Introduction
Hedonic Modeling of Spatially Structured House Prices and Rents
Parametric Hedonic Models
Implications of Spatial Dependence
Accounting for Spatial Dependence
Non-Parametric Hedonic Models and Cross-Validation
Implications of Spatial Dependence
Accounting for Spatial Dependence
Data and Methodology
Data Description
Variable | N | Mean | Median | SD | Min | Max |
---|---|---|---|---|---|---|
Continuous | ||||||
Rent per month [Euro] | 9256 | 1088.77 | 940.00 | 647.97 | 190.00 | 10,000.00 |
Living Area [sqm] | 9256 | 75.20 | 70.00 | 35.07 | 10.00 | 440.00 |
Age [years] | 9256 | 44.59 | 46.00 | 39.34 | −2.00 | 118.00 |
Entry date [years] | 9256 | 0.65 | 0.67 | 0.35 | 0.08 | 1.25 |
Latitude | 9256 | 50.12 | 50.12 | 0.02 | 50.08 | 50.21 |
Longitude | 9256 | 8.66 | 8.66 | 0.05 | 8.49 | 8.78 |
Discrete | ||||||
Rooms | 9256 | 2.55 | 2.50 | 1.00 | 1.00 | 8.50 |
Floor | 9256 | 2.54 | 2.00 | 2.82 | −0.50 | 39.00 |
Dummies [1 = yes, 0 = no] | ||||||
Bathtub | 9256 | 0.53 | 1.00 | 0.50 | 0.00 | 1.00 |
Refurbished | 9256 | 0.22 | 0.00 | 0.41 | 0.00 | 1.00 |
Built-in kitchen | 9256 | 0.71 | 1.00 | 0.45 | 0.00 | 1.00 |
Balcony | 9256 | 0.65 | 1.00 | 0.48 | 0.00 | 1.00 |
Parking | 9256 | 0.48 | 0.00 | 0.50 | 0.00 | 1.00 |
Elevator | 9256 | 0.50 | 1.00 | 0.50 | 0.00 | 1.00 |
Terrace | 9256 | 0.13 | 0.00 | 0.34 | 0.00 | 1.00 |
Distances | ||||||
NUTS centroid [km] | 9256 | 3.65 | 3.68 | 1.87 | 0.01 | 10.84 |
Bakery [km] | 9256 | 0.39 | 0.26 | 0.41 | 0.00 | 1.61 |
Bar [km] | 9256 | 0.73 | 0.52 | 0.64 | 0.00 | 2.54 |
Biergarten [km] | 9256 | 1.16 | 0.97 | 0.77 | 0.02 | 3.10 |
Café [km] | 9256 | 0.36 | 0.25 | 0.33 | 0.00 | 1.31 |
School [km] | 9256 | 0.31 | 0.28 | 0.17 | 0.02 | 0.75 |
Supermarket [km] | 9256 | 0.26 | 0.22 | 0.17 | 0.00 | 0.75 |
Bus station [km] | 9256 | 3.13 | 2.77 | 1.54 | 0.09 | 7.56 |
Entry Date | N | Mean Rent [Euro] | Mean Rent [Euro/sqm] |
---|---|---|---|
Jan-19 | 632 | 1129.57 | 14.57 |
Feb-19 | 586 | 1116.31 | 14.34 |
Mar-19 | 677 | 1081.28 | 14.20 |
Apr-19 | 576 | 1118.35 | 14.39 |
May-19 | 685 | 1135.66 | 14.54 |
Jun-19 | 602 | 1084.38 | 14.41 |
Jul-19 | 746 | 1083.23 | 14.22 |
Aug-19 | 755 | 1014.57 | 14.20 |
Sep-19 | 602 | 1177.08 | 14.37 |
Oct-19 | 633 | 1088.54 | 14.26 |
Nov-19 | 613 | 1005.09 | 13.85 |
Dec-19 | 392 | 1021.86 | 14.48 |
Jan-20 | 600 | 1101.10 | 14.82 |
Feb-20 | 611 | 1096.45 | 14.75 |
Mar-20 | 546 | 1068.98 | 14.73 |
Methodological Approach
Parametric Models
Non-Parametric Models
Performance Evaluation
Results
Model Selection
Fold | b (1) | b (2) | b (3) | b (4) | m (1) | m (2) | m (3) | m (4) | minnode (1) | minnode (2) | minnode (3) | minnode (4) |
Panel A1: Random Forest including Spatial Controls | ||||||||||||
1 | 650 | 500 | 400 | 400 | 9 | 9 | 12 | 14 | 2 | 2 | 8 | 7 |
