This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ForecastingRealizedVolatility
A. Nabbi
Introduction
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Forecasting Realized VolatilityUnivariate and Multivariate Heterogeneous Autoregressive
Model of Realized Volatility with Jump processes
A. Nabbi
Department of Quantitative EconomicsSchool of Business and Economics
Maastricht University
April 22, 2016
ForecastingRealizedVolatility
A. Nabbi
Introduction
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Outline
1 IntroductionFinancial Data, Models and IssuesVolatility EstimatorsVolatility Components
2 Long-Memory Models of RVHeterogeneous AR and AR Quarticity Models
3 Proposed Models
4 Forecast Measure
5 Univariate Models on S&P500Model EstimationsIn-sample Forecast Performance
6 What’s Next
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionFinancial Data, Models and Issues
Financial Data Characteristics:
Persistent Autocorrelation of square returns.
Return Distribution: fat-tailed and leptokurtic.
Slow convergence to Normal distribution.
Volatility Models:
Short-Memory Models: GARCH and SV Models.
Long-Memory Models: FIGARCH and ARFIMA.
Issues in Modeling:
Unable to reproduce data characteristics, Loss ofObservations, Lack of Economic interpretation.
Under-performance to estimate high-frequent data.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionFinancial Data, Models and Issues
Financial Data Characteristics:
Persistent Autocorrelation of square returns.
Return Distribution: fat-tailed and leptokurtic.
Slow convergence to Normal distribution.
Volatility Models:
Short-Memory Models: GARCH and SV Models.
Long-Memory Models: FIGARCH and ARFIMA.
Issues in Modeling:
Unable to reproduce data characteristics, Loss ofObservations, Lack of Economic interpretation.
Under-performance to estimate high-frequent data.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionFinancial Data, Models and Issues
Financial Data Characteristics:
Persistent Autocorrelation of square returns.
Return Distribution: fat-tailed and leptokurtic.
Slow convergence to Normal distribution.
Volatility Models:
Short-Memory Models: GARCH and SV Models.
Long-Memory Models: FIGARCH and ARFIMA.
Issues in Modeling:
Unable to reproduce data characteristics, Loss ofObservations, Lack of Economic interpretation.
Under-performance to estimate high-frequent data.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionVolatility Estimators
Daily Realized Variance
Let rt,i be high-frequency intraday return, then:
RV(d)t = RVt ≡
M∑i=1
r2t,i (1)
where ∆ = 1d/M and ∆-frequency return is defined byrt,i = log(Pt−1+i .∆)− log(Pt−1+(i−1).∆).
Jump robust estimators:
Bipower Variation: BPVt ≡ π2
(M−1M
)∑Mi=2 |rt,i ||rt,i−1|.
Median Truncated Realized Variance:MedRVt ≡ π
π−2
(M
M−1
)∑M−1i=2 Med(|rt,i+1|, |rt,i |, |rt,i−1|)2.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionVolatility Estimators
Daily Realized Variance
Let rt,i be high-frequency intraday return, then:
RV(d)t = RVt ≡
M∑i=1
r2t,i (1)
where ∆ = 1d/M and ∆-frequency return is defined byrt,i = log(Pt−1+i .∆)− log(Pt−1+(i−1).∆).
Jump robust estimators:
Bipower Variation: BPVt ≡ π2
(M−1M
)∑Mi=2 |rt,i ||rt,i−1|.
Median Truncated Realized Variance:MedRVt ≡ π
π−2
(M
M−1
)∑M−1i=2 Med(|rt,i+1|, |rt,i |, |rt,i−1|)2.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Financial Data,Models andIssues
VolatilityEstimators
VolatilityComponents
Long-MemoryModels of RV
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
IntroductionVolatility Components
Highly persistent process and consistent.
No access to Intraday data.
True long-memory processes versus simple componentmodels.
Volatility components: Short-, medium- and long-term.
Realized Variance over different time horizon
RV over time horizon h is defined by,
RV(h)t−i = RVt−i |t−h ≡ 1
h
h∑k=i
RVt−k (2)
For time horizons daily, weekly and monthly: h = 1, 5, 22.
ForecastingRealizedVolatility
A. Nabbi
Introduction
Long-MemoryModels of RV
HeterogeneousAR and ARQuarticityModels
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Long-Memory Models of Realized VolatilityHeterogeneous AR and AR Quarticity Models
Heterogeneous Autoregressive (HAR) - Corsi(2009)
The following model labeled as HAR(3)-RV which captures theapproximate long-memory dynamic dependencies conveniently.
RVt = β0 + β1RV(d)t−1 + β2RV
(w)t−1 + β3RV
(m)t−1 + ϵt (3)
RV refers to Realized Volatility.
Extensions to HAR:
HAR-J:
RVt = β0 + β1RV(d)t−1 + β2RV
(w)t−1 + β3RV
(m)t−1 + β4Jt−1 + ϵt
(4)
Continuous HAR:
RVt = β0+β1BPV(d)t−1+β2BPV
(w)t−1 +β3BPV
(m)t−1 + ϵt (5)
ForecastingRealizedVolatility
A. Nabbi
Introduction
Long-MemoryModels of RV
HeterogeneousAR and ARQuarticityModels
ProposedModels
ForecastMeasure
UnivariateModels onS&P500
What’s Next
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Long-Memory Models of Realized VolatilityHeterogeneous AR and AR Quarticity Models