Developing an Early Warning System for the Real Estate Market in Bulgaria
Developing an Early Warning Framework for Detecting of Crisis in Economic Sectors in Bulgaria: Based on Korean Economic Development Experience. Korean Development Institute, 1, MOEF Republic of Korea, 2020, ISBN:979-11-5832-567-0, 77-100-109-118
Posted: 14 Apr 2022
Date Written: 2020
This study develops an Early Warning Framework for the real estate sector of Bulgaria. We introduce the Early Warning System used in the real estate market of Korea and apply the methods to Bulgaria to develop an EWS for its real estate market. This report is structured as follows. Section 1 introduces the real estate EWS in Korea and Bulgaria. Section 2 overviews real estate markets of Bulgaria. Section 3 introduces the real estate EWS of Korea. Section 4 develops a real estate EWS for Bulgaria. Lastly section 5 concludes the chapter and discuss policy recommendations.
The real estate EWS model for Bulgaria is built based on the Korean model. However, there are similarities and differences between the Korean model and the Bulgarian model. We first summarize the similarities between them. First, we consider various sectors of real estate market. Focusing only on the housing price index does not adequately reflect the real estate market situation and may miss important potential crisis factors. To avoid this potential fallacy, we examine various sectors of real estate market and construct sectoral crisis indicators which will be combined into a composite EWS index. Second, a crisis is defined as the situation where sectoral indicators deviate from their normal status. Third, an early warning system alarms crisis signals six months ahead of future crisis. Fourth, an early warning report is systemized. Thus, once raw data are typed into the database, a final summary report is prepared.
Although the Bulgarian real estate EWS model shares the most important principles of the EWS model with Korean models, the real estate EWS for Bulgaria is different from that of Korea in many ways. First, the Korean model uses fixed thresholds to determine crisis levels while the Bulgarian model uses time-varying thresholds reflecting fast changing real estate market situations. Second, the Korean model employs expert surveys to determine relative weights of different sectors. In Bulgaria, we use the principal component analysis to find the relative weights of each sector. Third, the Korean model uses the exponential smoothing method to forecast the future crisis index. In the Bulgarian model, we use the ordered probit model to find the forecast of the crisis index. Fourth, the Korean model provides both national as well as regional EWS indices. In Bulgaria, due to data availability, we focus only on the national model. Fifth, the Korean model uses monthly data while the Bulgarian model uses quarterly data. Sixth, the Korean model determines crisis levels across five levels while the Bulgarian model uses three levels (normal/warning/crisis).
Developing an early warning framework for Bulgaria consists of various steps. First, we examine the Bulgarian real estate market from early 2000 to 2019 and identify various crisis situations throughout history. As the data show, several crises occurred during the period considered in this report. These crises will be used as benchmarks to define a crisis in the following steps.
Second, we collect data sets that will be used to define sectoral crisis indices. We consider three sectors (residential, business, and land), and a macro & finance factor. In each sector and a factor, we select variables that might help predict a crisis in the real estate market. We then transform the variables so that they provide signals for a crisis following the benchmark crisis discussed in the first step. For example, we use original values, take moving averages, or calculate the growth rates of the variables.
Third, we define crisis as the situation where each variable deviates from the normal status (or long-run trend). We define normal status as the three-year moving average of the transformed variables. Thus, we implicitly assume that economic structures slowly change over three years. Then, we define bands up and down of the slowly changing long-run trend with the bandwidth 1.0 and 1.5 standard deviations (SD) of the transformed variables. These bandwidths are chosen by trial and error to best represent the benchmark crises. A crisis is defined both upward and downward from the long-run trend because sudden and large drops as well as rises of variables are regarded as crisis situations in the real estate market. If values cross the 1.0 SD threshold, a warning signal is issued and if values cross the 1.5 SD threshold, crisis signal is issued.
Fourth, with the normal/warning/crisis signals defined for each variable, we construct a sectoral crisis index. In each sector, we have many variables and we need to combine the crisis signals from these variables in some way. Under the circumstance that there are no relative weights or exact dates of crisis periods in the real estate market, we need to use a way to extract common signals from individual variables and define a sectoral crisis index with common signals. For this purpose, we may use the principal component analysis (PCA). It is a technique often used in exploratory data analysis to find the best common factor from the given variables. If we apply the PCA to our problem, we can find a common factor that is a linear combination of variables in the sector with the relative weights given by the PCA. The common factor is known to best represent the market behavior of the sector. By rounding the common factor, we can construct sectoral crisis levels.
Fifth, we can forecast the sectoral crisis levels using the sectoral variables with a forecast horizon of six months. For this purpose, we use the ordered probit model, which is popularly used when the dependent variable is an ordered discrete data. The fitted values from this regression become the sectoral EWS index and level. For the sectoral EWS levels, we apply the PCA again to construct the composite EWS index and level.
Sixth, the early warning framework is completed by providing quarterly early warning reports to policymakers. We make the system as automatic as possible so that once the raw data is typed in the excel file, simply running the set of codes will provide the summary report. From the report, policymakers can identify the risk factors in each sector and carry out appropriate policy measures to solve the crisis situation. The role of the EWS is to provide alarms for potential future crises. Policymakers should prepare policy measures in each level of the EWS index.
Seventh, after operating early warning frameworks several months or over a year, we need to upgrade the EWS model because one model does not perfectly fit the data once and for all. Thus, we should evaluate the performances of the early warning framework and drop unnecessary variables, include new variables, and modify equations so that the new model better forecasts potential future crises.
Suggested Citation: Suggested Citation