Semi-parametric Realized Nonlinear Conditional Autoregressive Expectile and Expected Shortfall

Chao Wang; Richard Gerlach

arxiv: 1906.09961 · v1 · pith:XRCA75M4new · submitted 2019-06-21 · 💱 q-fin.RM · econ.EM

Semi-parametric Realized Nonlinear Conditional Autoregressive Expectile and Expected Shortfall

Chao Wang , Richard Gerlach This is my paper

Pith reviewed 2026-05-25 18:41 UTC · model grok-4.3

classification 💱 q-fin.RM econ.EM

keywords expectileexpected shortfallVaR forecastingES forecastingrealized measuresconditional autoregressivenonlinear thresholdBayesian MCMC

0 comments

The pith

A joint conditional autoregressive expectile and Expected Shortfall model uses realized measures and nonlinear thresholds to improve one-day-ahead VaR and ES forecasts.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a framework that jointly models conditional autoregressive expectiles and Expected Shortfall. It extends the setup with a measurement equation that directly links realized volatility measures to the latent conditional expectile and adds nonlinear threshold specifications. Bayesian MCMC estimation is developed and tested via simulation, then applied to one-day-ahead forecasting of Value-at-Risk and Expected Shortfall on seven market indices. The results indicate that these additions produce more accurate risk forecasts than prior approaches. This matters for financial risk management because better daily forecasts directly affect capital requirements and hedging decisions.

Core claim

The authors claim that a semi-parametric realized nonlinear conditional autoregressive expectile and Expected Shortfall framework, extended by a measurement equation capturing contemporaneous dependence between realized measures and latent conditional expectiles plus nonlinear threshold effects, yields improved one-day-ahead VaR and ES forecasts, as shown in empirical studies across seven market indices.

What carries the argument

The realized nonlinear conditional autoregressive expectile (RN-CARE) model with joint ES, using a measurement equation to link realized measures to the latent expectile.

If this is right

The joint framework produces more accurate one-day-ahead VaR and ES forecasts than standard alternatives.
Incorporating realized measures via the measurement equation improves estimation of the latent expectile.
Nonlinear threshold effects better capture asymmetric dynamics in financial returns.
Bayesian MCMC estimation reliably recovers parameters in both simulated and real data settings.
Empirical gains hold across multiple equity market indices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same measurement-equation approach could be tested on individual stocks or other asset classes to check whether the forecasting gains generalize.
If the nonlinear threshold improves fit, it might also reduce the number of exceedances in regulatory backtests.
Extending the model to multi-horizon forecasts would test whether the realized-measure link remains useful beyond one day.
The framework could be combined with portfolio optimization routines to assess economic value of the improved risk measures.

Load-bearing premise

The measurement equation must correctly capture the contemporaneous dependence between realized measures and the latent conditional expectile, and the nonlinear threshold must be the right functional form for the data-generating process.

What would settle it

One-day-ahead VaR and ES forecasts on the seven market indices show no accuracy gain, or show losses, when the full proposed model is used versus the version without the measurement equation or without the nonlinear thresholds.

Figures

Figures reproduced from arXiv: 1906.09961 by Chao Wang, Richard Gerlach.

**Figure 1.** Figure 1: In-sample wt plots estimated with ES-CAViaR-Add model with S&P 500. metric Least Squares (ALS) equation (Taylor, 2008): Xn t=1 |τ − I(rt < µτ )|(rt − µτ ) 2 , (4) No distributional assumption is required to estimate µτ here. As discussed in Section 1, ES is defined as ESα = E(Y |Y < Qα), which stands for the expected value of Y , conditional on the set of Y that is more extreme than the α-level quantile of… view at source ↗

**Figure 2.** Figure 2: S&P 500 VaR forecasts with GARCH-Skew-t, ES-CAViaR-M [PITH_FULL_IMAGE:figures/full_fig_p027_2.png] view at source ↗

**Figure 3.** Figure 3: S&P 500 VaR forecasts (zoomed in) with GARCH-Skew-t, E [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

**Figure 4.** Figure 4: For S&P 500 forecasting, top plot visualizes S&P 500 retur [PITH_FULL_IMAGE:figures/full_fig_p029_4.png] view at source ↗

