Preprint2026

Memory Choice and Tail Risk in Volatility Forecasting

Fuli Yang; Tingting Qiu; Shiyu Lin

Volatility & Financial Econometrics →

Abstract

In a paired archive of forecasts for four equity markets, extreme underprediction is more frequent under shorter memory. Relative to full memory, a half-life of 138 updates raises extreme QLIKE incidence from 1.03% to 2.03% (difference: 1.00 percentage point; 95% paired block interval: 0.39 to 1.75). The change comprises 118 entries and two exits, all on the underprediction side. Excluding 2020 and re-estimating the cutoffs leaves a 0.59-point increase. These comparisons describe stored forecasts whose original generation and information timing cannot be reconstructed. A separate, maturity-timed S&P 500 experiment delineates their scope: under an unchanged loss threshold, calibration changes the short-minus-full contrast from −2.55 to +0.62 points. The paired change is +3.17 points, with interval [−0.31, 8.89]. Of the 41-count shift in the short-minus-full exceedance contrast after calibration, 39 arose in 2022. Directional event accounting thus reveals both the underprediction cost observed in the archive and the dependence of memory comparisons on calibration. Selection criteria identify candidate shortfalls; they do not guarantee subsequent tail control.

Author abstract from the GARCH Press V1 manuscript.

Keywords

JEL: C22, C53, C58, G17

Cite this paper

Fuli Yang, Tingting Qiu, Shiyu Lin. Memory Choice and Tail Risk in Volatility Forecasting. GARCH Press. https://doi.org/10.69057/WP-0124-V1

Original source citation

Yang, F.; Qiu, T.; Lin, S. (2026). Memory Choice and Tail Risk in Volatility Forecasting. Zenodo. https://doi.org/10.5281/zenodo.24015305

Version & publication history

  1. Working Paper · V1

    GARCH Press version ↗
    10.69057/WP-0124-V1 ↗
  2. Original repository record

    Existing Zenodo record and DOI. Version labels follow the source; no additional version number is inferred.

    View source record ↗
  3. Added to GARCH Research Papers

    Bibliographic record added. This is not a new manuscript release or a new DOI.

Access & rights

Sign in to preview or download the GARCH Press manuscript. Access is available to authorized accounts.

Restricted full text at source.

See source record for manuscript access and rights.