Analysis: here's how a house's postcode, size, description, features and BER rating affects prices across the capital
In 1985, there was a furore over the creation of a Dublin 26 postcode. House owners in Dublin 6 claimed this would lead to property devaluation and the Dublin 6W postcode was created instead. The fascination with postcodes and the popularity of certain postcodes is nothing new in the capital. There is a long-standing theory that purchasers will pay more for a specific postcode and postcodes have been historically used as a wealth indicator and class status symbol. Perhaps this belief is still held by some individuals!
For those not familiar with the postcode landscape of Dublin, the city is is divided into 25 postcodes: D1 – D18, D20, D22, D24, D6W, North County Dublin, South County Dublin and West County Dublin. The Liffey splits the county into north and south, with even-numbered postcodes south of the river and odd-numbered postcodes to the north. The exceptions to this are Dublin 8 and Dublin 20 as they contain areas both north and south of the river. Nowadays, postcodes impact many areas of life including school catchment districts, hospital zones and local property tax.
The idea that properties are valued differently due to their postcode prompted our investigation into the impact of the postcode effect. To what degree does a postcode potentially impact a property’s price? Can using a more popular postcode, instead of your actual one, increase the resale value of your property?
We used data from January to November 2018 provided by 4PM Ltd to conduct our analysis. The dataset had 5,208 property transactions for the whole of Dublin over this time period, with sale prices obtained from the property price register and descriptions from the original property listing.
Property values are linked to property size, so we focused on price per square metre in order to compare values across the Dublin region. The expectation is that prices per square metre will be highest in areas with good employment, amenity or transport links and lowest in areas without these features. The map below shows shows the differences in price per square metre across Dublin. We can see that the higher values (dark red) are concentrated together around the city centre, to the south of the city and along the coast. Lower values (dark green) are observed as one moves away from the centre towards the county borders.
In order to isolate how much of postcode value is driven by perceived desire, we first need to establish the impact on value of attributes such as property type (apartment, house, duplex etc), energy rating, and individual house features (recently renovated, has parking or a big garden, etc.). There are a while host of local amenities such as schools, transport links and nearby parks that go into the value of a property.
We attempted to capture this through the use of a geospatial statistical model which enables us to estimate how much of the price per square metre is attributable to the GPS location of a property. After fitting our statistical models, we explore the remaining structure to see whether it can be linked to perceived affluence.
First of all, we show the property features used in our models with the vertical or y-axis providing the estimated effect of each, known as its relative scaling. The features shown are associated with the best fitting model. An estimated impact of 1 can be interpreted as the feature has no impact, lower than 1 indicates a reduction in value and greater than 1 an increase in value. For example, we estimate the premium for a renovated property to be between 3.2% and 5.3% over a non-renovated equivalent.
According to our model, properties that claim to have development potential for an extension or other works have an associated price premium between 2.6% and 5.7% over properties without. Interestingly, we see a premium for parking within 2km of O'Connell Bridge (Car Space.CC) in the order of 3.4%.
We also included property size in our model. When we initially analysed the data, we noticed that price did not always increase for increases in size. Price would increase to a certain point and then plateau. When we moved to a price per square metre setting, the plateauing effects were still present, but the relationship between price per square metre and size was negative. This makes sense as the bigger the property, the less it should cost per square metre. To allow for the plateauing, we allow a flexible relationship between price per square metre and size.
Here's the estimated relationships of price per square meter with property size. We see that smaller properties per square metre are more expensive than larger ones up to approximately 200 square metre. Property type has long been speculated to have an impact on the price, with different property types having different effects. Our statistical model allows us to examine the impact of property type on price per square metre while keeping all property attributes and features the same.
What about house description? Listing the property as a detached house (first point) has an estimated impact of 1.17, but this is 1.12 for a semi-detached property. As a result, the premium for a detached house over a semi-detached equivalent would be 4% (1 – 1.12/1.17).
On the other hand a duplex (the last point) is the least popular housing type. If there's a detached house and a duplex for sale in the same area - both 100 square metre, and with the same BER, number of bathrooms etc - and the detached house was valued at €250,000 (€2,500 per square metre), we'd expect the duplex to be €188,000 (€1,880 per square metre).
