Early Warning Indicator Model of Financial Developments Using an Ordered Logit – Defining Boom Phases

The definition of asset price booms overtakes the proposal of Gerdesmeier, Reimers and Roffia (2011). In the literature, several approaches to identify asset price booms have been used. For instance, Borio and Drehmann (2009) define a boom as a period in which the three-year moving average of the annual growth rate of asset prices is greater than the average growth rate (i.e. its mean) plus a multiple (1.3 in this specific case) of the standard deviation of the growth rates. By contrast, Alessi and Detken (2009) follow a different approach. In essence, they calculate the trend of the price variable using the one-sided Hodrick Proscott filter and then derive the gap between the actual values of the price variable and its trend measure. If the gap is greater than 1.75 times of the recursively determined standard deviation a boom will be identified. With respect to such a procedure, Detken, Gerdesmeier and Roffia (2010) note that these methods rest on some critical assumptions. First, there is an implicit acknowledgement that it is difficult to derive equilibrium asset prices with reference to the respective underlying fundamental variables. Second, the method relies on the use of a time-varying trend as a proxy for those underlying fundamentals. Third, significant deviations from the trend are then considered excessive and expected to be reversed at some point in future. Following these assumptions, a boom occurs when the “composite” asset market indicator development is greater than a pre-defined threshold.5 In this study, the trend is calculated by making use of the Christiano-Fitzgerald filter (2003), since the Hodrick-Prescott filter is well-known to suffer from an end-of-sample problem.
The emergence of a boom (i.e. a value of 1 of the “boom dummy” variable) is defined as a situation in which the gap between the actual composite indicator and the indicator’s trend has been greater than its mean (G) plus a factor 8 (equal to 1.75 and fixed across the sample period) multiplied by the standard deviation of the same indicator (sG), which are calculated over a rolling period of 60 quarters: Dumbot =1 iff Gt □(G |s +8sg |s), where s = t-60 for t > 60 or 1 for 0 < t < 60. 6 Putting both dummy definitions (2) and (4) together we determine the following ordered variable.
Figure 1 summarizes the behaviour of the changes of the composite indicator an of the yt of each country. As far as the euro area is concerned, only two booms can be detected, the first one from 1988 Q4 to 1991 Q4, which may be connected to the introduction of the common market and the second one from 2006 Q2 to 2007 Q4. It seems, however, that, at the aggregate level, developments in some countries are counterbalanced by movements in other regions of the euro area. Second; the overall number of booms seems to vary across countries. Third, the length of the booms also varies across countries, lasting from a few quarters up to, broadly speaking, two-three years. In addition, when looking at individual country’s experience, a few interesting issues also arise when considering the most recent developments in relation to the driving factors. For instance, it is interesting to note that in countries like Spain, Ireland and the United States booms in the composite asset price market are detected around 2006-2007, which is a period when all three countries experienced a strong house price boom which had started some years earlier. However, it is important to take into account that developments in the composite asset price indicator are also influenced by developments in stock prices, which can in principle counterbalance house prices developments over some periods of time. Therefore, opposite developments in the two individual asset markets may lead to a lack of a signal of a boom. This is, for instance, the case for Germany where no boom can be detected in the most recent episode, given the subdued developments in house prices.
The summarized exhibitions are given in Figure 2. Panel A gives the number of boom periods. The following observations seem worth noting. First, booms seem to be concentrated around three main periods. The first period is in the 1970s before the first oil price shock, the second period includes the end of the 1980s and the beginning of the 1990s, following the oil price trough in 1986, while the last cluster is around 2006-2007. Taken together, these observations lead to the conclusion that an analysis taking into account heterogeneities across countries and time has to be adopted. Panel B gives the number of bust periods. The busts seem to be less concentrated than booms and the number of busts periods (230) is lower than the number of booms (293). Moreover, they are shorter. There are two clusters. One exists after the first oil price shock. The second is around 2008-2009. Moreover, lot of busts of the second cluster directly succeeds a boom. (see also Figure 1).

Figure-1Figure-1Figure-1

Figure 1. Developments in the changes of the composite indicators and boom-normal-bust-periods in the main OECD countries and the euro area

Figure-2

Figure 2. Panel A gives the number of booms in each period; Panel B gives the number of busts in each period.