This thesis examines the regional house price patterns of the Norwegian housing market. Using a datasetwith monthly observations for all 19 counties, 10 urban areas and the aggregated index the analysis is limitedto the private housing market. This means that the analysis includes single estates, shared housing andapartments. The main objective is to analyze whether there are differences in the housing market that couldreflect the Oil industry employment pattern of Norway. Certain Norwegian regions have largeconcentrations of employment within the oil industry, while other regions are almost unexposed to thisindustry and rely more on export of services and goods. Additionally, as the Brent oil price has decreaseddramatically ever since 2014Q3 this paper hypothesizes that the housing markets of the more oil intensivecounties have been affected by the downturn.The regional analysis shows that the real house prices indexes of the western region have had anextraordinary development from 2003 to 2016. Stavanger and Sandnes, which have been in the center of theso-called Oil Adventure for 40 years, have had booming housing markets with returns no other urban areacan compete with. However, as the Brent oil price started to decrease in 2014 so did the house price indexof the Stavanger and Rogaland region. Three years of decreasing house prices in Rogaland could besuggestive of the western region not maintaining its extraordinary position. Conversely, the analysis findsthat all other counties and urban areas of the western region seem less affected by the structural changecompared to the Rogaland area. In addition, the indexes of Rogaland decreased for more than one yearbefore the Brent oil price decrease started. Interestingly, the house price growth has during the same periodaccelerated in parts of the northern region.Using stationarity tests such as the Augmented Dickey Fuller and Phillips-Perron tests evidence of unit rootsare found for all indexes. Detrending methods such as the Hodrick-Prescott filter and the first differencingapproach are applied to obtain stationary data series. The Durbin-Watson test is used to reject the hypothesisof autocorrelation for the differenced data. Based on the stationary indexes Autoregressive IntegratedMoving Average (ARIMA) models are individually fitted for each of the 19 counties and 10 urban areasbased on thorough diagnostics checking. The optimal number of Autoregressive lagged monthly valuesrange from 12 to 14. Using the optimized ARIMA models and HP filter based models forecasts are created to predict the expectedpath of all the respective indexes from 2016 to 2018. Even though the Norwegian Central Bank is knownfor its forecasts of the aggregated housing index, this is the first known attempt to forecast the regionaldevelopment in Norway based on county indexes. Both forecasting techniques are suggestive of thedownturn in Rogaland continuing throughout 2016. The two models do not, however, agree on the directionof the indexes of this county in 2017. Neither do the two methods agree with respect to the predicteddevelopment for the southern region from 2016 to 2018.The second objective of this thesis is to test for so-called ripple effects from the largest urban areas in thewestern and eastern region. The ARIMA forecast, which could be preferred over the HP method due to itslow Root Mean Squared Error, Mean Absolute Error and Bayes Information Criterion, was suggestive ofthe suburban areas around Oslo and Stavanger growing at lower rates in the future compared to theirhistorical yearly averages. An attempt is made using the Granger causality test to detect whether there areurban areas that seem to act as centers for house price changes affecting the rest of the country throughrippled patterns. Using this test, which is limited to the 10 urban areas of the sample with 3 different lagsettings, does not lead to sufficient evidence of ripple effects spreading from Oslo or Stavanger throughoutthe country. Evidence is, however, found of predictability among these large urban areas at the differentlagged intervals. This is suggestive of both Stavanger and Oslo having useful information for predicting thehouse price development of other regions and each other.At the longest lagged intervals, the standard Granger causality F-statistics is suggestive of only a few urbanareas having useful information in predicting other areas, whereas the heteroscedasticity robust statistics issuggestive of predictability in almost all possible directions. The latter result extends former research onripple effects in the Norwegian housing market. The first result, however, is more in line with the economicreasoning and expectations of this paper. Therefore, due to partially contradictory test results in theheteroscedasticity testing part of the paper, a plausible explanation is that the heteroscedasticity hypothesisassumed by the robust estimators could be rejected.
|Educations||MSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||92|