The present study extends the analysis of Gerdesmeier, Reimers and Roffia (2010, 2011) which focused on predicting asset price busts and booms using different probit models, respectively. This analysis determines one model to predict the development of a financial indicator which includes booms and busts as well as normal periods. This necessitates, as a first step, precise definition of booms and busts. Given that this study’s focus is on deriving a combined signal derived from several individual asset markets, a composite indicator combining stock and house price development is used. This is in line with the IMF (2010) which stresses the importance of the real estate market and the stock market to describe asset markets.

Once the respective boom and busts periods are selected, as a second step, we try to explain them by use of leading indicators represented by various financial, monetary and real indicators. This is also done by Gerdesmeier et al. (2010, 2011), Herwartz & Kholodilin (2011) or Dreger & Kholodilin (2011). As for the financial variables, we consider historical series of the short-term (three-month money market) and long-term (ten-year government bond yield) interest rates and their spreads. The short-term reflect the policy of the monetary authority. Relatively high rates show a restrictive policy and should reduce the probability of a boom and increase the probability of a bust. The long-term rate reflects the capital market. High rates indicate a high profitability of capital, which boost a boom in equity markets. The term spread, defined as difference between the long-term and the short-term rate may show the policy stance of the monetary policy. In the literature it has been shown that the term spread is a good measure to predict future output growth (see, for instance, Estrella and Hardouvelis 1991). Monetary indicators comprise broad money and credit to the private sector (or loans to the private sector whenever available). For example Borio et al. (1994) show that asset price booms are fuelled by money growth as well as credit growth. As for the real indicators, we consider real GDP growth and the investment-to-GDP ratio as well as the inflation rate measured by the growth rate of the consumer price index (CPI). High real GDP growth goes in line with high profits of firms and increase of stock prices. Moreover, economic agents obtain higher income and spend more money in the house markets. Furthermore, high output growth rates make companies optimistic and stimulate them to invest more. This positively affects the stock prices (see Gerdesmeier et al. 2011). In sum, the dataset contains quarterly data1 for 17 main industrial OECD countries (additionally, the euro area as a whole is included in the descriptive analysis) for the period 1969 Q1 — 2011 Q2.2

Our definition of busts follows the work of Gerdesmeier, Reimers and Roffia (2010). It relies on the methodologies developed by Berg and Pattillo (1999) and Andreou et al. (2007). A bust occurs when the “composite” asset market indicator declines by more than a pre-defined threshold.3 In line with this, a composite asset price indicator has been calculated by combining the stock price index with the house price index as follows: DC = fDStock prices + f2 DHouse prices (1)

where f is normalised to 1 and f2 = sDSP /sDHP (that is the ratio of the standard deviation of the two variables). The weight is calculated recursively throughout the sample period in order to take into account the information available up to each moment in time.

A bust is then defined on the basis of this composite indicator, and, generally speaking, it would be denoted as a situation in which this indicator declines by a certain amount at the end of a certain period with respect to its peak (see Andreou et al., 2007). In our case, we will denote as occurrence of a bust (i.e. a value of 1 of the “bust dummy” variable) a situation in which at the end period the composite indicator has declined by more than its mean (denoted as C) minus a factor (in our study 8 =1.5 is chosen as fixed across the sample period) multiplied by the standard deviation of the same indicator ( sC ) in the period from 1 to (t + r) with respect to its maximum reached in the same period, i.e.:

Dumbut = 1 iff DCt < (DC |1 ~8sdc lt) (2)

where C is the composite indicator (already expressed in terms annual rate of changes), DC = mean(DC) and 8 = 1.5 . In the empirical application there are busts periods are underbroken by one or two no busts periods. In such cases we set these periods to busts periods to get a less volatile dummy behavior.