• Keine Ergebnisse gefunden

Out-of-sample VaR backtesting results

Im Dokument Lévy copulae for stock returns (Seite 22-29)

3.3 Normalization and filtering of the observed data

4.1.2 Out-of-sample VaR backtesting results

The backtesting procedure conducted on the basis of the second half (another 200 days) of the available observed intraday data. The resulting failure rates with the corresponding Kupiec’s test p-values are presented in Table 2.

As we see the models with bigger T show better prediction ability in comparison with others, in particular the smallest quantiles are very well captured, the p-values 0.97 and 0.93 indicate a very strong result. This can be a consequence of the smaller amount of truncated trajectories, which leads to more flexible and accurate capability of capturing the quantiles.

Figures 1-4 present the exceedances within each of the models and con-sidered VaR-levels. All the figures display the findings of the previous table.

model/α 0.01 0.05 0.1 T = 1 0.0153 (0.4837) 0.0820 (0.0590) 0.1333 (0.1377) T = 2 0.0051 (0.4506) 0.0564 (0.6870) 0.1282 (0.2062) T = 3 0.0102 (0.9710) 0.0820 (0.0590) 0.1743 (0.0015) T = 4 0.0153 (0.4837) 0.0512 (0.9347) 0.1179 (0.4151)

Table 2: VaR performance of the IBM-Google portfolio. p-values of the Kupiec test in brackets.

It is clearly visible that the model does not exhibit a smooth quantile history, in turn responding very quickly to shocks. As on every step of the rolling window procedure the parameters of the model are re-estimated, the model possesses the ability to react rapidly to the changes. It is worth mentioning that increasing T leads to the smoothing of the behavior of the estimated VaR values, which reduces possible overestimation of the risk.

Figure 1: Exceedances plot forV aR(α) for the model withT = 1 on the IBM-Google portfolio. The dots represent the empirical portfolio value stated by the data. The three curves under the dots correspond to theα= 0.1 (yellow), α = 0.05 (green), α= 0.01 (blue)

Figure 2: Exceedances plot forV aR(α) for the model withT = 2 on the IBM-Google portfolio. The dots represent the empirical portfolio value stated by the data. The three curves under the dots correspond to theα= 0.1 (yellow), α = 0.05 (green), α= 0.01 (blue)

Figure 3: Exceedances plot forV aR(α) for the model withT = 3 on the IBM-Google portfolio. The dots represent the empirical portfolio value stated by the data. The three curves under the dots correspond to theα= 0.1 (yellow), α = 0.05 (green), α= 0.01 (blue)

Figure 4: Exceedances plot forV aR(α) for the model withT = 4 on the IBM-Google portfolio. The dots represent the empirical portfolio value stated by the data. The three curves under the dots correspond to theα= 0.1 (yellow), α = 0.05 (green), α= 0.01 (blue)

5 Conclusions

Based on assumptions of the form of the marginal tail integral and a L´evy copula family, we have introduced the model for intraday stock returns. The proposed model is dynamic, thus captures the time-varying aspects. Our empirical results demonstrate a good capability of the model for making the forecasts, in particular to predict the VaR of different levels.

The performed in this paper study shows that the more information we use for the estimation the smoother is the behavior of the simulated quantiles (VaR values) and the less becomes the amount of the simulated outliers.

However, using too much history makes the model very conservative and unable to react quickly to the incoming shocks and changes in economy, which is a serious problem, considering that we are dealing with the high frequency data. Thus, some payoff between these two features should be found in order to optimize the estimation-simulation procedure, maintaining a good capability of capturing the shocks. For the case of the intraday data it is often enough to use the observations of the last trading week.

As the obtained model allows for simulation of the arbitrary large amount of the price-paths of considered assets, one can use this property for predict-ing the value of any possible derivative of the underlypredict-ings. As well as VaR, the

systematic risk measures such as CoVaR (Adrian and Brunnermeier (2011)) could be easily calculated in order to measure the sensitivity of different financial institutions to the shocks in the economy.

References

Adrian, T. and Brunnermeier, M. K. (2011). Covar, http://www.

princeton.edu/~markus/research/papers/CoVaR.

Appelbaum, D. (2004). L´evy processes from probability to finance and quan-tum groups, Notices of the American Mathematical Society (Providence, RI: American Mathematical Society) 51(11): 1336–1347.

Appelbaum, D. (2011). Infinitely divisible central probability measures on compact lie groups - regularity, semigroups and transition kernels, The Annals of Probability 39(6): 2474–2496.

Applebaum, D. (2009). L´evy Processes and Stochastic Calculus, second edn, Cambridge University Press.

Barndorf-Nielsen, O., Hansen, P., Lunde, A. and Shepard, N. (2009). Re-alised kernels in practice: trades and quotes, The Econometrics Journal 12(3): 1–32.