2 | 300 | 350 | 550 | 250 | 9 | 9 | 10 | 10 | 1 | 1 | 5 | 8 |
3 | 300 | 500 | 200 | 500 | 9 | 9 | 10 | 10 | 1 | 1 | 6 | 6 |
4 | 300 | 600 | 500 | 450 | 7 | 9 | 9 | 10 | 2 | 1 | 2 | 1 |
5 | 500 | 450 | 300 | 650 | 7 | 7 | 10 | 10 | 1 | 2 | 6 | 5 |
6 | 350 | 500 | 450 | 600 | 9 | 9 | 9 | 12 | 2 | 2 | 7 | 6 |
7 | 600 | 300 | 500 | 500 | 9 | 7 | 12 | 9 | 1 | 1 | 1 | 7 |
8 | 650 | 650 | 650 | 550 | 7 | 9 | 9 | 10 | 1 | 1 | 3 | 7 |
9 | 550 | 650 | 600 | 400 | 9 | 9 | 10 | 9 | 1 | 3 | 4 | 6 |
10 | 550 | 500 | 650 | 500 | 9 | 10 | 9 | 10 | 2 | 1 | 2 | 3 |
Panel B1: Random Forest excluding Spatial Controls | ||||||||||||
1 | 550 | 650 | 200 | 500 | 5 | 6 | 5 | 6 | 2 | 1 | 9 | 10 |
2 | 250 | 250 | 450 | 550 | 5 | 5 | 5 | 6 | 1 | 1 | 9 | 8 |
3 | 600 | 450 | 500 | 250 | 5 | 5 | 6 | 5 | 2 | 1 | 10 | 8 |
4 | 500 | 550 | 200 | 600 | 5 | 5 | 5 | 6 | 2 | 1 | 8 | 10 |
5 | 600 | 650 | 500 | 500 | 5 | 5 | 5 | 6 | 1 | 2 | 10 | 8 |
6 | 600 | 350 | 450 | 300 | 5 | 5 | 6 | 6 | 2 | 2 | 9 | 9 |
7 | 450 | 650 | 350 | 350 | 5 | 6 | 6 | 6 | 1 | 1 | 9 | 9 |
8 | 400 | 200 | 450 | 600 | 5 | 5 | 6 | 5 | 1 | 1 | 10 | 9 |
9 | 450 | 450 | 600 | 600 | 5 | 5 | 5 | 6 | 1 | 1 | 8 | 10 |
10 | 450 | 550 | 500 | 300 | 6 | 6 | 6 | 6 | 2 | 1 | 9 | 9 |
Panel A2: Extreme Gradient Boosting Trees including Spatial Controls | ||||||||||||
Fold | nrounds (1) | nrounds (2) | nrounds (3) | nrounds (4) | m/p (1) | m/p (2) | m/p (3) | m/p (4) | η (1) | η (2) | η (3) | η (4) |
1 | 600 | 550 | 250 | 400 | 72% | 58% | 65% | 45% | 0.06 | 0.07 | 0.04 | 0.05 |
2 | 650 | 600 | 550 | 250 | 38% | 78% | 65% | 58% | 0.06 | 0.10 | 0.02 | 0.06 |
3 | 650 | 550 | 200 | 300 | 58% | 85% | 52% | 85% | 0.10 | 0.07 | 0.08 | 0.05 |
4 | 550 | 600 | 500 | 350 | 78% | 78% | 58% | 85% | 0.08 | 0.08 | 0.02 | 0.02 |
5 | 600 | 600 | 200 | 550 | 72% | 72% | 78% | 45% | 0.07 | 0.08 | 0.04 | 0.02 |
6 | 600 | 600 | 450 | 450 | 65% | 45% | 45% | 45% | 0.07 | 0.09 | 0.06 | 0.03 |
7 | 650 | 550 | 550 | 300 | 72% | 78% | 45% | 65% | 0.08 | 0.09 | 0.09 | 0.04 |
8 | 500 | 650 | 450 | 350 | 52% | 52% | 78% | 38% | 0.08 | 0.09 | 0.02 | 0.07 |
9 | 650 | 650 | 500 | 650 | 58% | 65% | 78% | 52% | 0.05 | 0.06 | 0.02 | 0.02 |
10 | 500 | 550 | 400 | 450 | 78% | 45% | 72% | 65% | 0.07 | 0.08 | 0.03 | 0.03 |
Panel B2: Extreme Gradient Boosting Trees excluding Spatial Controls | ||||||||||||
1 | 500 | 650 | 500 | 600 | 72% | 72% | 65% | 78% | 0.07 | 0.06 | 0.01 | 0.01 |
2 | 500 | 400 | 300 | 200 | 78% | 78% | 72% | 85% | 0.05 | 0.07 | 0.02 | 0.03 |
3 | 650 | 650 | 250 | 550 | 58% | 65% | 72% | 85% | 0.04 | 0.04 | 0.02 | 0.01 |
4 | 650 | 650 | 550 | 250 | 72% | 65% | 58% | 85% | 0.05 | 0.06 | 0.01 | 0.02 |
5 | 650 | 600 | 550 | 600 | 65% | 58% | 72% | 72% | 0.04 | 0.07 | 0.01 | 0.01 |
6 | 550 | 600 | 200 | 500 | 65% | 65% | 85% | 65% | 0.05 | 0.05 | 0.03 | 0.02 |