**Figure 5.** Figure 5: For S&P 500 forecasting, top plot visualizes S&P 500 retur [PITH_FULL_IMAGE:figures/full_fig_p030_5.png] view at source ↗

**Figure 6.** Figure 6: S&P 500 ES forecasts (zoomed in) with CARE, Threshold-G [PITH_FULL_IMAGE:figures/full_fig_p034_6.png] view at source ↗

read the original abstract

A joint conditional autoregressive expectile and Expected Shortfall framework is proposed. The framework is extended through incorporating a measurement equation which models the contemporaneous dependence between the realized measures and the latent conditional expectile. Nonlinear threshold specification is further incorporated into the proposed framework. A Bayesian Markov Chain Monte Carlo method is adapted for estimation, whose properties are assessed and compared with maximum likelihood via a simulation study. One-day-ahead VaR and ES forecasting studies, with seven market indices, provide empirical support to the proposed models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Extends conditional autoregressive expectile models with a measurement equation for realized measures and nonlinear thresholds, but any forecasting gains depend on those two choices being correctly specified without robustness checks shown.

read the letter

The paper proposes a joint framework for conditional expectiles and expected shortfall that adds a measurement equation linking realized measures to the latent expectile plus a nonlinear threshold specification. They adapt MCMC for estimation, run a simulation comparing it to maximum likelihood, and report one-day-ahead VaR and ES forecasts on seven market indices that appear to support the model over simpler versions. That is the core of what is new here—an incremental extension of existing CAViaR-style work rather than a fresh result. The setup is laid out clearly enough and the empirical exercise gives concrete numbers on real data, which is useful for anyone already working in this area. The simulation at least checks basic estimation properties. The soft spot is exactly what the stress-test note flags: the claimed improvements rest on the measurement equation correctly capturing the dependence and the nonlinear threshold being the right form. The abstract gives no sign of tests against linear baselines or alternative dependence structures, so if either piece is misspecified the gains could be artifacts and the model collapses back to standard CARE-ES. That is a real limitation for a forecasting paper, though not fatal on its own. This is for readers focused on realized volatility and risk measure forecasting in finance. Someone already using expectile or quantile autoregressive models might want to see the full specification and try the measurement equation on their data. It is coherent enough and has enough empirical content to deserve peer review rather than a desk reject, even if referees will likely press for more robustness on the functional form choices.

Referee Report

3 major / 2 minor

Summary. The paper proposes a joint conditional autoregressive expectile (CARE) and Expected Shortfall (ES) framework extended by a measurement equation capturing contemporaneous dependence between realized measures and the latent conditional expectile, plus a nonlinear threshold specification. Estimation uses an adapted Bayesian MCMC method whose properties are compared to MLE in a simulation study; one-day-ahead VaR and ES forecasting performance is evaluated empirically on seven market indices, with results claimed to support the proposed models.

Significance. If the measurement equation and nonlinear threshold demonstrably improve out-of-sample VaR/ES forecasts beyond standard CARE-ES specifications, the work would advance semi-parametric realized risk modeling by integrating high-frequency information and nonlinearity in a joint tail-risk framework. The simulation study comparing MCMC and MLE properties and the multi-index empirical forecasting exercise constitute clear strengths.

major comments (3)

[§3.2] §3.2 (Measurement equation): The paper does not report robustness checks against alternative dependence structures (e.g., lagged or nonlinear forms) between realized measures and the latent expectile; if the linear contemporaneous specification in the measurement equation is misspecified, the claimed forecasting gains relative to standard CARE-ES would not be attributable to the extension.
[§5] §5 (Empirical forecasting study): The one-day-ahead VaR/ES comparisons on the seven indices lack an explicit linear-threshold baseline; without isolating the incremental contribution of the nonlinear threshold (e.g., via direct comparison to the linear version of the same joint model), it is unclear whether reported improvements stem from the nonlinearity or from other modeling choices.
[§4] §4 (Simulation study): The Monte Carlo experiments evaluate MCMC convergence and bias under the assumed data-generating process but do not include experiments under misspecified measurement-equation or threshold forms; this leaves open whether the estimator remains reliable when the central modeling assumptions are violated, which is load-bearing for the empirical claims.

