Since office buildings vary in location, size, facility, rent and so on, it is difficult to illustrate their market with the average of such indices; segmentation according to those building specifications (specs) is needed. As an example of segmentation, some real estate-related companies grade office buildings according to their specs. Although their standards differ from each other, a few companies use the height and age of buildings. Conversely, attributes such as location, gross floor area, base floor area and building age are commonly used even though their values still differ for each company. Moreover, the number of grades differs from company to company, with there being two to four grades. These standards of grades have monotonicity, with wider floor area, newer buildings, etc. being better. In foreign countries, real estate-related companies also grade office buildings and make market reports on the basis of the grade. For example, Building Owners Management Association, which is an organization of office building owners in the United States, sets three grades of buildings, as A, B and C, in descending order and defines the overall characteristics of each grade. This grade differs from that of Japan in that rent is considered to be one of the building specs. Standards of the office building grade evaluation are subjective, sensory, and differ from company to company in that one company might deem a building to be A grade and the other might deem the same building to be B grade.
With the above discussion as the backdrop, we propose a quantitative grading method for office buildings. This method optimizes the thresholds of each spec of office buildings so as to maximize the variance ratio of contracted rents of targeted office buildings for each grade. We formulate this problem as parametric mixed integer programming and validate the method with the data of office buildings located in 23 Tokyo wards in 2013 and 2014. In the experiments, we tested 11 combinations of building specs.
The results of the experiments support the following conclusions. Compared with the existing typical office building grades published by some companies, the proposed method can grade office buildings that exhibit an increased variance ratio of rent. The best combination of building spaces derived by the proposed method comprises the following specifications: a location that is within the main five wards, building age, total floor area, base floor area and ceiling height. These specs differ slightly from those of the existing standards. The existing standards tend to classify buildings that have rents close to the average of different grades; overlaps of rents between different grades are often seen. Conversely, the proposed method sets the grade to classify buildings that have high rents in different grades.
Since the term of the data used for this experiment is only 2 years, we need longer-term data to validate the stability of the derived building specs. In addition, since we limited the combination of attributes to 11 due to the constraint of computational time, we should examine the combination of attributes exhaustively.
