Halo: Improving forecast accuracy through heteroscedastic estimation

The study asks whether estimating both location and scale in time-series forecasting can improve point-estimate accuracy.

Source: E001

The work introduces a modification that augments an existing deep forecaster with an output for distributional scale and trains it using the corresponding likelihood objective.

Source: E004

The proposed modification is intended to reuse the underlying forecasting architecture rather than require architecture-specific redesign.

Source: E022

Reported experiments show improvement on both accuracy measures across most model-market-metric comparisons.

Source: E004

The reported average reductions span a range for both squared-error and absolute-error measures, varying by model.

Source: E006

Each reported result comes from a single training run, and variability across seeds is not reported.

Source: E012

Testing on other benchmarks, forecast horizons, and seeds remains future work.

Source: E011

The work introduces a modification that augments an existing deep forecaster with an output for distributional scale and trains it using the corresponding likelihood objective.

Source: E004

The proposed modification is intended to reuse the underlying forecasting architecture rather than require architecture-specific redesign.

Source: E022

The study asks whether estimating both location and scale in time-series forecasting can improve point-estimate accuracy.

Source: E001

The work introduces a modification that augments an existing deep forecaster with an output for distributional scale and trains it using the corresponding likelihood objective.

Source: E004

The proposed modification is intended to reuse the underlying forecasting architecture rather than require architecture-specific redesign.

Source: E022

The modified forecaster produces location estimates together with positive scale estimates.

Source: E009

The implementation compares a shared-representation dual-output design with a parallel design that uses separate network copies for location and scale.

Source: E002, E010

The experiments adapt a transformer, a graph-based model paired with a variational autoencoder, and a shallow convolutional model.

Source: E019

The adaptations use distributional choices aligned with the original squared-error and absolute-error training objectives.

Source: E019

All models use series standardization, and the modified outputs are transformed back using the input-series scale, with the location also restored using its mean.

Source: E007, E008, E015, E018

Evaluation uses an electricity-price forecasting benchmark spanning several market datasets.

Source: E021

The comparison baseline retains each original forecasting model and reuses its previously tuned settings for the relevant model and market.

Source: E021

Modified variants are evaluated against those baselines, including validation-based tuning experiments for the modified models.

Source: E016

Accuracy is assessed with squared-error and absolute-error measures, aggregated over the forecast horizon.

Source: E014, E020

Tuning uses a validation holdout, while reported evaluation uses a separate test holdout reached after tuning.

Source: E012

Reported experiments show improvement on both accuracy measures across most model-market-metric comparisons.

Source: E004

The reported average reductions span a range for both squared-error and absolute-error measures, varying by model.

Source: E006

In the architecture comparisons, estimating scale appears more consequential for accuracy than choosing between the dual-output and parallel designs.

Source: E013

The reported improvement generally remains when the modified model uses settings already tuned for the point-estimate baseline, while retuning has mixed effects.

Source: E013

The tested setting withholds future exogenous inputs, assumes a single endogenous channel, and aligns exogenous and endogenous lookback windows.

Source: E003, E005, E017

The heteroscedastic distributions considered use a location-and-scale parameterization.

Source: E003

Training uses an adaptive optimizer, early stopping, a decaying learning-rate schedule, and seeds derived from a fixed base together with model and market identity.

Source: E012

Experiments run on a unified-memory personal-computing system using a machine-learning framework and its platform graphics backend.

Source: E012

Each reported result comes from a single training run, and variability across seeds is not reported.

Source: E012

Testing on other benchmarks, forecast horizons, and seeds remains future work.

Source: E011