minor comments (2)

Notation for the threshold parameters and the realized-measure scaling in the measurement equation could be clarified with an explicit parameter table.
Figure captions for the forecasting performance plots should include the exact loss functions and significance tests used for the reported rankings.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We respond to each major comment below, indicating whether revisions will be made to the manuscript.

read point-by-point responses

Referee: [§3.2] §3.2 (Measurement equation): The paper does not report robustness checks against alternative dependence structures (e.g., lagged or nonlinear forms) between realized measures and the latent expectile; if the linear contemporaneous specification in the measurement equation is misspecified, the claimed forecasting gains relative to standard CARE-ES would not be attributable to the extension.

Authors: We agree that robustness checks against alternative dependence structures would strengthen the attribution of forecasting gains. In the revised manuscript, we will add results using lagged realized measures as well as a nonlinear form in the measurement equation to evaluate sensitivity of the findings. revision: yes
Referee: [§5] §5 (Empirical forecasting study): The one-day-ahead VaR/ES comparisons on the seven indices lack an explicit linear-threshold baseline; without isolating the incremental contribution of the nonlinear threshold (e.g., via direct comparison to the linear version of the same joint model), it is unclear whether reported improvements stem from the nonlinearity or from other modeling choices.

Authors: We acknowledge that an explicit comparison to the linear-threshold version of the joint model is needed to isolate the nonlinearity contribution. We will incorporate such direct comparisons in the revised empirical forecasting study. revision: yes
Referee: [§4] §4 (Simulation study): The Monte Carlo experiments evaluate MCMC convergence and bias under the assumed data-generating process but do not include experiments under misspecified measurement-equation or threshold forms; this leaves open whether the estimator remains reliable when the central modeling assumptions are violated, which is load-bearing for the empirical claims.

Authors: The simulation study is designed to assess finite-sample properties (convergence, bias) of the MCMC estimator when the data-generating process matches the model assumptions, which is standard practice for validating an estimation procedure. Experiments under misspecification address model robustness rather than estimator reliability under correct specification. We maintain that the current design is appropriate and do not plan to expand the simulation section. revision: no

Circularity Check

0 steps flagged

No circularity: new model specification estimated from data with independent empirical validation

full rationale

The paper introduces a joint CARE-ES framework extended by a measurement equation for realized measures and a nonlinear threshold specification. Estimation uses Bayesian MCMC (compared to MLE in simulation), followed by one-day-ahead VaR/ES forecasting on seven market indices. No load-bearing step reduces a claimed prediction to a fitted parameter by construction, nor does any uniqueness theorem or ansatz rely on self-citation chains. The forecasting gains are presented as empirical outcomes conditional on the chosen functional forms, with no evidence that the central results are tautological with the inputs. This is a standard model-proposal paper whose derivation chain is self-contained against external data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; the model description implies several free parameters (autoregressive coefficients, threshold levels, measurement-equation loadings) and domain assumptions about the joint distribution of returns and realized measures, but none can be enumerated precisely.

pith-pipeline@v0.9.0 · 5603 in / 1146 out tokens · 24305 ms · 2026-05-25T18:41:09.609581+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages · 1 internal anchor

[1]

,Amemiya, T., and Poirier, D

Aigner, D.J. ,Amemiya, T., and Poirier, D. J. (1976). On the Estimation of Production Frontiers: Maximum Likelihood Estimation of the Parameters of a Disc ontinuous Density Function. International Economic Review , 17, 377-396. Andersen, T. G. and Bollerslev, T. (1998). Answering the skeptics : Yes, standard volatil- ity models do provide accurate forecas...

work page arXiv 1976
[2]

Bayesian Semi-parametric Realized-CARE Models for Tail Risk Forecasting Incorporating Realized Measures

Boca Raton, FL: CRC press. Gerlach, R, Chen, C.W.S. and Chan, N.Y. (2011). Bayesian time-vary ing quantile fore- casting for value-at-risk in ﬁnancial markets. Journal of Business & Economic Statistics, 29(4), 481-492. Gerlach, R. and Chen, C.W.S. (2016). Bayesian Expected Shortfall Forecasting Incor- porating the Intraday Range, Journal of Financial Econ...

