Steven Busuttil and Yuri Kalnishkan (2007) Weighted Kernel Regression for Predicting Changing Dependencies. Machine Learning: ECML 2007, 4701 (). pp. 535-542. ISSN 0302-9743
Full text access: Open
We want to make predictions in the online mode of learning for data where the dependence of the outcome y on the signal x can change with time. Standard regression techniques give all training examples the same weight; however, it is clear that older examples are less representative of the current dependency. Therefore, we require methods that consider the information content of examples to decay with time. We propose two methods for doing this: one naive and another, which is based on the Aggregating Algorithm (AA). Surprisingly these two techniques are computationally similar. To measure the empirical performance of these new methods, we perform experiments on options implied volatility data provided by the Russian Trading System Stock Exchange (RTSSE). In these experiments our methods perform better than the proprietary state-of-the-art technique currently used at the RTSSE.
This is a Draft version This version's date is: 08/09/2007 This item is peer reviewed
https://repository.royalholloway.ac.uk/items/90850a37-133a-4cdf-ba81-fb29f4a193f6/1/
Deposited by () on 30-Mar-2010 in Royal Holloway Research Online.Last modified on 06-Jan-2011
(C) 2007 Springer Verlag, whose permission to mount this version for private study and research is acknowledged. The repository version is the author's final draft.
1. Hull, J.C.: Options, Futures and Other Derivatives. 6th edn. Prentice Hall (2005)2. Vovk, V.: Competitive on-line statistics. International Statistical Review 69(2)(2001) 213–2483. Hoerl, A.E.: Application of ridge analysis to regression problems. Chemical EngineeringProgress 58 (1962) 54–594. Busuttil, S., Kalnishkan, Y., Gammerman, A.: Improving the aggregating algorithmfor regression. In: Proceedings of the 25th IASTED International Conference onArtificial Intelligence and Applications (AIA 2007), ACTA Press (2007) 347–3525. Aizerman, M., Braverman, E., Rozonoer, L.: Theoretical foundations of the potentialfunction method in pattern recognition learning. Automation and Remote Control25 (1964) 821–8376. Saunders, C., Gammerman, A., Vovk, V.: Ridge regression learning algorithm indual variables. In: Proceedings of the 15th International Conference on MachineLearning, Morgan Kaufmann (1998) 515–5217. Gammerman, A., Kalnishkan, Y., Vovk, V.: On-line prediction with kernels andthe complexity approximation principle. In: Proceedings of the 20th Conference onUncertainty in Artificial Intelligence, AUAI Press (2004) 170–1768. Sch¨olkopf, B., Smola, A.J.: Learning with Kernels — Support Vector Machines,Regularization, Optimization and Beyond. The MIT Press, USA (2002)9. Beckenbach, E.F., Bellman, R.: Inequalities. Springer (1961)