7Conclusion

Under consideration of the entire life cycle of real estate, a substantial potential of cost savings lies in the costs incurred during operation and occupancy. For a holistic assessment of the impact of decisions made during the planning process of construction, renovation, and modernisation measures, it is a crucial task to determine operating costs as early as possible. Therefore, the current investigation intends to reveal and describe the causal interrelationships between operating costs and a variety of variables with potential influence on these costs. The objective is the provision of an essential basis of information, statistical models, and adequate cost indicators for an accurate determination of operating costs for the practical application in the field of cost planning of real estate. The results aim to provide the foundation for a holistic assessment of planning alternatives and to support decision making and budgeting. The study is directed towards architects, planners, and the real estate management.

The quantitative approach employs the empirical data of a variety of operated facilities as a basis for the presented investigation. The data sample was collected as primary and secondary data in cooperation with a variety of project partners. On the foundation of an extensive review of relevant research studies and publications, a wide range of factors with potential influence on operating costs are defined as key variables and multiple methods for the statistical analysis are selected. Thereafter, regression, artificial neural network, and classification tree models are developed for the estimation of cost data of 15 cost groups according to the structure of the standard DIN 18960:2008-02. The main outcome can be summarised as the determination of adequate reference quantities, the identification of significant influential variables, and the introduction of categorised cost indicators. Furthermore, the most accurate operating cost estimation method is determined by an evaluation and comparison of the developed statistical models and cost indicators. Finally, the practical application of the introduced models is demonstrated in a detailed implementation example.

The main findings of the statistical analysis indicate a significant interrelationship between the utilisation of the facilities and their operating costs. For nearly all cost groups, the developed models reveal the largest effect on the variance of costs for the type of facility. The variable gives information on the utilisation of the facilities and differentiates the sample into the characteristics care retirement home, church facility, fire department, kindergarten, library, municipal facility, research/teaching facility, residential facility, school facility, sport facility, and town hall. Further significance can be observed for multiple predictor variables describing the condition of the facilities. Likewise, effect is indicated for variables giving information about the standards of the construction, the technical installations, and the outdoor facilities. The significant variables of both condition and standard are in general directly related to the type of costs under investigation. For example, effect on the costs of heat supply is detected for the condition and standard of the heating system, the heat distribution system, and the envelope of the building. Furthermore, the models show significant influence of various variables of the variable groups of quantities, location, and management strategy.

A comparison of the estimation performance of the developed statistical models and introduced cost indicators reveals for the analysed cost groups error rates of between 15.3% and 70.1% for the respective method with the most accurate cost estimation. For 8 of the total 15 cost groups, regression models with transformed variables for a correction of non-normality of the data distribution provide the highest cost estimation accuracy. Binary classification tree models can offer the best estimation performance for 4 of the analysed cost groups. This includes the cost-intensive cost groups of heating energy (about 23% of the operating costs), cleaning and care of buildings (about 35 %of the operating costs), and operation, inspection, and maintenance (about 14% of the operating costs). The cost estimation with median values of categorised cost indicators shows the lowest error rate and best performance for 3 costs groups. The cost estimation with the developed artificial neural network models can not provide the highest performance or lowest error rate for any of the analysed cost groups. Finally, a combination of the most accurate methods on the most detailed level of the standard DIN 18960:2008-02 can achieve an error rate of under 13 %for the estimation of the total operating costs.

Though the results of the current study are validated by multiple statistical measures and the general validity of the estimation methods is assessed by an independent test sample, the presented empirical investigation is subject to general limitations and restrictions. Therefore, it is essential to carry out a critical discussion to ensure an appropriate application and implementation of the results. As outlined in the description of the underlying data, various limitations and restrictions result from the consistency and representativeness of the sample. The cost data and further information was obtained as primary and secondary data from multiple project partners. Since the respective project partners selected the facilities that are included as observations in the investigation, the data sample has only restricted randomness in terms of statistics. It is nevertheless assumed, that the selected facilities represent the real estate portfolio of the respective project partners. The participating project partners are mainly public sector institutions as for example municipalities, universities, ecclesiastical administrations, social housing administrations, and social associations. Therefore, the results are generally only representative for facilities owned and operated by the public sector and only restricted inferences about private sector facilities can be drawn.

Further limitations and restrictions are expected in regard to the location of the analysed facilities. The data was primarily collected in the southwest area of Germany and a large amount of facilities included in the investigation is located in the city of Stuttgart. Therefore, the consideration of regional economic and climatic conditions is only available on a restricted level. Though the regional economic and climatic conditions can be taken into account by application of local indices, only restricted inferences can be drawn about facilities in other locations. Another restriction is expected for the consideration of management strategies as for example service level agreements and outsourced facility services. Since management concepts are only available as the outsourcing rate of cleaning services for the observations provided by the participating project partners, the analysis of the interrelationships between management strategies and operating costs is restricted.

The scope of costs and the cost types under investigation cause limitations and result in a restricted applicability as described in detail in the definition of the key variables. The analysed costs are classified according to the cost structure of the standard DIN 18960:2008-02. In order to implement the presented statistical models and cost indicators, in particular the results of the aggregated first and second level cost groups require a detailed consideration of their scope when practically applied. Further limitations and restrictions arise from the scope of the collected cost data. The underlying costs data are based on cash flows covering a maximum lifespan of five years. Irregular costs in particular may vary significantly across the different stages of the life cycle and can be directly related to the service life of individual components of the facilities. Though the scope of included facilities covers a variety of life cycle stages, the representativeness of costs incurred irregularly as for example inspection and maintenance costs may be restricted. Finally, the limited amount of observations with individual characteristics of variables causes restrictions. For a limited number of observations, the presence of outliers or errors in the underlying data may distort the outcome substantially. The negative impact of a limited data sample is significantly reduced by a detailed pre-analysis of the data sample.

The current investigation reveals substantial findings and the detailed presentation of model development can serve as a basis for future research approaches for operating cost planning. In order to increase the data foundation, future approaches may intend to include data of facilities owned and operated by the private sector. Therefore, the interrelationships between the operating costs and the management strategies can be examined in detail and a comparison of differences between private sector facilities and public sector facilities can be conducted. Likewise, the economic and climatic effect of the location may be analysed in detail in future research approaches by consideration of an extended data basis on a national or international level. An extension of the data sample with focus on characteristics with a limited number of observations may improve the results of a future investigation. Likewise, future research may consider additional types of facilities as for example industrial facilities, health service facilities, or laboratories for an investigation of operating costs. Furthermore, the scope of analysed costs may be extended and for example the expenditures for the repair of facilities or the real estate management may be included in future approaches. Though a relatively good fit to the underlying data is indicated for the developed statistical models, there remains a certain amount of unexplained variance of the analysed operating costs. Future research approaches may rely on a more detailed data basis and the cost data may cover a longer time span in the life cycle of the analysed facilities.

The evaluation of the applied statistical methods reveals a stable estimation performance for the regression models and for the non-linear regression models with transformed variables in particular. The transformations are conducted in order to compensate non-normality of the variables and result in a consistently higher estimation performance of the regression models. Since the estimation with artificial neural network models can not provide the highest performance and lowest error rate for any of the analysed cost groups, it is assumed that non-linear interrelationships between operating costs and influential variables are not significantly distinct for the underlying data. The classification tree models reveal in contrast a relatively good estimation performance for the most cost-intensive cost groups in particular. Therefore, future research may focus on the development of both regression and classification tree models. The ultimate objective of future research approaches should be the development of standardised cost models and applicable tools for the holistic determination of life cycle costs of real estate including the periods of planning, construction, operation, maintenance, repair, and reuse or demolition.