work page internal anchor Pith review Pith/arXiv arXiv 2011
[3]

and Laurent, S

Giot, P. and Laurent, S. (2004). Modelling daily value-at-risk using r ealized volatility and ARCH type models. Journal of Empirical Finance , 11(3), 379-398. Glosten, L.R., Jagannathan, R. and Runkle, D.E. (1993). On the rela tion between the expected value and the volatility of the nominal excess return on st ocks. The journal of ﬁnance , 48(5), 1779-18...

work page 2004
[4]

Harvey, A.C. (2013). Dynamic Models for Volatility and Heavy Tails, Ec onometric Society Monograph 52, Cambridge University Press, Cambridge. Koenker, R. and Machado, J.A. (1999). Goodness of ﬁt and relate d inference processes for quantile regression. Journal of the american statistical association , 94(448), 1296-1310. 38 Kupiec, P. H. (1995). Technique...

work page 2013
[5]

and Chen, R

Patton, A.J., Ziegel, J.F. and Chen, R. (2017). Dynamic semiparamet ric models for expected shortfall (and value-at-risk). arXiv preprint arXiv:170 7.05108. Roberts, G. O., Gelman, A. and Gilks, W. R. (1997). Weak convergen ce and optimal scaling of random walk Metropolis algorithms. The annals of applied probability , 7(1), 110-120. Taylor, J. (2008). Es...

work page doi:10.1080/07350015.2017.1281815 2017

[1] [1]

,Amemiya, T., and Poirier, D

Aigner, D.J. ,Amemiya, T., and Poirier, D. J. (1976). On the Estimation of Production Frontiers: Maximum Likelihood Estimation of the Parameters of a Disc ontinuous Density Function. International Economic Review , 17, 377-396. Andersen, T. G. and Bollerslev, T. (1998). Answering the skeptics : Yes, standard volatil- ity models do provide accurate forecas...

work page arXiv 1976

[2] [2]

Bayesian Semi-parametric Realized-CARE Models for Tail Risk Forecasting Incorporating Realized Measures

Boca Raton, FL: CRC press. Gerlach, R, Chen, C.W.S. and Chan, N.Y. (2011). Bayesian time-vary ing quantile fore- casting for value-at-risk in ﬁnancial markets. Journal of Business & Economic Statistics, 29(4), 481-492. Gerlach, R. and Chen, C.W.S. (2016). Bayesian Expected Shortfall Forecasting Incor- porating the Intraday Range, Journal of Financial Econ...

work page internal anchor Pith review Pith/arXiv arXiv 2011

[3] [3]

and Laurent, S

Giot, P. and Laurent, S. (2004). Modelling daily value-at-risk using r ealized volatility and ARCH type models. Journal of Empirical Finance , 11(3), 379-398. Glosten, L.R., Jagannathan, R. and Runkle, D.E. (1993). On the rela tion between the expected value and the volatility of the nominal excess return on st ocks. The journal of ﬁnance , 48(5), 1779-18...

work page 2004

[4] [4]

Harvey, A.C. (2013). Dynamic Models for Volatility and Heavy Tails, Ec onometric Society Monograph 52, Cambridge University Press, Cambridge. Koenker, R. and Machado, J.A. (1999). Goodness of ﬁt and relate d inference processes for quantile regression. Journal of the american statistical association , 94(448), 1296-1310. 38 Kupiec, P. H. (1995). Technique...

work page 2013

[5] [5]

and Chen, R

Patton, A.J., Ziegel, J.F. and Chen, R. (2017). Dynamic semiparamet ric models for expected shortfall (and value-at-risk). arXiv preprint arXiv:170 7.05108. Roberts, G. O., Gelman, A. and Gilks, W. R. (1997). Weak convergen ce and optimal scaling of random walk Metropolis algorithms. The annals of applied probability , 7(1), 110-120. Taylor, J. (2008). Es...

work page doi:10.1080/07350015.2017.1281815 2017