Spectral backtests of forecast distributions with application to risk management

Alexander John McNeil; Michael Gordy

doi:10.1016/j.jbankfin.2020.105817

Spectral backtests of forecast distributions with application to risk management

Alexander John McNeil, Michael Gordy

Management

Research output: Contribution to journal › Article › peer-review

Abstract

We study a class of backtests for forecast distributions in which the test statistic depends on a spectral transformation that weights exceedance events by a function of the modeled probability level. The weighting scheme is specified by a kernel measure which makes explicit the user’s priorities for model performance. The class of spectral backtests includes tests of unconditional coverage and tests of conditional coverage. We show how the class embeds a wide variety of backtests in the existing literature, and further propose novel variants which are easily implemented, well-sized and have good power. In an empirical application, we backtest forecast distributions for the overnight P&L of ten bank trading portfolios. For some portfolios, test results depend materially on the choice of kernel.

Original language	English
Article number	105817
Number of pages	13
Journal	Journal of Banking and Finance
Volume	116
Early online date	16 May 2020
DOIs	https://doi.org/10.1016/j.jbankfin.2020.105817
Publication status	Published - Jul 2020

Bibliographical note

Keywords

Backtesting; Volatility; Risk management

Access to Document

10.1016/j.jbankfin.2020.105817

spectraltest20200330_combined
© Elsevier, 2020. This is an author-produced version of the published paper. Uploaded in accordance with the publisher’s self-archiving policy.
Accepted author manuscript, 2.24 MBLicence: CC BY-NC-ND

Alexander John McNeil
- alexander.mcneilyork.acuk
- Management - Professor in Actuarial Science
Person: Academic

Cite this

@article{7c1ea36fe1ac47a789a998413049403c,

title = "Spectral backtests of forecast distributions with application to risk management",

abstract = "We study a class of backtests for forecast distributions in which the test statistic depends on a spectral transformation that weights exceedance events by a function of the modeled probability level. The weighting scheme is specified by a kernel measure which makes explicit the user{\textquoteright}s priorities for model performance. The class of spectral backtests includes tests of unconditional coverage and tests of conditional coverage. We show how the class embeds a wide variety of backtests in the existing literature, and further propose novel variants which are easily implemented, well-sized and have good power. In an empirical application, we backtest forecast distributions for the overnight P&L of ten bank trading portfolios. For some portfolios, test results depend materially on the choice of kernel.",

keywords = "Backtesting; Volatility; Risk management",

author = "McNeil, {Alexander John} and Michael Gordy",

note = "{\textcopyright} Elsevier, 2020. This is an author-produced version of the published paper. Uploaded in accordance with the publisher{\textquoteright}s self-archiving policy. ",

year = "2020",

month = jul,

doi = "10.1016/j.jbankfin.2020.105817",

language = "English",

volume = "116",

journal = "Journal of Banking & Finance",

issn = "0378-4266",

publisher = "Elsevier",

}

TY - JOUR

T1 - Spectral backtests of forecast distributions with application to risk management

AU - McNeil, Alexander John

AU - Gordy, Michael

N1 - © Elsevier, 2020. This is an author-produced version of the published paper. Uploaded in accordance with the publisher’s self-archiving policy.

PY - 2020/7

Y1 - 2020/7

N2 - We study a class of backtests for forecast distributions in which the test statistic depends on a spectral transformation that weights exceedance events by a function of the modeled probability level. The weighting scheme is specified by a kernel measure which makes explicit the user’s priorities for model performance. The class of spectral backtests includes tests of unconditional coverage and tests of conditional coverage. We show how the class embeds a wide variety of backtests in the existing literature, and further propose novel variants which are easily implemented, well-sized and have good power. In an empirical application, we backtest forecast distributions for the overnight P&L of ten bank trading portfolios. For some portfolios, test results depend materially on the choice of kernel.

AB - We study a class of backtests for forecast distributions in which the test statistic depends on a spectral transformation that weights exceedance events by a function of the modeled probability level. The weighting scheme is specified by a kernel measure which makes explicit the user’s priorities for model performance. The class of spectral backtests includes tests of unconditional coverage and tests of conditional coverage. We show how the class embeds a wide variety of backtests in the existing literature, and further propose novel variants which are easily implemented, well-sized and have good power. In an empirical application, we backtest forecast distributions for the overnight P&L of ten bank trading portfolios. For some portfolios, test results depend materially on the choice of kernel.

KW - Backtesting; Volatility; Risk management

U2 - 10.1016/j.jbankfin.2020.105817

DO - 10.1016/j.jbankfin.2020.105817

M3 - Article

SN - 0378-4266

VL - 116

JO - Journal of Banking & Finance

JF - Journal of Banking & Finance

M1 - 105817

ER -

Spectral backtests of forecast distributions with application to risk management

Abstract

Bibliographical note

Keywords

Access to Document

Profiles

Alexander John McNeil

Cite this