Using 5.8 million to buy a unit in Oslo, which one is worth?

Housing Price MCDA Model - a perspective from spatial contributions

The on-sale units of Oslo from Finn in Feb 2022. N = 950, Mean = 7.56M kr, Median = 5.77M kr. Which one is a good choice? it's up to your own preferences!

There are so many factors contributing to house prices. At the same location, the price varies by building’s attributes, e.g., the cost of construction, design, decoration, and furnishings. If a pair of twin buildings exist, which are at different locations, the surrounding environment will decide price. In this report, we defined surroundings as spatial contributions or spatial `scores.

Some like to live in the city center, enjoying night pubs, and shopping malls, whereas others prefer being near the forest. So, the unit’s price or value varies from different perspectives. Multi-criterion decision analysis (MCDA) is to support decision-makers in solving such problems (Meng et al., 2011; “Multiple-Criteria Decision Analysis,” 2022).**

This report proposed a Housing Price MCDA Model (HPMM) to value the spatial contributions by inputting user preferences, based on spatial information from open access databases (Norwegian Public Roads Administration, Statistics Norway), OpenStreetMap, and satellite images. The model is aimed to provide best-fitting options for house **buyers and **assessing the living** conditions in various areas of Oslo for better urban planning regulation**. In the end, I discussed the weakness of the model, and the several shopping tips revealed by the model.

This is a demonstration model that could be extended to other cities easily. The post-processed datasets in this project are good materials for automating GIS or WebGIS training courses. However, I had no time to make this model online due to limited time.

How it works

HPMM: Quantifying your living condition/city service (infrastructure) by `spatial score` and help to make decisions.

HPMM obtains (1) spatial information from statistics Norway, OpenStreetMap and optical satellite imagery, (2) user preferences, including point of interest (POI) and weights for all criteria. Once the model is set up, the on-sale units of Oslo in February from Finn.no are inputted to model. Then, the recommended units are output by model.

Oslo County polygon was resampled into a 50x50 meters grid first. All spatial information was accordingly assigned into each cell by location. The spatial information used in this scenario includes public transportation, kindergarten, school, stores, groceries, culture and sports building, parking places, vegetation index, and noise level. The spatial score is contributed by all these elements i on different distance a/b/c/d… and weights under certain reclassifying rules f(x).

Workflows.

The main tools used in this project:

The parameters used in the demo model: Table 1. The criteria and the model parameters (demo).

Code Criteria Type Spatial points by function Reclassifying Weight
1 Bus or metro Stop Numerical 1.5 pt: <250 m 1 pt: <500 m 0 pt: >=500 m 10 pt : >=24 pt 6 pt : >=12 pt 3pt : >=6 pt 1 pt : >=3 pt 0 pt : <3 pt 15
2 Parking place Boolean 10 pt: =<100 m 5 pt: =<250 m 0 pt: >=250 m - 5
3 Kindergarten Boolean 10 pt: <500 m 0 pt: >=500 m - 10
4 School Boolean 10 pt: <500 m 0 pt: >=500 m - 5
5 Supermarket Boolean 10 pt: <250 m 5 pt: <500 m 0 pt: >=500 m - 10
6 Café, Bakery, Bar, Restaurant, Pharmacy, Fast food, Convenience store, Beverage Numerical 1 pt: <500 m 0 pt: >=500 m 10 pt : >=50 pt 8 pt : >=25 pt 6pt : >=8 pt 4 pt : >=4 pt 0 pt : <4 pt 5
7 Library, Museum, Theatre, Stadium, Sports center, Cinema, Playground, Mall, Swimming pool Numerical 1 pt: <500 m 0 pt: >=500 m 10 pt : >=15 pt 6 pt : >=8 pt 3pt : >=3 pt 1 pt : >=1 pt 0 pt : <1 pt 5
8 College, University, Hospital Numerical 1 pt: <1000 m 0 pt: >=1000 m 10 pt : >=7 pt 5 pt: >=4 pt 1 pt: >=1 pt 0 pt : <1 pt 5
9 NDVI Numerical Average of 50 m 10 pt: 70% 6 pt: 50% 4 pt: 30% 0 pt: below 30% 20
10 Noise Numerical Average of 50 m 10 pt: <= 55db 2 pt: <= 60db 1 pt: <= 65db 0 pt: >65db 20

Some results

Taking public transportation or driving? Or half-half.

(left) The middle of the city center and the east side of Sankt Hanshaugen, are too far from T-bane, and have a relatively low density of road network due to terrain. Besides, the big park, University of Oslo, Ulleval Hospital, Aker Hospital have a negative impact on public transportation access. (right) If you prefer driving, Mjorstuen, Uranienborg, and Fagerborg are not easy parking area.

Shops and supermarkets.

