Housing prices in many countries have increased significantly over the past years, fueling a fear that speculative price bubbles will return. However, it can be difficult for policymakers to recognize when regulatory interventions in the market are necessary to counteract bubbles. This report shows how modern machine learning methods can be used to forecast speculative price bubbles at an early stage. Early warning models show that many OECD countries have an increased risk of speculative bubbles. In Germany, there are explosive price developments that have decoupled from real estate earnings. However, the forecast model indicates that the risk will decrease somewhat over the coming months at a high level. Unfortunately, the preventative measures in Germany remain insufficient. For example, there is a lack of intervention options involving household debt ceilings, and it is unclear when the Federal Financial Supervisory Authority (BaFin) can begin intervening in the market.
Keywords: Early warning system, speculative housing price bubble, panel logit,decision tree, random forest, support vector machine