risk assessment in commodity markets with semi-nonparametric specifications
TRANSCRIPT
MotivationModels and Methodology
ApplicationConclusion
Risk assessment in commodity markets withsemi-nonparametric specifications
Esther B. Del Brio, Andres Mora-Valencia, Javier Perote
I Network de Metodos Cuantitativos en Economıa - AFADECO, November 10,2017
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Outline
1 Motivation
2 Models and Methodology
3 Application
4 Conclusion
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Introduction
Commodity prices experienced a boom in the mid-2000s.
With the boom of commodity prices, several ExchangeTraded Funds (ETFs) were created in order to track the priceof main commodities such as gold, silver and oil.
Financial institutions and regulators - e.g. Financial StabilityBoard (FSB) and Bank for International Settlements (BIS) -expressed their concerns about the potential systemicimpact of the ETFs industry in 2011.
Commodity assets may present higher volatilities than equityassets even in “relatively” calm periods.
Recently, there has been a sharp decline in the price ofthese assets affecting developed markets based on mining andenergy industries like Canada and Australia, based metalsbusinesses in Deutsche Bank and JP Morgan, among others .
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Risk Measures
Value-at-risk (VaR) has been the standard market riskmeasure to assess regulatory capital requirements since 1996.
VaR has been criticized since they do not fulfill thesubadditivity property and its inability to accurately capturetail risk.
Expected Shortfall (ES) provides coherent risk measures.
Basel 2.5 (2012) proposes to replace VaR by ES.
ES is very sensitive to extreme events and may result inunstable capital numbers at high confidence levels.
ES does not satisfy the elicitability property, which generateddebate on whether ES can be backtestable.
VaR satisfies elicitability property and there are verywell-known backtesting methods for this measure.
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Literature Review
Authors Asset (Variable) Best Model
Giot & Laurent (2003) Several commodity markets APARCH + skewed-t innovations
Hung et al. (2008) Energy commodities GARCH + heavy tail model
Chiu et al. (2010) Brent and WTI crude oil prices Hull-White model
Aloui & Mabrouk (2010) Oil and gas commodities prices FIAPARCH + skewed-t innovations
Youssef et al. (2015) Crude oil and gasoline market FIAPARCH + EVT
AndriosopoulosNomikos(2015)
8 spot energy commodities and SEI MC simulations
Steen et al. (2015) 19 commotity futures Quantile regression
Aloui & Jammazi (2015) Oil-exchange rate pairs Wavelet-based models
Lu et al. (2014) Crude oil futures and natural gas futures t-Copula & skewed-t for mgnls
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
VaR/ES-ARMA(1,1)-EGARCH(1,1)
Rt+1 = µt+1 + σt+1Zt+1:
µt+1 = µ+ φ1µt + θεt + εt+1,
εt+1 = Zt+1σt+1 Zt+1 ∼ G (0, 1),
logσ2t+1 = ω + α (|zt | − E [|zt |]) + γzt + β log σ2
t ,
VaR = µt+1 + σt+1qγ(Zt+1),
ES = µt+1 + σt+1ESγ(Zt+1).
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Distributions (...for G)
1 Normal
2 Student’s t
3 Skewed t
4 GC Type A
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Normal, Student’s t and Skewed-t
Normal:
φ (zt) = 1√2π
exp{− z2
t2
}.
Student’s t:
t (zt) =Γ( ν+1
2 )√π(ν−2)Γ( ν
2 )
(1 + z2
tν−2
)− ν+12,
Skewed t (Fernandez and Steel, 1998):
g (zt) =
−2
γ+ 1γ
t (γzt) zt < 0,
2γ+ 1
γ
t(ztγ
)zt ≥ 0,
.
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Gram-Charlier Type A
GC Type A density:
f (zt ,d) =
(1 +
n∑s=1
dsHs (zt)
)φ (zt) ,
where:
φ (zt) is the normal pdf,
d = (d1, . . . , ds) ∈ Rs , and
Hs is the sth Hermite polynomial (HP) of order, which iscomputed in terms of the sth order derivative of the Gaussianpdf:
d sφ(zt)dzst
= (−1)s Hs (zt)φ (zt) .
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
GC Type A
First four HP:
H1 (zt) = zt
H2 (zt) = z2t − 1
H3 (zt) = z3t − 3zt
H4 (zt) = z4t − 6z2
t + 3
These polynomials form an orthonormal basis:∫Hs (zt)Hj (zt)φ (zt) dzt = 0 ∀s 6= j
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Fitted Densities
0
0,005
0,01
0,015
0,02
0,025
Fig. 2. Fitted densities (left tails)
Histogram Normal
Gram-Charlier Student's t
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
Fig. 1. Fitted densities
Histogram Normal
Gram-Charlier Student's t
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
GC Type A ES
ESα =
−φ(ϕ−1(α))α
[1 +
S∑s=2
ds{Hs
(ϕ−1 (α)
)+ sHs−2
(ϕ−1 (α)
)}]whereφ is the pdf of standard normal
andϕ−1 (α) is the α-quantile of the GC Type A distribution
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Backtesting methods
We compute one-step ahead forecasts for VaR and ES through a rollingwindow of size T and compare the model performance according to:
Backtesting methods for VaR
1 Bernoulli coverage test
2 Relative comparison for VaR
Backtesting methods for ES
1 t-test
2 Relative comparison for ES
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Commodity ETFs Prices
Sample: Daily prices form January 2007 to January 2016 for Gold, Silver,
Oil, Agriculture, Energy abd Broad Commodity ETFs
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Commodity ETFs Returns
Return data exhibits: volatility clustering, leptokurtosis, skewness,leverage effects
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Coverage test: 99%-VaR
Skewed-t and GC outperform the rest of the models. Result iscorroborated by the Diebold Mariano test for relative performance(pairwise comparison).
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
t-test: 97.5%-ES
Skewed-t and GC outperform the rest of the models. Result iscorroborated by the Diebold Mariano test for relative performance(pairwise comparison).
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Conclusion
We applied backtesting methods for both VaR and ES todifferent Commodity ETFs.
We compare the performance of different parametric andsemi-nonparametric specifications both in univariateframework.
Coverage and relative performance tests show that the skewedt and Gram-Charlier outperform other more traditional densityspecifications.
We show that the Gram-Charlier distribution is very tractablefor empirical purposes and provide a closed expression for ESwith GC distribution.
We recommend the application of this distribution to mitigateregulation concerns about global financial stability andcommodities risk assessment.
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Future Work
Applications of other risk measures such as median shortfall(it is also elicitable), spectral risk measures.
Compare results with other tests (Acerbi and Szekely, 2014).
Other commodity assets: Commodity Leveraged-ETFs.
Afadeco November 10,2017 Risk assessment in commodity markets
MotivationModels and Methodology
ApplicationConclusion
Thank you!
Andres Mora-Valencia
Afadeco November 10,2017 Risk assessment in commodity markets