By comparing the performance of the estimated actual housing price model and the modified housing price model where synthetic interest rates are included, the official bank rate’s effect on housing prices can be detected. Graphs displaying the housing price development of the actual and the modified housing price models are seen in figure 17. The graphs illustrating the respectable models compared to the true housing prices were presented in section 3.2.1, namely, figure 7, and figure 16 in section 3.2.5.
95
Figure 17: Development in Danish housing prices as predicted by the actual housing price model and the modified housing price model between the 1974Q4 and the 2017Q2 (Statistics Denmark and authors’ own creation).
There is a clear resemblance in the graphical interpretation of the two models. Both models follow the true housing price development relatively well, particularly in periods with low volatility in housing prices. Nonetheless, there are some significant deviations. The estimations of the two models are close to identical except from the change in two isolated parameters, namely the short- and long-term lending rates, these deviations can be associated exclusively with the indirect implementation of the Taylor rule. First and foremost, it can be drawn that the modified model appears more volatile with a higher frequency in shifts compared to the actual model. However, the range of the fluctuations is not as wide as seen in the actual model. This pattern is recognisable from the development in the Taylor rate and the synthetic lending rates observed in figure 12 and 14 in the empirical analysis.
However, the most substantial finding is that the housing prices explained by the modified model materialise on a consistently higher level than the actual model. This is directly imposed by the Taylor rule’s predication of a lower official bank rate over the estimation period. In analysing more thoroughly, there are identified some time periods of particular interest. As acknowledged in sub-sections 2.1.2 and 3.2.3, the second oil crisis from 1979 to 1982 caused great instability in the Danish economy, which led to a lower Taylor rate in comparison to the market observed official bank rate.
A lower Taylor rate predicts a less negative trend in housing prices in the modified model relative to the trend observed in the actual housing price model.
96 By observing figure 17, it is detected that the period from the 4th quarter of 1992 to the 3rd quarter of 1993 is a time span where the two models predict contradictive movements. While the housing prices in the actual model increase, the prices estimated by the modified model decrease. As identified in section 2.1.2, the general market conditions in Denmark from 1986 to 1993 were highly constrained with the Potato cure and the tax reform affecting households’ consumption. Particularly around the peak of this era in 1992 were there observed strikingly opposing tendencies between the estimated official bank rate and the Taylor rate, as outlined in section 3.2.2. The actual official bank rate was seen peaking in 3rd quarter of 1992 before it decreased in the following years. The decrease was also supported by the collapse of the ERM system in 1992, which contributed to a further reduction of the rate. For the housing prices in Denmark, a falling official bank rate induced lower lending rates and hence, an increase in housing prices due to an improvement in demand. The Taylor-approximated official bank rate, where the Danish market situation forms the basis for the level of the rate, did not predict this fall. The ERM breakdown affected Europe and ECB's monetary policy to a much larger extent than the Danish economy alone. Consequently, the modified housing price model does not experience a fall in synthetic interest rates and thus, not an increase in housing prices but a further decrease, as opposed to the actual model. This decrease lasted until the ceasing of the Potato cure in the second half of 1993. The increase in the Taylor-approximated official bank rate from the 1st quarter of 1993 was most likely motivated by the preceding negative rate, which encouraged the Danish economy in the right direction.
In the period between 1994 and 2008, the modified model seems to be the best model for explaining the true housing prices based on figure 7 and 16. The actual model has, unlike the modified one, some unusual spikes in 1998 and 2003. The actual official bank rate increased from the second half of 1997 until 2nd quarter of 1998, likely due to a response to the acceleration in real GDP growth in Europe (European Commission, 1998, p. 1). The Taylor rate predicted a more stable development around the same time. This might reflect the lower GDP growth in Denmark compared to the EU average. The EU average GDP growth was 3 % in 1998, while the Danish GDP growth was 2.2 % (“GDP growth (annual %)”, 2018). Nonetheless, the spike in housing prices detected in the actual model in 1998 is not evident in the true housing prices (see figure 7). Hence, the most likely explanation for the spike is of a technical matter. It was revealed in previous sections that a dummy eliminated the short-term effect of changes in bond yields and tax, as well as the imputed rate of property tax from the 4th quarter of 1998. The Cook’s distance test also captured this finding, where a significant Cook’s D value indicated that this 1998 observation was highly influential.