(left) In this report, the shop's layer contains a café, bakery, bar, restaurant, pharmacy, fast food, convenience store, and beverage. The hot zone for hanging out is between the Colosseum in Majorstuen, Frogner (west), Gamle Oslo (east), and Torshov (north). And Nydalen is a good choice as well. Vestre Aker and Grorud have several choices relative to other suburban areas. (right) There is no supermarket for the people living in the north of Frognerseterveien in at least 500m. Another supermarket blockhole is the T-bane station Smestad, which is located at a crossroad of Road 168 and 150, with the nearest supermarket 1.5 km in the east, 1.0 km in the south, 1.3 km in the north, and 0.8km in the west

When we use NDVI as a criterion, there is an obvious underestimation for units located near the water body, e.g., shoreline or lake. For example, in Huk, the better view to fjord, the more expensive the house would be. However, the beach does not have any vegetation and does not contribute to the spatial score. This issue could be fixed by adding another criterion.

Is your place greener than the average?

The average noise level below 55db over 24 hours would be thought harmless, 10 points, otherwise can only get 2 points (<60db), 1 point (<65db), 0 point (>65db). The noise exposure is the result from modeling, not indoor noise exposure, resulting in uncertainty about the final results. Since normally the noise model considers vegetation as an important parameter, the area close to parks and away from highways and railways get excellent scores from both two criteria.

The noise is mainly produced by traffic on roads and railways.

What it suggests

We put units from Finn on the spatial score map, the best-fit units are marked with cyan dots
Clearly, there is no correlation between price and spatial score under my preferences. The best-fit units are marked with cyan dots.

Aggregation was carried out under two scenarios, aggregation_1 based on demo parameters (table 1) and aggregation_2. When changing to the latter, shops, entertainment, NDVI and Noise were set from 5% to 10%, 5% to 10%, 20 % to 15% and 20% to 15%.

To be short, here is the recommended list of the units:

Table 2. The best-fits (Aggregation_1 > 8).

Finn_code District Size Price (Kr) Type Ppsm (Kr) Aggregation_1
249110832 Gamle Oslo 71 6029516 Andel • Leilighet • 3 soverom 84923 8.28
248981927 Frogner 76 6909620 Eier (Selveier) • Leilighet • 2 soverom 90916 8.15
248049271 Søndre Nordstrand 81 3588916 Eier (Selveier) • Leilighet • 2 soverom 44308 8.18
249057789 Sagene 50 5088694 Eier (Selveier) • Leilighet • 1 soverom 101774 8.14
243905166 Sagene 53 4763626 Andel • Leilighet • 1 soverom 89880 8.36
248207270 Sagene 60 5362570 Andel • Leilighet • 2 soverom 89376 8.45
248035518 Sagene 95 8151292 Eier (Selveier) • Leilighet • 3 soverom 85803 8.38

There is a sensitivity test, using the second group of parameters:

Table 3. The best-fits (Aggregation_2 > 8).

Finn_code District **Size ** Price (Kr) Type Aggregation 2 Aggregation 1
249110832 Gamle Oslo 71 6029516 Andel • Leilighet • 3 soverom 8.4 8.28
248219200 Frogner 84 9754142 Eier (Selveier) • Leilighet • 2 soverom 8.2 7.80
249057789 Sagene 50 5088694 Eier (Selveier) • Leilighet • 1 soverom 8.6 8.14
248999688 Frogner 36 3985861 Andel • Leilighet • 1 soverom 8.2 7.60
243905166 Sagene 53 4763626 Andel • Leilighet • 1 soverom 8.7 8.36
195629975 - 71 7483877 Eier (Selveier) • Leilighet • 2 soverom 8.1 7.50
248035518 Sagene 95 8151292 Eier (Selveier) • Leilighet • 3 soverom 8.6 8.38
Aggregation 2 result. In my own preferences, model strongly suggests units near the river and park in the city center. Secondly, Ullern, Vestre Aker, Bjerke, Grorud, Stovner, Nordstrand and Stenbraten also got high scores for good public transportation, nice environment and numerous shops.

Anything else?

A unit (finn code 248049271) at Søndre Nordstrand with extreme low price (3.6 million Kr) and good size (81 m2) got 8.18 points. Sadly, it is sold already. Most of the units recommend by model are sold already in the past month, which means the best-fit units are popular in some degree.

I know there are several issues in my method, but I have to limit my time on this project. Otherwise, a Web-based, automating GIS-MCDA model could be super fun for exploring dataset from Finn.

In general, the housing price MCDA model could provide the best-fitting options for house buyers but is not capable of suggesting selling price for landlords, because personnel preferences are totally different from the market average, and the model does not count the attributions of the house itself. But it is still possible and fun to do price estimation after gathering more information about market average and normalizing house attributions with coefficients.

(left) Spatial score and total price scatter plot (Aggregation_1). The dot size represents the unit size. (right) Spatial score and price per square meters scatter plot (Aggregation_1). The dot size represents unit size.