Multi-Horizon Price Forecasting for Second-Hand Electronics: A Systematic Benchmark of Deep Learning vs. Statistical Models
ORIGINAL / Deep Learning vs. Statistical Models for Multi-Horizon Price Forecasting of Second-Hand Electronics: A Systematic Benchmark
The first multi-horizon benchmark for second-hand electronics price forecasting, evaluating 11 statistical and deep learning models across six horizons from 1 to 365 days. N-BEATS outperforms statistical baselines at long horizons and a single model generalizes across horizons, offering methodological reference for subscription platform pricing.
01 ABSTRACT
The paper builds a large-scale multi-horizon forecasting benchmark for used electronics, addressing high volatility, sparse listings, and non-normal price dynamics. Experiments show N-BEATS achieves the lowest MAPE beyond 30 days, reaching 8.51% at 365 days versus 14.94% for the best statistical baseline, a 43% reduction. At short horizons (1-7 days) all models converge near 0.72% MAPE and the naive baseline remains competitive; a single N-BEATS trained at 365 days generalizes to all shorter horizons. These findings are based on daily price listings from Polish online marketplaces, and the authors report superior hyperparameter stability for N-BEATS and N-HiTS.
02 KEY FINDINGS
- N-BEATS achieves the lowest MAPE beyond 30 days, reaching 8.51% at 365 days, about a 43% reduction versus the best statistical baseline.
- At short horizons (1-7 days), all models converge to around 0.72% MAPE and the naive baseline remains competitive.
- A single N-BEATS model trained at 365 days generalizes to all shorter horizons, eliminating the need for horizon-specific models.
AI GENERATED SUMMARY / DISCOVERED BY ARXIV CS.LG · ARXIV Q-FIN