Can we estimate the uplift in property value by improving your energy rating? BER certificates were introduced in 2007 and indicate the energy performance of a house. They are needed in the event of the resale of residential housing, though some exemptions exist. We group the subcategories of the BERs together, i.e. 'A1' and ‘A2’ are grouped into ‘A’, ‘B1’, ‘B2’, and ‘B3’ are grouped into ‘B’, etc.. ‘A’ is the highest rating and is the most energy efficient, while ‘G’ is the lowest rating and is the least energy efficient.
In this graph, the x-axis shows the different ratings, with 'H' indicating exempt properties. Having an ‘A’ rated home has a relative scaling of 1.08 compared to the overall average effect of BER ratings. In contrast, a ‘G’ rating has an estimated impact of 0.94 compared to the average, reducing the price per square metre by 6%. For comparison, increasing the property from a ‘G’ to ‘A’ BER rating could increase the value by 13% (1-0.94/1.08). Though the results of the estimated impacts for property type and BER are not unexpected, by having these estimates we can quantify the value that they add when a homeowner is considering undertaking works. Keeping all attributes the same in our model, we also examine what can be associated with postcode structure.
Our first analysis of postcode effects used the postcodes that were listed on the property advertisement and these are show below. We have shortened some of the postcode names on the x axis, so NCD is North County Dublin, SCD is South County Dublin, WCD is West County Dublin, D1 is Dublin 1 and so on.
It is clear from this that the majority of southside postcodes (even numbers, shown in navy) have estimated impacts larger than 1 (above the dashed grey line), signifying postcodes add an additional premium per square metre over and above all other property attributes. Points symbolise our estimates for the different postcodes with the associated 95% confidence intervals. Points above the grey dashed line have a positive impact on price per square metre, points below have a negative impact on price per square metre and points on the line have no impact on price per square metre
But this does not tell the full story as our analysis showed some properties included the wrong postcode. Using GPS coordinates of the properties, we confirmed and corrected all the postcodes for the properties we studied. We found that 16% of properties studied changed postcode.
We attempted to quantify the changes that appeared most often. In the next graph, the text on the x-axis or horizontal shows the different postcode changes; 'D11.D9' can be interpreted as properties that were advertised as Dublin 11 but were actually in Dublin 9; ‘SCD.D24’ can be read as properties who were said to be in South County Dublin but are actually in Dublin 24, etc.. The bars around the dots appear to be longer than any of the previous plots – these are the 95% confidence intervals for our estimates. These larger intervals are due to the small number of transactions that fall into each category, so we are less certain of the true value.
Impacts larger than 1 (above the dotted line) show changes that have an additional premium associated with them. When we investigate the postcodes involved, this premium can be linked to claiming the property is in a neighbouring postcode with a higher median price. For example, our model predicts that if a property in Dublin 24, which correctly listed their postcode as D24, had a value of €250,000, the exact same property using the postcode South County Dublin would have an increased value between €253,000 and €293,000.
When we use our model to predict prices, 85% of the model's predictions are within 20% of the true sale price. One of the metrics we look at to determine whether a model is a good fit is R squared. R squared is a number that indicates how well the models fits the data, where a higher number generally indicates a better fit.
Our model has an R squared of 0.87 which means it accounts for 87% of the variability in prices. Importantly, we include the GPS coordinates of the property in our models, which allows us to include a value associated with the physical location of the property. We extend these values over the whole county, resulting in this surface map. We observe higher estimated values along the south coast and in areas with popular postcodes.
One potential application of this map is to assist in improving the current Local Property Tax calculations. The existing tax is based on the value of the residential property, presumably due to ease of calculation, which is a considerable disincentive towards the upgrading of a property.
One alternative would be to include location values from this map in calculating rates. The tax should include location values as well as dwelling values, rather than being based solely on the dwelling value. To the best of our knowledge, our models are the most complex valuation models that have been built for the Dublin region, enabling estimation of the contribution of individual house attributes and features to value a property.
Aoife Hurley is a PhD student at MACSI and the SFI Centre for Research Training in the Foundations of Data Science at the University of Limerick. Dr James Sweeney is a lecturer in the Department of Mathematics and Statistics and MACSI at the University of Limerick. He is a former Irish Research Council awardee
The views expressed here are those of the author and do not represent or reflect the views of RTÉ