To achieve carbon neutrality, expanding renewable energy power generation, such as solar and wind power, is being promoted everywhere. However, since the output of these renewable energy sources fluctuates greatly depending on weather conditions and other factors, it is difficult to balance abundant electricity supply and demand. As a countermeasure, demand response, which controls power consumption by consumers, is attracting attention. Among consumer equipment, building multi-type air-conditioners is a strong candidate for DR because of their large power consumption and high penetration in small and medium-sized office buildings. Customers who participate in DR receive incentives based on the amount of power curtailment (ΔkW). The limitation is how to measure the amount of electricity suppression. It is necessary to perform baseline estimation to measure the amount of power curtailment. One method for baseline estimation is the High 4 of 5 method, which averages demand data over the past four days. More mathematical methods using decision trees and neural networks have been studied. These nonparametric methods can construct an estimation model even when information on the target building is unavailable. In contrast, if operational information such as architectural drawings and occupants is available, the estimated model can be applied to another operation or case study for DR aggregation by changing the numerical values of certain parameters. Nonparametric estimation methods make reusing the estimated model by changing certain parameters difficult. Therefore, we described a baseline estimation method using a parametric mathematical model of building multi-type air-conditioners. Each parameter of the mathematical model, which simulates the minute unit power and room temperature, is fitted to match the power and room temperature of the building to be estimated. The true baseline after DR is implemented in an office building is impossible. Therefore, we constructed a fictitious five-story office building on a computer and fitted each parameter of the numerical model based on the normal operation data of this fictitious building. We showed the relationship between the time granularity of DR and the baseline estimation error using the constructed mathematical model.