7 | 400 | 600 | 600 | 550 | 78% | 58% | 72% | 65% | 0.09 | 0.08 | 0.01 | 0.01 |
8 | 450 | 350 | 500 | 500 | 52% | 58% | 72% | 78% | 0.07 | 0.07 | 0.01 | 0.01 |
9 | 450 | 650 | 200 | 300 | 78% | 65% | 72% | 85% | 0.06 | 0.04 | 0.03 | 0.02 |
10 | 450 | 450 | 500 | 600 | 65% | 45% | 65% | 65% | 0.07 | 0.06 | 0.01 | 0.01 |
Model Assessment
Panel A: Models including Spatial Controls | |||||||||
Method | Resampling Strategy | R2 | MAE | MAPE | MPE | RMSE | PE10 | IQR | COD |
OLS | holdout | 78.81% | 168.24 | 15.03% | −0.61% | 292.14 | 44.45% | 205.39 | 3.52% |
re-substitution | 85.13% | 149.79 | 13.91% | 1.64% | 251.01 | 47.21% | 199.63 | 13.73% | |
SDM | holdout | 81.62% | 153.56 | 13.77% | −0.91% | 272.09 | 50.54% | 182.57 | 3.25% |
re-substitution | 87.50% | 134.81 | 12.64% | 1.40% | 230.13 | 51.81% | 177.15 | 12.52% | |
SDEM | holdout | 79.08% | 166.51 | 14.70% | −1.57% | 290.29 | 45.65% | 198.80 | 3.45% |
re-substitution | 87.46% | 134.74 | 12.63% | 1.40% | 230.53 | 51.86% | 175.95 | 12.51% | |
RFR | holdout (non-spatial tuning) | 85.13% | 143.48 | 12.79% | −0.33% | 244.69 | 52.31% | 169.58 | 3.14% |
holdout (spatial tuning) | 85.14% | 143.58 | 12.81% | −0.30% | 244.64 | 52.48% | 171.37 | 3.15% | |
(1) non-spatial/non-spatial | 89.42% | 116.59 | 10.93% | 1.45% | 211.74 | 59.15% | 141.57 | 11.12% | |
(2) non-spatial/spatial | 82.62% | 157.30 | 14.35% | 0.67% | 271.40 | 45.83% | 208.53 | 14.69% | |
(3) spatial/spatial | 83.07% | 157.41 | 14.39% | 0.58% | 267.90 | 45.27% | 208.00 | 14.69% | |
(4) spatial/nonspatial | 89.49% | 117.93 | 11.08% | 1.42% | 211.07 | 58.65% | 144.64 | 11.26% | |
XGB | holdout (non-spatial tuning) | 85.20% | 142.45 | 12.89% | 0.71% | 244.15 | 53.04% | 165.43 | 3.15% |
holdout (spatial tuning) | 85.21% | 142.96 | 12.83% | 0.16% | 244.06 | 52.87% | 171.23 | 3.11% | |
(1) non-spatial/non-spatial | 90.93% | 112.66 | 10.53% | 1.13% | 196.08 | 60.71% | 140.46 | 10.67% | |
(2) non-spatial/spatial | 83.90% | 157.16 | 14.25% | −0.54% | 261.20 | 44.86% | 214.47 | 14.52% | |
(3) spatial/spatial | 84.54% | 152.67 | 13.95% | 0.00% | 255.98 | 46.21% | 202.73 | 14.19% | |
(4) spatial/nonspatial | 90.15% | 117.11 | 10.97% | 1.12% | 204.33 | 58.27% | 149.12 | 11.14% | |
Panel B: Models excluding Spatial Controls | |||||||||
OLS | holdout | 74.52% | 188.95 | 16.80% | −0.67% | 320.33 | 38.13% | 250.15 | 3.96% |
re-substitution | 80.58% | 173.36 | 15.99% | 2.08% | 286.89 | 40.23% | 235.13 | 15.66% | |
SDM | holdout | 81.56% | 154.06 | 13.82% | −0.91% | 272.49 | 49.91% | 184.02 | 3.25% |
re-substitution | 87.38% | 135.79 | 12.72% | 1.41% | 231.27 | 51.77% | 175.98 | 12.59% | |
SDEM | holdout | 75.90% | 177.71 | 15.41% | −2.07% | 311.53 | 42.17% | 218.24 | 3.67% |
