example, for rooms with twin beds, the model shows that the “with drinks” plan is effective; it also highlights the popularity of “souvenirs for women” among the various plans sold. By using this model, an analysis can be performed to uncover the reasons for popularity, even down to individual factors.

Using disclosed IR information to estimates sales from room-occupancy ratios

Moreover, the database is an accumulation of three years’ worth of information, so it can also ascertain seasonal variations in availability, as well as daily trends arising from, for instance, different days of the week. In addition, it can estimate the mean value of availability from a range of variables, such as room size or distance from a station. This data is also useful for comparing the monthly totals on availability at the main hotels, as announced by Kyoto City, and estimating the actual number of vacant rooms in the event that “there are vacancies.” When data on availability, whose accuracy has been improved in this way, is multiplied by the standard price calculated based on the price of plans for a bed without meals/with breakfast, sales can be estimated. At this point, the data can be used by banks for the actual risk management of hotel businesses, or by investment companies to evaluate hotels’ capitalized value. Thus, we can now attempt to combine this with IR information disclosed to investors by an RIET listed in the first section that invests only in hotels (a corporation-type investment trust company). We compare our estimates with the actual values disclosed by the RIET on the hotels, such as on availability, sales, and rates of return. If we can reduce the error margins, we can understand financial-statement information before it is actually disclosed. This is groundbreaking in two senses. First, we can isolate the conditions of financial statements before the hotels and REIT disclose this information; second, even though the information is high quality and could previously only be acquired by institutional investors, it now becomes open data that everyone can access and use.

How can we utilize the enormous data that has newly appeared in the financial field?

Similar to using Web data to estimate hotels’ availability, real-estate information on the Web can be applied to models in order to evaluate the capitalized value of commercial real estate, or to assess the credit risk of apartment loans, which is one of the leading products currently offered by banks. In actuality, estimates are currently being prepared experimentally, including those related to rental income from commercial real estate as a whole, such as rental condominiums and apartments in a region centered on Naha in Okinawa. Moreover, if, in the same way, we estimate the availability and sales of hotels not only in Kyoto, but in the whole of Japan, we will be able to discover a wider economic trend as one indicator of economic developments in Japan. Incidentally, management of the “Michibiki” four-satellite system of semi-zenith satellites that will pass above almost all of Japan will begin in 2018, and it is expected that it will be able to provide measurements accurate to 1cm. In the future, it will be possible to apply this sort of new, extremely detailed data for uses such as understanding conditions prior to disasters including earthquakes and landslides, and will make it possible to ascertain, to a high degree of accuracy, the extent of upheaval and subsidence of land. Moreover, I feel it will be highly possible to realize a system by which to simulate the economic loss from damage. Therefore, just as programs have replaced people to play the leading role in trading, it will become possible to conduct the analytical work of today, such as forecasts of company earnings that is currently carried out by human-wave tactics (e.g., analysts visiting companies to conduct interviews), completely automatically, from data collection through to estimates. Ultimately, I believe this is sure to result in improvements to services.

(Text in Japanese: Hiroshi Tsuda, Rue Ikeya. Photographs: Mitsuru Mizutani. Published: November 10, 2014)