97 In the beginning of 2000, the IT-bubble burst, which caused a decrease in both the Taylor-approximated official bank rate and the actual rate until 2003. However, the Taylor rate was argued to have a less steep downfall than the actual rate. The reason for that is likely due to the fact that the burst of the IT-bubble caused larger damages in the euro area in general than in Denmark on an individual level. Following the IT-bubble burst, a period of a more stable actual official bank rate was observed. This was impacted by the introduction of a new ECB strategy as of 2003. Observations made of figure 14 show no signs of significant movements in the actual official bank rate in 2003, which make the spike in housing prices in 2003 noteworthy. This spike can partly be attributed to the introduction of interest-only loans in October 2003. Based on figure 14, the Taylor-approximated official bank rate predicted an increase from a severe downfall at this point in time. This might be an expression for the positive effect of the easing in relation to the cost of mortgage on the Danish market, which was not expressed in the constant ECB-directed rate at this point in time. The ECB rate was however not reduced as much as predicted by the Taylor rate prior to this event.
Nevertheless, the upward trend in the estimated Taylor rate indicates a lower prediction of housing prices. When evaluating figure 16 and 17, it can be concluded that both the true housing prices and the actual model represent higher housing price levels. The spike as of the 4th quarter of 2003 is claimed to be unobservable in the modified housing price model's estimation results as the interest rate and mortgage effect were adjusted for when using the Danish market as a base. There is, however, some speculation associated with the extreme reaction in the actual housing price model compared to the market observed development.
It can be speculated whether the aforementioned spikes are reflections of the estimated actual housing price model’s tendency to emphasise interest rate’s impact on housing prices to a larger extent than what appears evident in the true market prices and the modified model. Seemingly, there are some trends the actual housing price model does not take into account, as it does not manage to follow all movements in market observed housing prices. For some events, such as the highlighted spikes in 1998 and 2003, explanations are improved in the modified model, despite this model’s lower R2 relative to the actual model.
In conclusion, it can be argued that the outcome of the actual housing price model and the modified housing price model describe more or less the same development, despite certain deviations. Both models provide significant coefficient estimates, except from the long-term variable in the modified model, which is close to reaching the 10 % significance level. Nevertheless, as the models with and without the impact of a Taylor rate do not differ substantially from each other, it can be argued that
98 the official bank rate’s influence on housing price determinations is somewhat limited. It is outlined throughout the analysis that there is dissimilarity between the actual official bank rate and the Taylor-predicted official bank rate, particularly in the years prior to 1999. Hence, a more extensive deviation in the housing price models would be expected if the official bank rate’s impact were more prominent than what is concluded in this thesis. This underlines the fact that other factors work highly determinative in housing price estimations and that interest rates alone are not necessarily decisive for the outcome, at least according to the chosen housing price model. However, both models seem to follow the market observed housing prices improving after 1999, which is in line with the Taylor rates' fit to the official bank rate displayed in figure 14. Additionally, the Taylor rate based on estimated coefficients also exhibits statistical significant estimation output from 1992 and throughout the estimation period. The estimated Taylor rate's graphical fit to data supports this conclusion, especially in the time period from 1999 to 2009. After 2014, the HP filtering effect is to some extent the factor to blame for the divergence between the estimated Taylor rate and the actual official bank rate (see figure 12). As explained in section 2.2.5, the latest economic fluctuations in the estimation period may be granted too large weight compared to earlier ones, which impacts the value of the output gap and hence the Taylor rate.
Finally, it is highly important to recognise that some of the effects of a predicted change in the official bank rate on housing prices may be compromised by the static transformation from official bank rate to lending rates. This pinpoints only one out of several uncertainties connected to the estimation process.