re-substitution | 87.08% | 136.36 | 12.74% | 1.42% | 233.96 | 51.74% | 175.61 | 12.60% | |
RFR | holdout (non-spatial tuning) | 80.66% | 171.26 | 15.14% | 0.17% | 279.07 | 44.05% | 219.94 | 3.76% |
holdout (spatial tuning) | 80.59% | 172.98 | 15.30% | 0.22% | 279.60 | 43.54% | 217.75 | 3.83% | |
(1) non-spatial/non-spatial | 85.80% | 143.98 | 13.40% | 1.67% | 245.33 | 49.17% | 185.16 | 13.68% | |
(2) non-spatial/spatial | 80.51% | 173.25 | 15.85% | 2.00% | 287.38 | 40.70% | 228.44 | 16.05% | |
(3) spatial/spatial | 80.29% | 172.41 | 15.73% | 1.96% | 289.02 | 40.75% | 226.66 | 15.94% | |
(4) spatial/nonspatial | 85.60% | 146.48 | 13.62% | 1.69% | 247.07 | 48.66% | 191.14 | 13.96% | |
XGB | holdout (non-spatial tuning) | 80.33% | 176.27 | 15.85% | 1.65% | 281.48 | 42.69% | 223.19 | 3.83% |
holdout (spatial tuning) | 79.47% | 173.97 | 15.28% | −1.17% | 287.56 | 43.26% | 225.95 | 3.79% | |
(1) non-spatial/non-spatial | 85.61% | 148.24 | 13.77% | 1.63% | 246.96 | 46.97% | 194.32 | 13.96% | |
(2) non-spatial/spatial | 81.26% | 172.33 | 15.78% | 1.97% | 281.83 | 40.39% | 226.85 | 15.93% | |
(3) spatial/spatial | 79.75% | 173.80 | 15.31% | −1.18% | 292.96 | 40.78% | 229.86 | 15.77% | |
(4) spatial/nonspatial | 84.24% | 154.54 | 14.03% | −0.52% | 258.41 | 45.02% | 208.12 | 14.49% |
Residual Spatial Autocorrelation
Panel A: Models including Spatial Controls | Panel B: Models excluding Spatial Controls | ||||||
---|---|---|---|---|---|---|---|
Method | Resampling Strategy | Morans’ I | Z-score | p value | Morans’ I | Z-score | p value |
OLS | holdout | 0.17 | 55.44 | 0.00 | 0.29 | 95.36 | 0.00 |
re-substitution | 0.27 | 19.30 | 0.00 | 0.39 | 27.48 | 0.00 | |
SDM | holdout | 0.03 | 10.17 | 0.00 | 0.03 | 10.23 | 0.00 |
re-substitution | 0.16 | 11.60 | 0.00 | 0.17 | 11.76 | 0.00 | |
SDEM | holdout | 0.03 | 10.75 | 0.00 | 0.04 | 13.58 | 0.00 |
re-substitution | 0.25 | 17.89 | 0.00 | 0.34 | 23.59 | 0.00 | |
RFR | holdout (non-spatial tuning) | 0.17 | 11.75 | 0.00 | 0.33 | 23.45 | 0.00 |
holdout (spatial tuning) | 0.17 | 11.70 | 0.00 | 0.32 | 22.60 | 0.00 | |
(1) non-spatial/non-spatial | 0.03 | 9.79 | 0.00 | 0.18 | 61.22 | 0.00 | |
(2) non-spatial/spatial | 0.19 | 64.70 | 0.00 | 0.26 | 87.78 | 0.00 | |
(3) spatial/spatial | 0.19 | 62.80 | 0.00 | 0.26 | 87.91 | 0.00 | |
(4) spatial/non-spatial | 0.03 | 9.97 | 0.00 | 0.19 | 63.13 | 0.00 | |
XGB | holdout (non-spatial tuning) | 0.14 | 9.52 | 0.00 | 0.26 | 18.53 | 0.00 |
holdout (spatial tuning) | 0.14 | 10.10 | 0.00 | 0.33 | 23.26 | 0.00 | |
(1) non-spatial/non-spatial | −0.01 | −1.90 | 0.06 | 0.15 | 51.03 | 0.00 | |
(2) non-spatial/spatial | 0.16 | 54.67 | 0.00 | 0.23 | 75.07 | 0.00 | |
(3) spatial/spatial | 0.16 | 53.59 | 0.00 | 0.28 | 92.21 | 0.00 | |
(4) spatial/non-spatial | 0.01 | 2.55 | 0.01 | 0.21 | 71.22 | 0.00 |