Forecasting Oil Price by Hierarchical Shrinkage in Dynamic Parameter Models
Table 10
Measures of forecast performance for log return of Brent with recursive rolling window results.
Constant variance
Stochastic variance
MAFE
MSFE
MLPL
MAFE
MSFE
MLPL
Model (h=1)
LASSO on constant and TVPs
0.066
0.008
3.273
0.083
0.012
2.438
LASSO only on constant coeff.
0.061
0.007
3.460
0.076
0.010
2.955
LASSO only on TVPs
0.078
0.010
2.750
0.088
0.013
2.423
TVP regression model
0.074
0.009
3.003
0.079
0.010
2.625
Constant coeff. model
0.074
0.009
2.991
0.077
0.010
2.696
Model (h=12)
LASSO on constant and TVPs
0.315
0.170
0.712
0.391
0.260
0.463
LASSO only on constant coeff.
0.306
0.166
0.592
0.328
0.206
0.702
LASSO only on TVPs
0.363
0.228
0.597
0.439
0.318
0.406
TVP regression model
0.411
0.270
0.481
0.392
0.253
0.486
Constant coeff. model
0.412
0.274
0.480
0.370
0.223
0.479
Note. The value noted in bold and underlined text indicates a model performing the best out of all models, while the bold and italic text represents a model performing the worst. MSFE, MAFE, and MLPL refer to the mean squared forecast error, mean absolute forecast error, and mean log predictive likelihood, respectively.