Barndorff-Nielsen, O. E., T., M. and S., R. (2001). L´evy Processes: Theory and Applications, Birkh¨auser.

Barndorff-Nielsen, O. and Shephard, N. (2012). Basics of l´evy processes, http://www.people.fas.harvard.edu/~shephard/introlevy120608.

pdf. Online; accessed 28-July-2015.

Basawa, I. and Brockwell, P. (1980). A note on estimation for gamma and stable processes,Biometrika 67(1): 234–236.

Berkowitz, J. and O’Brien, J. (2002). How accurate are value-at-risk models at commercial banks?, The Journal of Finance57: 1093–1112.

Bertoin, J. (1996). L´evy processes, Cambridge University Press, Cambridge.

B¨ocker, K. and Kl¨uppelberg, C. (2008). Modelling and measuring multivari-ate operational risk with l´evy copulas, The Journal of Operational Risk 3(2): 3–27.

Breymann, W., Dias, A. and Embrechts, P. (2003). Dependence structures for multivariate high-frequency data in finance,Quantitative Finance3(1): 11–

16.

Brunnermeier, M. K., A., C., C., G., A., P. and H., S. (2009). The Fun-damental Principles of Financial Regulation: 11th Geneva Report on the World Economy, Centre for Economic Policy Research.

Campbell, S. (2006). A review of backtesting and backtesting procedures, Journal of Risk 9(2): 1–17.

Cherubini, U., Luciano, E. and Vecchiato, W. (2004). Copula Methods in Finance, John Wiley & Sons.

Cherubini, U., Mulinacci, S., Gobbi, F. and Romagnoli, S. (2011). Dynamic Copula Methods in Finance, John Wiley & Sons.

Cont, R. and Tankov, P. (2004). Financial Modelling with Jump Processes, Chapman & Hall, Boca Raton.

De Lira Salvatierra, I. A. and Patton, A. J. (2013). Dynamic copula mod-els and high frequency data, http://ssrn.com/abstract=2284235. Eco-nomic Research Initiatives at Duke (ERID) Working Paper No. 165.

Dias, A. and Embrechts, P. (2004). Dynamic copula models for multivariate high-frequency data in finance,Technical report, ETH Z¨urich.

Dionne, G. (2013). Handbook of Insurance, Springer Science & Business Media.

Esmaeili, H. and Kl¨uppelberg, C. (2011). Parametric estimation of a bivariate stable l´evy process, Journal of Multivariate Analysis 102: 918–930.

Fengler, M. and Okhrin, O. (2012). Realized copula, Economics working paper series, University of St. Gallen, School of Economics and Political Science.

Gallo, G. M., Hong, Y. and Lee, T. (2001). Modeling the impact of overnight surprises on intra-daily stock returns, Australian Economic Pa-pers40(4): 567–580.

Grothe, O. (2013). Jump tail dependence in l´evy copula models, Extremes 16(3): 303–324.

Hu, W. (2005). Calibration of multivariate generalized hyperbolic distri-butions using the em algorithm, with applications in risk management, portfolio optimization and portfolio credit risk, http://diginole.lib.

fsu.edu/etd/3694. Electronic Theses, Treatises and Dissertations. Paper 3694.

Jin, X. (2009). Large portfolio risk management with dynamic copulas, Tech-nical report, McGill University.

Joe, H. (1997). Multivariate Models and Dependence Concepts, Chapman &

Hall, London.

Jorion, P. (2006). Value at Risk, third edn, McGraw-Hill.

Kallsen, J. and Tankov, P. (2006). Characterization of dependence of mul-tidimensional l´evy processes using l´evy copulas, Journal of Multivariate Analysis 97(7): 1551–1572.

Kupiec, P. (1995). Techniques for verifying the accuracy of risk management models, Journal of Derivatives 3: 73–84.

Lockwood, L. and Linn, S. (1990). An examination of stock market return volatility during overnight and intraday periods, 19641989,The Journal of Finance 45(2): 591–601.

McNeil, A., Frey, R. and Embrechts, P. (2015). Quantitative Risk Manage-ment: Concepts, Techniques and Tools, Princeton University Press.

McNeil, A. and Neˇslehov´a, J. (2009). Multivariate archimedean copulas, d-monotone functions andl1norm symmetric distributions,Annals of Statis-tics 37(5b): 3059–3097.

Nelsen, R. (2006). An Introduction to Copulas, second edn, Springer-Verlag, New York.

Nieppola, O. (2009). Backtesting value-at-risk models, Master’s thesis, Helsinki school of economics.

Sklar, A. (1959). Fonctions de r´epartition `a n dimensions et leurs marges, Publications de l’Institut de Statistique de l’Universit´e de Paris8: 229–231.

Im Dokument Lévy copulae for stock returns (Seite 22-29)